Add vecadd-loop

This is the same kernel as vecadd but repeated in a for-loop many times so that
the runtime overhead at the startup is amortized.

.gitignore

@@ -0,0 +1,10 @@
+*.dump
+*.o
+*.bin
+*.elf
+.depend
+*.a
+*.so
+*.log
+*.vcd
+blackbox.*.cache

.gitmodules

@@ -1,9 +1,12 @@
 [submodule "third_party/fpnew"]
 	path = third_party/fpnew
-	url = https://github.com/richardyrh/cvfpu.git
+	url = https://github.com/pulp-platform/fpnew.git
 [submodule "third_party/softfloat"]
 	path = third_party/softfloat
 	url = https://github.com/ucb-bar/berkeley-softfloat-3.git
+[submodule "third_party/cocogfx"]
+	path = third_party/cocogfx
+	url = https://github.com/gtcasl/cocogfx.git
 [submodule "third_party/ramulator"]
 	path = third_party/ramulator
 	url = https://github.com/CMU-SAFARI/ramulator.git

.travis.yml

@@ -1,90 +1,76 @@
 language: cpp
-dist: focal
+dist: bionic
 os: linux
 compiler: gcc
+addons:
+    apt:
+        sources:
+        - ubuntu-toolchain-r-test
+        packages:
+        - build-essential
+        - valgrind
+        - verilator
+        - yosys        
+        - libpng-dev
+        - libboost-serialization-dev
+        - libstdc++6
 
-addons:  
-  apt:
-    packages:
-      - build-essential
-      - valgrind
-      - libstdc++6
-
-env:
-  global:
-    - TOOLDIR=$HOME/tools
-
-cache:
-  directories:
-    - $TOOLDIR
-    - $HOME/build32
-    - $HOME/build64
-
-before_install:
-  - if [ ! -d "$TOOLDIR" ] || [ -z "$(ls -A $TOOLDIR)" ]; then  
-      mkdir -p $TOOLDIR;
-      OSDIR=ubuntu/focal ./ci/toolchain_install.sh --all;
-    fi
-  - source ./ci/toolchain_env.sh
+install:
+  # Set environments  
+  - export RISCV_TOOLCHAIN_PATH=/opt/riscv-gnu-toolchain
+  - export VERILATOR_ROOT=/opt/verilator
+  - export PATH=$VERILATOR_ROOT/bin:$PATH
+  # Install toolchain
+  - ci/toolchain_install.sh -all
+  # build project
+  - make -s
 
+# stages ordering
 stages:
-  - setup
   - test
 
 jobs:     
   include:
-    - stage: setup
-      script:
-        - rm -rf $HOME/build32 && cp -r $PWD $HOME/build32
-        - rm -rf $HOME/build64 && cp -r $PWD $HOME/build64
-        - make -C $HOME/build32 > /dev/null
-        - XLEN=64 make -C $HOME/build64 > /dev/null
-    - stage: test 
-      name: unittest
-      script: cp -r $HOME/build32 build && cd build && ./ci/travis_run.py ./ci/regression.sh --unittest
-    - stage: test 
-      name: isa
-      script: cp -r $HOME/build32 build && cd build && ./ci/travis_run.py ./ci/regression.sh --isa
-    - stage: test 
-      name: isa64
-      script: cp -r $HOME/build64 build && cd build && XLEN=64 ./ci/travis_run.py ./ci/regression.sh --isa
-    - stage: test 
-      name: regression
-      script: cp -r $HOME/build32 build && cd build && ./ci/travis_run.py ./ci/regression.sh --regression
-    - stage: test 
-      name: regression64
-      script: cp -r $HOME/build64 build && cd build && XLEN=64 ./ci/travis_run.py ./ci/regression.sh --regression
-    - stage: test 
-      name: opencl
-      script: cp -r $HOME/build32 build && cd build && ./ci/travis_run.py ./ci/regression.sh --opencl    
-    - stage: test 
+    - stage: test
+      name: coverage
+      script: cp -r $PWD ../build_coverage && cd ../build_coverage && ./ci/travis_run.py ./ci/regression.sh -coverage
+    - stage: test
+      name: coverage64
+      script: cp -r $PWD ../build_coverage64 && cd ../build_coverage64 && ./ci/travis_run.py ./ci/regression64.sh -coverage
+    - stage: test
+      name: tex
+      script: cp -r $PWD ../build_tex && cd ../build_tex && ./ci/travis_run.py ./ci/regression.sh -tex
+    - stage: test
       name: cluster
-      script: cp -r $HOME/build32 build && cd build && ./ci/travis_run.py ./ci/regression.sh --cluster    
-    - stage: test 
+      script: cp -r $PWD ../build_cluster && cd ../build_cluster && ./ci/travis_run.py ./ci/regression.sh -cluster    
+    - stage: test
       name: config
-      script: cp -r $HOME/build32 build && cd build && ./ci/travis_run.py ./ci/regression.sh --config
+      script: cp -r $PWD ../build_config && cd ../build_config && ./ci/travis_run.py ./ci/regression.sh -config
     - stage: test
       name: debug
-      script: cp -r $HOME/build32 build && cd build && ./ci/travis_run.py ./ci/regression.sh --debug
-    - stage: test 
+      script: cp -r $PWD ../build_debug && cd ../build_debug && ./ci/travis_run.py ./ci/regression.sh -debug
+    - stage: test
       name: stress0
-      script: cp -r $HOME/build32 build && cd build && ./ci/travis_run.py ./ci/regression.sh --stress0
-    - stage: test 
+      script: cp -r $PWD ../build_stress0 && cd ../build_stress0 && ./ci/travis_run.py ./ci/regression.sh -stress0
+    - stage: test
       name: stress1
-      script: cp -r $HOME/build32 build && cd build && ./ci/travis_run.py ./ci/regression.sh --stress1
-    - stage: test 
-      name: synthesis
-      script: cp -r $HOME/build32 build && cd build && ./ci/travis_run.py ./ci/regression.sh --synthesis
-    - stage: test 
-      name: synthesis64
-      script: cp -r $HOME/build64 build && cd build && XLEN=64 ./ci/travis_run.py ./ci/regression.sh --synthesis
-            
+      script: cp -r $PWD ../build_stress1 && cd ../build_stress1 && ./ci/travis_run.py ./ci/regression.sh -stress1
+    - stage: test
+      name: compiler
+      script: cp -r $PWD ../build_compiler && cd ../build_compiler && ./ci/travis_run.py ./ci/test_compiler.sh
+    - stage: test
+      name: tex
+      script: cp -r $PWD ../build_tex && cd ../build_tex && ./ci/travis_run.py ./ci/regression.sh -tex
+    - stage: test
+      name: unittest
+      script: cp -r $PWD ../build_unittest && cd ../build_unittest && ./ci/travis_run.py ./ci/regression.sh -unittest
+  
 after_success:
   # Gather code coverage
-  - lcov --directory runtime --capture --output-file runtime.cov # capture trace  
-  - lcov --directory sim --capture --output-file sim.cov # capture trace    
-  - lcov --list runtime.cov # output coverage data for debugging
-  - lcov --list sim.cov # output coverage data for debugging
+  - lcov --directory driver --capture --output-file driver.cov # capture trace  
+  - lcov --directory simx --capture --output-file simx.cov # capture trace    
+  - lcov --list driver.cov # output coverage data for debugging
+  - lcov --list simx.cov # output coverage data for debugging
   # Upload coverage report
-  - bash <(curl -s https://codecov.io/bash) -f runtime.cov
-  - bash <(curl -s https://codecov.io/bash) -f sim.cov
+  - bash <(curl -s https://codecov.io/bash) -f driver.cov
+  - bash <(curl -s https://codecov.io/bash) -f simx.cov

LICENSE

@@ -1,201 +1,24 @@
-                                 Apache License
-                           Version 2.0, January 2004
-                        http://www.apache.org/licenses/
+Copyright (c) <2020>, <Georgia Institute of Technology>
+All rights reserved.
 
-   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+    * Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in the
+      documentation and/or other materials provided with the distribution.
+    * Neither the name of the Georgia Institute of Technology nor the
+      names of its contributors may be used to endorse or promote products
+      derived from this software without specific prior written permission.
 
-   1. Definitions.
-
-      "License" shall mean the terms and conditions for use, reproduction,
-      and distribution as defined by Sections 1 through 9 of this document.
-
-      "Licensor" shall mean the copyright owner or entity authorized by
-      the copyright owner that is granting the License.
-
-      "Legal Entity" shall mean the union of the acting entity and all
-      other entities that control, are controlled by, or are under common
-      control with that entity. For the purposes of this definition,
-      "control" means (i) the power, direct or indirect, to cause the
-      direction or management of such entity, whether by contract or
-      otherwise, or (ii) ownership of fifty percent (50%) or more of the
-      outstanding shares, or (iii) beneficial ownership of such entity.
-
-      "You" (or "Your") shall mean an individual or Legal Entity
-      exercising permissions granted by this License.
-
-      "Source" form shall mean the preferred form for making modifications,
-      including but not limited to software source code, documentation
-      source, and configuration files.
-
-      "Object" form shall mean any form resulting from mechanical
-      transformation or translation of a Source form, including but
-      not limited to compiled object code, generated documentation,
-      and conversions to other media types.
-
-      "Work" shall mean the work of authorship, whether in Source or
-      Object form, made available under the License, as indicated by a
-      copyright notice that is included in or attached to the work
-      (an example is provided in the Appendix below).
-
-      "Derivative Works" shall mean any work, whether in Source or Object
-      form, that is based on (or derived from) the Work and for which the
-      editorial revisions, annotations, elaborations, or other modifications
-      represent, as a whole, an original work of authorship. For the purposes
-      of this License, Derivative Works shall not include works that remain
-      separable from, or merely link (or bind by name) to the interfaces of,
-      the Work and Derivative Works thereof.
-
-      "Contribution" shall mean any work of authorship, including
-      the original version of the Work and any modifications or additions
-      to that Work or Derivative Works thereof, that is intentionally
-      submitted to Licensor for inclusion in the Work by the copyright owner
-      or by an individual or Legal Entity authorized to submit on behalf of
-      the copyright owner. For the purposes of this definition, "submitted"
-      means any form of electronic, verbal, or written communication sent
-      to the Licensor or its representatives, including but not limited to
-      communication on electronic mailing lists, source code control systems,
-      and issue tracking systems that are managed by, or on behalf of, the
-      Licensor for the purpose of discussing and improving the Work, but
-      excluding communication that is conspicuously marked or otherwise
-      designated in writing by the copyright owner as "Not a Contribution."
-
-      "Contributor" shall mean Licensor and any individual or Legal Entity
-      on behalf of whom a Contribution has been received by Licensor and
-      subsequently incorporated within the Work.
-
-   2. Grant of Copyright License. Subject to the terms and conditions of
-      this License, each Contributor hereby grants to You a perpetual,
-      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
-      copyright license to reproduce, prepare Derivative Works of,
-      publicly display, publicly perform, sublicense, and distribute the
-      Work and such Derivative Works in Source or Object form.
-
-   3. Grant of Patent License. Subject to the terms and conditions of
-      this License, each Contributor hereby grants to You a perpetual,
-      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
-      (except as stated in this section) patent license to make, have made,
-      use, offer to sell, sell, import, and otherwise transfer the Work,
-      where such license applies only to those patent claims licensable
-      by such Contributor that are necessarily infringed by their
-      Contribution(s) alone or by combination of their Contribution(s)
-      with the Work to which such Contribution(s) was submitted. If You
-      institute patent litigation against any entity (including a
-      cross-claim or counterclaim in a lawsuit) alleging that the Work
-      or a Contribution incorporated within the Work constitutes direct
-      or contributory patent infringement, then any patent licenses
-      granted to You under this License for that Work shall terminate
-      as of the date such litigation is filed.
-
-   4. Redistribution. You may reproduce and distribute copies of the
-      Work or Derivative Works thereof in any medium, with or without
-      modifications, and in Source or Object form, provided that You
-      meet the following conditions:
-
-      (a) You must give any other recipients of the Work or
-          Derivative Works a copy of this License; and
-
-      (b) You must cause any modified files to carry prominent notices
-          stating that You changed the files; and
-
-      (c) You must retain, in the Source form of any Derivative Works
-          that You distribute, all copyright, patent, trademark, and
-          attribution notices from the Source form of the Work,
-          excluding those notices that do not pertain to any part of
-          the Derivative Works; and
-
-      (d) If the Work includes a "NOTICE" text file as part of its
-          distribution, then any Derivative Works that You distribute must
-          include a readable copy of the attribution notices contained
-          within such NOTICE file, excluding those notices that do not
-          pertain to any part of the Derivative Works, in at least one
-          of the following places: within a NOTICE text file distributed
-          as part of the Derivative Works; within the Source form or
-          documentation, if provided along with the Derivative Works; or,
-          within a display generated by the Derivative Works, if and
-          wherever such third-party notices normally appear. The contents
-          of the NOTICE file are for informational purposes only and
-          do not modify the License. You may add Your own attribution
-          notices within Derivative Works that You distribute, alongside
-          or as an addendum to the NOTICE text from the Work, provided
-          that such additional attribution notices cannot be construed
-          as modifying the License.
-
-      You may add Your own copyright statement to Your modifications and
-      may provide additional or different license terms and conditions
-      for use, reproduction, or distribution of Your modifications, or
-      for any such Derivative Works as a whole, provided Your use,
-      reproduction, and distribution of the Work otherwise complies with
-      the conditions stated in this License.
-
-   5. Submission of Contributions. Unless You explicitly state otherwise,
-      any Contribution intentionally submitted for inclusion in the Work
-      by You to the Licensor shall be under the terms and conditions of
-      this License, without any additional terms or conditions.
-      Notwithstanding the above, nothing herein shall supersede or modify
-      the terms of any separate license agreement you may have executed
-      with Licensor regarding such Contributions.
-
-   6. Trademarks. This License does not grant permission to use the trade
-      names, trademarks, service marks, or product names of the Licensor,
-      except as required for reasonable and customary use in describing the
-      origin of the Work and reproducing the content of the NOTICE file.
-
-   7. Disclaimer of Warranty. Unless required by applicable law or
-      agreed to in writing, Licensor provides the Work (and each
-      Contributor provides its Contributions) on an "AS IS" BASIS,
-      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
-      implied, including, without limitation, any warranties or conditions
-      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
-      PARTICULAR PURPOSE. You are solely responsible for determining the
-      appropriateness of using or redistributing the Work and assume any
-      risks associated with Your exercise of permissions under this License.
-
-   8. Limitation of Liability. In no event and under no legal theory,
-      whether in tort (including negligence), contract, or otherwise,
-      unless required by applicable law (such as deliberate and grossly
-      negligent acts) or agreed to in writing, shall any Contributor be
-      liable to You for damages, including any direct, indirect, special,
-      incidental, or consequential damages of any character arising as a
-      result of this License or out of the use or inability to use the
-      Work (including but not limited to damages for loss of goodwill,
-      work stoppage, computer failure or malfunction, or any and all
-      other commercial damages or losses), even if such Contributor
-      has been advised of the possibility of such damages.
-
-   9. Accepting Warranty or Additional Liability. While redistributing
-      the Work or Derivative Works thereof, You may choose to offer,
-      and charge a fee for, acceptance of support, warranty, indemnity,
-      or other liability obligations and/or rights consistent with this
-      License. However, in accepting such obligations, You may act only
-      on Your own behalf and on Your sole responsibility, not on behalf
-      of any other Contributor, and only if You agree to indemnify,
-      defend, and hold each Contributor harmless for any liability
-      incurred by, or claims asserted against, such Contributor by reason
-      of your accepting any such warranty or additional liability.
-
-   END OF TERMS AND CONDITIONS
-
-   APPENDIX: How to apply the Apache License to your work.
-
-      To apply the Apache License to your work, attach the following
-      boilerplate notice, with the fields enclosed by brackets "[]"
-      replaced with your own identifying information. (Don't include
-      the brackets!)  The text should be enclosed in the appropriate
-      comment syntax for the file format. We also recommend that a
-      file or class name and description of purpose be included on the
-      same "printed page" as the copyright notice for easier
-      identification within third-party archives.
-
-   Copyright [yyyy] [name of copyright owner]
-
-   Licensed under the Apache License, Version 2.0 (the "License");
-   you may not use this file except in compliance with the License.
-   You may obtain a copy of the License at
-
-       http://www.apache.org/licenses/LICENSE-2.0
-
-   Unless required by applicable law or agreed to in writing, software
-   distributed under the License is distributed on an "AS IS" BASIS,
-   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-   See the License for the specific language governing permissions and
-   limitations under the License.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL COPYRIGHT HOLDER BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Makefile

@@ -2,27 +2,13 @@ all:
 	$(MAKE) -C third_party
 	$(MAKE) -C hw
 	$(MAKE) -C sim
-	$(MAKE) -C kernel
+	$(MAKE) -C driver
 	$(MAKE) -C runtime
 	$(MAKE) -C tests
 
 clean:
 	$(MAKE) -C hw clean
 	$(MAKE) -C sim clean
-	$(MAKE) -C kernel clean
+	$(MAKE) -C driver clean
 	$(MAKE) -C runtime clean
-	$(MAKE) -C tests clean
-
-clean-all:
-	$(MAKE) -C third_party clean
-	$(MAKE) -C hw clean
-	$(MAKE) -C sim clean
-	$(MAKE) -C kernel clean
-	$(MAKE) -C runtime clean
-	$(MAKE) -C tests clean-all
-
-crtlsim:
-	$(MAKE) -C sim clean
-
-brtlsim:
-	$(MAKE) -C sim
+	$(MAKE) -C tests clean

README.md

@@ -1,25 +1,22 @@
 [![Build Status](https://travis-ci.com/vortexgpgpu/vortex.svg?branch=master)](https://travis-ci.com/vortexgpgpu/vortex)
 [![codecov](https://codecov.io/gh/vortexgpgpu/vortex/branch/master/graph/badge.svg)](https://codecov.io/gh/vortexgpgpu/vortex)
 
-# Vortex GPGPU
+# Vortex OpenGPU
 
-Vortex is a full-stack open-source RISC-V GPGPU.
+Vortex is a full-system RISCV-based GPGPU processor.
 
 ## Specifications
 
-- Support RISC-V RV32IMAF and RV64IMAFD
-- Microarchitecture:
-    - configurable number of cores, warps, and threads.
-    - configurable number of ALU, FPU, LSU, and SFU units per core.
-    - configurable pipeline issue width.
-    - optional shared memory, L1, L2, and L3 caches.
-- Software: 
-    - OpenCL 1.2 Support.
+- Support RISC-V RV32IMF ISA
+- Performance: 
+    - 1024 total threads running at 250 MHz
+    - 128 Gflops of compute bandwidth
+    - 16 GB/s of memory bandwidth
+- Scalability: up to 64 cores with optional L2 and L3 caches
+- Software: OpenCL 1.2 Support 
 - Supported FPGAs: 
-    - Altera Arria 10
-    - Altera Stratix 10
-    - Xilinx Alveo U50, U250, U280
-    - Xilinx Versal VCK5000
+    - Intel Arria 10
+    - Intel Stratix 10
 
 ## Directory structure
 
@@ -33,20 +30,14 @@ Vortex is a full-stack open-source RISC-V GPGPU.
 - `miscs`: Miscellaneous resources.
 
 ## Build Instructions
-More detailed build instructions can be found [here](docs/install_vortex.md).
 ### Supported OS Platforms
-- Ubuntu 18.04, 20.04
+- Ubuntu 18.04
 - Centos 7
 ### Toolchain Dependencies
 - [POCL](http://portablecl.org/)
 - [LLVM](https://llvm.org/)
 - [RISCV-GNU-TOOLCHAIN](https://github.com/riscv-collab/riscv-gnu-toolchain)
 - [Verilator](https://www.veripool.org/verilator)
-- [FpNew](https://github.com/pulp-platform/fpnew.git)
-- [SoftFloat](https://github.com/ucb-bar/berkeley-softfloat-3.git)
-- [Ramulator](https://github.com/CMU-SAFARI/ramulator.git)
-- [Yosys](https://github.com/YosysHQ/yosys)
-- [Sv2v](https://github.com/zachjs/sv2v)
 ### Install development tools 
     $ sudo apt-get install build-essential
     $ sudo apt-get install git
@@ -54,12 +45,8 @@ More detailed build instructions can be found [here](docs/install_vortex.md).
     $ git clone --recursive https://github.com/vortexgpgpu/vortex.git
     $ cd Vortex
 ### Install prebuilt toolchain
-    By default, the toolchain will install to /opt folder which requires sudo access. 
-    You can install the toolchain to a different location of your choice by setting TOOLDIR (e.g. export TOOLDIR=$HOME/tools).
-    $ export TOOLDIR=/opt
-    $ ./ci/toolchain_install.sh --all
-    $ source ./ci/toolchain_env.sh
+    $ ./ci/toolchain_install.sh -all
 ### Build Vortex sources
     $ make -s
 ### Quick demo running vecadd OpenCL kernel on 2 cores
-    $ ./ci/blackbox.sh --cores=2 --app=vecadd
+    $ ./ci/blackbox.sh --driver=rtlsim --cores=2 --app=vecadd

RELEASE

@@ -0,0 +1,4 @@
+
+Release Notes!
+
+* 07/01/2020 - LKG FPGA build - Passed basic, demo, vecadd kernels.

TODO

@@ -0,0 +1,23 @@
+
+
+
+Functionality:
+1) vx_cl_warpSpawn()
+   -> To be used by pocl->ops->run
+
+2) newlib Integration (LoadFile(""))
+   -> To be used by the Rhinio benchmarks
+
+3) POCL OPS Vortex Suite
+
+Performance:
+1) Icache doesn't need SEND_MEM_REQUEST Stage
+   -> Blocks are never dirty, so why not evict right away 
+
+2) Branch not taken speculation
+
+3) Runtime -02 not running on RTL, and -03 not running on RTL and Emulator
+
+
+Vector:
+1) Cycle accurate simulator (would require Cache Simulator)

ci/blackbox.sh

@@ -1,53 +1,26 @@
 #!/bin/sh
 
-# Copyright © 2019-2023
-# 
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-# http://www.apache.org/licenses/LICENSE-2.0
-# 
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
 show_usage()
 {
     echo "Vortex BlackBox Test Driver v1.0"
-    echo "Usage: $0 [[--clusters=#n] [--cores=#n] [--warps=#n] [--threads=#n] [--l2cache] [--l3cache] [[--driver=#name] [--app=#app] [--args=#args] [--debug=#level] [--scope] [--perf=#class] [--rebuild=#n] [--log=logfile] [--help]]"
-}
-
-show_help()
-{
-    show_usage
-    echo "  where"
-    echo "--driver: simx, rtlsim, oape, xrt"
-    echo "--app: any subfolder test under regression or opencl"
-    echo "--class: 0=disable, 1=pipeline, 2=memsys"
-    echo "--rebuild: 0=disable, 1=force, 2=auto, 3=temp"
+    echo "Usage: [[--clusters=#n] [--cores=#n] [--warps=#n] [--threads=#n] [--l2cache] [--l3cache] [[--driver=rtlsim|vlsim|simx] [--debug] [--scope] [--perf] [--app=vecadd|sgemm|basic|demo|dogfood] [--args=<args>] [--help]]"
 }
 
 SCRIPT_DIR=$(dirname "$0")
 VORTEX_HOME=$SCRIPT_DIR/..
 
-DRIVER=simx
+DRIVER=vlsim
 APP=sgemm
 CLUSTERS=1
 CORES=1
 WARPS=4
 THREADS=4
-L2=
-L3=
+L2=0
+L3=0
 DEBUG=0
-DEBUG_LEVEL=0
 SCOPE=0
 HAS_ARGS=0
-PERF_CLASS=0
-REBUILD=2
-TEMPBUILD=0
-LOGFILE=run.log
+DEBUG_LEVEL=1
 
 for i in "$@"
 do
@@ -77,15 +50,14 @@ case $i in
         shift
         ;;
     --l2cache)
-        L2=-DL2_ENABLE
+        L2=1
         shift
         ;;
     --l3cache)
-        L3=-DL3_ENABLE
+        L3=1
         shift
         ;;
-    --debug=*)
-        DEBUG_LEVEL=${i#*=}
+    --debug)
         DEBUG=1
         shift
         ;;
@@ -94,9 +66,8 @@ case $i in
         CORES=1        
         shift
         ;;
-    --perf=*)
+    --perf)
         PERF_FLAG=-DPERF_ENABLE
-        PERF_CLASS=${i#*=}    
         shift
         ;;
     --args=*)
@@ -104,43 +75,33 @@ case $i in
         HAS_ARGS=1
         shift
         ;;
-    --rebuild=*)
-        REBUILD=${i#*=}
-        shift
-        ;;
-    --log=*)
-        LOGFILE=${i#*=}
-        shift
-        ;;
     --help)
-        show_help
+        show_usage
         exit 0
         ;;
     *)
-        show_usage   
-        exit -1       
-        ;;
+    show_usage   
+    exit -1       
+    ;;
 esac
 done
 
-if [ $REBUILD -eq 3 ];
-then
-    REBUILD=1
-    TEMPBUILD=1
-fi
-
 case $DRIVER in
-    simx)
-        DRIVER_PATH=$VORTEX_HOME/runtime/simx
-        ;;
     rtlsim)
-        DRIVER_PATH=$VORTEX_HOME/runtime/rtlsim
+        DRIVER_PATH=$VORTEX_HOME/driver/rtlsim
         ;;
-    opae)
-        DRIVER_PATH=$VORTEX_HOME/runtime/opae
+    vlsim)
+        DRIVER_PATH=$VORTEX_HOME/driver/vlsim
         ;;
-    xrt)
-        DRIVER_PATH=$VORTEX_HOME/runtime/xrt
+    asesim)
+        DRIVER_PATH=$VORTEX_HOME/driver/asesim
+        ;;
+    fpga)
+        DRIVER_PATH=$VORTEX_HOME/driver/fpga
+        ;; 
+    simx)
+        DRIVER_PATH=$VORTEX_HOME/driver/simx
+        DEBUG_LEVEL=3
         ;;
     *)
         echo "invalid driver: $DRIVER"
@@ -155,156 +116,78 @@ elif [ -d "$VORTEX_HOME/tests/regression/$APP" ];
 then
     APP_PATH=$VORTEX_HOME/tests/regression/$APP
 else
-    echo "Application folder not found: $APP"
+    echo "Application folder found: $APP"
     exit -1
 fi
 
-CONFIGS="-DNUM_CLUSTERS=$CLUSTERS -DNUM_CORES=$CORES -DNUM_WARPS=$WARPS -DNUM_THREADS=$THREADS $L2 $L3 $PERF_FLAG $CONFIGS"
+CONFIGS="-DNUM_CLUSTERS=$CLUSTERS -DNUM_CORES=$CORES -DNUM_WARPS=$WARPS -DNUM_THREADS=$THREADS -DL2_ENABLE=$L2 -DL3_ENABLE=$L3 $PERF_FLAG $CONFIGS"
 
 echo "CONFIGS=$CONFIGS"
 
-if [ $REBUILD -ne 0 ] 
-then
-    BLACKBOX_CACHE=blackbox.$DRIVER.cache    
-    if [ -f "$BLACKBOX_CACHE" ]
-    then 
-        LAST_CONFIGS=`cat $BLACKBOX_CACHE`
-    fi
+BLACKBOX_CACHE=blackbox.$DRIVER.cache
 
-    if [ $REBUILD -eq 1 ] || [ "$CONFIGS+$DEBUG+$SCOPE" != "$LAST_CONFIGS" ];
-    then
-        make -C $DRIVER_PATH clean > /dev/null
-        echo "$CONFIGS+$DEBUG+$SCOPE" > $BLACKBOX_CACHE
-    fi
+if [ -f "$BLACKBOX_CACHE" ]
+then 
+    LAST_CONFIGS=`cat $BLACKBOX_CACHE`
 fi
 
-# export performance monitor class identifier
-export PERF_CLASS=$PERF_CLASS
+if [ "$CONFIGS+$DEBUG+$SCOPE" != "$LAST_CONFIGS" ]; 
+then
+    make -C $DRIVER_PATH clean
+fi
+
+echo "$CONFIGS+$DEBUG+$SCOPE" > $BLACKBOX_CACHE
 
 status=0
 
-# ensure config update
-make -C $VORTEX_HOME/hw config > /dev/null
-
-# ensure the stub driver is present
-make -C $VORTEX_HOME/runtime/stub > /dev/null
-
-if [ $DEBUG -ne 0 ]
+if [ $DEBUG -eq 1 ]
 then    
-    # running application
-    if [ $TEMPBUILD -eq 1 ]
+    if [ $SCOPE -eq 1 ]
     then
-        # setup temp directory
-        TEMPDIR=$(mktemp -d)
-        mkdir -p "$TEMPDIR/$DRIVER"
-
-        # driver initialization
-        if [ $SCOPE -eq 1 ]
-        then
-            echo "running: DESTDIR=$TEMPDIR/$DRIVER DEBUG=$DEBUG_LEVEL SCOPE=1 CONFIGS=$CONFIGS make -C $DRIVER_PATH"
-            DESTDIR="$TEMPDIR/$DRIVER" DEBUG=$DEBUG_LEVEL SCOPE=1 CONFIGS="$CONFIGS" make -C $DRIVER_PATH > /dev/null
-        else
-            echo "running: DESTDIR=$TEMPDIR/$DRIVER DEBUG=$DEBUG_LEVEL CONFIGS=$CONFIGS make -C $DRIVER_PATH"
-            DESTDIR="$TEMPDIR/$DRIVER" DEBUG=$DEBUG_LEVEL CONFIGS="$CONFIGS" make -C $DRIVER_PATH > /dev/null
-        fi
-
-        # running application
-        if [ $HAS_ARGS -eq 1 ]
-        then
-            echo "running: VORTEX_RT_PATH=$TEMPDIR OPTS=$ARGS make -C $APP_PATH run-$DRIVER > $LOGFILE 2>&1"
-            VORTEX_RT_PATH=$TEMPDIR OPTS=$ARGS make -C $APP_PATH run-$DRIVER > $LOGFILE 2>&1
-            status=$?
-        else
-            echo "running: VORTEX_RT_PATH=$TEMPDIR make -C $APP_PATH run-$DRIVER > $LOGFILE 2>&1"
-            VORTEX_RT_PATH=$TEMPDIR make -C $APP_PATH run-$DRIVER > $LOGFILE 2>&1
-            status=$?
-        fi
-
-        # cleanup temp directory
-        trap "rm -rf $TEMPDIR" EXIT
+        echo "running: DEBUG=$DEBUG_LEVEL SCOPE=1 CONFIGS="$CONFIGS" make -C $DRIVER_PATH"
+        DEBUG=$DEBUG_LEVEL SCOPE=1 CONFIGS="$CONFIGS" make -C $DRIVER_PATH
     else
-        # driver initialization
-        if [ $SCOPE -eq 1 ]
-        then
-            echo "running: DEBUG=$DEBUG_LEVEL SCOPE=1 CONFIGS=$CONFIGS make -C $DRIVER_PATH"
-            DEBUG=$DEBUG_LEVEL SCOPE=1 CONFIGS="$CONFIGS" make -C $DRIVER_PATH > /dev/null
-        else
-            echo "running: DEBUG=$DEBUG_LEVEL CONFIGS=$CONFIGS make -C $DRIVER_PATH"
-            DEBUG=$DEBUG_LEVEL CONFIGS="$CONFIGS" make -C $DRIVER_PATH > /dev/null
-        fi
-
-        # running application
-        if [ $HAS_ARGS -eq 1 ]
-        then
-            echo "running: OPTS=$ARGS make -C $APP_PATH run-$DRIVER > $LOGFILE 2>&1"
-            OPTS=$ARGS make -C $APP_PATH run-$DRIVER > $LOGFILE 2>&1
-            status=$?
-        else
-            echo "running: make -C $APP_PATH run-$DRIVER > $LOGFILE 2>&1"
-            make -C $APP_PATH run-$DRIVER > $LOGFILE 2>&1
-            status=$?
-        fi
+        echo "running: DEBUG=$DEBUG_LEVEL CONFIGS="$CONFIGS" make -C $DRIVER_PATH"
+        DEBUG=$DEBUG_LEVEL CONFIGS="$CONFIGS" make -C $DRIVER_PATH
+    fi    
+    
+    if [ $HAS_ARGS -eq 1 ]
+    then
+        echo "running: OPTS=$ARGS make -C $APP_PATH run-$DRIVER > run.log 2>&1"
+        OPTS=$ARGS make -C $APP_PATH run-$DRIVER > run.log 2>&1
+        status=$?
+    else
+        echo "running: make -C $APP_PATH run-$DRIVER > run.log 2>&1"
+        make -C $APP_PATH run-$DRIVER > run.log 2>&1
+        status=$?
     fi
     
     if [ -f "$APP_PATH/trace.vcd" ]
     then 
         mv -f $APP_PATH/trace.vcd .
     fi
-else    
-    if [ $TEMPBUILD -eq 1 ]
+else
+    echo "driver initialization..."
+    if [ $SCOPE -eq 1 ]
     then
-        # setup temp directory
-        TEMPDIR=$(mktemp -d)
-        mkdir -p "$TEMPDIR/$DRIVER"
-        
-        # driver initialization
-        if [ $SCOPE -eq 1 ]
-        then
-            echo "running: DESTDIR=$TEMPDIR/$DRIVER SCOPE=1 CONFIGS=$CONFIGS make -C $DRIVER_PATH"
-            DESTDIR="$TEMPDIR/$DRIVER" SCOPE=1 CONFIGS="$CONFIGS" make -C $DRIVER_PATH > /dev/null
-        else
-            echo "running: DESTDIR=$TEMPDIR/$DRIVER CONFIGS=$CONFIGS make -C $DRIVER_PATH"
-            DESTDIR="$TEMPDIR/$DRIVER" CONFIGS="$CONFIGS" make -C $DRIVER_PATH > /dev/null
-        fi
-        
-        # running application
-        if [ $HAS_ARGS -eq 1 ]
-        then
-            echo "running: VORTEX_RT_PATH=$TEMPDIR OPTS=$ARGS make -C $APP_PATH run-$DRIVER"
-            VORTEX_RT_PATH=$TEMPDIR OPTS=$ARGS make -C $APP_PATH run-$DRIVER
-            status=$?
-        else
-            echo "running: VORTEX_RT_PATH=$TEMPDIR make -C $APP_PATH run-$DRIVER"
-            VORTEX_RT_PATH=$TEMPDIR make -C $APP_PATH run-$DRIVER
-            status=$?
-        fi
-
-        # cleanup temp directory
-        trap "rm -rf $TEMPDIR" EXIT
+        echo "running: SCOPE=1 CONFIGS="$CONFIGS" make -C $DRIVER_PATH"
+        SCOPE=1 CONFIGS="$CONFIGS" make -C $DRIVER_PATH
     else
-        
-        # driver initialization
-        if [ $SCOPE -eq 1 ]
-        then
-            echo "running: SCOPE=1 CONFIGS=$CONFIGS make -C $DRIVER_PATH"
-            SCOPE=1 CONFIGS="$CONFIGS" make -C $DRIVER_PATH > /dev/null
-        else
-            echo "running: CONFIGS=$CONFIGS make -C $DRIVER_PATH"
-            CONFIGS="$CONFIGS" make -C $DRIVER_PATH > /dev/null
-        fi
-
-        # running application
-        if [ $HAS_ARGS -eq 1 ]
-        then
-            echo "running: OPTS=$ARGS make -C $APP_PATH run-$DRIVER"
-            OPTS=$ARGS make -C $APP_PATH run-$DRIVER
-            status=$?
-        else
-            echo "running: make -C $APP_PATH run-$DRIVER"
-            make -C $APP_PATH run-$DRIVER
-            status=$?
-        fi
+        echo "running: CONFIGS="$CONFIGS" make -C $DRIVER_PATH"
+        CONFIGS="$CONFIGS" make -C $DRIVER_PATH
+    fi
+    
+    echo "running application..."
+    if [ $HAS_ARGS -eq 1 ]
+    then
+        echo "running: OPTS=$ARGS make -C $APP_PATH run-$DRIVER"
+        OPTS=$ARGS make -C $APP_PATH run-$DRIVER
+        status=$?
+    else
+        echo "running: make -C $APP_PATH run-$DRIVER"
+        make -C $APP_PATH run-$DRIVER
+        status=$?
     fi
 fi
 
-exit $status
+exit $status

ci/prebuilt.sh

@@ -0,0 +1,73 @@
+#!/bin/bash
+
+# exit when any command fails
+set -e
+
+OS_DIR=${OS_DIR:-'ubuntu/bionic'}
+SRCDIR=${SRCDIR:-'/opt'}
+DESTDIR=${DESTDIR:-'.'}
+
+echo "OS_DIR=${OS_DIR}"
+echo "SRCDIR=${SRCDIR}"
+echo "DESTDIR=${DESTDIR}"
+
+riscv() 
+{
+    echo "prebuilt riscv-gnu-toolchain..."
+    tar -C $SRCDIR -cvjf riscv-gnu-toolchain.tar.bz2 riscv-gnu-toolchain
+    split -b 50M riscv-gnu-toolchain.tar.bz2 "riscv-gnu-toolchain.tar.bz2.part"    
+    mv riscv-gnu-toolchain.tar.bz2.part* $DESTDIR/riscv-gnu-toolchain/$OS_DIR
+    rm riscv-gnu-toolchain.tar.bz2
+}
+
+llvm() 
+{
+    echo "prebuilt llvm-riscv..."
+    tar -C $SRCDIR -cvjf llvm-vortex1.tar.bz2 llvm-riscv
+    split -b 50M llvm-vortex1.tar.bz2 "llvm-vortex1.tar.bz2.part"    
+    mv llvm-vortex1.tar.bz2.part* $DESTDIR/llvm-vortex/$OS_DIR
+    rm llvm-vortex1.tar.bz2
+}
+
+pocl() 
+{
+    echo "prebuilt pocl..."
+    tar -C $SRCDIR -cvjf pocl1.tar.bz2 pocl
+    mv pocl1.tar.bz2 $DESTDIR/pocl/$OS_DIR
+}
+
+verilator() 
+{
+    echo "prebuilt verilator..."
+    tar -C $SRCDIR -cvjf verilator.tar.bz2 verilator
+    mv verilator.tar.bz2 $DESTDIR/verilator/$OS_DIR
+}
+
+usage()
+{
+    echo "usage: prebuilt [[-riscv] [-llvm] [-pocl] [-verilator] [-all] [-h|--help]]"
+}
+
+while [ "$1" != "" ]; do
+    case $1 in
+        -pocl ) pocl
+                ;;
+        -verilator ) verilator
+                     ;;
+        -riscv ) riscv
+                 ;;
+        -llvm ) llvm
+                ;;
+        -all ) riscv
+               llvm
+               pocl
+               verilator
+               ;;
+        -h | --help ) usage
+                      exit
+                      ;;
+        * )           usage
+                      exit 1
+    esac
+    shift
+done

ci/regression.sh

@@ -1,106 +1,44 @@
 #!/bin/bash
 
-# Copyright © 2019-2023
-# 
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-# http://www.apache.org/licenses/LICENSE-2.0
-# 
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
 # exit when any command fails
 set -e
 
-# clear blackbox cache
-rm -f blackbox.*.cache
+# ensure build
+make -s
 
 unittest() 
 {
 make -C tests/unittest run
-make -C hw/unittest > /dev/null
 }
 
-isa() 
+coverage() 
 {
-echo "begin isa tests..."
+echo "begin coverage tests..."
 
+make -C tests/runtime run-rtlsim
+make -C tests/riscv/isa run-rtlsim
+make -C tests/regression run-vlsim
+make -C tests/opencl run-vlsim
+make -C tests/runtime run-simx
 make -C tests/riscv/isa run-simx
-make -C tests/riscv/isa run-rtlsim
-
-make -C sim/rtlsim clean && CONFIGS="-DDPI_DISABLE" make -C sim/rtlsim > /dev/null
-make -C tests/riscv/isa run-rtlsim
-
-make -C sim/rtlsim clean && CONFIGS="-DFPU_FPNEW" make -C sim/rtlsim > /dev/null
-make -C tests/riscv/isa run-rtlsim-32f
-
-make -C sim/rtlsim clean && CONFIGS="-DFPU_DPI" make -C sim/rtlsim > /dev/null
-make -C tests/riscv/isa run-rtlsim-32f
-
-make -C sim/rtlsim clean && CONFIGS="-DFPU_DSP" make -C sim/rtlsim > /dev/null
-make -C tests/riscv/isa run-rtlsim-32f
-
-if [ "$XLEN" == "64" ]
-then
-        make -C sim/rtlsim clean && CONFIGS="-DFPU_FPNEW" make -C sim/rtlsim > /dev/null
-        make -C tests/riscv/isa run-rtlsim-64f
-
-        make -C sim/rtlsim clean && CONFIGS="-DEXT_D_ENABLE -DFPU_FPNEW" make -C sim/rtlsim > /dev/null
-        make -C tests/riscv/isa run-rtlsim-64d || true
-
-        make -C sim/rtlsim clean && CONFIGS="-DFPU_DPI" make -C sim/rtlsim > /dev/null
-        make -C tests/riscv/isa run-rtlsim-64f
-
-        make -C sim/rtlsim clean && CONFIGS="-DFPU_DSP" make -C sim/rtlsim > /dev/null
-        make -C tests/riscv/isa run-rtlsim-64fx
-fi
-
-# restore default prebuilt configuration
-make -C sim/rtlsim clean && make -C sim/rtlsim > /dev/null
-
-echo "isa tests done!"
-}
-
-regression() 
-{
-echo "begin regression tests..."
-
-make -C tests/kernel run-simx
-make -C tests/kernel run-rtlsim
-
 make -C tests/regression run-simx
-make -C tests/regression run-rtlsim
+make -C tests/opencl run-simx
 
-# test FPU hardware implementations
-CONFIGS="-DFPU_DPI" ./ci/blackbox.sh --driver=rtlsim --app=dogfood
-CONFIGS="-DFPU_DSP" ./ci/blackbox.sh --driver=rtlsim --app=dogfood
-CONFIGS="-DFPU_FPNEW" ./ci/blackbox.sh --driver=rtlsim --app=dogfood
-
-# test local barrier
-./ci/blackbox.sh --driver=simx --app=dogfood --args="-n1 -t19"
-./ci/blackbox.sh --driver=rtlsim --app=dogfood --args="-n1 -t19"
-
-# test global barrier
-CONFIGS="-DGBAR_ENABLE" ./ci/blackbox.sh --driver=simx --app=dogfood --args="-n1 -t20" --cores=2
-CONFIGS="-DGBAR_ENABLE" ./ci/blackbox.sh --driver=rtlsim --app=dogfood --args="-n1 -t20" --cores=2
-
-# test FPU core
-
-echo "regression tests done!"
+echo "coverage tests done!"
 }
 
-opencl() 
+tex()
 {
-echo "begin opencl tests..."
+echo "begin texture tests..."
 
-make -C tests/opencl run-simx
-make -C tests/opencl run-rtlsim
+CONFIGS="-DEXT_TEX_ENABLE=1" ./ci/blackbox.sh --driver=vlsim --app=tex --args="-isoccer.png -osoccer_result.png -g0"
+CONFIGS="-DEXT_TEX_ENABLE=1" ./ci/blackbox.sh --driver=simx --app=tex --args="-isoccer.png -osoccer_result.png -g0"
+CONFIGS="-DEXT_TEX_ENABLE=1" ./ci/blackbox.sh --driver=rtlsim --app=tex --args="-itoad.png -otoad_result.png -g1"
+CONFIGS="-DEXT_TEX_ENABLE=1" ./ci/blackbox.sh --driver=simx --app=tex --args="-irainbow.png -orainbow_result.png -g2"
+CONFIGS="-DEXT_TEX_ENABLE=1" ./ci/blackbox.sh --driver=rtlsim --app=tex --args="-itoad.png -otoad_result.png -g1" --perf
+CONFIGS="-DEXT_TEX_ENABLE=1" ./ci/blackbox.sh --driver=simx --app=tex --args="-itoad.png -otoad_result.png -g1" --perf
 
-echo "opencl tests done!"
+echo "coverage texture done!"
 }
 
 cluster() 
@@ -108,26 +46,23 @@ cluster()
 echo "begin clustering tests..."
 
 # warp/threads configurations
-./ci/blackbox.sh --driver=rtlsim --cores=1 --warps=1 --threads=1 --app=diverge
-./ci/blackbox.sh --driver=rtlsim --cores=1 --warps=2 --threads=2 --app=diverge
-./ci/blackbox.sh --driver=rtlsim --cores=1 --warps=2 --threads=8 --app=diverge
-./ci/blackbox.sh --driver=rtlsim --cores=1 --warps=8 --threads=2 --app=diverge
-./ci/blackbox.sh --driver=simx --cores=1 --warps=1 --threads=1 --app=diverge
-./ci/blackbox.sh --driver=simx --cores=1 --warps=8 --threads=16 --app=diverge
+./ci/blackbox.sh --driver=rtlsim --cores=1 --warps=2 --threads=8 --app=demo
+./ci/blackbox.sh --driver=rtlsim --cores=1 --warps=8 --threads=2 --app=demo
+./ci/blackbox.sh --driver=simx --cores=1 --warps=8 --threads=16 --app=demo
 
 # cores clustering
-./ci/blackbox.sh --driver=rtlsim --cores=1 --clusters=1 --app=diverge --args="-n1"
-./ci/blackbox.sh --driver=rtlsim --cores=4 --clusters=1 --app=diverge --args="-n1"
-./ci/blackbox.sh --driver=rtlsim --cores=2 --clusters=2 --app=diverge --args="-n1"
-./ci/blackbox.sh --driver=simx --cores=4 --clusters=1 --app=diverge --args="-n1"
-./ci/blackbox.sh --driver=simx --cores=4 --clusters=2 --app=diverge --args="-n1"
+./ci/blackbox.sh --driver=rtlsim --cores=1 --clusters=1 --app=demo --args="-n1"
+./ci/blackbox.sh --driver=rtlsim --cores=4 --clusters=1 --app=demo --args="-n1"
+./ci/blackbox.sh --driver=rtlsim --cores=2 --clusters=2 --app=demo --args="-n1"
+./ci/blackbox.sh --driver=simx --cores=4 --clusters=1 --app=demo --args="-n1"
+./ci/blackbox.sh --driver=simx --cores=4 --clusters=2 --app=demo --args="-n1"
 
 # L2/L3
-./ci/blackbox.sh --driver=rtlsim --cores=2 --l2cache --app=diverge --args="-n1"
-./ci/blackbox.sh --driver=rtlsim --cores=2 --clusters=2 --l3cache --app=diverge --args="-n1"
+./ci/blackbox.sh --driver=rtlsim --cores=2 --l2cache --app=demo --args="-n1"
+./ci/blackbox.sh --driver=rtlsim --cores=2 --clusters=2 --l3cache --app=demo --args="-n1"
 ./ci/blackbox.sh --driver=rtlsim --cores=2 --clusters=2 --l2cache --l3cache --app=io_addr --args="-n1"
-./ci/blackbox.sh --driver=simx --cores=4 --clusters=2 --l2cache --app=diverge --args="-n1"
-./ci/blackbox.sh --driver=simx --cores=4 --clusters=4 --l2cache --l3cache --app=diverge --args="-n1"
+./ci/blackbox.sh --driver=simx --cores=4 --clusters=2 --l2cache --app=demo --args="-n1"
+./ci/blackbox.sh --driver=simx --cores=4 --clusters=4 --l2cache --l3cache --app=demo --args="-n1"
 
 echo "clustering tests done!"
 }
@@ -136,23 +71,11 @@ debug()
 {
 echo "begin debugging tests..."
 
-# test CSV trace generation
-make -C sim/simx clean && DEBUG=3 make -C sim/simx > /dev/null
-make -C sim/rtlsim clean && DEBUG=3 CONFIGS="-DGPR_RESET" make -C sim/rtlsim > /dev/null
-make -C tests/riscv/isa run-simx-32im > run_simx.log
-make -C tests/riscv/isa run-rtlsim-32im > run_rtlsim.log
-./ci/trace_csv.py -trtlsim run_rtlsim.log -otrace_rtlsim.csv
-./ci/trace_csv.py -tsimx run_simx.log -otrace_simx.csv
-diff trace_rtlsim.csv trace_simx.csv
-# restore default prebuilt configuration
-make -C sim/simx clean && make -C sim/simx > /dev/null
-make -C sim/rtlsim clean && make -C sim/rtlsim > /dev/null
-
-./ci/blackbox.sh --driver=opae --cores=2 --clusters=2 --l2cache --perf=1 --app=demo --args="-n1"
-./ci/blackbox.sh --driver=simx --cores=2 --clusters=2 --l2cache --perf=1 --app=demo --args="-n1"
-./ci/blackbox.sh --driver=opae --cores=2 --clusters=2 --l2cache --debug=1 --app=demo --args="-n1"
-./ci/blackbox.sh --driver=simx --cores=2 --clusters=2 --l2cache --debug=1 --app=demo --args="-n1"
-./ci/blackbox.sh --driver=opae --cores=1 --scope --app=basic --args="-t0 -n1"
+./ci/blackbox.sh --driver=vlsim --cores=2 --clusters=2 --l2cache --perf --app=demo --args="-n1"
+./ci/blackbox.sh --driver=simx --cores=2 --clusters=2 --l2cache --perf --app=demo --args="-n1"
+./ci/blackbox.sh --driver=vlsim --cores=2 --clusters=2 --l2cache --debug --app=demo --args="-n1"
+./ci/blackbox.sh --driver=simx --cores=2 --clusters=2 --l2cache --debug --app=demo --args="-n1"
+./ci/blackbox.sh --driver=vlsim --cores=1 --scope --app=basic --args="-t0 -n1"
 
 echo "debugging tests done!"
 }
@@ -161,77 +84,51 @@ config()
 {
 echo "begin configuration tests..."
 
-# disable DPI
-CONFIGS="-DDPI_DISABLE -DFPU_FPNEW" ./ci/blackbox.sh --driver=rtlsim --app=dogfood
-CONFIGS="-DDPI_DISABLE -DFPU_FPNEW" ./ci/blackbox.sh --driver=opae --app=dogfood
-
-# issue width
-CONFIGS="-DISSUE_WIDTH=1" ./ci/blackbox.sh --driver=rtlsim --app=diverge
-CONFIGS="-DISSUE_WIDTH=2" ./ci/blackbox.sh --driver=rtlsim --app=diverge
-CONFIGS="-DISSUE_WIDTH=1" ./ci/blackbox.sh --driver=simx --app=diverge
-CONFIGS="-DISSUE_WIDTH=2" ./ci/blackbox.sh --driver=simx --app=diverge
-
-# dispatch size
-CONFIGS="-DNUM_ALU_BLOCK=1 -DNUM_ALU_LANES=1" ./ci/blackbox.sh --driver=rtlsim --app=diverge
-CONFIGS="-DNUM_ALU_BLOCK=2 -DNUM_ALU_LANES=2" ./ci/blackbox.sh --driver=rtlsim --app=diverge
-CONFIGS="-DNUM_ALU_BLOCK=1 -DNUM_ALU_LANES=1" ./ci/blackbox.sh --driver=simx --app=diverge
-CONFIGS="-DNUM_ALU_BLOCK=2 -DNUM_ALU_LANES=2" ./ci/blackbox.sh --driver=simx --app=diverge
-
-# FPU scaling
-CONFIGS="-DNUM_ALU_BLOCK=4 -DNUM_FPU_LANES=2" ./ci/blackbox.sh --driver=rtlsim --app=sgemm
-CONFIGS="-DNUM_ALU_BLOCK=2 -DNUM_FPU_LANES=4" ./ci/blackbox.sh --driver=rtlsim --app=sgemm
-CONFIGS="-DNUM_ALU_BLOCK=4 -DNUM_FPU_LANES=4" ./ci/blackbox.sh --driver=rtlsim --app=sgemm
-
-# custom program startup address
-make -C tests/regression/dogfood clean-all
-STARTUP_ADDR=0x40000000 make -C tests/regression/dogfood
-CONFIGS="-DSTARTUP_ADDR=0x40000000" ./ci/blackbox.sh --driver=simx --app=dogfood
-CONFIGS="-DSTARTUP_ADDR=0x40000000" ./ci/blackbox.sh --driver=rtlsim --app=dogfood
-make -C tests/regression/dogfood clean-all
-make -C tests/regression/dogfood
-
 # disabling M extension
-CONFIGS="-DEXT_M_DISABLE" ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=no_mf_ext
+CONFIGS=-DEXT_M_DISABLE ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=no_mf_ext
 
 # disabling F extension
-CONFIGS="-DEXT_F_DISABLE" ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=no_mf_ext
-CONFIGS="-DEXT_F_DISABLE" ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=no_mf_ext --perf=1
-CONFIGS="-DEXT_F_DISABLE" ./ci/blackbox.sh --driver=simx --cores=1 --app=no_mf_ext --perf=1
+CONFIGS=-DEXT_F_DISABLE ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=no_mf_ext
+CONFIGS=-DEXT_F_DISABLE ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=no_mf_ext --perf
+CONFIGS=-DEXT_F_DISABLE ./ci/blackbox.sh --driver=simx --cores=1 --app=no_mf_ext --perf
 
 # disable shared memory
-CONFIGS="-DSM_DISABLE" ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=no_smem
-CONFIGS="-DSM_DISABLE" ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=no_smem --perf=1
-CONFIGS="-DSM_DISABLE" ./ci/blackbox.sh --driver=simx --cores=1 --app=no_smem --perf=1
+CONFIGS=-DSM_ENABLE=0 ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=no_smem
+CONFIGS=-DSM_ENABLE=0 ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=no_smem --perf
+CONFIGS=-DSM_ENABLE=0 ./ci/blackbox.sh --driver=simx --cores=1 --app=no_smem --perf
 
-# disable L1 cache
-CONFIGS="-DL1_DISABLE -DSM_DISABLE" ./ci/blackbox.sh --driver=rtlsim --app=sgemm
-CONFIGS="-DDCACHE_DISABLE" ./ci/blackbox.sh --driver=rtlsim --app=sgemm
+# using Default FPU core
+FPU_CORE=FPU_DEFAULT ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=dogfood
 
-# multiple L1 caches per cluster
-CONFIGS="-DNUM_DCACHES=2 -DNUM_ICACHES=2" ./ci/blackbox.sh --driver=rtlsim --app=sgemm --cores=8 --warps=1 --threads=2
+# using FPNEW FPU core
+FPU_CORE=FPU_FPNEW ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=dogfood
 
-# test AXI bus
+# using AXI bus
 AXI_BUS=1 ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=demo
 
 # adjust l1 block size to match l2
-CONFIGS="-DL1_LINE_SIZE=64" ./ci/blackbox.sh --driver=rtlsim --cores=2 --l2cache --app=io_addr --args="-n1"
+CONFIGS="-DL1_BLOCK_SIZE=64" ./ci/blackbox.sh --driver=rtlsim --cores=2 --l2cache --app=io_addr --args="-n1"
 
 # test cache banking
-CONFIGS="-DSMEM_NUM_BANKS=4 -DDCACHE_NUM_BANKS=1" ./ci/blackbox.sh --driver=rtlsim --app=sgemm
-CONFIGS="-DSMEM_NUM_BANKS=2 -DDCACHE_NUM_BANKS=2" ./ci/blackbox.sh --driver=rtlsim --app=sgemm
-CONFIGS="-DSMEM_NUM_BANKS=2 -DDCACHE_NUM_BANKS=2" ./ci/blackbox.sh --driver=simx --app=sgemm
-CONFIGS="-DDCACHE_NUM_BANKS=1" ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=sgemm
-CONFIGS="-DDCACHE_NUM_BANKS=2" ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=sgemm
-CONFIGS="-DDCACHE_NUM_BANKS=2" ./ci/blackbox.sh --driver=simx --cores=1 --app=sgemm
+CONFIGS="-DDNUM_BANKS=1" ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=io_addr
+CONFIGS="-DDNUM_BANKS=2" ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=io_addr
+CONFIGS="-DDNUM_BANKS=2" ./ci/blackbox.sh --driver=simx --cores=1 --app=io_addr
+
+# test cache multi-porting
+CONFIGS="-DDNUM_PORTS=2" ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=io_addr
+CONFIGS="-DDNUM_PORTS=2" ./ci/blackbox.sh --driver=rtlsim --cores=1 --app=demo --debug --args="-n1"
+CONFIGS="-DL2_NUM_PORTS=2 -DDNUM_PORTS=2" ./ci/blackbox.sh --driver=rtlsim --cores=2 --l2cache --app=io_addr
+CONFIGS="-DL2_NUM_PORTS=4 -DDNUM_PORTS=4" ./ci/blackbox.sh --driver=rtlsim --cores=4 --l2cache --app=io_addr
+CONFIGS="-DL2_NUM_PORTS=4 -DDNUM_PORTS=4" ./ci/blackbox.sh --driver=simx --cores=4 --l2cache --app=io_addr
 
 # test 128-bit MEM block
-CONFIGS="-DMEM_BLOCK_SIZE=16" ./ci/blackbox.sh --driver=opae --cores=1 --app=demo
+CONFIGS=-DMEM_BLOCK_SIZE=16 ./ci/blackbox.sh --driver=vlsim --cores=1 --app=demo
 
 # test single-bank DRAM
-CONFIGS="-DPLATFORM_PARAM_LOCAL_MEMORY_BANKS=1" ./ci/blackbox.sh --driver=opae --cores=1 --app=demo
+CONFIGS="-DPLATFORM_PARAM_LOCAL_MEMORY_BANKS=1" ./ci/blackbox.sh --driver=vlsim --cores=1 --app=demo
 
 # test 27-bit DRAM address
-CONFIGS="-DPLATFORM_PARAM_LOCAL_MEMORY_ADDR_WIDTH=27" ./ci/blackbox.sh --driver=opae --cores=1 --app=demo
+CONFIGS="-DPLATFORM_PARAM_LOCAL_MEMORY_ADDR_WIDTH=27" ./ci/blackbox.sh --driver=vlsim --cores=1 --app=demo
 
 echo "configuration tests done!"
 }
@@ -241,9 +138,14 @@ stress0()
 echo "begin stress0 tests..."
 
 # test verilator reset values
-CONFIGS="-DVERILATOR_RESET_VALUE=1" ./ci/blackbox.sh --driver=opae --cores=2 --clusters=2 --l2cache --l3cache --app=dogfood
-CONFIGS="-DVERILATOR_RESET_VALUE=1" ./ci/blackbox.sh --driver=opae --cores=2 --clusters=2 --l2cache --l3cache --app=io_addr
-CONFIGS="-DVERILATOR_RESET_VALUE=1" ./ci/blackbox.sh --driver=opae --app=printf
+CONFIGS="-DVERILATOR_RESET_VALUE=0" ./ci/blackbox.sh --driver=vlsim --cores=2 --clusters=2 --l2cache --l3cache --app=sgemm
+CONFIGS="-DVERILATOR_RESET_VALUE=1" ./ci/blackbox.sh --driver=vlsim --cores=2 --clusters=2 --l2cache --l3cache --app=sgemm
+FPU_CORE=FPU_DEFAULT CONFIGS="-DVERILATOR_RESET_VALUE=0" ./ci/blackbox.sh --driver=vlsim --cores=2 --clusters=2 --l2cache --l3cache --app=dogfood
+FPU_CORE=FPU_DEFAULT CONFIGS="-DVERILATOR_RESET_VALUE=1" ./ci/blackbox.sh --driver=vlsim --cores=2 --clusters=2 --l2cache --l3cache --app=dogfood
+CONFIGS="-DVERILATOR_RESET_VALUE=0" ./ci/blackbox.sh --driver=vlsim --cores=2 --clusters=2 --l2cache --l3cache --app=io_addr
+CONFIGS="-DVERILATOR_RESET_VALUE=1" ./ci/blackbox.sh --driver=vlsim --cores=2 --clusters=2 --l2cache --l3cache --app=io_addr
+CONFIGS="-DVERILATOR_RESET_VALUE=0" ./ci/blackbox.sh --driver=vlsim --app=printf
+CONFIGS="-DVERILATOR_RESET_VALUE=1" ./ci/blackbox.sh --driver=vlsim --app=printf
 
 echo "stress0 tests done!"
 }
@@ -252,75 +154,51 @@ stress1()
 {
 echo "begin stress1 tests..."
 
-./ci/blackbox.sh --driver=rtlsim --app=sgemm --args="-n128" --l2cache
+./ci/blackbox.sh --driver=rtlsim --cores=2 --l2cache --clusters=2 --l3cache --app=sgemm --args="-n256"
 
 echo "stress1 tests done!"
 }
 
-synthesis()
+usage()
 {
-echo "begin synthesis tests..."
-
-PREFIX=build_base make -C hw/syn/yosys clean
-PREFIX=build_base CONFIGS="-DDPI_DISABLE -DEXT_F_DISABLE" make -C hw/syn/yosys elaborate
-
-echo "synthesis tests done!"
+    echo "usage: regression [-unittest] [-coverage] [-tex] [-cluster] [-debug] [-config] [-stress[#n]] [-all] [-h|--help]"
 }
 
-show_usage()
-{
-    echo "Vortex Regression Test" 
-    echo "Usage: $0 [--unittest] [--isa] [--regression] [--opencl] [--cluster] [--debug] [--config] [--stress[#n]] [--synthesis] [--all] [--h|--help]"
-}
-
-start=$SECONDS
-
 while [ "$1" != "" ]; do
     case $1 in
-        --unittest ) unittest
+        -unittest ) unittest
                 ;;
-        --isa ) isa
+        -coverage ) coverage
                 ;;
-        --regression ) regression
+        -tex ) tex
                 ;;
-        --opencl ) opencl
+        -cluster ) cluster
                 ;;
-        --cluster ) cluster
+        -debug ) debug
                 ;;
-        --debug ) debug
+        -config ) config
                 ;;
-        --config ) config
+        -stress0 ) stress0
                 ;;
-        --stress0 ) stress0
+        -stress1 ) stress1
                 ;;
-        --stress1 ) stress1
-                ;;
-        --stress ) stress0
+        -stress ) stress0
                   stress1
                 ;;
-        --synthesis ) synthesis
-                ;;
-        --all ) unittest               
-               isa
-               regression
-               opencl
+        -all ) unittest
+               coverage
+               tex
                cluster
                debug
                config
                stress0
                stress1
-               synthesis
                 ;;
-        -h | --help ) show_usage
+        -h | --help ) usage
                       exit
                 ;;
-        * )           show_usage
+        * )           usage
                       exit 1
     esac
     shift
-done
-
-echo "Regression completed!"
-
-duration=$(( SECONDS - start ))
-awk -v t=$duration 'BEGIN{t=int(t*1000); printf "Elapsed Time: %d:%02d:%02d\n", t/3600000, t/60000%60, t/1000%60}'
+done

ci/regression64.sh

@@ -0,0 +1,38 @@
+#!/bin/bash
+
+# exit when any command fails
+set -e
+
+# ensure build
+make -s
+
+coverage() 
+{
+echo "begin coverage tests..."
+
+make -C sim/simx clean
+XLEN=64 make -C sim/simx
+XLEN=64 make -C tests/riscv/isa run-simx
+
+echo "coverage tests done!"
+}
+
+usage()
+{
+    echo "usage: regression [-coverage] [-all] [-h|--help]"
+}
+
+while [ "$1" != "" ]; do
+    case $1 in
+        -coverage ) coverage
+                ;;
+        -all ) coverage
+                ;;
+        -h | --help ) usage
+                      exit
+                ;;
+        * )           usage
+                      exit 1
+    esac
+    shift
+done

ci/test_compiler.sh

@@ -0,0 +1,30 @@
+#!/bin/bash
+
+# exit when any command fails
+set -e
+
+# ensure build
+make -s
+
+# clear POCL cache
+rm -rf ~/.cache/pocl
+
+# rebuild runtime
+make -C runtime clean
+make -C runtime
+
+# rebuild drivers
+make -C driver clean
+make -C driver
+
+# rebuild runtime tests
+make -C tests/runtime clean
+make -C tests/runtime
+
+# rebuild regression tests
+make -C tests/regression clean-all
+make -C tests/regression
+
+# rebuild opencl tests
+make -C tests/opencl clean-all
+make -C tests/opencl

ci/toolchain_env.sh

@@ -1,30 +0,0 @@
-#!/bin/sh
-
-# Copyright 2023 blaise
-# 
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-# 
-#     http://www.apache.org/licenses/LICENSE-2.0
-# 
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-TOOLDIR=${TOOLDIR:=/opt}
-
-export VERILATOR_ROOT=$TOOLDIR/verilator
-export PATH=$VERILATOR_ROOT/bin:$PATH
-
-export SV2V_PATH=$TOOLDIR/sv2v
-export PATH=$SV2V_PATH/bin:$PATH
-
-export YOSYS_PATH=$TOOLDIR/yosys
-export PATH=$YOSYS_PATH/bin:$PATH
-
-export LLVM_VORTEX=$TOOLDIR/llvm-vortex
-export POCL_CC_PATH=$TOOLDIR/pocl/compiler
-export POCL_RT_PATH=$TOOLDIR/pocl/runtime

ci/toolchain_install.sh

@@ -1,184 +1,97 @@
 #!/bin/bash
 
-# Copyright © 2019-2023
-# 
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-# http://www.apache.org/licenses/LICENSE-2.0
-# 
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
 # exit when any command fails
 set -e
 
 REPOSITORY=https://github.com/vortexgpgpu/vortex-toolchain-prebuilt/raw/master
-TOOLDIR=${TOOLDIR:=/opt}
-OSDIR=${OSDIR:=ubuntu/bionic}
 
-OS="${OS:=ubuntu/bionic}"
+DESTDIR="${DESTDIR:=/opt}"
 
 riscv()
 {
-    case $OSDIR in
-    "centos/7") parts=$(eval echo {a..h}) ;;
-    *)          parts=$(eval echo {a..j}) ;;
-    esac
-    rm -f riscv-gnu-toolchain.tar.bz2.parta*
-    for x in $parts
+    for x in {a..j} 
     do
-        wget $REPOSITORY/riscv-gnu-toolchain/$OSDIR/riscv-gnu-toolchain.tar.bz2.parta$x
+        wget $REPOSITORY/riscv-gnu-toolchain/ubuntu/bionic/riscv-gnu-toolchain.tar.bz2.parta$x
     done
     cat riscv-gnu-toolchain.tar.bz2.parta* > riscv-gnu-toolchain.tar.bz2
     tar -xvf riscv-gnu-toolchain.tar.bz2
-    cp -r riscv-gnu-toolchain $TOOLDIR
-    rm -f riscv-gnu-toolchain.tar.bz2*    
+    rm -f riscv-gnu-toolchain.tar.bz2*
+    cp -r riscv-gnu-toolchain $DESTDIR
     rm -rf riscv-gnu-toolchain
 }
 
 riscv64()
 {
-    case $OSDIR in
-    "centos/7") parts=$(eval echo {a..h}) ;;
-    *)          parts=$(eval echo {a..j}) ;;
-    esac
-    rm -f riscv64-gnu-toolchain.tar.bz2.parta*
-    for x in $parts
+    for x in {a..j} 
     do
-        wget $REPOSITORY/riscv64-gnu-toolchain/$OSDIR/riscv64-gnu-toolchain.tar.bz2.parta$x
+        wget $REPOSITORY/riscv64-gnu-toolchain/ubuntu/bionic/riscv64-gnu-toolchain.tar.bz2.parta$x
     done
     cat riscv64-gnu-toolchain.tar.bz2.parta* > riscv64-gnu-toolchain.tar.bz2
     tar -xvf riscv64-gnu-toolchain.tar.bz2
-    cp -r riscv64-gnu-toolchain $TOOLDIR
-    rm -f riscv64-gnu-toolchain.tar.bz2*    
+    rm -f riscv64-gnu-toolchain.tar.bz2*
+    cp -r riscv64-gnu-toolchain $DESTDIR
     rm -rf riscv64-gnu-toolchain
 }
 
-llvm-vortex()
+llvm()
 {
-    case $OSDIR in
-    "centos/7") parts=$(eval echo {a..b}) ;;
-    *)          parts=$(eval echo {a..b}) ;;
-    esac
-    echo $parts
-    rm -f llvm-vortex.tar.bz2.parta*
-    for x in $parts
+    for x in {a..b} 
     do
-        wget $REPOSITORY/llvm-vortex/$OSDIR/llvm-vortex.tar.bz2.parta$x
+        wget $REPOSITORY/llvm-vortex/ubuntu/bionic/llvm-vortex1.tar.bz2.parta$x
     done
-    cat llvm-vortex.tar.bz2.parta* > llvm-vortex.tar.bz2
-    tar -xvf llvm-vortex.tar.bz2
-    cp -r llvm-vortex $TOOLDIR
-    rm -f llvm-vortex.tar.bz2*    
-    rm -rf llvm-vortex
-}
-
-llvm-pocl()
-{
-    case $OSDIR in
-    "centos/7") parts=$(eval echo {a..b}) ;;
-    *)          parts=$(eval echo {a..b}) ;;
-    esac
-    echo $parts
-    rm -f llvm-pocl.tar.bz2.parta*
-    for x in $parts
-    do
-        wget $REPOSITORY/llvm-pocl/$OSDIR/llvm-pocl.tar.bz2.parta$x
-    done
-    cat llvm-pocl.tar.bz2.parta* > llvm-pocl.tar.bz2
-    tar -xvf llvm-pocl.tar.bz2
-    cp -r llvm-pocl $TOOLDIR
-    rm -f llvm-pocl.tar.bz2*    
-    rm -rf llvm-pocl
+    cat llvm-vortex1.tar.bz2.parta* > llvm-vortex1.tar.bz2
+    tar -xvf llvm-vortex1.tar.bz2
+    rm -f llvm-vortex1.tar.bz2*
+    cp -r llvm-riscv $DESTDIR
+    rm -rf llvm-riscv
 }
 
 pocl()
 {
-    wget $REPOSITORY/pocl/$OSDIR/pocl.tar.bz2
-    tar -xvf pocl.tar.bz2
-    rm -f pocl.tar.bz2
-    cp -r pocl $TOOLDIR
+    wget $REPOSITORY/pocl/ubuntu/bionic/pocl1.tar.bz2
+    tar -xvf pocl1.tar.bz2
+    rm -f pocl1.tar.bz2
+    cp -r pocl $DESTDIR
     rm -rf pocl
 }
 
 verilator()
 {
-    wget $REPOSITORY/verilator/$OSDIR/verilator.tar.bz2
+    wget $REPOSITORY/verilator/ubuntu/bionic/verilator.tar.bz2
     tar -xvf verilator.tar.bz2
-    cp -r verilator $TOOLDIR
-    rm -f verilator.tar.bz2    
+    rm -f verilator.tar.bz2
+    cp -r verilator $DESTDIR
     rm -rf verilator
 }
 
-sv2v() 
+usage()
 {
-    wget $REPOSITORY/sv2v/$OSDIR/sv2v.tar.bz2
-    tar -xvf sv2v.tar.bz2
-    rm -f sv2v.tar.bz2
-    cp -r sv2v $TOOLDIR
-    rm -rf sv2v
-}
-
-yosys()
-{
-    case $OSDIR in
-    "centos/7") parts=$(eval echo {a..c}) ;;
-    *)          parts=$(eval echo {a..c}) ;;
-    esac
-    echo $parts
-    rm -f yosys.tar.bz2.parta*
-    for x in $parts
-    do
-        wget $REPOSITORY/yosys/$OSDIR/yosys.tar.bz2.parta$x
-    done
-    cat yosys.tar.bz2.parta* > yosys.tar.bz2
-    tar -xvf yosys.tar.bz2
-    cp -r yosys $TOOLDIR
-    rm -f yosys.tar.bz2*    
-    rm -rf yosys
-}
-
-show_usage()
-{
-    echo "Install Pre-built Vortex Toolchain"
-    echo "Usage: $0 [[--riscv] [--riscv64] [--llvm-vortex] [--llvm-pocl] [--pocl] [--verilator] [--sv2v] [--yosys] [--all] [-h|--help]]"
+    echo "usage: toolchain_install [[-riscv] [-riscv64] [-llvm] [-pocl] [-verilator] [-all] [-h|--help]]"
 }
 
 while [ "$1" != "" ]; do
     case $1 in
-        --pocl ) pocl
+        -pocl ) pocl
                 ;;
-        --verilator ) verilator
+        -verilator ) verilator
+                     ;;
+        -riscv ) riscv
+                 ;;
+        -riscv64 ) riscv64
+                 ;;
+        -llvm ) llvm
                 ;;
-        --riscv ) riscv
-                ;;
-        --riscv64 ) riscv64
-                ;;
-        --llvm-vortex ) llvm-vortex
-                ;;
-        --llvm-pocl ) llvm-pocl                
-                ;;
-        --sv2v ) sv2v
-                ;;
-        --yosys ) yosys
-                ;;
-        --all ) pocl
-                verilator
-                sv2v
-                yosys
-                llvm-vortex
-                riscv
-                riscv64                
-                ;;
-        -h | --help ) show_usage
-                exit
-                ;;
-        * ) show_usage
-                exit 1
+        -all ) riscv
+               riscv64
+               llvm
+               pocl
+               verilator
+               ;;
+        -h | --help ) usage
+                      exit
+                      ;;
+        * )           usage
+                      exit 1
     esac
     shift
 done

ci/toolchain_prebuilt.sh

@@ -1,128 +0,0 @@
-#!/bin/bash
-
-# Copyright © 2019-2023
-# 
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-# http://www.apache.org/licenses/LICENSE-2.0
-# 
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-# exit when any command fails
-set -e
-
-TOOLDIR=${TOOLDIR:=/opt}
-OSDIR=${OSDIR:=ubuntu/bionic}
-
-riscv() 
-{
-    echo "prebuilt riscv-gnu-toolchain..."
-    tar -C $TOOLDIR -cvjf riscv-gnu-toolchain.tar.bz2 riscv-gnu-toolchain
-    split -b 50M riscv-gnu-toolchain.tar.bz2 "riscv-gnu-toolchain.tar.bz2.part"    
-    mv riscv-gnu-toolchain.tar.bz2.part* ./riscv-gnu-toolchain/$OSDIR
-    rm riscv-gnu-toolchain.tar.bz2
-}
-
-riscv64() 
-{
-    echo "prebuilt riscv64-gnu-toolchain..."
-    tar -C $TOOLDIR -cvjf riscv64-gnu-toolchain.tar.bz2 riscv64-gnu-toolchain
-    split -b 50M riscv64-gnu-toolchain.tar.bz2 "riscv64-gnu-toolchain.tar.bz2.part"    
-    mv riscv64-gnu-toolchain.tar.bz2.part* ./riscv64-gnu-toolchain/$OSDIR
-    rm riscv64-gnu-toolchain.tar.bz2
-}
-
-llvm-vortex() 
-{
-    echo "prebuilt llvm-vortex..."
-    tar -C $TOOLDIR -cvjf llvm-vortex.tar.bz2 llvm-vortex
-    split -b 50M llvm-vortex.tar.bz2 "llvm-vortex.tar.bz2.part"    
-    mv llvm-vortex.tar.bz2.part* ./llvm-vortex/$OSDIR
-    rm llvm-vortex.tar.bz2
-}
-
-llvm-pocl() 
-{
-    echo "prebuilt llvm-pocl..."
-    tar -C $TOOLDIR -cvjf llvm-pocl.tar.bz2 llvm-pocl
-    split -b 50M llvm-pocl.tar.bz2 "llvm-pocl.tar.bz2.part"    
-    mv llvm-pocl.tar.bz2.part* ./llvm-pocl/$OSDIR
-    rm llvm-pocl.tar.bz2
-}
-
-pocl() 
-{
-    echo "prebuilt pocl..."
-    tar -C $TOOLDIR -cvjf pocl.tar.bz2 pocl
-    mv pocl.tar.bz2 ./pocl/$OSDIR
-}
-
-verilator() 
-{
-    echo "prebuilt verilator..."
-    tar -C $TOOLDIR -cvjf verilator.tar.bz2 verilator
-    mv verilator.tar.bz2 ./verilator/$OSDIR
-}
-
-sv2v() 
-{
-    echo "prebuilt sv2v..."
-    tar -C $TOOLDIR -cvjf sv2v.tar.bz2 sv2v
-    mv sv2v.tar.bz2 ./sv2v/$OSDIR
-}
-
-yosys()
-{
-    echo "prebuilt yosys..."
-    tar -C $TOOLDIR -cvjf yosys.tar.bz2 yosys
-    split -b 50M yosys.tar.bz2 "yosys.tar.bz2.part"    
-    mv yosys.tar.bz2.part* ./yosys/$OSDIR
-    rm yosys.tar.bz2
-}
-
-show_usage()
-{
-    echo "Setup Pre-built Vortex Toolchain"
-    echo "Usage: $0 [[--riscv] [--llvm-vortex] [--llvm-pocl] [--pocl] [--verilator] [--sv2v] [-yosys] [--all] [-h|--help]]"
-}
-
-while [ "$1" != "" ]; do
-    case $1 in
-        --pocl ) pocl
-                ;;
-        --verilator ) verilator
-                ;;
-        --riscv ) riscv
-                ;;
-        --riscv64 ) riscv64
-                ;;
-        --llvm-vortex ) llvm-vortex
-                ;;
-        --llvm-pocl ) llvm-pocl
-                ;;
-        --sv2v ) sv2v
-                ;;
-        --yosys ) yosys
-                ;;
-        --all ) riscv
-                riscv64
-                llvm-vortex
-                llvm-pocl
-                pocl
-                verilator
-                sv2v
-                yosys
-                ;;
-        -h | --help ) show_usage
-                exit
-                ;;
-        * ) show_usage
-                exit 1
-    esac
-    shift
-done

ci/trace_csv.py

@@ -1,245 +0,0 @@
-#!/usr/bin/env python3
-
-# Copyright © 2019-2023
-# 
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-# http://www.apache.org/licenses/LICENSE-2.0
-# 
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import sys
-import argparse
-import csv
-import re
-            
-def parse_args():
-    parser = argparse.ArgumentParser(description='CPU trace log to CSV format converter.')
-    parser.add_argument('-t', '--type', default='simx', help='log type (rtlsim or simx)')
-    parser.add_argument('-o', '--csv', default='trace.csv', help='Output CSV file')
-    parser.add_argument('log', help='Input log file')
-    return parser.parse_args()
-
-def parse_simx(log_filename):    
-    pc_pattern = r"PC=(0x[0-9a-fA-F]+)"
-    instr_pattern = r"Instr (0x[0-9a-fA-F]+):"
-    opcode_pattern = r"Instr 0x[0-9a-fA-F]+: ([0-9a-zA-Z_\.]+)"
-    core_id_pattern = r"cid=(\d+)"
-    warp_id_pattern = r"wid=(\d+)"
-    tmask_pattern = r"tmask=(\d+)"
-    operands_pattern = r"Src\d+ Reg: (.+)"
-    destination_pattern = r"Dest Reg: (.+)"
-    uuid_pattern = r"#(\d+)"
-    entries = []
-    with open(log_filename, 'r') as log_file:
-        instr_data = None
-        for lineno, line in enumerate(log_file, start=1):
-            if line.startswith("DEBUG Fetch:"):
-                if instr_data:
-                    entries.append(instr_data)
-                instr_data = {}                       
-                instr_data["lineno"] = lineno
-                instr_data["PC"] = re.search(pc_pattern, line).group(1)
-                instr_data["core_id"] = re.search(core_id_pattern, line).group(1)
-                instr_data["warp_id"] = re.search(warp_id_pattern, line).group(1)
-                instr_data["tmask"] = re.search(tmask_pattern, line).group(1)            
-                instr_data["uuid"] = re.search(uuid_pattern, line).group(1)
-            elif line.startswith("DEBUG Instr"):
-                instr_data["instr"] = re.search(instr_pattern, line).group(1)
-                instr_data["opcode"] = re.search(opcode_pattern, line).group(1)
-            elif line.startswith("DEBUG Src"):
-                src_reg = re.search(operands_pattern, line).group(1)
-                instr_data["operands"] = (instr_data["operands"] + ', ' + src_reg) if 'operands' in instr_data else src_reg
-            elif line.startswith("DEBUG Dest"):
-                instr_data["destination"] = re.search(destination_pattern, line).group(1)
-        if instr_data:
-            entries.append(instr_data)
-    return entries
-        
-def reverse_binary(bin_str):
-    return bin_str[::-1]
-
-def bin_to_array(bin_str):
-    return [int(bit) for bit in bin_str]
-        
-def append_reg(text, value, sep):
-    if sep:
-        text += ", "
-    ivalue = int(value)
-    if (ivalue >= 32):
-        text += "f" + str(ivalue % 32)
-    else:
-        text += "x" + value
-    sep = True
-    return text, sep
-        
-def append_imm(text, value, sep):    
-    if sep:
-        text += ", "
-    text += value
-    sep = True
-    return text, sep
-    
-def append_value(text, reg, value, tmask_arr, sep):
-    text, sep = append_reg(text, reg, sep)
-    text += "={"
-    for i in range(len(tmask_arr)):
-        if i != 0:
-            text += ", "
-        if tmask_arr[i]:
-            text += value[i]
-        else:
-            text +="-"
-    text += "}"
-    return text, sep
-        
-def parse_rtlsim(log_filename):    
-    line_pattern = r"\d+: core(\d+)-(decode|issue|commit)"
-    pc_pattern = r"PC=(0x[0-9a-fA-F]+)"
-    instr_pattern = r"instr=(0x[0-9a-fA-F]+)"
-    ex_pattern = r"ex=([a-zA-Z]+)"
-    op_pattern = r"op=([\?0-9a-zA-Z_\.]+)"
-    warp_id_pattern = r"wid=(\d+)"
-    tmask_pattern = r"tmask=(\d+)"
-    wb_pattern = r"wb=(\d)"
-    opds_pattern = r"opds=(\d+)"
-    use_imm_pattern = r"use_imm=(\d)"
-    imm_pattern = r"imm=(0x[0-9a-fA-F]+)"
-    rd_pattern = r"rd=(\d+)"
-    rs1_pattern = r"rs1=(\d+)"
-    rs2_pattern = r"rs2=(\d+)"
-    rs3_pattern = r"rs3=(\d+)"
-    rs1_data_pattern = r"rs1_data=\{(.+?)\}"
-    rs2_data_pattern = r"rs2_data=\{(.+?)\}"
-    rs3_data_pattern = r"rs3_data=\{(.+?)\}"
-    rd_data_pattern = r"data=\{(.+?)\}"
-    eop_pattern = r"eop=(\d)"
-    uuid_pattern = r"#(\d+)"
-    entries = []    
-    with open(log_filename, 'r') as log_file:
-        instr_data = {}
-        for lineno, line in enumerate(log_file, start=1):
-            line_match = re.search(line_pattern, line)
-            if line_match:
-                PC = re.search(pc_pattern, line).group(1)
-                warp_id = re.search(warp_id_pattern, line).group(1)    
-                tmask = re.search(tmask_pattern, line).group(1)
-                uuid = re.search(uuid_pattern, line).group(1)            
-                core_id = line_match.group(1)
-                stage = line_match.group(2)
-                if stage == "decode":                
-                    trace = {}
-                    trace["uuid"] = uuid
-                    trace["PC"] = PC                
-                    trace["core_id"] = core_id
-                    trace["warp_id"] = warp_id    
-                    trace["tmask"] = reverse_binary(tmask)
-                    trace["instr"] = re.search(instr_pattern, line).group(1)
-                    trace["opcode"] = re.search(op_pattern, line).group(1)
-                    trace["opds"] = bin_to_array(re.search(opds_pattern, line).group(1))
-                    trace["rd"] = re.search(rd_pattern, line).group(1)
-                    trace["rs1"] = re.search(rs1_pattern, line).group(1)
-                    trace["rs2"] = re.search(rs2_pattern, line).group(1)
-                    trace["rs3"] = re.search(rs3_pattern, line).group(1)
-                    trace["use_imm"] = re.search(use_imm_pattern, line).group(1) == "1"
-                    trace["imm"] = re.search(imm_pattern, line).group(1)
-                    instr_data[uuid] = trace
-                elif stage == "issue":
-                    if uuid in instr_data:
-                        trace = instr_data[uuid]
-                        trace["lineno"] = lineno
-                        opds = trace["opds"]
-                        if opds[1]:
-                            trace["rs1_data"] = re.search(rs1_data_pattern, line).group(1).split(', ')[::-1]
-                        if opds[2]:
-                            trace["rs2_data"] = re.search(rs2_data_pattern, line).group(1).split(', ')[::-1]
-                        if opds[3]:
-                            trace["rs3_data"] = re.search(rs3_data_pattern, line).group(1).split(', ')[::-1]
-                        trace["issued"] = True
-                        instr_data[uuid] = trace
-                elif stage == "commit":     
-                    if uuid in instr_data:
-                        trace = instr_data[uuid]
-                        if "issued" in trace:
-                            opds = trace["opds"]
-                            dst_tmask_arr = bin_to_array(tmask)[::-1]
-                            wb = re.search(wb_pattern, line).group(1) == "1"
-                            if wb:
-                                rd_data = re.search(rd_data_pattern, line).group(1).split(', ')[::-1]
-                                if 'rd_data' in trace:
-                                    merged_rd_data = trace['rd_data']
-                                    for i in range(len(dst_tmask_arr)):
-                                        if dst_tmask_arr[i] == 1:
-                                            merged_rd_data[i] = rd_data[i]
-                                    trace['rd_data'] = merged_rd_data
-                                else:
-                                    trace['rd_data'] = rd_data
-                            instr_data[uuid] = trace
-                            eop = re.search(eop_pattern, line).group(1) == "1"
-                            if eop:
-                                tmask_arr = bin_to_array(trace["tmask"])
-                                destination = ''
-                                if wb:
-                                    destination, sep = append_value(destination, trace["rd"], trace['rd_data'], tmask_arr, False)
-                                    del trace['rd_data']
-                                trace["destination"] = destination
-                                operands = ''
-                                sep = False
-                                if opds[1]:
-                                    operands, sep = append_value(operands, trace["rs1"], trace["rs1_data"], tmask_arr, sep)
-                                    del trace["rs1_data"]
-                                if opds[2]:
-                                    operands, sep = append_value(operands, trace["rs2"], trace["rs2_data"], tmask_arr, sep)
-                                    del trace["rs2_data"]
-                                if opds[3]:
-                                    operands, sep = append_value(operands, trace["rs3"], trace["rs3_data"], tmask_arr, sep)
-                                    del trace["rs3_data"]
-                                trace["operands"] = operands
-                                del trace["opds"]
-                                del trace["rd"]
-                                del trace["rs1"]
-                                del trace["rs2"]
-                                del trace["rs3"]
-                                del trace["use_imm"]
-                                del trace["imm"]   
-                                del trace["issued"]                     
-                                del instr_data[uuid]
-                                entries.append(trace)
-    return entries                            
-
-def write_csv(log_filename, csv_filename, log_type):
-    entries = None
-    
-    # parse log file
-    if log_type == "rtlsim":
-        entries = parse_rtlsim(log_filename)
-    elif log_type == "simx":
-        entries = parse_simx(log_filename)
-    else:
-        print('Error: invalid log type')
-        sys.exit()
-        
-    # sort entries by uuid
-    entries.sort(key=lambda x: (int(x['core_id']), int(x['warp_id']), int(x['lineno'])))
-    for entry in entries:
-        del entry['lineno']
-    
-    # write to CSV
-    with open(csv_filename, 'w', newline='') as csv_file:
-        fieldnames = ["uuid", "PC", "opcode", "instr", "core_id", "warp_id", "tmask", "operands", "destination"]
-        writer = csv.DictWriter(csv_file, fieldnames=fieldnames)
-        writer.writeheader()
-        for entry in entries:
-            writer.writerow(entry)    
-
-def main():
-    args = parse_args()
-    write_csv(args.log, args.csv, args.type)
-
-if __name__ == "__main__":
-    main()

ci/travis_run.py

@@ -1,18 +1,4 @@
 #!/usr/bin/env python
-
-# Copyright 2019-2023
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-# http://www.apache.org/licenses/LICENSE-2.0
-# 
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
 import sys
 import time
 import threading
@@ -21,7 +7,7 @@ import subprocess
 # This script executes a long-running command while outputing "still running ..." periodically
 # to notify Travis build system that the program has not hanged
 
-PING_INTERVAL=300 # 5 minutes
+PING_INTERVAL=15
 
 def monitor(stop):
     wait_time = 0    
@@ -34,11 +20,11 @@ def monitor(stop):
             break
 
 def execute(command):
-    process = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
+    process = subprocess.Popen(command, stdout=subprocess.PIPE)
     while True:
         output = process.stdout.readline()
         if output:
-            line = output.decode('utf-8').rstrip()
+            line = output.decode('ascii').rstrip()
             print(">>> " + line)
             process.stdout.flush()
         ret = process.poll()

docs/cache_subsystem.md

@@ -2,26 +2,69 @@
 
 The Vortex Cache Sub-system has the following main properties:
 
-- High-bandwidth transfer with Multi-bank parallelism
-- Non-blocking pipelined architecture with local MSHR
-- Configurable design: Dcache, Icache, L2 cache, L3 cache
+- High-bandwidth with bank parallelism
+- Snoop protocol to flush data for CPU access
+- Generic design: Dcache, Icache, Shared Memory, L2 cache, L3 cache
 
-### Cache Microarchitecture 
+### Cache Hierarchy 
 
-![Image of Cache Hierarchy](./assets/img/cache_microarchitecture.png)
+![Image of Cache Hierarchy](./assets/img/cache_hierarchy.png)
 
-The Vortex cache is comprised of multiple parallel banks. It is comprised of the following modules:
-- **Bank request dispatch crossbar**: assign a bank to incoming requests and resolve collision using stalls.
-- **Bank response merge crossbar**: merge result from banks and forward to the core response.
-- **Memory request multiplexer**: arbitrate bank memory requests
-- **Memory response demultiplexer**: forward memory response to the corresponding bank.
-- **Flush Unit**: perform tag memory initialization.
+- Cache can be configured to be any level in the hierarchy
+- Caches communicate via snooping
+- Cache flush from AFU is passed down the hierarchy​
 
-Incoming requests entering the cache are sent to a dispatch crossbar that select the corresponding bank for each request, resolving bank collisions with stalls. The result output of each bank is merge back into outgoing response port via merger crossbar. Each bank intergates a non-blocking pipeline with a local Miss Status Holding Register (MSHR) to reduce the miss rate. The bank pipeline consists of the following stages:
+### VX_cache.v (Top Module)
 
-- **Schedule**: Selects the next request into the pipeline from the incoming core request, memory fill, or the MSHR entry, with priority given to the latter.
-- **Tag Access**: A single-port read/write access to the tag store.
-- **Data Access**: Single-port read/write access to the data store.
-- **Response Handling**: Core response back to the core.
+VX.cache.v is the top module of the cache verilog code located in the `/hw/rtl/cache` directory.
 
-Deadlocks inside the cache can occur when the MSHR is full and a new request is already in the pipeline. It can also occur when the memory request queue is full, and there is an incoming memory response. The cache mitigates MSHR deadlocks by using an early full signal before a new request is issued and similarly mitigates memory deadlocks by ensuring that its request queue never fills up.
+![Image of Vortex Cache](./assets/img/vortex_cache_top_module.png)
+
+- Configurable (Cache size, number of banks, bank line size, etc.)
+- I/O signals
+  - Core Request
+  - Core Rsp
+  - DRAM Req
+  - DRAM Rsp
+  - Snoop Rsp
+  - Snoop Rsp
+  - Snoop Forwarding Out
+  - Snoop Forwarding In
+- Bank Select
+  - Assigns valid and ready signals for each bank
+- Snoop Forwarder
+- DRAM Request Arbiter
+  - Prepares cache response for communication with DRAM
+- Snoop Response Arbiter
+  - Sends snoop response
+- Core Response Merge
+  - Cache accesses one line at a time. As a result, each request may not come back in the same response. This module tries to recombine the responses by thread ID. 
+
+### VX_bank.v
+
+VX_bank.v is the verilog code that handles cache bank functionality and is located in the `/hw/rtl/cache` directory.
+
+![Image of Vortex Cache Bank](./assets/img/vortex_bank.png)
+
+- Allows for high throughput​
+- Each bank contains queues to hold requests to the cache​
+- I/O signals
+  - Core request​
+  - Core Response​
+  - DRAM Fill Requests​
+  - DRAM Fill Response​
+  - DRAM WB Requests​
+  - Snp Request​
+  - Snp Response
+- Request Priority: DRAM fill, miss reserve, core request, snoop request​
+- Snoop Request Queue​
+- DRAM Fill Queue​
+- Core Req Arbiter​
+  - Requests to be processed by the bank
+- Tag Data Store​
+  - Registers for valid, dirty, dirtyb, tag, and data​
+  - Length of registers determined by lines in the bank​
+- Tag Data Access:​
+  - I/O: stall, snoop info, force request miss
+  - Writes to cache or sends read response; hit or miss determined here
+  - A missed request goes to the miss reserve if it is not a snoop request or DRAM fill 

docs/codebase.md

@@ -3,39 +3,38 @@
 The directory/file layout of the Vortex codebase is as followed:
 
 - `hw`:     
-  - `rtl`: hardware rtl sources    
-    - `core`: core pipeline
-    - `cache`: cache subsystem
-    - `mem`: memory subsystem
-    - `fpu`: floating point unit
+  - `rtl`: hardware rtl sources
+    - `cache`: cache subsystem code
+    - `fp_cores`: floating point unit code
     - `interfaces`: interfaces for inter-module communication
-    - `libs`: general-purpose RTL modules    
+    - `libs`: general-purpose RTL modules
   - `syn`: synthesis directory
-    - `altera`: Altera synthesis scripts
-    - `xilinx`: Xilinx synthesis scripts    
+    - `opae`: OPAE synthesis scripts
+    - `quartus`: Quartus synthesis scripts    
     - `synopsys`: Synopsys synthesis scripts
     - `modelsim`: Modelsim synthesis scripts
     - `yosys`: Yosys synthesis scripts
   - `unit_tests`: unit tests for some hardware components
-- `runtime`: host runtime software APIs
+- `driver`: host drivers repository
   - `include`: Vortex driver public headers
   - `stub`: Vortex stub driver library
-  - `opae`: software driver that uses Intel OPAE API with device targets=fpga|asesim|opaesim
-  - `xrt`: software driver that uses Xilinx XRT API with device targets=hw|hw_emu|sw_emu
+  - `fpga`: software driver that uses Intel OPAE FPGA
+  - `asesim`: software driver that uses Intel ASE simulator
+  - `vlsim`: software driver that uses vlsim simulator
   - `rtlsim`: software driver that uses rtlsim simulator
   - `simx`: software driver that uses simX simulator
-- `kernel`: GPU kernel software APIs
+- `runtime`: kernel runtime software
   - `include`: Vortex runtime public headers
   - `linker`: linker file for compiling kernels
   - `src`: runtime implementation
 - `sim`: 
-  - `opaesim`: Intel OPAE AFU RTL simulator
+  - `vlsim`: AFU RTL simulator
   - `rtlsim`: processor RTL simulator
   - `simX`: cycle approximate simulator for vortex
 - `tests`: tests repository.
-  - `riscv`: RISC-V conformance tests
-  - `kernel`: kernel tests
-  - `regression`: regression tests  
+  - `runtime`: runtime tests
+  - `regression`: regression tests
+  - `riscv`: RISC-V standard tests
   - `opencl`: opencl benchmarks and tests
 - `ci`: continuous integration scripts
 - `miscs`: miscellaneous resources.

docs/continuous_integration.md

@@ -1,36 +0,0 @@
-# Continuous Integration
-- Each time you push to the repo, the Continuous Integration pipeline will run
-- This pipeline consists of creating the correct development environment, building your code, and running all tests
-- This is an extensive pipeline so it might take some time to complete
-
-
-## Protecting Master Branch
-Navigate to your Repository:
-Open your repository on GitHub.
-
-Click on "Settings":
-In the upper-right corner of your repository page, click on the "Settings" tab.
-
-Select "Branches" in the left sidebar:
-On the left sidebar, look for the "Branches" option and click on it.
-
-Choose the Branch:
-Under "Branch protection rules," select the branch you want to protect. In this case, choose the main branch.
-
-Enable Branch Protection:``
-Check the box that says "Protect this branch."
-
-Configure Protection Settings:
-You can configure various protection settings. Some common settings include:
-
-Require pull request reviews before merging: This ensures that changes are reviewed before being merged.
-Require status checks to pass before merging: This ensures that automated tests and checks are passing.
-Require signed commits: This enforces that commits are signed with a verified signature.
-Restrict Who Can Push:
-You can further restrict who can push directly to the branch. You might want to limit this privilege to specific people or teams.
-
-Save Changes:
-Once you've configured the protection settings, scroll down and click on the "Save changes" button.
-
-Now, your main branch is protected, and certain criteria must be met before changes can be pushed directly to it. Contributors will need to create pull requests, have their changes reviewed, and meet other specified criteria before the changes can be merged into the main branch.
-

docs/contributing.md

@@ -1,18 +0,0 @@
-# Contributing to Vortex on Github
-
-## Github Details
-- There are two main repos, `vortex` (public, this one) and `vortex-dev` (private)
-- todo: Most current development is on `vortex`
-- If you have a legacy version of `vortex`, you can use the releases branch or tags to access the repo at that point in time
-
-## Contribution Process
-- You should create a new branch from develop that is clearly named with the feature that you want to add
-- Avoid pushing directly to the `master` branch instead you will need to make a Pull Request (PR)
-- There should be protections in place that prevent pushing directly to the main branch, but don't rely on it
-- When you make a PR it will be tested against the continuous integration (ci) pipeline (see `continuous_integration.md`)
-- It is not sufficient to just write some tests, they need to be incorporated into the ci pipeline to make sure they are run
-- During a PR, you might receive feedback regarding your changes and you might need to make further commits to your branch
-
-
-## Creating  and Adding Tests
-see `testing.md`

docs/debugging.md

@@ -1,37 +1,29 @@
-# Debugging Vortex GPU
-
-## Testing changes to the RTL or simulator GPU driver.
-
-The Blackbox utility script will not pick up your changes if the h/w configuration is the same as during teh last run.
-To force the utility to build the driver, you need pass the --rebuild=1 option when running tests. 
-Using --rebuild=0 will prevent the rebuild even if the h/w configuration is different from last run.
-
-    $ ./ci/blackbox.sh --driver=simx --app=demo --rebuild=1
+# Debugging Vortex Hardware
 
 ## SimX Debugging
 
 SimX cycle-approximate simulator allows faster debugging of Vortex kernels' execution. 
-The recommended method to enable debugging is to pass the `--debug=<level>` flag to `blackbox` tool when running a program.
+The recommended method to enable debugging is to pass the `--debug` flag to `blackbox` tool when running a program.
 
     // Running demo program on SimX in debug mode
-    $ ./ci/blackbox.sh --driver=simx --app=demo --debug=1
+    $ ./ci/blackbox.sh --driver=simx --app=demo --debug 
 
-A debug trace `run.log` is generated in the current directory during the program execution. The trace includes important states of the simulated processor (decoded instruction, register states, pipeline states, etc..). You can increase the verbosity of the trace by changing the debug level.
+A debug trace `run.log` is generated in the current directory during the program execution. The trace includes important states of the simulated processor (decoded instruction, register states, pipeline states, etc..). You can increase the verbosity level of the trace by changing the `DEBUG_LEVEL` variable to a value [1-5] (default is 3).
 
-    // Using SimX in debug mode with verbose level 3
-    $ ./ci/blackbox.sh --driver=simx --app=demo --debug=3
+    // Using SimX in debug mode with verbose level 4
+    $ CONFIGS=-DDEBUG_LEVEL=4 ./ci/blackbox.sh --driver=simx --app=demo --debug
 
 ## RTL Debugging
 
-To debug the processor RTL, you need to use VLSIM or RTLSIM driver. VLSIM simulates the full processor including the AFU command processor (using `/rtl/afu/opae/vortex_afu.sv` as top module). RTLSIM simulates the Vortex processor only (using `/rtl/Vortex.v` as top module).
+To debug the processor RTL, you need to use VLSIM or RTLSIM driver. VLSIM simulates the full processor including the AFU command processor (using `/rtl/afu/vortex_afu.sv` as top module). RTLSIM simulates the Vortex processor only (using `/rtl/Vortex.v` as top module).
 
 The recommended method to enable debugging is to pass the `--debug` flag to `blackbox` tool when running a program.
 
-    // Running demo program on the opae simulator in debug mode
-    $ TARGET=opaesim ./ci/blackbox.sh --driver=opae --app=demo --debug=1
+    // Running demo program on vlsim in debug mode
+    $ ./ci/blackbox.sh --driver=vlsim --app=demo --debug
 
     // Running demo program on rtlsim in debug mode
-    $ ./ci/blackbox.sh --driver=rtlsim --app=demo --debug=1
+    $ ./ci/blackbox.sh --driver=rtlsim --app=demo --debug
 
 A debug trace `run.log` is generated in the current directory during the program execution. The trace includes important states of the simulated processor (memory, caches, pipeline, stalls, etc..). A waveform trace `trace.vcd` is also generated in the current directory during the program execution. You can visualize the waveform trace using any tool that can open VCD files (Modelsim, Quartus, Vivado, etc..). [GTKwave] (http://gtkwave.sourceforge.net) is a great open-source scope analyzer that also works with VCD files.
 
@@ -40,7 +32,7 @@ A debug trace `run.log` is generated in the current directory during the program
 Debugging the FPGA directly may be necessary to investigate runtime bugs that the RTL simulation cannot catch. We have implemented an in-house scope analyzer for Vortex that works when the FPGA is running. To enable the FPGA scope analyzer, the FPGA bitstream should be built using `SCOPE=1` flag
 
     & cd /hw/syn/opae
-    $ CONFIGS="-DSCOPE=1" TARGET=fpga make
+    $ CONFIGS=-DSCOPE=1 make fpga-4c
 
 When running the program on the FPGA, you need to pass the `--scope` flag to the `blackbox` tool.
 
@@ -48,18 +40,4 @@ When running the program on the FPGA, you need to pass the `--scope` flag to the
     $ ./ci/blackbox.sh --driver=fpga --app=demo --scope
 
 
-A waveform trace `trace.vcd` will be generated in the current directory during the program execution. This trace includes a limited set of signals that are defined in `/hw/scripts/scope.json`. You can expand your signals' selection by updating the json file.
-
-## Analyzing Vortex trace log
-
-When debugging Vortex RTL or SimX Simulator, reading the trace run.log file can be overwhelming when the trace gets really large.
-We provide a trace sanitizer tool under ./hw/scripts/trace_csv.py that you can use to convert the large trace into a CSV file containing all the instructions that executed with their source and destination operands.
-
-    $ ./ci/blackbox.sh --driver=rtlsim --app=demo --debug=3 --log=run_rtlsim.log
-    $ ./ci/trace_csv.py -trtlsim run_rtlsim.log -otrace_rtlsim.csv
-
-    $ ./ci/blackbox.sh --driver=simx --app=demo --debug=3 --log=run_simx.log
-    $ ./ci/trace_csv.py -tsimx run_simx.log -otrace_simx.csv
-
-The first column in the CSV trace is UUID (universal unique identifier) of the instruction and the content is sorted by the UUID. You can use the UUID to trace the same instruction running on either the RTL hw or SimX simulator. 
-This can be very effective if you want to use SimX to debugging your RTL hardware by comparing CSV traces.
+A waveform trace `trace.vcd` will be generated in the current directory during the program execution. This trace includes a limited set of signals that are defined in `/hw/scripts/scope.json`. You can expand your signals' selection by updating the json file.

docs/environment_setup.md

@@ -1,45 +0,0 @@
-# Environment Setup
-These instructions apply to the development vortex repo using the updated toolchain. The updated toolchain is considered to be any commit of `master` pulled from July 2, 2023 onwards. The toolchain update in question can be viewed in this [commit](https://github.com/vortexgpgpu/vortex-dev/commit/0048496ba28d7b9a209a0e569d52d60f2b68fc04). Therefore, if you are unsure whether you are using the new toolchain or not, then you should check the `ci` folder for the existence of the `toolchain_prebuilt.sh` script. Furthermore, you should notice that the `toolchain_install.sh` script has the legacy `llvm()` split into `llvm-vortex()` and `llvm-pocl()`.
-
-
-## Set Up on Your Own System
-The toolchain binaries provided with Vortex are built on Ubuntu-based systems. To install Vortex on your own system, [follow these instructions](install_vortex.md).
-
-
-## Servers for Georgia Tech Students and Collaborators
-### Volvo
-Volvo is a 64-core server provided by HPArch. You need valid credentials to access it. If you don't already have access, you can get in contact with your mentor to ask about setting your account up.
-
-Setup on Volvo:
-1. Connect to Georgia Tech's VPN or ssh into another machine on campus
-2. `ssh volvo.cc.gatech.edu`
-3. Clone Vortex to your home directory: `git clone --recursive https://github.com/vortexgpgpu/vortex.git`
-4. `source /nethome/software/set_vortex_env.sh` to set up the necessary environment variables.
-5. `make -s` in the `vortex` root directory
-6. Run a test program: `./ci/blackbox.sh --cores=2 --app=dogfood`
-
-### Nio
-Nio is a 20-core desktop server provided by HPArch. If you have access to Volvo, you also have access to Nio.
-
-Setup on Nio:
-1. Connect to Georgia Tech's VPN or ssh into another machine on campus
-2. `ssh nio.cc.gatech.edu`
-3. Clone Vortex to your home directory: `git clone --recursive https://github.com/vortexgpgpu/vortex.git`
-4. `source /opt/set_vortex_env_dev.sh` to set up the necessary environment variables.
-5. `make -s` in the `vortex` root directory
-6. Run a test program: `./ci/blackbox.sh --cores=2 --app=dogfood`
-
-
-## Docker (Experimental)
-Docker allows for isolated pre-built environments to be created, shared and used. The emulation mode required for ARM-based processors will incur a decrease in performance. Currently, the dockerfile is not included with the official vortex repository and is not actively maintained or supported.
-
-### Setup with Docker
-1. Clone repo recursively onto your local machine: `git clone --recursive https://github.com/vortexgpgpu/vortex.git`
-2. Download the dockerfile from [here](https://github.gatech.edu/gist/usubramanya3/f1bf3e953faa38a6372e1292ffd0b65c) and place it in the root of the repo.
-3. Build the Dockerfile into an image: `docker build --platform=linux/amd64 -t vortex -f dockerfile .`
-4. Run a container based on the image: `docker run --rm -v ./:/root/vortex/ -it --name vtx-dev --privileged=true --platform=linux/amd64 vortex`
-5. Install the toolchain `./ci/toolchain_install.sh --all` (once per container)
-6. `make -s` in `vortex` root directory
-7. Run a test program: `./ci/blackbox.sh --cores=2 --app=dogfood`
-
-You may exit from a container and resume a container you have exited or start a second terminal session `docker exec -it <container-name> bash`

docs/execute_opencl_on_vortex.md

@@ -0,0 +1,128 @@
+# Execute OpenCL on Vortex backend
+
+## Requirements
+- [Vortex](https://github.com/vortexgpgpu/vortex)
+- [POCL for Vortex](https://github.com/vortexgpgpu/pocl)
+- [riscv-toolchain](https://github.com/riscv-collab/riscv-gnu-toolchain)
+- [llvm-riscv](https://github.com/llvm-mirror/llvm)
+
+For installation, please see [Build Instructions](../README.md) for more details.
+
+**For Ubuntu18.04 users, you can directly download pre-build toolchains with [toolchain_install.sh](https://github.com/vortexgpgpu/vortex/blob/master/ci/toolchain_install.sh) script.**
+```bash
+# please modify the DESTDIR variable in the script before execution
+bash toolchain_install.sh -all
+```
+Assuming we have installed all dependencies in `/opt` path, we can get the following environment:
+```bash
+tree -L 2 /opt
+'''
+/opt/
+├── llvm-riscv
+│   ├── bin
+│   ├── include
+│   ├── lib
+│   ├── libexec
+│   └── share
+├── pocl
+│   ├── compiler
+│   └── runtime
+├── riscv-gnu-toolchain
+│   ├── bin
+│   ├── drops
+│   ├── include
+│   ├── lib
+│   ├── libexec
+│   ├── riscv32-unknown-elf
+│   ├── share
+│   └── var
+└── verilator
+    ├── bin
+    ├── examples
+    ├── include
+    ├── verilator-config.cmake
+    └── verilator-config-version.cmake
+'''
+```
+## Execute OpenCL on Vortex
+In this tutorial, we show the example of executing a vecadd programs on SIMX backend. 
+To execute a OpenCL program on Vortex, we have the following steps:
+- Compile the [OpenCL kernels](https://github.com/vortexgpgpu/vortex/blob/master/tests/opencl/vecadd/kernel.cl) into risc-v binary by POCL compiler.
+- Compile the [OpenCL host](https://github.com/vortexgpgpu/vortex/blob/master/tests/opencl/vecadd/main.cc) and link with Vortex driver(```-lvortex```).
+- Execute the compiled host programs on a backend.
+
+Thus, we can write a Makefile as following:
+```Makefile
+LLVM_PREFIX ?= /opt/llvm-riscv
+RISCV_TOOLCHAIN_PATH ?= /opt/riscv-gnu-toolchain
+SYSROOT ?= $(RISCV_TOOLCHAIN_PATH)/riscv32-unknown-elf
+POCL_CC_PATH ?= /opt/pocl/compiler
+POCL_RT_PATH ?= /opt/pocl/runtime
+
+OPTS ?= -n64
+
+# please edit these two variable to your environment
+VORTEX_DRV_PATH ?= $(realpath ../../../driver)
+VORTEX_RT_PATH ?= $(realpath ../../../runtime)
+
+K_LLCFLAGS += "-O3 -march=riscv32 -target-abi=ilp32f -mcpu=generic-rv32 -mattr=+m,+f -mattr=+vortex -float-abi=hard -code-model=small"
+K_CFLAGS   += "-v -O3 --sysroot=$(SYSROOT) --gcc-toolchain=$(RISCV_TOOLCHAIN_PATH) -march=rv32imf -mabi=ilp32f -Xclang -target-feature -Xclang +vortex -I$(VORTEX_RT_PATH)/include -fno-rtti -fno-exceptions -ffreestanding -nostartfiles -fdata-sections -ffunction-sections"
+K_LDFLAGS  += "-Wl,-Bstatic,-T$(VORTEX_RT_PATH)/linker/vx_link.ld -Wl,--gc-sections $(VORTEX_RT_PATH)/libvortexrt.a -lm"
+
+CXXFLAGS += -std=c++11 -O2 -Wall -Wextra -Wfatal-errors
+
+CXXFLAGS += -Wno-deprecated-declarations -Wno-unused-parameter
+
+CXXFLAGS += -I$(POCL_RT_PATH)/include
+
+LDFLAGS += -L$(POCL_RT_PATH)/lib -L$(VORTEX_DRV_PATH)/stub -lOpenCL -lvortex
+
+PROJECT = vecadd
+
+SRCS = main.cc
+
+all: $(PROJECT) kernel.pocl
+
+kernel.pocl: kernel.cl
+	LLVM_PREFIX=$(LLVM_PREFIX) POCL_DEBUG=all LD_LIBRARY_PATH=$(LLVM_PREFIX)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -LLCFLAGS $(K_LLCFLAGS) -CFLAGS $(K_CFLAGS) -LDFLAGS $(K_LDFLAGS) -o kernel.pocl kernel.cl
+ 
+$(PROJECT): $(SRCS)
+	$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -o $@
+
+run-fpga: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/fpga:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
+
+run-asesim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/asesim:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
+	
+run-vlsim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/vlsim:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
+
+run-simx: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/simx:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
+
+run-rtlsim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
+
+.depend: $(SRCS)
+	$(CXX) $(CXXFLAGS) -MM $^ > .depend;
+
+clean:
+	rm -rf $(PROJECT) *.o .depend 
+
+clean-all: clean
+	rm -rf *.pocl *.dump
+
+ifneq ($(MAKECMDGOALS),clean)
+    -include .depend
+endif
+```
+
+First, build the host program.
+```bash
+make all
+```
+If we want to execute on SIMX, we can execute the command below.
+```bash
+make run-simx
+```

docs/fpga_setup.md

@@ -9,21 +9,30 @@ OPAE Environment Setup
     $ export C_INCLUDE_PATH=$OPAE_HOME/include:$C_INCLUDE_PATH
     $ export LIBRARY_PATH=$OPAE_HOME/lib:$LIBRARY_PATH
     $ export LD_LIBRARY_PATH=$OPAE_HOME/lib:$LD_LIBRARY_PATH
+    $ export RISCV_TOOLCHAIN_PATH=/opt/riscv-gnu-toolchain
+    $ export PATH=:/opt/verilator/bin:$PATH
+    $ export VERILATOR_ROOT=/opt/verilator
 
 OPAE Build
 ------------------
 
 The FPGA has to following configuration options:
-- DEVICE_FAMILY=arria10 | stratix10
-- NUM_CORES=#n
+- 1 core fpga (fpga-1c)
+- 2 cores fpga (fpga-2c)
+- 4 cores fpga (fpga-4c)
+- 8 cores fpga (fpga-8c)
+- 16 cores fpga (fpga-16c)
+- 32 cores fpga (fpga-32c)
+- 64 cores fpga (fpga-64c)
 
 Command line:
 
-    $ cd hw/syn/altera/opae
-    $ PREFIX=test1 TARGET=fpga NUM_CORES=4 make
+    $ cd hw/syn/opae
+    $ make fpga-<num-of-cores>c
 
-A new folder (ex: `test1_xxx_4c`) will be created and the build will start and take ~30-480 min to complete.
-Setting TARGET=ase will build the project for simulation using Intel ASE.
+Example: `make fpga-4c`
+
+A new folder (ex: `build_fpga_4c`) will be created and the build will start and take ~30-480 min to complete.
 
 
 OPAE Build Configuration
@@ -36,32 +45,35 @@ The hardware configuration file `/hw/rtl/VX_config.vh` defines all the hardware
 
 You configure the syntesis build from the command line:
 
-    $ CONFIGS="-DPERF_ENABLE -DNUM_THREADS=8" make
+    $ CONFIGS="-DPERF_ENABLE -DNUM_THREADS=8" make fpga-4c
 
 OPAE Build Progress
 -------------------
 
 You could check the last 10 lines in the build log for possible errors until build completion.
 
-    $ tail -n 10 <build_dir>/build.log
+    $ tail -n 10 ./build_fpga_<num-of-cores>c/build.log
 
 Check if the build is still running by looking for quartus_sh, quartus_syn, or quartus_fit programs.
 
     $ ps -u <username>
 
+
 If the build fails and you need to restart it, clean up the build folder using the following command:
 
-    $ make clean
+    $ make clean-fpga-<num-of-cores>c
+
+Example: `make clean-fpga-4c`
 
 The file `vortex_afu.gbs` should exist when the build is done:
 
-    $ ls -lsa <build_dir>/vortex_afu.gbs
+    $ ls -lsa ./build_fpga_<num-of-cores>c/vortex_afu.gbs
 
 
 Signing the bitstream and Programming the FPGA
 ----------------------------------------------
 
-    $ cd <build_dir>
+    $ cd ./build_fpga_<num-of-cores>c
     $ PACSign PR -t UPDATE -H openssl_manager -i vortex_afu.gbs -o vortex_afu_unsigned_ssl.gbs
     $ fpgasupdate vortex_afu_unsigned_ssl.gbs
 

docs/index.md

@@ -11,10 +11,10 @@
 - [Debugging](debugging.md)
 - [Useful Links](references.md)
 
+
 ## Installation
 
-- For the different environments Vortex supports, [read this document](environment_setup.md).
-- To install on your own system, [follow this document](install_vortex.md).
+- Refer to the build instructions in [README](../README.md). 
 
 ## Quick Start Scenarios
 
@@ -22,11 +22,9 @@ Running Vortex simulators with different configurations:
 - Run basic driver test with rtlsim driver and Vortex config of 2 clusters, 2 cores, 2 warps, 4 threads
 
     $ ./ci/blackbox.sh --driver=rtlsim --clusters=2 --cores=2 --warps=2 --threads=4  --app=basic
+- Run demo driver test with vlsim driver and Vortex config of 1 clusters, 4 cores, 4 warps, 2 threads
 
-- Run demo driver test with opae driver and Vortex config of 1 clusters, 4 cores, 4 warps, 2 threads
-
-    $ ./ci/blackbox.sh --driver=opae --clusters=1 --cores=4 --warps=4 --threads=2 --app=demo
-
+    $ ./ci/blackbox.sh --driver=vlsim --clusters=1 --cores=4 --warps=4 --threads=2 --app=demo
 - Run dogfood driver test with simx driver and Vortex config of 4 cluster, 4 cores, 8 warps, 6 threads 
 
-    $ ./ci/blackbox.sh --driver=simx --clusters=4 --cores=4 --warps=8 --threads=6  --app=dogfood
+    $ ./ci/blackbox.sh --driver=simx --clusters=4 --cores=4 --warps=8 --threads=6  --app=dogfood

docs/install_vortex.md

@@ -1,124 +0,0 @@
-# Installing and Setting Up the Vortex Environment
-
-## Ubuntu 18.04, 20.04
-
-1. Install the following dependencies:
-
-   ```
-   sudo apt-get install build-essential zlib1g-dev libtinfo-dev libncurses5 uuid-dev libboost-serialization-dev libpng-dev libhwloc-dev
-   ```
-
-2. Upgrade gcc to 11:
-
-   ```
-   sudo apt-get install gcc-11 g++-11
-   ```
-   
-   Multiple gcc versions on Ubuntu can be managed with update-alternatives, e.g.:
-   
-   ```
-   sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-9 9
-   sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-9 9
-   sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-11 11
-   sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-11 11
-   ```
-
-3. Download the Vortex codebase:
-
-   ```
-   git clone --recursive https://github.com/vortexgpgpu/vortex.git
-   ```
-
-4. Install Vortex's prebuilt toolchain:
-
-    ```
-    cd vortex
-    sudo ./ci/toolchain_install.sh -all
-	
-    # By default, the toolchain will install to /opt folder. This is recommended, but you can install the toolchain to a different directory by setting DESTDIR.
-    DESTDIR=$TOOLDIR ./ci/toolchain_install.sh -all
-    ```
-
-5. Set up environment:
-
-    ```
-    export VORTEX_HOME=$TOOLDIR/vortex
-    export LLVM_VORTEX=$TOOLDIR/llvm-vortex
-    export LLVM_POCL=$TOOLDIR/llvm-pocl
-    export POCL_CC_PATH=$TOOLDIR/pocl/compiler
-    export POCL_RT_PATH=$TOOLDIR/pocl/runtime
-    export RISCV_TOOLCHAIN_PATH=$TOOLDIR/riscv-gnu-toolchain
-    export VERILATOR_ROOT=$TOOLDIR/verilator
-    export SV2V_PATH=$TOOLDIR/sv2v
-    export YOSYS_PATH=$TOOLDIR/yosys
-	
-    export PATH=$YOSYS_PATH/bin:$SV2V_PATH/bin:$VERILATOR_ROOT/bin:$PATH
-    ```
-
-6. Build Vortex
-
-    ```
-    make
-    ```
-
-
-## RHEL 8
-Note: depending on the system, some of the toolchain may need to be recompiled for non-Ubuntu Linux. The source for the tools can be found [here](https://github.com/vortexgpgpu/).
-
-1. Install the following dependencies:
-
-   ```
-   sudo yum install libpng-devel boost boost-devel boost-serialization libuuid-devel opencl-headers hwloc hwloc-devel gmp-devel compat-hwloc1
-   ```
-
-2. Upgrade gcc to 11:
-
-    ```
-    sudo yum install gcc-toolset-11
-    ```
-	
-   Multiple gcc versions on Red Hat can be managed with scl
-
-3. Install MPFR 4.2.0:
-
-   Download [the source](https://ftp.gnu.org/gnu/mpfr/) and follow [the installation documentation](https://www.mpfr.org/mpfr-current/mpfr.html#How-to-Install).
-
-4. Download the Vortex codebase:
-
-   ```
-   git clone --recursive https://github.com/vortexgpgpu/vortex.git
-   ```
-
-5. Install Vortex's prebuilt toolchain:
-
-    ```
-    cd vortex
-    sudo ./ci/toolchain_install.sh -all
-	
-    # By default, the toolchain will install to /opt folder. This is recommended, but you can install the toolchain to a different directory by setting DESTDIR.
-    DESTDIR=$TOOLDIR ./ci/toolchain_install.sh -all
-    ```
-
-6. Set up environment:
-
-    ```
-    export VORTEX_HOME=$TOOLDIR/vortex
-    export LLVM_VORTEX=$TOOLDIR/llvm-vortex
-    export LLVM_POCL=$TOOLDIR/llvm-pocl
-    export POCL_CC_PATH=$TOOLDIR/pocl/compiler
-    export POCL_RT_PATH=$TOOLDIR/pocl/runtime
-    export RISCV_TOOLCHAIN_PATH=$TOOLDIR/riscv-gnu-toolchain
-    export VERILATOR_ROOT=$TOOLDIR/verilator
-    export SV2V_PATH=$TOOLDIR/sv2v
-    export YOSYS_PATH=$TOOLDIR/yosys
-	
-    export PATH=$YOSYS_PATH/bin:$SV2V_PATH/bin:$VERILATOR_ROOT/bin:$PATH
-	
-    export LD_LIBRARY_PATH=<path to mpfr>/src/.libs:$LD_LIBRARY_PATH
-    ```
-
-7. Build Vortex
-
-    ```
-    make
-    ```

docs/microarchitecture.md

@@ -24,57 +24,71 @@ Vortex uses the SIMT (Single Instruction, Multiple Threads) execution model with
   - Control the number of warps to activate during execution
   - `WSPAWN` *count, addr*: activate count warps and jump to addr location
 - **Control-Flow Divergence**
-  - Control threads activation when a branch diverges
-    - `SPLIT` *taken, predicate*: apply predicate thread mask and save current state into IPDOM stack
-    - `JOIN`: pop IPDOM stack to restore thread mask
-    - `PRED` *predicate, restore_mask*: thread predicate instruction
+  - Control threads to activate when a branch diverges
+    - `SPLIT` *predicate*: apply 'taken' predicate thread mask adn save 'not-taken' into IPDOM stack
+    - `JOIN`: restore 'not-taken' thread mask
 - **Warp Synchronization**
   - `BAR` *id, count*: stall warps entering barrier *id* until count is reached
 
 ### Vortex Pipeline/Datapath
 
-![Image of Vortex Microarchitecture](./assets/img/vortex_microarchitecture.png)
+![Image of Vortex Microarchitecture](./assets/img/vortex_microarchitecture_v2.png)
 
-Vortex has a 6-stage pipeline:
-
-- **Schedule**
-  - Warp Scheduler
-    - Schedule the next PC into the pipeline
-    - Track stalled, active warps
-  - IPDOM Stack
-    - Save split/join states for divergent threads
-  - Inflight Tracker
-    - Track in-flight instructions
+Vortex has a 5-stage pipeline: FI | ID | Issue | EX | WB.
 
 - **Fetch**
-  - Retrieve instructions from memory
-  - Handle I-cache requests/responses
+  - Warp Scheduler
+    - Track stalled & active warps, resolve branches and barriers, maintain split/join IPDOM stack
+  - Instruction Cache
+    - Retrieve instruction from cache, issue I-cache requests/responses
 - **Decode**
-  - Decode fetched instructions
-  - Notify warp scheduler on control instructions
+  - Decode fetched instructions, notify warp scheduler when the following instructions are decoded:
+    - Branch, tmc, split/join, wspawn
+  - Precompute used_regs mask (needed for Issue stage)
 - **Issue**
+  - Scheduling
+    - In-order issue (operands/execute unit ready), out-of-order commit
   - IBuffer
-    - Store decoded instructions in separate per-warp queues
+    - Store fetched instructions, separate queues per-warp, selects next warp through round-robin scheduling
   - Scoreboard
     - Track in-use registers
-    - Check register use for decoded instructions
-  - Operands Collector
-    - Fetch the operands for issued instructions from the register file
+  - GPRs (General-Purpose Registers) stage
+    - Fetch issued instruction operands and send operands to execute unit
 - **Execute**
   - ALU Unit
-    - Handle arithmetic and branch operations
+    - Single-cycle operations (+,-,>>,<<,&,|,^), Branch instructions (Share ALU resources)
+  - MULDIV Unit
+    - Multiplier - done in 2 cycles
+    - Divider - division and remainder, done in 32 cycles
+      - Implements serial alogrithm (Stalls the pipeline)
   - FPU Unit
-    - Handle floating-point operations
+    - Multi-cycle operations, uses `FPnew` Library on ASIC, uses hard DSPs on FPGA
+  - CSR Unit
+    - Store constant status registers - device caps, FPU status flags, performance counters
+    - Handle external CSR requests (requests from host CPU)
   - LSU Unit
-    - Handle load/store operations
-  - SFU Unit
-    - Handle warp control operations
-    - Handle Control Status Registers (CSRs) operations
+    - Handle load/store operations, issue D-cache requests, handle D-cache responses
+    - Commit load responses - saves storage, Scoreboard tracks completion
+  - GPGPU Unit
+    - Handle GPGPU instructions
+      - TMC, WSPAWN, SPLIT, BAR
+    - JOIN is handled by Warp Scheduler (upon SPLIT response)
 - **Commit**
-  - Write result back to the register file and update the Scoreboard.
-
-### Vortex clustering architecture
-- Sockets
-  - Grouping multiple cores sharing L1 cache
-- Clusters
-  - Grouping of sockets sharing L2 cache
+  - Commit
+    - Update CSR flags, update performance counters
+  - Writeback
+    - Write result back to GPRs, notify Scoreboard (release in-use register), select candidate instruction (ALU unit has highest priority)
+- **Clustering**
+  - Group mulitple cores into clusters (optionally share L2 cache)
+  - Group multiple clusters (optionally share L3 cache)
+  - Configurable at build time
+  - Default configuration:
+    - #Clusters = 1
+    - #Cores = 4
+    - #Warps = 4
+    - #Threads = 4
+- **FPGA AFU Interface**
+  - Manage CPU-GPU comunication
+    - Query devices caps, load kernel instructions and resource buffers, start kernel execution, read destination buffers
+  - Local Memory - GPU access to local DRAM
+  - Reserved I/O addresses - redirect to host CPU, console output

docs/simulation.md

@@ -10,7 +10,7 @@ SimX is a C++ cycle-level in-house simulator developed for Vortex. The relevant
 
 ### FGPA Simulation
 
-The current target FPGA for simulation is the Arria10 Intel Accelerator Card v1.0. The guide to build the fpga with specific configurations is located [here.](fpga_setup.md)
+The current target FPGA for simulation is the Arria10 Intel Accelerator Card v1.0. The guide to build the fpga with specific configurations is located [here.](https://github.com/vortexgpgpu/vortex-dev/blob/master/doc/FPGA_Startup_Guide.md)
 
 ### How to Test
 
@@ -22,15 +22,15 @@ Running tests under specific drivers (rtlsim,simx,fpga) is done using the script
 - *Threads* - used to specify the number of threads (smallest unit of computation) within a configuration.
 - *L2cache* - used to enable the shard l2cache among the Vortex cores.
 - *L3cache* - used to enable the shared l3cache among the Vortex clusters.
-- *Driver* - used to specify which driver to run the Vortex simulation (either rtlsim, opae, xrt, simx).
+- *Driver* - used to specify which driver to run the Vortex simulation (either rtlsim, vlsim, fpga, or simx).
 - *Debug* - used to enable debug mode for the Vortex simulation.
 - *Perf* - used to enable the detailed performance counters within the Vortex simulation.
 - *App* - used to specify which test/benchmark to run in the Vortex simulation. The main choices are vecadd, sgemm, basic, demo, and dogfood. Other tests/benchmarks are located in the `/benchmarks/opencl` folder though not all of them work wit the current version of Vortex.
 - *Args* - used to pass additional arguments to the application.
 
-Example use of command line arguments: Run the sgemm benchmark using the opae driver with a Vortex configuration of 1 cluster, 4 cores, 4 warps, and 4 threads.
+Example use of command line arguments: Run the sgemm benchmark using the vlsim driver with a Vortex configuration of 1 cluster, 4 cores, 4 warps, and 4 threads.
 
-    $ ./ci/blackbox.sh --clusters=1 --cores=4 --warps=4 --threads=4 --driver=opae --app=sgemm
+    $ ./ci/blackbox.sh --clusters=1 --cores=4 --warps=4 --threads=4 --driver=vlsim --app=sgemm
 
 Output from terminal:
 ```

docs/testing.md

@@ -1,47 +0,0 @@
-# Testing
-
-## Running a Vortex application
-
-The framework provides a utility script: blackbox.sh under the /ci/ folder for executing applications in the tests tree.
-You can query the commandline options of the tool using:
-
-    $ ./ci/blackbox.sh --help
-
-To execute sgemm test program on the simx driver and passing "-n10" as argument to sgemm:
-
-    $ ./ci/blackbox.sh --driver=simx --app=sgemm --args="-n10"
-
-You can execute the same application of a GPU architecture with 2 cores:
-
-    $ ./ci/blackbox.sh --core=2 --driver=simx --app=sgemm --args="-n10"
-
-When excuting, Blackbox needs to recompile the driver if the desired architecture changes. 
-It tracks the latest configuration in a file under the current directory blackbox.<driver>.cache.
-To avoid having to rebuild the driver all the time, Blackbox checks if the latest cached configuration matches the current.
-
-## Running Benchmarks
-
-The Vortex test suite is located under the /test/ folder
-You can execute the default regression suite by running the following commands at the root folder.
-
-    $ make -C tests/regression run-simx 
-    $ make -C tests/regression run-rtlsim
-
-You can execute the default opncl suite by running the following commands at the root folder.
-
-    $ make -C tests/opencl run-simx 
-    $ make -C tests/opencl run-rtlsim
-
-## Creating Your Own Regression Tests
-- Inside `test/` you will find a series of folders which are named based on what they test
-- You can view the tests to see which ones have tests similar to what you are trying to create new tests for
-- once you have found a similar baseline, you can copy the folder and rename it to what you are planning to test
-- `testcases.h` contains each of the test case templates
-- `main.cpp` contains the implementation of each of the test cases and builds a test suite of all the tests cases you want
-
-Compile the test case: `make -C tests/regression/<testcase-name>/ clean-all && make -C tests/regression/<testcase-name>/`
-
-Run the test case: `./ci/blackbox.sh --driver=simx --cores=4 --app=<testcase-name> --debug`
-
-## Adding Your Tests to the CI Pipeline
-see `continuous_integration.md`

driver/Makefile

@@ -0,0 +1,29 @@
+all: stub rtlsim simx vlsim
+
+stub:
+	$(MAKE) -C stub
+
+fpga:
+	$(MAKE) -C fpga
+
+asesim:
+	$(MAKE) -C asesim
+
+vlsim:
+	$(MAKE) -C vlsim
+
+rtlsim:
+	$(MAKE) -C rtlsim
+
+simx:
+	$(MAKE) -C simx
+
+clean:
+	$(MAKE) clean -C stub
+	$(MAKE) clean -C fpga
+	$(MAKE) clean -C asesim
+	$(MAKE) clean -C vlsim
+	$(MAKE) clean -C rtlsim
+	$(MAKE) clean -C simx
+
+.PHONY: all stub fpga asesim vlsim rtlsim simx clean

driver/asesim/Makefile

@@ -0,0 +1,73 @@
+OPAE_HOME ?= /tools/opae/1.4.0
+
+RTL_DIR=../../hw/rtl
+
+SCRIPT_DIR=../../hw/scripts
+
+OPAE_SYN_DIR=../../hw/syn/opae
+
+CXXFLAGS += -std=c++11 -Wall -Wextra -pedantic -Wfatal-errors
+
+CXXFLAGS += -I. -I../include -I../../hw -I$(OPAE_HOME)/include -I$(OPAE_SYN_DIR)
+
+LDFLAGS += -L$(OPAE_HOME)/lib -luuid -lopae-c-ase
+
+# stack execution protection
+LDFLAGS +=-z noexecstack
+
+# data relocation and projection
+LDFLAGS +=-z relro -z now
+
+# stack buffer overrun detection
+CXXFLAGS +=-fstack-protector
+
+# Position independent code
+CXXFLAGS += -fPIC
+
+# Add external configuration
+CXXFLAGS += $(CONFIGS)
+
+# Dump perf stats
+CXXFLAGS += -DDUMP_PERF_STATS
+
+LDFLAGS += -shared
+
+PROJECT = libvortex.so
+
+SRCS = ../common/opae.cpp ../common/vx_utils.cpp
+
+# Debugigng
+ifdef DEBUG
+	CXXFLAGS += -g -O0
+else    
+	CXXFLAGS += -O2 -DNDEBUG
+endif
+
+# Enable scope analyzer
+ifdef SCOPE
+	CXXFLAGS += -DSCOPE	
+	SRCS += ../common/vx_scope.cpp
+	SCOPE_H = scope-defs.h 
+endif
+
+# Enable perf counters
+ifdef PERF
+	CXXFLAGS += -DPERF_ENABLE
+endif
+
+all: $(PROJECT)
+
+$(OPAE_SYN_DIR)/vortex_afu.h:
+	$(MAKE) -C $(OPAE_SYN_DIR) vortex_afu.h
+
+scope-defs.h: $(SCRIPT_DIR)/scope.json
+	$(SCRIPT_DIR)/scope.py $(CONFIGS) -cc scope-defs.h -vl $(RTL_DIR)/scope-defs.vh $(SCRIPT_DIR)/scope.json
+
+# generate scope data
+scope: scope-defs.h	
+
+$(PROJECT): $(SRCS) $(OPAE_SYN_DIR)/vortex_afu.h $(SCOPE_H)
+	$(CXX) $(CXXFLAGS) -DUSE_ASE $(SRCS) $(LDFLAGS) -o $(PROJECT)
+
+clean:
+	rm -rf $(PROJECT) *.o scope-defs.h

driver/common/opae.cpp

@@ -0,0 +1,535 @@
+#include <stdint.h>
+#include <iostream>
+#include <stdio.h>
+#include <stdlib.h>
+#include <cstdlib>
+#include <unistd.h>
+#include <assert.h>
+#include <cmath>
+#include <sstream>
+#include <unordered_map>
+#include <list>
+
+#if defined(USE_FPGA) || defined(USE_ASE) 
+#include <opae/fpga.h>
+#include <uuid/uuid.h>
+#elif defined(USE_VLSIM)
+#include <fpga.h>
+#endif
+
+#include "vx_utils.h"
+#include "vx_malloc.h"
+#include <vortex.h>
+#include <VX_config.h>
+#include "vortex_afu.h"
+
+#ifdef SCOPE
+#include "vx_scope.h"
+#endif
+
+#define CHECK_RES(_expr)                                \
+   do {                                                 \
+     fpga_result res = _expr;                           \
+     if (res == FPGA_OK)                                \
+       break;                                           \
+     printf("[VXDRV] Error: '%s' returned %d, %s!\n",   \
+            #_expr, (int)res, fpgaErrStr(res));         \
+     return -1;                                         \
+   } while (false)
+
+///////////////////////////////////////////////////////////////////////////////
+
+#define CMD_MEM_READ        AFU_IMAGE_CMD_MEM_READ
+#define CMD_MEM_WRITE       AFU_IMAGE_CMD_MEM_WRITE
+#define CMD_RUN             AFU_IMAGE_CMD_RUN
+
+#define MMIO_CMD_TYPE       (AFU_IMAGE_MMIO_CMD_TYPE * 4)
+#define MMIO_IO_ADDR        (AFU_IMAGE_MMIO_IO_ADDR * 4)
+#define MMIO_MEM_ADDR       (AFU_IMAGE_MMIO_MEM_ADDR * 4)
+#define MMIO_DATA_SIZE      (AFU_IMAGE_MMIO_DATA_SIZE * 4)
+#define MMIO_DEV_CAPS       (AFU_IMAGE_MMIO_DEV_CAPS * 4)
+#define MMIO_STATUS         (AFU_IMAGE_MMIO_STATUS * 4)
+
+#define STATUS_STATE_BITS   8
+
+///////////////////////////////////////////////////////////////////////////////
+
+class vx_device {
+public:
+    vx_device() 
+        : mem_allocator(
+            ALLOC_BASE_ADDR, 
+            ALLOC_BASE_ADDR + LOCAL_MEM_SIZE,
+            4096,            
+            CACHE_BLOCK_SIZE)
+    {}
+    
+    ~vx_device() {}
+
+    fpga_handle fpga;
+    vortex::MemoryAllocator mem_allocator;
+    unsigned version;
+    unsigned num_cores;
+    unsigned num_warps;
+    unsigned num_threads;
+};
+
+typedef struct vx_buffer_ {
+    uint64_t wsid;
+    void* host_ptr;
+    uint64_t io_addr;
+    vx_device_h hdevice;
+    uint64_t size;
+} vx_buffer_t;
+
+///////////////////////////////////////////////////////////////////////////////
+
+#ifdef DUMP_PERF_STATS
+class AutoPerfDump {
+private:
+    std::list<vx_device_h> devices_;
+
+public:
+    AutoPerfDump() {} 
+
+    ~AutoPerfDump() {
+        for (auto device : devices_) {
+            vx_dump_perf(device, stdout);
+        }
+    }
+
+    void add_device(vx_device_h device) {
+        devices_.push_back(device);
+    }
+
+    void remove_device(vx_device_h device) {
+        devices_.remove(device);
+    }    
+};
+
+AutoPerfDump gAutoPerfDump;
+#endif
+
+///////////////////////////////////////////////////////////////////////////////
+
+extern int vx_dev_caps(vx_device_h hdevice, uint32_t caps_id, uint64_t *value) {
+    if (nullptr == hdevice)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+
+    switch (caps_id) {
+    case VX_CAPS_VERSION:
+        *value = device->version;
+        break;
+    case VX_CAPS_MAX_CORES:
+        *value = device->num_cores;
+        break;
+    case VX_CAPS_MAX_WARPS:
+        *value = device->num_warps;
+        break;
+    case VX_CAPS_MAX_THREADS:
+        *value = device->num_threads;
+        break;
+    case VX_CAPS_CACHE_LINE_SIZE:
+        *value = CACHE_BLOCK_SIZE;
+        break;
+    case VX_CAPS_LOCAL_MEM_SIZE:
+        *value = LOCAL_MEM_SIZE;
+        break;
+    case VX_CAPS_ALLOC_BASE_ADDR:
+        *value = ALLOC_BASE_ADDR;
+        break;
+    case VX_CAPS_KERNEL_BASE_ADDR:
+        *value = STARTUP_ADDR;
+        break;
+    default:
+        fprintf(stderr, "[VXDRV] Error: invalid caps id: %d\n", caps_id);
+        std::abort();
+        return -1;
+    }
+
+    return 0;
+}
+
+extern int vx_dev_open(vx_device_h* hdevice) {
+    if (nullptr == hdevice)
+        return  -1;
+
+    fpga_handle accel_handle;    
+    vx_device* device;   
+
+#ifndef USE_VLSIM
+    fpga_result res;    
+    fpga_token accel_token;
+    fpga_properties filter = nullptr;    
+    fpga_guid guid; 
+    uint32_t num_matches;
+    
+    // Set up a filter that will search for an accelerator
+    CHECK_RES(fpgaGetProperties(nullptr, &filter));
+    res = fpgaPropertiesSetObjectType(filter, FPGA_ACCELERATOR);
+    if (res != FPGA_OK) {
+        fprintf(stderr, "[VXDRV] Error: fpgaGetProperties() returned %d, %s!\n", (int)res, fpgaErrStr(res));
+        fpgaDestroyProperties(&filter);
+        return -1;
+    }
+
+    // Add the desired UUID to the filter
+    uuid_parse(AFU_ACCEL_UUID, guid);
+    res = fpgaPropertiesSetGUID(filter, guid);
+    if (res != FPGA_OK) {
+        fprintf(stderr, "[VXDRV] Error: fpgaPropertiesSetGUID() returned %d, %s!\n", (int)res, fpgaErrStr(res));
+        fpgaDestroyProperties(&filter);
+        return -1;
+    }
+
+    // Do the search across the available FPGA contexts
+    num_matches = 1;
+    res = fpgaEnumerate(&filter, 1, &accel_token, 1, &num_matches);
+    if (res != FPGA_OK) {
+        fprintf(stderr, "[VXDRV] Error: fpgaEnumerate() returned %d, %s!\n", (int)res, fpgaErrStr(res));
+        fpgaDestroyProperties(&filter);
+        return -1;
+    }
+
+    // Not needed anymore
+    fpgaDestroyProperties(&filter);
+
+    if (num_matches < 1) {
+        fprintf(stderr, "[VXDRV] Error: accelerator %s not found!\n", AFU_ACCEL_UUID);
+        fpgaDestroyToken(&accel_token);
+        return -1;
+    }
+
+    // Open accelerator
+    res = fpgaOpen(accel_token, &accel_handle, 0);
+    if (res != FPGA_OK) {
+        fprintf(stderr, "[VXDRV] Error: fpgaOpen() returned %d, %s!\n", (int)res, fpgaErrStr(res));
+        fpgaDestroyToken(&accel_token);
+        return -1;
+    }
+
+    // Done with token
+    fpgaDestroyToken(&accel_token);
+#else
+    // Open accelerator
+    CHECK_RES(fpgaOpen(NULL, &accel_handle, 0));
+#endif
+
+    // allocate device object
+    device = new vx_device();
+    if (nullptr == device) {
+        fpgaClose(accel_handle);
+        return -1;
+    }
+
+    device->fpga = accel_handle;
+    
+    {   
+        // Load device CAPS
+        uint64_t dev_caps;
+        int ret = fpgaReadMMIO64(device->fpga, 0, MMIO_DEV_CAPS, &dev_caps);        
+        if (ret != FPGA_OK) {
+            fpgaClose(accel_handle);
+            return ret;
+        }
+        device->version     = (dev_caps >> 0)  & 0xffff;
+        device->num_cores   = (dev_caps >> 16) & 0xffff;
+        device->num_warps   = (dev_caps >> 32) & 0xffff;
+        device->num_threads = (dev_caps >> 48) & 0xffff;
+    #ifndef NDEBUG    
+        fprintf(stdout, "[VXDRV] DEVCAPS: version=%d, num_cores=%d, num_warps=%d, num_threads=%d\n", 
+                device->version, device->num_cores, device->num_warps, device->num_threads);
+    #endif
+    }
+    
+#ifdef SCOPE
+    {
+        int ret = vx_scope_start(accel_handle, 0, -1);
+        if (ret != 0) {
+            fpgaClose(accel_handle);
+            return ret;
+        }
+    }
+#endif    
+
+    *hdevice = device;
+
+#ifdef DUMP_PERF_STATS
+    gAutoPerfDump.add_device(*hdevice);
+#endif
+
+    return 0;
+}
+
+extern int vx_dev_close(vx_device_h hdevice) {
+    if (nullptr == hdevice)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+
+#ifdef SCOPE
+    vx_scope_stop(device->fpga);
+#endif
+
+#ifdef DUMP_PERF_STATS
+    gAutoPerfDump.remove_device(hdevice);
+    vx_dump_perf(hdevice, stdout);
+#endif
+
+    fpgaClose(device->fpga);
+
+    delete device;
+
+    return 0;
+}
+
+extern int vx_mem_alloc(vx_device_h hdevice, uint64_t size, uint64_t* dev_maddr) {
+    if (nullptr == hdevice 
+     || nullptr == dev_maddr
+     || 0 >= size)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+    return device->mem_allocator.allocate(size, dev_maddr);
+}
+
+extern int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr) {
+    if (nullptr == hdevice)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+    return device->mem_allocator.release(dev_maddr);
+}
+
+extern int vx_buf_alloc(vx_device_h hdevice, uint64_t size, vx_buffer_h* hbuffer) {
+    fpga_result res;
+    void* host_ptr;
+    uint64_t wsid;
+    uint64_t io_addr;
+    vx_buffer_t* buffer;
+
+    if (nullptr == hdevice
+     || 0 >= size
+     || nullptr == hbuffer)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+
+    size_t asize = aligned_size(size, CACHE_BLOCK_SIZE);
+
+    res = fpgaPrepareBuffer(device->fpga, asize, &host_ptr, &wsid, 0);
+    if (FPGA_OK != res) {
+        return -1;
+    }
+
+    // Get the physical address of the buffer in the accelerator
+    res = fpgaGetIOAddress(device->fpga, wsid, &io_addr);
+    if (FPGA_OK != res) {
+        fpgaReleaseBuffer(device->fpga, wsid);
+        return -1;
+    }
+
+    // allocate buffer object
+    buffer = (vx_buffer_t*)malloc(sizeof(vx_buffer_t));
+    if (nullptr == buffer) {
+        fpgaReleaseBuffer(device->fpga, wsid);
+        return -1;
+    }
+
+    buffer->wsid     = wsid;
+    buffer->host_ptr = host_ptr;
+    buffer->io_addr  = io_addr;
+    buffer->hdevice  = hdevice;
+    buffer->size     = asize;
+
+    *hbuffer = buffer;
+
+    return 0;
+}
+
+extern void* vx_host_ptr(vx_buffer_h hbuffer) {
+    if (nullptr == hbuffer)
+        return nullptr;
+
+    vx_buffer_t* buffer = ((vx_buffer_t*)hbuffer);
+    return buffer->host_ptr;
+}
+
+extern int vx_buf_free(vx_buffer_h hbuffer) {
+    if (nullptr == hbuffer)
+        return -1;
+
+    vx_buffer_t* buffer = ((vx_buffer_t*)hbuffer);
+    vx_device *device = ((vx_device*)buffer->hdevice);
+
+    fpgaReleaseBuffer(device->fpga, buffer->wsid);
+
+    free(buffer);
+
+    return 0;
+}
+
+extern int vx_ready_wait(vx_device_h hdevice, uint64_t timeout) {
+    if (nullptr == hdevice)
+        return -1;
+
+    std::unordered_map<uint32_t, std::stringstream> print_bufs;
+    
+    vx_device *device = ((vx_device*)hdevice);
+
+    struct timespec sleep_time; 
+
+#if defined(USE_ASE)
+    sleep_time.tv_sec = 1;
+    sleep_time.tv_nsec = 0;
+#else
+    sleep_time.tv_sec = 0;
+    sleep_time.tv_nsec = 1000000;
+#endif
+
+    // to milliseconds
+    uint64_t sleep_time_ms = (sleep_time.tv_sec * 1000) + (sleep_time.tv_nsec / 1000000);
+    
+    for (;;) {
+        uint64_t status;
+        CHECK_RES(fpgaReadMMIO64(device->fpga, 0, MMIO_STATUS, &status));
+
+        // check for console data
+        uint32_t cout_data = status >> STATUS_STATE_BITS;
+        if (cout_data & 0x1) {
+            // retrieve console data
+            do {
+                char cout_char = (cout_data >> 1) & 0xff;
+                uint32_t cout_tid = (cout_data >> 9) & 0xff;
+                auto& ss_buf = print_bufs[cout_tid];
+                ss_buf << cout_char;
+                if (cout_char == '\n') {
+                    std::cout << std::dec << "#" << cout_tid << ": " << ss_buf.str() << std::flush;
+                    ss_buf.str("");
+                }
+                CHECK_RES(fpgaReadMMIO64(device->fpga, 0, MMIO_STATUS, &status));
+                cout_data = status >> STATUS_STATE_BITS;
+            } while (cout_data & 0x1);
+        }
+
+        uint32_t state = status & ((1 << STATUS_STATE_BITS)-1);
+
+        if (0 == state || 0 == timeout) {
+            for (auto& buf : print_bufs) {
+                auto str = buf.second.str();
+                if (!str.empty()) {
+                std::cout << "#" << buf.first << ": " << str << std::endl;
+                }
+            }
+            if (state != 0) {
+                fprintf(stdout, "[VXDRV] ready-wait timed out: state=%d\n", state);
+            }
+            break;
+        }
+
+        nanosleep(&sleep_time, nullptr);
+        timeout -= sleep_time_ms;
+    };
+
+    return 0;
+}
+
+extern int vx_copy_to_dev(vx_buffer_h hbuffer, uint64_t dev_maddr, uint64_t size, uint64_t src_offset) {
+    if (nullptr == hbuffer 
+     || 0 >= size)
+        return -1;
+
+    vx_buffer_t *buffer = ((vx_buffer_t*)hbuffer);
+    vx_device *device = ((vx_device*)buffer->hdevice);
+
+    uint64_t dev_mem_size = LOCAL_MEM_SIZE; 
+    uint64_t asize = aligned_size(size, CACHE_BLOCK_SIZE);
+
+    // check alignment
+    if (!is_aligned(dev_maddr, CACHE_BLOCK_SIZE))
+        return -1;
+    if (!is_aligned(buffer->io_addr + src_offset, CACHE_BLOCK_SIZE))
+        return -1;
+
+    // bound checking
+    if (src_offset + asize > buffer->size)
+        return -1;
+    if (dev_maddr + asize > dev_mem_size)
+        return -1;
+
+    // Ensure ready for new command
+    if (vx_ready_wait(buffer->hdevice, MAX_TIMEOUT) != 0)
+        return -1;
+
+    auto ls_shift = (int)std::log2(CACHE_BLOCK_SIZE);
+
+    CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_IO_ADDR, (buffer->io_addr + src_offset) >> ls_shift));
+    CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_MEM_ADDR, dev_maddr >> ls_shift));
+    CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_DATA_SIZE, asize >> ls_shift));   
+    CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CMD_TYPE, CMD_MEM_WRITE));
+
+    // Wait for the write operation to finish
+    if (vx_ready_wait(buffer->hdevice, MAX_TIMEOUT) != 0)
+        return -1;
+
+    return 0;
+}
+
+extern int vx_copy_from_dev(vx_buffer_h hbuffer, uint64_t dev_maddr, uint64_t size, uint64_t dest_offset) {
+    if (nullptr == hbuffer 
+     || 0 >= size)
+        return -1;
+
+    vx_buffer_t *buffer = ((vx_buffer_t*)hbuffer);
+    vx_device *device = ((vx_device*)buffer->hdevice);
+
+    uint64_t dev_mem_size = LOCAL_MEM_SIZE;  
+    uint64_t asize = aligned_size(size, CACHE_BLOCK_SIZE);
+
+    // check alignment
+    if (!is_aligned(dev_maddr, CACHE_BLOCK_SIZE))
+        return -1;
+    if (!is_aligned(buffer->io_addr + dest_offset, CACHE_BLOCK_SIZE))
+        return -1; 
+
+    // bound checking
+    if (dest_offset + asize > buffer->size)
+        return -1;
+    if (dev_maddr + asize > dev_mem_size)
+        return -1;
+
+    // Ensure ready for new command
+    if (vx_ready_wait(buffer->hdevice, MAX_TIMEOUT) != 0)
+        return -1;
+
+    auto ls_shift = (int)std::log2(CACHE_BLOCK_SIZE);
+
+    CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_IO_ADDR, (buffer->io_addr + dest_offset) >> ls_shift));
+    CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_MEM_ADDR, dev_maddr >> ls_shift));    
+    CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_DATA_SIZE, asize >> ls_shift));   
+    CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CMD_TYPE, CMD_MEM_READ));
+
+    // Wait for the write operation to finish
+    if (vx_ready_wait(buffer->hdevice, MAX_TIMEOUT) != 0)
+        return -1;
+
+    return 0;
+}
+
+extern int vx_start(vx_device_h hdevice) {
+    if (nullptr == hdevice)
+        return -1;   
+
+    vx_device *device = ((vx_device*)hdevice);
+
+    // Ensure ready for new command
+    if (vx_ready_wait(hdevice, MAX_TIMEOUT) != 0)
+        return -1;    
+  
+    // start execution    
+    CHECK_RES(fpgaWriteMMIO64(device->fpga, 0, MMIO_CMD_TYPE, CMD_RUN));
+
+    return 0;
+}

driver/common/vx_malloc.h

@@ -1,38 +1,22 @@
-// Copyright © 2019-2023
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// 
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
 #pragma once
 
 #include <cstdint>
 #include <assert.h>
-#include <stdio.h>
 
 namespace vortex {
 
 class MemoryAllocator {
 public:
     MemoryAllocator(
-        uint64_t baseAddress,
-        uint64_t capacity,
+        uint64_t minAddress,
+        uint64_t maxAddress,
         uint32_t pageAlign, 
         uint32_t blockAlign) 
-        : baseAddress_(baseAddress)
-        , capacity_(capacity)
+        : nextAddress_(minAddress)
+        , maxAddress_(maxAddress)
         , pageAlign_(pageAlign)
         , blockAlign_(blockAlign)
         , pages_(nullptr)
-        , nextAddress_(0)
-        , allocated_(0)
     {}
 
     ~MemoryAllocator() {
@@ -44,28 +28,14 @@ public:
             currPage = nextPage;
         }
     }
-    
-    uint32_t baseAddress() const {
-        return baseAddress_;
-    }
-
-    uint32_t capacity() const {
-        return capacity_;
-    }
-
-    uint64_t free() const {
-        return (capacity_ - allocated_);
-    }
-
-    uint64_t allocated() const {
-        return allocated_;
-    }
 
+    // NOTE(hansung): This is code running on the CPU, but CPU is still the one
+    // that keeps track of allocation of the GPU memory.  GPU kernel simply runs
+    // assuming that CPU has done the right thing and returned a safe and valid
+    // chunk of memory.
     int allocate(uint64_t size, uint64_t* addr) {
-        if (size == 0 || addr == nullptr) {
-            printf("error: invalid argurments\n");
+        if (size == 0 || addr == nullptr)
             return -1;
-        }
 
         // Align allocation size
         size = AlignSize(size, blockAlign_);
@@ -95,17 +65,15 @@ public:
         if (nullptr == freeBlock) {
             // Allocate a new page for this request
             currPage = this->NewPage(size);
-            if (nullptr == currPage) {
-                printf("error: out of memory\n");
+            if (nullptr == currPage)
                 return -1;
-            }
             freeBlock = currPage->freeSList;
         }   
 
         // Remove the block from the free lists
         assert(freeBlock->size >= size);
-        currPage->RemoveFreeMList(freeBlock);
-        currPage->RemoveFreeSList(freeBlock);
+        currPage->RemoveFreeMBlock(freeBlock);
+        currPage->RemoveFreeSBlock(freeBlock);
 
         // If the free block we have found is larger than what we are looking for,
         // we may be able to split our free block in two.
@@ -119,33 +87,29 @@ public:
             auto newBlock = new block_t(nextAddr, extraBytes);
 
             // Add the new block to the free lists
-            currPage->InsertFreeMList(newBlock);
-            currPage->InsertFreeSList(newBlock);
+            currPage->InsertFreeMBlock(newBlock);
+            currPage->InsertFreeSBlock(newBlock);
         }
 
         // Insert the free block into the used list
-        currPage->InsertUsedList(freeBlock);
+        currPage->InsertUsedBlock(freeBlock);
 
         // Return the free block address
-        *addr = baseAddress_ + freeBlock->addr;
-
-        // Update allocated size
-        allocated_ += size;
+        *addr = freeBlock->addr;
 
         return 0;
     }
 
     int release(uint64_t addr) {
         // Walk all pages to find the pointer
-        uint64_t local_addr = addr - baseAddress_;
         block_t* usedBlock = nullptr;
         auto currPage = pages_;
         while (currPage) {
-            if (local_addr >= currPage->addr
-            &&  local_addr < (currPage->addr + currPage->size)) {
+            if (addr >= currPage->addr
+            &&  addr < (currPage->addr + currPage->size)) {
                 auto currBlock = currPage->usedList;
                 while (currBlock) {
-                    if (currBlock->addr == local_addr) {
+                    if (currBlock->addr == addr) {
                         usedBlock = currBlock;
                         break;
                     }
@@ -157,18 +121,14 @@ public:
         }
 
         // found the corresponding block?
-        if (nullptr == usedBlock) {
-            printf("error: invalid address to release: 0x%lx\n", addr);
+        if (nullptr == usedBlock)
             return -1;
-        }
-
-        auto size = usedBlock->size;
 
         // Remove the block from the used list
-        currPage->RemoveUsedList(usedBlock);
+        currPage->RemoveUsedBlock(usedBlock);
 
         // Insert the block into the free M-list.
-        currPage->InsertFreeMList(usedBlock);
+        currPage->InsertFreeMBlock(usedBlock);
 
         // Check if we can merge adjacent free blocks from the left.        
         if (usedBlock->prevFreeM) {
@@ -185,7 +145,7 @@ public:
                 }
 
                 // Detach previous block from the free S-list since size increased
-                currPage->RemoveFreeSList(prevBlock);
+                currPage->RemoveFreeSBlock(prevBlock);
 
                 // reset usedBlock
                 delete usedBlock;
@@ -208,26 +168,23 @@ public:
                 }
 
                 // Delete next block
-                currPage->RemoveFreeSList(nextBlock);
+                currPage->RemoveFreeSBlock(nextBlock);
                 delete nextBlock;
             }
         }
 
         // Insert the block into the free S-list.
-        currPage->InsertFreeSList(usedBlock);
+        currPage->InsertFreeSBlock(usedBlock);
 
         // Check if we can free empty pages
         if (nullptr == currPage->usedList) {
             // Try to delete the page
             while (currPage && this->DeletePage(currPage)) {
-                currPage = this->FindNextEmptyPage();
+                currPage = this->NextEmptyPage();
             }
 
         }
 
-        // update allocated size
-        allocated_ -= size;
-
         return 0;
     }
 
@@ -283,7 +240,7 @@ private:
             freeSList = freeMList = new block_t(addr, size);
         }
 
-        void InsertUsedList(block_t* block) {
+        void InsertUsedBlock(block_t* block) {
             block->nextUsed = usedList;
             if (usedList) {
                 usedList->prevUsed = block;
@@ -291,7 +248,7 @@ private:
             usedList = block;
         }
 
-        void RemoveUsedList(block_t* block) {
+        void RemoveUsedBlock(block_t* block) {
             if (block->prevUsed) {
                 block->prevUsed->nextUsed = block->nextUsed;
             } else {
@@ -304,7 +261,7 @@ private:
             block->prevUsed = nullptr;
         }
 
-        void InsertFreeMList(block_t* block) {
+        void InsertFreeMBlock(block_t* block) {
             block_t* currBlock = freeMList;
             block_t* prevBlock = nullptr;
             while (currBlock && (currBlock->addr < block->addr)) {
@@ -323,7 +280,7 @@ private:
             }    
         }
 
-        void RemoveFreeMList(block_t* block) {
+        void RemoveFreeMBlock(block_t* block) {
             if (block->prevFreeM) {
                 block->prevFreeM->nextFreeM = block->nextFreeM;
             } else {
@@ -336,7 +293,7 @@ private:
             block->prevFreeM = nullptr;
         }
 
-        void InsertFreeSList(block_t* block) {
+        void InsertFreeSBlock(block_t* block) {
             block_t* currBlock = this->freeSList;
             block_t* prevBlock = nullptr;
             while (currBlock && (currBlock->size > block->size)) {
@@ -355,7 +312,7 @@ private:
             }
         }
 
-        void RemoveFreeSList(block_t* block) {
+        void RemoveFreeSBlock(block_t* block) {
             if (block->prevFreeS) {
                 block->prevFreeS->nextFreeS = block->nextFreeS;
             } else {
@@ -371,7 +328,7 @@ private:
 
     page_t* NewPage(uint64_t size) {
         // Increase buffer size to include the page and first block size
-        // also add padding to ensure page alignment
+        // also add padding to ensure page aligment
         size = AlignSize(size, pageAlign_);
 
         // Allocate page memory
@@ -379,7 +336,7 @@ private:
         nextAddress_ += size;
 
         // Overflow check
-        if (nextAddress_ > capacity_)
+        if (nextAddress_ > maxAddress_)
             return nullptr;
 
         // Allocate object
@@ -428,7 +385,7 @@ private:
         return true;
     }
 
-    page_t* FindNextEmptyPage() {
+    page_t* NextEmptyPage() {
        auto currPage = pages_;
         while (currPage) {
             if (nullptr == currPage->usedList)
@@ -443,13 +400,11 @@ private:
         return (size + alignment - 1) & ~(alignment - 1);
     }
 
-    uint64_t baseAddress_;
-    uint64_t capacity_;
-    uint32_t pageAlign_;    
-    uint32_t blockAlign_;    
-    page_t*  pages_;
     uint64_t nextAddress_;
-    uint64_t allocated_;
+    uint64_t maxAddress_;
+    uint32_t pageAlign_;    
+    uint32_t blockAlign_;
+    page_t* pages_;    
 };
 
-} // namespace vortex
+} // namespace vortex

driver/common/vx_scope.cpp

@@ -0,0 +1,250 @@
+#include "vx_scope.h"
+#include <iostream>
+#include <fstream>
+#include <thread>
+#include <chrono>
+#include <vector>
+#include <assert.h>
+#include <chrono>
+#include <thread>
+#include <mutex>
+#include <VX_config.h>
+#include <vortex_afu.h>
+#include <scope-defs.h>
+
+#define FRAME_FLUSH_SIZE 100
+
+#define CHECK_RES(_expr)                            \
+   do {                                             \
+     fpga_result res = _expr;                       \
+     if (res == FPGA_OK)                            \
+       break;                                       \
+     printf("OPAE Error: '%s' returned %d, %s!\n",  \
+            #_expr, (int)res, fpgaErrStr(res));     \
+     return -1;                                     \
+   } while (false)
+
+#define MMIO_SCOPE_READ     (AFU_IMAGE_MMIO_SCOPE_READ * 4)
+#define MMIO_SCOPE_WRITE    (AFU_IMAGE_MMIO_SCOPE_WRITE * 4)
+
+#define CMD_GET_VALID   0 
+#define CMD_GET_DATA    1 
+#define CMD_GET_WIDTH   2
+#define CMD_GET_COUNT   3 
+#define CMD_SET_START   4
+#define CMD_SET_STOP    5
+#define CMD_GET_OFFSET  6
+
+static constexpr int num_modules = sizeof(scope_modules) / sizeof(scope_module_t);
+
+static constexpr int num_taps = sizeof(scope_taps) / sizeof(scope_tap_t);
+
+constexpr int calcFrameWidth(int index = 0) {
+    return (index < num_taps) ? (scope_taps[index].width + calcFrameWidth(index + 1)) : 0;
+}
+
+static constexpr int fwidth = calcFrameWidth();
+
+#ifdef HANG_TIMEOUT
+static std::thread g_timeout_thread;
+static std::mutex g_timeout_mutex;
+
+static void timeout_callback(fpga_handle fpga) {
+    std::this_thread::sleep_for(std::chrono::seconds{HANG_TIMEOUT});
+    vx_scope_stop(fpga);
+    fpgaClose(fpga);
+    exit(0);
+}
+#endif
+
+uint64_t print_clock(std::ofstream& ofs, uint64_t delta, uint64_t timestamp) {
+    while (delta != 0) {
+        ofs << '#' << timestamp++ << std::endl;
+        ofs << "b0 0" << std::endl;
+        ofs << '#' << timestamp++ << std::endl;
+        ofs << "b1 0" << std::endl;
+        --delta;
+    }
+    return timestamp;
+}
+
+void dump_taps(std::ofstream& ofs, int module) {    
+     for (int i = 0; i < num_taps; ++i) {
+        auto& tap = scope_taps[i];
+        if (tap.module != module)
+            continue;
+        ofs << "$var reg " << tap.width << " " << (i + 1) << " " << tap.name << " $end" << std::endl;                
+    }
+}
+
+void dump_module(std::ofstream& ofs, int parent) {
+    for (auto& module : scope_modules) {
+        if (module.parent != parent)
+            continue;
+        if (module.name[0] == '*') {
+            ofs << "$var reg 1 0 clk $end" << std::endl;
+        } else {
+            ofs << "$scope module " << module.name << " $end" << std::endl;
+        }        
+        dump_module(ofs, module.index);
+        dump_taps(ofs, module.index);
+        if (module.name[0] != '*') {
+            ofs << "$upscope $end" << std::endl;
+        }
+    }
+}
+
+int vx_scope_start(fpga_handle hfpga, uint64_t start_time, uint64_t stop_time) {    
+    if (nullptr == hfpga)
+        return -1;  
+    
+    if (stop_time != uint64_t(-1)) {
+        // set stop time
+        uint64_t cmd_stop = ((stop_time << 3) | CMD_SET_STOP);
+        CHECK_RES(fpgaWriteMMIO64(hfpga, 0, MMIO_SCOPE_WRITE, cmd_stop));
+        std::cout << "scope stop time: " << std::dec << stop_time << "s" << std::endl;
+    }
+    
+    // start recording
+    uint64_t cmd_delay = ((start_time << 3) | CMD_SET_START);
+    CHECK_RES(fpgaWriteMMIO64(hfpga, 0, MMIO_SCOPE_WRITE, cmd_delay));
+    std::cout << "scope start time: " << std::dec << start_time << "s" << std::endl;
+
+#ifdef HANG_TIMEOUT
+    g_timeout_thread = std::thread(timeout_callback, hfpga);
+    g_timeout_thread.detach();
+#endif
+
+    return 0;
+}
+
+int vx_scope_stop(fpga_handle hfpga) {    
+#ifdef HANG_TIMEOUT
+    if (!g_timeout_mutex.try_lock())
+        return 0;
+#endif
+
+    if (nullptr == hfpga)
+        return -1;
+
+    // forced stop
+    uint64_t cmd_stop = ((0 << 3) | CMD_SET_STOP);
+    CHECK_RES(fpgaWriteMMIO64(hfpga, 0, MMIO_SCOPE_WRITE, cmd_stop));
+
+    std::cout << "scope trace dump begin..." << std::endl;
+
+    std::ofstream ofs("trace.vcd");
+
+    ofs << "$version Generated by Vortex Scope $end" << std::endl;
+    ofs << "$timescale 1 ns $end" << std::endl;
+    ofs << "$scope module TOP $end" << std::endl;
+    
+    dump_module(ofs, -1);
+    dump_taps(ofs, -1);
+    ofs << "$upscope $end" << std::endl;
+    ofs << "enddefinitions $end" << std::endl;
+
+    uint64_t frame_width, max_frames, data_valid, offset, delta;   
+    uint64_t timestamp = 0;
+    uint64_t frame_offset = 0;
+    uint64_t frame_no = 0;  
+    int signal_id = 0;
+    int signal_offset = 0; 
+
+    // wait for recording to terminate
+    CHECK_RES(fpgaWriteMMIO64(hfpga, 0, MMIO_SCOPE_WRITE, CMD_GET_VALID));
+    do {        
+        CHECK_RES(fpgaReadMMIO64(hfpga, 0, MMIO_SCOPE_READ, &data_valid));        
+        if (data_valid)
+            break;
+        std::this_thread::sleep_for(std::chrono::seconds(1));
+    } while (true);
+
+    // get frame width
+    CHECK_RES(fpgaWriteMMIO64(hfpga, 0, MMIO_SCOPE_WRITE, CMD_GET_WIDTH));
+    CHECK_RES(fpgaReadMMIO64(hfpga, 0, MMIO_SCOPE_READ, &frame_width));
+    std::cout << "scope::frame_width=" << std::dec << frame_width << std::endl;  
+
+    if (fwidth != (int)frame_width) {   
+        std::cerr << "invalid frame_width: expecting " << std::dec << fwidth << "!" << std::endl;
+        std::abort();
+    }
+
+    // get max frames
+    CHECK_RES(fpgaWriteMMIO64(hfpga, 0, MMIO_SCOPE_WRITE, CMD_GET_COUNT));
+    CHECK_RES(fpgaReadMMIO64(hfpga, 0, MMIO_SCOPE_READ, &max_frames));
+    std::cout << "scope::max_frames=" << std::dec << max_frames << std::endl;    
+
+    // get offset    
+    CHECK_RES(fpgaWriteMMIO64(hfpga, 0, MMIO_SCOPE_WRITE, CMD_GET_OFFSET));    
+    CHECK_RES(fpgaReadMMIO64(hfpga, 0, MMIO_SCOPE_READ, &offset));
+
+    // get data
+    CHECK_RES(fpgaWriteMMIO64(hfpga, 0, MMIO_SCOPE_WRITE, CMD_GET_DATA));
+
+    // print clock header
+    CHECK_RES(fpgaReadMMIO64(hfpga, 0, MMIO_SCOPE_READ, &delta));
+    timestamp = print_clock(ofs, offset + delta + 2, timestamp);
+    signal_id = num_taps;
+
+    std::vector<char> signal_data(frame_width+1);
+
+    do {
+        if (frame_no == (max_frames-1)) {
+            // verify last frame is valid
+            CHECK_RES(fpgaWriteMMIO64(hfpga, 0, MMIO_SCOPE_WRITE, CMD_GET_VALID));
+            CHECK_RES(fpgaReadMMIO64(hfpga, 0, MMIO_SCOPE_READ, &data_valid));  
+            assert(data_valid == 1);
+            CHECK_RES(fpgaWriteMMIO64(hfpga, 0, MMIO_SCOPE_WRITE, CMD_GET_DATA));
+        }
+
+        // read next data words
+        uint64_t word;
+        CHECK_RES(fpgaReadMMIO64(hfpga, 0, MMIO_SCOPE_READ, &word));
+        
+        do {          
+            int signal_width = scope_taps[signal_id-1].width;
+            int word_offset = frame_offset % 64;
+
+            signal_data[signal_width - signal_offset - 1] = ((word >> word_offset) & 0x1) ? '1' : '0';
+
+            ++signal_offset;
+            ++frame_offset;
+
+            if (signal_offset == signal_width) {
+                signal_data[signal_width] = 0; // string null termination
+                ofs << 'b' << signal_data.data() << ' ' << signal_id << std::endl;
+                signal_offset = 0;            
+                --signal_id;
+            }
+
+            if (frame_offset == frame_width) {   
+                assert(0 == signal_offset);   
+                frame_offset = 0;
+                ++frame_no;
+
+                if (frame_no != max_frames) {    
+                    // print clock header
+                    CHECK_RES(fpgaReadMMIO64(hfpga, 0, MMIO_SCOPE_READ, &delta));
+                    timestamp = print_clock(ofs, delta + 1, timestamp);            
+                    signal_id = num_taps;
+                    if (0 == (frame_no % FRAME_FLUSH_SIZE)) {
+                        ofs << std::flush;
+                        std::cout << "*** " << frame_no << "/" << max_frames << " frames" << std::endl;
+                    }
+                }                     
+            }
+            
+        } while ((frame_offset % 64) != 0);
+
+    } while (frame_no != max_frames);
+
+    std::cout << "scope trace dump done! - " << (timestamp/2) << " cycles" << std::endl;
+
+    // verify data not valid
+    CHECK_RES(fpgaWriteMMIO64(hfpga, 0, MMIO_SCOPE_WRITE, CMD_GET_VALID));
+    CHECK_RES(fpgaReadMMIO64(hfpga, 0, MMIO_SCOPE_READ, &data_valid));  
+    assert(data_valid == 0);
+
+    return 0;
+}

driver/common/vx_scope.h

@@ -0,0 +1,19 @@
+#pragma once
+
+#include <stdint.h>
+
+#ifdef USE_VLSIM
+#include <fpga.h>
+#else
+#include <opae/fpga.h>
+#endif
+
+#if defined(USE_FPGA)
+#define HANG_TIMEOUT 60
+#else
+#define HANG_TIMEOUT (30*60)
+#endif
+
+int vx_scope_start(fpga_handle hfpga, uint64_t start_time = 0, uint64_t stop_time = -1);
+
+int vx_scope_stop(fpga_handle hfpga);

driver/common/vx_utils.cpp

@@ -0,0 +1,356 @@
+#include "vx_utils.h"
+#include <iostream>
+#include <fstream>
+#include <cstring>
+#include <vortex.h>
+#include <VX_config.h>
+#include <assert.h>
+
+uint64_t aligned_size(uint64_t size, uint64_t alignment) {        
+    assert(0 == (alignment & (alignment - 1)));
+    return (size + alignment - 1) & ~(alignment - 1);
+}
+
+bool is_aligned(uint64_t addr, uint64_t alignment) {
+    assert(0 == (alignment & (alignment - 1)));
+    return 0 == (addr & (alignment - 1));
+}
+
+extern int vx_upload_kernel_bytes(vx_device_h device, const void* content, uint64_t size) {
+  int err = 0;
+
+  if (NULL == content || 0 == size)
+    return -1;
+
+  uint32_t buffer_transfer_size = 65536; // 64 KB
+  uint64_t kernel_base_addr;
+  err = vx_dev_caps(device, VX_CAPS_KERNEL_BASE_ADDR, &kernel_base_addr);
+  if (err != 0)
+    return -1;
+
+  // allocate device buffer
+  vx_buffer_h buffer;
+  err = vx_buf_alloc(device, buffer_transfer_size, &buffer);
+  if (err != 0)
+    return -1; 
+
+  // get buffer address
+  auto buf_ptr = (uint8_t*)vx_host_ptr(buffer);
+
+  //
+  // upload content
+  //
+
+  uint64_t offset = 0;
+  while (offset < size) {
+    auto chunk_size = std::min<uint64_t>(buffer_transfer_size, size - offset);
+    std::memcpy(buf_ptr, (uint8_t*)content + offset, chunk_size);
+
+    /*printf("***  Upload Kernel to 0x%0x: data=", kernel_base_addr + offset);
+    for (int i = 0, n = ((chunk_size+7)/8); i < n; ++i) {
+      printf("%08x", ((uint64_t*)((uint8_t*)content + offset))[n-1-i]);
+    }
+    printf("\n");*/
+
+    err = vx_copy_to_dev(buffer, kernel_base_addr + offset, chunk_size, 0);
+    if (err != 0) {
+      vx_buf_free(buffer);
+      return err;
+    }
+    offset += chunk_size;
+  }
+
+  vx_buf_free(buffer);
+
+  return 0;
+}
+
+extern int vx_upload_kernel_file(vx_device_h device, const char* filename) {
+  std::ifstream ifs(filename);
+  if (!ifs) {
+    std::cout << "error: " << filename << " not found" << std::endl;
+    return -1;
+  }
+
+  // read file content
+  ifs.seekg(0, ifs.end);
+  auto size = ifs.tellg();
+  auto content = new char [size];   
+  ifs.seekg(0, ifs.beg);
+  ifs.read(content, size);
+
+  // upload
+  int err = vx_upload_kernel_bytes(device, content, size);
+
+  // release buffer
+  delete[] content;
+
+  return err;
+}
+
+/*static uint32_t get_csr_32(const uint32_t* buffer, int addr) {
+  uint32_t value_lo = buffer[addr - CSR_MPM_BASE];
+  return value_lo;
+}*/
+
+static uint64_t get_csr_64(const uint32_t* buffer, int addr) {
+  uint32_t value_lo = buffer[addr - CSR_MPM_BASE];
+  uint32_t value_hi = buffer[addr - CSR_MPM_BASE + 32];
+  return (uint64_t(value_hi) << 32) | value_lo;
+}
+
+extern int vx_dump_perf(vx_device_h device, FILE* stream) {
+  int ret = 0;
+
+  uint64_t instrs = 0;
+  uint64_t cycles = 0;
+
+#ifdef PERF_ENABLE    
+  // PERF: pipeline stalls
+  uint64_t ibuffer_stalls = 0;
+  uint64_t scoreboard_stalls = 0;
+  uint64_t lsu_stalls = 0;
+  uint64_t fpu_stalls = 0;
+  uint64_t csr_stalls = 0;
+  uint64_t alu_stalls = 0;
+  uint64_t gpu_stalls = 0;
+  // PERF: decode
+  uint64_t loads = 0;
+  uint64_t stores = 0;
+  uint64_t branches = 0;
+  // PERF: Icache 
+  uint64_t icache_reads = 0;
+  uint64_t icache_read_misses = 0;
+  // PERF: Dcache 
+  uint64_t dcache_reads = 0;
+  uint64_t dcache_writes = 0;
+  uint64_t dcache_read_misses = 0;
+  uint64_t dcache_write_misses = 0;
+  uint64_t dcache_bank_stalls = 0;  
+  uint64_t dcache_mshr_stalls = 0;
+  // PERF: shared memory
+  uint64_t smem_reads = 0;
+  uint64_t smem_writes = 0;
+  uint64_t smem_bank_stalls = 0;
+  // PERF: memory
+  uint64_t mem_reads = 0;
+  uint64_t mem_writes = 0;
+  uint64_t mem_lat = 0;
+#ifdef EXT_TEX_ENABLE
+  // PERF: texunit
+  uint64_t tex_mem_reads = 0;
+  uint64_t tex_mem_lat = 0;
+#endif
+#endif     
+
+  uint64_t num_cores;
+  ret = vx_dev_caps(device, VX_CAPS_MAX_CORES, &num_cores);
+  if (ret != 0)
+    return ret;
+
+  vx_buffer_h staging_buf;
+  ret = vx_buf_alloc(device, 64 * sizeof(uint32_t), &staging_buf);
+  if (ret != 0)
+    return ret;
+
+  auto staging_ptr = (uint32_t*)vx_host_ptr(staging_buf);
+      
+  for (unsigned core_id = 0; core_id < num_cores; ++core_id) {
+    ret = vx_copy_from_dev(staging_buf, IO_CSR_ADDR + 64 * sizeof(uint32_t) * core_id, 64 * sizeof(uint32_t), 0);
+    if (ret != 0) {
+      vx_buf_free(staging_buf);
+      return ret;
+    }
+
+    uint64_t instrs_per_core = get_csr_64(staging_ptr, CSR_MINSTRET);
+    uint64_t cycles_per_core = get_csr_64(staging_ptr, CSR_MCYCLE);
+    float IPC = (float)(double(instrs_per_core) / double(cycles_per_core));
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: instrs=%ld, cycles=%ld, IPC=%f\n", core_id, instrs_per_core, cycles_per_core, IPC);            
+    instrs += instrs_per_core;
+    cycles = std::max<uint64_t>(cycles_per_core, cycles);
+
+  #ifdef PERF_ENABLE
+    // PERF: pipeline    
+    // ibuffer_stall
+    uint64_t ibuffer_stalls_per_core = get_csr_64(staging_ptr, CSR_MPM_IBUF_ST);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: ibuffer stalls=%ld\n", core_id, ibuffer_stalls_per_core);
+    ibuffer_stalls += ibuffer_stalls_per_core;
+    // scoreboard_stall
+    uint64_t scoreboard_stalls_per_core = get_csr_64(staging_ptr, CSR_MPM_SCRB_ST);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: scoreboard stalls=%ld\n", core_id, scoreboard_stalls_per_core);
+    scoreboard_stalls += scoreboard_stalls_per_core;
+    // alu_stall
+    uint64_t alu_stalls_per_core = get_csr_64(staging_ptr, CSR_MPM_ALU_ST);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: alu unit stalls=%ld\n", core_id, alu_stalls_per_core);
+    alu_stalls += alu_stalls_per_core;      
+    // lsu_stall
+    uint64_t lsu_stalls_per_core = get_csr_64(staging_ptr, CSR_MPM_LSU_ST);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: lsu unit stalls=%ld\n", core_id, lsu_stalls_per_core);
+    lsu_stalls += lsu_stalls_per_core;
+    // csr_stall
+    uint64_t csr_stalls_per_core = get_csr_64(staging_ptr, CSR_MPM_CSR_ST);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: csr unit stalls=%ld\n", core_id, csr_stalls_per_core);
+    csr_stalls += csr_stalls_per_core;    
+    // fpu_stall
+    uint64_t fpu_stalls_per_core = get_csr_64(staging_ptr, CSR_MPM_FPU_ST);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: fpu unit stalls=%ld\n", core_id, fpu_stalls_per_core);
+    fpu_stalls += fpu_stalls_per_core;      
+    // gpu_stall
+    uint64_t gpu_stalls_per_core = get_csr_64(staging_ptr, CSR_MPM_GPU_ST);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: gpu unit stalls=%ld\n", core_id, gpu_stalls_per_core);
+    gpu_stalls += gpu_stalls_per_core;  
+
+    // PERF: decode
+    // loads
+    uint64_t loads_per_core = get_csr_64(staging_ptr, CSR_MPM_LOADS);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: loads=%ld\n", core_id, loads_per_core);
+    loads += loads_per_core;
+    // stores
+    uint64_t stores_per_core = get_csr_64(staging_ptr, CSR_MPM_STORES);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: stores=%ld\n", core_id, stores_per_core);
+    stores += stores_per_core;
+    // branches
+    uint64_t branches_per_core = get_csr_64(staging_ptr, CSR_MPM_BRANCHES);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: branches=%ld\n", core_id, branches_per_core);
+    branches += branches_per_core;
+
+    // PERF: Icache
+    // total reads
+    uint64_t icache_reads_per_core = get_csr_64(staging_ptr, CSR_MPM_ICACHE_READS);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: icache reads=%ld\n", core_id, icache_reads_per_core);
+    icache_reads += icache_reads_per_core;
+    // read misses
+    uint64_t icache_miss_r_per_core = get_csr_64(staging_ptr, CSR_MPM_ICACHE_MISS_R);
+    int icache_read_hit_ratio = (int)((1.0 - (double(icache_miss_r_per_core) / double(icache_reads_per_core))) * 100);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: icache misses=%ld (hit ratio=%d%%)\n", core_id, icache_miss_r_per_core, icache_read_hit_ratio);
+    icache_read_misses += icache_miss_r_per_core;
+
+    // PERF: Dcache
+    // total reads
+    uint64_t dcache_reads_per_core = get_csr_64(staging_ptr, CSR_MPM_DCACHE_READS);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: dcache reads=%ld\n", core_id, dcache_reads_per_core);
+    dcache_reads += dcache_reads_per_core;
+    // total write
+    uint64_t dcache_writes_per_core = get_csr_64(staging_ptr, CSR_MPM_DCACHE_WRITES);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: dcache writes=%ld\n", core_id, dcache_writes_per_core);
+    dcache_writes += dcache_writes_per_core;
+    // read misses
+    uint64_t dcache_miss_r_per_core = get_csr_64(staging_ptr, CSR_MPM_DCACHE_MISS_R);
+    int dcache_read_hit_ratio = (int)((1.0 - (double(dcache_miss_r_per_core) / double(dcache_reads_per_core))) * 100);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: dcache read misses=%ld (hit ratio=%d%%)\n", core_id, dcache_miss_r_per_core, dcache_read_hit_ratio);
+    dcache_read_misses += dcache_miss_r_per_core;
+    // read misses
+    uint64_t dcache_miss_w_per_core = get_csr_64(staging_ptr, CSR_MPM_DCACHE_MISS_W);
+    int dcache_write_hit_ratio = (int)((1.0 - (double(dcache_miss_w_per_core) / double(dcache_writes_per_core))) * 100);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: dcache write misses=%ld (hit ratio=%d%%)\n", core_id, dcache_miss_w_per_core, dcache_write_hit_ratio);
+    dcache_write_misses += dcache_miss_w_per_core;
+    // bank_stalls
+    uint64_t dcache_bank_st_per_core = get_csr_64(staging_ptr, CSR_MPM_DCACHE_BANK_ST);
+    int dcache_bank_utilization = (int)((double(dcache_reads_per_core + dcache_writes_per_core) / double(dcache_reads_per_core + dcache_writes_per_core + dcache_bank_st_per_core)) * 100);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: dcache bank stalls=%ld (utilization=%d%%)\n", core_id, dcache_bank_st_per_core, dcache_bank_utilization);
+    dcache_bank_stalls += dcache_bank_st_per_core;
+    // mshr_stalls
+    uint64_t dcache_mshr_st_per_core = get_csr_64(staging_ptr, CSR_MPM_DCACHE_MSHR_ST);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: dcache mshr stalls=%ld\n", core_id, dcache_mshr_st_per_core);
+    dcache_mshr_stalls += dcache_mshr_st_per_core; 
+
+    // PERF: SMEM
+    // total reads
+    uint64_t smem_reads_per_core = get_csr_64(staging_ptr, CSR_MPM_SMEM_READS);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: smem reads=%ld\n", core_id, smem_reads_per_core);
+    smem_reads += smem_reads_per_core;
+    // total write
+    uint64_t smem_writes_per_core = get_csr_64(staging_ptr, CSR_MPM_SMEM_WRITES);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: smem writes=%ld\n", core_id, smem_writes_per_core);
+    smem_writes += smem_writes_per_core;
+    // bank_stalls
+    uint64_t smem_bank_st_per_core = get_csr_64(staging_ptr, CSR_MPM_SMEM_BANK_ST);
+    int smem_bank_utilization = (int)((double(smem_reads_per_core + smem_writes_per_core) / double(smem_reads_per_core + smem_writes_per_core + smem_bank_st_per_core)) * 100);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: smem bank stalls=%ld (utilization=%d%%)\n", core_id, smem_bank_st_per_core, smem_bank_utilization);
+    smem_bank_stalls += smem_bank_st_per_core;
+
+    // PERF: memory
+    uint64_t mem_reads_per_core  = get_csr_64(staging_ptr, CSR_MPM_MEM_READS);
+    uint64_t mem_writes_per_core = get_csr_64(staging_ptr, CSR_MPM_MEM_WRITES);
+    uint64_t mem_lat_per_core    = get_csr_64(staging_ptr, CSR_MPM_MEM_LAT);      
+    int mem_avg_lat = (int)(double(mem_lat_per_core) / double(mem_reads_per_core));       
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: memory requests=%ld (reads=%ld, writes=%ld)\n", core_id, (mem_reads_per_core + mem_writes_per_core), mem_reads_per_core, mem_writes_per_core);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: memory latency=%d cycles\n", core_id, mem_avg_lat);
+    mem_reads  += mem_reads_per_core;
+    mem_writes += mem_writes_per_core;
+    mem_lat    += mem_lat_per_core;    
+  
+  #ifdef EXT_TEX_ENABLE
+    // total reads
+    uint64_t tex_reads_per_core = get_csr_64(staging_ptr, CSR_MPM_TEX_READS);
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: tex memory reads=%ld\n", core_id, tex_reads_per_core);
+    tex_mem_reads += tex_reads_per_core;
+    
+    // read latency
+    uint64_t tex_lat_per_core = get_csr_64(staging_ptr, CSR_MPM_TEX_LAT);
+    int tex_avg_lat = (int)(double(tex_lat_per_core) / double(tex_reads_per_core));
+    if (num_cores > 1) fprintf(stream, "PERF: core%d: tex memory latency=%d cycles\n", core_id, tex_avg_lat);
+    tex_mem_lat += tex_lat_per_core;
+  #endif
+  #endif
+  }  
+  
+  float IPC = (float)(double(instrs) / double(cycles));
+  fprintf(stream, "PERF: instrs=%ld, cycles=%ld, IPC=%f\n", instrs, cycles, IPC);    
+      
+#ifdef PERF_ENABLE
+  int icache_read_hit_ratio = (int)((1.0 - (double(icache_read_misses) / double(icache_reads))) * 100);
+  int dcache_read_hit_ratio = (int)((1.0 - (double(dcache_read_misses) / double(dcache_reads))) * 100);
+  int dcache_write_hit_ratio = (int)((1.0 - (double(dcache_write_misses) / double(dcache_writes))) * 100);
+  int dcache_bank_utilization = (int)((double(dcache_reads + dcache_writes) / double(dcache_reads + dcache_writes + dcache_bank_stalls)) * 100);
+  int smem_bank_utilization = (int)((double(smem_reads + smem_writes) / double(smem_reads + smem_writes + smem_bank_stalls)) * 100);
+  int mem_avg_lat = (int)(double(mem_lat) / double(mem_reads));
+  fprintf(stream, "PERF: ibuffer stalls=%ld\n", ibuffer_stalls);
+  fprintf(stream, "PERF: scoreboard stalls=%ld\n", scoreboard_stalls);
+  fprintf(stream, "PERF: alu unit stalls=%ld\n", alu_stalls);
+  fprintf(stream, "PERF: lsu unit stalls=%ld\n", lsu_stalls);
+  fprintf(stream, "PERF: csr unit stalls=%ld\n", csr_stalls);
+  fprintf(stream, "PERF: fpu unit stalls=%ld\n", fpu_stalls);
+  fprintf(stream, "PERF: gpu unit stalls=%ld\n", gpu_stalls);
+  fprintf(stream, "PERF: loads=%ld\n", loads);
+  fprintf(stream, "PERF: stores=%ld\n", stores);
+  fprintf(stream, "PERF: branches=%ld\n", branches);
+  fprintf(stream, "PERF: icache reads=%ld\n", icache_reads);
+  fprintf(stream, "PERF: icache read misses=%ld (hit ratio=%d%%)\n", icache_read_misses, icache_read_hit_ratio);
+  fprintf(stream, "PERF: dcache reads=%ld\n", dcache_reads);
+  fprintf(stream, "PERF: dcache writes=%ld\n", dcache_writes);
+  fprintf(stream, "PERF: dcache read misses=%ld (hit ratio=%d%%)\n", dcache_read_misses, dcache_read_hit_ratio);
+  fprintf(stream, "PERF: dcache write misses=%ld (hit ratio=%d%%)\n", dcache_write_misses, dcache_write_hit_ratio);  
+  fprintf(stream, "PERF: dcache bank stalls=%ld (utilization=%d%%)\n", dcache_bank_stalls, dcache_bank_utilization);
+  fprintf(stream, "PERF: dcache mshr stalls=%ld\n", dcache_mshr_stalls);
+  fprintf(stream, "PERF: smem reads=%ld\n", smem_reads);
+  fprintf(stream, "PERF: smem writes=%ld\n", smem_writes); 
+  fprintf(stream, "PERF: smem bank stalls=%ld (utilization=%d%%)\n", smem_bank_stalls, smem_bank_utilization);
+  fprintf(stream, "PERF: memory requests=%ld (reads=%ld, writes=%ld)\n", (mem_reads + mem_writes), mem_reads, mem_writes);
+  fprintf(stream, "PERF: memory average latency=%d cycles\n", mem_avg_lat);
+#ifdef EXT_TEX_ENABLE
+  int tex_avg_lat = (int)(double(tex_mem_lat) / double(tex_mem_reads));
+  fprintf(stream, "PERF: tex memory reads=%ld\n", tex_mem_reads);
+  fprintf(stream, "PERF: tex memory latency=%d cycles\n", tex_avg_lat);
+#endif
+#endif
+
+  // release allocated resources
+  vx_buf_free(staging_buf);
+
+  return ret;
+}
+
+// Deprecated API functions
+
+extern int vx_alloc_shared_mem(vx_device_h hdevice, uint64_t size, vx_buffer_h* hbuffer) {
+  return vx_buf_alloc(hdevice, size, hbuffer);
+}
+
+extern int vx_buf_release(vx_buffer_h hbuffer) {
+  return vx_buf_free(hbuffer);
+}
+
+extern int vx_alloc_dev_mem(vx_device_h hdevice, uint64_t size, uint64_t* dev_maddr) {
+  return vx_mem_alloc(hdevice, size, dev_maddr);
+}

driver/common/vx_utils.h

@@ -0,0 +1,18 @@
+#pragma once
+
+#include <cstdint>
+
+uint64_t aligned_size(uint64_t size, uint64_t alignment);
+
+bool is_aligned(uint64_t addr, uint64_t alignment);
+
+#define CACHE_BLOCK_SIZE    64
+// NOTE(hansung): This is changed to something more akin to be in a heap area
+// for a CPU userspace program, since that works better with Chipyard's default
+// memory mapping scheme (0x80000000 and above).  This gives us a pretty small
+// space though.
+#define ALLOC_BASE_ADDR     0xc0000000ul
+#define LOCAL_MEM_SIZE      0x40000000ul     // 1 GB 
+// #define ALLOC_BASE_ADDR     0x00000000
+// #define LOCAL_MEM_SIZE      4294967296     // 4 GB 
+#define DEVICE_MAX_ADDR     0xfffffffful

driver/fpga/Makefile

@@ -0,0 +1,75 @@
+OPAE_HOME ?= /tools/opae/1.4.0
+
+RTL_DIR=../../hw/rtl
+
+SCRIPT_DIR=../../hw/scripts
+
+OPAE_SYN_DIR=../../hw/syn/opae
+
+CXXFLAGS += -std=c++11 -Wall -Wextra -pedantic -Wfatal-errors
+
+CXXFLAGS += -I. -I../include -I../../hw -I$(OPAE_HOME)/include -I$(OPAE_SYN_DIR)
+
+LDFLAGS += -L$(OPAE_HOME)/lib -luuid -lopae-c
+
+#SCOPE=1
+
+# stack execution protection
+LDFLAGS +=-z noexecstack
+
+# data relocation and projection
+LDFLAGS +=-z relro -z now
+
+# stack buffer overrun detection
+CXXFLAGS +=-fstack-protector
+
+# Position independent code
+CXXFLAGS += -fPIC
+
+# Add external configuration
+CXXFLAGS += $(CONFIGS)
+
+# Dump perf stats
+CXXFLAGS += -DDUMP_PERF_STATS
+
+LDFLAGS += -shared
+
+PROJECT = libvortex.so
+
+SRCS = ../common/opae.cpp ../common/vx_utils.cpp
+
+# Debugigng
+ifdef DEBUG
+	CXXFLAGS += -g -O0
+else    
+	CXXFLAGS += -O2 -DNDEBUG
+endif
+
+# Enable scope analyzer
+ifdef SCOPE
+	CXXFLAGS += -DSCOPE	
+	SRCS += ../common/vx_scope.cpp
+	SCOPE_H = scope-defs.h 
+endif
+
+# Enable perf counters
+ifdef PERF
+	CXXFLAGS += -DPERF_ENABLE
+endif
+
+all: $(PROJECT)
+
+$(OPAE_SYN_DIR)/vortex_afu.h:
+	$(MAKE) -C $(OPAE_SYN_DIR) vortex_afu.h
+
+scope-defs.h: $(SCRIPT_DIR)/scope.json
+	$(SCRIPT_DIR)/scope.py $(CONFIGS) -cc scope-defs.h -vl $(RTL_DIR)/scope-defs.vh $(SCRIPT_DIR)/scope.json
+
+# generate scope data
+scope: scope-defs.h
+
+$(PROJECT): $(SRCS) $(OPAE_SYN_DIR)/vortex_afu.h $(SCOPE_H)
+	$(CXX) $(CXXFLAGS) -DUSE_FPGA $^ $(LDFLAGS) -o $(PROJECT)
+
+clean:
+	rm -rf $(PROJECT) *.o scope-defs.h

driver/include/vortex.h

@@ -0,0 +1,84 @@
+#ifndef __VX_DRIVER_H__
+#define __VX_DRIVER_H__
+
+#include <stddef.h>
+#include <stdint.h>
+#include <stdio.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef void* vx_device_h;
+
+typedef void* vx_buffer_h;
+
+// device caps ids
+#define VX_CAPS_VERSION           0x0 
+#define VX_CAPS_MAX_CORES         0x1
+#define VX_CAPS_MAX_WARPS         0x2
+#define VX_CAPS_MAX_THREADS       0x3
+#define VX_CAPS_CACHE_LINE_SIZE   0x4
+#define VX_CAPS_LOCAL_MEM_SIZE    0x5
+#define VX_CAPS_ALLOC_BASE_ADDR   0x6
+#define VX_CAPS_KERNEL_BASE_ADDR  0x7
+
+#define MAX_TIMEOUT               (60*60*1000)   // 1hr 
+
+// open the device and connect to it
+int vx_dev_open(vx_device_h* hdevice);
+
+// Close the device when all the operations are done
+int vx_dev_close(vx_device_h hdevice);
+
+// return device configurations
+int vx_dev_caps(vx_device_h hdevice, uint32_t caps_id, uint64_t *value);
+
+// Allocate shared buffer with device
+int vx_buf_alloc(vx_device_h hdevice, uint64_t size, vx_buffer_h* hbuffer);
+
+// release buffer
+int vx_buf_free(vx_buffer_h hbuffer);
+
+// Get host pointer address  
+void* vx_host_ptr(vx_buffer_h hbuffer);
+
+// allocate device memory and return address
+int vx_mem_alloc(vx_device_h hdevice, uint64_t size, uint64_t* dev_maddr);
+
+// release device memory
+int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr);
+
+// Copy bytes from buffer to device local memory
+int vx_copy_to_dev(vx_buffer_h hbuffer, uint64_t dev_maddr, uint64_t size, uint64_t src_offset);
+
+// Copy bytes from device local memory to buffer
+int vx_copy_from_dev(vx_buffer_h hbuffer, uint64_t dev_maddr, uint64_t size, uint64_t dst_offset);
+
+// Start device execution
+int vx_start(vx_device_h hdevice);
+
+// Wait for device ready with milliseconds timeout
+int vx_ready_wait(vx_device_h hdevice, uint64_t timeout);
+
+////////////////////////////// UTILITY FUNCIONS ///////////////////////////////
+
+// upload kernel bytes to device
+int vx_upload_kernel_bytes(vx_device_h device, const void* content, uint64_t size);
+
+// upload kernel file to device
+int vx_upload_kernel_file(vx_device_h device, const char* filename);
+
+// dump performance counters
+int vx_dump_perf(vx_device_h device, FILE* stream);
+
+//////////////////////////// DEPRECATED FUNCTIONS /////////////////////////////
+int vx_alloc_dev_mem(vx_device_h hdevice, uint64_t size, uint64_t* dev_maddr);
+int vx_alloc_shared_mem(vx_device_h hdevice, uint64_t size, vx_buffer_h* hbuffer);
+int vx_buf_release(vx_buffer_h hbuffer);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // __VX_DRIVER_H__

driver/rtlsim/.gitignore

@@ -0,0 +1,2 @@
+obj_dir
+*.so

driver/rtlsim/Makefile

@@ -0,0 +1,43 @@
+RTLSIM_DIR = ../../sim/rtlsim
+
+CXXFLAGS += -std=c++11 -Wall -Wextra -pedantic -Wfatal-errors
+
+CXXFLAGS += -I../include -I../common -I../../hw -I$(RTLSIM_DIR) -I$(RTLSIM_DIR)/../common
+
+# Position independent code
+CXXFLAGS += -fPIC
+
+# Add external configuration
+CXXFLAGS += $(CONFIGS)
+
+# Dump perf stats
+CXXFLAGS += -DDUMP_PERF_STATS
+
+LDFLAGS += -shared -pthread
+LDFLAGS += -L. -lrtlsim
+
+SRCS = vortex.cpp ../common/vx_utils.cpp
+
+# Debugigng
+ifdef DEBUG
+	CXXFLAGS += -g -O0
+else    
+	CXXFLAGS += -O2 -DNDEBUG
+endif
+
+# Enable perf counters
+ifdef PERF
+	CXXFLAGS += -DPERF_ENABLE
+endif
+
+PROJECT = libvortex.so
+
+all: $(PROJECT)
+	
+$(PROJECT): $(SRCS)
+	DESTDIR=../../driver/rtlsim $(MAKE) -C $(RTLSIM_DIR) ../../driver/rtlsim/librtlsim.so
+	$(CXX) $(CXXFLAGS) $(SRCS) $(LDFLAGS) -o $(PROJECT)
+
+clean:
+	DESTDIR=../../driver/rtlsim $(MAKE) -C $(RTLSIM_DIR) clean
+	rm -rf $(PROJECT) *.o

driver/rtlsim/vortex.cpp

@@ -0,0 +1,355 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <assert.h>
+#include <iostream>
+#include <future>
+#include <list>
+#include <chrono>
+
+#include <vortex.h>
+#include <vx_malloc.h>
+#include <vx_utils.h>
+#include <VX_config.h>
+#include <mem.h>
+#include <util.h>
+#include <processor.h>
+
+#define RAM_PAGE_SIZE 4096
+
+using namespace vortex;
+
+///////////////////////////////////////////////////////////////////////////////
+
+class vx_device;
+class vx_buffer {
+public:
+    vx_buffer(uint64_t size, vx_device* device) 
+        : size_(size)
+        , device_(device) {
+        auto aligned_asize = aligned_size(size, CACHE_BLOCK_SIZE);
+        data_ = malloc(aligned_asize);
+    }
+
+    ~vx_buffer() {
+        if (data_) {
+            free(data_);
+        }
+    }
+
+    void* data() const {
+        return data_;
+    }
+
+    uint64_t size() const {
+        return size_;
+    }
+
+    vx_device* device() const {
+        return device_;
+    }
+
+private:
+    uint64_t size_;
+    vx_device* device_;
+    void* data_;
+};
+
+///////////////////////////////////////////////////////////////////////////////
+
+class vx_device {    
+public:
+    vx_device() 
+        : ram_(RAM_PAGE_SIZE)
+        , mem_allocator_(
+            ALLOC_BASE_ADDR,
+            ALLOC_BASE_ADDR + LOCAL_MEM_SIZE,
+            RAM_PAGE_SIZE,
+            CACHE_BLOCK_SIZE) 
+    {
+        processor_.attach_ram(&ram_);
+    }
+
+    ~vx_device() {    
+        if (future_.valid()) {
+            future_.wait();
+        }
+    }
+
+    int alloc_local_mem(uint64_t size, uint64_t* dev_maddr) {
+        return mem_allocator_.allocate(size, dev_maddr);
+    }
+
+    int free_local_mem(uint64_t dev_maddr) {
+        return mem_allocator_.release(dev_maddr);
+    }
+
+    int upload(const void* src, uint64_t dest_addr, uint64_t size, uint64_t src_offset) {
+        uint64_t asize = aligned_size(size, CACHE_BLOCK_SIZE);
+        if (dest_addr + asize > DEVICE_MAX_ADDR)
+            return -1;
+
+        /*printf("VXDRV: upload %ld bytes from 0x%lx:", size, uintptr_t((uint8_t*)src + src_offset));
+        for (int i = 0;  i < (asize / CACHE_BLOCK_SIZE); ++i) {
+            printf("\n0x%08lx=", dest_addr + i * CACHE_BLOCK_SIZE);
+            for (int j = 0;  j < CACHE_BLOCK_SIZE; ++j) {
+                printf("%02x", *((uint8_t*)src + src_offset + i * CACHE_BLOCK_SIZE + CACHE_BLOCK_SIZE - 1 - j));
+            }
+        }
+        printf("\n");*/
+        
+        ram_.write((const uint8_t*)src + src_offset, dest_addr, asize);
+        return 0;
+    }
+
+    int download(void* dest, uint64_t src_addr, uint64_t size, uint64_t dest_offset) {
+        uint64_t asize = aligned_size(size, CACHE_BLOCK_SIZE);
+        if (src_addr + asize > DEVICE_MAX_ADDR)
+            return -1;
+
+        ram_.read((uint8_t*)dest + dest_offset, src_addr, asize);
+        
+        /*printf("VXDRV: download %ld bytes to 0x%lx:", size, uintptr_t((uint8_t*)dest + dest_offset));
+        for (int i = 0;  i < (asize / CACHE_BLOCK_SIZE); ++i) {
+            printf("\n0x%08lx=", src_addr + i * CACHE_BLOCK_SIZE);
+            for (int j = 0;  j < CACHE_BLOCK_SIZE; ++j) {
+                printf("%02x", *((uint8_t*)dest + dest_offset + i * CACHE_BLOCK_SIZE + CACHE_BLOCK_SIZE - 1 - j));
+            }
+        }
+        printf("\n");*/
+        
+        return 0;
+    }
+
+    int start() {   
+        // ensure prior run completed
+        if (future_.valid()) {
+            future_.wait();
+        }
+        // start new run
+        future_ = std::async(std::launch::async, [&]{
+            processor_.run();
+        });
+        return 0;
+    }
+
+    int wait(uint64_t timeout) {
+        if (!future_.valid())
+            return 0;
+        uint64_t timeout_sec = timeout / 1000;
+        std::chrono::seconds wait_time(1);
+        for (;;) {
+            // wait for 1 sec and check status
+            auto status = future_.wait_for(wait_time);
+            if (status == std::future_status::ready 
+             || 0 == timeout_sec--)
+                break;
+        }
+        return 0;
+    }
+
+private:
+
+    RAM ram_;
+    Processor processor_;
+    MemoryAllocator mem_allocator_;     
+    std::future<void> future_;
+};
+
+///////////////////////////////////////////////////////////////////////////////
+
+#ifdef DUMP_PERF_STATS
+class AutoPerfDump {
+private:
+    std::list<vx_device_h> devices_;
+
+public:
+    AutoPerfDump() {} 
+
+    ~AutoPerfDump() {
+        for (auto device : devices_) {
+            vx_dump_perf(device, stdout);
+        }
+    }
+
+    void add_device(vx_device_h device) {
+        devices_.push_back(device);
+    }
+
+    void remove_device(vx_device_h device) {
+        devices_.remove(device);
+    }    
+};
+
+AutoPerfDump gAutoPerfDump;
+#endif
+
+///////////////////////////////////////////////////////////////////////////////
+
+extern int vx_dev_caps(vx_device_h hdevice, uint32_t caps_id, uint64_t *value) {
+   if (nullptr == hdevice)
+        return  -1;
+
+    switch (caps_id) {
+    case VX_CAPS_VERSION:
+        *value = IMPLEMENTATION_ID;
+        break;
+    case VX_CAPS_MAX_CORES:
+        *value = NUM_CORES * NUM_CLUSTERS;        
+        break;
+    case VX_CAPS_MAX_WARPS:
+        *value = NUM_WARPS;
+        break;
+    case VX_CAPS_MAX_THREADS:
+        *value = NUM_THREADS;
+        break;
+    case VX_CAPS_CACHE_LINE_SIZE:
+        *value = CACHE_BLOCK_SIZE;
+        break;
+    case VX_CAPS_LOCAL_MEM_SIZE:
+        *value = LOCAL_MEM_SIZE;
+        break;
+    case VX_CAPS_ALLOC_BASE_ADDR:
+        *value = ALLOC_BASE_ADDR;
+        break;
+    case VX_CAPS_KERNEL_BASE_ADDR:
+        *value = STARTUP_ADDR;
+        break;
+    default:
+        std::cout << "invalid caps id: " << caps_id << std::endl;
+        std::abort();
+        return -1;
+    }
+
+    return 0;
+}
+
+extern int vx_dev_open(vx_device_h* hdevice) {
+    if (nullptr == hdevice)
+        return  -1;
+
+    *hdevice = new vx_device();
+
+#ifdef DUMP_PERF_STATS
+    gAutoPerfDump.add_device(*hdevice);
+#endif
+
+    return 0;
+}
+
+extern int vx_dev_close(vx_device_h hdevice) {
+    if (nullptr == hdevice)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+    
+#ifdef DUMP_PERF_STATS
+    gAutoPerfDump.remove_device(hdevice);
+    vx_dump_perf(hdevice, stdout);
+#endif
+
+    delete device;
+
+    return 0;
+}
+
+extern int vx_mem_alloc(vx_device_h hdevice, uint64_t size, uint64_t* dev_maddr) {
+    if (nullptr == hdevice 
+     || nullptr == dev_maddr
+     || 0 >= size)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+    return device->alloc_local_mem(size, dev_maddr);
+}
+
+extern int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr) {
+    if (nullptr == hdevice)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+    return device->free_local_mem(dev_maddr);
+}
+
+extern int vx_buf_alloc(vx_device_h hdevice, uint64_t size, vx_buffer_h* hbuffer) {
+    if (nullptr == hdevice 
+     || 0 >= size
+     || nullptr == hbuffer)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+
+    auto buffer = new vx_buffer(size, device);
+    if (nullptr == buffer->data()) {
+        delete buffer;
+        return -1;
+    }
+
+    *hbuffer = buffer;
+
+    return 0;
+}
+
+extern void* vx_host_ptr(vx_buffer_h hbuffer) {
+    if (nullptr == hbuffer)
+        return nullptr;
+
+    vx_buffer* buffer = ((vx_buffer*)hbuffer);
+
+    return buffer->data();
+}
+
+extern int vx_buf_free(vx_buffer_h hbuffer) {
+    if (nullptr == hbuffer)
+        return -1;
+
+    vx_buffer* buffer = ((vx_buffer*)hbuffer);
+
+    delete buffer;
+
+    return 0;
+}
+
+extern int vx_copy_to_dev(vx_buffer_h hbuffer, uint64_t dev_maddr, uint64_t size, uint64_t src_offset) {
+    if (nullptr == hbuffer 
+     || 0 >= size)
+        return -1;
+
+    auto buffer = (vx_buffer*)hbuffer;
+
+    if (size + src_offset > buffer->size())
+        return -1;
+
+    return buffer->device()->upload(buffer->data(), dev_maddr, size, src_offset);
+}
+
+extern int vx_copy_from_dev(vx_buffer_h hbuffer, uint64_t dev_maddr, uint64_t size, uint64_t dest_offset) {
+     if (nullptr == hbuffer 
+      || 0 >= size)
+        return -1;
+
+    auto buffer = (vx_buffer*)hbuffer;
+
+    if (size + dest_offset > buffer->size())
+        return -1;    
+
+    return buffer->device()->download(buffer->data(), dev_maddr, size, dest_offset);
+}
+
+extern int vx_start(vx_device_h hdevice) {
+    if (nullptr == hdevice)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+
+    return device->start();
+}
+
+extern int vx_ready_wait(vx_device_h hdevice, uint64_t timeout) {
+    if (nullptr == hdevice)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+
+    return device->wait(timeout);
+}

driver/simx/Makefile

@@ -0,0 +1,32 @@
+SIMX_DIR = ../../sim/simx
+
+CXXFLAGS += -std=c++11 -Wall -Wextra -Wfatal-errors
+
+CXXFLAGS += -fPIC -Wno-maybe-uninitialized
+CXXFLAGS += -I../include -I../common -I../../hw -I$(SIMX_DIR) -I$(SIMX_DIR)/../common
+CXXFLAGS += $(CONFIGS)
+CXXFLAGS += -DDUMP_PERF_STATS
+
+LDFLAGS += -shared -pthread
+LDFLAGS += -L. -lsimx
+
+SRCS = vortex.cpp ../common/vx_utils.cpp
+
+# Debugigng
+ifdef DEBUG
+	CXXFLAGS += -g -O0
+else    
+	CXXFLAGS += -O2 -DNDEBUG
+endif
+
+PROJECT = libvortex.so
+
+all: $(PROJECT)
+
+$(PROJECT): $(SRCS)
+	DESTDIR=../../driver/simx $(MAKE) -C $(SIMX_DIR) ../../driver/simx/libsimx.so
+	$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -o $@
+
+clean:
+	DESTDIR=../../driver/simx $(MAKE) -C $(SIMX_DIR) clean
+	rm -rf libsimx.so $(PROJECT) *.o

driver/simx/vortex.cpp

@@ -0,0 +1,357 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <assert.h>
+#include <iostream>
+#include <future>
+#include <chrono>
+
+#include <vortex.h>
+#include <vx_utils.h>
+#include <vx_malloc.h>
+
+#include <VX_config.h>
+
+#include <util.h>
+
+#include <processor.h>
+#include <archdef.h>
+#include <mem.h>
+#include <constants.h>
+
+
+using namespace vortex;
+
+///////////////////////////////////////////////////////////////////////////////
+
+class vx_device;
+
+class vx_buffer {
+public:
+    vx_buffer(uint64_t size, vx_device* device) 
+        : size_(size)
+        , device_(device) {
+        uint64_t aligned_asize = aligned_size(size, CACHE_BLOCK_SIZE);
+        data_ = malloc(aligned_asize);
+    }
+
+    ~vx_buffer() {
+        if (data_) {
+            free(data_);
+        }
+    }
+
+    void* data() const {
+        return data_;
+    }
+
+    uint64_t size() const {
+        return size_;
+    }
+
+    vx_device* device() const {
+        return device_;
+    }
+
+private:
+    uint64_t size_;
+    vx_device* device_;
+    void* data_;
+};
+
+///////////////////////////////////////////////////////////////////////////////
+
+class vx_device {    
+public:
+    vx_device() 
+        : arch_(NUM_CORES * NUM_CLUSTERS, NUM_WARPS, NUM_THREADS)
+        , ram_(RAM_PAGE_SIZE)
+        , processor_(arch_)
+        , mem_allocator_(
+            ALLOC_BASE_ADDR,
+            ALLOC_BASE_ADDR + LOCAL_MEM_SIZE,
+            RAM_PAGE_SIZE,
+            CACHE_BLOCK_SIZE) 
+    {
+        // attach memory module
+        processor_.attach_ram(&ram_);
+    }
+
+    ~vx_device() {
+        if (future_.valid()) {
+            future_.wait();
+        }
+    }    
+
+    int alloc_local_mem(uint64_t size, uint64_t* dev_maddr) {
+        return mem_allocator_.allocate(size, dev_maddr);
+    }
+
+    int free_local_mem(uint64_t dev_maddr) {
+        return mem_allocator_.release(dev_maddr);
+    }
+
+    int upload(const void* src, uint64_t dest_addr, uint64_t size, uint64_t src_offset) {
+        uint64_t asize = aligned_size(size, CACHE_BLOCK_SIZE);
+        if (dest_addr + asize > DEVICE_MAX_ADDR)
+            return -1;
+
+        ram_.write((const uint8_t*)src + src_offset, dest_addr, asize);
+        
+        /*printf("VXDRV: upload %d bytes to 0x%x\n", size, dest_addr);
+        for (int i = 0; i < size; i += 4) {
+            printf("mem-write: 0x%x <- 0x%x\n", dest_addr + i, *(uint32_t*)((uint8_t*)src + src_offset + i));
+        }*/
+        
+        return 0;
+    }
+
+    int download(void* dest, uint64_t src_addr, uint64_t size, uint64_t dest_offset) {
+        uint64_t asize = aligned_size(size, CACHE_BLOCK_SIZE);
+        if (src_addr + asize > DEVICE_MAX_ADDR)
+            return -1;
+
+        ram_.read((uint8_t*)dest + dest_offset, src_addr, asize);
+        
+        /*printf("VXDRV: download %d bytes from 0x%x\n", size, src_addr);
+        for (int i = 0; i < size; i += 4) {
+            printf("mem-read: 0x%x -> 0x%x\n", src_addr + i, *(uint32_t*)((uint8_t*)dest + dest_offset + i));
+        }*/
+        
+        return 0;
+    }
+
+    int start() {  
+        // ensure prior run completed
+        if (future_.valid()) {
+            future_.wait();
+        }
+        
+        // start new run
+        future_ = std::async(std::launch::async, [&]{
+            processor_.run();
+        });
+        
+        return 0;
+    }
+
+    int wait(uint64_t timeout) {
+        if (!future_.valid())
+            return 0;
+        uint64_t timeout_sec = timeout / 1000;
+        std::chrono::seconds wait_time(1);
+        for (;;) {
+            // wait for 1 sec and check status
+            auto status = future_.wait_for(wait_time);
+            if (status == std::future_status::ready 
+             || 0 == timeout_sec--)
+                break;
+        }
+        return 0;
+    } 
+
+private:
+    ArchDef arch_;
+    RAM ram_;
+    Processor processor_;
+    MemoryAllocator mem_allocator_;       
+    std::future<void> future_;
+};
+
+///////////////////////////////////////////////////////////////////////////////
+
+#ifdef DUMP_PERF_STATS
+class AutoPerfDump {
+private:
+    std::list<vx_device_h> devices_;
+
+public:
+    AutoPerfDump() {} 
+
+    ~AutoPerfDump() {
+        for (auto device : devices_) {
+            vx_dump_perf(device, stdout);
+        }
+    }
+
+    void add_device(vx_device_h device) {
+        devices_.push_back(device);
+    }
+
+    void remove_device(vx_device_h device) {
+        devices_.remove(device);
+    }    
+};
+
+AutoPerfDump gAutoPerfDump;
+#endif
+
+///////////////////////////////////////////////////////////////////////////////
+
+extern int vx_dev_open(vx_device_h* hdevice) {
+    if (nullptr == hdevice)
+        return  -1;
+
+    *hdevice = new vx_device();    
+
+#ifdef DUMP_PERF_STATS
+    gAutoPerfDump.add_device(*hdevice);
+#endif
+
+    return 0;
+}
+
+extern int vx_dev_close(vx_device_h hdevice) {
+    if (nullptr == hdevice)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+
+#ifdef DUMP_PERF_STATS
+    gAutoPerfDump.remove_device(hdevice);
+    vx_dump_perf(hdevice, stdout);
+#endif
+
+    delete device;
+
+    return 0;
+}
+
+extern int vx_dev_caps(vx_device_h hdevice, uint32_t caps_id, uint64_t *value) {
+    if (nullptr == hdevice)
+        return  -1;
+
+    switch (caps_id) {
+    case VX_CAPS_VERSION:
+        *value = IMPLEMENTATION_ID;
+        break;
+    case VX_CAPS_MAX_CORES:
+        *value = NUM_CORES * NUM_CLUSTERS;        
+        break;
+    case VX_CAPS_MAX_WARPS:
+        *value = NUM_WARPS;
+        break;
+    case VX_CAPS_MAX_THREADS:
+        *value = NUM_THREADS;
+        break;
+    case VX_CAPS_CACHE_LINE_SIZE:
+        *value = CACHE_BLOCK_SIZE;
+        break;
+    case VX_CAPS_LOCAL_MEM_SIZE:
+        *value = LOCAL_MEM_SIZE;
+        break;
+    case VX_CAPS_ALLOC_BASE_ADDR:
+        *value = ALLOC_BASE_ADDR;
+        break;
+    case VX_CAPS_KERNEL_BASE_ADDR:
+        *value = STARTUP_ADDR;
+        break;
+    default:
+        std::cout << "invalid caps id: " << caps_id << std::endl;
+        std::abort();
+        return -1;
+    }
+
+    return 0;
+}
+
+extern int vx_mem_alloc(vx_device_h hdevice, uint64_t size, uint64_t* dev_maddr) {
+    if (nullptr == hdevice 
+     || nullptr == dev_maddr
+     || 0 >= size)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+    return device->alloc_local_mem(size, dev_maddr);
+}
+
+extern int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr) {
+    if (nullptr == hdevice)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+    return device->free_local_mem(dev_maddr);
+}
+
+extern int vx_buf_alloc(vx_device_h hdevice, uint64_t size, vx_buffer_h* hbuffer) {
+    if (nullptr == hdevice 
+     || 0 >= size
+     || nullptr == hbuffer)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+
+    auto buffer = new vx_buffer(size, device);
+    if (nullptr == buffer->data()) {
+        delete buffer;
+        return -1;
+    }
+
+    *hbuffer = buffer;
+
+    return 0;
+}
+
+extern void* vx_host_ptr(vx_buffer_h hbuffer) {
+    if (nullptr == hbuffer)
+        return nullptr;
+
+    vx_buffer* buffer = ((vx_buffer*)hbuffer);
+
+    return buffer->data();
+}
+
+extern int vx_buf_free(vx_buffer_h hbuffer) {
+    if (nullptr == hbuffer)
+        return -1;
+
+    vx_buffer* buffer = ((vx_buffer*)hbuffer);
+
+    delete buffer;
+
+    return 0;
+}
+
+extern int vx_copy_to_dev(vx_buffer_h hbuffer, uint64_t dev_maddr, uint64_t size, uint64_t src_offset) {
+    if (nullptr == hbuffer 
+     || 0 >= size)
+        return -1;
+
+    auto buffer = (vx_buffer*)hbuffer;
+
+    if (size + src_offset > buffer->size())
+        return -1;
+
+    return buffer->device()->upload(buffer->data(), dev_maddr, size, src_offset);
+}
+
+extern int vx_copy_from_dev(vx_buffer_h hbuffer, uint64_t dev_maddr, uint64_t size, uint64_t dest_offset) {
+     if (nullptr == hbuffer 
+      || 0 >= size)
+        return -1;
+
+    auto buffer = (vx_buffer*)hbuffer;
+
+    if (size + dest_offset > buffer->size())
+        return -1;    
+
+    return buffer->device()->download(buffer->data(), dev_maddr, size, dest_offset);
+}
+
+extern int vx_start(vx_device_h hdevice) {
+    if (nullptr == hdevice)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+
+    return device->start();
+}
+
+extern int vx_ready_wait(vx_device_h hdevice, uint64_t timeout) {
+    if (nullptr == hdevice)
+        return -1;
+
+    vx_device *device = ((vx_device*)hdevice);
+
+    return device->wait(timeout);
+}

driver/stub/Makefile

@@ -1,17 +1,12 @@
-XLEN ?= 32
-DESTDIR ?= $(CURDIR)
-SIM_DIR = $(abspath ../../sim)
-HW_DIR = $(abspath ../../hw)
-
 CXXFLAGS += -std=c++11 -O2 -Wall -Wextra -pedantic -Wfatal-errors
 
-CXXFLAGS += -I../include -I../common -I$(HW_DIR) -I$(SIM_DIR)/common
+CXXFLAGS += -I../include -I../../runtime -I../../hw
 
 CXXFLAGS += -fPIC
 
 LDFLAGS += -shared -pthread
 
-SRCS = vortex.cpp ../common/utils.cpp
+SRCS = vortex.cpp ../common/vx_utils.cpp
 
 PROJECT = libvortex.so
 

driver/stub/vortex.cpp

@@ -0,0 +1,49 @@
+#include <vortex.h>
+
+extern int vx_dev_open(vx_device_h* /*hdevice*/) {
+    return -1;
+}
+
+extern int vx_dev_close(vx_device_h /*hdevice*/) {
+    return -1;
+}
+
+extern int vx_dev_caps(vx_device_h /*hdevice*/, uint32_t /*caps_id*/, uint64_t* /*value*/) {
+    return -1;
+}
+
+extern int vx_mem_alloc(vx_device_h /*hdevice*/, uint64_t /*size*/, uint64_t* /*dev_maddr*/) {
+    return -1;
+}
+
+int vx_mem_free(vx_device_h /*hdevice*/, uint64_t /*dev_maddr*/) {
+    return -1;
+}
+
+extern int vx_buf_alloc(vx_device_h /*hdevice*/, uint64_t /*size*/, vx_buffer_h* /*hbuffer*/) {
+    return -1;
+}
+
+extern void* vx_host_ptr(vx_buffer_h /*hbuffer*/) {
+    return nullptr;
+}
+
+extern int vx_buf_free(vx_buffer_h /*hbuffer*/) {
+    return -1;
+}
+
+extern int vx_copy_to_dev(vx_buffer_h /*hbuffer*/, uint64_t /*dev_maddr*/, uint64_t /*size*/, uint64_t /*src_offset*/) {
+    return -1;
+}
+
+extern int vx_copy_from_dev(vx_buffer_h /*hbuffer*/, uint64_t /*dev_maddr*/, uint64_t /*size*/, uint64_t /*dest_offset*/) {
+     return -1;
+}
+
+extern int vx_start(vx_device_h /*hdevice*/) {
+    return -1;
+}
+
+extern int vx_ready_wait(vx_device_h /*hdevice*/, uint64_t /*timeout*/) {
+    return -1;
+}

driver/vlsim/Makefile

@@ -0,0 +1,60 @@
+VLSIM_DIR = ../../sim/vlsim
+
+RTL_DIR=../../hw/rtl
+
+SCRIPT_DIR=../../hw/scripts
+
+CXXFLAGS += -std=c++11 -Wall -Wextra -pedantic -Wfatal-errors
+
+CXXFLAGS += -I. -I../include -I../../hw -I$(VLSIM_DIR)
+
+# Position independent code
+CXXFLAGS += -fPIC
+
+# Add external configuration
+CXXFLAGS += $(CONFIGS)
+
+# Dump perf stats
+CXXFLAGS += -DDUMP_PERF_STATS
+
+LDFLAGS += -shared -pthread
+LDFLAGS += -L. -lopae-c-vlsim
+
+SRCS = ../common/opae.cpp ../common/vx_utils.cpp
+
+# Debugigng
+ifdef DEBUG
+	CXXFLAGS += -g -O0
+else    
+	CXXFLAGS += -O2 -DNDEBUG
+endif
+
+# Enable scope analyzer
+ifdef SCOPE
+	CXXFLAGS += -DSCOPE	
+	SRCS += ../common/vx_scope.cpp
+	SCOPE_H = scope-defs.h 
+endif
+
+# Enable perf counters
+ifdef PERF
+	CXXFLAGS += -DPERF_ENABLE
+endif
+
+PROJECT = libvortex.so
+
+all: $(PROJECT)
+
+scope-defs.h: $(SCRIPT_DIR)/scope.json
+	$(SCRIPT_DIR)/scope.py $(CONFIGS) -cc scope-defs.h -vl $(RTL_DIR)/scope-defs.vh $(SCRIPT_DIR)/scope.json
+
+# generate scope data
+scope: scope-defs.h
+
+$(PROJECT): $(SRCS) $(SCOPE_H)
+	DESTDIR=../../driver/vlsim $(MAKE) -C $(VLSIM_DIR) ../../driver/vlsim/libopae-c-vlsim.so
+	$(CXX) $(CXXFLAGS) -DUSE_VLSIM $(SRCS) $(LDFLAGS) -o $(PROJECT)
+
+clean:
+	DESTDIR=../../driver/vlsim $(MAKE) -C $(VLSIM_DIR) clean
+	rm -rf libopae-c-vlsim.so $(PROJECT) *.o scope-defs.h

env.my-pocl-riscv32.sh

@@ -0,0 +1,7 @@
+export VORTEX_ENV="my-pocl-riscv32"
+export LLVM_PREFIX=/scratch/hansung/build/llvm-riscv32-unknown-linux-gnu-10.0.1
+export POCL_CC_PATH=/scratch/hansung/build/pocl-riscv32/compiler
+export POCL_RT_PATH=/scratch/hansung/build/pocl-riscv32/runtime
+export VERILATOR_ROOT=/scratch/hansung/build/vortex-toolchain-prebuilt/verilator
+export RISCV_TOOLCHAIN_PATH=/scratch/hansung/build/vortex-toolchain-prebuilt
+export PATH="/scratch/hansung/build/vortex-toolchain-prebuilt/verilator/bin:$PATH"

env.my-pocl.sh

@@ -0,0 +1,7 @@
+export VORTEX_ENV="my-pocl"
+export LLVM_PREFIX=/scratch/hansung/build/vortex-toolchain-prebuilt/llvm-riscv/
+export POCL_CC_PATH=/scratch/hansung/build/pocl-vortex/compiler
+export POCL_RT_PATH=/scratch/hansung/build/pocl-vortex/runtime
+export VERILATOR_ROOT=/scratch/hansung/build/vortex-toolchain-prebuilt/verilator
+export RISCV_TOOLCHAIN_PATH=/scratch/hansung/build/vortex-toolchain-prebuilt
+export PATH="/scratch/hansung/build/vortex-toolchain-prebuilt/verilator/bin:$PATH"

env.vortex-prebuilt.sh

@@ -0,0 +1,17 @@
+if [ -n "$VORTEX_ENV" ]
+then
+    echo "VORTEX_ENV already set. Exiting."
+    return
+fi
+
+# PREBUILT_DIR=/scratch/hansung/build/vortex-toolchain-prebuilt-d2ba5df-230831
+PREBUILT_DIR=/scratch/hansung/build/vortex-toolchain-prebuilt-230831
+
+export VORTEX_ENV="vortex-prebuilt"
+export LLVM_PREFIX=$PREBUILT_DIR/llvm-riscv/
+export POCL_CC_PATH=$PREBUILT_DIR/pocl/compiler
+export POCL_RT_PATH=$PREBUILT_DIR/pocl/runtime
+export VERILATOR_ROOT=$PREBUILT_DIR/verilator
+export RISCV_TOOLCHAIN_PATH=$PREBUILT_DIR/
+export PATH="$BUILDDIR/vortex-toolchain-prebuilt-d2ba5df-230831/verilator/bin:$PATH"
+export PS1="($VORTEX_ENV) $PS1"

env.vortex-prebuilt2.sh

@@ -0,0 +1,7 @@
+export VORTEX_ENV="vortex-prebuilt2"
+export LLVM_PREFIX=/scratch/hansung/build/vortex-toolchain-prebuilt/llvm-riscv2
+export POCL_CC_PATH=/scratch/hansung/build/vortex-toolchain-prebuilt/pocl2/compiler
+export POCL_RT_PATH=/scratch/hansung/build/vortex-toolchain-prebuilt/pocl2/runtime
+export VERILATOR_ROOT=/scratch/hansung/build/vortex-toolchain-prebuilt/verilator
+export RISCV_TOOLCHAIN_PATH=/scratch/hansung/build/vortex-toolchain-prebuilt
+export PATH="/scratch/hansung/build/vortex-toolchain-prebuilt/verilator/bin:$PATH"

hw/.gitignore

@@ -1,2 +1 @@
-VX_config.h
-VX_types.h
+obj_dir/*

hw/Makefile

@@ -1,17 +1,12 @@
 RTL_DIR=./rtl
 SCRIPT_DIR=./scripts
 
-all: config
-
-config: VX_config.h VX_types.h
+all: VX_config.h
 
 VX_config.h: $(RTL_DIR)/VX_config.vh	 
 	$(SCRIPT_DIR)/gen_config.py -i $(RTL_DIR)/VX_config.vh -o VX_config.h
 
-VX_types.h: $(RTL_DIR)/VX_types.vh	 
-	$(SCRIPT_DIR)/gen_config.py -i $(RTL_DIR)/VX_types.vh -o VX_types.h
-
 clean:
-	rm -f VX_config.h VX_types.h
+	rm -f VX_config.h
 
-.PHONY: VX_config.h VX_types.h
+.PHONY: VX_config.h

hw/VX_config.h

@@ -1,684 +0,0 @@
-// auto-generated by gen_config.py. DO NOT EDIT
-// Generated at 2024-05-07 13:55:58.398687
-
-// Translated from ./rtl/VX_config.vh:
-
-// Copyright © 2019-2023
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// 
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-#ifndef VX_CONFIG_VH
-#define VX_CONFIG_VH
-
-#ifndef MIN
-#define MIN(x, y)   (((x) < (y)) ? (x) : (y))
-#endif
-
-#ifndef MAX
-#define MAX(x, y)   (((x) > (y)) ? (x) : (y))
-#endif
-
-#ifndef CLAMP
-#define CLAMP(x, lo, hi)   (((x) > (hi)) ? (hi) : (((x) < (lo)) ? (lo) : (x)))
-#endif
-
-#ifndef UP
-#define UP(x)   (((x) != 0) ? (x) : 1)
-#endif
-
-///////////////////////////////////////////////////////////////////////////////
-
-#ifndef EXT_M_DISABLE
-#define EXT_M_ENABLE
-#endif
-
-#ifndef EXT_F_DISABLE
-#define EXT_F_ENABLE
-#endif
-
-#ifndef XLEN_32
-#ifndef XLEN_64
-#define XLEN_32
-#endif
-#endif
-
-#ifdef XLEN_64
-#define XLEN 64
-#endif
-
-#ifdef XLEN_32
-#define XLEN 32
-#endif
-
-#ifdef EXT_D_ENABLE
-#define FLEN_64
-#else
-#define FLEN_32
-#endif
-
-#ifdef FLEN_64
-#define FLEN 64
-#endif
-
-#ifdef FLEN_32
-#define FLEN 32
-#endif
-
-#ifdef XLEN_64
-#ifdef FLEN_32
-    #define FPU_RV64F
-#endif
-#endif
-
-#ifndef NUM_CLUSTERS
-#define NUM_CLUSTERS 1
-#endif
-
-#ifndef NUM_CORES
-#define NUM_CORES 1
-#endif
-
-#ifndef NUM_WARPS
-#define NUM_WARPS 4
-#endif
-
-#ifndef NUM_THREADS
-#define NUM_THREADS 4
-#endif
-
-#ifndef NUM_BARRIERS
-#define NUM_BARRIERS 8
-#endif
-
-#ifndef SOCKET_SIZE
-#define SOCKET_SIZE MIN(4, NUM_CORES)
-#endif
-#define NUM_SOCKETS UP(NUM_CORES / SOCKET_SIZE)
-
-#ifdef L2_ENABLE
-    #define L2_ENABLED   1
-#else
-    #define L2_ENABLED   0
-#endif
-
-#ifdef L3_ENABLE
-    #define L3_ENABLED   1
-#else
-    #define L3_ENABLED   0
-#endif
-
-#ifdef L1_DISABLE
-    #define ICACHE_DISABLE
-    #define DCACHE_DISABLE
-#endif
-
-#ifndef MEM_BLOCK_SIZE
-#define MEM_BLOCK_SIZE 64
-#endif
-
-#ifndef MEM_ADDR_WIDTH
-#ifdef XLEN_64
-#define MEM_ADDR_WIDTH 48
-#else
-#define MEM_ADDR_WIDTH 32
-#endif
-#endif
-
-#ifndef L1_LINE_SIZE
-#ifdef L1_DISABLE
-#define L1_LINE_SIZE ((L2_ENABLED || L3_ENABLED) ? 4 : MEM_BLOCK_SIZE)
-#else
-#define L1_LINE_SIZE ((L2_ENABLED || L3_ENABLED) ? 16 : MEM_BLOCK_SIZE)
-#endif
-#endif
-
-#ifdef L2_ENABLE
-#define L2_LINE_SIZE MEM_BLOCK_SIZE
-#else
-#define L2_LINE_SIZE L1_LINE_SIZE
-#endif
-
-#ifdef L3_ENABLE
-#define L3_LINE_SIZE MEM_BLOCK_SIZE
-#else
-#define L3_LINE_SIZE L2_LINE_SIZE
-#endif
-
-#ifdef XLEN_64
-
-#ifndef STARTUP_ADDR
-#define STARTUP_ADDR 0x180000000
-#endif
-
-#ifndef STACK_BASE_ADDR
-#define STACK_BASE_ADDR 0x1FF000000
-#endif
-
-#else
-
-#ifndef STARTUP_ADDR
-#define STARTUP_ADDR 0x80000000
-#endif
-
-#ifndef STACK_BASE_ADDR
-#define STACK_BASE_ADDR 0xFF000000
-#endif
-
-#endif
-
-#ifndef SMEM_BASE_ADDR
-#define SMEM_BASE_ADDR STACK_BASE_ADDR
-#endif
-
-#ifndef SMEM_LOG_SIZE
-#define SMEM_LOG_SIZE   17
-#endif
-
-#ifndef IO_BASE_ADDR
-#define IO_BASE_ADDR (SMEM_BASE_ADDR + (1 << SMEM_LOG_SIZE))
-#endif
-
-#ifndef IO_COUT_ADDR
-#define IO_COUT_ADDR IO_BASE_ADDR
-#endif
-#define IO_COUT_SIZE MEM_BLOCK_SIZE
-
-#ifndef IO_CSR_ADDR
-#define IO_CSR_ADDR (IO_COUT_ADDR + IO_COUT_SIZE)
-#endif
-#define IO_CSR_SIZE (4 * 64 * NUM_CORES * NUM_CLUSTERS)
-
-#ifndef STACK_LOG2_SIZE
-#define STACK_LOG2_SIZE 13
-#endif
-#define STACK_SIZE (1 << STACK_LOG2_SIZE)
-
-#define RESET_DELAY 8
-
-#ifndef STALL_TIMEOUT
-#define STALL_TIMEOUT (100000 * (1 ** (L2_ENABLED + L3_ENABLED)))
-#endif
-
-#ifndef SV_DPI
-#define DPI_DISABLE
-#endif
-
-#ifndef FPU_FPNEW
-#ifndef FPU_DSP
-#ifndef FPU_DPI
-#ifndef SYNTHESIS
-#ifndef DPI_DISABLE
-#define FPU_DPI
-#else
-#define FPU_DSP
-#endif
-#else
-#define FPU_DSP
-#endif
-#endif
-#endif
-#endif
-
-#ifndef SYNTHESIS
-#ifndef DPI_DISABLE
-#define IMUL_DPI
-#define IDIV_DPI
-#endif
-#endif
-
-#ifndef DEBUG_LEVEL
-#define DEBUG_LEVEL 3
-#endif
-
-// Pipeline Configuration /////////////////////////////////////////////////////
-
-// Issue width
-#ifndef ISSUE_WIDTH
-#define ISSUE_WIDTH     NUM_WARPS
-#endif
-
-// Number of ALU units
-#ifndef NUM_ALU_LANES
-#define NUM_ALU_LANES   NUM_THREADS
-#endif
-#ifndef NUM_ALU_BLOCKS
-#define NUM_ALU_BLOCKS  4
-#endif
-
-// Number of FPU units
-#ifndef NUM_FPU_LANES
-#define NUM_FPU_LANES   NUM_THREADS
-#endif
-#ifndef NUM_FPU_BLOCKS
-#define NUM_FPU_BLOCKS  2
-#endif
-
-// Number of LSU units
-#ifndef NUM_LSU_LANES
-#define NUM_LSU_LANES   NUM_THREADS
-#endif
-
-// Number of SFU units
-#ifndef NUM_SFU_LANES
-#define NUM_SFU_LANES   MIN(NUM_THREADS, 4)
-#endif
-
-// Size of Instruction Buffer
-#ifndef IBUF_SIZE
-#define IBUF_SIZE   (4 * ISSUE_WIDTH)
-#endif
-
-// Size of LSU Request Queue
-#ifndef LSUQ_SIZE
-#define LSUQ_SIZE   (4 * NUM_WARPS * (NUM_THREADS / NUM_LSU_LANES))
-#endif
-
-// LSU Duplicate Address Check
-#ifndef LSU_DUP_DISABLE
-#define LSU_DUP_ENABLE
-#endif
-#ifdef LSU_DUP_ENABLE
-#define LSU_DUP_ENABLED 1
-#else
-#define LSU_DUP_ENABLED 0
-#endif
-
-#ifdef GBAR_ENABLE
-#define GBAR_ENABLED 1
-#else
-#define GBAR_ENABLED 0
-#endif
-
-#ifndef LATENCY_IMUL
-#ifdef VIVADO
-#define LATENCY_IMUL 4
-#endif
-#ifdef QUARTUS
-#define LATENCY_IMUL 3
-#endif
-#ifndef LATENCY_IMUL
-#define LATENCY_IMUL 4
-#endif
-#endif
-
-// Floating-Point Units ///////////////////////////////////////////////////////
-
-// Size of FPU Request Queue
-#ifndef FPUQ_SIZE
-#define FPUQ_SIZE (2 * (NUM_THREADS / NUM_FPU_LANES))
-#endif
-
-// FNCP Latency
-#ifndef LATENCY_FNCP
-#define LATENCY_FNCP 2
-#endif
-
-// FMA Latency
-#ifndef LATENCY_FMA
-#ifdef FPU_DPI
-#define LATENCY_FMA 4    
-#endif
-#ifdef FPU_FPNEW
-#define LATENCY_FMA 4    
-#endif
-#ifdef FPU_DSP
-#ifdef QUARTUS
-#define LATENCY_FMA 4
-#endif
-#ifdef VIVADO
-#define LATENCY_FMA 16    
-#endif
-#ifndef LATENCY_FMA
-#define LATENCY_FMA 4    
-#endif
-#endif
-#endif
-
-// FDIV Latency
-#ifndef LATENCY_FDIV
-#ifdef FPU_DPI
-#define LATENCY_FDIV 15    
-#endif
-#ifdef FPU_FPNEW
-#define LATENCY_FDIV 16    
-#endif
-#ifdef FPU_DSP
-#ifdef QUARTUS
-#define LATENCY_FDIV 15
-#endif
-#ifdef VIVADO
-#define LATENCY_FDIV 28    
-#endif
-#ifndef LATENCY_FDIV
-#define LATENCY_FDIV 16
-#endif
-#endif
-#endif
-
-// FSQRT Latency
-#ifndef LATENCY_FSQRT
-#ifdef FPU_DPI
-#define LATENCY_FSQRT 10    
-#endif
-#ifdef FPU_FPNEW
-#define LATENCY_FSQRT 16    
-#endif
-#ifdef FPU_DSP
-#ifdef QUARTUS
-#define LATENCY_FSQRT 10
-#endif
-#ifdef VIVADO
-#define LATENCY_FSQRT 28    
-#endif
-#ifndef LATENCY_FSQRT
-#define LATENCY_FSQRT 16    
-#endif
-#endif
-#endif
-
-// FCVT Latency
-#ifndef LATENCY_FCVT
-#define LATENCY_FCVT 5
-#endif
-
-// Icache Configurable Knobs //////////////////////////////////////////////////
-
-// Cache Enable
-#ifndef ICACHE_DISABLE
-#define ICACHE_ENABLE
-#endif
-#ifdef ICACHE_ENABLE
-    #define ICACHE_ENABLED 1
-#else
-    #define ICACHE_ENABLED 0
-    #define NUM_ICACHES 0
-#endif
-
-// Number of Cache Units
-#ifndef NUM_ICACHES
-#define NUM_ICACHES UP(SOCKET_SIZE / 4)
-#endif
-
-// Cache Size
-#ifndef ICACHE_SIZE
-#define ICACHE_SIZE 16384
-#endif
-
-// Core Response Queue Size
-#ifndef ICACHE_CRSQ_SIZE
-#define ICACHE_CRSQ_SIZE 2
-#endif
-
-// Miss Handling Register Size
-#ifndef ICACHE_MSHR_SIZE
-#define ICACHE_MSHR_SIZE 16
-#endif
-
-// Memory Request Queue Size
-#ifndef ICACHE_MREQ_SIZE
-#define ICACHE_MREQ_SIZE 4
-#endif
-
-// Memory Response Queue Size
-#ifndef ICACHE_MRSQ_SIZE
-#define ICACHE_MRSQ_SIZE 0
-#endif
-
-// Number of Associative Ways
-#ifndef ICACHE_NUM_WAYS
-#define ICACHE_NUM_WAYS 1
-#endif
-
-// Dcache Configurable Knobs //////////////////////////////////////////////////
-
-// Cache Enable
-#ifndef DCACHE_DISABLE
-#define DCACHE_ENABLE
-#endif
-#ifdef DCACHE_ENABLE
-    #define DCACHE_ENABLED 1
-#else
-    #define DCACHE_ENABLED 0
-    #define NUM_DCACHES 0
-    #define DCACHE_NUM_BANKS 1
-#endif
-
-// Number of Cache Units
-#ifndef NUM_DCACHES
-#define NUM_DCACHES UP(SOCKET_SIZE / 4)
-#endif
-
-// Cache Size
-#ifndef DCACHE_SIZE
-#define DCACHE_SIZE 16384
-#endif
-
-// Number of Banks
-#ifndef DCACHE_NUM_BANKS
-#define DCACHE_NUM_BANKS NUM_LSU_LANES
-#endif
-
-// Core Response Queue Size
-#ifndef DCACHE_CRSQ_SIZE
-#define DCACHE_CRSQ_SIZE 2
-#endif
-
-// Miss Handling Register Size
-#ifndef DCACHE_MSHR_SIZE
-#define DCACHE_MSHR_SIZE 8
-#endif
-
-// Memory Request Queue Size
-#ifndef DCACHE_MREQ_SIZE
-#define DCACHE_MREQ_SIZE 4
-#endif
-
-// Memory Response Queue Size
-#ifndef DCACHE_MRSQ_SIZE
-#define DCACHE_MRSQ_SIZE 0
-#endif
-
-// Number of Associative Ways
-#ifndef DCACHE_NUM_WAYS
-#define DCACHE_NUM_WAYS 1
-#endif
-
-// SM Configurable Knobs //////////////////////////////////////////////////////
-
-#ifndef SM_DISABLE
-#define SM_ENABLE
-#endif
-
-#ifdef SM_ENABLE
-    #define SM_ENABLED   1
-#else
-    #define SM_ENABLED   0
-    #define SMEM_NUM_BANKS 1
-#endif
-
-// Number of Banks
-#ifndef SMEM_NUM_BANKS
-#define SMEM_NUM_BANKS (NUM_LSU_LANES)
-#endif
-
-// L2cache Configurable Knobs /////////////////////////////////////////////////
-
-// Cache Size
-#ifndef L2_CACHE_SIZE
-#ifdef ALTERA_S10
-#define L2_CACHE_SIZE 2097152
-#else
-#define L2_CACHE_SIZE 1048576
-#endif
-#endif
-
-// Number of Banks
-#ifndef L2_NUM_BANKS
-#define L2_NUM_BANKS MIN(4, NUM_SOCKETS)
-#endif
-
-// Core Response Queue Size
-#ifndef L2_CRSQ_SIZE
-#define L2_CRSQ_SIZE 2
-#endif
-
-// Miss Handling Register Size
-#ifndef L2_MSHR_SIZE
-#define L2_MSHR_SIZE 16
-#endif
-
-// Memory Request Queue Size
-#ifndef L2_MREQ_SIZE
-#define L2_MREQ_SIZE 4
-#endif
-
-// Memory Response Queue Size
-#ifndef L2_MRSQ_SIZE
-#define L2_MRSQ_SIZE 0
-#endif
-
-// Number of Associative Ways
-#ifndef L2_NUM_WAYS
-#define L2_NUM_WAYS 2
-#endif
-
-// L3cache Configurable Knobs /////////////////////////////////////////////////
-
-// Cache Size
-#ifndef L3_CACHE_SIZE
-#ifdef ALTERA_S10
-#define L3_CACHE_SIZE 2097152
-#else
-#define L3_CACHE_SIZE 1048576
-#endif
-#endif
-
-// Number of Banks
-#ifndef L3_NUM_BANKS
-#define L3_NUM_BANKS MIN(4, NUM_CLUSTERS)
-#endif
-
-// Core Response Queue Size
-#ifndef L3_CRSQ_SIZE
-#define L3_CRSQ_SIZE 2
-#endif
-
-// Miss Handling Register Size
-#ifndef L3_MSHR_SIZE
-#define L3_MSHR_SIZE 16
-#endif
-
-// Memory Request Queue Size
-#ifndef L3_MREQ_SIZE
-#define L3_MREQ_SIZE 4
-#endif
-
-// Memory Response Queue Size
-#ifndef L3_MRSQ_SIZE
-#define L3_MRSQ_SIZE 0
-#endif
-
-// Number of Associative Ways
-#ifndef L3_NUM_WAYS
-#define L3_NUM_WAYS 4
-#endif
-
-// ISA Extensions /////////////////////////////////////////////////////////////
-
-#ifdef EXT_A_ENABLE
-    #define EXT_A_ENABLED   1
-#else
-    #define EXT_A_ENABLED   0
-#endif
-
-#ifdef EXT_C_ENABLE
-    #define EXT_C_ENABLED   1
-#else
-    #define EXT_C_ENABLED   0
-#endif
-
-#ifdef EXT_D_ENABLE
-    #define EXT_D_ENABLED   1
-#else
-    #define EXT_D_ENABLED   0
-#endif
-
-#ifdef EXT_F_ENABLE
-    #define EXT_F_ENABLED   1
-#else
-    #define EXT_F_ENABLED   0
-#endif
-
-#ifdef EXT_M_ENABLE
-    #define EXT_M_ENABLED   1
-#else
-    #define EXT_M_ENABLED   0
-#endif
-
-#define ISA_STD_A           0
-#define ISA_STD_C           2
-#define ISA_STD_D           3
-#define ISA_STD_E           4
-#define ISA_STD_F           5
-#define ISA_STD_H           7
-#define ISA_STD_I           8
-#define ISA_STD_N           13
-#define ISA_STD_Q           16
-#define ISA_STD_S           18
-#define ISA_STD_U           20
-
-#define ISA_EXT_ICACHE      0
-#define ISA_EXT_DCACHE      1
-#define ISA_EXT_L2CACHE     2
-#define ISA_EXT_L3CACHE     3
-#define ISA_EXT_SMEM        4
-
-#define MISA_EXT  (ICACHE_ENABLED  << ISA_EXT_ICACHE) \
-                | (DCACHE_ENABLED  << ISA_EXT_DCACHE) \
-                | (L2_ENABLED      << ISA_EXT_L2CACHE) \
-                | (L3_ENABLED      << ISA_EXT_L3CACHE) \
-                | (SM_ENABLED      << ISA_EXT_SMEM)
-
-#define MISA_STD  (EXT_A_ENABLED <<  0) /* A - Atomic Instructions extension */ \
-                | (0 <<  1) /* B - Tentatively reserved for Bit operations extension */ \
-                | (EXT_C_ENABLED <<  2) /* C - Compressed extension */ \
-                | (EXT_D_ENABLED <<  3) /* D - Double precsision floating-point extension */ \
-                | (0 <<  4) /* E - RV32E base ISA */ \
-                | (EXT_F_ENABLED << 5) /* F - Single precsision floating-point extension */ \
-                | (0 <<  6) /* G - Additional standard extensions present */ \
-                | (0 <<  7) /* H - Hypervisor mode implemented */ \
-                | (1 <<  8) /* I - RV32I/64I/128I base ISA */ \
-                | (0 <<  9) /* J - Reserved */ \
-                | (0 << 10) /* K - Reserved */ \
-                | (0 << 11) /* L - Tentatively reserved for Bit operations extension */ \
-                | (EXT_M_ENABLED << 12) /* M - Integer Multiply/Divide extension */ \
-                | (0 << 13) /* N - User level interrupts supported */ \
-                | (0 << 14) /* O - Reserved */ \
-                | (0 << 15) /* P - Tentatively reserved for Packed-SIMD extension */ \
-                | (0 << 16) /* Q - Quad-precision floating-point extension */ \
-                | (0 << 17) /* R - Reserved */ \
-                | (0 << 18) /* S - Supervisor mode implemented */ \
-                | (0 << 19) /* T - Tentatively reserved for Transactional Memory extension */ \
-                | (1 << 20) /* U - User mode implemented */ \
-                | (0 << 21) /* V - Tentatively reserved for Vector extension */ \
-                | (0 << 22) /* W - Reserved */ \
-                | (1 << 23) /* X - Non-standard extensions present */ \
-                | (0 << 24) /* Y - Reserved */ \
-                | (0 << 25) /* Z - Reserved */
-
-// Device identification //////////////////////////////////////////////////////
-
-#define VENDOR_ID           0
-#define ARCHITECTURE_ID     0
-#define IMPLEMENTATION_ID   0
-
-#endif // VX_CONFIG_VH

hw/dpi/float_dpi.cpp

@@ -1,16 +1,3 @@
-// Copyright © 2019-2023
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// 
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
 #include <stdio.h>
 #include <math.h>
 #include <unordered_map>
@@ -18,563 +5,167 @@
 #include <mutex>
 #include <iostream>
 #include <rvfloats.h>
-#include <util.h>
 #include "svdpi.h"
-// #include "verilated_vpi.h"
+#include "verilated_vpi.h"
 #include "VX_config.h"
 
-#include <bit>
-#include "half.h"
-
 extern "C" {
-  void dpi_fadd(bool enable, int dst_fmt, int64_t a, int64_t b, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
-  void dpi_fsub(bool enable, int dst_fmt, int64_t a, int64_t b, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
-  void dpi_fmul(bool enable, int dst_fmt, int64_t a, int64_t b, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
-  void dpi_fmadd(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t c, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
-  void dpi_fmsub(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t c, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
-  void dpi_fnmadd(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t c, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
-  void dpi_fnmsub(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t c, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
+  void dpi_fadd(bool enable, int a, int b, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
+  void dpi_fsub(bool enable, int a, int b, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
+  void dpi_fmul(bool enable, int a, int b, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
+  void dpi_fmadd(bool enable, int a, int b, int c, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
+  void dpi_fmsub(bool enable, int a, int b, int c, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
+  void dpi_fnmadd(bool enable, int a, int b, int c, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
+  void dpi_fnmsub(bool enable, int a, int b, int c, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
 
-  void dpi_fdiv(bool enable, int dst_fmt, int64_t a, int64_t b, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
-  void dpi_fsqrt(bool enable, int dst_fmt, int64_t a, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
-
-  void dpi_ftoi(bool enable, int dst_fmt, int src_fmt, int64_t a, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
-  void dpi_ftou(bool enable, int dst_fmt, int src_fmt, int64_t a, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
-  void dpi_itof(bool enable, int dst_fmt, int src_fmt, int64_t a, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
-  void dpi_utof(bool enable, int dst_fmt, int src_fmt, int64_t a, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags);
-  void dpi_f2f(bool enable, int dst_fmt, int64_t a, int64_t* result);
+  void dpi_fdiv(bool enable, int a, int b, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
+  void dpi_fsqrt(bool enable, int a, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
   
-  void dpi_fclss(bool enable, int dst_fmt, int64_t a, int64_t* result);
-  void dpi_fsgnj(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result);
-  void dpi_fsgnjn(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result);
-  void dpi_fsgnjx(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result);
+  void dpi_ftoi(bool enable, int a, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
+  void dpi_ftou(bool enable, int a, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
+  void dpi_itof(bool enable, int a, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
+  void dpi_utof(bool enable, int a, const svBitVecVal* frm, int* result, svBitVecVal* fflags);
 
-  void dpi_flt(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result, svBitVecVal* fflags);
-  void dpi_fle(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result, svBitVecVal* fflags);
-  void dpi_feq(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result, svBitVecVal* fflags);
-  void dpi_fmin(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result, svBitVecVal* fflags);
-  void dpi_fmax(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result, svBitVecVal* fflags);
+  void dpi_fclss(bool enable, int a, int* result);
+  void dpi_fsgnj(bool enable, int a, int b, int* result);
+  void dpi_fsgnjn(bool enable, int a, int b, int* result);
+  void dpi_fsgnjx(bool enable, int a, int b, int* result);
 
-  void dpi_hmma(bool enable, const svBitVecVal* A_tile, const svBitVecVal* B_tile, const svBitVecVal* C_tile, svBitVecVal* D_tile);
-  void dpi_print_results(int wid, int octet, const svBitVecVal* A_tile, const svBitVecVal* B_tile, const svBitVecVal* C_tile, const svBitVecVal* D_tile);
+  void dpi_flt(bool enable, int a, int b, int* result, svBitVecVal* fflags);
+  void dpi_fle(bool enable, int a, int b, int* result, svBitVecVal* fflags);
+  void dpi_feq(bool enable, int a, int b, int* result, svBitVecVal* fflags);
+  void dpi_fmin(bool enable, int a, int b, int* result, svBitVecVal* fflags);
+  void dpi_fmax(bool enable, int a, int b, int* result, svBitVecVal* fflags);
 }
 
-inline uint64_t nan_box(uint32_t value) {
-#ifdef FPU_RV64F
-  return value | 0xffffffff00000000;
-#else 
-  return value;
-#endif
-}
-
-inline bool is_nan_boxed(uint64_t value) {
-#ifdef FPU_RV64F
-  return (uint32_t(value >> 32) == 0xffffffff);
-#else
-  __unused (value);
-  return true;
-#endif
-}
-
-inline int64_t check_boxing(int64_t a) {  
-  if (!is_nan_boxed(a)) {
-    return nan_box(0x7fc00000); // NaN
-  }
-  return a;
-}
-
-void dpi_fadd(bool enable, int dst_fmt, int64_t a, int64_t b, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_fadd(bool enable, int a, int b, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) {
-    *result = rv_fadd_d(a, b, (*frm & 0x7), fflags);
-  } else {
-    *result = nan_box(rv_fadd_s(check_boxing(a), check_boxing(b), (*frm & 0x7), fflags));
-  }
+  *result = rv_fadd_s(a, b, (*frm & 0x7), fflags);
 }
 
-void dpi_fsub(bool enable, int dst_fmt, int64_t a, int64_t b, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_fsub(bool enable, int a, int b, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) {
-    *result = rv_fsub_d(a, b, (*frm & 0x7), fflags);
-  } else {
-    *result = nan_box(rv_fsub_s(check_boxing(a), check_boxing(b), (*frm & 0x7), fflags));
-  }
+  *result = rv_fsub_s(a, b, (*frm & 0x7), fflags);
 }
 
-void dpi_fmul(bool enable, int dst_fmt, int64_t a, int64_t b, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_fmul(bool enable, int a, int b, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fmul_d(a, b, (*frm & 0x7), fflags); 
-  } else {
-    *result = nan_box(rv_fmul_s(check_boxing(a), check_boxing(b), (*frm & 0x7), fflags));
-  }
+  *result = rv_fmul_s(a, b, (*frm & 0x7), fflags);
 }
 
-void dpi_fmadd(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t c, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_fmadd(bool enable, int a, int b, int c, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fmadd_d(a, b, c, (*frm & 0x7), fflags);
-  } else {
-    *result = nan_box(rv_fmadd_s(check_boxing(a), check_boxing(b), check_boxing(c), (*frm & 0x7), fflags));
-  }
+  *result = rv_fmadd_s(a, b, c, (*frm & 0x7), fflags);
 }
 
-void dpi_fmsub(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t c, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_fmsub(bool enable, int a, int b, int c, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fmsub_d(a, b, c, (*frm & 0x7), fflags);
-  } else {
-    *result = nan_box(rv_fmsub_s(check_boxing(a), check_boxing(b), check_boxing(c), (*frm & 0x7), fflags));
-  }
+  *result = rv_fmsub_s(a, b, c, (*frm & 0x7), fflags);
 }
 
-void dpi_fnmadd(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t c, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_fnmadd(bool enable, int a, int b, int c, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fnmadd_d(a, b, c, (*frm & 0x7), fflags);
-  } else {
-    *result = nan_box(rv_fnmadd_s(check_boxing(a), check_boxing(b), check_boxing(c), (*frm & 0x7), fflags));
-  }
+  *result = rv_fnmadd_s(a, b, c, (*frm & 0x7), fflags);
 }
 
-void dpi_fnmsub(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t c, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_fnmsub(bool enable, int a, int b, int c, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fnmsub_d(a, b, c, (*frm & 0x7), fflags);
-  } else {
-    *result = nan_box(rv_fnmsub_s(check_boxing(a), check_boxing(b), check_boxing(c), (*frm & 0x7), fflags));
-  }
+  *result = rv_fnmsub_s(a, b, c, (*frm & 0x7), fflags);
 }
 
-void dpi_fdiv(bool enable, int dst_fmt, int64_t a, int64_t b, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_fdiv(bool enable, int a, int b, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fdiv_d(a, b, (*frm & 0x7), fflags); 
-  } else {
-    *result = nan_box(rv_fdiv_s(check_boxing(a), check_boxing(b), (*frm & 0x7), fflags));
-  }
+  *result = rv_fdiv_s(a, b, (*frm & 0x7), fflags);
 }
 
-void dpi_fsqrt(bool enable, int dst_fmt, int64_t a, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_fsqrt(bool enable, int a, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fsqrt_d(a, (*frm & 0x7), fflags); 
-  } else {
-    *result = nan_box(rv_fsqrt_s(check_boxing(a), (*frm & 0x7), fflags));
-  }
+  *result = rv_fsqrt_s(a, (*frm & 0x7), fflags);
 }
 
-void dpi_ftoi(bool enable, int dst_fmt, int src_fmt, int64_t a, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_ftoi(bool enable, int a, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) {
-    if (src_fmt) { 
-      *result = rv_ftol_d(a, (*frm & 0x7), fflags);
-    } else {
-      *result = rv_ftol_s(check_boxing(a), (*frm & 0x7), fflags);
-    }
-  } else {    
-    if (src_fmt) { 
-      *result = sext<uint64_t>(rv_ftoi_d(a, (*frm & 0x7), fflags), 32);
-    } else {
-      *result = sext<uint64_t>(rv_ftoi_s(check_boxing(a), (*frm & 0x7), fflags), 32);
-    }
-  }
+  *result = rv_ftoi_s(a, (*frm & 0x7), fflags);
 }
 
-void dpi_ftou(bool enable, int dst_fmt, int src_fmt, int64_t a, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_ftou(bool enable, int a, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) {
-    if (src_fmt) { 
-      *result = rv_ftolu_d(a, (*frm & 0x7), fflags);
-    } else {
-      *result = rv_ftolu_s(check_boxing(a), (*frm & 0x7), fflags);
-    }
-  } else {    
-    if (src_fmt) { 
-      *result = sext<uint64_t>(rv_ftou_d(a, (*frm & 0x7), fflags), 32);
-    } else {
-      *result = sext<uint64_t>(rv_ftou_s(check_boxing(a), (*frm & 0x7), fflags), 32); 
-    }
-  }
+  *result = rv_ftou_s(a, (*frm & 0x7), fflags);
 }
 
-void dpi_itof(bool enable, int dst_fmt, int src_fmt, int64_t a, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_itof(bool enable, int a, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) {
-    if (src_fmt) { 
-      *result = rv_ltof_d(a, (*frm & 0x7), fflags);
-    } else { 
-      *result = rv_itof_d(a, (*frm & 0x7), fflags);
-    }
-  } else {
-    if (src_fmt) { 
-      *result = nan_box(rv_ltof_s(a, (*frm & 0x7), fflags)); 
-    } else { 
-      *result = nan_box(rv_itof_s(a, (*frm & 0x7), fflags)); 
-    }
-  }
+  *result = rv_itof_s(a, (*frm & 0x7), fflags);
 }
 
-void dpi_utof(bool enable, int dst_fmt, int src_fmt, int64_t a, const svBitVecVal* frm, int64_t* result, svBitVecVal* fflags) {
+void dpi_utof(bool enable, int a, const svBitVecVal* frm, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) {
-    if (src_fmt) { 
-      *result = rv_lutof_d(a, (*frm & 0x7), fflags);
-    } else { 
-      *result = rv_utof_d(a, (*frm & 0x7), fflags);
-    }
-  } else {
-    if (src_fmt) { 
-      *result = nan_box(rv_lutof_s(a, (*frm & 0x7), fflags));
-    } else { 
-      *result = nan_box(rv_utof_s(a, (*frm & 0x7), fflags));
-    }
-  }
+  *result = rv_utof_s(a, (*frm & 0x7), fflags);
 }
 
-void dpi_f2f(bool enable, int dst_fmt, int64_t a, int64_t* result) {
+void dpi_flt(bool enable, int a, int b, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) {
-    *result = rv_ftod((int32_t)check_boxing(a));
-  } else {
-    *result = nan_box(rv_dtof(a));
-  }
+  *result = rv_flt_s(a, b, fflags);
 }
 
-void dpi_fclss(bool enable, int dst_fmt, int64_t a, int64_t* result) {
+void dpi_fle(bool enable, int a, int b, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fclss_d(a); 
-  } else { 
-    *result = rv_fclss_s(check_boxing(a)); 
-  }
+  *result = rv_fle_s(a, b, fflags);
 }
 
-void dpi_fsgnj(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result) {
+void dpi_feq(bool enable, int a, int b, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fsgnj_d(a, b); 
-  } else {
-    *result = nan_box(rv_fsgnj_s(check_boxing(a), check_boxing(b)));
-  }
+  *result = rv_feq_s(a, b, fflags);
 }
 
-void dpi_fsgnjn(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result) {
+void dpi_fmin(bool enable, int a, int b, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fsgnjn_d(a, b); 
-  } else {
-    *result = nan_box(rv_fsgnjn_s(check_boxing(a), check_boxing(b)));
-  }
+  *result = rv_fmin_s(a, b, fflags);
 }
 
-void dpi_fsgnjx(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result) {
+void dpi_fmax(bool enable, int a, int b, int* result, svBitVecVal* fflags) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fsgnjx_d(a, b); 
-  } else {
-    *result = nan_box(rv_fsgnjx_s(check_boxing(a), check_boxing(b)));
-  }
+  *result = rv_fmax_s(a, b, fflags);
 }
 
-void dpi_flt(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result, svBitVecVal* fflags) {
+void dpi_fclss(bool enable, int a, int* result) {
   if (!enable) 
     return;
-  if (dst_fmt) {
-    *result = rv_flt_d(a, b, fflags); 
-  } else {
-    *result = rv_flt_s(check_boxing(a), check_boxing(b), fflags);
-  }
+  *result = rv_fclss_s(a);
 }
 
-void dpi_fle(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result, svBitVecVal* fflags) {
+void dpi_fsgnj(bool enable, int a, int b, int* result) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fle_d(a, b, fflags); 
-  } else {
-    *result = rv_fle_s(check_boxing(a), check_boxing(b), fflags);
-  }
+  *result = rv_fsgnj_s(a, b);
 }
 
-void dpi_feq(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result, svBitVecVal* fflags) {
+void dpi_fsgnjn(bool enable, int a, int b, int* result) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_feq_d(a, b, fflags); 
-  } else {
-    *result = rv_feq_s(check_boxing(a), check_boxing(b), fflags);
-  }
+  *result = rv_fsgnjn_s(a, b);
 }
 
-void dpi_fmin(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result, svBitVecVal* fflags) {
+void dpi_fsgnjx(bool enable, int a, int b, int* result) {
   if (!enable) 
     return;
-  if (dst_fmt) { 
-    *result = rv_fmin_d(a, b, fflags); 
-  } else {
-    *result = nan_box(rv_fmin_s(check_boxing(a), check_boxing(b), fflags));
-  }
-}
-
-void dpi_fmax(bool enable, int dst_fmt, int64_t a, int64_t b, int64_t* result, svBitVecVal* fflags) {
-  if (!enable) 
-    return;
-  if (dst_fmt) { 
-    *result = rv_fmax_d(a, b, fflags); 
-  } else {
-    *result = nan_box(rv_fmax_s(check_boxing(a), check_boxing(b), fflags));
-  }
-}
-
-// A is M * K, B is K * M, C is M * M, D is M * M
-#define M 4
-#define K 2 // FIXME: 4x4x1 / cycle / octet!
-
-// all row major
-float c_A_tile[M][K];
-float c_B_tile[K][M];
-float c_C_tile[M][M];
-float c_D_tile[M][M];
-
-// code assumes that svBitVecVal is basically a uint32_t
-static_assert(sizeof(svBitVecVal) == 4);
-
-void clear_float_array(float* c_tile, int rows, int cols) {
-  for (int i = 0; i < rows; i += 1) {
-    for (int j = 0; j < cols; j += 1) {
-      int index = i * cols + j;
-      c_tile[index] = 0.0f;
-    }
-  }
-}
-
-void fill_float_array(const svBitVecVal* sv_tile, float* c_tile, int rows, int cols) {
-  
-  for (int i = 0; i < rows; i += 1) {
-    for (int j = 0; j < cols; j += 1) {
-      int index = i * cols + j;
-      svBitVecVal sv_val = sv_tile[index];
-
-      uint32_t c_val = sv_val;
-      float c_float;
-
-      memcpy(&c_float, &c_val, sizeof(c_float));
-      c_tile[index] = c_float;
-      
-      // std::cout << c_float << " ";
-    }
-    // std::cout << std::endl;
-  }
-}
-
-void write_float_array(svBitVecVal* sv_tile, float* c_tile, int rows, int cols) {
-  for (int i = 0; i < rows; i += 1) {
-    for (int j = 0; j < cols; j += 1) {
-      int index = i * cols + j;
-      svBitVecVal* sv_val = &sv_tile[index];
-
-      float c_float = c_tile[index];
-      memcpy(sv_val, &c_float, sizeof(c_float));
-
-      // std::cout << c_float << " ";
-    }
-    // std::cout << std::endl;
-  }
-}
-
-void dpi_hmma(bool enable, const svBitVecVal* A_tile, const svBitVecVal* B_tile, const svBitVecVal* C_tile, svBitVecVal* D_tile) {
-  if (!enable) {
-    return;
-  }
-  clear_float_array(&c_A_tile[0][0], M, K);
-  clear_float_array(&c_B_tile[0][0], K, M);
-  clear_float_array(&c_C_tile[0][0], M, M);
-  clear_float_array(&c_D_tile[0][0], M, M);
-
-  // std::cout << "A: " << std::endl;
-  fill_float_array(A_tile, &c_A_tile[0][0], M, K);
-  // std::cout << "B: " << std::endl;
-  fill_float_array(B_tile, &c_B_tile[0][0], K, M);
-  // std::cout << "C: " << std::endl;
-  fill_float_array(C_tile, &c_C_tile[0][0], M, M);
-  
-  for (int i = 0; i < M; i += 1) {
-    for (int j = 0; j < M; j += 1) {
-      float accum = c_C_tile[i][j];
-      for (int k = 0; k < K; k += 1) {
-        accum += c_A_tile[i][k] * c_B_tile[k][j];
-      }
-      c_D_tile[i][j] = accum;
-    }
-  }
-
-  write_float_array(D_tile, &c_D_tile[0][0], M, M);
-}
-
-// 1 copy per warp
-float A_tile_full[4][16][8];
-float B_tile_full[4][8][16];
-float C_tile_full[4][16][16];
-float D_tile_full[4][16][16];
-int steps[4];
-
-void print_array(float* array, int rows, int cols) {
-  for (int i = 0; i < rows; i += 1) {
-    for (int j = 0; j < cols; j += 1) {
-      std::cout << array[i*cols+j] << " ";
-    }
-    std::cout << "\n";
-  }
-  std::cout << std::endl;
-}
-
-void dpi_print_results(int wid, int octet, const svBitVecVal* A_tile, const svBitVecVal* B_tile, const svBitVecVal* C_tile, const svBitVecVal* D_tile) {
-  // std::cout << "A: " << std::endl;
-  fill_float_array(A_tile, &c_A_tile[0][0], M, K);
-  // std::cout << "B: " << std::endl;
-  fill_float_array(B_tile, &c_B_tile[0][0], K, M);
-  // std::cout << "C: " << std::endl;
-  fill_float_array(C_tile, &c_C_tile[0][0], M, M);
-  // for some reason this still holds onto old value? very strange
-  // std::cout << "D: " << std::endl;
-  fill_float_array(D_tile, &c_D_tile[0][0], M, M);
-
-  int octet_row_offset;
-  int octet_col_offset;
-  switch(octet) {
-  case 0:
-    octet_row_offset = 0;
-    octet_col_offset = 0;
-    break;
-  case 1:
-    octet_row_offset = 8;
-    octet_col_offset = 0;
-    break;
-  case 2:
-    octet_row_offset = 0;
-    octet_col_offset = 8;
-    break;
-  case 3:
-    octet_row_offset = 8;
-    octet_col_offset = 8;
-    break;
-  }
-
-  int step_row_offset;
-  int step_col_offset;
-  int step = (steps[wid] % 16) / 4;
-  int set = (steps[wid] / 16);
-  switch(step) {
-  case 0:
-    step_row_offset = 0;
-    step_col_offset = 0;
-    break;
-  case 1:
-    step_row_offset = 2;
-    step_col_offset = 0;
-    break;
-  case 2:
-    step_row_offset = 0;
-    step_col_offset = 4;
-    break;
-  case 3:
-    step_row_offset = 2;
-    step_col_offset = 4;
-    break;
-  }
-  
-  if (steps[0] >= 48) {
-    // std::cout << "octet " << octet << " step " << steps[0] << "\n";
-    // print_array(&c_D_tile[0][0], 4, 4); 
-  }
-
-  D_tile_full[wid][octet_row_offset+step_row_offset+0][octet_col_offset+step_col_offset+0] = c_D_tile[0][0];
-  D_tile_full[wid][octet_row_offset+step_row_offset+0][octet_col_offset+step_col_offset+1] = c_D_tile[0][1];
-  D_tile_full[wid][octet_row_offset+step_row_offset+0][octet_col_offset+step_col_offset+2] = c_D_tile[0][2];
-  D_tile_full[wid][octet_row_offset+step_row_offset+0][octet_col_offset+step_col_offset+3] = c_D_tile[0][3];
-  D_tile_full[wid][octet_row_offset+step_row_offset+1][octet_col_offset+step_col_offset+0] = c_D_tile[1][0];
-  D_tile_full[wid][octet_row_offset+step_row_offset+1][octet_col_offset+step_col_offset+1] = c_D_tile[1][1];
-  D_tile_full[wid][octet_row_offset+step_row_offset+1][octet_col_offset+step_col_offset+2] = c_D_tile[1][2];
-  D_tile_full[wid][octet_row_offset+step_row_offset+1][octet_col_offset+step_col_offset+3] = c_D_tile[1][3];
-  D_tile_full[wid][octet_row_offset+step_row_offset+4][octet_col_offset+step_col_offset+0] = c_D_tile[2][0];
-  D_tile_full[wid][octet_row_offset+step_row_offset+4][octet_col_offset+step_col_offset+1] = c_D_tile[2][1];
-  D_tile_full[wid][octet_row_offset+step_row_offset+4][octet_col_offset+step_col_offset+2] = c_D_tile[2][2];
-  D_tile_full[wid][octet_row_offset+step_row_offset+4][octet_col_offset+step_col_offset+3] = c_D_tile[2][3];
-  D_tile_full[wid][octet_row_offset+step_row_offset+5][octet_col_offset+step_col_offset+0] = c_D_tile[3][0];
-  D_tile_full[wid][octet_row_offset+step_row_offset+5][octet_col_offset+step_col_offset+1] = c_D_tile[3][1];
-  D_tile_full[wid][octet_row_offset+step_row_offset+5][octet_col_offset+step_col_offset+2] = c_D_tile[3][2];
-  D_tile_full[wid][octet_row_offset+step_row_offset+5][octet_col_offset+step_col_offset+3] = c_D_tile[3][3];
-
-  if (octet == 0 || octet == 1) {
-    octet_row_offset = octet * 8;
-    if (step == 0) {
-      step_row_offset = 0; 
-    }
-    if (step == 1) {
-      step_row_offset = 2;
-    }
-    if (step == 0 || step == 1) {
-      A_tile_full[wid][octet_row_offset+step_row_offset+0][set*2+0] = c_A_tile[0][0];
-      A_tile_full[wid][octet_row_offset+step_row_offset+0][set*2+1] = c_A_tile[0][1];
-      A_tile_full[wid][octet_row_offset+step_row_offset+1][set*2+0] = c_A_tile[1][0];
-      A_tile_full[wid][octet_row_offset+step_row_offset+1][set*2+1] = c_A_tile[1][1];
-      A_tile_full[wid][octet_row_offset+step_row_offset+4][set*2+0] = c_A_tile[2][0];
-      A_tile_full[wid][octet_row_offset+step_row_offset+4][set*2+1] = c_A_tile[2][1];
-      A_tile_full[wid][octet_row_offset+step_row_offset+5][set*2+0] = c_A_tile[3][0];
-      A_tile_full[wid][octet_row_offset+step_row_offset+5][set*2+1] = c_A_tile[3][1];
-    }
-  }
-
-  if (octet == 0 || octet == 2) {
-    octet_col_offset = octet * 4;
-    if (step == 0) {
-      step_col_offset = 0; 
-    }
-    else if (step == 2) {
-      step_col_offset = 4;
-    }
-    if (step == 0 || step == 2) {
-      B_tile_full[wid][set*2+0][octet_col_offset+step_col_offset+0] = c_B_tile[0][0];
-      B_tile_full[wid][set*2+0][octet_col_offset+step_col_offset+1] = c_B_tile[0][1];
-      B_tile_full[wid][set*2+0][octet_col_offset+step_col_offset+2] = c_B_tile[0][2];
-      B_tile_full[wid][set*2+0][octet_col_offset+step_col_offset+3] = c_B_tile[0][3];
-      B_tile_full[wid][set*2+1][octet_col_offset+step_col_offset+0] = c_B_tile[1][0];
-      B_tile_full[wid][set*2+1][octet_col_offset+step_col_offset+1] = c_B_tile[1][1];
-      B_tile_full[wid][set*2+1][octet_col_offset+step_col_offset+2] = c_B_tile[1][2];
-      B_tile_full[wid][set*2+1][octet_col_offset+step_col_offset+3] = c_B_tile[1][3];
-    }
-  }
-
-  steps[wid] += 1;
-  if (steps[wid] % 32 == 0) {
-    steps[wid] = 0;
-    std::cout << "warp " << wid << " finished wmma\n";
-    std::cout << "A tile" << "\n";
-    print_array(&A_tile_full[wid][0][0], 16, 8);
-    std::cout << "B tile" << "\n";
-    print_array(&B_tile_full[wid][0][0], 8, 16);
-    // std::cout << "C tile" << "\n";
-    // print_array(&C_tile_full[wid][0][0], 16, 16);
-    std::cout << "D tile" << "\n";
-    print_array(&D_tile_full[wid][0][0], 16, 16);
-  }
-}
+  *result = rv_fsgnjx_s(a, b);
+}

hw/dpi/float_dpi.vh

@@ -1,50 +1,31 @@
-// Copyright © 2019-2023
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// 
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
+`ifndef FLOAT_DPI
+`define FLOAT_DPI
 
-`ifndef FLOAT_DPI_VH
-`define FLOAT_DPI_VH
+import "DPI-C" function void dpi_fadd(input logic enable, input int a, input int b, input bit[2:0] frm, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_fsub(input logic enable, input int a, input int b, input bit[2:0] frm, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_fmul(input logic enable, input int a, input int b, input bit[2:0] frm, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_fmadd(input logic enable, input int a, input int b, input int c, input bit[2:0] frm, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_fmsub(input logic enable, input int a, input int b, input int c, input bit[2:0] frm, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_fnmadd(input logic enable, input int a, input int b, input int c, input bit[2:0] frm, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_fnmsub(input logic enable, input int a, input int b, input int c, input bit[2:0] frm, output int result, output bit[4:0] fflags);
 
-`include "VX_config.vh"
+import "DPI-C" function void dpi_fdiv(input logic enable, input int a, input int b, input bit[2:0] frm, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_fsqrt(input logic enable, input int a, input bit[2:0] frm, output int result, output bit[4:0] fflags);
 
-import "DPI-C" function void dpi_fadd(input logic enable, input int dst_fmt, input longint a, input longint b, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_fsub(input logic enable, input int dst_fmt, input longint a, input longint b, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_fmul(input logic enable, input int dst_fmt, input longint a, input longint b, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_fmadd(input logic enable, input int dst_fmt, input longint a, input longint b, input longint c, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_fmsub(input logic enable, input int dst_fmt, input longint a, input longint b, input longint c, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_fnmadd(input logic enable, input int dst_fmt, input longint a, input longint b, input longint c, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_fnmsub(input logic enable, input int dst_fmt, input longint a, input longint b, input longint c, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_ftoi(input logic enable, input int a, input bit[2:0] frm, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_ftou(input logic enable, input int a, input bit[2:0] frm, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_itof(input logic enable, input int a, input bit[2:0] frm, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_utof(input logic enable, input int a, input bit[2:0] frm, output int result, output bit[4:0] fflags);
 
-import "DPI-C" function void dpi_fdiv(input logic enable, input int dst_fmt, input longint a, input longint b, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_fsqrt(input logic enable, input int dst_fmt, input longint a, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_fclss(input logic enable, input int a, output int result);
+import "DPI-C" function void dpi_fsgnj(input logic enable, input int a, input int b, output int result);
+import "DPI-C" function void dpi_fsgnjn(input logic enable, input int a, input int b, output int result);
+import "DPI-C" function void dpi_fsgnjx(input logic enable, input int a, input int b, output int result);
 
-import "DPI-C" function void dpi_ftoi(input logic enable, input int dst_fmt, input int src_fmt, input longint a, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_ftou(input logic enable, input int dst_fmt, input int src_fmt, input longint a, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_itof(input logic enable, input int dst_fmt, input int src_fmt, input longint a, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_utof(input logic enable, input int dst_fmt, input int src_fmt, input longint a, input bit[2:0] frm, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_f2f(input logic enable, input int dst_fmt, input longint a, output longint result);
+import "DPI-C" function void dpi_flt(input logic enable, input int a, input int b, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_fle(input logic enable, input int a, input int b, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_feq(input logic enable, input int a, input int b, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_fmin(input logic enable, input int a, input int b, output int result, output bit[4:0] fflags);
+import "DPI-C" function void dpi_fmax(input logic enable, input int a, input int b, output int result, output bit[4:0] fflags);
 
-import "DPI-C" function void dpi_fclss(input logic enable, input int dst_fmt, input longint a, output longint result);
-import "DPI-C" function void dpi_fsgnj(input logic enable, input int dst_fmt, input longint a, input longint b, output longint result);
-import "DPI-C" function void dpi_fsgnjn(input logic enable, input int dst_fmt, input longint a, input longint b, output longint result);
-import "DPI-C" function void dpi_fsgnjx(input logic enable, input int dst_fmt, input longint a, input longint b, output longint result);
-
-import "DPI-C" function void dpi_flt(input logic enable, input int dst_fmt, input longint a, input longint b, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_fle(input logic enable, input int dst_fmt, input longint a, input longint b, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_feq(input logic enable, input int dst_fmt, input longint a, input longint b, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_fmin(input logic enable, input int dst_fmt, input longint a, input longint b, output longint result, output bit[4:0] fflags);
-import "DPI-C" function void dpi_fmax(input logic enable, input int dst_fmt, input longint a, input longint b, output longint result, output bit[4:0] fflags);
-
-import "DPI-C" function void dpi_hmma(input logic enable, input bit[3:0][1:0][31:0] A_tile, input bit[1:0][3:0][31:0] B_tile, input bit[3:0][3:0][31:0] C_tile, output bit[3:0][3:0][31:0] D_tile);
-import "DPI-C" function void dpi_print_results(input int wid, input int octet, input bit[3:0][1:0][31:0] A_tile, input bit[1:0][3:0][31:0] B_tile, input bit[3:0][3:0][31:0] C_tile, input bit[3:0][3:0][31:0] D_tile);
-
-`endif
+`endif

hw/dpi/util_dpi.cpp

@@ -1,58 +1,23 @@
-// Copyright © 2019-2023
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// 
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
 #include <stdio.h>
-#include <stdarg.h>
 #include <math.h>
 #include <unordered_map>
 #include <vector>
 #include <mutex>
-#include <memory>
 #include <iostream>
-
 #include "svdpi.h"
-// #include "verilated_vpi.h"
-
-#include "uuid_gen.h"
-
-#ifdef XLEN_64
-#define iword_t   int64_t
-#define uword_t   uint64_t
-#define idword_t  __int128_t
-#define udword_t  __uint128_t
-#else
-#define iword_t   int32_t
-#define uword_t   uint32_t
-#define idword_t  int64_t
-#define udword_t  uint64_t
-#endif
-
-#ifndef DEBUG_LEVEL
-#define DEBUG_LEVEL 3
-#endif
+#include "verilated_vpi.h"
+#include "VX_config.h"
 
 extern "C" {
-  void dpi_imul(bool enable, bool is_signed_a, bool is_signed_b, iword_t a, iword_t b, iword_t* resultl, iword_t* resulth);
-  void dpi_idiv(bool enable, bool is_signed, iword_t a, iword_t b, iword_t* quotient, iword_t* remainder);
+  void dpi_imul(bool enable, int a, int b, bool is_signed_a, bool is_signed_b, int* resultl, int* resulth);
+  void dpi_idiv(bool enable, int a, int b, bool is_signed, int* quotient, int* remainder);
 
   int dpi_register();
   void dpi_assert(int inst, bool cond, int delay);
 
-  void dpi_trace(int level, const char* format, ...);
+  void dpi_trace(const char* format, ...);
   void dpi_trace_start();
   void dpi_trace_stop();
-
-  uint64_t dpi_uuid_gen(bool reset, int wid, uint64_t PC);
 }
 
 bool sim_trace_enabled();
@@ -128,54 +93,49 @@ void dpi_assert(int inst, bool cond, int delay) {
   }
 }
 
-///////////////////////////////////////////////////////////////////////////////
-
-void dpi_imul(bool enable, bool is_signed_a, bool is_signed_b, iword_t a, iword_t b, iword_t* resultl, iword_t* resulth) {
+void dpi_imul(bool enable, int a, int b, bool is_signed_a, bool is_signed_b, int* resultl, int* resulth) {
   if (!enable)
     return;
-  udword_t first  = *(uword_t*)&a;
-  udword_t second = *(uword_t*)&b;
-
-  udword_t mask = udword_t(-1) << (8 * sizeof(iword_t));
     
-  if (is_signed_a && a < 0) {
-    first |= mask;
+  uint64_t first  = *(uint32_t*)&a;
+  uint64_t second = *(uint32_t*)&b;
+    
+  if (is_signed_a && (first & 0x80000000)) {
+    first |= 0xFFFFFFFF00000000;
   }
 
-  if (is_signed_b && b < 0) {
-    second |= mask;
+  if (is_signed_b && (second & 0x80000000)) {
+    second |= 0xFFFFFFFF00000000;
   }
 
-  udword_t result;
+  uint64_t result;
   if (is_signed_a || is_signed_b) {
-    result = idword_t(first) * idword_t(second);
+    result = (int64_t)first * (int64_t)second;
   } else {
     result = first * second;
-  }
-
-  *resultl = iword_t(result);
-  *resulth = iword_t(result >> (8 * sizeof(iword_t)));
+  }    
+    
+  *resultl = result & 0xFFFFFFFF;
+  *resulth = (result >> 32) & 0xFFFFFFFF;
 }
 
-void dpi_idiv(bool enable, bool is_signed, iword_t a, iword_t b, iword_t* quotient, iword_t* remainder) {
+void dpi_idiv(bool enable, int a, int b, bool is_signed, int* quotient, int* remainder) {
   if (!enable)
     return;
 
-  uword_t dividen = a;
-  uword_t divisor = b;
-
-  auto inf_neg = uword_t(1) << (8 * sizeof(iword_t) - 1);
+  uint32_t dividen = *(uint32_t*)&a;
+  uint32_t divisor = *(uint32_t*)&b;
 
   if (is_signed) {
     if (b == 0) {
       *quotient  = -1;
       *remainder = dividen;
-    } else if (dividen == inf_neg && divisor == -1) {
+    } else if (dividen == 0x80000000 && divisor == 0xffffffff) {
       *remainder = 0;
       *quotient  = dividen;
     } else { 
-      *quotient  = (iword_t)dividen / (iword_t)divisor;
-      *remainder = (iword_t)dividen % (iword_t)divisor;      
+      *quotient  = (int32_t)dividen / (int32_t)divisor;
+      *remainder = (int32_t)dividen % (int32_t)divisor;      
     }
   } else {    
     if (b == 0) {
@@ -188,13 +148,9 @@ void dpi_idiv(bool enable, bool is_signed, iword_t a, iword_t b, iword_t* quotie
   }
 }
 
-///////////////////////////////////////////////////////////////////////////////
-
-void dpi_trace(int level, const char* format, ...) { 
-  if (level > DEBUG_LEVEL)
+void dpi_trace(const char* format, ...) { 
+  if (!sim_trace_enabled())
     return;
-  // if (!sim_trace_enabled())
-  //   return;
   va_list va;
 	va_start(va, format);  
 	vprintf(format, va);
@@ -202,33 +158,9 @@ void dpi_trace(int level, const char* format, ...) {
 }
 
 void dpi_trace_start() { 
-  // sim_trace_enable(true);
+  sim_trace_enable(true);
 }
 
 void dpi_trace_stop() { 
-  // sim_trace_enable(false);
-}
-
-///////////////////////////////////////////////////////////////////////////////
-
-std::unordered_map<uint32_t, std::shared_ptr<vortex::UUIDGenerator>> g_uuid_gens;
-
-uint64_t dpi_uuid_gen(bool reset, int wid, uint64_t PC) {
-  if (reset) {
-    g_uuid_gens.clear();
-    return 0;
-  }
-  std::shared_ptr<vortex::UUIDGenerator> uuid_gen;
-  auto it = g_uuid_gens.find(wid);
-  if (it == g_uuid_gens.end()) {
-    uuid_gen = std::make_shared<vortex::UUIDGenerator>();
-    g_uuid_gens.emplace(wid, uuid_gen);
-  } else {
-    uuid_gen = it->second;
-  }
-  uint32_t instr_uuid = uuid_gen->get_uuid(PC);
-  uint32_t instr_id  = instr_uuid & 0xffff;
-  uint32_t instr_ref = instr_uuid >> 16;
-  uint64_t uuid = (uint64_t(instr_ref) << 32) | (wid << 16) | instr_id;
-  return uuid;
-}
+  sim_trace_enable(false);
+}

hw/dpi/util_dpi.vh

@@ -1,37 +1,14 @@
-// Copyright © 2019-2023
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// 
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
+`ifndef UTIL_DPI
+`define UTIL_DPI
 
-`ifndef UTIL_DPI_VH
-`define UTIL_DPI_VH
-
-`include "VX_config.vh"
-
-`ifdef XLEN_64
-`define INT_TYPE longint
-`else
-`define INT_TYPE int
-`endif
-
-import "DPI-C" function void dpi_imul(input logic enable, input logic is_signed_a, input logic is_signed_b, input `INT_TYPE a, input `INT_TYPE b, output `INT_TYPE resultl, output `INT_TYPE resulth);
-import "DPI-C" function void dpi_idiv(input logic enable, input logic is_signed, input `INT_TYPE a, input `INT_TYPE b, output `INT_TYPE quotient, output `INT_TYPE remainder);
+import "DPI-C" function void dpi_imul(input logic enable, input int a, input int b, input logic is_signed_a, input logic is_signed_b, output int resultl, output int resulth);
+import "DPI-C" function void dpi_idiv(input logic enable, input int a, input int b, input logic is_signed, output int quotient, output int remainder);
 
 import "DPI-C" function int dpi_register();
 import "DPI-C" function void dpi_assert(int inst, input logic cond, input int delay);
 
-import "DPI-C" function void dpi_trace(input int level, input string format /*verilator sformat*/);
+import "DPI-C" function void dpi_trace(input string format /*verilator sformat*/);
 import "DPI-C" function void dpi_trace_start();
 import "DPI-C" function void dpi_trace_stop();
 
-import "DPI-C" function longint dpi_uuid_gen(input logic reset, input int wid, input longint PC);
-
-`endif
+`endif

hw/rtl/.gitignore

@@ -0,0 +1 @@
+/VX_user_config.vh

hw/rtl/VX_alu_unit.sv

@@ -0,0 +1,235 @@
+`include "VX_define.vh"
+
+module VX_alu_unit #(
+    parameter CORE_ID = 0
+) (
+    input wire              clk,
+    input wire              reset,
+    
+    // Inputs
+    VX_alu_req_if.slave     alu_req_if,
+
+    // Outputs
+    VX_branch_ctl_if.master branch_ctl_if,
+    VX_commit_if.master     alu_commit_if    
+);   
+
+    `UNUSED_PARAM (CORE_ID)
+    
+    reg [`NUM_THREADS-1:0][31:0]  alu_result;    
+    wire [`NUM_THREADS-1:0][31:0] add_result;   
+    wire [`NUM_THREADS-1:0][32:0] sub_result;
+    wire [`NUM_THREADS-1:0][31:0] shr_result;
+    reg [`NUM_THREADS-1:0][31:0]  msc_result;
+
+    wire ready_in;    
+
+    `UNUSED_VAR (alu_req_if.op_mod)
+    wire                    is_br_op = `INST_ALU_IS_BR(alu_req_if.op_mod);
+    wire [`INST_ALU_BITS-1:0] alu_op = `INST_ALU_BITS'(alu_req_if.op_type);
+    wire [`INST_BR_BITS-1:0]   br_op = `INST_BR_BITS'(alu_req_if.op_type);
+    wire                  alu_signed = `INST_ALU_SIGNED(alu_op);   
+    wire [1:0]          alu_op_class = `INST_ALU_OP_CLASS(alu_op); 
+    wire                      is_sub = (alu_op == `INST_ALU_SUB);
+
+    wire [`NUM_THREADS-1:0][31:0] alu_in1 = alu_req_if.rs1_data;
+    wire [`NUM_THREADS-1:0][31:0] alu_in2 = alu_req_if.rs2_data;
+
+    wire [`NUM_THREADS-1:0][31:0] alu_in1_PC   = alu_req_if.use_PC ? {`NUM_THREADS{alu_req_if.PC}} : alu_in1;
+    wire [`NUM_THREADS-1:0][31:0] alu_in2_imm  = alu_req_if.use_imm ? {`NUM_THREADS{alu_req_if.imm}} : alu_in2;
+    wire [`NUM_THREADS-1:0][31:0] alu_in2_less = (alu_req_if.use_imm && ~is_br_op) ? {`NUM_THREADS{alu_req_if.imm}} : alu_in2;
+
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin
+        assign add_result[i] = alu_in1_PC[i] + alu_in2_imm[i];
+    end
+
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin
+        wire [32:0] sub_in1 = {alu_signed & alu_in1[i][31], alu_in1[i]};
+        wire [32:0] sub_in2 = {alu_signed & alu_in2_less[i][31], alu_in2_less[i]};
+        assign sub_result[i] = sub_in1 - sub_in2;
+    end
+
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin    
+        wire [32:0] shr_in1 = {alu_signed & alu_in1[i][31], alu_in1[i]};
+        assign shr_result[i] = 32'($signed(shr_in1) >>> alu_in2_imm[i][4:0]);
+    end        
+
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin 
+        always @(*) begin
+            case (alu_op)
+                `INST_ALU_AND: msc_result[i] = alu_in1[i] & alu_in2_imm[i];
+                `INST_ALU_OR:  msc_result[i] = alu_in1[i] | alu_in2_imm[i];
+                `INST_ALU_XOR: msc_result[i] = alu_in1[i] ^ alu_in2_imm[i];                
+                //`INST_ALU_SLL,
+                default:  msc_result[i] = alu_in1[i] << alu_in2_imm[i][4:0];
+            endcase
+        end
+    end
+            
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin
+        always @(*) begin
+            case (alu_op_class)                        
+                2'b00: alu_result[i] = add_result[i];               // ADD, LUI, AUIPC 
+                2'b01: alu_result[i] = {31'b0, sub_result[i][32]};  // SLTU, SLT
+                2'b10: alu_result[i] = is_sub ? sub_result[i][31:0] // SUB
+                                              : shr_result[i];      // SRL, SRA
+                // 2'b11,
+                default: alu_result[i] = msc_result[i];             // AND, OR, XOR, SLL
+            endcase
+        end       
+    end
+
+    // branch
+    
+    wire is_jal = is_br_op && (br_op == `INST_BR_JAL || br_op == `INST_BR_JALR);
+    wire [`NUM_THREADS-1:0][31:0] alu_jal_result = is_jal ? {`NUM_THREADS{alu_req_if.next_PC}} : alu_result; 
+
+    wire [31:0] br_dest    = add_result[alu_req_if.tid]; 
+    wire [32:0] cmp_result = sub_result[alu_req_if.tid];   
+    
+    wire is_less  = cmp_result[32];
+    wire is_equal = ~(| cmp_result[31:0]);        
+
+    // output
+
+    wire                          alu_valid_in;
+    wire                          alu_ready_in;
+    wire                          alu_valid_out;
+    wire                          alu_ready_out;
+    wire [`UUID_BITS-1:0]         alu_uuid;
+    wire [`NW_BITS-1:0]           alu_wid;
+    wire [`NUM_THREADS-1:0]       alu_tmask;
+    wire [31:0]                   alu_PC;
+    wire [`NR_BITS-1:0]           alu_rd;   
+    wire                          alu_wb; 
+    wire [`NUM_THREADS-1:0][31:0] alu_data;
+
+    wire [`INST_BR_BITS-1:0] br_op_r;
+    wire [31:0] br_dest_r;
+    wire is_less_r;
+    wire is_equal_r;
+    wire is_br_op_r;
+
+    assign alu_ready_in = alu_ready_out || ~alu_valid_out;
+
+    VX_pipe_register #(
+        .DATAW  (1 + `UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `NR_BITS + 1 + (`NUM_THREADS * 32) + 1 + `INST_BR_BITS + 1 + 1 + 32),
+        .RESETW (1)
+    ) pipe_reg (
+        .clk      (clk),
+        .reset    (reset),
+        .enable   (alu_ready_in),
+        .data_in  ({alu_valid_in,  alu_req_if.uuid, alu_req_if.wid, alu_req_if.tmask, alu_req_if.PC, alu_req_if.rd, alu_req_if.wb, alu_jal_result, is_br_op,   br_op,   is_less,   is_equal,   br_dest}),
+        .data_out ({alu_valid_out, alu_uuid,        alu_wid,        alu_tmask,        alu_PC,        alu_rd,        alu_wb,        alu_data,       is_br_op_r, br_op_r, is_less_r, is_equal_r, br_dest_r})
+    );
+
+    `UNUSED_VAR (br_op_r)
+    wire br_neg    = `INST_BR_NEG(br_op_r);
+    wire br_less   = `INST_BR_LESS(br_op_r);
+    wire br_static = `INST_BR_STATIC(br_op_r);
+
+    assign branch_ctl_if.valid = alu_valid_out && alu_ready_out && is_br_op_r;
+    assign branch_ctl_if.taken = ((br_less ? is_less_r : is_equal_r) ^ br_neg) | br_static;
+    assign branch_ctl_if.wid   = alu_wid;
+    assign branch_ctl_if.dest  = br_dest_r;
+
+`ifdef EXT_M_ENABLE
+
+    wire                          mul_valid_in;
+    wire                          mul_ready_in;
+    wire                          mul_valid_out;    
+    wire                          mul_ready_out;
+    wire [`UUID_BITS-1:0]         mul_uuid;
+    wire [`NW_BITS-1:0]           mul_wid;
+    wire [`NUM_THREADS-1:0]       mul_tmask;
+    wire [31:0]                   mul_PC;
+    wire [`NR_BITS-1:0]           mul_rd;
+    wire                          mul_wb;
+    wire [`NUM_THREADS-1:0][31:0] mul_data;
+
+    wire [`INST_MUL_BITS-1:0] mul_op = `INST_MUL_BITS'(alu_req_if.op_type);
+
+    VX_muldiv muldiv (
+        .clk        (clk),
+        .reset      (reset),
+        
+        // Inputs
+        .alu_op     (mul_op),
+        .uuid_in    (alu_req_if.uuid),
+        .wid_in     (alu_req_if.wid),
+        .tmask_in   (alu_req_if.tmask),
+        .PC_in      (alu_req_if.PC),
+        .rd_in      (alu_req_if.rd),
+        .wb_in      (alu_req_if.wb),
+        .alu_in1    (alu_req_if.rs1_data), 
+        .alu_in2    (alu_req_if.rs2_data),
+
+        // Outputs
+        .wid_out    (mul_wid),
+        .uuid_out   (mul_uuid),
+        .tmask_out  (mul_tmask),
+        .PC_out     (mul_PC),
+        .rd_out     (mul_rd),
+        .wb_out     (mul_wb),
+        .data_out   (mul_data),
+
+        // handshake
+        .valid_in   (mul_valid_in),
+        .ready_in   (mul_ready_in),
+        .valid_out  (mul_valid_out),
+        .ready_out  (mul_ready_out)
+    );
+
+    wire is_mul_op = `INST_ALU_IS_MUL(alu_req_if.op_mod);
+
+    assign ready_in = is_mul_op ? mul_ready_in : alu_ready_in;
+
+    assign alu_valid_in = alu_req_if.valid && ~is_mul_op;
+    assign mul_valid_in = alu_req_if.valid && is_mul_op;
+
+    assign alu_commit_if.valid = alu_valid_out || mul_valid_out;
+    assign alu_commit_if.uuid  = alu_valid_out ? alu_uuid  : mul_uuid;
+    assign alu_commit_if.wid   = alu_valid_out ? alu_wid   : mul_wid;
+    assign alu_commit_if.tmask = alu_valid_out ? alu_tmask : mul_tmask;
+    assign alu_commit_if.PC    = alu_valid_out ? alu_PC    : mul_PC;
+    assign alu_commit_if.rd    = alu_valid_out ? alu_rd    : mul_rd;
+    assign alu_commit_if.wb    = alu_valid_out ? alu_wb    : mul_wb;
+    assign alu_commit_if.data  = alu_valid_out ? alu_data  : mul_data;
+
+    assign alu_ready_out = alu_commit_if.ready;
+    assign mul_ready_out = alu_commit_if.ready & ~alu_valid_out; // ALU takes priority
+
+`else 
+
+    assign ready_in = alu_ready_in;
+
+    assign alu_valid_in = alu_req_if.valid;
+
+    assign alu_commit_if.valid = alu_valid_out;
+    assign alu_commit_if.uuid  = alu_uuid;
+    assign alu_commit_if.wid   = alu_wid;
+    assign alu_commit_if.tmask = alu_tmask;
+    assign alu_commit_if.PC    = alu_PC; 
+    assign alu_commit_if.rd    = alu_rd;    
+    assign alu_commit_if.wb    = alu_wb;
+    assign alu_commit_if.data  = alu_data;
+
+    assign alu_ready_out = alu_commit_if.ready;
+
+`endif
+
+    assign alu_commit_if.eop = 1'b1;
+
+    // can accept new request?
+    assign alu_req_if.ready = ready_in;
+
+`ifdef DBG_TRACE_CORE_PIPELINE
+    always @(posedge clk) begin
+        if (branch_ctl_if.valid) begin
+            dpi_trace("%d: core%0d-branch: wid=%0d, PC=%0h, taken=%b, dest=%0h (#%0d)\n", 
+                $time, CORE_ID, branch_ctl_if.wid, alu_commit_if.PC, branch_ctl_if.taken, branch_ctl_if.dest, alu_uuid);
+        end
+    end
+`endif
+
+endmodule

hw/rtl/VX_cache_arb.sv

@@ -0,0 +1,159 @@
+`include "VX_define.vh"
+
+module VX_cache_arb #(    
+    parameter NUM_REQS      = 1, 
+    parameter LANES         = 1,
+    parameter DATA_SIZE     = 1,
+    parameter TAG_IN_WIDTH  = 1,
+    parameter TAG_SEL_IDX   = 0,   
+    parameter BUFFERED_REQ  = 0,
+    parameter BUFFERED_RSP  = 0,
+    parameter TYPE          = "R",
+
+    localparam ADDR_WIDTH   = (32-`CLOG2(DATA_SIZE)),
+    localparam DATA_WIDTH   = (8 * DATA_SIZE),
+    localparam LOG_NUM_REQS = `CLOG2(NUM_REQS),
+    localparam TAG_OUT_WIDTH = TAG_IN_WIDTH + LOG_NUM_REQS
+) (
+    input wire clk,
+    input wire reset,
+
+    // input requests    
+    input wire [NUM_REQS-1:0][LANES-1:0]                    req_valid_in, 
+    input wire [NUM_REQS-1:0][LANES-1:0]                    req_rw_in,   
+    input wire [NUM_REQS-1:0][LANES-1:0][DATA_SIZE-1:0]     req_byteen_in, 
+    input wire [NUM_REQS-1:0][LANES-1:0][ADDR_WIDTH-1:0]    req_addr_in, 
+    input wire [NUM_REQS-1:0][LANES-1:0][DATA_WIDTH-1:0]    req_data_in,    
+    input wire [NUM_REQS-1:0][LANES-1:0][TAG_IN_WIDTH-1:0]  req_tag_in,  
+    output wire [NUM_REQS-1:0][LANES-1:0]                   req_ready_in,
+
+    // output request
+    output wire [LANES-1:0]                                 req_valid_out,
+    output wire [LANES-1:0]                                 req_rw_out,  
+    output wire [LANES-1:0][DATA_SIZE-1:0]                  req_byteen_out,  
+    output wire [LANES-1:0][ADDR_WIDTH-1:0]                 req_addr_out, 
+    output wire [LANES-1:0][DATA_WIDTH-1:0]                 req_data_out,   
+    output wire [LANES-1:0][TAG_OUT_WIDTH-1:0]              req_tag_out,    
+    input wire  [LANES-1:0]                                 req_ready_out,
+
+    // input response
+    input wire                                              rsp_valid_in,    
+    input wire [LANES-1:0]                                  rsp_tmask_in,    
+    input wire [LANES-1:0][DATA_WIDTH-1:0]                  rsp_data_in,
+    input wire [TAG_OUT_WIDTH-1:0]                          rsp_tag_in,
+    output wire                                             rsp_ready_in,
+
+    // output responses
+    output wire [NUM_REQS-1:0]                              rsp_valid_out,
+    output wire [NUM_REQS-1:0][LANES-1:0]                   rsp_tmask_out,
+    output wire [NUM_REQS-1:0][LANES-1:0][DATA_WIDTH-1:0]   rsp_data_out,
+    output wire [NUM_REQS-1:0][TAG_IN_WIDTH-1:0]            rsp_tag_out,
+    input wire  [NUM_REQS-1:0]                              rsp_ready_out    
+);  
+    localparam REQ_DATAW = TAG_OUT_WIDTH + ADDR_WIDTH + 1 + DATA_SIZE + DATA_WIDTH;
+    localparam RSP_DATAW = LANES * (1 + DATA_WIDTH) + TAG_IN_WIDTH;
+
+    if (NUM_REQS > 1) begin
+
+        wire [NUM_REQS-1:0][LANES-1:0][REQ_DATAW-1:0] req_data_in_merged;
+        wire [LANES-1:0][REQ_DATAW-1:0] req_data_out_merged;
+
+        for (genvar i = 0; i < NUM_REQS; i++) begin
+            for (genvar j = 0; j < LANES; ++j) begin            
+                wire [TAG_OUT_WIDTH-1:0] req_tag_in_w;
+
+                VX_bits_insert #( 
+                    .N   (TAG_IN_WIDTH),
+                    .S   (LOG_NUM_REQS),
+                    .POS (TAG_SEL_IDX)
+                ) bits_insert (
+                    .data_in  (req_tag_in[i][j]),
+                    .sel_in   (LOG_NUM_REQS'(i)),
+                    .data_out (req_tag_in_w)
+                );
+
+                assign req_data_in_merged[i][j] = {req_tag_in_w, req_addr_in[i][j], req_rw_in[i][j], req_byteen_in[i][j], req_data_in[i][j]};
+            end
+        end
+
+        VX_stream_arbiter #(            
+            .NUM_REQS (NUM_REQS),
+            .LANES    (LANES),
+            .DATAW    (REQ_DATAW),
+            .BUFFERED (BUFFERED_REQ),
+            .TYPE     (TYPE)
+        ) req_arb (
+            .clk       (clk),
+            .reset     (reset),
+            .valid_in  (req_valid_in),
+            .data_in   (req_data_in_merged),
+            .ready_in  (req_ready_in),
+            .valid_out (req_valid_out),
+            .data_out  (req_data_out_merged),
+            .ready_out (req_ready_out)
+        );
+
+        for (genvar i = 0; i < LANES; ++i) begin
+            assign {req_tag_out[i], req_addr_out[i], req_rw_out[i], req_byteen_out[i], req_data_out[i]} = req_data_out_merged[i];
+        end
+
+        ///////////////////////////////////////////////////////////////////////
+
+        wire [NUM_REQS-1:0][RSP_DATAW-1:0] rsp_data_out_merged;
+
+        wire [LOG_NUM_REQS-1:0] rsp_sel = rsp_tag_in[TAG_SEL_IDX +: LOG_NUM_REQS];
+
+        wire [TAG_IN_WIDTH-1:0] rsp_tag_in_w;
+
+        VX_bits_remove #( 
+            .N   (TAG_OUT_WIDTH),
+            .S   (LOG_NUM_REQS),
+            .POS (TAG_SEL_IDX)
+        ) bits_remove (
+            .data_in  (rsp_tag_in),
+            .data_out (rsp_tag_in_w)
+        );
+
+        VX_stream_demux #(
+            .NUM_REQS (NUM_REQS),
+            .LANES    (1),
+            .DATAW    (RSP_DATAW),
+            .BUFFERED (BUFFERED_RSP)
+        ) rsp_demux (
+            .clk       (clk),
+            .reset     (reset),
+            .sel_in    (rsp_sel),
+            .valid_in  (rsp_valid_in),
+            .data_in   ({rsp_tmask_in, rsp_tag_in_w, rsp_data_in}),
+            .ready_in  (rsp_ready_in),
+            .valid_out (rsp_valid_out),
+            .data_out  (rsp_data_out_merged),
+            .ready_out (rsp_ready_out)
+        );
+        
+        for (genvar i = 0; i < NUM_REQS; i++) begin
+            assign {rsp_tmask_out[i], rsp_tag_out[i], rsp_data_out[i]} = rsp_data_out_merged[i];
+        end
+
+    end else begin
+
+        `UNUSED_VAR (clk)
+        `UNUSED_VAR (reset)
+
+        assign req_valid_out  = req_valid_in;
+        assign req_tag_out    = req_tag_in;
+        assign req_addr_out   = req_addr_in;
+        assign req_rw_out     = req_rw_in;
+        assign req_byteen_out = req_byteen_in;
+        assign req_data_out   = req_data_in;
+        assign req_ready_in   = req_ready_out;
+
+        assign rsp_valid_out  = rsp_valid_in;
+        assign rsp_tmask_out  = rsp_tmask_in;
+        assign rsp_tag_out    = rsp_tag_in;
+        assign rsp_data_out   = rsp_data_in;
+        assign rsp_ready_in   = rsp_ready_out;
+
+    end
+
+endmodule

hw/rtl/VX_cluster.sv

@@ -1,171 +1,195 @@
-// Copyright © 2019-2023
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// 
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
 `include "VX_define.vh"
 
-module VX_cluster import VX_gpu_pkg::*; #(
+module VX_cluster #(
     parameter CLUSTER_ID = 0
 ) ( 
-    `SCOPE_IO_DECL
+    `SCOPE_IO_VX_cluster
 
     // Clock
-    input  wire                 clk,
-    input  wire                 reset,
+    input  wire                             clk,
+    input  wire                             reset,
 
-`ifdef PERF_ENABLE
-    VX_mem_perf_if.slave        mem_perf_if,
-`endif
+    // Memory request
+    output wire                             mem_req_valid,
+    output wire                             mem_req_rw,    
+    output wire [`L2_MEM_BYTEEN_WIDTH-1:0]  mem_req_byteen,    
+    output wire [`L2_MEM_ADDR_WIDTH-1:0]    mem_req_addr,
+    output wire [`L2_MEM_DATA_WIDTH-1:0]    mem_req_data,
+    output wire [`L2_MEM_TAG_WIDTH-1:0]     mem_req_tag,
+    input  wire                             mem_req_ready,
 
-    // DCRs
-    VX_dcr_bus_if.slave         dcr_bus_if,
-
-    // Memory
-    VX_mem_bus_if.master        mem_bus_if,
-
-    // simulation helper signals
-    output wire                 sim_ebreak,
-    output wire [`NUM_REGS-1:0][`XLEN-1:0] sim_wb_value,
+    // Memory response    
+    input wire                              mem_rsp_valid,        
+    input wire [`L2_MEM_DATA_WIDTH-1:0]     mem_rsp_data,
+    input wire [`L2_MEM_TAG_WIDTH-1:0]      mem_rsp_tag,
+    output wire                             mem_rsp_ready,
 
     // Status
-    output wire                 busy
-);
+    output wire                             busy
+); 
+    `STATIC_ASSERT((`L2_ENABLE == 0 || `NUM_CORES > 1), ("invalid parameter"))
 
-`ifdef SCOPE
-    localparam scope_socket = 0;
-    `SCOPE_IO_SWITCH (scope_socket + `NUM_SOCKETS);
-`endif    
+    wire [`NUM_CORES-1:0]                       per_core_mem_req_valid;
+    wire [`NUM_CORES-1:0]                       per_core_mem_req_rw;    
+    wire [`NUM_CORES-1:0][`DCACHE_MEM_BYTEEN_WIDTH-1:0] per_core_mem_req_byteen;    
+    wire [`NUM_CORES-1:0][`DCACHE_MEM_ADDR_WIDTH-1:0] per_core_mem_req_addr;
+    wire [`NUM_CORES-1:0][`DCACHE_MEM_DATA_WIDTH-1:0] per_core_mem_req_data;
+    wire [`NUM_CORES-1:0][`L1_MEM_TAG_WIDTH-1:0] per_core_mem_req_tag;
+    wire [`NUM_CORES-1:0]                       per_core_mem_req_ready;
 
-`ifdef PERF_ENABLE
-    VX_mem_perf_if mem_perf_tmp_if();    
-    assign mem_perf_tmp_if.icache  = 'x;
-    assign mem_perf_tmp_if.dcache  = 'x;
-    assign mem_perf_tmp_if.l3cache = mem_perf_if.l3cache;
-    assign mem_perf_tmp_if.smem    = 'x;
-    assign mem_perf_tmp_if.mem     = mem_perf_if.mem;
-`endif    
+    wire [`NUM_CORES-1:0]                       per_core_mem_rsp_valid;            
+    wire [`NUM_CORES-1:0][`DCACHE_MEM_DATA_WIDTH-1:0] per_core_mem_rsp_data;
+    wire [`NUM_CORES-1:0][`L1_MEM_TAG_WIDTH-1:0] per_core_mem_rsp_tag;
+    wire [`NUM_CORES-1:0]                       per_core_mem_rsp_ready;
 
-`ifdef GBAR_ENABLE
+    wire [`NUM_CORES-1:0]                       per_core_busy;
 
-    VX_gbar_bus_if per_socket_gbar_bus_if[`NUM_SOCKETS]();
-    VX_gbar_bus_if gbar_bus_if();
+    for (genvar i = 0; i < `NUM_CORES; i++) begin
 
-    `RESET_RELAY (gbar_reset, reset);
+        `RESET_RELAY (core_reset);
 
-    VX_gbar_arb #(
-        .NUM_REQS (`NUM_SOCKETS),
-        .OUT_REG  ((`NUM_SOCKETS > 2) ? 1 : 0) // bgar_unit has no backpressure
-    ) gbar_arb (
-        .clk        (clk),
-        .reset      (gbar_reset),
-        .bus_in_if  (per_socket_gbar_bus_if),
-        .bus_out_if (gbar_bus_if)
-    );
-
-    VX_gbar_unit #(
-        .INSTANCE_ID ($sformatf("gbar%0d", CLUSTER_ID))
-    ) gbar_unit (
-        .clk         (clk),
-        .reset       (gbar_reset),
-        .gbar_bus_if (gbar_bus_if)
-    );
-
-`endif
-
-    VX_mem_bus_if #(
-        .DATA_SIZE (`L1_LINE_SIZE),
-        .TAG_WIDTH (L1_MEM_ARB_TAG_WIDTH)
-    ) per_socket_mem_bus_if[`NUM_SOCKETS]();
-
-    `RESET_RELAY (l2_reset, reset);
-
-    VX_cache_wrap #(
-        .INSTANCE_ID    ("l2cache"),
-        .CACHE_SIZE     (`L2_CACHE_SIZE),
-        .LINE_SIZE      (`L2_LINE_SIZE),
-        .NUM_BANKS      (`L2_NUM_BANKS),
-        .NUM_WAYS       (`L2_NUM_WAYS),
-        .WORD_SIZE      (L2_WORD_SIZE),
-        .NUM_REQS       (L2_NUM_REQS),
-        .CRSQ_SIZE      (`L2_CRSQ_SIZE),
-        .MSHR_SIZE      (`L2_MSHR_SIZE),
-        .MRSQ_SIZE      (`L2_MRSQ_SIZE),
-        .MREQ_SIZE      (`L2_MREQ_SIZE),
-        .TAG_WIDTH      (L2_TAG_WIDTH),
-        .WRITE_ENABLE   (1),
-        .UUID_WIDTH     (`UUID_WIDTH),  
-        .CORE_OUT_REG   (2),
-        .MEM_OUT_REG    (2),
-        .NC_ENABLE      (1),
-        .PASSTHRU       (!`L2_ENABLED)
-    ) l2cache (
-        .clk            (clk),
-        .reset          (l2_reset),
-    `ifdef PERF_ENABLE
-        .cache_perf     (mem_perf_tmp_if.l2cache),
-    `endif
-        .core_bus_if    (per_socket_mem_bus_if),
-        .mem_bus_if     (mem_bus_if)
-    );
-
-    ///////////////////////////////////////////////////////////////////////////
-
-    wire [`NUM_SOCKETS-1:0] per_socket_sim_ebreak;
-    wire [`NUM_SOCKETS-1:0][`NUM_REGS-1:0][`XLEN-1:0] per_socket_sim_wb_value;
-    assign sim_ebreak = per_socket_sim_ebreak[0];
-    assign sim_wb_value = per_socket_sim_wb_value[0];
-    `UNUSED_VAR (per_socket_sim_ebreak)
-    `UNUSED_VAR (per_socket_sim_wb_value)
-
-    VX_dcr_bus_if socket_dcr_bus_tmp_if();
-    assign socket_dcr_bus_tmp_if.write_valid = dcr_bus_if.write_valid && (dcr_bus_if.write_addr >= `VX_DCR_BASE_STATE_BEGIN && dcr_bus_if.write_addr < `VX_DCR_BASE_STATE_END);
-    assign socket_dcr_bus_tmp_if.write_addr  = dcr_bus_if.write_addr;
-    assign socket_dcr_bus_tmp_if.write_data  = dcr_bus_if.write_data;
-
-    wire [`NUM_SOCKETS-1:0] per_socket_busy;
-
-    `BUFFER_DCR_BUS_IF (socket_dcr_bus_if, socket_dcr_bus_tmp_if, (`NUM_SOCKETS > 1));
-
-    // Generate all sockets
-    for (genvar i = 0; i < `NUM_SOCKETS; ++i) begin
-
-        `RESET_RELAY (socket_reset, reset);
-
-        VX_socket #(
-            .SOCKET_ID ((CLUSTER_ID * `NUM_SOCKETS) + i)
-        ) socket (
-            `SCOPE_IO_BIND  (scope_socket+i)
+        VX_core #(
+            .CORE_ID(i + (CLUSTER_ID * `NUM_CORES))
+        ) core (
+            `SCOPE_BIND_VX_cluster_core(i)
 
             .clk            (clk),
-            .reset          (socket_reset),
+            .reset          (core_reset),
 
-        `ifdef PERF_ENABLE
-            .mem_perf_if    (mem_perf_tmp_if),
-        `endif
-            
-            .dcr_bus_if     (socket_dcr_bus_if),
+            .mem_req_valid  (per_core_mem_req_valid[i]),
+            .mem_req_rw     (per_core_mem_req_rw   [i]),                
+            .mem_req_byteen (per_core_mem_req_byteen[i]),                
+            .mem_req_addr   (per_core_mem_req_addr [i]),
+            .mem_req_data   (per_core_mem_req_data [i]),
+            .mem_req_tag    (per_core_mem_req_tag  [i]),
+            .mem_req_ready  (per_core_mem_req_ready[i]),
+                     
+            .mem_rsp_valid  (per_core_mem_rsp_valid[i]),                
+            .mem_rsp_data   (per_core_mem_rsp_data [i]),
+            .mem_rsp_tag    (per_core_mem_rsp_tag  [i]),
+            .mem_rsp_ready  (per_core_mem_rsp_ready[i]),
 
-            .mem_bus_if     (per_socket_mem_bus_if[i]),
-
-        `ifdef GBAR_ENABLE
-            .gbar_bus_if    (per_socket_gbar_bus_if[i]),
-        `endif
-
-            .sim_ebreak     (per_socket_sim_ebreak[i]),
-            .sim_wb_value   (per_socket_sim_wb_value[i]),
-            .busy           (per_socket_busy[i])
+            .busy           (per_core_busy         [i])
         );
     end
+    
+    assign busy = (| per_core_busy);
 
-    `BUFFER_EX(busy, (| per_socket_busy), 1'b1, (`NUM_SOCKETS > 1));
+    if (`L2_ENABLE) begin
+    `ifdef PERF_ENABLE
+        VX_perf_cache_if perf_l2cache_if();
+    `endif
+
+        `RESET_RELAY (l2_reset);
+
+        VX_cache #(
+            .CACHE_ID           (`L2_CACHE_ID),
+            .CACHE_SIZE         (`L2_CACHE_SIZE),
+            .CACHE_LINE_SIZE    (`L2_CACHE_LINE_SIZE),
+            .NUM_BANKS          (`L2_NUM_BANKS),
+            .NUM_PORTS          (`L2_NUM_PORTS),
+            .WORD_SIZE          (`L2_WORD_SIZE),
+            .NUM_REQS           (`L2_NUM_REQS),
+            .CREQ_SIZE          (`L2_CREQ_SIZE),
+            .CRSQ_SIZE          (`L2_CRSQ_SIZE),
+            .MSHR_SIZE          (`L2_MSHR_SIZE),
+            .MRSQ_SIZE          (`L2_MRSQ_SIZE),
+            .MREQ_SIZE          (`L2_MREQ_SIZE),
+            .WRITE_ENABLE       (1),          
+            .CORE_TAG_WIDTH     (`L1_MEM_TAG_WIDTH),
+            .CORE_TAG_ID_BITS   (0),
+            .MEM_TAG_WIDTH      (`L2_MEM_TAG_WIDTH),
+            .NC_ENABLE          (1)
+        ) l2cache (
+            `SCOPE_BIND_VX_cluster_l2cache
+              
+            .clk                (clk),
+            .reset              (l2_reset),
+
+        `ifdef PERF_ENABLE
+            .perf_cache_if      (perf_l2cache_if),
+        `endif
+
+            // Core request
+            .core_req_valid     (per_core_mem_req_valid),
+            .core_req_rw        (per_core_mem_req_rw),
+            .core_req_byteen    (per_core_mem_req_byteen),
+            .core_req_addr      (per_core_mem_req_addr),
+            .core_req_data      (per_core_mem_req_data),  
+            .core_req_tag       (per_core_mem_req_tag),  
+            .core_req_ready     (per_core_mem_req_ready),
+
+            // Core response
+            .core_rsp_valid     (per_core_mem_rsp_valid),
+            .core_rsp_data      (per_core_mem_rsp_data),
+            .core_rsp_tag       (per_core_mem_rsp_tag),
+            .core_rsp_ready     (per_core_mem_rsp_ready),
+            `UNUSED_PIN (core_rsp_tmask),
+
+            // Memory request
+            .mem_req_valid      (mem_req_valid),
+            .mem_req_rw         (mem_req_rw),        
+            .mem_req_byteen     (mem_req_byteen),
+            .mem_req_addr       (mem_req_addr),
+            .mem_req_data       (mem_req_data),
+            .mem_req_tag        (mem_req_tag),
+            .mem_req_ready      (mem_req_ready),
+            
+            // Memory response
+            .mem_rsp_valid      (mem_rsp_valid),
+            .mem_rsp_tag        (mem_rsp_tag),
+            .mem_rsp_data       (mem_rsp_data),
+            .mem_rsp_ready      (mem_rsp_ready)
+        );
+
+    end else begin
+
+        `RESET_RELAY (mem_arb_reset);
+
+        VX_mem_arb #(
+            .NUM_REQS     (`NUM_CORES),
+            .DATA_WIDTH   (`DCACHE_MEM_DATA_WIDTH),
+            .ADDR_WIDTH   (`DCACHE_MEM_ADDR_WIDTH),           
+            .TAG_IN_WIDTH (`L1_MEM_TAG_WIDTH),            
+            .TYPE         ("R"),
+            .TAG_SEL_IDX  (1), // Skip 0 for NC flag
+            .BUFFERED_REQ (1),
+            .BUFFERED_RSP (1)
+        ) mem_arb (
+            .clk            (clk),
+            .reset          (mem_arb_reset),
+
+            // Core request
+            .req_valid_in   (per_core_mem_req_valid),
+            .req_rw_in      (per_core_mem_req_rw),
+            .req_byteen_in  (per_core_mem_req_byteen),
+            .req_addr_in    (per_core_mem_req_addr),
+            .req_data_in    (per_core_mem_req_data),  
+            .req_tag_in     (per_core_mem_req_tag),  
+            .req_ready_in   (per_core_mem_req_ready),
+
+            // Memory request
+            .req_valid_out  (mem_req_valid),
+            .req_rw_out     (mem_req_rw),        
+            .req_byteen_out (mem_req_byteen),        
+            .req_addr_out   (mem_req_addr),
+            .req_data_out   (mem_req_data),
+            .req_tag_out    (mem_req_tag),
+            .req_ready_out  (mem_req_ready),
+
+            // Core response
+            .rsp_valid_out  (per_core_mem_rsp_valid),
+            .rsp_data_out   (per_core_mem_rsp_data),
+            .rsp_tag_out    (per_core_mem_rsp_tag),
+            .rsp_ready_out  (per_core_mem_rsp_ready),
+            
+            // Memory response
+            .rsp_valid_in   (mem_rsp_valid),
+            .rsp_tag_in     (mem_rsp_tag),
+            .rsp_data_in    (mem_rsp_data),
+            .rsp_ready_in   (mem_rsp_ready)
+        );
+
+    end
 
 endmodule

hw/rtl/VX_commit.sv

@@ -0,0 +1,138 @@
+`include "VX_define.vh"
+
+module VX_commit #(
+    parameter CORE_ID = 0
+) (
+    input wire              clk,
+    input wire              reset,
+
+    // inputs
+    VX_commit_if.slave      alu_commit_if,
+    VX_commit_if.slave      ld_commit_if,
+    VX_commit_if.slave      st_commit_if, 
+    VX_commit_if.slave      csr_commit_if,
+`ifdef EXT_F_ENABLE
+    VX_commit_if.slave      fpu_commit_if,
+`endif
+    VX_commit_if.slave      gpu_commit_if,
+
+    // outputs
+    VX_writeback_if.master  writeback_if,
+    VX_cmt_to_csr_if.master cmt_to_csr_if
+);
+    // CSRs update
+
+    wire alu_commit_fire = alu_commit_if.valid && alu_commit_if.ready;
+    wire ld_commit_fire  = ld_commit_if.valid && ld_commit_if.ready;
+    wire st_commit_fire  = st_commit_if.valid && st_commit_if.ready;
+    wire csr_commit_fire = csr_commit_if.valid && csr_commit_if.ready;
+`ifdef EXT_F_ENABLE
+    wire fpu_commit_fire = fpu_commit_if.valid && fpu_commit_if.ready;
+`endif
+    wire gpu_commit_fire = gpu_commit_if.valid && gpu_commit_if.ready;
+
+    wire commit_fire = alu_commit_fire
+                    || ld_commit_fire
+                    || st_commit_fire
+                    || csr_commit_fire
+                `ifdef EXT_F_ENABLE
+                    || fpu_commit_fire
+                `endif
+                    || gpu_commit_fire;
+
+`ifdef EXT_F_ENABLE
+    wire [(6*`NUM_THREADS)-1:0] commit_tmask;
+`else
+    wire [(5*`NUM_THREADS)-1:0] commit_tmask;
+`endif
+
+    wire [$clog2($bits(commit_tmask)+1)-1:0] commit_size;
+
+    assign commit_tmask = {
+        {`NUM_THREADS{alu_commit_fire}} & alu_commit_if.tmask,
+        {`NUM_THREADS{ld_commit_fire}}  & ld_commit_if.tmask, 
+        {`NUM_THREADS{st_commit_fire}}  & st_commit_if.tmask,
+        {`NUM_THREADS{csr_commit_fire}} & csr_commit_if.tmask,
+    `ifdef EXT_F_ENABLE
+        {`NUM_THREADS{fpu_commit_fire}} & fpu_commit_if.tmask,
+    `endif
+        {`NUM_THREADS{gpu_commit_fire}} & gpu_commit_if.tmask
+    };
+    
+    `POP_COUNT(commit_size, commit_tmask);
+
+    VX_pipe_register #(
+        .DATAW  (1 + $bits(commit_size)),
+        .RESETW (1)
+    ) pipe_reg (
+        .clk      (clk),
+        .reset    (reset),
+        .enable   (1'b1),
+        .data_in  ({commit_fire,         commit_size}),
+        .data_out ({cmt_to_csr_if.valid, cmt_to_csr_if.commit_size})
+    );
+
+    // Writeback
+
+    VX_writeback #(
+        .CORE_ID(CORE_ID)
+    ) writeback (
+        .clk            (clk),
+        .reset          (reset),
+
+        .alu_commit_if  (alu_commit_if),
+        .ld_commit_if   (ld_commit_if),        
+        .csr_commit_if  (csr_commit_if),
+    `ifdef EXT_F_ENABLE
+        .fpu_commit_if  (fpu_commit_if),
+    `endif
+        .gpu_commit_if  (gpu_commit_if),
+        .writeback_if   (writeback_if)
+    );
+
+    // store and gpu commits don't writeback  
+    assign st_commit_if.ready  = 1'b1;
+
+`ifdef DBG_TRACE_CORE_PIPELINE
+    always @(posedge clk) begin
+        if (alu_commit_if.valid && alu_commit_if.ready) begin
+             dpi_trace("%d: core%0d-commit: wid=%0d, PC=%0h, ex=ALU, tmask=%b, wb=%0d, rd=%0d, data=", $time, CORE_ID, alu_commit_if.wid, alu_commit_if.PC, alu_commit_if.tmask, alu_commit_if.wb, alu_commit_if.rd);
+            `TRACE_ARRAY1D(alu_commit_if.data, `NUM_THREADS);
+             dpi_trace(" (#%0d)\n", alu_commit_if.uuid);
+        end
+        if (ld_commit_if.valid && ld_commit_if.ready) begin
+             dpi_trace("%d: core%0d-commit: wid=%0d, PC=%0h, ex=LSU, tmask=%b, wb=%0d, rd=%0d, data=", $time, CORE_ID, ld_commit_if.wid, ld_commit_if.PC, ld_commit_if.tmask, ld_commit_if.wb, ld_commit_if.rd);
+            `TRACE_ARRAY1D(ld_commit_if.data, `NUM_THREADS);
+             dpi_trace(" (#%0d)\n", ld_commit_if.uuid);
+        end
+        if (st_commit_if.valid && st_commit_if.ready) begin
+            dpi_trace("%d: core%0d-commit: wid=%0d, PC=%0h, ex=LSU, tmask=%b, wb=%0d, rd=%0d (#%0d)\n", $time, CORE_ID, st_commit_if.wid, st_commit_if.PC, st_commit_if.tmask, st_commit_if.wb, st_commit_if.rd, st_commit_if.uuid);
+        end
+        if (csr_commit_if.valid && csr_commit_if.ready) begin
+             dpi_trace("%d: core%0d-commit: wid=%0d, PC=%0h, ex=CSR, tmask=%b, wb=%0d, rd=%0d, data=", $time, CORE_ID, csr_commit_if.wid, csr_commit_if.PC, csr_commit_if.tmask, csr_commit_if.wb, csr_commit_if.rd);
+            `TRACE_ARRAY1D(csr_commit_if.data, `NUM_THREADS);
+             dpi_trace(" (#%0d)\n", csr_commit_if.uuid);
+        end      
+    `ifdef EXT_F_ENABLE
+        if (fpu_commit_if.valid && fpu_commit_if.ready) begin
+             dpi_trace("%d: core%0d-commit: wid=%0d, PC=%0h, ex=FPU, tmask=%b, wb=%0d, rd=%0d, data=", $time, CORE_ID, fpu_commit_if.wid, fpu_commit_if.PC, fpu_commit_if.tmask, fpu_commit_if.wb, fpu_commit_if.rd);
+            `TRACE_ARRAY1D(fpu_commit_if.data, `NUM_THREADS);
+             dpi_trace(" (#%0d)\n", fpu_commit_if.uuid);
+        end
+    `endif
+        if (gpu_commit_if.valid && gpu_commit_if.ready) begin
+             dpi_trace("%d: core%0d-commit: wid=%0d, PC=%0h, ex=GPU, tmask=%b, wb=%0d, rd=%0d, data=", $time, CORE_ID, gpu_commit_if.wid, gpu_commit_if.PC, gpu_commit_if.tmask, gpu_commit_if.wb, gpu_commit_if.rd);
+            `TRACE_ARRAY1D(gpu_commit_if.data, `NUM_THREADS);
+             dpi_trace(" (#%0d)\n", gpu_commit_if.uuid);
+        end
+    end
+`endif
+
+endmodule
+
+
+
+
+
+
+

hw/rtl/VX_config.vh

@@ -1,88 +1,10 @@
-// Copyright © 2019-2023
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// 
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
+`ifndef VX_CONFIG
+`define VX_CONFIG
 
-`ifndef VX_CONFIG_VH
-`define VX_CONFIG_VH
-
-`ifndef MIN
-`define MIN(x, y)   (((x) < (y)) ? (x) : (y))
-`endif
-
-`ifndef MAX
-`define MAX(x, y)   (((x) > (y)) ? (x) : (y))
-`endif
-
-`ifndef CLAMP
-`define CLAMP(x, lo, hi)   (((x) > (hi)) ? (hi) : (((x) < (lo)) ? (lo) : (x)))
-`endif
-
-`ifndef UP
-`define UP(x)   (((x) != 0) ? (x) : 1)
-`endif
-
-///////////////////////////////////////////////////////////////////////////////
-
-`ifndef EXT_M_DISABLE
-`define EXT_M_ENABLE
-`endif
-
-`ifndef EXT_F_DISABLE
-`define EXT_F_ENABLE
-`endif
-
-// core-coupled tensor core
-`ifndef EXT_T_DISABLE
-`define EXT_T_ENABLE
-// decoupled Hopper-style tensor core
-// `ifndef EXT_T_HOPPER
-// `define EXT_T_HOPPER
-// `endif
-`endif
-
-`ifndef XLEN_32
-`ifndef XLEN_64
-`define XLEN_32
-`endif
-`endif
-
-`ifdef XLEN_64
-`define XLEN 64
-`endif
-
-`ifdef XLEN_32
+`ifndef XLEN
 `define XLEN 32
 `endif
 
-`ifdef EXT_D_ENABLE
-`define FLEN_64
-`else
-`define FLEN_32
-`endif
-
-`ifdef FLEN_64
-`define FLEN 64
-`endif
-
-`ifdef FLEN_32
-`define FLEN 32
-`endif
-
-`ifdef XLEN_64
-`ifdef FLEN_32
-    `define FPU_RV64F
-`endif
-`endif
-
 `ifndef NUM_CLUSTERS
 `define NUM_CLUSTERS 1
 `endif
@@ -95,366 +17,296 @@
 `define NUM_WARPS 4
 `endif
 
-`ifndef NUM_TENSOR_WARPS
-`define NUM_TENSOR_WARPS 2
-`endif
-
-`define NUM_SCALAR_WARPS (`NUM_WARPS - `NUM_TENSOR_WARPS)
-
-`define WU_CONFIG_STATIC_ASSERTS \
-    generate \
-        if (!(`NUM_WARPS > 0)) begin : g_wu_num_warps_gt_zero \
-            invalid_NUM_WARPS_must_be_greater_than_zero __wu_config_error(); \
-        end \
-        if (!(`NUM_TENSOR_WARPS > 0)) begin : g_wu_num_tensor_warps_gt_zero \
-            invalid_NUM_TENSOR_WARPS_must_be_greater_than_zero __wu_config_error(); \
-        end \
-        if (!(`NUM_TENSOR_WARPS < `NUM_WARPS)) begin : g_wu_num_tensor_warps_lt_num_warps \
-            invalid_NUM_TENSOR_WARPS_must_be_smaller_than_NUM_WARPS __wu_config_error(); \
-        end \
-        if (!(`NUM_SCALAR_WARPS > 0)) begin : g_wu_num_scalar_warps_gt_zero \
-            invalid_NUM_SCALAR_WARPS_must_be_greater_than_zero __wu_config_error(); \
-        end \
-    endgenerate
-
-`define IS_SCALAR_WARP(wid) ((wid) < `NUM_SCALAR_WARPS)
-`define IS_TENSOR_WARP(wid) ((wid) >= `NUM_SCALAR_WARPS)
-
-`ifndef TENSOR_NUM_GPRS
-`define TENSOR_NUM_GPRS 8
-`endif
-
-`ifndef TENSOR_NUM_FPRS
-`define TENSOR_NUM_FPRS 8
-`endif
-
 `ifndef NUM_THREADS
 `define NUM_THREADS 4
 `endif
 
 `ifndef NUM_BARRIERS
-`define NUM_BARRIERS 8
+`define NUM_BARRIERS 4
 `endif
 
-`ifndef SOCKET_SIZE
-`define SOCKET_SIZE `MIN(4, `NUM_CORES)
-`endif
-`define NUM_SOCKETS `UP(`NUM_CORES / `SOCKET_SIZE)
-
-`ifdef L2_ENABLE
-    `define L2_ENABLED   1
-`else
-    `define L2_ENABLED   0
+`ifndef L2_ENABLE
+`define L2_ENABLE 0
 `endif
 
-`ifdef L3_ENABLE
-    `define L3_ENABLED   1
-`else
-    `define L3_ENABLED   0
+`ifndef L3_ENABLE
+`define L3_ENABLE 0
 `endif
 
-`ifdef L1_DISABLE
-    `define ICACHE_DISABLE
-    `define DCACHE_DISABLE
+`ifndef SM_ENABLE
+`define SM_ENABLE 1
 `endif
 
 `ifndef MEM_BLOCK_SIZE
 `define MEM_BLOCK_SIZE 64
 `endif
 
-`ifndef MEM_ADDR_WIDTH
-`ifdef XLEN_64
-`define MEM_ADDR_WIDTH 48
-`else
-`define MEM_ADDR_WIDTH 32
+`ifndef L1_BLOCK_SIZE
+`define L1_BLOCK_SIZE ((`L2_ENABLE || `L3_ENABLE) ? 16 : `MEM_BLOCK_SIZE)
 `endif
-`endif
-
-`ifndef L1_LINE_SIZE
-`ifdef L1_DISABLE
-`define L1_LINE_SIZE ((`L2_ENABLED || `L3_ENABLED) ? 4 : `MEM_BLOCK_SIZE)
-`else
-`define L1_LINE_SIZE ((`L2_ENABLED || `L3_ENABLED) ? 16 : `MEM_BLOCK_SIZE)
-`endif
-`endif
-
-`ifdef L2_ENABLE
-`define L2_LINE_SIZE `MEM_BLOCK_SIZE
-`else
-`define L2_LINE_SIZE `L1_LINE_SIZE
-`endif
-
-`ifdef L3_ENABLE
-`define L3_LINE_SIZE `MEM_BLOCK_SIZE
-`else
-`define L3_LINE_SIZE `L2_LINE_SIZE
-`endif
-
-`ifdef XLEN_64
-
-`ifndef STARTUP_ADDR
-`define STARTUP_ADDR 64'h180000000
-`endif
-
-`ifndef STACK_BASE_ADDR
-`define STACK_BASE_ADDR 64'h1FF000000
-`endif
-
-`else
 
 `ifndef STARTUP_ADDR
 `define STARTUP_ADDR 32'h80000000
 `endif
 
-`ifndef STACK_BASE_ADDR
-`define STACK_BASE_ADDR 32'hFF000000
-`endif
-
-`endif
-
-`ifndef SMEM_BASE_ADDR
-`define SMEM_BASE_ADDR `STACK_BASE_ADDR
-`endif
-
-`ifndef SMEM_LOG_SIZE
-`define SMEM_LOG_SIZE   19
-`endif
-
 `ifndef IO_BASE_ADDR
-`define IO_BASE_ADDR (`SMEM_BASE_ADDR + (1 << `SMEM_LOG_SIZE))
+`define IO_BASE_ADDR 32'hFF000000
+`endif
+
+`ifndef IO_ADDR_SIZE
+`define IO_ADDR_SIZE (32'hFFFFFFFF - `IO_BASE_ADDR + 1)
 `endif
 
 `ifndef IO_COUT_ADDR
-`define IO_COUT_ADDR `IO_BASE_ADDR
+`define IO_COUT_ADDR (32'hFFFFFFFF - `MEM_BLOCK_SIZE + 1)
 `endif
+
+`ifndef IO_COUT_SIZE
 `define IO_COUT_SIZE `MEM_BLOCK_SIZE
+`endif
 
 `ifndef IO_CSR_ADDR
-`define IO_CSR_ADDR (`IO_COUT_ADDR + `IO_COUT_SIZE)
-`endif
-`define IO_CSR_SIZE (4 * 64 * `NUM_CORES * `NUM_CLUSTERS)
-
-`ifndef STACK_LOG2_SIZE
-`define STACK_LOG2_SIZE 13
-`endif
-`define STACK_SIZE (1 << `STACK_LOG2_SIZE)
-
-`define RESET_DELAY 8
-
-`ifndef STALL_TIMEOUT
-`define STALL_TIMEOUT (100000 * (1 ** (`L2_ENABLED + `L3_ENABLED)))
+`define IO_CSR_ADDR `IO_BASE_ADDR
 `endif
 
-`ifndef SV_DPI
-`define DPI_DISABLE
+`ifndef SMEM_BASE_ADDR
+`define SMEM_BASE_ADDR `IO_BASE_ADDR
 `endif
 
-`ifndef FPU_FPNEW
-`ifndef FPU_DSP
-`ifndef FPU_DPI
-`ifndef SYNTHESIS
-`ifndef DPI_DISABLE
-`define FPU_DPI
-`else
-`define FPU_DSP
-`endif
-`else
-`define FPU_DSP
-`endif
-`endif
-`endif
+`ifndef EXT_M_DISABLE
+`define EXT_M_ENABLE
 `endif
 
-`ifndef SYNTHESIS
-`ifndef DPI_DISABLE
-`define IMUL_DPI
-`define IDIV_DPI
-`endif
+`ifndef EXT_F_DISABLE
+`define EXT_F_ENABLE
 `endif
 
-`ifndef DEBUG_LEVEL
-`define DEBUG_LEVEL 3
-`endif
+// Device identification
+`define VENDOR_ID           0
+`define ARCHITECTURE_ID     0
+`define IMPLEMENTATION_ID   0
 
-// Pipeline Configuration /////////////////////////////////////////////////////
-
-// Issue width
-`ifndef ISSUE_WIDTH
-`define ISSUE_WIDTH     `NUM_WARPS
-`endif
-
-// Number of ALU units
-`ifndef NUM_ALU_LANES
-`define NUM_ALU_LANES   `NUM_THREADS
-`endif
-`ifndef NUM_ALU_BLOCKS
-`define NUM_ALU_BLOCKS  2
-`endif
-
-// Number of FPU units
-`ifndef NUM_FPU_LANES
-`define NUM_FPU_LANES   `NUM_THREADS
-`endif
-`ifndef NUM_FPU_BLOCKS
-`define NUM_FPU_BLOCKS  1
-`endif
-
-// Number of LSU units
-`ifndef NUM_LSU_LANES
-`define NUM_LSU_LANES   `NUM_THREADS
-`endif
-
-// Number of SFU units
-`ifndef NUM_SFU_LANES
-`define NUM_SFU_LANES   `MIN(`NUM_THREADS, 4)
-`endif
-
-// Size of Instruction Buffer
-`ifndef IBUF_SIZE
-`define IBUF_SIZE   (4 * `ISSUE_WIDTH)
-`endif
-
-// Size of LSU Request Queue
-`ifndef LSUQ_SIZE
-`define LSUQ_SIZE   (4 * `NUM_WARPS * (`NUM_THREADS / `NUM_LSU_LANES))
-`endif
-
-// LSU Duplicate Address Check
-`ifndef LSU_DUP_DISABLE
-`define LSU_DUP_ENABLE
-`endif
-`ifdef LSU_DUP_ENABLE
-`define LSU_DUP_ENABLED 1
-`else
-`define LSU_DUP_ENABLED 0
-`endif
-
-`ifdef GBAR_ENABLE
-`define GBAR_ENABLED 1
-`else
-`define GBAR_ENABLED 0
-`endif
+///////////////////////////////////////////////////////////////////////////////
 
 `ifndef LATENCY_IMUL
-`ifdef VIVADO
-`define LATENCY_IMUL 4
-`endif
-`ifdef QUARTUS
 `define LATENCY_IMUL 3
 `endif
-`ifndef LATENCY_IMUL
-`define LATENCY_IMUL 4
-`endif
-`endif
 
-// Floating-Point Units ///////////////////////////////////////////////////////
-
-// Size of FPU Request Queue
-`ifndef FPUQ_SIZE
-`define FPUQ_SIZE (8 * (`NUM_THREADS / `NUM_FPU_LANES))
-`endif
-
-// FNCP Latency
 `ifndef LATENCY_FNCP
 `define LATENCY_FNCP 2
 `endif
 
-// FMA Latency
 `ifndef LATENCY_FMA
-`ifdef FPU_DPI
-`define LATENCY_FMA 4    
-`endif
-`ifdef FPU_FPNEW
-`define LATENCY_FMA 4    
-`endif
-`ifdef FPU_DSP
-`ifdef QUARTUS
 `define LATENCY_FMA 4
 `endif
-`ifdef VIVADO
-`define LATENCY_FMA 16    
-`endif
-`ifndef LATENCY_FMA
-`define LATENCY_FMA 4    
-`endif
-`endif
-`endif
 
-// FDIV Latency
 `ifndef LATENCY_FDIV
-`ifdef FPU_DPI
-`define LATENCY_FDIV 15    
-`endif
-`ifdef FPU_FPNEW
-`define LATENCY_FDIV 16    
-`endif
-`ifdef FPU_DSP
-`ifdef QUARTUS
+`ifdef ALTERA_S10
+`define LATENCY_FDIV 34
+`else
 `define LATENCY_FDIV 15
 `endif
-`ifdef VIVADO
-`define LATENCY_FDIV 28    
-`endif
-`ifndef LATENCY_FDIV
-`define LATENCY_FDIV 16
-`endif
-`endif
 `endif
 
-// FSQRT Latency
 `ifndef LATENCY_FSQRT
-`ifdef FPU_DPI
-`define LATENCY_FSQRT 10    
-`endif
-`ifdef FPU_FPNEW
-`define LATENCY_FSQRT 16    
-`endif
-`ifdef FPU_DSP
-`ifdef QUARTUS
+`ifdef ALTERA_S10
+`define LATENCY_FSQRT 25
+`else
 `define LATENCY_FSQRT 10
 `endif
-`ifdef VIVADO
-`define LATENCY_FSQRT 28    
-`endif
-`ifndef LATENCY_FSQRT
-`define LATENCY_FSQRT 16    
-`endif
-`endif
 `endif
 
-// FCVT Latency
+`ifndef LATENCY_FDIVSQRT
+`define LATENCY_FDIVSQRT 32
+`endif
+
 `ifndef LATENCY_FCVT
 `define LATENCY_FCVT 5
 `endif
 
-// Tensor Core Latency
-`ifndef LATENCY_HMMA
-`define LATENCY_HMMA 4
+`define RESET_DELAY 6
+
+// CSR Addresses //////////////////////////////////////////////////////////////
+
+// User Floating-Point CSRs
+`define CSR_FFLAGS      12'h001
+`define CSR_FRM         12'h002
+`define CSR_FCSR        12'h003
+
+`define CSR_SATP        12'h180
+
+`define CSR_PMPCFG0     12'h3A0
+`define CSR_PMPADDR0    12'h3B0
+
+`define CSR_MSTATUS     12'h300
+`define CSR_MISA        12'h301
+`define CSR_MEDELEG     12'h302
+`define CSR_MIDELEG     12'h303
+`define CSR_MIE         12'h304
+`define CSR_MTVEC       12'h305
+
+`define CSR_MEPC        12'h341
+
+// Machine Performance-monitoring counters
+`define CSR_MPM_BASE                12'hB00
+`define CSR_MPM_BASE_H              12'hB80
+// PERF: pipeline
+`define CSR_MCYCLE                  12'hB00
+`define CSR_MCYCLE_H                12'hB80
+`define CSR_MPM_RESERVED            12'hB01
+`define CSR_MPM_RESERVED_H          12'hB81
+`define CSR_MINSTRET                12'hB02
+`define CSR_MINSTRET_H              12'hB82
+`define CSR_MPM_IBUF_ST             12'hB03
+`define CSR_MPM_IBUF_ST_H           12'hB83
+`define CSR_MPM_SCRB_ST             12'hB04
+`define CSR_MPM_SCRB_ST_H           12'hB84
+`define CSR_MPM_ALU_ST              12'hB05
+`define CSR_MPM_ALU_ST_H            12'hB85
+`define CSR_MPM_LSU_ST              12'hB06
+`define CSR_MPM_LSU_ST_H            12'hB86
+`define CSR_MPM_CSR_ST              12'hB07
+`define CSR_MPM_CSR_ST_H            12'hB87
+`define CSR_MPM_FPU_ST              12'hB08
+`define CSR_MPM_FPU_ST_H            12'hB88
+`define CSR_MPM_GPU_ST              12'hB09
+`define CSR_MPM_GPU_ST_H            12'hB89
+// PERF: decode
+`define CSR_MPM_LOADS               12'hB0A
+`define CSR_MPM_LOADS_H             12'hB8A
+`define CSR_MPM_STORES              12'hB0B
+`define CSR_MPM_STORES_H            12'hB8B
+`define CSR_MPM_BRANCHES            12'hB0C
+`define CSR_MPM_BRANCHES_H          12'hB8C
+// PERF: icache
+`define CSR_MPM_ICACHE_READS        12'hB0D     // total reads
+`define CSR_MPM_ICACHE_READS_H      12'hB8D
+`define CSR_MPM_ICACHE_MISS_R       12'hB0E     // read misses
+`define CSR_MPM_ICACHE_MISS_R_H     12'hB8E
+// PERF: dcache
+`define CSR_MPM_DCACHE_READS        12'hB0F     // total reads
+`define CSR_MPM_DCACHE_READS_H      12'hB8F
+`define CSR_MPM_DCACHE_WRITES       12'hB10     // total writes
+`define CSR_MPM_DCACHE_WRITES_H     12'hB90
+`define CSR_MPM_DCACHE_MISS_R       12'hB11     // read misses
+`define CSR_MPM_DCACHE_MISS_R_H     12'hB91
+`define CSR_MPM_DCACHE_MISS_W       12'hB12     // write misses
+`define CSR_MPM_DCACHE_MISS_W_H     12'hB92
+`define CSR_MPM_DCACHE_BANK_ST      12'hB13     // bank conflicts
+`define CSR_MPM_DCACHE_BANK_ST_H    12'hB93
+`define CSR_MPM_DCACHE_MSHR_ST      12'hB14     // MSHR stalls
+`define CSR_MPM_DCACHE_MSHR_ST_H    12'hB94
+// PERF: smem
+`define CSR_MPM_SMEM_READS          12'hB15     // total reads
+`define CSR_MPM_SMEM_READS_H        12'hB95
+`define CSR_MPM_SMEM_WRITES         12'hB16     // total writes
+`define CSR_MPM_SMEM_WRITES_H       12'hB96
+`define CSR_MPM_SMEM_BANK_ST        12'hB17     // bank conflicts
+`define CSR_MPM_SMEM_BANK_ST_H      12'hB97
+// PERF: memory
+`define CSR_MPM_MEM_READS           12'hB18     // memory reads
+`define CSR_MPM_MEM_READS_H         12'hB98
+`define CSR_MPM_MEM_WRITES          12'hB19     // memory writes
+`define CSR_MPM_MEM_WRITES_H        12'hB99
+`define CSR_MPM_MEM_LAT             12'hB1A     // memory latency
+`define CSR_MPM_MEM_LAT_H           12'hB9A
+// PERF: texunit
+`define CSR_MPM_TEX_READS           12'hB1B     // texture accesses
+`define CSR_MPM_TEX_READS_H         12'hB9B
+`define CSR_MPM_TEX_LAT             12'hB1C     // texture latency
+`define CSR_MPM_TEX_LAT_H           12'hB9C
+
+// Machine Information Registers
+`define CSR_MVENDORID   12'hF11
+`define CSR_MARCHID     12'hF12
+`define CSR_MIMPID      12'hF13
+`define CSR_MHARTID     12'hF14
+
+// User SIMT CSRs
+`define CSR_WTID        12'hCC0
+`define CSR_LTID        12'hCC1
+`define CSR_GTID        12'hCC2
+`define CSR_LWID        12'hCC3
+`define CSR_GWID        `CSR_MHARTID
+`define CSR_GCID        12'hCC5
+`define CSR_TMASK       12'hCC4
+
+// Machine SIMT CSRs
+`define CSR_NT          12'hFC0
+`define CSR_NW          12'hFC1
+`define CSR_NC          12'hFC2
+
+////////// Texture Units //////////////////////////////////////////////////////
+
+`define NUM_TEX_UNITS           2
+`define TEX_SUBPIXEL_BITS       8
+
+`define TEX_DIM_BITS            15
+`define TEX_LOD_MAX             `TEX_DIM_BITS
+`define TEX_LOD_BITS            4
+
+`define TEX_FXD_BITS            32
+`define TEX_FXD_FRAC            (`TEX_DIM_BITS+`TEX_SUBPIXEL_BITS)
+
+`define TEX_STATE_ADDR          0
+`define TEX_STATE_WIDTH         1
+`define TEX_STATE_HEIGHT        2
+`define TEX_STATE_FORMAT        3
+`define TEX_STATE_FILTER        4
+`define TEX_STATE_WRAPU         5
+`define TEX_STATE_WRAPV         6
+`define TEX_STATE_MIPOFF(lod)   (7+(lod))
+`define NUM_TEX_STATES          (`TEX_STATE_MIPOFF(`TEX_LOD_MAX)+1)
+
+`define CSR_TEX_UNIT            12'hFD0
+
+`define CSR_TEX_STATE_BEGIN     12'hFD1
+`define CSR_TEX_ADDR            (`CSR_TEX_STATE_BEGIN+`TEX_STATE_ADDR)
+`define CSR_TEX_WIDTH           (`CSR_TEX_STATE_BEGIN+`TEX_STATE_WIDTH)
+`define CSR_TEX_HEIGHT          (`CSR_TEX_STATE_BEGIN+`TEX_STATE_HEIGHT)
+`define CSR_TEX_FORMAT          (`CSR_TEX_STATE_BEGIN+`TEX_STATE_FORMAT)
+`define CSR_TEX_FILTER          (`CSR_TEX_STATE_BEGIN+`TEX_STATE_FILTER)
+`define CSR_TEX_WRAPU           (`CSR_TEX_STATE_BEGIN+`TEX_STATE_WRAPU)
+`define CSR_TEX_WRAPV           (`CSR_TEX_STATE_BEGIN+`TEX_STATE_WRAPV)
+`define CSR_TEX_MIPOFF(lod)     (`CSR_TEX_STATE_BEGIN+`TEX_STATE_MIPOFF(lod))
+`define CSR_TEX_STATE_END       (`CSR_TEX_STATE_BEGIN+`NUM_TEX_STATES)
+
+`define CSR_TEX_STATE(addr)     ((addr) - `CSR_TEX_STATE_BEGIN)
+
+// Pipeline Queues ////////////////////////////////////////////////////////////
+
+// Size of Instruction Buffer
+`ifndef IBUF_SIZE
+`define IBUF_SIZE 2
+`endif
+
+// Size of LSU Request Queue
+`ifndef LSUQ_SIZE
+`define LSUQ_SIZE (`NUM_WARPS * 2)
+`endif
+
+// Size of FPU Request Queue
+`ifndef FPUQ_SIZE
+`define FPUQ_SIZE 8
+`endif
+
+// Texture Unit Request Queue
+`ifndef TEXQ_SIZE
+`define TEXQ_SIZE (`NUM_WARPS * 2)
 `endif
 
 // Icache Configurable Knobs //////////////////////////////////////////////////
 
-// Cache Enable
-`ifndef ICACHE_DISABLE
-`define ICACHE_ENABLE
-`endif
-`ifdef ICACHE_ENABLE
-    `define ICACHE_ENABLED 1
-`else
-    `define ICACHE_ENABLED 0
-    `define NUM_ICACHES 0
-`endif
-
-// Number of Cache Units
-`ifndef NUM_ICACHES
-`define NUM_ICACHES `UP(`SOCKET_SIZE / 4)
-`endif
-
-// Cache Size
+// Size of cache in bytes
 `ifndef ICACHE_SIZE
 `define ICACHE_SIZE 16384
 `endif
 
+// Core Request Queue Size
+`ifndef ICACHE_CREQ_SIZE
+`define ICACHE_CREQ_SIZE 0
+`endif
+
 // Core Response Queue Size
 `ifndef ICACHE_CRSQ_SIZE
 `define ICACHE_CRSQ_SIZE 2
@@ -462,7 +314,7 @@
 
 // Miss Handling Register Size
 `ifndef ICACHE_MSHR_SIZE
-`define ICACHE_MSHR_SIZE 16
+`define ICACHE_MSHR_SIZE `NUM_WARPS
 `endif
 
 // Memory Request Queue Size
@@ -475,38 +327,26 @@
 `define ICACHE_MRSQ_SIZE 0
 `endif
 
-// Number of Associative Ways
-`ifndef ICACHE_NUM_WAYS
-`define ICACHE_NUM_WAYS 1
-`endif
-
 // Dcache Configurable Knobs //////////////////////////////////////////////////
 
-// Cache Enable
-`ifndef DCACHE_DISABLE
-`define DCACHE_ENABLE
-`endif
-`ifdef DCACHE_ENABLE
-    `define DCACHE_ENABLED 1
-`else
-    `define DCACHE_ENABLED 0
-    `define NUM_DCACHES 0
-    `define DCACHE_NUM_BANKS 1
-`endif
-
-// Number of Cache Units
-`ifndef NUM_DCACHES
-`define NUM_DCACHES `UP(`SOCKET_SIZE / 4)
-`endif
-
-// Cache Size
+// Size of cache in bytes
 `ifndef DCACHE_SIZE
 `define DCACHE_SIZE 16384
 `endif
 
-// Number of Banks
+// Number of banks
 `ifndef DCACHE_NUM_BANKS
-`define DCACHE_NUM_BANKS `NUM_LSU_LANES
+`define DCACHE_NUM_BANKS `NUM_THREADS
+`endif
+
+// Number of ports per bank
+`ifndef DCACHE_NUM_PORTS
+`define DCACHE_NUM_PORTS 1
+`endif
+
+// Core Request Queue Size
+`ifndef DCACHE_CREQ_SIZE
+`define DCACHE_CREQ_SIZE 0
 `endif
 
 // Core Response Queue Size
@@ -516,7 +356,7 @@
 
 // Miss Handling Register Size
 `ifndef DCACHE_MSHR_SIZE
-`define DCACHE_MSHR_SIZE 8
+`define DCACHE_MSHR_SIZE `LSUQ_SIZE
 `endif
 
 // Memory Request Queue Size
@@ -529,43 +369,54 @@
 `define DCACHE_MRSQ_SIZE 0
 `endif
 
-// Number of Associative Ways
-`ifndef DCACHE_NUM_WAYS
-`define DCACHE_NUM_WAYS 1
-`endif
-
 // SM Configurable Knobs //////////////////////////////////////////////////////
 
-`ifndef SM_DISABLE
-`define SM_ENABLE
+// per thread stack size
+`ifndef STACK_LOG2_SIZE
+`define STACK_LOG2_SIZE 10
+`endif
+`define STACK_SIZE (1 << `STACK_LOG2_SIZE)
+
+// Size of cache in bytes
+`ifndef SMEM_SIZE
+`define SMEM_SIZE (`STACK_SIZE * `NUM_WARPS * `NUM_THREADS)
 `endif
 
-`ifdef SM_ENABLE
-    `define SM_ENABLED   1
-`else
-    `define SM_ENABLED   0
-    `define SMEM_NUM_BANKS 1
-`endif
-
-// Number of Banks
+// Number of banks
 `ifndef SMEM_NUM_BANKS
-`define SMEM_NUM_BANKS (`NUM_LSU_LANES)
+`define SMEM_NUM_BANKS `NUM_THREADS
+`endif
+
+// Core Request Queue Size
+`ifndef SMEM_CREQ_SIZE
+`define SMEM_CREQ_SIZE 2
+`endif
+
+// Core Response Queue Size
+`ifndef SMEM_CRSQ_SIZE
+`define SMEM_CRSQ_SIZE 2
 `endif
 
 // L2cache Configurable Knobs /////////////////////////////////////////////////
 
-// Cache Size
+// Size of cache in bytes
 `ifndef L2_CACHE_SIZE
-`ifdef ALTERA_S10
-`define L2_CACHE_SIZE 2097152
-`else
-`define L2_CACHE_SIZE 1048576
-`endif
+`define L2_CACHE_SIZE 131072
 `endif
 
-// Number of Banks
+// Number of banks
 `ifndef L2_NUM_BANKS
-`define L2_NUM_BANKS `MIN(4, `NUM_SOCKETS)
+`define L2_NUM_BANKS ((`NUM_CORES < 4) ? `NUM_CORES : 4)
+`endif
+
+// Number of ports per bank
+`ifndef L2_NUM_PORTS
+`define L2_NUM_PORTS 1
+`endif
+
+// Core Request Queue Size
+`ifndef L2_CREQ_SIZE
+`define L2_CREQ_SIZE 0
 `endif
 
 // Core Response Queue Size
@@ -588,25 +439,26 @@
 `define L2_MRSQ_SIZE 0
 `endif
 
-// Number of Associative Ways
-`ifndef L2_NUM_WAYS
-`define L2_NUM_WAYS 2
-`endif
-
 // L3cache Configurable Knobs /////////////////////////////////////////////////
 
-// Cache Size
+// Size of cache in bytes
 `ifndef L3_CACHE_SIZE
-`ifdef ALTERA_S10
-`define L3_CACHE_SIZE 2097152
-`else
 `define L3_CACHE_SIZE 1048576
 `endif
+
+// Number of banks
+`ifndef L3_NUM_BANKS
+`define L3_NUM_BANKS ((`NUM_CLUSTERS < 4) ? `NUM_CORES : 4)
 `endif
 
-// Number of Banks
-`ifndef L3_NUM_BANKS
-`define L3_NUM_BANKS `MIN(4, `NUM_CLUSTERS)
+// Number of ports per bank
+`ifndef L3_NUM_PORTS
+`define L3_NUM_PORTS 1
+`endif
+
+// Core Request Queue Size
+`ifndef L3_CREQ_SIZE
+`define L3_CREQ_SIZE 0
 `endif
 
 // Core Response Queue Size
@@ -629,104 +481,4 @@
 `define L3_MRSQ_SIZE 0
 `endif
 
-// Number of Associative Ways
-`ifndef L3_NUM_WAYS
-`define L3_NUM_WAYS 4
-`endif
-
-// ISA Extensions /////////////////////////////////////////////////////////////
-
-`ifdef EXT_A_ENABLE
-    `define EXT_A_ENABLED   1
-`else
-    `define EXT_A_ENABLED   0
-`endif
-
-`ifdef EXT_C_ENABLE
-    `define EXT_C_ENABLED   1
-`else
-    `define EXT_C_ENABLED   0
-`endif
-
-`ifdef EXT_D_ENABLE
-    `define EXT_D_ENABLED   1
-`else
-    `define EXT_D_ENABLED   0
-`endif
-
-`ifdef EXT_F_ENABLE
-    `define EXT_F_ENABLED   1
-`else
-    `define EXT_F_ENABLED   0
-`endif
-
-`ifdef EXT_T_ENABLE
-    `define EXT_T_ENABLED   1
-`else
-    `define EXT_T_ENABLED   0
-`endif
-
-`ifdef EXT_M_ENABLE
-    `define EXT_M_ENABLED   1
-`else
-    `define EXT_M_ENABLED   0
-`endif
-
-`define ISA_STD_A           0
-`define ISA_STD_C           2
-`define ISA_STD_D           3
-`define ISA_STD_E           4
-`define ISA_STD_F           5
-`define ISA_STD_H           7
-`define ISA_STD_I           8
-`define ISA_STD_N           13
-`define ISA_STD_Q           16
-`define ISA_STD_S           18
-`define ISA_STD_U           20
-
-`define ISA_EXT_ICACHE      0
-`define ISA_EXT_DCACHE      1
-`define ISA_EXT_L2CACHE     2
-`define ISA_EXT_L3CACHE     3
-`define ISA_EXT_SMEM        4
-
-`define MISA_EXT  (`ICACHE_ENABLED  << `ISA_EXT_ICACHE) \
-                | (`DCACHE_ENABLED  << `ISA_EXT_DCACHE) \
-                | (`L2_ENABLED      << `ISA_EXT_L2CACHE) \
-                | (`L3_ENABLED      << `ISA_EXT_L3CACHE) \
-                | (`SM_ENABLED      << `ISA_EXT_SMEM)
-
-`define MISA_STD  (`EXT_A_ENABLED <<  0) /* A - Atomic Instructions extension */ \
-                | (0 <<  1) /* B - Tentatively reserved for Bit operations extension */ \
-                | (`EXT_C_ENABLED <<  2) /* C - Compressed extension */ \
-                | (`EXT_D_ENABLED <<  3) /* D - Double precsision floating-point extension */ \
-                | (0 <<  4) /* E - RV32E base ISA */ \
-                | (`EXT_F_ENABLED << 5) /* F - Single precsision floating-point extension */ \
-                | (0 <<  6) /* G - Additional standard extensions present */ \
-                | (0 <<  7) /* H - Hypervisor mode implemented */ \
-                | (1 <<  8) /* I - RV32I/64I/128I base ISA */ \
-                | (0 <<  9) /* J - Reserved */ \
-                | (0 << 10) /* K - Reserved */ \
-                | (0 << 11) /* L - Tentatively reserved for Bit operations extension */ \
-                | (`EXT_M_ENABLED << 12) /* M - Integer Multiply/Divide extension */ \
-                | (0 << 13) /* N - User level interrupts supported */ \
-                | (0 << 14) /* O - Reserved */ \
-                | (0 << 15) /* P - Tentatively reserved for Packed-SIMD extension */ \
-                | (0 << 16) /* Q - Quad-precision floating-point extension */ \
-                | (0 << 17) /* R - Reserved */ \
-                | (0 << 18) /* S - Supervisor mode implemented */ \
-                | (0 << 19) /* T - Tentatively reserved for Transactional Memory extension */ \
-                | (1 << 20) /* U - User mode implemented */ \
-                | (0 << 21) /* V - Tentatively reserved for Vector extension */ \
-                | (0 << 22) /* W - Reserved */ \
-                | (1 << 23) /* X - Non-standard extensions present */ \
-                | (0 << 24) /* Y - Reserved */ \
-                | (0 << 25) /* Z - Reserved */
-
-// Device identification //////////////////////////////////////////////////////
-
-`define VENDOR_ID           0
-`define ARCHITECTURE_ID     0
-`define IMPLEMENTATION_ID   0
-
-`endif // VX_CONFIG_VH
+`endif

hw/rtl/VX_core.sv

@@ -0,0 +1,156 @@
+`include "VX_define.vh"
+
+module VX_core #( 
+    parameter CORE_ID = 0
+) (        
+    `SCOPE_IO_VX_core
+    
+    // Clock
+    input  wire                             clk,
+    input  wire                             reset,
+
+    // Memory request
+    output wire                             mem_req_valid,
+    output wire                             mem_req_rw,    
+    output wire [`DCACHE_MEM_BYTEEN_WIDTH-1:0] mem_req_byteen,
+    output wire [`DCACHE_MEM_ADDR_WIDTH-1:0] mem_req_addr,
+    output wire [`DCACHE_MEM_DATA_WIDTH-1:0] mem_req_data,
+    output wire [`L1_MEM_TAG_WIDTH-1:0]     mem_req_tag,
+    input  wire                             mem_req_ready,
+
+    // Memory reponse    
+    input  wire                             mem_rsp_valid,
+    input  wire [`DCACHE_MEM_DATA_WIDTH-1:0] mem_rsp_data,
+    input  wire [`L1_MEM_TAG_WIDTH-1:0]     mem_rsp_tag,
+    output wire                             mem_rsp_ready,
+
+    // Status
+    output wire                             busy
+);
+`ifdef PERF_ENABLE
+    VX_perf_memsys_if perf_memsys_if();
+`endif
+
+    VX_mem_req_if #(
+        .DATA_WIDTH (`DCACHE_MEM_DATA_WIDTH),
+        .ADDR_WIDTH (`DCACHE_MEM_ADDR_WIDTH),
+        .TAG_WIDTH  (`L1_MEM_TAG_WIDTH)
+    ) mem_req_if();
+
+    VX_mem_rsp_if #(
+        .DATA_WIDTH (`DCACHE_MEM_DATA_WIDTH),
+        .TAG_WIDTH  (`L1_MEM_TAG_WIDTH)
+    ) mem_rsp_if();
+
+    assign mem_req_valid = mem_req_if.valid;
+    assign mem_req_rw    = mem_req_if.rw;
+    assign mem_req_byteen= mem_req_if.byteen;
+    assign mem_req_addr  = mem_req_if.addr;
+    assign mem_req_data  = mem_req_if.data;
+    assign mem_req_tag   = mem_req_if.tag;
+    assign mem_req_if.ready = mem_req_ready;
+
+    assign mem_rsp_if.valid = mem_rsp_valid;
+    assign mem_rsp_if.data  = mem_rsp_data;
+    assign mem_rsp_if.tag   = mem_rsp_tag;
+    assign mem_rsp_ready = mem_rsp_if.ready;
+
+    //--
+
+    VX_dcache_req_if #(
+        .NUM_REQS  (`DCACHE_NUM_REQS), 
+        .WORD_SIZE (`DCACHE_WORD_SIZE), 
+        .TAG_WIDTH (`DCACHE_CORE_TAG_WIDTH)
+    ) dcache_req_if();
+
+    VX_dcache_rsp_if #(
+        .NUM_REQS  (`DCACHE_NUM_REQS), 
+        .WORD_SIZE (`DCACHE_WORD_SIZE), 
+        .TAG_WIDTH (`DCACHE_CORE_TAG_WIDTH)
+    ) dcache_rsp_if();
+    
+    VX_icache_req_if #(
+        .WORD_SIZE (`ICACHE_WORD_SIZE), 
+        .TAG_WIDTH (`ICACHE_CORE_TAG_WIDTH)
+    ) icache_req_if();
+
+    VX_icache_rsp_if #(
+        .WORD_SIZE (`ICACHE_WORD_SIZE), 
+        .TAG_WIDTH (`ICACHE_CORE_TAG_WIDTH)
+    ) icache_rsp_if();
+    
+    VX_pipeline #(
+        .CORE_ID(CORE_ID)
+    ) pipeline (
+        `SCOPE_BIND_VX_core_pipeline
+    `ifdef PERF_ENABLE
+        .perf_memsys_if (perf_memsys_if),
+    `endif
+
+        .clk(clk),
+        .reset(reset),
+
+        // Dcache core request
+        .dcache_req_valid   (dcache_req_if.valid),
+        .dcache_req_rw      (dcache_req_if.rw),
+        .dcache_req_byteen  (dcache_req_if.byteen),
+        .dcache_req_addr    (dcache_req_if.addr),
+        .dcache_req_data    (dcache_req_if.data),
+        .dcache_req_tag     (dcache_req_if.tag),
+        .dcache_req_ready   (dcache_req_if.ready),
+
+        // Dcache core reponse    
+        .dcache_rsp_valid   (dcache_rsp_if.valid),
+        .dcache_rsp_tmask   (dcache_rsp_if.tmask),
+        .dcache_rsp_data    (dcache_rsp_if.data),
+        .dcache_rsp_tag     (dcache_rsp_if.tag),
+        .dcache_rsp_ready   (dcache_rsp_if.ready),
+
+        // Icache core request
+        .icache_req_valid   (icache_req_if.valid),
+        .icache_req_addr    (icache_req_if.addr),
+        .icache_req_tag     (icache_req_if.tag),
+        .icache_req_ready   (icache_req_if.ready),
+
+        // Icache core reponse    
+        .icache_rsp_valid   (icache_rsp_if.valid),
+        .icache_rsp_data    (icache_rsp_if.data),
+        .icache_rsp_tag     (icache_rsp_if.tag),
+        .icache_rsp_ready   (icache_rsp_if.ready),
+
+        // Status
+        .busy(busy)
+    );  
+
+    //--
+
+    VX_mem_unit #(
+        .CORE_ID(CORE_ID)
+    ) mem_unit (
+        `SCOPE_BIND_VX_core_mem_unit
+    `ifdef PERF_ENABLE
+        .perf_memsys_if (perf_memsys_if),
+    `endif
+
+        .clk            (clk),
+        .reset          (reset),
+
+        // Core <-> Dcache
+        .dcache_req_if  (dcache_req_if),
+        .dcache_rsp_if  (dcache_rsp_if),
+        
+        // Core <-> Icache
+        .icache_req_if  (icache_req_if),
+        .icache_rsp_if  (icache_rsp_if),
+
+        // Memory
+        .mem_req_if     (mem_req_if),
+        .mem_rsp_if     (mem_rsp_if)
+    );
+    
+endmodule
+
+
+
+
+

hw/rtl/VX_core_wrapper.sv

@@ -1,654 +0,0 @@
-`include "VX_define.vh"
-`include "VX_gpu_pkg.sv"
-// TODO: move VX_define constants to parameters, and then parameterize in blackbox
-
-module Vortex import VX_gpu_pkg::*; #(
-    parameter CORE_ID = 0,
-    parameter TENSOR_FP16 = 0,
-    parameter logic [63:0] STARTUP_ADDR = 64'h0000_0000_0001_0100,
-    parameter NUM_THREADS = 0,
-    parameter NUM_TENSOR_CORES = 1,
-    parameter TC_DATA_WIDTH = 256,
-    parameter TC_TAG_WIDTH = 4
-) (
-
-    /* adapt to CoreIO bundle at src/main/scala/tile/Core.scala */
-
-    input         clock,
-    input         reset,
-    // input         hartid,
-    input  [31:0] reset_vector,
-    input         interrupts_debug,
-    input         interrupts_mtip,
-    input         interrupts_msip,
-    input         interrupts_meip,
-    input         interrupts_seip,
-
-    // imem ------------------------------------------------
-
-    input                         imem_0_a_ready,
-    input                         imem_0_d_valid,
-    input  [2:0]                  imem_0_d_bits_opcode,
-    input  [3:0]                  imem_0_d_bits_size,
-    input  [ICACHE_TAG_WIDTH-1:0] imem_0_d_bits_source,
-    input  [31:0]                 imem_0_d_bits_data,
-    output                        imem_0_a_valid,
-    output [2:0]                  imem_0_a_bits_opcode,
-    output [3:0]                  imem_0_a_bits_size,
-    output [ICACHE_TAG_WIDTH-1:0] imem_0_a_bits_source,
-    output [31:0]                 imem_0_a_bits_address,
-    output [3:0]                  imem_0_a_bits_mask,
-    output [31:0]                 imem_0_a_bits_data,
-    output                        imem_0_d_ready,
-
-    // dmem ------------------------------------------------
-
-    input  [DCACHE_NUM_REQS - 1:0]                           dmem_d_valid,
-    input  [(DCACHE_NUM_REQS * 3) - 1:0]                     dmem_d_bits_opcode,
-    input  [(DCACHE_NUM_REQS * 4) - 1:0]                     dmem_d_bits_size,
-    input  [(DCACHE_NUM_REQS * DCACHE_NOSM_TAG_WIDTH) - 1:0] dmem_d_bits_source,
-    input  [(DCACHE_NUM_REQS * 32) - 1:0]                    dmem_d_bits_data,
-    output [DCACHE_NUM_REQS - 1:0]                           dmem_d_ready,
-
-    input  [DCACHE_NUM_REQS - 1:0]                           dmem_a_ready,
-    output [DCACHE_NUM_REQS - 1:0]                           dmem_a_valid,
-    output [(DCACHE_NUM_REQS * 3) - 1:0]                     dmem_a_bits_opcode,
-    output [(DCACHE_NUM_REQS * 4) - 1:0]                     dmem_a_bits_size,
-    output [(DCACHE_NUM_REQS * DCACHE_NOSM_TAG_WIDTH) - 1:0] dmem_a_bits_source,
-    output [(DCACHE_NUM_REQS * 32) - 1:0]                    dmem_a_bits_address,
-    output [(DCACHE_NUM_REQS * 4) - 1:0]                     dmem_a_bits_mask,
-    output [(DCACHE_NUM_REQS * 32) - 1:0]                    dmem_a_bits_data,
-
-    // smem ------------------------------------------------
-
-    input  [DCACHE_NUM_REQS - 1:0]                           smem_d_valid,
-    input  [(DCACHE_NUM_REQS * 3) - 1:0]                     smem_d_bits_opcode,
-    input  [(DCACHE_NUM_REQS * 4) - 1:0]                     smem_d_bits_size,
-    input  [(DCACHE_NUM_REQS * DCACHE_NOSM_TAG_WIDTH) - 1:0] smem_d_bits_source,
-    input  [(DCACHE_NUM_REQS * 32) - 1:0]                    smem_d_bits_data,
-    output [DCACHE_NUM_REQS - 1:0]                           smem_d_ready,
-
-    input  [DCACHE_NUM_REQS - 1:0]                           smem_a_ready,
-    output [DCACHE_NUM_REQS - 1:0]                           smem_a_valid,
-    output [(DCACHE_NUM_REQS * 3) - 1:0]                     smem_a_bits_opcode,
-    output [(DCACHE_NUM_REQS * 4) - 1:0]                     smem_a_bits_size,
-    output [(DCACHE_NUM_REQS * DCACHE_NOSM_TAG_WIDTH) - 1:0] smem_a_bits_source,
-    output [(DCACHE_NUM_REQS * 32) - 1:0]                    smem_a_bits_address,
-    output [(DCACHE_NUM_REQS * 4) - 1:0]                     smem_a_bits_mask,
-    output [(DCACHE_NUM_REQS * 32) - 1:0]                    smem_a_bits_data,
-
-    // tc --------------------------------------------------
-    input  [NUM_TENSOR_CORES * 3 - 1:0]   tc_a_ready,
-    output [NUM_TENSOR_CORES * 3 - 1:0]   tc_a_valid,
-    output [NUM_TENSOR_CORES * 3 - 1:0]   tc_a_bits_write,
-    output [NUM_TENSOR_CORES * 3 * 32 - 1:0]  tc_a_bits_address,
-    output [NUM_TENSOR_CORES * 3 * TC_TAG_WIDTH - 1:0] tc_a_bits_tag,
-    output [NUM_TENSOR_CORES * 3 * 32 - 1:0] tc_a_bits_mask,
-    output [NUM_TENSOR_CORES * 3 * TC_DATA_WIDTH - 1:0] tc_a_bits_data,
-    output [NUM_TENSOR_CORES * 3 - 1:0]   tc_d_ready,
-    input  [NUM_TENSOR_CORES * 3 - 1:0]   tc_d_valid,
-    input  [NUM_TENSOR_CORES * 3 * TC_DATA_WIDTH - 1:0] tc_d_bits_data,
-    input  [NUM_TENSOR_CORES * 3 * TC_TAG_WIDTH - 1:0] tc_d_bits_tag,
-
-    // shared tmem direct SRAM ports
-    output [NUM_TENSOR_CORES-1:0]                         tc_tmem_A_ren,
-    input  [NUM_TENSOR_CORES-1:0]                         tc_tmem_A_rready,
-    output [NUM_TENSOR_CORES*9-1:0]                       tc_tmem_A_raddr,
-    input  [NUM_TENSOR_CORES*`NUM_THREADS*`XLEN-1:0]      tc_tmem_A_rdata,
-    output [NUM_TENSOR_CORES-1:0]                         tc_tmem_C_ren,
-    input  [NUM_TENSOR_CORES-1:0]                         tc_tmem_C_rready,
-    output [NUM_TENSOR_CORES*9-1:0]                       tc_tmem_C_raddr,
-    input  [NUM_TENSOR_CORES*`NUM_THREADS*`XLEN-1:0]      tc_tmem_C_rdata,
-    output [NUM_TENSOR_CORES-1:0]                         tc_tmem_C_wen,
-    input  [NUM_TENSOR_CORES-1:0]                         tc_tmem_C_wready,
-    output [NUM_TENSOR_CORES*9-1:0]                       tc_tmem_C_waddr,
-    output [NUM_TENSOR_CORES*`NUM_THREADS*`XLEN-1:0]      tc_tmem_C_wdata,
-    output [NUM_TENSOR_CORES*`NUM_THREADS*`XLEN/8-1:0]    tc_tmem_C_mask,
-
-    // gbar ------------------------------------------------
-
-    output                   gbar_req_valid,
-    output [`NB_WIDTH - 1:0] gbar_req_id,
-    output [`NC_WIDTH - 1:0] gbar_req_size_m1,
-    output [`NC_WIDTH - 1:0] gbar_req_core_id,
-    input                    gbar_req_ready,
-    input                    gbar_rsp_valid,
-    input  [`NB_WIDTH - 1:0] gbar_rsp_id,
-
-    // fpu (unused) ----------------------------------------
-    //
-    // input         fpu_fcsr_flags_valid,
-    // input  [4:0]  fpu_fcsr_flags_bits,
-    // // input  [63:0] fpu_store_data,
-    // input  [31:0] fpu_toint_data,
-    // input         fpu_fcsr_rdy,
-    // input         fpu_nack_mem,
-    // input         fpu_illegal_rm,
-    // input         fpu_dec_wen,
-    // input         fpu_dec_ldst,
-    // input         fpu_dec_ren1,
-    // input         fpu_dec_ren2,
-    // input         fpu_dec_ren3,
-    // input         fpu_dec_swap12,
-    // input         fpu_dec_swap23,
-    // input  [1:0]  fpu_dec_typeTagIn,
-    // input  [1:0]  fpu_dec_typeTagOut,
-    // input         fpu_dec_fromint,
-    // input         fpu_dec_toint,
-    // input         fpu_dec_fastpipe,
-    // input         fpu_dec_fma,
-    // input         fpu_dec_div,
-    // input         fpu_dec_sqrt,
-    // input         fpu_dec_wflags,
-    // input         fpu_sboard_set,
-    // input         fpu_sboard_clr,
-    // input  [4:0]  fpu_sboard_clra,
-
-    // output        fpu_hartid,
-    // output [31:0] fpu_time,
-    // output [31:0] fpu_inst,
-    // output [31:0] fpu_fromint_data,
-    // output [2:0]  fpu_fcsr_rm,
-    // output        fpu_dmem_resp_val,
-    // output [2:0]  fpu_dmem_resp_type,
-    // output [4:0]  fpu_dmem_resp_tag,
-    // output        fpu_valid,
-    // output        fpu_killx,
-    // output        fpu_killm,
-    // output        fpu_keep_clock_enabled,
-
-    // accelerator cisc csr --------------------------------
-
-    input wire [31:0]       acc_read_in,
-    output wire [31:0]      acc_write_out,
-    output wire             acc_write_en,
-
-    input         downstream_mem_busy,
-
-    output        finished,
-
-    input         traceStall,
-    output        wfi
-);
-
-    logic [3:0] intr_counter;
-    logic msip_1d, intr_reset;
-    logic busy;
-    reg busy_prev;
-    reg finished_reg;
-
-    assign intr_reset = |intr_counter;
-    /* busy and interrupts */
-    always @(posedge clock) begin
-      msip_1d <= interrupts_msip;
-      if (reset) begin
-        busy_prev <= 1'b0;
-        finished_reg <= 1'b0;
-        intr_counter <= 4'h8;
-      end else begin
-        // Vortex core's busy signal goes up some cycles after the reset,
-        // so we can't simply use ~busy as finished because of the initial
-        // ephemeral state.  Instead detect the *negedge* of the busy
-        // signal and use that to indicate finish.
-        busy_prev <= busy;
-        if (busy_prev && !busy) begin
-          finished_reg <= 1'b1;
-        end
-
-        if (~msip_1d && interrupts_msip) begin
-          // rising edge
-          intr_counter <= 4'h7;
-        end else if (intr_counter <= 4'h7) begin
-          intr_counter <= intr_counter > 0 ? intr_counter - 4'h1 : 4'h0;
-        end
-      end
-    end
-
-    assign finished = finished_reg;
-    assign wfi = 1'b0; // FIXME: unused
-
-    // ------------------------------------------------------------------------
-    // TL <-> Vortex core-cache interface adapter
-    // ------------------------------------------------------------------------
-
-    VX_mem_bus_if #(
-        .DATA_SIZE (ICACHE_WORD_SIZE), 
-        .TAG_WIDTH (ICACHE_TAG_WIDTH)
-    ) icache_bus_if();
-
-    // NOTE(hansung): need to use DCACHE_NOSM_TAG_WIDTH here instead of
-    // DCACHE_TAG_WIDTH; the latter is only used inside the core to
-    // differentiate between requests going to the cache vs. sharedmem.
-    // FIXME: DCACHE_NUM_REQS is assumed to be the same as NUM_LANES as of
-    // now.
-    VX_mem_bus_if #(
-        .DATA_SIZE (DCACHE_WORD_SIZE), 
-        .TAG_WIDTH (DCACHE_NOSM_TAG_WIDTH)
-    ) dcache_bus_if[DCACHE_NUM_REQS]();
-
-    VX_mem_bus_if #(
-        .DATA_SIZE (DCACHE_WORD_SIZE), 
-        .TAG_WIDTH (DCACHE_NOSM_TAG_WIDTH)
-    ) smem_bus_if[DCACHE_NUM_REQS]();
-
-    // always @(posedge clock) begin
-    //   `ASSERT(DCACHE_NUM_REQS == NUM_THREADS, "DCACHE_NUM_REQS doesn't match NUM_THREADS");
-    // end
-
-    // imem -------------------------------------------------------------------
-
-    assign icache_bus_if.rsp_valid = imem_0_d_valid;
-    // TODO: hardcoded DCACHE_WORD_SIZE = 4
-    assign icache_bus_if.rsp_data.data = imem_0_d_bits_data;
-    assign icache_bus_if.rsp_data.tag = imem_0_d_bits_source[ICACHE_TAG_WIDTH-1:0];
-    assign imem_0_d_ready = icache_bus_if.rsp_ready;
-
-    // always @(posedge clock) begin
-    //     if (icache_req_if.valid && icache_req_if.ready)
-    //         icache_rsp_if.tag <= icache_req_if.tag;
-    // end
-    assign imem_0_a_bits_source = {32'b0, icache_bus_if.req_data.tag}[ICACHE_TAG_WIDTH-1:0];
-    assign imem_0_a_valid = icache_bus_if.req_valid;
-    assign imem_0_a_bits_address = {icache_bus_if.req_data.addr, 2'b0};
-    assign icache_bus_if.req_ready = imem_0_a_ready;
-
-    assign imem_0_a_bits_data = 32'd0;
-    assign imem_0_a_bits_mask = 4'hf;
-    // assign imem_0_a_bits_corrupt = 1'b0;
-    // assign imem_0_a_bits_param = 3'd0;
-    assign imem_0_a_bits_size = 4'd2; // 32b
-    assign imem_0_a_bits_opcode = 3'd4; // Get
-
-    // dmem -------------------------------------------------------------------
-
-    // Vortex core does not accept write acks; filter them out here
-    generate
-        for (genvar i = 0; i < DCACHE_NUM_REQS; i++) begin
-            assign dcache_bus_if[i].rsp_valid =
-                (dmem_d_valid[i] && (dmem_d_bits_opcode[i * 3 +: 3] !== 3'd0 /*AccessAck*/));
-            // Data and tag assignment for dcache
-            assign dcache_bus_if[i].rsp_data.data = dmem_d_bits_data[i * 32 +: 32];
-            assign dcache_bus_if[i].rsp_data.tag = dmem_d_bits_source[i * DCACHE_NOSM_TAG_WIDTH +: DCACHE_NOSM_TAG_WIDTH];
-
-            // Handling write ACKs, setting ready bit for dcache
-            assign dmem_d_ready[i] = dcache_bus_if[i].rsp_ready ||
-                                    (dmem_d_valid[i] && (dmem_d_bits_opcode[i * 3 +: 3] == 3'd0 /*AccessAck*/));
-
-            // Request validity and address/data/source assignment for dcache
-            assign dmem_a_valid[i] = dcache_bus_if[i].req_valid;
-            assign dmem_a_bits_address[i * 32 +: 32] = {dcache_bus_if[i].req_data.addr, 2'b0};
-            assign dmem_a_bits_data[i * 32 +: 32] = dcache_bus_if[i].req_data.data;
-            assign dmem_a_bits_source[i * DCACHE_NOSM_TAG_WIDTH +: DCACHE_NOSM_TAG_WIDTH] = dcache_bus_if[i].req_data.tag;
-
-            // Opcode, size, and mask assignment for dcache
-            assign dmem_a_bits_opcode[i * 3 +: 3] =
-                dcache_bus_if[i].req_data.rw ?
-                (&dcache_bus_if[i].req_data.byteen ? 3'd0 /*PutFull*/ : 3'd1 /*PutPartial*/)
-                : 3'd4 /*Get*/;
-            assign dmem_a_bits_size[i * 4 +: 4] = 4'd2; // Fixed size
-            assign dmem_a_bits_mask[i * 4 +: 4] = dcache_bus_if[i].req_data.byteen;
-
-            // Setting request ready signal for dcache
-            assign dcache_bus_if[i].req_ready = dmem_a_ready[i];
-
-            // Data and tag assignment for smem
-            assign smem_bus_if[i].rsp_valid =
-                (smem_d_valid[i] && (smem_d_bits_opcode[i * 3 +: 3] !== 3'd0 /*AccessAck*/));
-            assign smem_bus_if[i].rsp_data.data = smem_d_bits_data[i * 32 +: 32];
-            assign smem_bus_if[i].rsp_data.tag = smem_d_bits_source[i * DCACHE_NOSM_TAG_WIDTH +: DCACHE_NOSM_TAG_WIDTH];
-
-            // Handling write ACKs, setting ready bit for smem
-            assign smem_d_ready[i] = smem_bus_if[i].rsp_ready ||
-                                    (smem_d_valid[i] && (smem_d_bits_opcode[i * 3 +: 3] == 3'd0 /*AccessAck*/));
-
-            // Request validity and address/data/source assignment for smem
-            assign smem_a_valid[i] = smem_bus_if[i].req_valid;
-            assign smem_a_bits_address[i * 32 +: 32] = {smem_bus_if[i].req_data.addr, 2'b0};
-            assign smem_a_bits_data[i * 32 +: 32] = smem_bus_if[i].req_data.data;
-            assign smem_a_bits_source[i * DCACHE_NOSM_TAG_WIDTH +: DCACHE_NOSM_TAG_WIDTH] = smem_bus_if[i].req_data.tag;
-
-            // Opcode, size, and mask assignment for smem
-            assign smem_a_bits_opcode[i * 3 +: 3] =
-                smem_bus_if[i].req_data.rw ?
-                (&smem_bus_if[i].req_data.byteen ? 3'd0 /*PutFull*/ : 3'd1 /*PutPartial*/)
-                : 3'd4 /*Get*/;
-            assign smem_a_bits_size[i * 4 +: 4] = 4'd2; // Fixed size
-            assign smem_a_bits_mask[i * 4 +: 4] = smem_bus_if[i].req_data.byteen;
-
-            // Setting request ready signal for smem
-            assign smem_bus_if[i].req_ready = smem_a_ready[i];
-        end
-    endgenerate
-
-    // tc ---------------------------------------------------------------------
-    VX_tc_bus_if #(.TAG_WIDTH(TC_TAG_WIDTH)) tc_p0_bus_if[NUM_TENSOR_CORES]();
-    VX_tc_bus_if #(.TAG_WIDTH(TC_TAG_WIDTH)) tc_p2_bus_if[NUM_TENSOR_CORES]();
-
-    for (genvar tc = 0; tc < NUM_TENSOR_CORES; ++tc) begin : g_tc_ports
-        localparam P0 = tc * 3;
-        localparam P1 = tc * 3 + 1;
-        localparam P2 = tc * 3 + 2;
-
-        assign tc_a_valid[P0] = tc_p0_bus_if[tc].req_valid;
-        assign tc_a_valid[P1] = 1'b0;
-        assign tc_a_valid[P2] = tc_p2_bus_if[tc].req_valid;
-
-        assign tc_a_bits_write[P0] = tc_p0_bus_if[tc].req_data.rw;
-        assign tc_a_bits_write[P1] = 1'b0;
-        assign tc_a_bits_write[P2] = tc_p2_bus_if[tc].req_data.rw;
-
-        assign tc_a_bits_address[P0 * 32 +: 32] = tc_p0_bus_if[tc].req_data.addr;
-        assign tc_a_bits_address[P1 * 32 +: 32] = 32'b0;
-        assign tc_a_bits_address[P2 * 32 +: 32] = tc_p2_bus_if[tc].req_data.addr;
-
-        assign tc_a_bits_tag[P0 * TC_TAG_WIDTH +: TC_TAG_WIDTH] = tc_p0_bus_if[tc].req_data.tag;
-        assign tc_a_bits_tag[P1 * TC_TAG_WIDTH +: TC_TAG_WIDTH] = '0;
-        assign tc_a_bits_tag[P2 * TC_TAG_WIDTH +: TC_TAG_WIDTH] = tc_p2_bus_if[tc].req_data.tag;
-
-        assign tc_a_bits_mask[P0 * 32 +: 32] = tc_p0_bus_if[tc].req_data.byteen;
-        assign tc_a_bits_mask[P1 * 32 +: 32] = '0;
-        assign tc_a_bits_mask[P2 * 32 +: 32] = tc_p2_bus_if[tc].req_data.byteen;
-
-        assign tc_a_bits_data[P0 * TC_DATA_WIDTH +: TC_DATA_WIDTH] = tc_p0_bus_if[tc].req_data.data;
-        assign tc_a_bits_data[P1 * TC_DATA_WIDTH +: TC_DATA_WIDTH] = '0;
-        assign tc_a_bits_data[P2 * TC_DATA_WIDTH +: TC_DATA_WIDTH] = tc_p2_bus_if[tc].req_data.data;
-
-        assign tc_p0_bus_if[tc].req_ready = tc_a_ready[P0];
-        assign tc_p0_bus_if[tc].rsp_valid = tc_d_valid[P0];
-        assign tc_p0_bus_if[tc].rsp_data.data = tc_d_bits_data[P0 * TC_DATA_WIDTH +: TC_DATA_WIDTH];
-        assign tc_p0_bus_if[tc].rsp_data.tag = tc_d_bits_tag[P0 * TC_TAG_WIDTH +: TC_TAG_WIDTH];
-
-        assign tc_p2_bus_if[tc].req_ready = tc_a_ready[P2];
-        assign tc_p2_bus_if[tc].rsp_valid = tc_d_valid[P2];
-        assign tc_p2_bus_if[tc].rsp_data.data = tc_d_bits_data[P2 * TC_DATA_WIDTH +: TC_DATA_WIDTH];
-        assign tc_p2_bus_if[tc].rsp_data.tag = tc_d_bits_tag[P2 * TC_TAG_WIDTH +: TC_TAG_WIDTH];
-
-        assign tc_d_ready[P0] = tc_p0_bus_if[tc].rsp_ready;
-        assign tc_d_ready[P1] = 1'b0;
-        assign tc_d_ready[P2] = tc_p2_bus_if[tc].rsp_ready;
-    end
-
-    // gbar -------------------------------------------------------------------
-`ifdef GBAR_ENABLE
-    VX_gbar_bus_if gbar_bus_if();
-    assign gbar_req_valid = gbar_bus_if.req_valid;
-    assign gbar_req_id = gbar_bus_if.req_id;
-    assign gbar_req_size_m1 = gbar_bus_if.req_size_m1;
-    assign gbar_req_core_id = gbar_bus_if.req_core_id;
-    assign gbar_bus_if.req_ready = gbar_req_ready;
-    assign gbar_bus_if.rsp_valid = gbar_rsp_valid;
-    assign gbar_bus_if.rsp_id = gbar_rsp_id;
-`endif
-
-    // fpu --------------------------------------------------------------------
-
-    // assign {fpu_hartid, fpu_time, fpu_inst, fpu_fromint_data, fpu_fcsr_rm, fpu_dmem_resp_val, fpu_dmem_resp_type,
-    //         fpu_dmem_resp_tag, fpu_valid, fpu_killx, fpu_killm, fpu_keep_clock_enabled} = '0;
-
-    logic sim_ebreak;
-    logic [`NUM_REGS-1:0][`XLEN-1:0] sim_wb_value;
-
-    logic [3:0] reset_start_counter;
-    logic core_reset;
-
-    always @(posedge clock) begin
-      if (reset) begin
-        reset_start_counter <= 4'ha;
-      end else begin
-        if (reset_start_counter > 4'h0) begin
-          reset_start_counter <= reset_start_counter - 4'h1;
-        end
-      end
-    end
-    // Delay reset signal by a few cycles to make time for resetting the DCR
-    // (device configuration registers).
-    assign core_reset = reset || (reset_start_counter != 4'h0); // || intr_reset;
-
-    // A small FSM that tries to set DCR "properly" in the same order as
-    // defined in VX_types.vh.
-    //
-    // DCR is a device configuration register that holds (among other things)
-    // the startup address for the kernel, nominally set to 0x80000000.
-    // TODO: Original Vortex code buffers dcr_bus by one cycle when
-    // SOCKET_SIZE > 1, as below.  Might want to check if we need to do the
-    // same
-    //   `BUFFER_DCR_BUS_IF (core_dcr_bus_if, dcr_bus_if, (`SOCKET_SIZE > 1));
-    logic [`VX_DCR_ADDR_BITS-1:0]  dcr_state;
-    logic [`VX_DCR_ADDR_BITS-1:0]  dcr_state_n;
-    logic                          dcr_write_valid;
-    logic [`VX_DCR_ADDR_WIDTH-1:0] dcr_write_addr;
-    logic [`VX_DCR_DATA_WIDTH-1:0] dcr_write_data;
-    always @(posedge clock) begin
-      if (reset) begin
-        dcr_state <= `VX_DCR_ADDR_BITS'h000;
-      end else begin
-        dcr_state <= dcr_state_n;
-      end
-    end
-    always @(*) begin
-      dcr_state_n = dcr_state;
-      dcr_write_valid = 1'b0;
-      dcr_write_addr = `VX_DCR_ADDR_WIDTH'b0;
-      dcr_write_data = `VX_DCR_DATA_WIDTH'b0;
-
-      case (dcr_state)
-        `VX_DCR_ADDR_BITS'h000: begin
-          dcr_state_n = `VX_DCR_BASE_STATE_BEGIN;
-        end
-        `VX_DCR_BASE_STATE_BEGIN: begin
-          dcr_state_n = `VX_DCR_BASE_STARTUP_ADDR1;
-
-          dcr_write_valid = 1'b1;
-          dcr_write_addr = `VX_DCR_BASE_STARTUP_ADDR0;
-          dcr_write_data = STARTUP_ADDR[31:0];
-        end
-        `VX_DCR_BASE_STARTUP_ADDR1: begin
-          dcr_state_n = `VX_DCR_BASE_MPM_CLASS;
-
-          dcr_write_valid = 1'b1;
-          dcr_write_addr = `VX_DCR_BASE_STARTUP_ADDR1;
-          dcr_write_data = STARTUP_ADDR[63:32];
-        end
-        `VX_DCR_BASE_MPM_CLASS: begin
-          dcr_state_n = `VX_DCR_BASE_STATE_END;
-
-          dcr_write_valid = 1'b1;
-          dcr_write_addr = `VX_DCR_BASE_MPM_CLASS;
-          dcr_write_data = `VX_DCR_DATA_WIDTH'h0;
-        end
-        `VX_DCR_BASE_STATE_END: begin
-          dcr_state_n = dcr_state;
-
-          dcr_write_valid = 1'b0;
-        end
-      endcase
-    end
-
-    VX_dcr_bus_if dcr_bus_if();
-    assign dcr_bus_if.write_valid = dcr_write_valid;
-    assign dcr_bus_if.write_addr  = dcr_write_addr;
-    assign dcr_bus_if.write_data  = dcr_write_data;
-
-    VX_mem_perf_if mem_perf_if();
-
-    // TODO: SCOPE_IO_BIND should be socket id
-    VX_core #(
-        .CORE_ID (CORE_ID),
-        .TENSOR_FP16 (TENSOR_FP16),
-        .NUM_TENSOR_CORES (NUM_TENSOR_CORES)
-    ) core (
-        `SCOPE_IO_BIND  (0)
-
-        .clk            (clock),
-        .reset          (core_reset),
-
-    `ifdef PERF_ENABLE
-        // NOTE unused
-        .mem_perf_if    (mem_perf_if),
-    `endif
-        
-        .dcr_bus_if     (dcr_bus_if),
-
-        .smem_bus_if    (smem_bus_if),
-
-        .dcache_bus_if  (dcache_bus_if),
-
-        .icache_bus_if  (icache_bus_if),
-
-    `ifdef GBAR_ENABLE
-        .gbar_bus_if    (gbar_bus_if),
-    `endif
-
-        .tensor_smem_A_if (tc_p0_bus_if),
-    `ifdef EXT_T_BLACKWELL
-        .tensor_tmem_A_ren(tc_tmem_A_ren),
-        .tensor_tmem_A_rready(tc_tmem_A_rready),
-        .tensor_tmem_A_raddr(tc_tmem_A_raddr),
-        .tensor_tmem_A_rdata(tc_tmem_A_rdata),
-        .tensor_tmem_C_ren(tc_tmem_C_ren),
-        .tensor_tmem_C_rready(tc_tmem_C_rready),
-        .tensor_tmem_C_raddr(tc_tmem_C_raddr),
-        .tensor_tmem_C_rdata(tc_tmem_C_rdata),
-        .tensor_tmem_C_wen(tc_tmem_C_wen),
-        .tensor_tmem_C_wready(tc_tmem_C_wready),
-        .tensor_tmem_C_waddr(tc_tmem_C_waddr),
-        .tensor_tmem_C_wdata(tc_tmem_C_wdata),
-        .tensor_tmem_C_mask(tc_tmem_C_mask),
-        .tensor_smem_B_if (tc_p2_bus_if),
-    `else
-        .tensor_tmem_A_ren(tc_tmem_A_ren),
-        .tensor_tmem_A_rready(tc_tmem_A_rready),
-        .tensor_tmem_A_raddr(tc_tmem_A_raddr),
-        .tensor_tmem_A_rdata(tc_tmem_A_rdata),
-        .tensor_tmem_C_ren(tc_tmem_C_ren),
-        .tensor_tmem_C_rready(tc_tmem_C_rready),
-        .tensor_tmem_C_raddr(tc_tmem_C_raddr),
-        .tensor_tmem_C_rdata(tc_tmem_C_rdata),
-        .tensor_tmem_C_wen(tc_tmem_C_wen),
-        .tensor_tmem_C_wready(tc_tmem_C_wready),
-        .tensor_tmem_C_waddr(tc_tmem_C_waddr),
-        .tensor_tmem_C_wdata(tc_tmem_C_wdata),
-        .tensor_tmem_C_mask(tc_tmem_C_mask),
-        .tensor_smem_B_if (tc_p2_bus_if),
-    `endif
-
-        .sim_ebreak     (sim_ebreak),
-        .sim_wb_value   (sim_wb_value),
-        .busy           (busy),
-        .downstream_mem_busy(downstream_mem_busy),
-
-        .acc_read_in    (acc_read_in),
-        .acc_write_out  (acc_write_out),
-        .acc_write_en   (acc_write_en)
-    );
-
-    // VX_dcache_req_if #(
-    //     .NUM_REQS  (`DCACHE_NUM_REQS),
-    //     .WORD_SIZE (`DCACHE_WORD_SIZE),
-    //     .TAG_WIDTH (`DCACHE_CORE_TAG_WIDTH)
-    // ) dcache_req_if();
-
-    // VX_dcache_rsp_if #(
-    //     .NUM_REQS  (`DCACHE_NUM_REQS),
-    //     .WORD_SIZE (`DCACHE_WORD_SIZE), 
-    //     .TAG_WIDTH (`DCACHE_CORE_TAG_WIDTH)
-    // ) dcache_rsp_if();
-    // 
-    // VX_icache_req_if #(
-    //     .WORD_SIZE (`ICACHE_WORD_SIZE), 
-    //     .TAG_WIDTH (`ICACHE_CORE_TAG_WIDTH)
-    // ) icache_req_if();
-
-    // VX_icache_rsp_if #(
-    //     .WORD_SIZE (`ICACHE_WORD_SIZE), 
-    //     .TAG_WIDTH (`ICACHE_CORE_TAG_WIDTH)
-    // ) icache_rsp_if();
-
-    // VX_pipeline #(
-    //     .CORE_ID(CORE_ID)
-    // ) pipeline (
-    //     `SCOPE_BIND_VX_core_pipeline
-    // `ifdef PERF_ENABLE
-    //     .perf_memsys_if (perf_memsys_if),
-    // `endif
-
-    //     .clk(clock),
-    //     .reset(reset || intr_reset),
-
-    //     .irq(1'b0/*intr_reset*/),
-
-    //     // Dcache core request
-    //     .dcache_req_valid   (dcache_req_if.valid),
-    //     .dcache_req_rw      (dcache_req_if.rw),
-    //     .dcache_req_byteen  (dcache_req_if.byteen),
-    //     .dcache_req_addr    (dcache_req_if.addr),
-    //     .dcache_req_data    (dcache_req_if.data),
-    //     .dcache_req_tag     (dcache_req_if.tag),
-    //     .dcache_req_ready   (dcache_req_if.ready),
-
-    //     // Dcache core reponse    
-    //     .dcache_rsp_valid   (dcache_rsp_if.valid),
-    //     .dcache_rsp_tmask   (dcache_rsp_if.tmask),
-    //     .dcache_rsp_data    (dcache_rsp_if.data),
-    //     .dcache_rsp_tag     (dcache_rsp_if.tag),
-    //     .dcache_rsp_ready   (dcache_rsp_if.ready),
-
-    //     // Icache core request
-    //     .icache_req_valid   (icache_req_if.valid),
-    //     .icache_req_addr    (icache_req_if.addr),
-    //     .icache_req_tag     (icache_req_if.tag),
-    //     .icache_req_ready   (icache_req_if.ready),
-
-    //     // Icache core reponse    
-    //     .icache_rsp_valid   (icache_rsp_if.valid),
-    //     .icache_rsp_data    (icache_rsp_if.data),
-    //     .icache_rsp_tag     (icache_rsp_if.tag),
-    //     .icache_rsp_ready   (icache_rsp_if.ready),
-
-    //     // Status
-    //     .busy(busy)
-    // );
-
-    logic [31:0] finish_counter;
-
-    always @(posedge clock) begin
-      if (reset) begin
-        finish_counter <= 32'd0;
-      end else begin
-        if (finished) begin
-          finish_counter <= finish_counter + 32'd1;
-        end
-      end
-    end
-
-    // give slack for other cores to finish
-    wire all_cores_finished = (finish_counter > 32'd10000);
-
-`ifdef SIMULATION
-    always @(posedge clock) begin
-        if (!reset) begin
-            if ((CORE_ID == '0) && all_cores_finished) begin
-                $display("simulation has probably ended. exiting");
-                $finish();
-            end
-            if (busy_prev && !busy) begin
-                $display("---------------- core%2d has no more active warps ----------------", CORE_ID);
-                // TODO: lane assumed to be 4
-                // `ifndef SYNTHESIS
-                // for (integer j = 0; j < `NUM_WARPS; j++) begin
-                //     $display("warp %2d", j);
-                //     for (integer k = 0; k < `NUM_REGS; k += 1)
-                //         $display("x%2d: %08x  %08x  %08x  %08x", k,
-                //             pipeline.issue.gpr_stage.iports[/*thread*/0].dp_ram1.not_out_reg.reg_dump.ram[j * `NUM_REGS + k],
-                //             pipeline.issue.gpr_stage.iports[/*thread*/1].dp_ram1.not_out_reg.reg_dump.ram[j * `NUM_REGS + k],
-                //             pipeline.issue.gpr_stage.iports[/*thread*/2].dp_ram1.not_out_reg.reg_dump.ram[j * `NUM_REGS + k],
-                //             pipeline.issue.gpr_stage.iports[/*thread*/3].dp_ram1.not_out_reg.reg_dump.ram[j * `NUM_REGS + k]);
-                //     end
-                // `endif
-                // @(posedge clock) $finish();
-            end
-        end
-    end
-`endif
-
-endmodule : Vortex

hw/rtl/VX_csr_data.sv

@@ -0,0 +1,265 @@
+`include "VX_define.vh"
+
+module VX_csr_data #(
+    parameter CORE_ID = 0
+) (
+    input wire clk,
+    input wire reset,
+
+`ifdef PERF_ENABLE
+`ifdef EXT_TEX_ENABLE
+    VX_perf_tex_if.slave            perf_tex_if,
+`endif
+    VX_perf_memsys_if.slave         perf_memsys_if,
+    VX_perf_pipeline_if.slave       perf_pipeline_if,
+`endif
+
+    VX_cmt_to_csr_if.slave          cmt_to_csr_if,
+    VX_fetch_to_csr_if.slave        fetch_to_csr_if,
+
+`ifdef EXT_F_ENABLE
+    VX_fpu_to_csr_if.slave          fpu_to_csr_if,
+`endif
+`ifdef EXT_TEX_ENABLE
+    VX_tex_csr_if.master            tex_csr_if,
+`endif 
+
+    input wire                      read_enable,
+    input wire [`UUID_BITS-1:0]     read_uuid,
+    input wire[`CSR_ADDR_BITS-1:0]  read_addr,
+    input wire[`NW_BITS-1:0]        read_wid,
+    output wire[31:0]               read_data,
+
+    input wire                      write_enable, 
+    input wire [`UUID_BITS-1:0]     write_uuid,
+    input wire[`CSR_ADDR_BITS-1:0]  write_addr,
+    input wire[`NW_BITS-1:0]        write_wid,
+    input wire[31:0]                write_data,
+    
+    input wire                      busy
+);
+    import fpu_types::*;
+    
+    reg [`CSR_WIDTH-1:0] csr_satp;
+    reg [`CSR_WIDTH-1:0] csr_mstatus;
+    reg [`CSR_WIDTH-1:0] csr_medeleg;
+    reg [`CSR_WIDTH-1:0] csr_mideleg;
+    reg [`CSR_WIDTH-1:0] csr_mie;
+    reg [`CSR_WIDTH-1:0] csr_mtvec;
+    reg [`CSR_WIDTH-1:0] csr_mepc;    
+    reg [`CSR_WIDTH-1:0] csr_pmpcfg [0:0];
+    reg [`CSR_WIDTH-1:0] csr_pmpaddr [0:0];
+    reg [63:0] csr_cycle;
+    reg [63:0] csr_instret;
+    
+    reg [`NUM_WARPS-1:0][`INST_FRM_BITS+`FFLAGS_BITS-1:0] fcsr;
+
+    always @(posedge clk) begin
+        if (reset) begin
+            fcsr <= '0;
+        end else begin
+        `ifdef EXT_F_ENABLE
+            if (fpu_to_csr_if.write_enable) begin
+                fcsr[fpu_to_csr_if.write_wid][`FFLAGS_BITS-1:0] <= fcsr[fpu_to_csr_if.write_wid][`FFLAGS_BITS-1:0]
+                                                                 | fpu_to_csr_if.write_fflags;
+            end
+        `endif
+            if (write_enable) begin
+                case (write_addr)
+                    `CSR_FFLAGS:   fcsr[write_wid][`FFLAGS_BITS-1:0] <= write_data[`FFLAGS_BITS-1:0];
+                    `CSR_FRM:      fcsr[write_wid][`INST_FRM_BITS+`FFLAGS_BITS-1:`FFLAGS_BITS] <= write_data[`INST_FRM_BITS-1:0];
+                    `CSR_FCSR:     fcsr[write_wid] <= write_data[`FFLAGS_BITS+`INST_FRM_BITS-1:0];
+                    `CSR_SATP:     csr_satp       <= write_data[`CSR_WIDTH-1:0];
+                    `CSR_MSTATUS:  csr_mstatus    <= write_data[`CSR_WIDTH-1:0];
+                    `CSR_MEDELEG:  csr_medeleg    <= write_data[`CSR_WIDTH-1:0];
+                    `CSR_MIDELEG:  csr_mideleg    <= write_data[`CSR_WIDTH-1:0];
+                    `CSR_MIE:      csr_mie        <= write_data[`CSR_WIDTH-1:0];
+                    `CSR_MTVEC:    csr_mtvec      <= write_data[`CSR_WIDTH-1:0];
+                    `CSR_MEPC:     csr_mepc       <= write_data[`CSR_WIDTH-1:0];
+                    `CSR_PMPCFG0:  csr_pmpcfg[0]  <= write_data[`CSR_WIDTH-1:0];
+                    `CSR_PMPADDR0: csr_pmpaddr[0] <= write_data[`CSR_WIDTH-1:0];
+                    default: begin
+                    `ifdef EXT_TEX_ENABLE
+                        `ASSERT((write_addr == `CSR_TEX_UNIT)
+                             || (write_addr >= `CSR_TEX_STATE_BEGIN 
+                              && write_addr < `CSR_TEX_STATE_END),
+                                ("%t: *** invalid CSR write address: %0h (#%0d)", $time, write_addr, write_uuid));
+                    `else
+                        `ASSERT(~write_enable, ("%t: *** invalid CSR write address: %0h (#%0d)", $time, write_addr, write_uuid));
+                    `endif
+                    end
+                endcase
+            end
+        end
+    end
+
+    `UNUSED_VAR (write_data)
+
+    // TEX CSRs
+`ifdef EXT_TEX_ENABLE    
+    assign tex_csr_if.write_enable = write_enable;
+    assign tex_csr_if.write_addr   = write_addr;
+    assign tex_csr_if.write_data   = write_data;
+    assign tex_csr_if.write_uuid   = write_uuid;
+`endif
+
+    always @(posedge clk) begin
+       if (reset) begin
+            csr_cycle   <= 0;
+            csr_instret <= 0;
+        end else begin
+            if (busy) begin
+                csr_cycle <= csr_cycle + 1;
+            end
+            if (cmt_to_csr_if.valid) begin
+                csr_instret <= csr_instret + 64'(cmt_to_csr_if.commit_size);
+            end
+        end
+    end
+
+    reg [31:0] read_data_r;
+    reg read_addr_valid_r;
+
+    always @(*) begin
+        read_data_r = 'x;
+        read_addr_valid_r = 1;
+        case (read_addr)
+            `CSR_FFLAGS     : read_data_r = 32'(fcsr[read_wid][`FFLAGS_BITS-1:0]);
+            `CSR_FRM        : read_data_r = 32'(fcsr[read_wid][`INST_FRM_BITS+`FFLAGS_BITS-1:`FFLAGS_BITS]);
+            `CSR_FCSR       : read_data_r = 32'(fcsr[read_wid]);
+
+            `CSR_WTID       ,            
+            `CSR_LTID       ,
+            `CSR_LWID       : read_data_r = 32'(read_wid);            
+            `CSR_GTID       ,
+            /*`CSR_MHARTID ,*/
+            `CSR_GWID       : read_data_r = CORE_ID * `NUM_WARPS + 32'(read_wid);
+            `CSR_GCID       : read_data_r = CORE_ID;
+
+            `CSR_TMASK      : read_data_r = 32'(fetch_to_csr_if.thread_masks[read_wid]);
+
+            `CSR_NT         : read_data_r = `NUM_THREADS;
+            `CSR_NW         : read_data_r = `NUM_WARPS;
+            `CSR_NC         : read_data_r = `NUM_CORES * `NUM_CLUSTERS;
+            
+            `CSR_MCYCLE     : read_data_r = csr_cycle[31:0];
+            `CSR_MCYCLE_H   : read_data_r = 32'(csr_cycle[`PERF_CTR_BITS-1:32]);
+            `CSR_MINSTRET   : read_data_r = csr_instret[31:0];
+            `CSR_MINSTRET_H : read_data_r = 32'(csr_instret[`PERF_CTR_BITS-1:32]);
+            
+        `ifdef PERF_ENABLE
+            // PERF: pipeline
+            `CSR_MPM_IBUF_ST    : read_data_r = perf_pipeline_if.ibf_stalls[31:0];
+            `CSR_MPM_IBUF_ST_H  : read_data_r = 32'(perf_pipeline_if.ibf_stalls[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_SCRB_ST    : read_data_r = perf_pipeline_if.scb_stalls[31:0];
+            `CSR_MPM_SCRB_ST_H  : read_data_r = 32'(perf_pipeline_if.scb_stalls[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_ALU_ST     : read_data_r = perf_pipeline_if.alu_stalls[31:0];
+            `CSR_MPM_ALU_ST_H   : read_data_r = 32'(perf_pipeline_if.alu_stalls[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_LSU_ST     : read_data_r = perf_pipeline_if.lsu_stalls[31:0];
+            `CSR_MPM_LSU_ST_H   : read_data_r = 32'(perf_pipeline_if.lsu_stalls[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_CSR_ST     : read_data_r = perf_pipeline_if.csr_stalls[31:0];
+            `CSR_MPM_CSR_ST_H   : read_data_r = 32'(perf_pipeline_if.csr_stalls[`PERF_CTR_BITS-1:32]);
+        `ifdef EXT_F_ENABLE    
+            `CSR_MPM_FPU_ST     : read_data_r = perf_pipeline_if.fpu_stalls[31:0];
+            `CSR_MPM_FPU_ST_H   : read_data_r = 32'(perf_pipeline_if.fpu_stalls[`PERF_CTR_BITS-1:32]);
+        `else        
+            `CSR_MPM_FPU_ST     : read_data_r = '0;
+            `CSR_MPM_FPU_ST_H   : read_data_r = '0;
+        `endif
+            `CSR_MPM_GPU_ST     : read_data_r = perf_pipeline_if.gpu_stalls[31:0];
+            `CSR_MPM_GPU_ST_H   : read_data_r = 32'(perf_pipeline_if.gpu_stalls[`PERF_CTR_BITS-1:32]);
+            // PERF: decode
+            `CSR_MPM_LOADS      : read_data_r = perf_pipeline_if.loads[31:0];
+            `CSR_MPM_LOADS_H    : read_data_r = 32'(perf_pipeline_if.loads[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_STORES     : read_data_r = perf_pipeline_if.stores[31:0];
+            `CSR_MPM_STORES_H   : read_data_r = 32'(perf_pipeline_if.stores[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_BRANCHES   : read_data_r = perf_pipeline_if.branches[31:0];
+            `CSR_MPM_BRANCHES_H : read_data_r = 32'(perf_pipeline_if.branches[`PERF_CTR_BITS-1:32]);
+            // PERF: icache
+            `CSR_MPM_ICACHE_READS       : read_data_r = perf_memsys_if.icache_reads[31:0];
+            `CSR_MPM_ICACHE_READS_H     : read_data_r = 32'(perf_memsys_if.icache_reads[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_ICACHE_MISS_R      : read_data_r = perf_memsys_if.icache_read_misses[31:0];
+            `CSR_MPM_ICACHE_MISS_R_H    : read_data_r = 32'(perf_memsys_if.icache_read_misses[`PERF_CTR_BITS-1:32]);
+            // PERF: dcache
+            `CSR_MPM_DCACHE_READS       : read_data_r = perf_memsys_if.dcache_reads[31:0];
+            `CSR_MPM_DCACHE_READS_H     : read_data_r = 32'(perf_memsys_if.dcache_reads[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_DCACHE_WRITES      : read_data_r = perf_memsys_if.dcache_writes[31:0];
+            `CSR_MPM_DCACHE_WRITES_H    : read_data_r = 32'(perf_memsys_if.dcache_writes[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_DCACHE_MISS_R      : read_data_r = perf_memsys_if.dcache_read_misses[31:0];
+            `CSR_MPM_DCACHE_MISS_R_H    : read_data_r = 32'(perf_memsys_if.dcache_read_misses[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_DCACHE_MISS_W      : read_data_r = perf_memsys_if.dcache_write_misses[31:0];
+            `CSR_MPM_DCACHE_MISS_W_H    : read_data_r = 32'(perf_memsys_if.dcache_write_misses[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_DCACHE_BANK_ST     : read_data_r = perf_memsys_if.dcache_bank_stalls[31:0];
+            `CSR_MPM_DCACHE_BANK_ST_H   : read_data_r = 32'(perf_memsys_if.dcache_bank_stalls[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_DCACHE_MSHR_ST     : read_data_r = perf_memsys_if.dcache_mshr_stalls[31:0];
+            `CSR_MPM_DCACHE_MSHR_ST_H   : read_data_r = 32'(perf_memsys_if.dcache_mshr_stalls[`PERF_CTR_BITS-1:32]);
+            // PERF: smem          
+            `CSR_MPM_SMEM_READS     : read_data_r = perf_memsys_if.smem_reads[31:0];
+            `CSR_MPM_SMEM_READS_H   : read_data_r = 32'(perf_memsys_if.smem_reads[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_SMEM_WRITES    : read_data_r = perf_memsys_if.smem_writes[31:0];
+            `CSR_MPM_SMEM_WRITES_H  : read_data_r = 32'(perf_memsys_if.smem_writes[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_SMEM_BANK_ST   : read_data_r = perf_memsys_if.smem_bank_stalls[31:0];
+            `CSR_MPM_SMEM_BANK_ST_H : read_data_r = 32'(perf_memsys_if.smem_bank_stalls[`PERF_CTR_BITS-1:32]);
+            // PERF: memory
+            `CSR_MPM_MEM_READS      : read_data_r = perf_memsys_if.mem_reads[31:0];
+            `CSR_MPM_MEM_READS_H    : read_data_r = 32'(perf_memsys_if.mem_reads[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_MEM_WRITES     : read_data_r = perf_memsys_if.mem_writes[31:0];
+            `CSR_MPM_MEM_WRITES_H   : read_data_r = 32'(perf_memsys_if.mem_writes[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_MEM_LAT        : read_data_r = perf_memsys_if.mem_latency[31:0];
+            `CSR_MPM_MEM_LAT_H      : read_data_r = 32'(perf_memsys_if.mem_latency[`PERF_CTR_BITS-1:32]);
+        `ifdef EXT_TEX_ENABLE
+            // PERF: texunit
+            `CSR_MPM_TEX_READS      : read_data_r = perf_tex_if.mem_reads[31:0];
+            `CSR_MPM_TEX_READS_H    : read_data_r = 32'(perf_tex_if.mem_reads[`PERF_CTR_BITS-1:32]);
+            `CSR_MPM_TEX_LAT        : read_data_r = perf_tex_if.mem_latency[31:0];
+            `CSR_MPM_TEX_LAT_H      : read_data_r = 32'(perf_tex_if.mem_latency[`PERF_CTR_BITS-1:32]);
+        `endif
+            // PERF: reserved            
+            `CSR_MPM_RESERVED       : read_data_r = '0;
+            `CSR_MPM_RESERVED_H     : read_data_r = '0;
+        `endif
+            
+            `CSR_SATP      : read_data_r = 32'(csr_satp);
+            
+            `CSR_MSTATUS   : read_data_r = 32'(csr_mstatus);
+            `CSR_MISA      : read_data_r = `ISA_CODE;
+            `CSR_MEDELEG   : read_data_r = 32'(csr_medeleg);
+            `CSR_MIDELEG   : read_data_r = 32'(csr_mideleg);
+            `CSR_MIE       : read_data_r = 32'(csr_mie);
+            `CSR_MTVEC     : read_data_r = 32'(csr_mtvec);
+
+            `CSR_MEPC      : read_data_r = 32'(csr_mepc);
+
+            `CSR_PMPCFG0   : read_data_r = 32'(csr_pmpcfg[0]);
+            `CSR_PMPADDR0  : read_data_r = 32'(csr_pmpaddr[0]);
+            
+            `CSR_MVENDORID : read_data_r = `VENDOR_ID;
+            `CSR_MARCHID   : read_data_r = `ARCHITECTURE_ID;
+            `CSR_MIMPID    : read_data_r = `IMPLEMENTATION_ID;
+
+            default: begin
+                if ((read_addr >= `CSR_MPM_BASE && read_addr < (`CSR_MPM_BASE + 32))
+                 || (read_addr >= `CSR_MPM_BASE_H && read_addr < (`CSR_MPM_BASE_H + 32))) begin
+                     read_addr_valid_r = 1;
+                end else     
+            `ifdef EXT_TEX_ENABLE    
+                if ((read_addr == `CSR_TEX_UNIT)
+                 || (read_addr >= `CSR_TEX_STATE_BEGIN
+                  && read_addr < `CSR_TEX_STATE_END)) begin
+                    read_addr_valid_r = 1;
+                end else
+            `endif
+                    read_addr_valid_r = 0;
+            end
+        endcase
+    end 
+
+    `RUNTIME_ASSERT(~read_enable || read_addr_valid_r, ("%t: *** invalid CSR read address: %0h (#%0d)", $time, read_addr, read_uuid))
+
+    assign read_data = read_data_r;
+
+`ifdef EXT_F_ENABLE    
+    assign fpu_to_csr_if.read_frm = fcsr[fpu_to_csr_if.read_wid][`INST_FRM_BITS+`FFLAGS_BITS-1:`FFLAGS_BITS];
+`endif
+
+endmodule

hw/rtl/VX_csr_unit.sv

@@ -0,0 +1,151 @@
+`include "VX_define.vh"
+
+module VX_csr_unit #(
+    parameter CORE_ID = 0
+) (
+    input wire                  clk,
+    input wire                  reset,
+
+`ifdef PERF_ENABLE
+`ifdef EXT_TEX_ENABLE
+    VX_perf_tex_if.slave        perf_tex_if,
+`endif
+    VX_perf_memsys_if.slave     perf_memsys_if,
+    VX_perf_pipeline_if.slave   perf_pipeline_if,
+`endif
+
+    VX_cmt_to_csr_if.slave      cmt_to_csr_if,
+    VX_fetch_to_csr_if.slave    fetch_to_csr_if,
+    VX_csr_req_if.slave         csr_req_if,
+    VX_commit_if.master         csr_commit_if,
+    
+`ifdef EXT_F_ENABLE
+    VX_fpu_to_csr_if.slave      fpu_to_csr_if,
+    input wire[`NUM_WARPS-1:0]  fpu_pending,
+`endif
+`ifdef EXT_TEX_ENABLE
+    VX_tex_csr_if.master        tex_csr_if,
+`endif
+
+    output wire[`NUM_WARPS-1:0] pending,
+    input wire                  busy
+);    
+    wire csr_we_s1;
+    wire [`CSR_ADDR_BITS-1:0] csr_addr_s1;    
+    wire [31:0] csr_read_data;
+    wire [31:0] csr_read_data_s1;
+    wire [31:0] csr_updated_data_s1;  
+
+    wire write_enable = csr_commit_if.valid && csr_we_s1;
+
+    wire [31:0] csr_req_data = csr_req_if.use_imm ? 32'(csr_req_if.imm) : csr_req_if.rs1_data;
+
+    VX_csr_data #(
+        .CORE_ID(CORE_ID)
+    ) csr_data (
+        .clk            (clk),
+        .reset          (reset),
+    `ifdef PERF_ENABLE
+    `ifdef EXT_TEX_ENABLE
+        .perf_tex_if    (perf_tex_if),
+    `endif
+        .perf_memsys_if (perf_memsys_if),
+        .perf_pipeline_if(perf_pipeline_if),
+    `endif
+        .cmt_to_csr_if  (cmt_to_csr_if),
+        .fetch_to_csr_if(fetch_to_csr_if),
+    `ifdef EXT_F_ENABLE
+        .fpu_to_csr_if  (fpu_to_csr_if), 
+    `endif
+    `ifdef EXT_TEX_ENABLE
+        .tex_csr_if     (tex_csr_if),
+    `endif
+        .read_enable    (csr_req_if.valid),
+        .read_uuid      (csr_req_if.uuid),
+        .read_addr      (csr_req_if.addr),
+        .read_wid       (csr_req_if.wid),      
+        .read_data      (csr_read_data),
+        .write_enable   (write_enable),        
+        .write_uuid     (csr_commit_if.uuid),
+        .write_addr     (csr_addr_s1), 
+        .write_wid      (csr_commit_if.wid),
+        .write_data     (csr_updated_data_s1),
+        .busy           (busy)
+    );    
+
+    wire write_hazard = (csr_addr_s1 == csr_req_if.addr)
+                     && (csr_commit_if.wid == csr_req_if.wid) 
+                     && csr_commit_if.valid;
+
+    wire [31:0] csr_read_data_qual = write_hazard ? csr_updated_data_s1 : csr_read_data; 
+
+    reg [31:0] csr_updated_data;
+    reg csr_we_s0_unqual; 
+    
+    always @(*) begin        
+        csr_we_s0_unqual = (csr_req_data != 0);
+        case (csr_req_if.op_type)
+            `INST_CSR_RW: begin
+                csr_updated_data = csr_req_data;
+                csr_we_s0_unqual = 1;
+            end
+            `INST_CSR_RS: begin
+                csr_updated_data = csr_read_data_qual | csr_req_data;
+            end
+            //`INST_CSR_RC
+            default: begin
+                csr_updated_data = csr_read_data_qual & ~csr_req_data;
+            end
+        endcase
+    end
+
+`ifdef EXT_F_ENABLE
+    wire stall_in = fpu_pending[csr_req_if.wid];
+`else 
+    wire stall_in = 0;
+`endif
+
+    wire csr_req_valid = csr_req_if.valid && !stall_in;  
+
+    wire stall_out = ~csr_commit_if.ready && csr_commit_if.valid;
+
+    VX_pipe_register #(
+        .DATAW  (1 + `UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `NR_BITS + 1 + 1 + `CSR_ADDR_BITS + 32 + 32),
+        .RESETW (1)
+    ) pipe_reg (
+        .clk      (clk),
+        .reset    (reset),
+        .enable   (!stall_out),
+        .data_in  ({csr_req_valid,       csr_req_if.uuid,    csr_req_if.wid,    csr_req_if.tmask,    csr_req_if.PC,    csr_req_if.rd,    csr_req_if.wb,    csr_we_s0_unqual, csr_req_if.addr, csr_read_data_qual, csr_updated_data}),
+        .data_out ({csr_commit_if.valid, csr_commit_if.uuid, csr_commit_if.wid, csr_commit_if.tmask, csr_commit_if.PC, csr_commit_if.rd, csr_commit_if.wb, csr_we_s1,        csr_addr_s1,     csr_read_data_s1,   csr_updated_data_s1})
+    );
+
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin
+        assign csr_commit_if.data[i] = (csr_addr_s1 == `CSR_WTID) ? i : 
+                                         (csr_addr_s1 == `CSR_LTID 
+                                       || csr_addr_s1 == `CSR_GTID) ? (csr_read_data_s1 * `NUM_THREADS + i) : 
+                                                                      csr_read_data_s1;
+    end         
+
+    assign csr_commit_if.eop = 1'b1;
+
+    // can accept new request?
+    assign csr_req_if.ready = ~(stall_out || stall_in);
+
+    // pending request
+    reg [`NUM_WARPS-1:0] pending_r;
+    always @(posedge clk) begin
+        if (reset) begin
+            pending_r <= 0;
+        end else begin
+            if (csr_commit_if.valid && csr_commit_if.ready) begin
+                 pending_r[csr_commit_if.wid] <= 0;
+            end          
+            if (csr_req_if.valid && csr_req_if.ready) begin
+                 pending_r[csr_req_if.wid] <= 1;
+            end
+        end
+    end
+    assign pending = pending_r;
+
+endmodule

hw/rtl/VX_decode.sv

@@ -0,0 +1,495 @@
+`include "VX_define.vh"
+`ifdef DBG_TRACE_CORE_PIPELINE
+`include "VX_trace_instr.vh"
+`endif
+
+`ifdef EXT_F_ENABLE
+    `define USED_IREG(r) \
+        r``_r = {1'b0, ``r}
+
+    `define USED_FREG(r) \
+        r``_r = {1'b1, ``r}
+`else
+    `define USED_IREG(r) \
+        r``_r = ``r
+`endif
+
+module VX_decode  #(
+    parameter CORE_ID = 0
+) (
+    input  wire         clk,
+    input  wire         reset,
+
+`ifdef PERF_ENABLE
+    VX_perf_pipeline_if.decode perf_decode_if,
+`endif
+
+    // inputs
+    VX_ifetch_rsp_if.slave ifetch_rsp_if,
+
+    // outputs      
+    VX_decode_if.master decode_if,
+    VX_wstall_if.master wstall_if,
+    VX_join_if.master   join_if
+);
+    `UNUSED_PARAM (CORE_ID)
+    `UNUSED_VAR (clk)
+    `UNUSED_VAR (reset)
+    
+    reg [`EX_BITS-1:0]  ex_type;    
+    reg [`INST_OP_BITS-1:0] op_type; 
+    reg [`INST_MOD_BITS-1:0] op_mod;
+    reg [`NR_BITS-1:0]  rd_r, rs1_r, rs2_r, rs3_r;
+    reg [31:0]          imm;    
+    reg use_rd, use_PC, use_imm;
+    reg is_join, is_wstall;
+
+    wire [31:0] instr = ifetch_rsp_if.data;
+    wire [6:0] opcode = instr[6:0];  
+    wire [1:0] func2  = instr[26:25];
+    wire [2:0] func3  = instr[14:12];
+    wire [6:0] func7  = instr[31:25];
+    wire [11:0] u_12  = instr[31:20]; 
+
+    wire [4:0] rd  = instr[11:7];
+    wire [4:0] rs1 = instr[19:15];
+    wire [4:0] rs2 = instr[24:20];
+    wire [4:0] rs3 = instr[31:27];
+
+    wire [19:0] upper_imm = {func7, rs2, rs1, func3};
+    wire [11:0] alu_imm   = (func3[0] && ~func3[1]) ? {{7{1'b0}}, rs2} : u_12;
+    wire [11:0] s_imm     = {func7, rd};
+    wire [12:0] b_imm     = {instr[31], instr[7], instr[30:25], instr[11:8], 1'b0};
+    wire [20:0] jal_imm   = {instr[31], instr[19:12], instr[20], instr[30:21], 1'b0};
+
+    `UNUSED_VAR (rs3)
+    
+    always @(*) begin
+
+        ex_type   = 0;
+        op_type   = 'x;
+        op_mod    = 0;
+        rd_r      = 0;
+        rs1_r     = 0;
+        rs2_r     = 0;
+        rs3_r     = 0;
+        imm       = 'x;
+        use_imm   = 0;
+        use_PC    = 0;
+        use_rd    = 0;
+        is_join   = 0;
+        is_wstall = 0;
+
+        case (opcode)            
+            `INST_I: begin
+                ex_type = `EX_ALU;
+                case (func3)
+                    3'h0: op_type = `INST_OP_BITS'(`INST_ALU_ADD);
+                    3'h1: op_type = `INST_OP_BITS'(`INST_ALU_SLL);
+                    3'h2: op_type = `INST_OP_BITS'(`INST_ALU_SLT);
+                    3'h3: op_type = `INST_OP_BITS'(`INST_ALU_SLTU);
+                    3'h4: op_type = `INST_OP_BITS'(`INST_ALU_XOR);
+                    3'h5: op_type = (func7[5]) ? `INST_OP_BITS'(`INST_ALU_SRA) : `INST_OP_BITS'(`INST_ALU_SRL);
+                    3'h6: op_type = `INST_OP_BITS'(`INST_ALU_OR);
+                    3'h7: op_type = `INST_OP_BITS'(`INST_ALU_AND);
+                    default:;
+                endcase
+                use_rd  = 1;
+                use_imm = 1;
+                imm     = {{20{alu_imm[11]}}, alu_imm};
+                `USED_IREG (rd);
+                `USED_IREG (rs1);
+            end
+            `INST_R: begin 
+                ex_type = `EX_ALU;
+            `ifdef EXT_F_ENABLE
+                if (func7[0]) begin
+                    case (func3)
+                        3'h0: op_type = `INST_OP_BITS'(`INST_MUL_MUL);
+                        3'h1: op_type = `INST_OP_BITS'(`INST_MUL_MULH);
+                        3'h2: op_type = `INST_OP_BITS'(`INST_MUL_MULHSU);
+                        3'h3: op_type = `INST_OP_BITS'(`INST_MUL_MULHU);
+                        3'h4: op_type = `INST_OP_BITS'(`INST_MUL_DIV);
+                        3'h5: op_type = `INST_OP_BITS'(`INST_MUL_DIVU);
+                        3'h6: op_type = `INST_OP_BITS'(`INST_MUL_REM);
+                        3'h7: op_type = `INST_OP_BITS'(`INST_MUL_REMU);
+                        default:; 
+                    endcase
+                    op_mod = 2;
+                end else 
+            `endif
+                begin
+                    case (func3)
+                        3'h0: op_type = (func7[5]) ? `INST_OP_BITS'(`INST_ALU_SUB) : `INST_OP_BITS'(`INST_ALU_ADD);
+                        3'h1: op_type = `INST_OP_BITS'(`INST_ALU_SLL);
+                        3'h2: op_type = `INST_OP_BITS'(`INST_ALU_SLT);
+                        3'h3: op_type = `INST_OP_BITS'(`INST_ALU_SLTU);
+                        3'h4: op_type = `INST_OP_BITS'(`INST_ALU_XOR);
+                        3'h5: op_type = (func7[5]) ? `INST_OP_BITS'(`INST_ALU_SRA) : `INST_OP_BITS'(`INST_ALU_SRL);
+                        3'h6: op_type = `INST_OP_BITS'(`INST_ALU_OR);
+                        3'h7: op_type = `INST_OP_BITS'(`INST_ALU_AND);
+                        default:;
+                    endcase
+                end          
+                use_rd = 1;
+                `USED_IREG (rd);
+                `USED_IREG (rs1);
+                `USED_IREG (rs2);
+            end
+            `INST_LUI: begin 
+                ex_type = `EX_ALU;
+                op_type = `INST_OP_BITS'(`INST_ALU_LUI);
+                use_rd  = 1;
+                use_imm = 1;
+                imm     = {upper_imm, 12'(0)};
+                `USED_IREG (rd);
+                rs1_r   = 0;
+            end
+            `INST_AUIPC: begin 
+                ex_type = `EX_ALU;
+                op_type = `INST_OP_BITS'(`INST_ALU_AUIPC);
+                use_rd  = 1;
+                use_imm = 1;
+                use_PC  = 1;
+                imm     = {upper_imm, 12'(0)};
+                `USED_IREG (rd);
+            end
+            `INST_JAL: begin 
+                ex_type = `EX_ALU;
+                op_type = `INST_OP_BITS'(`INST_BR_JAL);
+                op_mod  = 1;
+                use_rd  = 1;
+                use_imm = 1;
+                use_PC  = 1;
+                is_wstall = 1;
+                imm     = {{11{jal_imm[20]}}, jal_imm};
+                `USED_IREG (rd);
+            end
+            `INST_JALR: begin 
+                ex_type = `EX_ALU;
+                op_type = `INST_OP_BITS'(`INST_BR_JALR);
+                op_mod  = 1;
+                use_rd  = 1;
+                use_imm = 1;
+                is_wstall = 1;
+                imm     = {{20{u_12[11]}}, u_12};
+                `USED_IREG (rd);
+                `USED_IREG (rs1);
+            end
+            `INST_B: begin 
+                ex_type = `EX_ALU;
+                case (func3)
+                    3'h0: op_type = `INST_OP_BITS'(`INST_BR_EQ);
+                    3'h1: op_type = `INST_OP_BITS'(`INST_BR_NE);
+                    3'h4: op_type = `INST_OP_BITS'(`INST_BR_LT);
+                    3'h5: op_type = `INST_OP_BITS'(`INST_BR_GE);
+                    3'h6: op_type = `INST_OP_BITS'(`INST_BR_LTU);
+                    3'h7: op_type = `INST_OP_BITS'(`INST_BR_GEU);
+                    default:;
+                endcase
+                op_mod  = 1;
+                use_imm = 1;
+                use_PC  = 1;
+                is_wstall = 1;
+                imm     = {{19{b_imm[12]}}, b_imm};
+                `USED_IREG (rs1);
+                `USED_IREG (rs2);
+            end
+            `INST_FENCE: begin
+                ex_type = `EX_LSU;
+                op_mod  = `INST_MOD_BITS'(1);
+            end
+            `INST_SYS : begin 
+                if (func3[1:0] != 0) begin                    
+                    ex_type = `EX_CSR;
+                    op_type = `INST_OP_BITS'(func3[1:0]);
+                    use_rd  = 1;
+                    use_imm = func3[2]; 
+                    imm[`CSR_ADDR_BITS-1:0] = u_12; // addr
+                    `USED_IREG (rd);
+                    if (func3[2]) begin
+                        imm[`CSR_ADDR_BITS +: `NRI_BITS] = rs1; // imm
+                    end else begin
+                        `USED_IREG (rs1);
+                    end                    
+                end else begin
+                    ex_type = `EX_ALU;
+                    case (u_12)
+                        12'h000: op_type = `INST_OP_BITS'(`INST_BR_ECALL);
+                        12'h001: op_type = `INST_OP_BITS'(`INST_BR_EBREAK);             
+                        12'h002: op_type = `INST_OP_BITS'(`INST_BR_URET);                        
+                        12'h102: op_type = `INST_OP_BITS'(`INST_BR_SRET);                        
+                        12'h302: op_type = `INST_OP_BITS'(`INST_BR_MRET);
+                        default:;
+                    endcase
+                    op_mod  = 1;
+                    use_rd  = 1;
+                    use_imm = 1;
+                    use_PC  = 1;
+                    is_wstall = 1;
+                    imm     = 32'd4;
+                    `USED_IREG (rd);
+                end
+            end
+        `ifdef EXT_F_ENABLE
+            `INST_FL, 
+        `endif
+            `INST_L: begin 
+                ex_type = `EX_LSU;
+                op_type = `INST_OP_BITS'({1'b0, func3});
+                use_rd  = 1;
+                imm     = {{20{u_12[11]}}, u_12};
+            `ifdef EXT_F_ENABLE
+                if (opcode[2]) begin
+                    `USED_FREG (rd);
+                end else
+            `endif
+                `USED_IREG (rd);
+                `USED_IREG (rs1);
+            end
+        `ifdef EXT_F_ENABLE
+            `INST_FS, 
+        `endif
+            `INST_S: begin 
+                ex_type = `EX_LSU;
+                op_type = `INST_OP_BITS'({1'b1, func3});
+                imm     = {{20{s_imm[11]}}, s_imm};
+                `USED_IREG (rs1);
+            `ifdef EXT_F_ENABLE
+                if (opcode[2]) begin
+                    `USED_FREG (rs2);
+                end else
+            `endif
+                `USED_IREG (rs2);
+            end
+        `ifdef EXT_F_ENABLE
+            `INST_FMADD,
+            `INST_FMSUB,
+            `INST_FNMSUB,
+            `INST_FNMADD: begin 
+                ex_type = `EX_FPU;
+                op_type = `INST_OP_BITS'(opcode[3:0]);
+                op_mod  = func3;
+                use_rd  = 1;
+                `USED_FREG (rd);              
+                `USED_FREG (rs1);
+                `USED_FREG (rs2);
+                `USED_FREG (rs3);
+            end
+            `INST_FCI: begin 
+                ex_type = `EX_FPU;
+                op_mod  = func3;
+                use_rd  = 1;                
+                case (func7)
+                    7'h00, // FADD
+                    7'h04, // FSUB
+                    7'h08, // FMUL
+                    7'h0C: begin // FDIV
+                        op_type = `INST_OP_BITS'(func7[3:0]);
+                        `USED_FREG (rd);
+                        `USED_FREG (rs1);
+                        `USED_FREG (rs2);
+                    end
+                    7'h2C: begin
+                        op_type = `INST_OP_BITS'(`INST_FPU_SQRT);
+                        `USED_FREG (rd);
+                        `USED_FREG (rs1);
+                    end
+                    7'h50: begin
+                        op_type = `INST_OP_BITS'(`INST_FPU_CMP);
+                        `USED_IREG (rd);
+                        `USED_FREG (rs1);
+                        `USED_FREG (rs2);
+                    end
+                    7'h60: begin
+                        op_type = (instr[20]) ? `INST_OP_BITS'(`INST_FPU_CVTWUS) : `INST_OP_BITS'(`INST_FPU_CVTWS);
+                        `USED_IREG (rd);
+                        `USED_FREG (rs1);
+                    end
+                    7'h68: begin
+                        op_type = (instr[20]) ? `INST_OP_BITS'(`INST_FPU_CVTSWU) : `INST_OP_BITS'(`INST_FPU_CVTSW);
+                        `USED_FREG (rd);
+                        `USED_IREG (rs1);
+                    end
+                    7'h10: begin
+                        // FSGNJ=0, FSGNJN=1, FSGNJX=2
+                        op_type = `INST_OP_BITS'(`INST_FPU_MISC);
+                        op_mod  = {1'b0, func3[1:0]};
+                        `USED_FREG (rd);
+                        `USED_FREG (rs1);
+                        `USED_FREG (rs2);
+                    end
+                    7'h14: begin
+                        // FMIN=3, FMAX=4
+                        op_type = `INST_OP_BITS'(`INST_FPU_MISC);
+                        op_mod  = func3[0] ? 4 : 3;
+                        `USED_FREG (rd);
+                        `USED_FREG (rs1);
+                        `USED_FREG (rs2);
+                    end
+                    7'h70: begin 
+                        if (func3[0]) begin
+                            // FCLASS
+                            op_type = `INST_OP_BITS'(`INST_FPU_CLASS);                                     
+                        end else begin
+                            // FMV.X.W=5
+                            op_type = `INST_OP_BITS'(`INST_FPU_MISC);
+                            op_mod  = 5;
+                        end
+                        `USED_IREG (rd);
+                        `USED_FREG (rs1);                                           
+                    end 
+                    7'h78: begin 
+                        // FMV.W.X=6
+                        op_type = `INST_OP_BITS'(`INST_FPU_MISC); 
+                        op_mod  = 6;
+                        `USED_FREG (rd);
+                        `USED_IREG (rs1);
+                    end
+                default:;
+                endcase
+            end
+        `endif
+            `INST_GPGPU: begin 
+                ex_type = `EX_GPU;
+                case (func3)
+                    3'h0: begin
+                        op_type = rs2[0] ? `INST_OP_BITS'(`INST_GPU_PRED) : `INST_OP_BITS'(`INST_GPU_TMC);
+                        is_wstall = 1;
+                        `USED_IREG (rs1);
+                    end
+                    3'h1: begin
+                        op_type = `INST_OP_BITS'(`INST_GPU_WSPAWN);
+                        `USED_IREG (rs1);
+                        `USED_IREG (rs2);
+                    end
+                    3'h2: begin
+                        op_type = `INST_OP_BITS'(`INST_GPU_SPLIT);
+                        is_wstall = 1;
+                        `USED_IREG (rs1);
+                    end
+                    3'h3: begin 
+                        op_type = `INST_OP_BITS'(`INST_GPU_JOIN);
+                        is_join = 1;
+                    end
+                    3'h4: begin 
+                        op_type = `INST_OP_BITS'(`INST_GPU_BAR);
+                        is_wstall = 1;
+                        `USED_IREG (rs1);
+                        `USED_IREG (rs2);
+                    end                
+                    3'h5: begin
+                        ex_type = `EX_LSU;
+                        op_type = `INST_OP_BITS'(`INST_LSU_LW);
+                        op_mod  = `INST_MOD_BITS'(2);
+                        `USED_IREG (rs1);
+                    end
+                    default:;
+                endcase
+            end
+            `INST_GPU: begin                
+                case (func3)
+                `ifdef EXT_TEX_ENABLE
+                    3'h0: begin
+                        ex_type = `EX_GPU;
+                        op_type = `INST_OP_BITS'(`INST_GPU_TEX);
+                        op_mod  = `INST_MOD_BITS'(func2);
+                        use_rd  = 1;       
+                        `USED_IREG (rd);              
+                        `USED_IREG (rs1);
+                        `USED_IREG (rs2);
+                        `USED_IREG (rs3);
+                    end
+                `endif
+                    default:;
+                endcase
+            end
+            default:;
+        endcase
+    end
+
+    `UNUSED_VAR (func2)
+
+    // disable write to integer register r0
+    wire wb = use_rd && (| rd_r);
+
+    assign decode_if.valid     = ifetch_rsp_if.valid;
+    assign decode_if.uuid      = ifetch_rsp_if.uuid;
+    assign decode_if.wid       = ifetch_rsp_if.wid;
+    assign decode_if.tmask     = ifetch_rsp_if.tmask;
+    assign decode_if.PC        = ifetch_rsp_if.PC;
+    assign decode_if.ex_type   = ex_type;
+    assign decode_if.op_type   = op_type;
+    assign decode_if.op_mod    = op_mod;
+    assign decode_if.wb        = wb;
+    assign decode_if.rd        = rd_r;
+    assign decode_if.rs1       = rs1_r;
+    assign decode_if.rs2       = rs2_r;
+    assign decode_if.rs3       = rs3_r;
+    assign decode_if.imm       = imm;
+    assign decode_if.use_PC    = use_PC;
+    assign decode_if.use_imm   = use_imm;
+
+    ///////////////////////////////////////////////////////////////////////////
+
+    wire ifetch_rsp_fire = ifetch_rsp_if.valid && ifetch_rsp_if.ready;
+
+    assign join_if.valid = ifetch_rsp_fire && is_join;
+    assign join_if.wid = ifetch_rsp_if.wid;
+
+    assign wstall_if.valid = ifetch_rsp_fire;
+    assign wstall_if.wid = ifetch_rsp_if.wid;
+    assign wstall_if.stalled = is_wstall;
+
+    assign ifetch_rsp_if.ready = decode_if.ready;
+
+`ifdef PERF_ENABLE
+    wire [$clog2(`NUM_THREADS+1)-1:0] perf_loads_per_cycle;
+    wire [$clog2(`NUM_THREADS+1)-1:0] perf_stores_per_cycle;
+    wire [$clog2(`NUM_THREADS+1)-1:0] perf_branches_per_cycle;
+
+    wire [`NUM_THREADS-1:0] perf_loads_per_mask = decode_if.tmask & {`NUM_THREADS{decode_if.ex_type  == `EX_LSU && `INST_LSU_IS_MEM(decode_if.op_mod) && decode_if.wb}};
+    wire [`NUM_THREADS-1:0] perf_stores_per_mask = decode_if.tmask & {`NUM_THREADS{decode_if.ex_type == `EX_LSU && `INST_LSU_IS_MEM(decode_if.op_mod) && ~decode_if.wb}};
+    wire [`NUM_THREADS-1:0] perf_branches_per_mask = decode_if.tmask & {`NUM_THREADS{decode_if.ex_type == `EX_ALU && `INST_ALU_IS_BR(decode_if.op_mod)}};
+
+    `POP_COUNT(perf_loads_per_cycle, perf_loads_per_mask);
+    `POP_COUNT(perf_stores_per_cycle, perf_stores_per_mask);
+    `POP_COUNT(perf_branches_per_cycle, perf_branches_per_mask);
+
+    reg [`PERF_CTR_BITS-1:0] perf_loads;
+    reg [`PERF_CTR_BITS-1:0] perf_stores;
+    reg [`PERF_CTR_BITS-1:0] perf_branches;
+
+    always @(posedge clk) begin
+        if (reset) begin
+            perf_loads    <= 0;
+            perf_stores   <= 0;
+            perf_branches <= 0;
+        end else begin
+            if (decode_if.valid && decode_if.ready) begin
+                perf_loads    <= perf_loads + `PERF_CTR_BITS'(perf_loads_per_cycle);
+                perf_stores   <= perf_stores + `PERF_CTR_BITS'(perf_stores_per_cycle);
+                perf_branches <= perf_branches + `PERF_CTR_BITS'(perf_branches_per_cycle);
+            end
+        end
+    end
+    
+    assign perf_decode_if.loads    = perf_loads;
+    assign perf_decode_if.stores   = perf_stores;
+    assign perf_decode_if.branches = perf_branches;
+`endif
+
+`ifdef DBG_TRACE_CORE_PIPELINE
+    always @(posedge clk) begin
+        if (decode_if.valid && decode_if.ready) begin
+            dpi_trace("%d: core%0d-decode: wid=%0d, PC=%0h, ex=", $time, CORE_ID, decode_if.wid, decode_if.PC);
+            trace_ex_type(decode_if.ex_type);
+            dpi_trace(", op=");
+            trace_ex_op(decode_if.ex_type, decode_if.op_type, decode_if.op_mod);
+            dpi_trace(", mod=%0d, tmask=%b, wb=%b, rd=%0d, rs1=%0d, rs2=%0d, rs3=%0d, imm=%0h, use_pc=%b, use_imm=%b (#%0d)\n",
+                decode_if.op_mod, decode_if.tmask, decode_if.wb, decode_if.rd, decode_if.rs1, decode_if.rs2, decode_if.rs3, decode_if.imm, decode_if.use_PC, decode_if.use_imm, decode_if.uuid);
+        end
+    end
+`endif
+
+endmodule

hw/rtl/VX_define.vh

@@ -1,40 +1,24 @@
-// Copyright © 2019-2023
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// 
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-`ifndef VX_DEFINE_VH
-`define VX_DEFINE_VH
+`ifndef VX_DEFINE
+`define VX_DEFINE
 
 `include "VX_platform.vh"
 `include "VX_config.vh"
-`include "VX_types.vh"
 
 ///////////////////////////////////////////////////////////////////////////////
 
-`define NW_BITS         `CLOG2(`NUM_WARPS)
-`define NC_WIDTH        `UP(`NC_BITS)
+`define NW_BITS         `LOG2UP(`NUM_WARPS)
 
-`define NT_BITS         `CLOG2(`NUM_THREADS)
-`define NW_WIDTH        `UP(`NW_BITS)
+`define NT_BITS         `LOG2UP(`NUM_THREADS)
 
-`define NC_BITS         `CLOG2(`NUM_CORES)
-`define NT_WIDTH        `UP(`NT_BITS)
+`define NC_BITS         `LOG2UP(`NUM_CORES)
 
-`define NB_BITS         `CLOG2(`NUM_BARRIERS)
-`define NB_WIDTH        `UP(`NB_BITS)
+`define NB_BITS         `LOG2UP(`NUM_BARRIERS)
 
 `define NUM_IREGS       32
 
-`define NRI_BITS        `CLOG2(`NUM_IREGS)
+`define NRI_BITS        `LOG2UP(`NUM_IREGS)
+
+`define NTEX_BITS       `LOG2UP(`NUM_TEX_UNITS)
 
 `ifdef EXT_F_ENABLE
 `define NUM_REGS        (2 * `NUM_IREGS)
@@ -42,34 +26,25 @@
 `define NUM_REGS        `NUM_IREGS
 `endif
 
-`define NR_BITS         `CLOG2(`NUM_REGS)
+`define NR_BITS         `LOG2UP(`NUM_REGS)
+
+`define CSR_ADDR_BITS   12
+
+`define CSR_WIDTH       12
 
 `define PERF_CTR_BITS   44
 
-`ifndef NDEBUG
-`define UUID_WIDTH      44
-`else
-`define UUID_WIDTH      1
-`endif
+`define UUID_BITS       44
 
 ///////////////////////////////////////////////////////////////////////////////
 
-`define EX_ALU          0
-`define EX_LSU          1
-`define EX_SFU          2
-`define EX_FPU          (`EX_SFU + `EXT_F_ENABLED)
-`define EX_TENSOR       (`EX_FPU + `EXT_T_ENABLED)
-
-`define NUM_EX_UNITS    (3 + `EXT_F_ENABLED + `EXT_T_ENABLED)
-`define EX_BITS         `CLOG2(`NUM_EX_UNITS)
-`define EX_WIDTH        `UP(`EX_BITS)
-
-`define SFU_CSRS        0
-`define SFU_WCTL        1
-
-`define NUM_SFU_UNITS   (2)
-`define SFU_BITS        `CLOG2(`NUM_SFU_UNITS)
-`define SFU_WIDTH       `UP(`SFU_BITS)
+`define EX_NOP          3'h0
+`define EX_ALU          3'h1
+`define EX_LSU          3'h2
+`define EX_CSR          3'h3
+`define EX_FPU          3'h4
+`define EX_GPU          3'h5
+`define EX_BITS         3
 
 ///////////////////////////////////////////////////////////////////////////////
 
@@ -85,10 +60,6 @@
 `define INST_FENCE      7'b0001111 // Fence instructions
 `define INST_SYS        7'b1110011 // system instructions
 
-// RV64I instruction specific opcodes (for any W instruction)
-`define INST_I_W        7'b0011011 // W type immediate instructions
-`define INST_R_W        7'b0111011 // W type register instructions
-
 `define INST_FL         7'b0000111 // float load instruction
 `define INST_FS         7'b0100111 // float store  instruction
 `define INST_FMADD      7'b1000011  
@@ -97,11 +68,10 @@
 `define INST_FNMADD     7'b1001111 
 `define INST_FCI        7'b1010011 // float common instructions
 
-// Custom extension opcodes
-`define INST_EXT1       7'b0001011 // 0x0B
-`define INST_EXT2       7'b0101011 // 0x2B
-`define INST_EXT3       7'b1011011 // 0x5B
-`define INST_EXT4       7'b1111011 // 0x7B
+`define INST_GPGPU      7'b1101011
+`define INST_GPU        7'b1011011
+
+`define INST_TEX       7'b0101011
 
 ///////////////////////////////////////////////////////////////////////////////
 
@@ -116,8 +86,7 @@
 ///////////////////////////////////////////////////////////////////////////////
 
 `define INST_OP_BITS    4
-`define INST_MOD_BITS   4
-`define INST_FMT_BITS   2
+`define INST_MOD_BITS   3
 
 ///////////////////////////////////////////////////////////////////////////////
 
@@ -126,22 +95,20 @@
 `define INST_ALU_AUIPC       4'b0011
 `define INST_ALU_SLTU        4'b0100
 `define INST_ALU_SLT         4'b0101
-`define INST_ALU_SUB         4'b0111
 `define INST_ALU_SRL         4'b1000
 `define INST_ALU_SRA         4'b1001
+`define INST_ALU_SUB         4'b1011
 `define INST_ALU_AND         4'b1100
 `define INST_ALU_OR          4'b1101
 `define INST_ALU_XOR         4'b1110
 `define INST_ALU_SLL         4'b1111
 `define INST_ALU_OTHER       4'b0111
 `define INST_ALU_BITS        4
-`define INST_ALU_CLASS(op)   op[3:2]
-`define INST_ALU_SIGNED(op)  op[0]
-`define INST_ALU_IS_SUB(op)  op[1]
-`define INST_ALU_IS_BR(mod)  mod[0]
-`define INST_ALU_IS_M(mod)   mod[1]
-`define INST_ALU_IS_W(mod)   mod[2]
-`define INST_ALU_IS_RED(mod) mod[3]
+`define INST_ALU_OP(x)       x[`INST_ALU_BITS-1:0]
+`define INST_ALU_OP_CLASS(x) x[3:2]
+`define INST_ALU_SIGNED(x)   x[0]
+`define INST_ALU_IS_BR(x)    x[0]
+`define INST_ALU_IS_MUL(x)   x[1]
 
 `define INST_BR_EQ           4'b0000
 `define INST_BR_NE           4'b0010
@@ -158,307 +125,292 @@
 `define INST_BR_MRET         4'b1110
 `define INST_BR_OTHER        4'b1111
 `define INST_BR_BITS         4
-`define INST_BR_CLASS(op)    {1'b0, ~op[3]}
-`define INST_BR_IS_NEG(op)   op[1]
-`define INST_BR_IS_LESS(op)  op[2]
-`define INST_BR_IS_STATIC(op) op[3]
+`define INST_BR_NEG(x)       x[1]
+`define INST_BR_LESS(x)      x[2]
+`define INST_BR_STATIC(x)    x[3]
 
-`define INST_M_MUL           3'b000
-`define INST_M_MULHU         3'b001
-`define INST_M_MULH          3'b010
-`define INST_M_MULHSU        3'b011
-`define INST_M_DIV           3'b100
-`define INST_M_DIVU          3'b101
-`define INST_M_REM           3'b110
-`define INST_M_REMU          3'b111
-`define INST_M_BITS          3
-`define INST_M_SIGNED(op)    (~op[0])
-`define INST_M_IS_MULX(op)   (~op[2])
-`define INST_M_IS_MULH(op)   (op[1:0] != 0)
-`define INST_M_SIGNED_A(op)  (op[1:0] != 1)
-`define INST_M_IS_REM(op)    op[1]
-
-`define INST_RED_ADD         4'b0000
-`define INST_RED_ADDU        4'b1000
-`define INST_RED_MIN         4'b0001
-`define INST_RED_MINU        4'b1001
-`define INST_RED_MAX         4'b0010
-`define INST_RED_MAXU        4'b1010
-`define INST_RED_AND         4'b0011
-`define INST_RED_OR          4'b0100
-`define INST_RED_XOR         4'b0101
-`define INST_RED_BITS        4
+`define INST_MUL_MUL         3'h0
+`define INST_MUL_MULH        3'h1
+`define INST_MUL_MULHSU      3'h2
+`define INST_MUL_MULHU       3'h3
+`define INST_MUL_DIV         3'h4
+`define INST_MUL_DIVU        3'h5
+`define INST_MUL_REM         3'h6
+`define INST_MUL_REMU        3'h7
+`define INST_MUL_BITS        3
+`define INST_MUL_IS_DIV(x)   x[2]
 
 `define INST_FMT_B           3'b000
 `define INST_FMT_H           3'b001
 `define INST_FMT_W           3'b010
-`define INST_FMT_D           3'b011
 `define INST_FMT_BU          3'b100
 `define INST_FMT_HU          3'b101
-`define INST_FMT_WU          3'b110
 
 `define INST_LSU_LB          4'b0000 
 `define INST_LSU_LH          4'b0001
 `define INST_LSU_LW          4'b0010
-`define INST_LSU_LD          4'b0011 // new for RV64I LD
 `define INST_LSU_LBU         4'b0100
 `define INST_LSU_LHU         4'b0101
-`define INST_LSU_LWU         4'b0110 // new for RV64I LWU
 `define INST_LSU_SB          4'b1000 
 `define INST_LSU_SH          4'b1001
 `define INST_LSU_SW          4'b1010
-`define INST_LSU_SD          4'b1011 // new for RV64I SD
-`define INST_LSU_FENCE       4'b1111
 `define INST_LSU_BITS        4
-`define INST_LSU_FMT(op)     op[2:0]
-`define INST_LSU_WSIZE(op)   op[1:0]
-`define INST_LSU_IS_FENCE(op) (op[3:2] == 3)
+`define INST_LSU_FMT(x)      x[2:0]
+`define INST_LSU_WSIZE(x)    x[1:0]
+`define INST_LSU_IS_MEM(x)   (3'h0 == x)
+`define INST_LSU_IS_FENCE(x) (3'h1 == x)
+`define INST_LSU_IS_PREFETCH(x) (3'h2 == x)
 
 `define INST_FENCE_BITS      1
 `define INST_FENCE_D         1'h0
 `define INST_FENCE_I         1'h1
 
-`define INST_FPU_ADD         4'b0000 
-`define INST_FPU_SUB         4'b0001 
-`define INST_FPU_MUL         4'b0010 
-`define INST_FPU_DIV         4'b0011
-`define INST_FPU_SQRT        4'b0100
-`define INST_FPU_CMP         4'b0101 // mod: LE=0, LT=1, EQ=2
-`define INST_FPU_F2F         4'b0110
-`define INST_FPU_MISC        4'b0111 // mod: SGNJ=0, SGNJN=1, SGNJX=2, CLASS=3, MVXW=4, MVWX=5, FMIN=6, FMAX=7
-`define INST_FPU_F2I         4'b1000
-`define INST_FPU_F2U         4'b1001
-`define INST_FPU_I2F         4'b1010
-`define INST_FPU_U2F         4'b1011
-`define INST_FPU_MADD        4'b1100 
-`define INST_FPU_MSUB        4'b1101   
-`define INST_FPU_NMSUB       4'b1110   
-`define INST_FPU_NMADD       4'b1111
+`define INST_CSR_RW          2'h1
+`define INST_CSR_RS          2'h2
+`define INST_CSR_RC          2'h3
+`define INST_CSR_OTHER       2'h0
+`define INST_CSR_BITS        2
+
+`define INST_FPU_ADD         4'h0 
+`define INST_FPU_SUB         4'h4 
+`define INST_FPU_MUL         4'h8 
+`define INST_FPU_DIV         4'hC
+`define INST_FPU_CVTWS       4'h1  // FCVT.W.S
+`define INST_FPU_CVTWUS      4'h5  // FCVT.WU.S
+`define INST_FPU_CVTSW       4'h9  // FCVT.S.W
+`define INST_FPU_CVTSWU      4'hD  // FCVT.S.WU
+`define INST_FPU_SQRT        4'h2
+`define INST_FPU_CLASS       4'h6  
+`define INST_FPU_CMP         4'hA
+`define INST_FPU_MISC        4'hE  // SGNJ, SGNJN, SGNJX, FMIN, FMAX, MVXW, MVWX 
+`define INST_FPU_MADD        4'h3 
+`define INST_FPU_MSUB        4'h7   
+`define INST_FPU_NMSUB       4'hB   
+`define INST_FPU_NMADD       4'hF
 `define INST_FPU_BITS        4
-`define INST_FPU_IS_W(mod)   (mod[4])
-`define INST_FPU_IS_CLASS(op, mod) (op == `INST_FPU_MISC && mod == 3)
-`define INST_FPU_IS_MVXW(op, mod) (op == `INST_FPU_MISC && mod == 4)
 
-`define INST_SFU_TMC         4'h0
-`define INST_SFU_WSPAWN      4'h1 
-`define INST_SFU_SPLIT       4'h2
-`define INST_SFU_JOIN        4'h3
-`define INST_SFU_BAR         4'h4
-`define INST_SFU_PRED        4'h5
-`define INST_SFU_CSRRW       4'h6
-`define INST_SFU_CSRRS       4'h7
-`define INST_SFU_CSRRC       4'h8
-`define INST_SFU_CMOV        4'h9
-`define INST_SFU_BAR_MASK    4'ha
-`define INST_SFU_BITS        4
-`define INST_SFU_CSR(f3)     (4'h6 + 4'(f3) - 4'h1)
-`define INST_SFU_IS_WCTL(op) ((op <= 5) || (op == `INST_SFU_BAR_MASK))
-`define INST_SFU_IS_CSR(op)  (op >= 6 && op <= 8)
+`define INST_GPU_TMC         4'h0
+`define INST_GPU_WSPAWN      4'h1 
+`define INST_GPU_SPLIT       4'h2
+`define INST_GPU_JOIN        4'h3
+`define INST_GPU_BAR         4'h4
+`define INST_GPU_PRED        4'h5
+`define INST_GPU_TEX         4'h6
+`define INST_GPU_BITS        4
 
-`define INST_TENSOR_HMMA       4'b0000
-// Hopper WGMMA-style asynchronous op
-`define INST_TENSOR_HGMMA      4'b0001
-`define INST_TENSOR_HGMMA_WAIT 4'b0010
-`define INST_TENSOR_TCGEN05_CP      4'b0011
-`define INST_TENSOR_TCGEN05_CP_WAIT 4'b0100
-`define INST_TENSOR_BWGMMA          4'b0101
-`define INST_TENSOR_BWGMMA_WAIT     4'b0110
-`define INST_TENSOR_TCGEN05_LD      4'b0111
-`define INST_TENSOR_TCGEN05_ST      4'b1000
-`define INST_TENSOR_TCGEN05_CB      4'b1001
+///////////////////////////////////////////////////////////////////////////////
 
-`ifdef EXT_T_HOPPER
-`define EXT_T_ASYNC
-`elsif EXT_T_BLACKWELL
-`define EXT_T_ASYNC
+`ifdef EXT_M_ENABLE
+    `define ISA_EXT_M   (1 << 12)
+`else
+    `define ISA_EXT_M   0
 `endif
 
+`ifdef EXT_F_ENABLE
+    `define ISA_EXT_F   (1 << 5)
+`else
+    `define ISA_EXT_F   0
+`endif
+
+`define ISA_CODE  (0 <<  0) // A - Atomic Instructions extension \
+                | (0 <<  1) // B - Tentatively reserved for Bit operations extension  \
+                | (0 <<  2) // C - Compressed extension \
+                | (0 <<  3) // D - Double precsision floating-point extension \
+                | (0 <<  4) // E - RV32E base ISA \
+                |`ISA_EXT_F // F - Single precsision floating-point extension \
+                | (0 <<  6) // G - Additional standard extensions present \
+                | (0 <<  7) // H - Hypervisor mode implemented \
+                | (1 <<  8) // I - RV32I/64I/128I base ISA \
+                | (0 <<  9) // J - Reserved \
+                | (0 << 10) // K - Reserved \
+                | (0 << 11) // L - Tentatively reserved for Bit operations extension \
+                |`ISA_EXT_M // M - Integer Multiply/Divide extension \
+                | (0 << 13) // N - User level interrupts supported \
+                | (0 << 14) // O - Reserved \
+                | (0 << 15) // P - Tentatively reserved for Packed-SIMD extension \
+                | (0 << 16) // Q - Quad-precision floating-point extension \
+                | (0 << 17) // R - Reserved \
+                | (0 << 18) // S - Supervisor mode implemented \
+                | (0 << 19) // T - Tentatively reserved for Transactional Memory extension \
+                | (1 << 20) // U - User mode implemented \
+                | (0 << 21) // V - Tentatively reserved for Vector extension \
+                | (0 << 22) // W - Reserved \
+                | (1 << 23) // X - Non-standard extensions present \
+                | (0 << 24) // Y - Reserved \
+                | (0 << 25) // Z - Reserved
+
 ///////////////////////////////////////////////////////////////////////////////
 
 // non-cacheable tag bits
-`define NC_TAG_BITS             1
+`define NC_TAG_BIT              1
+
+// texture tag bits
+`define TEX_TAG_BIT             1
 
 // cache address type bits
-`ifdef SM_ENABLE
-`define CACHE_ADDR_TYPE_BITS    (`NC_TAG_BITS + 1)
-`else
-`define CACHE_ADDR_TYPE_BITS    `NC_TAG_BITS
+`define CACHE_ADDR_TYPE_BITS    (`NC_TAG_BIT + `SM_ENABLE)
+
+////////////////////////// Icache Configurable Knobs //////////////////////////
+
+// Cache ID
+`define ICACHE_ID               (32'(`L3_ENABLE) + 32'(`L2_ENABLE) * `NUM_CLUSTERS + CORE_ID * 3 + 0)
+
+// Word size in bytes
+`define ICACHE_WORD_SIZE        4
+
+// Block size in bytes
+`define ICACHE_LINE_SIZE        `L1_BLOCK_SIZE
+
+// TAG sharing enable       
+`define ICACHE_CORE_TAG_ID_BITS `NW_BITS
+
+// Core request tag bits
+`define ICACHE_CORE_TAG_WIDTH   (`UUID_BITS + `ICACHE_CORE_TAG_ID_BITS)
+
+// Memory request data bits
+`define ICACHE_MEM_DATA_WIDTH   (`ICACHE_LINE_SIZE * 8)
+
+// Memory request address bits
+`define ICACHE_MEM_ADDR_WIDTH   (32 - `CLOG2(`ICACHE_LINE_SIZE))
+
+// Memory request tag bits
+`define ICACHE_MEM_TAG_WIDTH    `CLOG2(`ICACHE_MSHR_SIZE)
+
+////////////////////////// Dcache Configurable Knobs //////////////////////////
+
+// Cache ID
+`define DCACHE_ID               (32'(`L3_ENABLE) + 32'(`L2_ENABLE) * `NUM_CLUSTERS + CORE_ID * 3 + 1)
+
+// Word size in bytes
+`define DCACHE_WORD_SIZE        4
+
+// Block size in bytes
+`define DCACHE_LINE_SIZE        `L1_BLOCK_SIZE
+
+// Core request tag bits
+`define LSUQ_ADDR_BITS          `LOG2UP(`LSUQ_SIZE)
+`ifdef EXT_TEX_ENABLE
+`define LSU_TAG_ID_BITS         `MAX(`LSUQ_ADDR_BITS, 2)
+`define LSU_TEX_DCACHE_TAG_BITS (`UUID_BITS + `LSU_TAG_ID_BITS + `CACHE_ADDR_TYPE_BITS)
+`define DCACHE_CORE_TAG_ID_BITS (`LSU_TAG_ID_BITS + `CACHE_ADDR_TYPE_BITS + `TEX_TAG_BIT)
+`else 
+`define LSU_TAG_ID_BITS         `LSUQ_ADDR_BITS
+`define DCACHE_CORE_TAG_ID_BITS (`LSU_TAG_ID_BITS + `CACHE_ADDR_TYPE_BITS)
 `endif
+`define DCACHE_CORE_TAG_WIDTH   (`UUID_BITS + `DCACHE_CORE_TAG_ID_BITS)
+ 
+// Memory request data bits
+`define DCACHE_MEM_DATA_WIDTH   (`DCACHE_LINE_SIZE * 8)
 
-`define ARB_SEL_BITS(I, O)      ((I > O) ? `CLOG2((I + O - 1) / O) : 0)
+// Memory request address bits
+`define DCACHE_MEM_ADDR_WIDTH   (32 - `CLOG2(`DCACHE_LINE_SIZE))
+
+// Memory byte enable bits
+`define DCACHE_MEM_BYTEEN_WIDTH `DCACHE_LINE_SIZE
+
+// Input request size
+`define DCACHE_NUM_REQS         `NUM_THREADS
+
+// Memory request tag bits
+`define _DMEM_ADDR_RATIO_W      $clog2(`DCACHE_LINE_SIZE / `DCACHE_WORD_SIZE)
+`define _DNC_MEM_TAG_WIDTH      ($clog2(`DCACHE_NUM_REQS) + `_DMEM_ADDR_RATIO_W + `DCACHE_CORE_TAG_WIDTH)
+`define DCACHE_MEM_TAG_WIDTH    `MAX((`CLOG2(`DCACHE_NUM_BANKS) + `CLOG2(`DCACHE_MSHR_SIZE) + `NC_TAG_BIT), `_DNC_MEM_TAG_WIDTH)
+
+// Merged D-cache/I-cache memory tag
+`define L1_MEM_TAG_WIDTH        (`MAX(`ICACHE_MEM_TAG_WIDTH, `DCACHE_MEM_TAG_WIDTH) + `CLOG2(2))
+
+////////////////////////// SM Configurable Knobs //////////////////////////////
+
+// Cache ID
+`define SMEM_ID                 (32'(`L3_ENABLE) + 32'(`L2_ENABLE) * `NUM_CLUSTERS + CORE_ID * 3 + 2)
+
+// Word size in bytes
+`define SMEM_WORD_SIZE          4
+
+// bank address offset
+`define SMEM_BANK_ADDR_OFFSET   `CLOG2(`STACK_SIZE / `SMEM_WORD_SIZE)
+
+// Input request size
+`define SMEM_NUM_REQS           `NUM_THREADS
+
+////////////////////////// L2cache Configurable Knobs /////////////////////////
+
+// Cache ID
+`define L2_CACHE_ID              (32'(`L3_ENABLE) + CLUSTER_ID)
+
+// Word size in bytes
+`define L2_WORD_SIZE             `DCACHE_LINE_SIZE
+
+// Block size in bytes
+`define L2_CACHE_LINE_SIZE       ((`L2_ENABLE) ? `MEM_BLOCK_SIZE : `L2_WORD_SIZE)
+
+// Input request tag bits
+`define L2_CORE_TAG_WIDTH        (`DCACHE_CORE_TAG_WIDTH + `CLOG2(`NUM_CORES))
+
+// Memory request data bits
+`define L2_MEM_DATA_WIDTH        (`L2_CACHE_LINE_SIZE * 8)
+
+// Memory request address bits
+`define L2_MEM_ADDR_WIDTH        (32 - `CLOG2(`L2_CACHE_LINE_SIZE))
+
+// Memory byte enable bits
+`define L2_MEM_BYTEEN_WIDTH      `L2_CACHE_LINE_SIZE
+
+// Input request size
+`define L2_NUM_REQS              `NUM_CORES
+
+// Memory request tag bits
+`define _L2_MEM_ADDR_RATIO_W     $clog2(`L2_CACHE_LINE_SIZE / `L2_WORD_SIZE)
+`define _L2_NC_MEM_TAG_WIDTH     ($clog2(`L2_NUM_REQS) + `_L2_MEM_ADDR_RATIO_W + `L1_MEM_TAG_WIDTH)
+`define _L2_MEM_TAG_WIDTH        `MAX((`CLOG2(`L2_NUM_BANKS) + `CLOG2(`L2_MSHR_SIZE) + `NC_TAG_BIT), `_L2_NC_MEM_TAG_WIDTH)
+`define L2_MEM_TAG_WIDTH         ((`L2_ENABLE) ? `_L2_MEM_TAG_WIDTH : (`L1_MEM_TAG_WIDTH + `CLOG2(`L2_NUM_REQS)))
+
+////////////////////////// L3cache Configurable Knobs /////////////////////////
+
+// Cache ID
+`define L3_CACHE_ID              0
+
+// Word size in bytes
+`define L3_WORD_SIZE             `L2_CACHE_LINE_SIZE
+
+// Block size in bytes
+`define L3_CACHE_LINE_SIZE       ((`L3_ENABLE) ? `MEM_BLOCK_SIZE : `L3_WORD_SIZE)
+
+// Input request tag bits
+`define L3_CORE_TAG_WIDTH        (`L2_CORE_TAG_WIDTH + `CLOG2(`NUM_CLUSTERS))
+
+// Memory request data bits
+`define L3_MEM_DATA_WIDTH        (`L3_CACHE_LINE_SIZE * 8)
+
+// Memory request address bits
+`define L3_MEM_ADDR_WIDTH        (32 - `CLOG2(`L3_CACHE_LINE_SIZE))
+
+// Memory byte enable bits
+`define L3_MEM_BYTEEN_WIDTH      `L3_CACHE_LINE_SIZE
+
+// Input request size
+`define L3_NUM_REQS              `NUM_CLUSTERS
+
+// Memory request tag bits
+`define _L3_MEM_ADDR_RATIO_W     $clog2(`L3_CACHE_LINE_SIZE / `L3_WORD_SIZE)
+`define _L3_NC_MEM_TAG_WIDTH     ($clog2(`L3_NUM_REQS) + `_L3_MEM_ADDR_RATIO_W + `L2_MEM_TAG_WIDTH)
+`define _L3_MEM_TAG_WIDTH        `MAX((`CLOG2(`L3_NUM_BANKS) + `CLOG2(`L3_MSHR_SIZE) + `NC_TAG_BIT), `_L3_NC_MEM_TAG_WIDTH)
+`define L3_MEM_TAG_WIDTH         ((`L3_ENABLE) ? `_L3_MEM_TAG_WIDTH : (`L2_MEM_TAG_WIDTH + `CLOG2(`L3_NUM_REQS)))
 
 ///////////////////////////////////////////////////////////////////////////////
 
-`define CACHE_MEM_TAG_WIDTH(mshr_size, num_banks) \
-        (`CLOG2(mshr_size) + `CLOG2(num_banks) + `NC_TAG_BITS)
-        
-`define CACHE_NC_BYPASS_TAG_WIDTH(num_reqs, line_size, word_size, tag_width) \
-        (`CLOG2(num_reqs) + `CLOG2(line_size / word_size) + tag_width)
+`define VX_MEM_BYTEEN_WIDTH     `L3_MEM_BYTEEN_WIDTH   
+`define VX_MEM_ADDR_WIDTH       `L3_MEM_ADDR_WIDTH
+`define VX_MEM_DATA_WIDTH       `L3_MEM_DATA_WIDTH
+`define VX_MEM_TAG_WIDTH        `L3_MEM_TAG_WIDTH
+`define VX_CORE_TAG_WIDTH       `L3_CORE_TAG_WIDTH 
+`define VX_CSR_ID_WIDTH         `LOG2UP(`NUM_CLUSTERS * `NUM_CORES)
 
-`define CACHE_BYPASS_TAG_WIDTH(num_reqs, line_size, word_size, tag_width) \
-        (`CACHE_NC_BYPASS_TAG_WIDTH(num_reqs, line_size, word_size, tag_width) + `NC_TAG_BITS)
-
-`define CACHE_NC_MEM_TAG_WIDTH(mshr_size, num_banks, num_reqs, line_size, word_size, tag_width) \
-        `MAX(`CACHE_MEM_TAG_WIDTH(mshr_size, num_banks), `CACHE_NC_BYPASS_TAG_WIDTH(num_reqs, line_size, word_size, tag_width))
+`define TO_FULL_ADDR(x)         {x, (32-$bits(x))'(0)}
 
 ///////////////////////////////////////////////////////////////////////////////
 
-`define CACHE_CLUSTER_CORE_ARB_TAG(tag_width, num_inputs, num_caches) \
-        (tag_width + `ARB_SEL_BITS(num_inputs, `UP(num_caches)))  
+`include "VX_fpu_types.vh"
+`include "VX_gpu_types.vh"
 
-`define CACHE_CLUSTER_MEM_ARB_TAG(tag_width, num_caches) \
-        (tag_width + `ARB_SEL_BITS(`UP(num_caches), 1))
-
-`define CACHE_CLUSTER_MEM_TAG_WIDTH(mshr_size, num_banks, num_caches) \
-        `CACHE_CLUSTER_MEM_ARB_TAG(`CACHE_MEM_TAG_WIDTH(mshr_size, num_banks),  num_caches)
-
-`define CACHE_CLUSTER_NC_BYPASS_TAG_WIDTH(num_reqs, line_size, word_size, tag_width, num_inputs, num_caches) \
-        `CACHE_CLUSTER_MEM_ARB_TAG((`CLOG2(num_reqs) + `CLOG2(line_size / word_size) + `CACHE_CLUSTER_CORE_ARB_TAG(tag_width, num_inputs, num_caches)), num_caches)
-
-`define CACHE_CLUSTER_BYPASS_TAG_WIDTH(num_reqs, line_size, word_size, tag_width, num_inputs, num_caches) \
-        `CACHE_CLUSTER_MEM_ARB_TAG((`CACHE_NC_BYPASS_TAG_WIDTH(num_reqs, line_size, word_size, `CACHE_CLUSTER_CORE_ARB_TAG(tag_width, num_inputs, num_caches)) + `NC_TAG_BITS), num_caches)
-
-`define CACHE_CLUSTER_NC_MEM_TAG_WIDTH(mshr_size, num_banks, num_reqs, line_size, word_size, tag_width, num_inputs, num_caches) \
-        `CACHE_CLUSTER_MEM_ARB_TAG(`MAX(`CACHE_MEM_TAG_WIDTH(mshr_size, num_banks), `CACHE_NC_BYPASS_TAG_WIDTH(num_reqs, line_size, word_size, `CACHE_CLUSTER_CORE_ARB_TAG(tag_width, num_inputs, num_caches))), num_caches)
-
-///////////////////////////////////////////////////////////////////////////////
-
-`ifdef ICACHE_ENABLE
-`define L1_ENABLE
 `endif
-`ifdef DCACHE_ENABLE
-`define L1_ENABLE
-`endif
-
-`define VX_MEM_BYTEEN_WIDTH     `L3_LINE_SIZE   
-`define VX_MEM_ADDR_WIDTH       (`MEM_ADDR_WIDTH - `CLOG2(`L3_LINE_SIZE))
-`define VX_MEM_DATA_WIDTH       (`L3_LINE_SIZE * 8)
-`define VX_MEM_TAG_WIDTH        L3_MEM_TAG_WIDTH
-
-`define VX_DCR_ADDR_WIDTH       `VX_DCR_ADDR_BITS
-`define VX_DCR_DATA_WIDTH       32
-
-`define TO_FULL_ADDR(x)         {x, (`MEM_ADDR_WIDTH-$bits(x))'(0)}
-
-///////////////////////////////////////////////////////////////////////////////
-
-`define BUFFER_EX(dst, src, ena, latency) \
-    VX_pipe_register #( \
-        .DATAW  ($bits(dst)), \
-        .RESETW ($bits(dst)), \
-        .DEPTH  (latency) \
-    ) __``dst ( \
-        .clk      (clk), \
-        .reset    (reset), \
-        .enable   (ena), \
-        .data_in  (src), \
-        .data_out (dst) \
-    )
-
-`define BUFFER(dst, src) `BUFFER_EX(dst, src, 1'b1, 1)
-
-`define POP_COUNT_EX(out, in, model) \
-    VX_popcount #( \
-        .N ($bits(in)), \
-        .MODEL (model) \
-    ) __``out ( \
-        .data_in  (in), \
-        .data_out (out) \
-    )
-
-`define POP_COUNT(out, in) `POP_COUNT_EX(out, in, 1)
-
-`define ASSIGN_VX_MEM_BUS_IF(dst, src) \
-    assign dst.req_valid  = src.req_valid; \
-    assign dst.req_data   = src.req_data; \
-    assign src.req_ready  = dst.req_ready; \
-    assign src.rsp_valid  = dst.rsp_valid; \
-    assign src.rsp_data   = dst.rsp_data; \
-    assign dst.rsp_ready  = src.rsp_ready
-
-`define ASSIGN_VX_MEM_BUS_IF_X(dst, src, TD, TS) \
-    assign dst.req_valid = src.req_valid; \
-    assign dst.req_data.rw = src.req_data.rw; \
-    assign dst.req_data.byteen = src.req_data.byteen; \
-    assign dst.req_data.addr = src.req_data.addr; \
-    assign dst.req_data.data = src.req_data.data; \
-    if (TD != TS) \
-        assign dst.req_data.tag = {src.req_data.tag, {(TD-TS){1'b0}}}; \
-    else \
-        assign dst.req_data.tag = src.req_data.tag; \
-    assign src.req_ready = dst.req_ready; \
-    assign src.rsp_valid = dst.rsp_valid; \
-    assign src.rsp_data.data = dst.rsp_data.data; \
-    assign src.rsp_data.tag = dst.rsp_data.tag[TD-1 -: TS]; \
-    assign dst.rsp_ready = src.rsp_ready
-
-`define BUFFER_DCR_BUS_IF(dst, src, enable) \
-    logic [(1 + `VX_DCR_ADDR_WIDTH + `VX_DCR_DATA_WIDTH)-1:0] __``dst; \
-    if (enable) begin \
-        always @(posedge clk) begin \
-            __``dst <= {src.write_valid, src.write_addr, src.write_data}; \
-        end \
-    end else begin \
-        assign __``dst = {src.write_valid, src.write_addr, src.write_data}; \
-    end \
-    VX_dcr_bus_if dst(); \
-    assign {dst.write_valid, dst.write_addr, dst.write_data} = __``dst
-
-`define PERF_COUNTER_ADD(dst, src, field, width, dst_count, src_count, reg_enable) \
-    for (genvar __d = 0; __d < dst_count; ++__d) begin \
-        localparam __count = ((src_count > dst_count) ? ((src_count + dst_count - 1) / dst_count) : 1); \
-        wire [__count-1:0][width-1:0] __reduce_add_i_``src``field; \
-        wire [width-1:0] __reduce_add_o_``dst``field; \
-        for (genvar __i = 0; __i < __count; ++__i) begin \
-            assign __reduce_add_i_``src``field[__i] = ``src[__d * __count + __i].``field; \
-        end \
-        VX_reduce #(.DATAW_IN(width), .N(__count), .OP("+")) __reduce_add_``dst``field ( \
-            __reduce_add_i_``src``field, \
-            __reduce_add_o_``dst``field \
-        ); \
-        if (reg_enable) begin \
-            reg [width-1:0] __reduce_add_r_``dst``field; \
-            always @(posedge clk) begin \
-                if (reset) begin \
-                    __reduce_add_r_``dst``field <= '0; \
-                end else begin \
-                    __reduce_add_r_``dst``field <= __reduce_add_o_``dst``field; \
-                end \
-            end \
-            assign ``dst[__d].``field = __reduce_add_r_``dst``field; \
-        end else begin \
-            assign ``dst[__d].``field = __reduce_add_o_``dst``field; \
-        end \
-    end
-
-`define ASSIGN_BLOCKED_WID(dst, src, block_idx, block_size) \
-    if (block_size != 1) begin \
-        if (block_size != `NUM_WARPS) begin \
-            assign dst = {src[`NW_WIDTH-1:`CLOG2(block_size)], `CLOG2(block_size)'(block_idx)}; \
-        end else begin \
-            assign dst = `NW_WIDTH'(block_idx); \
-        end \
-    end else begin \
-        assign dst = src; \
-    end
-
-`define TO_DISPATCH_DATA(data, tid) { \
-    data.uuid, \
-    data.wis, \
-    data.tmask, \
-    data.op_type, \
-    data.op_mod, \
-    data.wb, \
-    data.use_PC, \
-    data.use_imm, \
-    data.PC, \
-    data.imm, \
-    data.rd, \
-    tid, \
-    data.rs1_data, \
-    data.rs2_data, \
-    data.rs3_data}
-
-///////////////////////////////////////////////////////////////////////////////
-
-`endif // VX_DEFINE_VH

hw/rtl/VX_dispatch.sv

@@ -0,0 +1,159 @@
+`include "VX_define.vh"
+
+module VX_dispatch (
+    input wire              clk,
+    input wire              reset,
+
+    // inputs
+    VX_ibuffer_if.slave     ibuffer_if,
+    VX_gpr_rsp_if.slave     gpr_rsp_if,
+
+    // outputs
+    VX_alu_req_if.master    alu_req_if,
+    VX_lsu_req_if.master    lsu_req_if,
+    VX_csr_req_if.master    csr_req_if,
+`ifdef EXT_F_ENABLE
+    VX_fpu_req_if.master    fpu_req_if,
+`endif
+    VX_gpu_req_if.master    gpu_req_if    
+);
+    wire [`NT_BITS-1:0] tid;
+    wire alu_req_ready;
+    wire lsu_req_ready;
+    wire csr_req_ready;
+`ifdef EXT_F_ENABLE
+    wire fpu_req_ready;
+`endif
+    wire gpu_req_ready;
+
+    VX_lzc #(
+        .N (`NUM_THREADS)
+    ) tid_select (
+        .in_i       (ibuffer_if.tmask),
+        .cnt_o      (tid),
+        `UNUSED_PIN (valid_o)
+    );
+
+    wire [31:0] next_PC = ibuffer_if.PC + 4;
+
+    // ALU unit
+
+    wire alu_req_valid = ibuffer_if.valid && (ibuffer_if.ex_type == `EX_ALU);
+    wire [`INST_ALU_BITS-1:0] alu_op_type = `INST_ALU_BITS'(ibuffer_if.op_type);
+    
+    VX_skid_buffer #(
+        .DATAW   (`UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + 32 + `INST_ALU_BITS + `INST_MOD_BITS + 32 + 1 + 1 + `NR_BITS + 1 + `NT_BITS + (2 * `NUM_THREADS * 32)),
+        .OUT_REG (1)
+    ) alu_buffer (
+        .clk       (clk),
+        .reset     (reset),
+        .valid_in  (alu_req_valid),
+        .ready_in  (alu_req_ready),
+        .data_in   ({ibuffer_if.uuid, ibuffer_if.wid, ibuffer_if.tmask, ibuffer_if.PC, next_PC,            alu_op_type,        ibuffer_if.op_mod, ibuffer_if.imm, ibuffer_if.use_PC, ibuffer_if.use_imm, ibuffer_if.rd, ibuffer_if.wb, tid,            gpr_rsp_if.rs1_data, gpr_rsp_if.rs2_data}),
+        .data_out  ({alu_req_if.uuid, alu_req_if.wid, alu_req_if.tmask, alu_req_if.PC, alu_req_if.next_PC, alu_req_if.op_type, alu_req_if.op_mod, alu_req_if.imm, alu_req_if.use_PC, alu_req_if.use_imm, alu_req_if.rd, alu_req_if.wb, alu_req_if.tid, alu_req_if.rs1_data, alu_req_if.rs2_data}),
+        .valid_out (alu_req_if.valid),
+        .ready_out (alu_req_if.ready)
+    );
+
+    // lsu unit
+
+    wire lsu_req_valid = ibuffer_if.valid && (ibuffer_if.ex_type == `EX_LSU);
+    wire [`INST_LSU_BITS-1:0] lsu_op_type = `INST_LSU_BITS'(ibuffer_if.op_type);
+    wire lsu_is_fence = `INST_LSU_IS_FENCE(ibuffer_if.op_mod);
+    wire lsu_is_prefetch = `INST_LSU_IS_PREFETCH(ibuffer_if.op_mod);
+
+    VX_skid_buffer #(
+        .DATAW   (`UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `INST_LSU_BITS + 1 + 32 + `NR_BITS + 1 + (2 * `NUM_THREADS * 32) + 1),
+        .OUT_REG (1)
+    ) lsu_buffer (
+        .clk       (clk),
+        .reset     (reset),
+        .valid_in  (lsu_req_valid),
+        .ready_in  (lsu_req_ready),
+        .data_in   ({ibuffer_if.uuid, ibuffer_if.wid, ibuffer_if.tmask, ibuffer_if.PC, lsu_op_type,        lsu_is_fence,        ibuffer_if.imm,    ibuffer_if.rd, ibuffer_if.wb, gpr_rsp_if.rs1_data,  gpr_rsp_if.rs2_data, lsu_is_prefetch}),
+        .data_out  ({lsu_req_if.uuid, lsu_req_if.wid, lsu_req_if.tmask, lsu_req_if.PC, lsu_req_if.op_type, lsu_req_if.is_fence, lsu_req_if.offset, lsu_req_if.rd, lsu_req_if.wb, lsu_req_if.base_addr, lsu_req_if.store_data, lsu_req_if.is_prefetch}),
+        .valid_out (lsu_req_if.valid),
+        .ready_out (lsu_req_if.ready)
+    );
+
+    // csr unit
+
+    wire csr_req_valid = ibuffer_if.valid && (ibuffer_if.ex_type == `EX_CSR);
+    wire [`INST_CSR_BITS-1:0] csr_op_type = `INST_CSR_BITS'(ibuffer_if.op_type);
+    wire [`CSR_ADDR_BITS-1:0] csr_addr = ibuffer_if.imm[`CSR_ADDR_BITS-1:0];
+    wire [`NRI_BITS-1:0] csr_imm = ibuffer_if.imm[`CSR_ADDR_BITS +: `NRI_BITS];
+    wire [31:0] csr_rs1_data = gpr_rsp_if.rs1_data[tid];
+
+    VX_skid_buffer #(
+        .DATAW   (`UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `INST_CSR_BITS + `CSR_ADDR_BITS + `NR_BITS + 1 + 1 + `NRI_BITS + 32),
+        .OUT_REG (1)
+    ) csr_buffer (
+        .clk       (clk),
+        .reset     (reset),
+        .valid_in  (csr_req_valid),
+        .ready_in  (csr_req_ready),
+        .data_in   ({ibuffer_if.uuid, ibuffer_if.wid, ibuffer_if.tmask, ibuffer_if.PC, csr_op_type,        csr_addr,        ibuffer_if.rd, ibuffer_if.wb, ibuffer_if.use_imm, csr_imm,        csr_rs1_data}),
+        .data_out  ({csr_req_if.uuid, csr_req_if.wid, csr_req_if.tmask, csr_req_if.PC, csr_req_if.op_type, csr_req_if.addr, csr_req_if.rd, csr_req_if.wb, csr_req_if.use_imm, csr_req_if.imm, csr_req_if.rs1_data}),
+        .valid_out (csr_req_if.valid),
+        .ready_out (csr_req_if.ready)
+    );
+
+    // fpu unit
+
+`ifdef EXT_F_ENABLE
+    wire fpu_req_valid = ibuffer_if.valid && (ibuffer_if.ex_type == `EX_FPU);
+    wire [`INST_FPU_BITS-1:0] fpu_op_type = `INST_FPU_BITS'(ibuffer_if.op_type);
+        
+    VX_skid_buffer #(
+        .DATAW   (`UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `INST_FPU_BITS + `INST_MOD_BITS + `NR_BITS + 1 + (3 * `NUM_THREADS * 32)),
+        .OUT_REG (1)
+    ) fpu_buffer (
+        .clk       (clk),
+        .reset     (reset),
+        .valid_in  (fpu_req_valid),
+        .ready_in  (fpu_req_ready),
+        .data_in   ({ibuffer_if.uuid, ibuffer_if.wid, ibuffer_if.tmask, ibuffer_if.PC, fpu_op_type,        ibuffer_if.op_mod, ibuffer_if.rd, ibuffer_if.wb, gpr_rsp_if.rs1_data, gpr_rsp_if.rs2_data, gpr_rsp_if.rs3_data}),
+        .data_out  ({fpu_req_if.uuid, fpu_req_if.wid, fpu_req_if.tmask, fpu_req_if.PC, fpu_req_if.op_type, fpu_req_if.op_mod, fpu_req_if.rd, fpu_req_if.wb, fpu_req_if.rs1_data, fpu_req_if.rs2_data, fpu_req_if.rs3_data}),
+        .valid_out (fpu_req_if.valid),
+        .ready_out (fpu_req_if.ready)
+    );
+`else
+    `UNUSED_VAR (gpr_rsp_if.rs3_data)
+`endif
+
+    // gpu unit
+
+    wire gpu_req_valid = ibuffer_if.valid && (ibuffer_if.ex_type == `EX_GPU);
+    wire [`INST_GPU_BITS-1:0] gpu_op_type = `INST_GPU_BITS'(ibuffer_if.op_type);
+
+    VX_skid_buffer #(
+        .DATAW   (`UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + 32 + `INST_GPU_BITS + `INST_MOD_BITS + `NR_BITS + 1 + `NT_BITS  + (3 * `NUM_THREADS * 32)),
+        .OUT_REG (1)
+    ) gpu_buffer (
+        .clk       (clk),
+        .reset     (reset),
+        .valid_in  (gpu_req_valid),
+        .ready_in  (gpu_req_ready),
+        .data_in   ({ibuffer_if.uuid, ibuffer_if.wid, ibuffer_if.tmask, ibuffer_if.PC, next_PC,            gpu_op_type,        ibuffer_if.op_mod, ibuffer_if.rd, ibuffer_if.wb, tid,            gpr_rsp_if.rs1_data, gpr_rsp_if.rs2_data, gpr_rsp_if.rs3_data}),
+        .data_out  ({gpu_req_if.uuid, gpu_req_if.wid, gpu_req_if.tmask, gpu_req_if.PC, gpu_req_if.next_PC, gpu_req_if.op_type, gpu_req_if.op_mod, gpu_req_if.rd, gpu_req_if.wb, gpu_req_if.tid, gpu_req_if.rs1_data, gpu_req_if.rs2_data, gpu_req_if.rs3_data}),
+        .valid_out (gpu_req_if.valid),
+        .ready_out (gpu_req_if.ready)
+    ); 
+
+    // can take next request?
+    reg ready_r;
+    always @(*) begin
+        case (ibuffer_if.ex_type) 
+        `EX_ALU: ready_r = alu_req_ready;
+        `EX_LSU: ready_r = lsu_req_ready;
+        `EX_CSR: ready_r = csr_req_ready;
+    `ifdef EXT_F_ENABLE
+        `EX_FPU: ready_r = fpu_req_ready;
+    `endif
+        `EX_GPU: ready_r = gpu_req_ready;
+        default: ready_r = 1'b1; // ignore NOPs
+        endcase
+    end
+    assign ibuffer_if.ready = ready_r;
+    
+endmodule

hw/rtl/VX_execute.sv

@@ -0,0 +1,238 @@
+`include "VX_define.vh"
+
+module VX_execute #(
+    parameter CORE_ID = 0
+) (
+    `SCOPE_IO_VX_execute
+
+    input wire clk, 
+    input wire reset,    
+
+    // Dcache interface
+    // NOTE(hansung): this comes out of VX_lsu_unit
+    VX_dcache_req_if.master dcache_req_if,
+    VX_dcache_rsp_if.slave dcache_rsp_if,
+
+    // commit interface
+    VX_cmt_to_csr_if.slave  cmt_to_csr_if,
+
+    // fetch interface
+    VX_fetch_to_csr_if.slave fetch_to_csr_if,
+
+`ifdef PERF_ENABLE
+    VX_perf_memsys_if.slave perf_memsys_if,
+    VX_perf_pipeline_if.slave perf_pipeline_if,
+ `endif
+    
+    // inputs    
+    VX_alu_req_if.slave     alu_req_if,
+    VX_lsu_req_if.slave     lsu_req_if,    
+    VX_csr_req_if.slave     csr_req_if,  
+`ifdef EXT_F_ENABLE
+    VX_fpu_req_if.slave     fpu_req_if,    
+`endif
+    VX_gpu_req_if.slave     gpu_req_if,
+    
+    // outputs
+    VX_branch_ctl_if.master branch_ctl_if,    
+    VX_warp_ctl_if.master   warp_ctl_if,
+    VX_commit_if.master     alu_commit_if,
+    VX_commit_if.master     ld_commit_if,
+    VX_commit_if.master     st_commit_if,
+    VX_commit_if.master     csr_commit_if,
+`ifdef EXT_F_ENABLE
+    VX_commit_if.master     fpu_commit_if,
+`endif
+    VX_commit_if.master     gpu_commit_if,
+    
+    input wire              busy
+);     
+
+`ifdef EXT_TEX_ENABLE
+
+    VX_dcache_req_if #(
+        .NUM_REQS  (`NUM_THREADS), 
+        .WORD_SIZE (4), 
+        .TAG_WIDTH (`LSU_TEX_DCACHE_TAG_BITS)
+    ) lsu_dcache_req_if();
+
+    VX_dcache_rsp_if #(
+        .NUM_REQS  (`NUM_THREADS), 
+        .WORD_SIZE (4), 
+        .TAG_WIDTH (`LSU_TEX_DCACHE_TAG_BITS)
+    ) lsu_dcache_rsp_if();
+
+    VX_dcache_req_if #(
+        .NUM_REQS  (`NUM_THREADS), 
+        .WORD_SIZE (4), 
+        .TAG_WIDTH (`LSU_TEX_DCACHE_TAG_BITS)
+    ) tex_dcache_req_if();
+
+    VX_dcache_rsp_if #(
+        .NUM_REQS  (`NUM_THREADS), 
+        .WORD_SIZE (4), 
+        .TAG_WIDTH (`LSU_TEX_DCACHE_TAG_BITS)
+    ) tex_dcache_rsp_if();
+
+    VX_tex_csr_if tex_csr_if();
+
+`ifdef PERF_ENABLE
+    VX_perf_tex_if perf_tex_if();
+`endif
+
+    VX_cache_arb #(
+        .NUM_REQS      (2),
+        .LANES         (`NUM_THREADS),
+        .DATA_SIZE     (4),            
+        .TAG_IN_WIDTH  (`LSU_TEX_DCACHE_TAG_BITS),
+        .TAG_SEL_IDX   (`NC_TAG_BIT + `SM_ENABLE)
+    ) tex_lsu_arb (
+        .clk            (clk),
+        .reset          (reset),
+
+        // Tex/LSU request
+        .req_valid_in   ({tex_dcache_req_if.valid,  lsu_dcache_req_if.valid}),
+        .req_rw_in      ({tex_dcache_req_if.rw,     lsu_dcache_req_if.rw}),
+        .req_byteen_in  ({tex_dcache_req_if.byteen, lsu_dcache_req_if.byteen}),
+        .req_addr_in    ({tex_dcache_req_if.addr,   lsu_dcache_req_if.addr}),
+        .req_data_in    ({tex_dcache_req_if.data,   lsu_dcache_req_if.data}),  
+        .req_tag_in     ({tex_dcache_req_if.tag,    lsu_dcache_req_if.tag}),  
+        .req_ready_in   ({tex_dcache_req_if.ready,  lsu_dcache_req_if.ready}),
+
+        // Dcache request
+        .req_valid_out  (dcache_req_if.valid),
+        .req_rw_out     (dcache_req_if.rw),        
+        .req_byteen_out (dcache_req_if.byteen),        
+        .req_addr_out   (dcache_req_if.addr),
+        .req_data_out   (dcache_req_if.data),
+        .req_tag_out    (dcache_req_if.tag),
+        .req_ready_out  (dcache_req_if.ready),
+        
+        // Dcache response
+        .rsp_valid_in   (dcache_rsp_if.valid),
+        .rsp_tmask_in   (dcache_rsp_if.tmask),
+        .rsp_tag_in     (dcache_rsp_if.tag),
+        .rsp_data_in    (dcache_rsp_if.data),
+        .rsp_ready_in   (dcache_rsp_if.ready),
+
+        // Tex/LSU response
+        .rsp_valid_out  ({tex_dcache_rsp_if.valid, lsu_dcache_rsp_if.valid}),
+        .rsp_tmask_out  ({tex_dcache_rsp_if.tmask, lsu_dcache_rsp_if.tmask}),
+        .rsp_data_out   ({tex_dcache_rsp_if.data,  lsu_dcache_rsp_if.data}),
+        .rsp_tag_out    ({tex_dcache_rsp_if.tag,   lsu_dcache_rsp_if.tag}),
+        .rsp_ready_out  ({tex_dcache_rsp_if.ready, lsu_dcache_rsp_if.ready})
+    );
+
+`endif
+
+`ifdef EXT_F_ENABLE
+    wire [`NUM_WARPS-1:0] csr_pending;
+    wire [`NUM_WARPS-1:0] fpu_pending;
+    VX_fpu_to_csr_if fpu_to_csr_if();
+`endif
+
+    `RESET_RELAY (alu_reset);
+    `RESET_RELAY (lsu_reset);
+    `RESET_RELAY (csr_reset);
+    `RESET_RELAY (gpu_reset);
+    
+    VX_alu_unit #(
+        .CORE_ID(CORE_ID)
+    ) alu_unit (
+        .clk            (clk),
+        .reset          (alu_reset),
+        .alu_req_if     (alu_req_if),
+        .branch_ctl_if  (branch_ctl_if),
+        .alu_commit_if  (alu_commit_if)
+    );
+
+    VX_lsu_unit #(
+        .CORE_ID(CORE_ID)
+    ) lsu_unit (
+        `SCOPE_BIND_VX_execute_lsu_unit
+        .clk            (clk),
+        .reset          (lsu_reset),
+    `ifdef EXT_TEX_ENABLE
+        .dcache_req_if  (lsu_dcache_req_if),
+        .dcache_rsp_if  (lsu_dcache_rsp_if),
+    `else 
+        .dcache_req_if  (dcache_req_if),
+        .dcache_rsp_if  (dcache_rsp_if),
+    `endif
+        .lsu_req_if     (lsu_req_if),
+        .ld_commit_if   (ld_commit_if),
+        .st_commit_if   (st_commit_if)
+    );
+
+    VX_csr_unit #(
+        .CORE_ID(CORE_ID)
+    ) csr_unit (
+        .clk            (clk),
+        .reset          (csr_reset),   
+    `ifdef PERF_ENABLE
+    `ifdef EXT_TEX_ENABLE
+        .perf_tex_if    (perf_tex_if),
+    `endif
+        .perf_memsys_if (perf_memsys_if),
+        .perf_pipeline_if(perf_pipeline_if),
+    `endif    
+        .cmt_to_csr_if  (cmt_to_csr_if),    
+        .fetch_to_csr_if(fetch_to_csr_if),
+        .csr_req_if     (csr_req_if),   
+        .csr_commit_if  (csr_commit_if),
+    `ifdef EXT_F_ENABLE  
+        .fpu_to_csr_if  (fpu_to_csr_if),
+        .fpu_pending    (fpu_pending),        
+        .pending        (csr_pending),
+    `else
+        `UNUSED_PIN (pending),
+    `endif        
+    `ifdef EXT_TEX_ENABLE
+        .tex_csr_if     (tex_csr_if),
+    `endif
+        .busy           (busy)
+    );
+
+`ifdef EXT_F_ENABLE
+    `RESET_RELAY (fpu_reset);
+
+    VX_fpu_unit #(
+        .CORE_ID(CORE_ID)
+    ) fpu_unit (
+        .clk            (clk),
+        .reset          (fpu_reset),        
+        .fpu_req_if     (fpu_req_if), 
+        .fpu_to_csr_if  (fpu_to_csr_if), 
+        .fpu_commit_if  (fpu_commit_if),
+        .csr_pending    (csr_pending),
+        .pending        (fpu_pending) 
+    );
+`endif
+
+    VX_gpu_unit #(
+        .CORE_ID(CORE_ID)
+    ) gpu_unit (
+        `SCOPE_BIND_VX_execute_gpu_unit
+        .clk            (clk),
+        .reset          (gpu_reset),    
+        .gpu_req_if     (gpu_req_if),
+    `ifdef EXT_TEX_ENABLE
+    `ifdef PERF_ENABLE
+        .perf_tex_if    (perf_tex_if),
+    `endif
+        .tex_csr_if     (tex_csr_if),
+        .dcache_req_if  (tex_dcache_req_if),
+        .dcache_rsp_if  (tex_dcache_rsp_if),
+    `endif
+        .warp_ctl_if    (warp_ctl_if),
+        .gpu_commit_if  (gpu_commit_if)
+    );
+
+    // special workaround to get RISC-V tests Pass/Fail status
+    wire ebreak /* verilator public */;
+    assign ebreak = alu_req_if.valid && alu_req_if.ready
+                 && `INST_ALU_IS_BR(alu_req_if.op_mod)
+                 && (`INST_BR_BITS'(alu_req_if.op_type) == `INST_BR_EBREAK 
+                  || `INST_BR_BITS'(alu_req_if.op_type) == `INST_BR_ECALL);
+
+endmodule

hw/rtl/VX_fetch.sv

@@ -0,0 +1,68 @@
+`include "VX_define.vh"
+
+module VX_fetch #(
+    parameter CORE_ID = 0
+) (
+    `SCOPE_IO_VX_fetch
+
+    input wire clk,
+    input wire reset,
+
+    // Icache interface
+    VX_icache_req_if.master icache_req_if,
+    VX_icache_rsp_if.slave  icache_rsp_if,
+
+    // inputs
+    VX_wstall_if.slave      wstall_if,
+    VX_join_if.slave        join_if,
+    VX_branch_ctl_if.slave  branch_ctl_if,
+    VX_warp_ctl_if.slave    warp_ctl_if,
+
+    // outputs
+    VX_ifetch_rsp_if.master ifetch_rsp_if,
+
+    // csr interface
+    VX_fetch_to_csr_if.master fetch_to_csr_if,
+
+    // busy status
+    output wire             busy
+);
+
+    VX_ifetch_req_if  ifetch_req_if();
+
+    VX_warp_sched #(
+        .CORE_ID(CORE_ID)
+    ) warp_sched (
+        `SCOPE_BIND_VX_fetch_warp_sched
+
+        .clk              (clk),
+        .reset            (reset),     
+
+        .warp_ctl_if      (warp_ctl_if),
+        .wstall_if        (wstall_if),
+        .join_if          (join_if),
+        .branch_ctl_if    (branch_ctl_if),
+
+        .ifetch_req_if    (ifetch_req_if),
+
+        .fetch_to_csr_if  (fetch_to_csr_if),
+        
+        .busy             (busy)
+    ); 
+
+    VX_icache_stage #(
+        .CORE_ID(CORE_ID)
+    ) icache_stage (
+        `SCOPE_BIND_VX_fetch_icache_stage
+
+        .clk            (clk),
+        .reset          (reset),        
+        
+        .icache_rsp_if  (icache_rsp_if),
+        .icache_req_if  (icache_req_if),
+
+        .ifetch_req_if  (ifetch_req_if),
+        .ifetch_rsp_if  (ifetch_rsp_if)   
+    );
+
+endmodule

hw/rtl/VX_fpu_unit.sv

@@ -0,0 +1,219 @@
+`include "VX_define.vh"
+
+module VX_fpu_unit #(
+    parameter CORE_ID = 0
+) (
+    input wire clk,
+    input wire reset,
+
+    VX_fpu_req_if.slave     fpu_req_if,
+    VX_fpu_to_csr_if.master fpu_to_csr_if,
+    VX_commit_if.master     fpu_commit_if,
+
+    input wire[`NUM_WARPS-1:0] csr_pending,
+    output wire[`NUM_WARPS-1:0] pending
+); 
+    import fpu_types::*;
+    
+    `UNUSED_PARAM (CORE_ID)    
+    localparam FPUQ_BITS = `LOG2UP(`FPUQ_SIZE);
+
+    wire ready_in;
+    wire valid_out;
+    wire ready_out;
+
+    wire [`UUID_BITS-1:0] rsp_uuid;
+    wire [`NW_BITS-1:0] rsp_wid;
+    wire [`NUM_THREADS-1:0] rsp_tmask;
+    wire [31:0] rsp_PC;
+    wire [`NR_BITS-1:0] rsp_rd;
+    wire rsp_wb;
+
+    wire has_fflags;
+    fflags_t [`NUM_THREADS-1:0] fflags;
+    wire [`NUM_THREADS-1:0][31:0] result;
+
+    wire [FPUQ_BITS-1:0] tag_in, tag_out;    
+    wire fpuq_full;
+
+    wire fpuq_push = fpu_req_if.valid && fpu_req_if.ready;
+    wire fpuq_pop  = valid_out && ready_out;
+
+    VX_index_buffer #(
+        .DATAW   (`UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `NR_BITS + 1),
+        .SIZE    (`FPUQ_SIZE)
+    ) req_metadata  (
+        .clk          (clk),
+        .reset        (reset),
+        .acquire_slot (fpuq_push),       
+        .write_addr   (tag_in),                
+        .read_addr    (tag_out),
+        .release_addr (tag_out),        
+        .write_data   ({fpu_req_if.uuid, fpu_req_if.wid, fpu_req_if.tmask, fpu_req_if.PC, fpu_req_if.rd, fpu_req_if.wb}),                    
+        .read_data    ({rsp_uuid,        rsp_wid,        rsp_tmask,        rsp_PC,        rsp_rd,        rsp_wb}), 
+        .release_slot (fpuq_pop),     
+        .full         (fpuq_full),
+        `UNUSED_PIN (empty)
+    );
+    
+    // can accept new request?
+    assign fpu_req_if.ready = ready_in && ~fpuq_full && !csr_pending[fpu_req_if.wid];
+
+    wire valid_in = fpu_req_if.valid && ~fpuq_full && !csr_pending[fpu_req_if.wid];
+
+    // resolve dynamic FRM from CSR   
+    assign fpu_to_csr_if.read_wid = fpu_req_if.wid;
+    wire [`INST_FRM_BITS-1:0] fpu_frm = (fpu_req_if.op_mod == `INST_FRM_DYN) ? fpu_to_csr_if.read_frm : fpu_req_if.op_mod;
+
+`ifdef FPU_DPI
+
+    VX_fpu_dpi #(
+        .TAGW (FPUQ_BITS)
+    ) fpu_dpi (
+        .clk        (clk),
+        .reset      (reset),   
+
+        .valid_in   (valid_in),
+        .ready_in   (ready_in),        
+
+        .tag_in     (tag_in),
+        
+        .op_type    (fpu_req_if.op_type),
+        .frm        (fpu_frm),
+
+        .dataa      (fpu_req_if.rs1_data),
+        .datab      (fpu_req_if.rs2_data),
+        .datac      (fpu_req_if.rs3_data),
+        .result     (result), 
+
+        .has_fflags (has_fflags),
+        .fflags     (fflags),
+
+        .tag_out    (tag_out),
+
+        .ready_out  (ready_out),
+        .valid_out  (valid_out)
+    );
+
+`elsif FPU_FPNEW
+
+    VX_fpu_fpnew #(
+        .FMULADD  (1),
+        .FDIVSQRT (1),
+        .FNONCOMP (1),
+        .FCONV    (1),
+        .TAGW     (FPUQ_BITS)
+    ) fpu_fpnew (
+        .clk        (clk),
+        .reset      (reset),   
+
+        .valid_in   (valid_in),
+        .ready_in   (ready_in),        
+
+        .tag_in     (tag_in),
+        
+        .op_type    (fpu_req_if.op_type),
+        .frm        (fpu_frm),
+
+        .dataa      (fpu_req_if.rs1_data),
+        .datab      (fpu_req_if.rs2_data),
+        .datac      (fpu_req_if.rs3_data),
+        .result     (result), 
+
+        .has_fflags (has_fflags),
+        .fflags     (fflags),
+
+        .tag_out    (tag_out),
+
+        .ready_out  (ready_out),
+        .valid_out  (valid_out)
+    );   
+
+`else
+
+    VX_fpu_fpga #(
+        .TAGW (FPUQ_BITS)
+    ) fpu_fpga (
+        .clk        (clk),
+        .reset      (reset),   
+
+        .valid_in   (valid_in),
+        .ready_in   (ready_in),        
+
+        .tag_in     (tag_in),
+        
+        .op_type    (fpu_req_if.op_type),
+        .frm        (fpu_frm),
+
+        .dataa      (fpu_req_if.rs1_data),
+        .datab      (fpu_req_if.rs2_data),
+        .datac      (fpu_req_if.rs3_data),
+        .result     (result), 
+
+        .has_fflags (has_fflags),
+        .fflags     (fflags),
+
+        .tag_out    (tag_out),
+
+        .ready_out  (ready_out),
+        .valid_out  (valid_out)
+    );
+    
+`endif
+
+    reg has_fflags_r;
+    fflags_t fflags_r;
+    
+    fflags_t rsp_fflags;
+    always @(*) begin
+        rsp_fflags = '0;        
+        for (integer i = 0; i < `NUM_THREADS; i++) begin
+            if (rsp_tmask[i]) begin
+                rsp_fflags.NX |= fflags[i].NX;
+                rsp_fflags.UF |= fflags[i].UF;
+                rsp_fflags.OF |= fflags[i].OF;
+                rsp_fflags.DZ |= fflags[i].DZ;
+                rsp_fflags.NV |= fflags[i].NV;
+            end
+        end
+    end
+
+    wire stall_out = ~fpu_commit_if.ready && fpu_commit_if.valid;
+
+    VX_pipe_register #(
+        .DATAW  (1 + `UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `NR_BITS + 1 + (`NUM_THREADS * 32) + 1 + `FFLAGS_BITS),
+        .RESETW (1)
+    ) pipe_reg (
+        .clk      (clk),
+        .reset    (reset),
+        .enable   (!stall_out),
+        .data_in  ({valid_out,           rsp_uuid,           rsp_wid,           rsp_tmask,           rsp_PC,           rsp_rd,           rsp_wb,           result,             has_fflags,   rsp_fflags}),
+        .data_out ({fpu_commit_if.valid, fpu_commit_if.uuid, fpu_commit_if.wid, fpu_commit_if.tmask, fpu_commit_if.PC, fpu_commit_if.rd, fpu_commit_if.wb, fpu_commit_if.data, has_fflags_r, fflags_r})
+    );
+
+    assign fpu_commit_if.eop = 1'b1;
+
+    assign ready_out = ~stall_out;
+
+    // CSR fflags Update    
+    assign fpu_to_csr_if.write_enable = fpu_commit_if.valid && fpu_commit_if.ready && has_fflags_r;
+    assign fpu_to_csr_if.write_wid    = fpu_commit_if.wid;     
+    assign fpu_to_csr_if.write_fflags = fflags_r;
+
+    // pending request
+    reg [`NUM_WARPS-1:0] pending_r;
+    always @(posedge clk) begin
+        if (reset) begin
+            pending_r <= 0;
+        end else begin
+            if (fpu_commit_if.valid && fpu_commit_if.ready) begin
+                 pending_r[fpu_commit_if.wid] <= 0;
+            end          
+            if (fpu_req_if.valid && fpu_req_if.ready) begin
+                 pending_r[fpu_req_if.wid] <= 1;
+            end
+        end
+    end
+    assign pending = pending_r;
+
+endmodule

hw/rtl/VX_gpr_stage.sv

@@ -0,0 +1,91 @@
+`include "VX_define.vh"
+
+module VX_gpr_stage #(
+    parameter CORE_ID = 0
+) (
+    input wire              clk,
+    input wire              reset,
+
+    // inputs    
+    VX_writeback_if.slave   writeback_if,  
+    VX_gpr_req_if.slave     gpr_req_if,
+
+    // outputs
+    VX_gpr_rsp_if.master    gpr_rsp_if
+);
+
+    `UNUSED_PARAM (CORE_ID)
+    `UNUSED_VAR (reset)    
+
+    localparam RAM_SIZE = `NUM_WARPS * `NUM_REGS;
+
+    // ensure r0 never gets written, which can happen before the reset
+    wire write_enable = writeback_if.valid && (writeback_if.rd != 0);
+    
+    wire [`NUM_THREADS-1:0] wren;
+    for (genvar i = 0; i < `NUM_THREADS; ++i) begin
+        assign wren[i] = write_enable && writeback_if.tmask[i];
+    end
+
+    wire [$clog2(RAM_SIZE)-1:0] waddr, raddr1, raddr2;
+    assign waddr  = {writeback_if.wid, writeback_if.rd};
+    assign raddr1 = {gpr_req_if.wid, gpr_req_if.rs1};
+    assign raddr2 = {gpr_req_if.wid, gpr_req_if.rs2};
+
+    for (genvar i = 0; i < `NUM_THREADS; ++i) begin
+        VX_dp_ram #(
+            .DATAW       (32),
+            .SIZE        (RAM_SIZE),
+            .INIT_ENABLE (1),
+            .INIT_VALUE  (0)
+        ) dp_ram1 (
+            .clk   (clk),
+            .wren  (wren[i]),
+            .waddr (waddr),
+            .wdata (writeback_if.data[i]),
+            .raddr (raddr1),
+            .rdata (gpr_rsp_if.rs1_data[i])
+        );
+
+        VX_dp_ram #(
+            .DATAW       (32),
+            .SIZE        (RAM_SIZE),
+            .INIT_ENABLE (1),
+            .INIT_VALUE  (0)
+        ) dp_ram2 (
+            .clk   (clk),
+            .wren  (wren[i]),
+            .waddr (waddr),
+            .wdata (writeback_if.data[i]),
+            .raddr (raddr2),
+            .rdata (gpr_rsp_if.rs2_data[i])
+        );
+    end
+    
+`ifdef EXT_F_ENABLE
+    wire [$clog2(RAM_SIZE)-1:0] raddr3;
+    assign raddr3 = {gpr_req_if.wid, gpr_req_if.rs3};    
+
+    for (genvar i = 0; i < `NUM_THREADS; ++i) begin
+        VX_dp_ram #(
+            .DATAW       (32),
+            .SIZE        (RAM_SIZE),
+            .INIT_ENABLE (1),
+            .INIT_VALUE  (0)
+        ) dp_ram3 (
+            .clk   (clk),
+            .wren  (wren[i]),
+            .waddr (waddr),
+            .wdata (writeback_if.data[i]),
+            .raddr (raddr3),
+            .rdata (gpr_rsp_if.rs3_data[i])
+        );
+    end
+`else    
+    `UNUSED_VAR (gpr_req_if.rs3)    
+    assign gpr_rsp_if.rs3_data = 'x;
+`endif
+    
+    assign writeback_if.ready = 1'b1;
+
+endmodule

hw/rtl/VX_gpu_pkg.sv

@@ -1,248 +0,0 @@
-// Copyright © 2019-2023
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// 
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-`ifndef VX_GPU_PKG_VH
-`define VX_GPU_PKG_VH
-
-`include "VX_define.vh"
-
-package VX_gpu_pkg;
-
-    typedef struct packed {
-        logic                    valid;
-        logic [`NUM_THREADS-1:0] tmask;
-    } tmc_t;
-
-    typedef struct packed {
-        logic                   valid;
-        logic [`NUM_WARPS-1:0]  wmask;
-        logic [`XLEN-1:0]       pc;
-    } wspawn_t;
-
-    typedef struct packed {
-        logic                    valid;
-        logic                    is_dvg;
-        logic [`NUM_THREADS-1:0] then_tmask;
-        logic [`NUM_THREADS-1:0] else_tmask;
-        logic [`XLEN-1:0]        next_pc;
-    } split_t;
-
-    typedef struct packed {
-        logic valid;
-        logic is_dvg;
-    } join_t;
-
-    typedef struct packed {
-        logic                   valid;
-        logic [`NB_WIDTH-1:0]   id;
-        logic                   is_global;
-        logic [1:0]             domain;
-        logic [`NUM_WARPS-1:0]  mask;
-    `ifdef GBAR_ENABLE
-        logic [`MAX(`NW_WIDTH, `NC_WIDTH)-1:0] size_m1;
-    `else
-        logic [`NW_WIDTH-1:0]   size_m1;
-    `endif
-    } barrier_t;
-
-    localparam logic [1:0] BARRIER_ALL    = 2'd0;
-    localparam logic [1:0] BARRIER_SCALAR = 2'd1;
-    localparam logic [1:0] BARRIER_TENSOR = 2'd2;
-    localparam logic [1:0] BARRIER_MASK   = 2'd3;
-    localparam logic       WU_DOMAIN_SCALAR = 1'b0;
-    localparam logic       WU_DOMAIN_TENSOR = 1'b1;
-
-    typedef struct packed {
-        logic [`XLEN-1:0]   startup_addr;
-        logic [7:0]         mpm_class;
-    } base_dcrs_t;
-
-    typedef struct packed {
-        logic [`PERF_CTR_BITS-1:0] reads;
-        logic [`PERF_CTR_BITS-1:0] writes;
-        logic [`PERF_CTR_BITS-1:0] read_misses;
-        logic [`PERF_CTR_BITS-1:0] write_misses;
-        logic [`PERF_CTR_BITS-1:0] bank_stalls;
-        logic [`PERF_CTR_BITS-1:0] mshr_stalls;
-        logic [`PERF_CTR_BITS-1:0] mem_stalls;
-        logic [`PERF_CTR_BITS-1:0] crsp_stalls;
-    } cache_perf_t;
-
-    typedef struct packed {
-        logic [`PERF_CTR_BITS-1:0] reads;
-        logic [`PERF_CTR_BITS-1:0] writes;
-        logic [`PERF_CTR_BITS-1:0] latency;
-    } mem_perf_t;
-
-    /* verilator lint_off UNUSED */
-
-    ////////////////////////// Icache Parameters //////////////////////////////
-
-    // Word size in bytes
-    localparam ICACHE_WORD_SIZE	    = 4;
-    localparam ICACHE_ADDR_WIDTH	= (`MEM_ADDR_WIDTH - `CLOG2(ICACHE_WORD_SIZE));
-
-    // Block size in bytes
-    localparam ICACHE_LINE_SIZE	    = `L1_LINE_SIZE;
-
-    // Core request tag Id bits       
-    localparam ICACHE_TAG_ID_BITS	= `NW_WIDTH;
-
-    // Core request tag bits
-    localparam ICACHE_TAG_WIDTH	    = (1 + `UUID_WIDTH + ICACHE_TAG_ID_BITS);
-
-    // Memory request data bits
-    localparam ICACHE_MEM_DATA_WIDTH = (ICACHE_LINE_SIZE * 8);
-
-    // Memory request tag bits
-    `ifdef ICACHE_ENABLE
-    localparam ICACHE_MEM_TAG_WIDTH = `CACHE_CLUSTER_MEM_TAG_WIDTH(`ICACHE_MSHR_SIZE, 1, `NUM_ICACHES);
-    `else
-    localparam ICACHE_MEM_TAG_WIDTH = `CACHE_CLUSTER_BYPASS_TAG_WIDTH(1, ICACHE_LINE_SIZE, ICACHE_WORD_SIZE, ICACHE_TAG_WIDTH, `SOCKET_SIZE, `NUM_ICACHES);
-    `endif
-
-    ////////////////////////// Dcache Parameters //////////////////////////////
-
-    // Word size in bytes
-    localparam DCACHE_WORD_SIZE	    = (`XLEN / 8);
-    localparam DCACHE_ADDR_WIDTH	= (`MEM_ADDR_WIDTH - `CLOG2(DCACHE_WORD_SIZE));
-
-    // Block size in bytes
-    localparam DCACHE_LINE_SIZE 	= `L1_LINE_SIZE;
-
-    // Input request size
-    localparam DCACHE_NUM_REQS	    = `MAX(`DCACHE_NUM_BANKS, `SMEM_NUM_BANKS);
-
-    // Memory request size
-    localparam LSU_MEM_REQS	        = `NUM_LSU_LANES;
-
-    // Batch select bits
-    localparam DCACHE_NUM_BATCHES	= ((LSU_MEM_REQS + DCACHE_NUM_REQS - 1) / DCACHE_NUM_REQS);
-    localparam DCACHE_BATCH_SEL_BITS = `CLOG2(DCACHE_NUM_BATCHES);
-
-    // Core request tag Id bits
-    localparam LSUQ_TAG_BITS	    = (`CLOG2(`LSUQ_SIZE) + DCACHE_BATCH_SEL_BITS);
-    localparam DCACHE_TAG_ID_BITS	= (LSUQ_TAG_BITS + `CACHE_ADDR_TYPE_BITS);
-
-    // Core request tag bits
-    localparam DCACHE_TAG_WIDTH	    = (`UUID_WIDTH + DCACHE_TAG_ID_BITS);
-    localparam DCACHE_NOSM_TAG_WIDTH = (DCACHE_TAG_WIDTH - `SM_ENABLED);
-    
-    // Memory request data bits
-    localparam DCACHE_MEM_DATA_WIDTH = (DCACHE_LINE_SIZE * 8);
-
-    // Memory request tag bits
-    `ifdef DCACHE_ENABLE
-    localparam DCACHE_MEM_TAG_WIDTH = `CACHE_CLUSTER_NC_MEM_TAG_WIDTH(`DCACHE_MSHR_SIZE, `DCACHE_NUM_BANKS, DCACHE_NUM_REQS, DCACHE_LINE_SIZE, DCACHE_WORD_SIZE, DCACHE_NOSM_TAG_WIDTH, `SOCKET_SIZE, `NUM_DCACHES);
-    `else
-    localparam DCACHE_MEM_TAG_WIDTH = `CACHE_CLUSTER_NC_BYPASS_TAG_WIDTH(DCACHE_NUM_REQS, DCACHE_LINE_SIZE, DCACHE_WORD_SIZE, DCACHE_NOSM_TAG_WIDTH, `SOCKET_SIZE, `NUM_DCACHES);
-    `endif
-
-    /////////////////////////////// L1 Parameters /////////////////////////////
-
-    localparam L1_MEM_TAG_WIDTH     = `MAX(ICACHE_MEM_TAG_WIDTH, DCACHE_MEM_TAG_WIDTH);
-    localparam L1_MEM_ARB_TAG_WIDTH = (L1_MEM_TAG_WIDTH + `CLOG2(2));
-    
-    /////////////////////////////// L2 Parameters /////////////////////////////
-
-    localparam ICACHE_MEM_ARB_IDX = 0;
-    localparam DCACHE_MEM_ARB_IDX = ICACHE_MEM_ARB_IDX + 1;
-
-    // Word size in bytes
-    localparam L2_WORD_SIZE	        = `L1_LINE_SIZE;
-
-    // Input request size
-    localparam L2_NUM_REQS	        = `NUM_SOCKETS;
-
-    // Core request tag bits
-    localparam L2_TAG_WIDTH	        = L1_MEM_ARB_TAG_WIDTH;
-
-    // Memory request data bits
-    localparam L2_MEM_DATA_WIDTH	= (`L2_LINE_SIZE * 8);
-
-    // Memory request tag bits
-    `ifdef L2_ENABLE
-    localparam L2_MEM_TAG_WIDTH     = `CACHE_NC_MEM_TAG_WIDTH(`L2_MSHR_SIZE, `L2_NUM_BANKS, L2_NUM_REQS, `L2_LINE_SIZE, L2_WORD_SIZE, L2_TAG_WIDTH);
-    `else
-    localparam L2_MEM_TAG_WIDTH     = `CACHE_NC_BYPASS_TAG_WIDTH(L2_NUM_REQS, `L2_LINE_SIZE, L2_WORD_SIZE, L2_TAG_WIDTH);
-    `endif
-
-    /////////////////////////////// L3 Parameters /////////////////////////////
-
-    // Word size in bytes
-    localparam L3_WORD_SIZE	        = `L2_LINE_SIZE;
-
-    // Input request size
-    localparam L3_NUM_REQS	        = `NUM_CLUSTERS;
-
-    // Core request tag bits
-    localparam L3_TAG_WIDTH	        = L2_MEM_TAG_WIDTH;
-
-    // Memory request data bits
-    localparam L3_MEM_DATA_WIDTH	= (`L3_LINE_SIZE * 8);
-
-    // Memory request tag bits
-    `ifdef L3_ENABLE
-    localparam L3_MEM_TAG_WIDTH     = `CACHE_NC_MEM_TAG_WIDTH(`L3_MSHR_SIZE, `L3_NUM_BANKS, L3_NUM_REQS, `L3_LINE_SIZE, L3_WORD_SIZE, L3_TAG_WIDTH);
-    `else
-    localparam L3_MEM_TAG_WIDTH     = `CACHE_NC_BYPASS_TAG_WIDTH(L3_NUM_REQS, `L3_LINE_SIZE, L3_WORD_SIZE, L3_TAG_WIDTH);
-    `endif
-
-    /* verilator lint_on UNUSED */
-
-    /////////////////////////////// Issue parameters //////////////////////////
-
-    localparam ISSUE_ISW   = `CLOG2(`ISSUE_WIDTH);
-    localparam ISSUE_ISW_W = `UP(ISSUE_ISW);   
-    localparam ISSUE_RATIO = `NUM_WARPS / `ISSUE_WIDTH;
-    localparam ISSUE_WIS   = `CLOG2(ISSUE_RATIO);
-    localparam ISSUE_WIS_W = `UP(ISSUE_WIS);
-    
-`IGNORE_UNUSED_BEGIN
-    function logic [`NW_WIDTH-1:0] wis_to_wid(
-        input logic [ISSUE_WIS_W-1:0] wis, 
-        input logic [ISSUE_ISW_W-1:0] isw
-    );
-        if (ISSUE_WIS == 0) begin
-            wis_to_wid = `NW_WIDTH'(isw);
-        end else if (ISSUE_ISW == 0) begin
-            wis_to_wid = `NW_WIDTH'(wis);
-        end else begin 
-            wis_to_wid = `NW_WIDTH'({wis, isw});
-        end
-    endfunction
-
-    function logic [ISSUE_ISW_W-1:0] wid_to_isw(
-        input logic [`NW_WIDTH-1:0] wid
-    );
-        if (ISSUE_ISW != 0) begin    
-            wid_to_isw = wid[ISSUE_ISW_W-1:0];
-        end else begin
-            wid_to_isw = 0;
-        end
-    endfunction
-
-    function logic [ISSUE_WIS_W-1:0] wid_to_wis(
-        input logic [`NW_WIDTH-1:0] wid
-    );
-        if (ISSUE_WIS != 0) begin
-            wid_to_wis = ISSUE_WIS_W'(wid >> ISSUE_ISW);
-        end else begin
-            wid_to_wis = 0;
-        end
-    endfunction
-`IGNORE_UNUSED_END
-
-endpackage
-
-`endif // VX_GPU_PKG_VH

hw/rtl/VX_gpu_types.vh

@@ -0,0 +1,43 @@
+`ifndef VX_GPU_TYPES
+`define VX_GPU_TYPES
+
+`include "VX_define.vh"
+
+package gpu_types;
+
+typedef struct packed {
+    logic                    valid;
+    logic [`NUM_THREADS-1:0] tmask;
+} gpu_tmc_t;
+
+`define GPU_TMC_BITS $bits(gpu_types::gpu_tmc_t)
+
+typedef struct packed {
+    logic                   valid;
+    logic [`NUM_WARPS-1:0]  wmask;
+    logic [31:0]            pc;
+} gpu_wspawn_t;
+
+`define GPU_WSPAWN_BITS $bits(gpu_types::gpu_wspawn_t)
+
+typedef struct packed {
+    logic                   valid;
+    logic                   diverged;
+    logic [`NUM_THREADS-1:0] then_tmask;
+    logic [`NUM_THREADS-1:0] else_tmask;
+    logic [31:0]            pc;
+} gpu_split_t;
+
+`define GPU_SPLIT_BITS $bits(gpu_types::gpu_split_t)
+
+typedef struct packed {
+    logic                   valid;
+    logic [`NB_BITS-1:0]    id;
+    logic [`NW_BITS-1:0]    size_m1;
+} gpu_barrier_t;
+
+`define GPU_BARRIER_BITS $bits(gpu_types::gpu_barrier_t)
+
+endpackage
+
+`endif

hw/rtl/VX_gpu_unit.sv

@@ -0,0 +1,220 @@
+`include "VX_define.vh"
+
+module VX_gpu_unit #(
+    parameter CORE_ID = 0
+) (
+    `SCOPE_IO_VX_gpu_unit
+    
+    input wire          clk,
+    input wire          reset,
+
+    // Inputs
+    VX_gpu_req_if.slave gpu_req_if,
+
+`ifdef EXT_TEX_ENABLE
+    // PERF
+`ifdef PERF_ENABLE
+    VX_perf_tex_if.master perf_tex_if,
+`endif
+    VX_dcache_req_if.master dcache_req_if,
+    VX_dcache_rsp_if.slave  dcache_rsp_if,
+    VX_tex_csr_if.slave tex_csr_if,
+`endif
+
+    // Outputs
+    VX_warp_ctl_if.master warp_ctl_if,
+    VX_commit_if.master gpu_commit_if
+);
+    import gpu_types::*;
+
+    `UNUSED_PARAM (CORE_ID)
+
+    localparam WCTL_DATAW = `GPU_TMC_BITS + `GPU_WSPAWN_BITS + `GPU_SPLIT_BITS + `GPU_BARRIER_BITS;
+    localparam RSP_DATAW  = `MAX(`NUM_THREADS * 32, WCTL_DATAW);
+
+    wire                    rsp_valid;
+    wire [`UUID_BITS-1:0]   rsp_uuid;
+    wire [`NW_BITS-1:0]     rsp_wid;
+    wire [`NUM_THREADS-1:0] rsp_tmask;
+    wire [31:0]             rsp_PC;
+    wire [`NR_BITS-1:0]     rsp_rd;   
+    wire                    rsp_wb;
+
+    wire [RSP_DATAW-1:0] rsp_data, rsp_data_r;
+
+    gpu_tmc_t       tmc;
+    gpu_wspawn_t    wspawn;
+    gpu_barrier_t   barrier;
+    gpu_split_t     split;
+    
+    wire [WCTL_DATAW-1:0] warp_ctl_data;
+    wire is_warp_ctl;
+    
+    wire stall_in, stall_out;
+    
+    wire is_wspawn = (gpu_req_if.op_type == `INST_GPU_WSPAWN);
+    wire is_tmc    = (gpu_req_if.op_type == `INST_GPU_TMC);
+    wire is_split  = (gpu_req_if.op_type == `INST_GPU_SPLIT);
+    wire is_bar    = (gpu_req_if.op_type == `INST_GPU_BAR);
+    wire is_pred   = (gpu_req_if.op_type == `INST_GPU_PRED);
+
+    wire [31:0] rs1_data = gpu_req_if.rs1_data[gpu_req_if.tid];
+    wire [31:0] rs2_data = gpu_req_if.rs2_data[gpu_req_if.tid];
+    
+    wire [`NUM_THREADS-1:0] taken_tmask;
+    wire [`NUM_THREADS-1:0] not_taken_tmask;
+
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin
+        wire taken = (gpu_req_if.rs1_data[i] != 0);
+        assign taken_tmask[i]     = gpu_req_if.tmask[i] & taken;
+        assign not_taken_tmask[i] = gpu_req_if.tmask[i] & ~taken;
+    end
+
+    // tmc
+
+    wire [`NUM_THREADS-1:0] pred_mask = (taken_tmask != 0) ? taken_tmask : gpu_req_if.tmask;
+
+    assign tmc.valid = is_tmc || is_pred;
+    assign tmc.tmask = is_pred ? pred_mask : rs1_data[`NUM_THREADS-1:0];
+
+    // wspawn
+
+    wire [31:0] wspawn_pc = rs2_data;
+    wire [`NUM_WARPS-1:0] wspawn_wmask;
+    for (genvar i = 0; i < `NUM_WARPS; i++) begin
+        assign wspawn_wmask[i] = (i < rs1_data);
+    end
+    assign wspawn.valid = is_wspawn;
+    assign wspawn.wmask = wspawn_wmask;
+    assign wspawn.pc    = wspawn_pc;
+
+    // split
+
+    assign split.valid      = is_split;
+    assign split.diverged   = (| taken_tmask) && (| not_taken_tmask);
+    assign split.then_tmask = taken_tmask;
+    assign split.else_tmask = not_taken_tmask;
+    assign split.pc         = gpu_req_if.next_PC;
+
+    // barrier
+    
+    assign barrier.valid   = is_bar;
+    assign barrier.id      = rs1_data[`NB_BITS-1:0];
+    assign barrier.size_m1 = (`NW_BITS)'(rs2_data - 1);       
+
+    // pack warp ctl result
+    assign warp_ctl_data = {tmc, wspawn, split, barrier};
+
+    // texture
+
+`ifdef EXT_TEX_ENABLE
+
+    `UNUSED_VAR (gpu_req_if.op_mod)
+    
+    VX_tex_req_if   tex_req_if();
+    VX_tex_rsp_if   tex_rsp_if();    
+
+    wire is_tex = (gpu_req_if.op_type == `INST_GPU_TEX);
+
+    assign tex_req_if.valid = gpu_req_if.valid && is_tex;
+    assign tex_req_if.uuid  = gpu_req_if.uuid;
+    assign tex_req_if.wid   = gpu_req_if.wid;
+    assign tex_req_if.tmask = gpu_req_if.tmask;
+    assign tex_req_if.PC    = gpu_req_if.PC;
+    assign tex_req_if.rd    = gpu_req_if.rd;
+    assign tex_req_if.wb    = gpu_req_if.wb;
+    
+    assign tex_req_if.unit      = gpu_req_if.op_mod[`NTEX_BITS-1:0];
+    assign tex_req_if.coords[0] = gpu_req_if.rs1_data;
+    assign tex_req_if.coords[1] = gpu_req_if.rs2_data;
+    assign tex_req_if.lod       = gpu_req_if.rs3_data;        
+
+    VX_tex_unit #(
+        .CORE_ID(CORE_ID)
+    ) tex_unit (
+        .clk           (clk),
+        .reset         (reset),
+    `ifdef PERF_ENABLE
+        .perf_tex_if   (perf_tex_if),
+    `endif
+        .tex_req_if    (tex_req_if),
+        .tex_csr_if    (tex_csr_if),
+        .tex_rsp_if    (tex_rsp_if),
+        .dcache_req_if (dcache_req_if),
+        .dcache_rsp_if (dcache_rsp_if)
+    );
+
+    assign tex_rsp_if.ready = !stall_out;
+
+    assign stall_in = (is_tex && ~tex_req_if.ready)
+                   || (~is_tex && (tex_rsp_if.valid || stall_out));
+
+    assign is_warp_ctl = !(is_tex || tex_rsp_if.valid);
+
+    assign rsp_valid = tex_rsp_if.valid || (gpu_req_if.valid && ~is_tex);
+    assign rsp_uuid  = tex_rsp_if.valid ? tex_rsp_if.uuid : gpu_req_if.uuid;
+    assign rsp_wid   = tex_rsp_if.valid ? tex_rsp_if.wid : gpu_req_if.wid;
+    assign rsp_tmask = tex_rsp_if.valid ? tex_rsp_if.tmask : gpu_req_if.tmask;
+    assign rsp_PC    = tex_rsp_if.valid ? tex_rsp_if.PC : gpu_req_if.PC;
+    assign rsp_rd    = tex_rsp_if.rd;
+    assign rsp_wb    = tex_rsp_if.valid && tex_rsp_if.wb;
+    assign rsp_data  = tex_rsp_if.valid ? RSP_DATAW'(tex_rsp_if.data) : RSP_DATAW'(warp_ctl_data);
+    
+`else   
+
+    `UNUSED_VAR (gpu_req_if.op_mod)
+    `UNUSED_VAR (gpu_req_if.rs3_data)
+    `UNUSED_VAR (gpu_req_if.wb)
+    `UNUSED_VAR (gpu_req_if.rd)
+
+    assign stall_in = stall_out;
+    assign is_warp_ctl = 1;
+
+    assign rsp_valid = gpu_req_if.valid;
+    assign rsp_uuid  = gpu_req_if.uuid;
+    assign rsp_wid   = gpu_req_if.wid;
+    assign rsp_tmask = gpu_req_if.tmask;
+    assign rsp_PC    = gpu_req_if.PC;
+    assign rsp_rd    = 0;
+    assign rsp_wb    = 0;
+    assign rsp_data  = RSP_DATAW'(warp_ctl_data);
+
+`endif
+
+    wire is_warp_ctl_r;
+
+    // output
+    assign stall_out = ~gpu_commit_if.ready && gpu_commit_if.valid;
+
+    VX_pipe_register #(
+        .DATAW  (1 + `UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `NR_BITS + 1 + RSP_DATAW + 1),
+        .RESETW (1)
+    ) pipe_reg (
+        .clk      (clk),
+        .reset    (reset),
+        .enable   (!stall_out),
+        .data_in  ({rsp_valid,           rsp_uuid,           rsp_wid,           rsp_tmask,           rsp_PC,           rsp_rd,           rsp_wb,           rsp_data,   is_warp_ctl}),
+        .data_out ({gpu_commit_if.valid, gpu_commit_if.uuid, gpu_commit_if.wid, gpu_commit_if.tmask, gpu_commit_if.PC, gpu_commit_if.rd, gpu_commit_if.wb, rsp_data_r, is_warp_ctl_r})
+    );  
+
+    assign gpu_commit_if.data = rsp_data_r[(`NUM_THREADS * 32)-1:0];
+    assign gpu_commit_if.eop  = 1'b1;
+
+    // warp control reponse
+     
+    assign {warp_ctl_if.tmc, warp_ctl_if.wspawn, warp_ctl_if.split, warp_ctl_if.barrier} = rsp_data_r[WCTL_DATAW-1:0];
+    
+    assign warp_ctl_if.valid = gpu_commit_if.valid && gpu_commit_if.ready && is_warp_ctl_r;
+    assign warp_ctl_if.wid   = gpu_commit_if.wid;    
+
+    // can accept new request?
+    assign gpu_req_if.ready = ~stall_in;
+
+    `SCOPE_ASSIGN (gpu_rsp_valid, warp_ctl_if.valid);
+    `SCOPE_ASSIGN (gpu_rsp_uuid, gpu_commit_if.uuid);
+    `SCOPE_ASSIGN (gpu_rsp_tmc, warp_ctl_if.tmc.valid);
+    `SCOPE_ASSIGN (gpu_rsp_wspawn, warp_ctl_if.wspawn.valid);          
+    `SCOPE_ASSIGN (gpu_rsp_split, warp_ctl_if.split.valid);
+    `SCOPE_ASSIGN (gpu_rsp_barrier, warp_ctl_if.barrier.valid);
+
+endmodule

hw/rtl/VX_ibuffer.sv

@@ -0,0 +1,210 @@
+`include "VX_define.vh"
+
+module VX_ibuffer #(
+    parameter CORE_ID = 0
+) (
+    input wire clk,
+    input wire reset,
+
+    // inputs
+    VX_decode_if.slave  decode_if,  
+
+    // outputs
+    VX_ibuffer_if.master ibuffer_if
+);
+
+    `UNUSED_PARAM (CORE_ID)
+    
+    localparam DATAW   = `UUID_BITS + `NUM_THREADS + 32 + `EX_BITS + `INST_OP_BITS + `INST_FRM_BITS + 1 + (`NR_BITS * 4) + 32 + 1 + 1;
+    localparam ADDRW   = $clog2(`IBUF_SIZE+1);
+    localparam NWARPSW = $clog2(`NUM_WARPS+1);
+
+    reg [`NUM_WARPS-1:0][ADDRW-1:0] used_r;
+    reg [`NUM_WARPS-1:0] full_r, empty_r, alm_empty_r;
+    
+    wire [`NUM_WARPS-1:0] q_full, q_empty, q_alm_empty;
+    wire [DATAW-1:0] q_data_in;
+    wire [`NUM_WARPS-1:0][DATAW-1:0] q_data_prev;    
+    reg [`NUM_WARPS-1:0][DATAW-1:0] q_data_out;
+
+    wire enq_fire = decode_if.valid && decode_if.ready;
+    wire deq_fire = ibuffer_if.valid && ibuffer_if.ready;
+
+    for (genvar i = 0; i < `NUM_WARPS; ++i) begin
+
+        wire writing = enq_fire && (i == decode_if.wid); 
+        wire reading = deq_fire && (i == ibuffer_if.wid);
+
+        wire going_empty = empty_r[i] || (alm_empty_r[i] && reading);
+
+        VX_elastic_buffer #(
+            .DATAW   (DATAW),
+            .SIZE    (`IBUF_SIZE),
+            .OUT_REG (1)
+        ) queue (
+            .clk      (clk),
+            .reset    (reset),
+            .valid_in (writing && !going_empty),
+            .data_in  (q_data_in),            
+            .ready_out(reading),
+            .data_out (q_data_prev[i]),            
+            `UNUSED_PIN (ready_in),
+            `UNUSED_PIN (valid_out)
+        );
+
+        always @(posedge clk) begin
+            if (reset) begin            
+                used_r[i]      <= 0;
+                full_r[i]      <= 0; 
+                empty_r[i]     <= 1; 
+                alm_empty_r[i] <= 1;
+            end else begin  
+                if (writing) begin
+                    if (!reading) begin
+                        empty_r[i] <= 0;
+                        if (used_r[i] == 1)
+                            alm_empty_r[i] <= 0;
+                        if (used_r[i] == ADDRW'(`IBUF_SIZE))
+                            full_r[i] <= 1;
+                    end
+                end else if (reading) begin
+                    full_r[i] <= 0; 
+                    if (used_r[i] == ADDRW'(1))
+                        empty_r[i] <= 1;
+                    if (used_r[i] == ADDRW'(2))
+                        alm_empty_r[i] <= 1;
+                end
+                used_r[i] <= used_r[i] + ADDRW'($signed(2'(writing) - 2'(reading)));
+            end 
+
+            if (writing && going_empty) begin                                                       
+                q_data_out[i] <= q_data_in;
+            end else if (reading) begin
+                q_data_out[i] <= q_data_prev[i];
+            end                  
+        end
+        
+        assign q_full[i]      = full_r[i];
+        assign q_empty[i]     = empty_r[i];
+        assign q_alm_empty[i] = alm_empty_r[i];
+    end
+
+    ///////////////////////////////////////////////////////////////////////////
+
+    reg [`NUM_WARPS-1:0] valid_table, valid_table_n;
+    reg [`NW_BITS-1:0] deq_wid, deq_wid_n;
+    reg [`NW_BITS-1:0] deq_wid_rr, deq_wid_rr_n;
+    reg deq_valid, deq_valid_n;
+    reg [DATAW-1:0] deq_instr, deq_instr_n;
+    reg [NWARPSW-1:0] num_warps;
+
+    `UNUSED_VAR (deq_instr)
+
+    // calculate valid table
+    always @(*) begin
+        valid_table_n = valid_table;        
+        if (deq_fire) begin
+            valid_table_n[deq_wid] = !q_alm_empty[deq_wid];
+        end
+        if (enq_fire) begin
+            valid_table_n[decode_if.wid] = 1;
+        end
+    end
+
+    // round-robin warp scheduling
+    VX_rr_arbiter #(
+        .NUM_REQS (`NUM_WARPS)
+    ) rr_arbiter (
+        .clk      (clk),
+        .reset    (reset),          
+        .requests (valid_table_n), 
+        .grant_index (deq_wid_rr_n),
+        `UNUSED_PIN (grant_valid),
+        `UNUSED_PIN (grant_onehot),
+        `UNUSED_PIN (enable)
+    );
+
+    // schedule the next instruction to issue
+    always @(*) begin
+        if (num_warps > 1) begin
+            deq_valid_n = 1;
+            deq_wid_n   = deq_wid_rr;          
+            deq_instr_n = q_data_out[deq_wid_rr];
+        end else if (1 == num_warps && !(deq_fire && q_alm_empty[deq_wid])) begin
+            deq_valid_n = 1;
+            deq_wid_n   = deq_wid;
+            deq_instr_n = deq_fire ? q_data_prev[deq_wid] : q_data_out[deq_wid];
+        end else begin
+            deq_valid_n = enq_fire;
+            deq_wid_n   = decode_if.wid;
+            deq_instr_n = q_data_in;
+        end
+    end
+
+    wire warp_added   = enq_fire && q_empty[decode_if.wid];
+    wire warp_removed = deq_fire && ~(enq_fire && decode_if.wid == deq_wid) && q_alm_empty[deq_wid];
+    
+    always @(posedge clk) begin
+        if (reset)  begin            
+            valid_table <= 0;
+            deq_valid   <= 0;  
+            num_warps   <= 0;
+        end else begin
+            valid_table <= valid_table_n;            
+            deq_valid   <= deq_valid_n;
+            
+
+            if (warp_added && !warp_removed) begin
+                num_warps <= num_warps + NWARPSW'(1);
+            end else if (warp_removed && !warp_added) begin
+                num_warps <= num_warps - NWARPSW'(1);                
+            end
+        end
+            
+        deq_wid    <= deq_wid_n;
+        deq_wid_rr <= deq_wid_rr_n;
+        deq_instr  <= deq_instr_n;
+    end
+    
+    assign decode_if.ready = ~q_full[decode_if.wid];
+    
+    assign q_data_in = {decode_if.uuid,
+                        decode_if.tmask, 
+                        decode_if.PC, 
+                        decode_if.ex_type, 
+                        decode_if.op_type, 
+                        decode_if.op_mod, 
+                        decode_if.wb,
+                        decode_if.use_PC,
+                        decode_if.use_imm,
+                        decode_if.imm,
+                        decode_if.rd, 
+                        decode_if.rs1, 
+                        decode_if.rs2, 
+                        decode_if.rs3};
+
+    assign ibuffer_if.valid = deq_valid;
+    assign ibuffer_if.wid   = deq_wid;
+    assign {ibuffer_if.uuid,
+            ibuffer_if.tmask, 
+            ibuffer_if.PC, 
+            ibuffer_if.ex_type, 
+            ibuffer_if.op_type, 
+            ibuffer_if.op_mod, 
+            ibuffer_if.wb,
+            ibuffer_if.use_PC,
+            ibuffer_if.use_imm,
+            ibuffer_if.imm,
+            ibuffer_if.rd, 
+            ibuffer_if.rs1, 
+            ibuffer_if.rs2, 
+            ibuffer_if.rs3} = deq_instr;
+
+    // scoreboard forwarding
+    assign ibuffer_if.wid_n = deq_wid_n;
+    assign ibuffer_if.rd_n  = deq_instr_n[3*`NR_BITS +: `NR_BITS];
+    assign ibuffer_if.rs1_n = deq_instr_n[2*`NR_BITS +: `NR_BITS];    
+    assign ibuffer_if.rs2_n = deq_instr_n[1*`NR_BITS +: `NR_BITS];
+    assign ibuffer_if.rs3_n = deq_instr_n[0*`NR_BITS +: `NR_BITS];
+
+endmodule

hw/rtl/VX_icache_stage.sv

@@ -0,0 +1,102 @@
+`include "VX_define.vh"
+
+module VX_icache_stage #(
+    parameter CORE_ID = 0
+) (
+    `SCOPE_IO_VX_icache_stage
+
+    input  wire             clk,
+    input  wire             reset,
+    
+    // Icache interface
+    VX_icache_req_if.master icache_req_if,
+    VX_icache_rsp_if.slave  icache_rsp_if,
+    
+    // request
+    VX_ifetch_req_if.slave  ifetch_req_if,
+
+    // reponse
+    VX_ifetch_rsp_if.master ifetch_rsp_if
+);
+
+    `UNUSED_PARAM (CORE_ID)
+    `UNUSED_VAR (reset)
+
+    localparam OUT_REG = 0;
+
+    wire [`NW_BITS-1:0] req_tag, rsp_tag;
+
+    wire icache_req_fire = icache_req_if.valid && icache_req_if.ready;
+    
+    assign req_tag = ifetch_req_if.wid;
+    assign rsp_tag = icache_rsp_if.tag[`NW_BITS-1:0];
+
+    wire [`UUID_BITS-1:0] rsp_uuid;
+    wire [31:0] rsp_PC;
+    wire [`NUM_THREADS-1:0] rsp_tmask;
+
+    VX_dp_ram #(
+        .DATAW  (32 + `NUM_THREADS + `UUID_BITS),
+        .SIZE   (`NUM_WARPS),
+        .LUTRAM (1)
+    ) req_metadata (
+        .clk   (clk),        
+        .wren  (icache_req_fire),
+        .waddr (req_tag),
+        .wdata ({ifetch_req_if.PC, ifetch_req_if.tmask, ifetch_req_if.uuid}),
+        .raddr (rsp_tag),
+        .rdata ({rsp_PC, rsp_tmask, rsp_uuid})
+    );
+
+    `RUNTIME_ASSERT((!ifetch_req_if.valid || ifetch_req_if.PC >= `STARTUP_ADDR), 
+        ("%t: *** invalid PC=%0h, wid=%0d, tmask=%b (#%0d)", $time, ifetch_req_if.PC, ifetch_req_if.wid, ifetch_req_if.tmask, ifetch_req_if.uuid))
+
+    // Icache Request
+    assign icache_req_if.valid = ifetch_req_if.valid;
+    assign icache_req_if.addr  = ifetch_req_if.PC[31:2];
+    assign icache_req_if.tag   = {ifetch_req_if.uuid, req_tag};
+
+    // Can accept new request?
+    assign ifetch_req_if.ready = icache_req_if.ready;
+
+    wire [`NW_BITS-1:0] rsp_wid = rsp_tag;
+
+    wire stall_out = ~ifetch_rsp_if.ready && (0 == OUT_REG && ifetch_rsp_if.valid);
+
+    VX_pipe_register #(
+        .DATAW  (1 + `NW_BITS + `NUM_THREADS + 32 + 32 + `UUID_BITS),
+        .RESETW (1),
+        .DEPTH  (OUT_REG)
+    ) pipe_reg (
+        .clk      (clk),
+        .reset    (reset),
+        .enable   (!stall_out),
+        .data_in  ({icache_rsp_if.valid, rsp_wid,           rsp_tmask,           rsp_PC,           icache_rsp_if.data, rsp_uuid}),
+        .data_out ({ifetch_rsp_if.valid, ifetch_rsp_if.wid, ifetch_rsp_if.tmask, ifetch_rsp_if.PC, ifetch_rsp_if.data, ifetch_rsp_if.uuid})
+    );     
+    
+    // Can accept new response?
+    assign icache_rsp_if.ready = ~stall_out;
+
+    `SCOPE_ASSIGN (icache_req_fire, icache_req_fire);
+    `SCOPE_ASSIGN (icache_req_uuid, ifetch_req_if.uuid);
+    `SCOPE_ASSIGN (icache_req_addr, {icache_req_if.addr, 2'b0});    
+    `SCOPE_ASSIGN (icache_req_tag,  req_tag);
+
+    `SCOPE_ASSIGN (icache_rsp_fire, icache_rsp_if.valid && icache_rsp_if.ready);
+    `SCOPE_ASSIGN (icache_rsp_uuid, rsp_uuid);
+    `SCOPE_ASSIGN (icache_rsp_data, icache_rsp_if.data);
+    `SCOPE_ASSIGN (icache_rsp_tag,  rsp_tag);
+
+`ifdef DBG_TRACE_CORE_ICACHE
+    always @(posedge clk) begin
+        if (icache_req_fire) begin
+            dpi_trace("%d: I$%0d req: wid=%0d, PC=%0h (#%0d)\n", $time, CORE_ID, ifetch_req_if.wid, ifetch_req_if.PC, ifetch_req_if.uuid);
+        end
+        if (ifetch_rsp_if.valid && ifetch_rsp_if.ready) begin
+            dpi_trace("%d: I$%0d rsp: wid=%0d, PC=%0h, data=%0h (#%0d)\n", $time, CORE_ID, ifetch_rsp_if.wid, ifetch_rsp_if.PC, ifetch_rsp_if.data, ifetch_rsp_if.uuid);
+        end
+    end
+`endif
+
+endmodule

hw/rtl/VX_ipdom_stack.sv

@@ -0,0 +1,68 @@
+`include "VX_platform.vh"
+
+module VX_ipdom_stack #(
+    parameter WIDTH = 1,
+    parameter DEPTH = 1
+) (
+    input  wire               clk,
+    input  wire               reset,
+    input  wire               pair,
+    input  wire [WIDTH - 1:0] q1,
+    input  wire [WIDTH - 1:0] q2,
+    output wire [WIDTH - 1:0] d,
+    input  wire               push,
+    input  wire               pop,
+    output wire               index,
+    output wire               empty,
+    output wire               full
+);
+    localparam ADDRW = $clog2(DEPTH);
+
+    reg is_part [DEPTH-1:0];
+    
+    reg [ADDRW-1:0] rd_ptr, wr_ptr;
+
+    wire [WIDTH-1:0] d1, d2;
+
+    always @(posedge clk) begin
+        if (reset) begin   
+            rd_ptr <= 0;
+            wr_ptr <= 0;
+        end else begin
+            if (push) begin
+                rd_ptr <= wr_ptr;
+                wr_ptr <= wr_ptr + ADDRW'(1);
+            end else if (pop) begin            
+                wr_ptr <= wr_ptr - ADDRW'(is_part[rd_ptr]);
+                rd_ptr <= rd_ptr - ADDRW'(is_part[rd_ptr]);
+            end
+        end
+    end    
+
+    VX_dp_ram #(
+        .DATAW  (WIDTH * 2),
+        .SIZE   (DEPTH),
+        .LUTRAM (1)
+    ) store (
+        .clk   (clk),
+        .wren  (push),
+        .waddr (wr_ptr),
+        .wdata ({q2, q1}),
+        .raddr (rd_ptr),
+        .rdata ({d2, d1})
+    );
+    
+    always @(posedge clk) begin
+        if (push) begin
+            is_part[wr_ptr] <= ~pair;   
+        end else if (pop) begin            
+            is_part[rd_ptr] <= 1;
+        end
+    end
+
+    assign index = is_part[rd_ptr];
+    assign d     = index ? d1 : d2;
+    assign empty = (ADDRW'(0) == wr_ptr);
+    assign full  = (ADDRW'(DEPTH-1) == wr_ptr);
+
+endmodule

hw/rtl/VX_issue.sv

@@ -0,0 +1,256 @@
+`include "VX_define.vh"
+
+module VX_issue #(
+    parameter CORE_ID = 0
+) (
+    `SCOPE_IO_VX_issue
+
+    input wire      clk,
+    input wire      reset,
+
+`ifdef PERF_ENABLE
+    VX_perf_pipeline_if.issue perf_issue_if,
+`endif
+
+    VX_decode_if.slave      decode_if,
+    VX_writeback_if.slave   writeback_if,   
+    
+    VX_alu_req_if.master    alu_req_if,
+    VX_lsu_req_if.master    lsu_req_if,    
+    VX_csr_req_if.master    csr_req_if,
+`ifdef EXT_F_ENABLE
+    VX_fpu_req_if.master    fpu_req_if,    
+`endif
+    VX_gpu_req_if.master    gpu_req_if
+);
+    VX_ibuffer_if   ibuffer_if();    
+    VX_gpr_req_if   gpr_req_if();
+    VX_gpr_rsp_if   gpr_rsp_if();
+    VX_writeback_if sboard_wb_if();
+    VX_ibuffer_if   scoreboard_if();
+    VX_ibuffer_if   dispatch_if();
+
+    // GPR request interface
+    assign gpr_req_if.wid       = ibuffer_if.wid;
+    assign gpr_req_if.rs1       = ibuffer_if.rs1;
+    assign gpr_req_if.rs2       = ibuffer_if.rs2;
+    assign gpr_req_if.rs3       = ibuffer_if.rs3;
+
+    // scoreboard writeback interface
+    assign sboard_wb_if.valid   = writeback_if.valid;
+    assign sboard_wb_if.uuid    = writeback_if.uuid;
+    assign sboard_wb_if.wid     = writeback_if.wid;
+    assign sboard_wb_if.PC      = writeback_if.PC;
+    assign sboard_wb_if.rd      = writeback_if.rd;
+    assign sboard_wb_if.eop     = writeback_if.eop;
+        
+    // scoreboard interface
+    assign scoreboard_if.valid  = ibuffer_if.valid && dispatch_if.ready;
+    assign scoreboard_if.uuid   = ibuffer_if.uuid;
+    assign scoreboard_if.wid    = ibuffer_if.wid;
+    assign scoreboard_if.PC     = ibuffer_if.PC;   
+    assign scoreboard_if.wb     = ibuffer_if.wb;      
+    assign scoreboard_if.rd     = ibuffer_if.rd;
+    assign scoreboard_if.rd_n   = ibuffer_if.rd_n;        
+    assign scoreboard_if.rs1_n  = ibuffer_if.rs1_n;        
+    assign scoreboard_if.rs2_n  = ibuffer_if.rs2_n;        
+    assign scoreboard_if.rs3_n  = ibuffer_if.rs3_n;        
+    assign scoreboard_if.wid_n  = ibuffer_if.wid_n;
+    
+    // dispatch interface
+    assign dispatch_if.valid    = ibuffer_if.valid && scoreboard_if.ready;
+    assign dispatch_if.uuid     = ibuffer_if.uuid;
+    assign dispatch_if.wid      = ibuffer_if.wid;
+    assign dispatch_if.tmask    = ibuffer_if.tmask;
+    assign dispatch_if.PC       = ibuffer_if.PC;
+    assign dispatch_if.ex_type  = ibuffer_if.ex_type;    
+    assign dispatch_if.op_type  = ibuffer_if.op_type; 
+    assign dispatch_if.op_mod   = ibuffer_if.op_mod;    
+    assign dispatch_if.wb       = ibuffer_if.wb;
+    assign dispatch_if.rd       = ibuffer_if.rd;
+    assign dispatch_if.rs1      = ibuffer_if.rs1;
+    assign dispatch_if.imm      = ibuffer_if.imm;        
+    assign dispatch_if.use_PC   = ibuffer_if.use_PC;
+    assign dispatch_if.use_imm  = ibuffer_if.use_imm;
+
+    // issue the instruction
+    assign ibuffer_if.ready = scoreboard_if.ready && dispatch_if.ready;
+
+    `RESET_RELAY (ibuf_reset);
+    `RESET_RELAY (scoreboard_reset);
+    `RESET_RELAY (gpr_reset);
+    `RESET_RELAY (dispatch_reset);
+
+    VX_ibuffer #(
+        .CORE_ID(CORE_ID)
+    ) ibuffer (
+        .clk        (clk),
+        .reset      (ibuf_reset), 
+        .decode_if  (decode_if),
+        .ibuffer_if (ibuffer_if) 
+    );
+
+    VX_scoreboard #(
+        .CORE_ID(CORE_ID)
+    ) scoreboard (
+        .clk        (clk),
+        .reset      (scoreboard_reset),         
+        .writeback_if(sboard_wb_if),
+        .ibuffer_if (scoreboard_if)
+    );
+
+    VX_gpr_stage #(
+        .CORE_ID(CORE_ID)
+    ) gpr_stage (
+        .clk          (clk),      
+        .reset        (gpr_reset),          
+        .writeback_if (writeback_if),
+        .gpr_req_if   (gpr_req_if),
+        .gpr_rsp_if   (gpr_rsp_if)
+    );
+
+    VX_dispatch dispatch (
+        .clk        (clk),      
+        .reset      (dispatch_reset),
+        .ibuffer_if (dispatch_if),
+        .gpr_rsp_if (gpr_rsp_if),
+        .alu_req_if (alu_req_if),
+        .lsu_req_if (lsu_req_if),        
+        .csr_req_if (csr_req_if),
+    `ifdef EXT_F_ENABLE
+        .fpu_req_if (fpu_req_if),
+    `endif
+        .gpu_req_if (gpu_req_if)
+    );
+
+    `SCOPE_ASSIGN (issue_fire,        ibuffer_if.valid && ibuffer_if.ready);
+    `SCOPE_ASSIGN (issue_uuid,        ibuffer_if.uuid);
+    `SCOPE_ASSIGN (issue_tmask,       ibuffer_if.tmask);
+    `SCOPE_ASSIGN (issue_ex_type,     ibuffer_if.ex_type);
+    `SCOPE_ASSIGN (issue_op_type,     ibuffer_if.op_type);
+    `SCOPE_ASSIGN (issue_op_mod,      ibuffer_if.op_mod);
+    `SCOPE_ASSIGN (issue_wb,          ibuffer_if.wb);
+    `SCOPE_ASSIGN (issue_rd,          ibuffer_if.rd);
+    `SCOPE_ASSIGN (issue_rs1,         ibuffer_if.rs1);
+    `SCOPE_ASSIGN (issue_rs2,         ibuffer_if.rs2);
+    `SCOPE_ASSIGN (issue_rs3,         ibuffer_if.rs3);
+    `SCOPE_ASSIGN (issue_imm,         ibuffer_if.imm);
+    `SCOPE_ASSIGN (issue_use_pc,      ibuffer_if.use_PC);
+    `SCOPE_ASSIGN (issue_use_imm,     ibuffer_if.use_imm);
+    `SCOPE_ASSIGN (scoreboard_delay,  !scoreboard_if.ready); 
+    `SCOPE_ASSIGN (dispatch_delay,    !dispatch_if.ready);    
+    `SCOPE_ASSIGN (gpr_rs1,           gpr_rsp_if.rs1_data);
+    `SCOPE_ASSIGN (gpr_rs2,           gpr_rsp_if.rs2_data);
+    `SCOPE_ASSIGN (gpr_rs3,           gpr_rsp_if.rs3_data);
+    `SCOPE_ASSIGN (writeback_valid,   writeback_if.valid);
+    `SCOPE_ASSIGN (writeback_uuid,    writeback_if.uuid);
+    `SCOPE_ASSIGN (writeback_tmask,   writeback_if.tmask);
+    `SCOPE_ASSIGN (writeback_rd,      writeback_if.rd);
+    `SCOPE_ASSIGN (writeback_data,    writeback_if.data);
+    `SCOPE_ASSIGN (writeback_eop,     writeback_if.eop);
+
+`ifdef PERF_ENABLE
+    reg [`PERF_CTR_BITS-1:0] perf_ibf_stalls;
+    reg [`PERF_CTR_BITS-1:0] perf_scb_stalls;
+    reg [`PERF_CTR_BITS-1:0] perf_alu_stalls;
+    reg [`PERF_CTR_BITS-1:0] perf_lsu_stalls;
+    reg [`PERF_CTR_BITS-1:0] perf_csr_stalls;
+    reg [`PERF_CTR_BITS-1:0] perf_gpu_stalls;
+`ifdef EXT_F_ENABLE
+    reg [`PERF_CTR_BITS-1:0] perf_fpu_stalls;
+`endif
+
+    always @(posedge clk) begin
+        if (reset) begin
+            perf_ibf_stalls <= 0;
+            perf_scb_stalls <= 0;
+            perf_alu_stalls <= 0;
+            perf_lsu_stalls <= 0;
+            perf_csr_stalls <= 0;
+            perf_gpu_stalls <= 0;
+        `ifdef EXT_F_ENABLE
+            perf_fpu_stalls <= 0;
+        `endif
+        end else begin
+            if (decode_if.valid & ~decode_if.ready) begin
+                perf_ibf_stalls <= perf_ibf_stalls  + `PERF_CTR_BITS'd1;
+            end
+            if (scoreboard_if.valid & ~scoreboard_if.ready) begin 
+                perf_scb_stalls <= perf_scb_stalls  + `PERF_CTR_BITS'd1;
+            end
+            if (dispatch_if.valid & ~dispatch_if.ready) begin
+                case (dispatch_if.ex_type)
+                `EX_ALU: perf_alu_stalls <= perf_alu_stalls + `PERF_CTR_BITS'd1;
+            `ifdef EXT_F_ENABLE
+                `EX_FPU: perf_fpu_stalls <= perf_fpu_stalls + `PERF_CTR_BITS'd1;
+            `endif
+                `EX_LSU: perf_lsu_stalls <= perf_lsu_stalls + `PERF_CTR_BITS'd1;
+                `EX_CSR: perf_csr_stalls <= perf_csr_stalls + `PERF_CTR_BITS'd1;
+                //`EX_GPU:
+                default: perf_gpu_stalls <= perf_gpu_stalls + `PERF_CTR_BITS'd1;
+                endcase
+            end
+        end
+    end
+    
+    assign perf_issue_if.ibf_stalls = perf_ibf_stalls;
+    assign perf_issue_if.scb_stalls = perf_scb_stalls; 
+    assign perf_issue_if.alu_stalls = perf_alu_stalls;
+    assign perf_issue_if.lsu_stalls = perf_lsu_stalls;
+    assign perf_issue_if.csr_stalls = perf_csr_stalls;
+    assign perf_issue_if.gpu_stalls = perf_gpu_stalls;
+`ifdef EXT_F_ENABLE
+    assign perf_issue_if.fpu_stalls = perf_fpu_stalls;
+`endif
+`endif
+
+`ifdef DBG_TRACE_CORE_PIPELINE
+    always @(posedge clk) begin
+        if (alu_req_if.valid && alu_req_if.ready) begin
+            dpi_trace("%d: core%0d-issue: wid=%0d, PC=%0h, ex=ALU, tmask=%b, rd=%0d, rs1_data=", 
+                $time, CORE_ID, alu_req_if.wid, alu_req_if.PC, alu_req_if.tmask, alu_req_if.rd);
+            `TRACE_ARRAY1D(alu_req_if.rs1_data, `NUM_THREADS);
+            dpi_trace(", rs2_data=");
+            `TRACE_ARRAY1D(alu_req_if.rs2_data, `NUM_THREADS);
+            dpi_trace(" (#%0d)\n", alu_req_if.uuid);
+        end
+        if (lsu_req_if.valid && lsu_req_if.ready) begin
+            dpi_trace("%d: core%0d-issue: wid=%0d, PC=%0h, ex=LSU, tmask=%b, rd=%0d, offset=%0h, addr=", 
+                $time, CORE_ID, lsu_req_if.wid, lsu_req_if.PC, lsu_req_if.tmask, lsu_req_if.rd, lsu_req_if.offset); 
+            `TRACE_ARRAY1D(lsu_req_if.base_addr, `NUM_THREADS);
+            dpi_trace(", data=");
+            `TRACE_ARRAY1D(lsu_req_if.store_data, `NUM_THREADS);
+            dpi_trace(" (#%0d)\n", lsu_req_if.uuid);
+        end
+        if (csr_req_if.valid && csr_req_if.ready) begin
+            dpi_trace("%d: core%0d-issue: wid=%0d, PC=%0h, ex=CSR, tmask=%b, rd=%0d, addr=%0h, rs1_data=", 
+                $time, CORE_ID, csr_req_if.wid, csr_req_if.PC, csr_req_if.tmask, csr_req_if.rd, csr_req_if.addr);   
+            `TRACE_ARRAY1D(csr_req_if.rs1_data, `NUM_THREADS);
+            dpi_trace(" (#%0d)\n", csr_req_if.uuid);
+        end
+    `ifdef EXT_F_ENABLE
+        if (fpu_req_if.valid && fpu_req_if.ready) begin
+            dpi_trace("%d: core%0d-issue: wid=%0d, PC=%0h, ex=FPU, tmask=%b, rd=%0d, rs1_data=", 
+                $time, CORE_ID, fpu_req_if.wid, fpu_req_if.PC, fpu_req_if.tmask, fpu_req_if.rd);   
+            `TRACE_ARRAY1D(fpu_req_if.rs1_data, `NUM_THREADS);
+            dpi_trace(", rs2_data=");
+            `TRACE_ARRAY1D(fpu_req_if.rs2_data, `NUM_THREADS);
+            dpi_trace(", rs3_data=");
+            `TRACE_ARRAY1D(fpu_req_if.rs3_data, `NUM_THREADS);
+            dpi_trace(" (#%0d)\n", fpu_req_if.uuid);
+        end
+    `endif
+        if (gpu_req_if.valid && gpu_req_if.ready) begin
+            dpi_trace("%d: core%0d-issue: wid=%0d, PC=%0h, ex=GPU, tmask=%b, rd=%0d, rs1_data=", 
+                $time, CORE_ID, gpu_req_if.wid, gpu_req_if.PC, gpu_req_if.tmask, gpu_req_if.rd);   
+            `TRACE_ARRAY1D(gpu_req_if.rs1_data, `NUM_THREADS);
+            dpi_trace(", rs2_data=");
+            `TRACE_ARRAY1D(gpu_req_if.rs2_data, `NUM_THREADS);
+            dpi_trace(", rs3_data=");
+            `TRACE_ARRAY1D(gpu_req_if.rs3_data, `NUM_THREADS);
+            dpi_trace(" (#%0d)\n", gpu_req_if.uuid);   
+        end
+    end
+`endif
+
+endmodule

hw/rtl/VX_lsu_unit.sv

@@ -0,0 +1,372 @@
+`include "VX_define.vh"
+
+module VX_lsu_unit #(
+    parameter CORE_ID = 0
+) (    
+    `SCOPE_IO_VX_lsu_unit
+
+    input wire              clk,
+    input wire              reset,
+
+   // Dcache interface
+    VX_dcache_req_if.master dcache_req_if,
+    VX_dcache_rsp_if.slave  dcache_rsp_if,
+
+    // inputs
+    VX_lsu_req_if.slave     lsu_req_if,
+
+    // outputs
+    VX_commit_if.master     ld_commit_if,
+    VX_commit_if.master     st_commit_if
+);
+    localparam MEM_ASHIFT = `CLOG2(`MEM_BLOCK_SIZE);    
+    localparam MEM_ADDRW  = 32 - MEM_ASHIFT;
+    localparam REQ_ASHIFT = `CLOG2(`DCACHE_WORD_SIZE);
+
+    `STATIC_ASSERT(0 == (`IO_BASE_ADDR % MEM_ASHIFT), ("invalid parameter"))
+    `STATIC_ASSERT(0 == (`SMEM_BASE_ADDR % MEM_ASHIFT), ("invalid parameter"))
+    `STATIC_ASSERT(`SMEM_SIZE == `MEM_BLOCK_SIZE * (`SMEM_SIZE / `MEM_BLOCK_SIZE), ("invalid parameter"))
+    
+    wire                          req_valid;
+    wire [`UUID_BITS-1:0]         req_uuid;
+    wire [`NUM_THREADS-1:0]       req_tmask;
+    wire [`NUM_THREADS-1:0][31:0] req_addr;       
+    wire [`INST_LSU_BITS-1:0]     req_type;
+    wire [`NUM_THREADS-1:0][31:0] req_data;   
+    wire [`NR_BITS-1:0]           req_rd;
+    wire                          req_wb; // NOTE(hansung): 0:load, 1:store
+    wire [`NW_BITS-1:0]           req_wid;
+    wire [31:0]                   req_pc;
+    wire                          req_is_dup;
+    wire                          req_is_prefetch;
+    
+    wire mbuf_empty;
+
+    wire [`NUM_THREADS-1:0][`CACHE_ADDR_TYPE_BITS-1:0] lsu_addr_type, req_addr_type;
+
+    // full address calculation
+    wire [`NUM_THREADS-1:0][31:0] full_addr;    
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin
+        assign full_addr[i] = lsu_req_if.base_addr[i] + lsu_req_if.offset;
+    end
+
+    // detect duplicate addresses
+    wire [`NUM_THREADS-2:0] addr_matches;
+    for (genvar i = 0; i < (`NUM_THREADS-1); i++) begin
+        assign addr_matches[i] = (lsu_req_if.base_addr[i+1] == lsu_req_if.base_addr[0]) || ~lsu_req_if.tmask[i+1];
+    end    
+
+    wire lsu_is_dup = lsu_req_if.tmask[0] && (& addr_matches);
+
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin
+        // is non-cacheable address
+        wire is_addr_nc = (full_addr[i][MEM_ASHIFT +: MEM_ADDRW] >= MEM_ADDRW'(`IO_BASE_ADDR >> MEM_ASHIFT));
+        if (`SM_ENABLE) begin
+            // is shared memory address
+            wire is_addr_sm = (full_addr[i][MEM_ASHIFT +: MEM_ADDRW] >= MEM_ADDRW'((`SMEM_BASE_ADDR - `SMEM_SIZE) >> MEM_ASHIFT))
+                            & (full_addr[i][MEM_ASHIFT +: MEM_ADDRW] < MEM_ADDRW'(`SMEM_BASE_ADDR >> MEM_ASHIFT));            
+            assign lsu_addr_type[i] = {is_addr_nc, is_addr_sm};
+        end else begin
+            assign lsu_addr_type[i] = is_addr_nc;
+        end
+    end
+
+    // fence stalls the pipeline until all pending requests are sent
+    wire fence_wait = lsu_req_if.is_fence && (req_valid || !mbuf_empty);
+    
+    wire ready_in;
+    wire stall_in = ~ready_in && req_valid; 
+    
+    wire lsu_valid = lsu_req_if.valid && ~fence_wait;
+
+    wire lsu_wb = lsu_req_if.wb | lsu_req_if.is_prefetch;
+
+    VX_pipe_register #(
+        .DATAW  (1 + 1 + 1 + `UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + (`NUM_THREADS * 32) + (`NUM_THREADS * `CACHE_ADDR_TYPE_BITS) + `INST_LSU_BITS + `NR_BITS + 1 + (`NUM_THREADS * 32)),
+        .RESETW (1)
+    ) req_pipe_reg (
+        .clk      (clk),
+        .reset    (reset),
+        .enable   (!stall_in),
+        .data_in  ({lsu_valid, lsu_is_dup, lsu_req_if.is_prefetch, lsu_req_if.uuid, lsu_req_if.wid, lsu_req_if.tmask, lsu_req_if.PC, full_addr, lsu_addr_type, lsu_req_if.op_type, lsu_req_if.rd, lsu_wb, lsu_req_if.store_data}),
+        .data_out ({req_valid, req_is_dup, req_is_prefetch,        req_uuid,        req_wid,        req_tmask,        req_pc,        req_addr,  req_addr_type, req_type,           req_rd,        req_wb, req_data})
+    );
+
+    // Can accept new request?
+    assign lsu_req_if.ready = ~stall_in && ~fence_wait;
+
+    wire [`UUID_BITS-1:0] rsp_uuid;
+    wire [`NW_BITS-1:0] rsp_wid;
+    wire [31:0] rsp_pc;
+    wire [`NR_BITS-1:0] rsp_rd;
+    wire rsp_wb;
+    wire [`INST_LSU_BITS-1:0] rsp_type;
+    wire rsp_is_dup;
+    wire rsp_is_prefetch;
+
+    reg [`LSUQ_SIZE-1:0][`NUM_THREADS-1:0] rsp_rem_mask;         
+    wire [`NUM_THREADS-1:0] rsp_rem_mask_n;
+    wire [`NUM_THREADS-1:0] rsp_tmask;
+
+    reg [`NUM_THREADS-1:0] req_sent_mask;
+    reg is_req_start;
+
+    wire [`LSUQ_ADDR_BITS-1:0] mbuf_waddr, mbuf_raddr;    
+    wire mbuf_full;
+
+    `UNUSED_VAR (rsp_type)
+    `UNUSED_VAR (rsp_is_prefetch)
+
+    wire [`NUM_THREADS-1:0][REQ_ASHIFT-1:0] req_offset, rsp_offset;
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin  
+        assign req_offset[i] = req_addr[i][1:0];
+    end
+
+    wire [`NUM_THREADS-1:0] dcache_req_fire = dcache_req_if.valid & dcache_req_if.ready;
+
+    wire dcache_rsp_fire = dcache_rsp_if.valid && dcache_rsp_if.ready;
+
+    wire [`NUM_THREADS-1:0] req_tmask_dup = req_tmask & {{(`NUM_THREADS-1){~req_is_dup}}, 1'b1};
+
+    wire mbuf_push = ~mbuf_full
+                  && (| ({`NUM_THREADS{req_valid}} & req_tmask_dup & dcache_req_if.ready))
+                  && is_req_start   // first submission only                  
+                  && req_wb;        // loads only
+
+    wire mbuf_pop = dcache_rsp_fire && (0 == rsp_rem_mask_n);
+    
+    assign mbuf_raddr = dcache_rsp_if.tag[`CACHE_ADDR_TYPE_BITS +: `LSUQ_ADDR_BITS];
+    `UNUSED_VAR (dcache_rsp_if.tag)    
+
+    // do not writeback from software prefetch
+    wire req_wb2 = req_wb && ~req_is_prefetch;
+
+    VX_index_buffer #(
+        .DATAW (`UUID_BITS + `NW_BITS + 32 + `NUM_THREADS + `NR_BITS + 1 + `INST_LSU_BITS + (`NUM_THREADS * REQ_ASHIFT) + 1 + 1),
+        .SIZE  (`LSUQ_SIZE)
+    ) req_metadata (
+        .clk          (clk),
+        .reset        (reset),
+        .write_addr   (mbuf_waddr),  
+        .acquire_slot (mbuf_push),       
+        .read_addr    (mbuf_raddr),
+        .write_data   ({req_uuid, req_wid, req_pc, req_tmask, req_rd, req_wb2, req_type, req_offset, req_is_dup, req_is_prefetch}),                    
+        .read_data    ({rsp_uuid, rsp_wid, rsp_pc, rsp_tmask, rsp_rd, rsp_wb,  rsp_type, rsp_offset, rsp_is_dup, rsp_is_prefetch}),
+        .release_addr (mbuf_raddr),
+        .release_slot (mbuf_pop),     
+        .full         (mbuf_full),
+        .empty        (mbuf_empty)
+    );
+
+    wire dcache_req_ready = &(dcache_req_if.ready | req_sent_mask | ~req_tmask_dup);
+
+    wire [`NUM_THREADS-1:0] req_sent_mask_n = req_sent_mask | dcache_req_fire;
+
+    always @(posedge clk) begin
+        if (reset) begin
+            req_sent_mask <= 0;
+            is_req_start  <= 1;
+        end else begin
+            if (dcache_req_ready) begin
+                req_sent_mask <= 0;
+                is_req_start  <= 1;
+            end else begin
+                req_sent_mask <= req_sent_mask_n;
+                is_req_start  <= (0 == req_sent_mask_n);
+            end
+        end
+    end
+
+    // need to hold the acquired tag index until the full request is submitted
+    reg [`LSUQ_ADDR_BITS-1:0] req_tag_hold;
+    wire [`LSUQ_ADDR_BITS-1:0] req_tag = is_req_start ? mbuf_waddr : req_tag_hold;
+    always @(posedge clk) begin
+        if (mbuf_push) begin            
+            req_tag_hold <= mbuf_waddr;
+        end
+    end    
+
+    assign rsp_rem_mask_n = rsp_rem_mask[mbuf_raddr] & ~dcache_rsp_if.tmask;
+
+    always @(posedge clk) begin
+        if (mbuf_push)  begin
+            rsp_rem_mask[mbuf_waddr] <= req_tmask_dup;
+        end    
+        if (dcache_rsp_fire) begin
+            rsp_rem_mask[mbuf_raddr] <= rsp_rem_mask_n;
+        end
+    end
+
+    // ensure all dependencies for the requests are resolved
+    wire req_dep_ready = (req_wb && ~(mbuf_full && is_req_start)) 
+                      || (~req_wb && st_commit_if.ready);
+
+    // DCache Request
+
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin
+
+        reg [3:0]  mem_req_byteen;    
+        reg [31:0] mem_req_data;
+
+        always @(*) begin
+            mem_req_byteen = {4{req_wb}};
+            case (`INST_LSU_WSIZE(req_type))
+                0: mem_req_byteen[req_offset[i]] = 1;
+                1: begin
+                    mem_req_byteen[req_offset[i]] = 1;
+                    mem_req_byteen[{req_offset[i][1], 1'b1}] = 1;
+                end
+                default : mem_req_byteen = {4{1'b1}};
+            endcase
+        end
+
+        always @(*) begin
+            mem_req_data = req_data[i];
+            case (req_offset[i])
+                1: mem_req_data[31:8]  = req_data[i][23:0];
+                2: mem_req_data[31:16] = req_data[i][15:0];
+                3: mem_req_data[31:24] = req_data[i][7:0];
+                default:;
+            endcase
+        end
+
+        assign dcache_req_if.valid[i]  = req_valid && req_dep_ready && req_tmask_dup[i] && !req_sent_mask[i];
+        assign dcache_req_if.rw[i]     = ~req_wb;
+        assign dcache_req_if.addr[i]   = req_addr[i][31:2];
+        assign dcache_req_if.byteen[i] = mem_req_byteen;
+        assign dcache_req_if.data[i]   = mem_req_data;
+        assign dcache_req_if.tag[i]    = {req_uuid, `LSU_TAG_ID_BITS'(req_tag), req_addr_type[i]};
+    end
+
+    assign ready_in = req_dep_ready && dcache_req_ready;
+
+    // send store commit
+
+    wire is_store_rsp = req_valid && ~req_wb && dcache_req_ready;
+
+    assign st_commit_if.valid = is_store_rsp;
+    assign st_commit_if.uuid  = req_uuid;
+    assign st_commit_if.wid   = req_wid;
+    assign st_commit_if.tmask = req_tmask;
+    assign st_commit_if.PC    = req_pc;
+    assign st_commit_if.rd    = 0;
+    assign st_commit_if.wb    = 0;
+    assign st_commit_if.eop   = 1'b1;
+    assign st_commit_if.data  = 0;
+
+    // load response formatting
+
+    reg [`NUM_THREADS-1:0][31:0] rsp_data;
+    wire [`NUM_THREADS-1:0] rsp_tmask_qual;
+
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin     
+        wire [31:0] rsp_data32 = (i == 0 || rsp_is_dup) ? dcache_rsp_if.data[0] : dcache_rsp_if.data[i];
+        wire [15:0] rsp_data16 = rsp_offset[i][1] ? rsp_data32[31:16] : rsp_data32[15:0];
+        wire [7:0]  rsp_data8  = rsp_offset[i][0] ? rsp_data16[15:8] : rsp_data16[7:0];
+
+        always @(*) begin
+            case (`INST_LSU_FMT(rsp_type))
+            `INST_FMT_B:  rsp_data[i] = 32'(signed'(rsp_data8));
+            `INST_FMT_H:  rsp_data[i] = 32'(signed'(rsp_data16));
+            `INST_FMT_BU: rsp_data[i] = 32'(unsigned'(rsp_data8));
+            `INST_FMT_HU: rsp_data[i] = 32'(unsigned'(rsp_data16));
+            default: rsp_data[i] = rsp_data32;
+            endcase
+        end        
+    end   
+
+    assign rsp_tmask_qual = rsp_is_dup ? rsp_tmask : dcache_rsp_if.tmask;
+
+    // send load commit
+
+    wire load_rsp_stall = ~ld_commit_if.ready && ld_commit_if.valid;
+    
+    VX_pipe_register #(
+        .DATAW  (1 + `UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `NR_BITS + 1 + (`NUM_THREADS * 32) + 1),
+        .RESETW (1)
+    ) rsp_pipe_reg (
+        .clk      (clk),
+        .reset    (reset),
+        .enable   (!load_rsp_stall),
+        .data_in  ({dcache_rsp_if.valid, rsp_uuid,          rsp_wid,          rsp_tmask_qual,     rsp_pc,          rsp_rd,          rsp_wb,          rsp_data,          mbuf_pop}),
+        .data_out ({ld_commit_if.valid,  ld_commit_if.uuid, ld_commit_if.wid, ld_commit_if.tmask, ld_commit_if.PC, ld_commit_if.rd, ld_commit_if.wb, ld_commit_if.data, ld_commit_if.eop})
+    );
+
+    // Can accept new cache response?
+    assign dcache_rsp_if.ready = ~load_rsp_stall;
+
+    // scope registration
+    `SCOPE_ASSIGN (dcache_req_fire,  dcache_req_fire);
+    `SCOPE_ASSIGN (dcache_req_uuid,  req_uuid);
+    `SCOPE_ASSIGN (dcache_req_addr,  req_addr);    
+    `SCOPE_ASSIGN (dcache_req_rw,    ~req_wb);
+    `SCOPE_ASSIGN (dcache_req_byteen,dcache_req_if.byteen);
+    `SCOPE_ASSIGN (dcache_req_data,  dcache_req_if.data);
+    `SCOPE_ASSIGN (dcache_req_tag,   req_tag);
+    `SCOPE_ASSIGN (dcache_rsp_fire,  dcache_rsp_if.tmask & {`NUM_THREADS{dcache_rsp_fire}});
+    `SCOPE_ASSIGN (dcache_rsp_uuid,  rsp_uuid);    
+    `SCOPE_ASSIGN (dcache_rsp_data,  dcache_rsp_if.data);
+    `SCOPE_ASSIGN (dcache_rsp_tag,   mbuf_raddr);
+
+`ifndef SYNTHESIS
+    reg [`LSUQ_SIZE-1:0][(`NW_BITS + 32 + `NR_BITS + `UUID_BITS + 64 + 1)-1:0] pending_reqs;
+    wire [63:0] delay_timeout = 10000 * (1 ** (`L2_ENABLE + `L3_ENABLE));
+
+    always @(posedge clk) begin
+        if (reset) begin
+            pending_reqs <= '0;
+        end begin
+            if (mbuf_push) begin            
+                pending_reqs[mbuf_waddr] <= {req_wid, req_pc, req_rd, req_uuid, $time, 1'b1};
+            end
+            if (mbuf_pop) begin
+                pending_reqs[mbuf_raddr] <= '0;
+            end
+        end
+
+        for (integer i = 0; i < `LSUQ_SIZE; ++i) begin
+            if (pending_reqs[i][0]) begin 
+                `ASSERT(($time - pending_reqs[i][1 +: 64]) < delay_timeout, 
+                    ("%t: *** D$%0d response timeout: remaining=%b, wid=%0d, PC=%0h, rd=%0d (#%0d)", 
+                        $time, CORE_ID, rsp_rem_mask[i], pending_reqs[i][1+64+`UUID_BITS+`NR_BITS+32 +: `NW_BITS], 
+                                                         pending_reqs[i][1+64+`UUID_BITS+`NR_BITS +: 32], 
+                                                         pending_reqs[i][1+64+`UUID_BITS +: `NR_BITS],
+                                                         pending_reqs[i][1+64 +: `UUID_BITS]));
+            end
+        end
+    end
+`endif
+    
+`ifdef DBG_TRACE_CORE_DCACHE
+    wire dcache_req_fire_any = (| dcache_req_fire);
+    always @(posedge clk) begin    
+        if (lsu_req_if.valid && fence_wait) begin
+            dpi_trace("%d: *** D$%0d fence wait\n", $time, CORE_ID);
+        end
+        if (dcache_req_fire_any) begin
+            if (dcache_req_if.rw[0]) begin
+                dpi_trace("%d: D$%0d Wr Req: wid=%0d, PC=%0h, tmask=%b, addr=", $time, CORE_ID, req_wid, req_pc, dcache_req_fire);
+                `TRACE_ARRAY1D(req_addr, `NUM_THREADS);
+                dpi_trace(", tag=%0h, byteen=%0h, type=", req_tag, dcache_req_if.byteen);
+                `TRACE_ARRAY1D(req_addr_type, `NUM_THREADS);
+                dpi_trace(", data=");
+                `TRACE_ARRAY1D(dcache_req_if.data, `NUM_THREADS);
+                dpi_trace(", (#%0d)\n", req_uuid);
+            end else begin
+                dpi_trace("%d: D$%0d Rd Req: prefetch=%b, wid=%0d, PC=%0h, tmask=%b, addr=", $time, CORE_ID, req_is_prefetch, req_wid, req_pc, dcache_req_fire);
+                `TRACE_ARRAY1D(req_addr, `NUM_THREADS);
+                dpi_trace(", tag=%0h, byteen=%0h, type=", req_tag, dcache_req_if.byteen);
+                `TRACE_ARRAY1D(req_addr_type, `NUM_THREADS);
+                dpi_trace(", rd=%0d, is_dup=%b (#%0d)\n", req_rd, req_is_dup, req_uuid);
+            end
+        end
+        if (dcache_rsp_fire) begin
+            dpi_trace("%d: D$%0d Rsp: prefetch=%b, wid=%0d, PC=%0h, tmask=%b, tag=%0h, rd=%0d, data=", 
+                $time, CORE_ID, rsp_is_prefetch, rsp_wid, rsp_pc, dcache_rsp_if.tmask, mbuf_raddr, rsp_rd);
+            `TRACE_ARRAY1D(dcache_rsp_if.data, `NUM_THREADS);
+            dpi_trace(", is_dup=%b (#%0d)\n", rsp_is_dup, rsp_uuid);
+        end
+    end
+`endif
+    
+endmodule

hw/rtl/VX_mem_arb.sv

@@ -0,0 +1,146 @@
+`include "VX_define.vh"
+
+module VX_mem_arb #(    
+    parameter NUM_REQS      = 1, 
+    parameter DATA_WIDTH    = 1,
+    parameter ADDR_WIDTH    = 1,
+    parameter TAG_IN_WIDTH  = 1,    
+    parameter TAG_SEL_IDX   = 0,
+    parameter BUFFERED_REQ  = 0,
+    parameter BUFFERED_RSP  = 0,
+    parameter TYPE          = "P",
+    
+    parameter DATA_SIZE     = (DATA_WIDTH / 8),
+    parameter LOG_NUM_REQS  = `CLOG2(NUM_REQS),
+    parameter TAG_OUT_WIDTH = TAG_IN_WIDTH + LOG_NUM_REQS
+) (
+    input wire clk,
+    input wire reset,
+
+    // input requests    
+    input wire [NUM_REQS-1:0]                   req_valid_in,    
+    input wire [NUM_REQS-1:0][TAG_IN_WIDTH-1:0] req_tag_in,  
+    input wire [NUM_REQS-1:0][ADDR_WIDTH-1:0]   req_addr_in,
+    input wire [NUM_REQS-1:0]                   req_rw_in,  
+    input wire [NUM_REQS-1:0][DATA_SIZE-1:0]    req_byteen_in,  
+    input wire [NUM_REQS-1:0][DATA_WIDTH-1:0]   req_data_in,  
+    output wire [NUM_REQS-1:0]                  req_ready_in,
+
+    // output request
+    output wire                                 req_valid_out,
+    output wire [TAG_OUT_WIDTH-1:0]             req_tag_out,   
+    output wire [ADDR_WIDTH-1:0]                req_addr_out, 
+    output wire                                 req_rw_out,  
+    output wire [DATA_SIZE-1:0]                 req_byteen_out,  
+    output wire [DATA_WIDTH-1:0]                req_data_out,    
+    input wire                                  req_ready_out,
+
+    // input response
+    input wire                                  rsp_valid_in,
+    input wire [TAG_OUT_WIDTH-1:0]              rsp_tag_in,
+    input wire [DATA_WIDTH-1:0]                 rsp_data_in,
+    output wire                                 rsp_ready_in,
+
+    // output responses
+    output wire [NUM_REQS-1:0]                  rsp_valid_out,
+    output wire [NUM_REQS-1:0][TAG_IN_WIDTH-1:0] rsp_tag_out,
+    output wire [NUM_REQS-1:0][DATA_WIDTH-1:0]  rsp_data_out,
+    input wire  [NUM_REQS-1:0]                  rsp_ready_out    
+);
+    localparam REQ_DATAW = TAG_OUT_WIDTH + ADDR_WIDTH + 1 + DATA_SIZE + DATA_WIDTH;
+    localparam RSP_DATAW = TAG_IN_WIDTH + DATA_WIDTH;
+
+    if (NUM_REQS > 1) begin
+
+        wire [NUM_REQS-1:0][REQ_DATAW-1:0] req_data_in_merged;
+
+        for (genvar i = 0; i < NUM_REQS; i++) begin
+            wire [TAG_OUT_WIDTH-1:0] req_tag_in_w;
+
+            VX_bits_insert #( 
+                .N   (TAG_IN_WIDTH),
+                .S   (LOG_NUM_REQS),
+                .POS (TAG_SEL_IDX)
+            ) bits_insert (
+                .data_in  (req_tag_in[i]),
+                .sel_in   (LOG_NUM_REQS'(i)),
+                .data_out (req_tag_in_w)
+            );
+
+            assign req_data_in_merged[i] = {req_tag_in_w, req_addr_in[i], req_rw_in[i], req_byteen_in[i], req_data_in[i]};
+        end
+
+        VX_stream_arbiter #(            
+            .NUM_REQS (NUM_REQS),
+            .DATAW    (REQ_DATAW),
+            .BUFFERED (BUFFERED_REQ),
+            .TYPE     (TYPE)
+        ) req_arb (
+            .clk       (clk),
+            .reset     (reset),
+            .valid_in  (req_valid_in),
+            .data_in   (req_data_in_merged),
+            .ready_in  (req_ready_in),
+            .valid_out (req_valid_out),
+            .data_out  ({req_tag_out, req_addr_out, req_rw_out, req_byteen_out, req_data_out}),
+            .ready_out (req_ready_out)
+        );
+
+        ///////////////////////////////////////////////////////////////////////
+
+        wire [NUM_REQS-1:0][RSP_DATAW-1:0] rsp_data_out_merged;
+
+        wire [LOG_NUM_REQS-1:0] rsp_sel = rsp_tag_in[TAG_SEL_IDX +: LOG_NUM_REQS];
+
+        wire [TAG_IN_WIDTH-1:0] rsp_tag_in_w;
+
+        VX_bits_remove #( 
+            .N   (TAG_OUT_WIDTH),
+            .S   (LOG_NUM_REQS),
+            .POS (TAG_SEL_IDX)
+        ) bits_remove (
+            .data_in  (rsp_tag_in),
+            .data_out (rsp_tag_in_w)
+        );
+
+        VX_stream_demux #(
+            .NUM_REQS (NUM_REQS),
+            .DATAW    (RSP_DATAW),
+            .BUFFERED (BUFFERED_RSP)
+        ) rsp_demux (
+            .clk       (clk),
+            .reset     (reset),
+            .sel_in    (rsp_sel),
+            .valid_in  (rsp_valid_in),
+            .data_in   ({rsp_tag_in_w, rsp_data_in}),
+            .ready_in  (rsp_ready_in),
+            .valid_out (rsp_valid_out),
+            .data_out  (rsp_data_out_merged),
+            .ready_out (rsp_ready_out)
+        );
+        
+        for (genvar i = 0; i < NUM_REQS; i++) begin
+            assign {rsp_tag_out[i], rsp_data_out[i]} = rsp_data_out_merged[i];
+        end        
+
+    end else begin
+
+        `UNUSED_VAR (clk)
+        `UNUSED_VAR (reset)
+
+        assign req_valid_out  = req_valid_in;
+        assign req_tag_out    = req_tag_in;
+        assign req_addr_out   = req_addr_in;
+        assign req_rw_out     = req_rw_in;
+        assign req_byteen_out = req_byteen_in;
+        assign req_data_out   = req_data_in;
+        assign req_ready_in   = req_ready_out;
+
+        assign rsp_valid_out  = rsp_valid_in;
+        assign rsp_tag_out    = rsp_tag_in;
+        assign rsp_data_out   = rsp_data_in;
+        assign rsp_ready_in   = rsp_ready_out;
+
+    end
+
+endmodule

hw/rtl/VX_mem_unit.sv

@@ -0,0 +1,420 @@
+`include "VX_define.vh"
+
+module VX_mem_unit # (
+    parameter CORE_ID = 0
+) (
+    `SCOPE_IO_VX_mem_unit
+
+    input wire              clk,
+    input wire              reset,
+    
+`ifdef PERF_ENABLE
+    VX_perf_memsys_if.master perf_memsys_if,
+`endif
+
+    // Core <-> Dcache    
+    VX_dcache_req_if.slave  dcache_req_if,
+    VX_dcache_rsp_if.master dcache_rsp_if,
+    
+    // Core <-> Icache    
+    VX_icache_req_if.slave  icache_req_if,  
+    VX_icache_rsp_if.master icache_rsp_if,
+
+    // Memory
+    VX_mem_req_if.master    mem_req_if,
+    VX_mem_rsp_if.slave     mem_rsp_if
+);
+    
+`ifdef PERF_ENABLE
+    VX_perf_cache_if perf_icache_if(), perf_dcache_if(), perf_smem_if();
+`endif
+
+    VX_mem_req_if #(
+        .DATA_WIDTH (`ICACHE_MEM_DATA_WIDTH),
+        .ADDR_WIDTH (`ICACHE_MEM_ADDR_WIDTH),
+        .TAG_WIDTH  (`ICACHE_MEM_TAG_WIDTH)
+    ) icache_mem_req_if();
+
+    VX_mem_rsp_if #(
+        .DATA_WIDTH (`ICACHE_MEM_DATA_WIDTH),
+        .TAG_WIDTH  (`ICACHE_MEM_TAG_WIDTH)
+    ) icache_mem_rsp_if();
+
+    VX_mem_req_if #(
+        .DATA_WIDTH (`DCACHE_MEM_DATA_WIDTH),
+        .ADDR_WIDTH (`DCACHE_MEM_ADDR_WIDTH),
+        .TAG_WIDTH  (`DCACHE_MEM_TAG_WIDTH)
+    ) dcache_mem_req_if();
+
+    VX_mem_rsp_if #(
+        .DATA_WIDTH (`DCACHE_MEM_DATA_WIDTH),
+        .TAG_WIDTH  (`DCACHE_MEM_TAG_WIDTH)
+    ) dcache_mem_rsp_if();
+
+    VX_dcache_req_if #(
+        .NUM_REQS  (`DCACHE_NUM_REQS), 
+        .WORD_SIZE (`DCACHE_WORD_SIZE), 
+        .TAG_WIDTH (`DCACHE_CORE_TAG_WIDTH-`SM_ENABLE)
+    ) dcache_req_tmp_if();
+
+    VX_dcache_rsp_if #(
+        .NUM_REQS  (`DCACHE_NUM_REQS), 
+        .WORD_SIZE (`DCACHE_WORD_SIZE), 
+        .TAG_WIDTH (`DCACHE_CORE_TAG_WIDTH-`SM_ENABLE)
+    ) dcache_rsp_tmp_if();
+
+    `RESET_RELAY (icache_reset);
+    `RESET_RELAY (dcache_reset);
+    `RESET_RELAY (mem_arb_reset);
+
+    VX_cache #(
+        .CACHE_ID           (`ICACHE_ID),
+        .CACHE_SIZE         (`ICACHE_SIZE),
+        .CACHE_LINE_SIZE    (`ICACHE_LINE_SIZE),
+        .NUM_BANKS          (1),
+        .WORD_SIZE          (`ICACHE_WORD_SIZE),
+        .NUM_REQS           (1),
+        .CREQ_SIZE          (`ICACHE_CREQ_SIZE),
+        .CRSQ_SIZE          (`ICACHE_CRSQ_SIZE),
+        .MSHR_SIZE          (`ICACHE_MSHR_SIZE),
+        .MRSQ_SIZE          (`ICACHE_MRSQ_SIZE),
+        .MREQ_SIZE          (`ICACHE_MREQ_SIZE),
+        .WRITE_ENABLE       (0),
+        .CORE_TAG_WIDTH     (`ICACHE_CORE_TAG_WIDTH),
+        .CORE_TAG_ID_BITS   (`ICACHE_CORE_TAG_ID_BITS),
+        .MEM_TAG_WIDTH      (`ICACHE_MEM_TAG_WIDTH)
+    ) icache (
+        `SCOPE_BIND_VX_mem_unit_icache
+
+        .clk                (clk),
+        .reset              (icache_reset),
+
+        // Core request
+        .core_req_valid     (icache_req_if.valid),
+        .core_req_rw        (1'b0),
+        .core_req_byteen    ('b0),
+        .core_req_addr      (icache_req_if.addr),
+        .core_req_data      ('x),        
+        .core_req_tag       (icache_req_if.tag),
+        .core_req_ready     (icache_req_if.ready),
+
+        // Core response
+        .core_rsp_valid     (icache_rsp_if.valid),
+        .core_rsp_data      (icache_rsp_if.data),
+        .core_rsp_tag       (icache_rsp_if.tag),
+        .core_rsp_ready     (icache_rsp_if.ready),
+        `UNUSED_PIN (core_rsp_tmask),
+
+    `ifdef PERF_ENABLE
+        .perf_cache_if      (perf_icache_if),
+    `endif
+
+        // Memory Request
+        .mem_req_valid     (icache_mem_req_if.valid),
+        .mem_req_rw        (icache_mem_req_if.rw),        
+        .mem_req_byteen    (icache_mem_req_if.byteen),        
+        .mem_req_addr      (icache_mem_req_if.addr),
+        .mem_req_data      (icache_mem_req_if.data),
+        .mem_req_tag       (icache_mem_req_if.tag),
+        .mem_req_ready     (icache_mem_req_if.ready),        
+
+        // Memory response
+        .mem_rsp_valid     (icache_mem_rsp_if.valid),        
+        .mem_rsp_data      (icache_mem_rsp_if.data),
+        .mem_rsp_tag       (icache_mem_rsp_if.tag),
+        .mem_rsp_ready     (icache_mem_rsp_if.ready)
+    );
+
+    VX_cache #(
+        .CACHE_ID           (`DCACHE_ID),
+        .CACHE_SIZE         (`DCACHE_SIZE),
+        .CACHE_LINE_SIZE    (`DCACHE_LINE_SIZE),
+        .NUM_BANKS          (`DCACHE_NUM_BANKS),
+        .NUM_PORTS          (`DCACHE_NUM_PORTS),
+        .WORD_SIZE          (`DCACHE_WORD_SIZE),
+        .NUM_REQS           (`DCACHE_NUM_REQS),
+        .CREQ_SIZE          (`DCACHE_CREQ_SIZE),
+        .CRSQ_SIZE          (`DCACHE_CRSQ_SIZE),
+        .MSHR_SIZE          (`DCACHE_MSHR_SIZE),
+        .MRSQ_SIZE          (`DCACHE_MRSQ_SIZE),
+        .MREQ_SIZE          (`DCACHE_MREQ_SIZE),
+        .WRITE_ENABLE       (1),
+        .CORE_TAG_WIDTH     (`DCACHE_CORE_TAG_WIDTH-`SM_ENABLE),
+        .CORE_TAG_ID_BITS   (`DCACHE_CORE_TAG_ID_BITS-`SM_ENABLE),
+        .MEM_TAG_WIDTH      (`DCACHE_MEM_TAG_WIDTH),
+        .NC_ENABLE          (1)
+    ) dcache (
+        `SCOPE_BIND_VX_mem_unit_dcache
+        
+        .clk                (clk),
+        .reset              (dcache_reset),
+
+        // Core req
+        .core_req_valid     (dcache_req_tmp_if.valid),
+        .core_req_rw        (dcache_req_tmp_if.rw),
+        .core_req_byteen    (dcache_req_tmp_if.byteen),
+        .core_req_addr      (dcache_req_tmp_if.addr),
+        .core_req_data      (dcache_req_tmp_if.data),        
+        .core_req_tag       (dcache_req_tmp_if.tag),
+        .core_req_ready     (dcache_req_tmp_if.ready),
+
+        // Core response
+        .core_rsp_valid     (dcache_rsp_tmp_if.valid),
+        .core_rsp_tmask     (dcache_rsp_tmp_if.tmask),
+        .core_rsp_data      (dcache_rsp_tmp_if.data),
+        .core_rsp_tag       (dcache_rsp_tmp_if.tag),
+        .core_rsp_ready     (dcache_rsp_tmp_if.ready),
+
+    `ifdef PERF_ENABLE
+        .perf_cache_if      (perf_dcache_if),
+    `endif
+
+        // Memory request
+        .mem_req_valid      (dcache_mem_req_if.valid),
+        .mem_req_rw         (dcache_mem_req_if.rw),        
+        .mem_req_byteen     (dcache_mem_req_if.byteen),        
+        .mem_req_addr       (dcache_mem_req_if.addr),
+        .mem_req_data       (dcache_mem_req_if.data),
+        .mem_req_tag        (dcache_mem_req_if.tag),
+        .mem_req_ready      (dcache_mem_req_if.ready),
+
+        // Memory response
+        .mem_rsp_valid      (dcache_mem_rsp_if.valid),        
+        .mem_rsp_data       (dcache_mem_rsp_if.data),
+        .mem_rsp_tag        (dcache_mem_rsp_if.tag),
+        .mem_rsp_ready      (dcache_mem_rsp_if.ready)
+    ); 
+
+    if (`SM_ENABLE) begin                
+        VX_dcache_req_if #(
+            .NUM_REQS  (`DCACHE_NUM_REQS), 
+            .WORD_SIZE (`DCACHE_WORD_SIZE), 
+            .TAG_WIDTH (`DCACHE_CORE_TAG_WIDTH-`SM_ENABLE)
+        ) smem_req_if();
+
+        VX_dcache_rsp_if #(
+            .NUM_REQS  (`DCACHE_NUM_REQS), 
+            .WORD_SIZE (`DCACHE_WORD_SIZE), 
+            .TAG_WIDTH (`DCACHE_CORE_TAG_WIDTH-`SM_ENABLE)
+        ) smem_rsp_if();
+
+        `RESET_RELAY (smem_arb_reset);
+        `RESET_RELAY (smem_reset);
+
+        VX_smem_arb #(
+            .NUM_REQS      (2),
+            .LANES         (`NUM_THREADS),
+            .DATA_SIZE     (4),            
+            .TAG_IN_WIDTH  (`DCACHE_CORE_TAG_WIDTH),
+            .TAG_SEL_IDX   (0), // SM flag
+            .TYPE          ("P"),
+            .BUFFERED_REQ  (2),
+            .BUFFERED_RSP  (1)
+        ) smem_arb (
+            .clk          (clk),
+            .reset        (smem_arb_reset),
+
+            // input request
+            .req_valid_in   (dcache_req_if.valid),
+            .req_rw_in      (dcache_req_if.rw),        
+            .req_byteen_in  (dcache_req_if.byteen),        
+            .req_addr_in    (dcache_req_if.addr),
+            .req_data_in    (dcache_req_if.data),
+            .req_tag_in     (dcache_req_if.tag),
+            .req_ready_in   (dcache_req_if.ready),
+            
+            // output requests
+            .req_valid_out  ({smem_req_if.valid,  dcache_req_tmp_if.valid}),
+            .req_rw_out     ({smem_req_if.rw,     dcache_req_tmp_if.rw}),
+            .req_byteen_out ({smem_req_if.byteen, dcache_req_tmp_if.byteen}),
+            .req_addr_out   ({smem_req_if.addr,   dcache_req_tmp_if.addr}),
+            .req_data_out   ({smem_req_if.data,   dcache_req_tmp_if.data}),  
+            .req_tag_out    ({smem_req_if.tag,    dcache_req_tmp_if.tag}),  
+            .req_ready_out  ({smem_req_if.ready,  dcache_req_tmp_if.ready}),            
+            
+            // input responses
+            .rsp_valid_in   ({smem_rsp_if.valid, dcache_rsp_tmp_if.valid}),
+            .rsp_tmask_in   ({smem_rsp_if.tmask, dcache_rsp_tmp_if.tmask}),
+            .rsp_data_in    ({smem_rsp_if.data,  dcache_rsp_tmp_if.data}),
+            .rsp_tag_in     ({smem_rsp_if.tag,   dcache_rsp_tmp_if.tag}),
+            .rsp_ready_in   ({smem_rsp_if.ready, dcache_rsp_tmp_if.ready}),
+
+            // output response
+            .rsp_valid_out  (dcache_rsp_if.valid),
+            .rsp_tmask_out  (dcache_rsp_if.tmask),
+            .rsp_tag_out    (dcache_rsp_if.tag),
+            .rsp_data_out   (dcache_rsp_if.data),
+            .rsp_ready_out  (dcache_rsp_if.ready)
+        );
+
+        VX_shared_mem #(
+            .CACHE_ID           (`SMEM_ID),
+            .CACHE_SIZE         (`SMEM_SIZE),
+            .NUM_BANKS          (`SMEM_NUM_BANKS),
+            .WORD_SIZE          (`SMEM_WORD_SIZE),
+            .NUM_REQS           (`SMEM_NUM_REQS),
+            .CREQ_SIZE          (`SMEM_CREQ_SIZE),
+            .CRSQ_SIZE          (`SMEM_CRSQ_SIZE),
+            .CORE_TAG_WIDTH     (`DCACHE_CORE_TAG_WIDTH-`SM_ENABLE),
+            .CORE_TAG_ID_BITS   (`DCACHE_CORE_TAG_ID_BITS-`SM_ENABLE),
+            .BANK_ADDR_OFFSET   (`SMEM_BANK_ADDR_OFFSET)
+        ) smem (            
+            .clk                (clk),
+            .reset              (smem_reset),
+
+        `ifdef PERF_ENABLE
+            .perf_cache_if      (perf_smem_if),
+        `endif
+
+            // Core request
+            .core_req_valid     (smem_req_if.valid),
+            .core_req_rw        (smem_req_if.rw),
+            .core_req_byteen    (smem_req_if.byteen),
+            .core_req_addr      (smem_req_if.addr),
+            .core_req_data      (smem_req_if.data),        
+            .core_req_tag       (smem_req_if.tag),
+            .core_req_ready     (smem_req_if.ready),
+
+            // Core response
+            .core_rsp_valid     (smem_rsp_if.valid),
+            .core_rsp_tmask     (smem_rsp_if.tmask),
+            .core_rsp_data      (smem_rsp_if.data),
+            .core_rsp_tag       (smem_rsp_if.tag),
+            .core_rsp_ready     (smem_rsp_if.ready)
+        );    
+    end else begin
+        // core to D-cache request
+        for (genvar i = 0; i < `DCACHE_NUM_REQS; ++i) begin
+            VX_skid_buffer #(
+                .DATAW ((32-`CLOG2(`DCACHE_WORD_SIZE)) + 1 + `DCACHE_WORD_SIZE + (8*`DCACHE_WORD_SIZE) + `DCACHE_CORE_TAG_WIDTH)
+            ) req_buf (
+                .clk       (clk),
+                .reset     (reset),
+                .valid_in  (dcache_req_if.valid[i]),        
+                .data_in   ({dcache_req_if.addr[i], dcache_req_if.rw[i], dcache_req_if.byteen[i], dcache_req_if.data[i], dcache_req_if.tag[i]}),
+                .ready_in  (dcache_req_if.ready[i]),
+                .valid_out (dcache_req_tmp_if.valid[i]),
+                .data_out  ({dcache_req_tmp_if.addr[i], dcache_req_tmp_if.rw[i], dcache_req_tmp_if.byteen[i], dcache_req_tmp_if.data[i], dcache_req_tmp_if.tag[i]}),
+                .ready_out (dcache_req_tmp_if.ready[i])
+            );
+        end
+        
+        // D-cache to core reponse
+        assign dcache_rsp_if.valid  = dcache_rsp_tmp_if.valid;
+        assign dcache_rsp_if.tmask  = dcache_rsp_tmp_if.tmask;
+        assign dcache_rsp_if.tag    = dcache_rsp_tmp_if.tag;
+        assign dcache_rsp_if.data   = dcache_rsp_tmp_if.data;
+        assign dcache_rsp_tmp_if.ready = dcache_rsp_if.ready;
+    end
+
+    wire [`DCACHE_MEM_TAG_WIDTH-1:0] icache_mem_req_tag = `DCACHE_MEM_TAG_WIDTH'(icache_mem_req_if.tag);
+    wire [`DCACHE_MEM_TAG_WIDTH-1:0] icache_mem_rsp_tag;
+    assign icache_mem_rsp_if.tag = icache_mem_rsp_tag[`ICACHE_MEM_TAG_WIDTH-1:0];
+    `UNUSED_VAR (icache_mem_rsp_tag)
+
+    VX_mem_arb #(
+        .NUM_REQS      (2),
+        .DATA_WIDTH    (`DCACHE_MEM_DATA_WIDTH),
+        .ADDR_WIDTH    (`DCACHE_MEM_ADDR_WIDTH),
+        .TAG_IN_WIDTH  (`DCACHE_MEM_TAG_WIDTH),
+        .TYPE          ("R"),
+        .TAG_SEL_IDX   (1), // Skip 0 for NC flag
+        .BUFFERED_REQ  (1),
+        .BUFFERED_RSP  (2)
+    ) mem_arb (
+        .clk            (clk),
+        .reset          (mem_arb_reset),
+
+        // Source request
+        .req_valid_in   ({dcache_mem_req_if.valid,  icache_mem_req_if.valid}),
+        .req_rw_in      ({dcache_mem_req_if.rw,     icache_mem_req_if.rw}),
+        .req_byteen_in  ({dcache_mem_req_if.byteen, icache_mem_req_if.byteen}),
+        .req_addr_in    ({dcache_mem_req_if.addr,   icache_mem_req_if.addr}),
+        .req_data_in    ({dcache_mem_req_if.data,   icache_mem_req_if.data}),  
+        .req_tag_in     ({dcache_mem_req_if.tag,    icache_mem_req_tag}),  
+        .req_ready_in   ({dcache_mem_req_if.ready,  icache_mem_req_if.ready}),
+
+        // Memory request
+        .req_valid_out  (mem_req_if.valid),
+        .req_rw_out     (mem_req_if.rw),        
+        .req_byteen_out (mem_req_if.byteen),        
+        .req_addr_out   (mem_req_if.addr),
+        .req_data_out   (mem_req_if.data),
+        .req_tag_out    (mem_req_if.tag),
+        .req_ready_out  (mem_req_if.ready),
+
+        // Source response
+        .rsp_valid_out  ({dcache_mem_rsp_if.valid, icache_mem_rsp_if.valid}),
+        .rsp_data_out   ({dcache_mem_rsp_if.data,  icache_mem_rsp_if.data}),
+        .rsp_tag_out    ({dcache_mem_rsp_if.tag,   icache_mem_rsp_tag}),
+        .rsp_ready_out  ({dcache_mem_rsp_if.ready, icache_mem_rsp_if.ready}),
+        
+        // Memory response
+        .rsp_valid_in   (mem_rsp_if.valid),
+        .rsp_tag_in     (mem_rsp_if.tag),
+        .rsp_data_in    (mem_rsp_if.data),
+        .rsp_ready_in   (mem_rsp_if.ready)
+    );
+
+`ifdef PERF_ENABLE
+    
+    `UNUSED_VAR (perf_dcache_if.mem_stalls)
+    `UNUSED_VAR (perf_dcache_if.crsp_stalls)
+
+    assign perf_memsys_if.icache_reads       = perf_icache_if.reads;
+    assign perf_memsys_if.icache_read_misses = perf_icache_if.read_misses;
+    assign perf_memsys_if.dcache_reads       = perf_dcache_if.reads;
+    assign perf_memsys_if.dcache_writes      = perf_dcache_if.writes;
+    assign perf_memsys_if.dcache_read_misses = perf_dcache_if.read_misses;
+    assign perf_memsys_if.dcache_write_misses= perf_dcache_if.write_misses;
+    assign perf_memsys_if.dcache_bank_stalls = perf_dcache_if.bank_stalls;
+    assign perf_memsys_if.dcache_mshr_stalls = perf_dcache_if.mshr_stalls;    
+
+if (`SM_ENABLE) begin
+    assign perf_memsys_if.smem_reads         = perf_smem_if.reads;
+    assign perf_memsys_if.smem_writes        = perf_smem_if.writes;
+    assign perf_memsys_if.smem_bank_stalls   = perf_smem_if.bank_stalls;    
+end else begin
+    assign perf_memsys_if.smem_reads         = 0;
+    assign perf_memsys_if.smem_writes        = 0;
+    assign perf_memsys_if.smem_bank_stalls   = 0;
+end
+    
+    reg [`PERF_CTR_BITS-1:0] perf_mem_pending_reads;
+
+    always @(posedge clk) begin
+        if (reset) begin
+            perf_mem_pending_reads <= 0;
+        end else begin
+            perf_mem_pending_reads <= perf_mem_pending_reads + 
+                `PERF_CTR_BITS'($signed(2'((mem_req_if.valid && mem_req_if.ready && !mem_req_if.rw) && !(mem_rsp_if.valid && mem_rsp_if.ready)) - 
+                    2'((mem_rsp_if.valid && mem_rsp_if.ready) && !(mem_req_if.valid && mem_req_if.ready && !mem_req_if.rw))));
+        end
+    end
+    
+    reg [`PERF_CTR_BITS-1:0] perf_mem_reads;
+    reg [`PERF_CTR_BITS-1:0] perf_mem_writes;
+    reg [`PERF_CTR_BITS-1:0] perf_mem_lat;
+
+    always @(posedge clk) begin
+        if (reset) begin       
+            perf_mem_reads  <= 0;     
+            perf_mem_writes <= 0;            
+            perf_mem_lat    <= 0;
+        end else begin  
+            if (mem_req_if.valid && mem_req_if.ready && !mem_req_if.rw) begin
+                perf_mem_reads <= perf_mem_reads + `PERF_CTR_BITS'd1;
+            end
+            if (mem_req_if.valid && mem_req_if.ready && mem_req_if.rw) begin
+                perf_mem_writes <= perf_mem_writes + `PERF_CTR_BITS'd1;
+            end      
+            perf_mem_lat <= perf_mem_lat + perf_mem_pending_reads;
+        end
+    end
+
+    assign perf_memsys_if.mem_reads   = perf_mem_reads;       
+    assign perf_memsys_if.mem_writes  = perf_mem_writes;
+    assign perf_memsys_if.mem_latency = perf_mem_lat;
+`endif
+    
+endmodule

hw/rtl/VX_muldiv.sv

@@ -0,0 +1,226 @@
+`include "VX_define.vh"
+
+module VX_muldiv (
+    input wire clk,
+    input wire reset,
+    
+    // Inputs    
+    input wire [`INST_MUL_BITS-1:0]     alu_op,
+    input wire [`UUID_BITS-1:0]         uuid_in,
+    input wire [`NW_BITS-1:0]           wid_in,
+    input wire [`NUM_THREADS-1:0]       tmask_in,
+    input wire [31:0]                   PC_in,
+    input wire [`NR_BITS-1:0]           rd_in,
+    input wire                          wb_in,
+    input wire [`NUM_THREADS-1:0][31:0] alu_in1, 
+    input wire [`NUM_THREADS-1:0][31:0] alu_in2,
+
+    // Outputs
+    output wire [`UUID_BITS-1:0]         uuid_out,
+    output wire [`NW_BITS-1:0]           wid_out,
+    output wire [`NUM_THREADS-1:0]       tmask_out,
+    output wire [31:0]                   PC_out,
+    output wire [`NR_BITS-1:0]           rd_out,
+    output wire                          wb_out,
+    output wire [`NUM_THREADS-1:0][31:0] data_out,
+
+    // handshake
+    input wire  valid_in,
+    output wire ready_in,
+    output wire valid_out,
+    input wire  ready_out
+); 
+
+    wire is_div_op = `INST_MUL_IS_DIV(alu_op);
+
+    wire [`NUM_THREADS-1:0][31:0] mul_result;
+    wire [`UUID_BITS-1:0] mul_uuid_out;
+    wire [`NW_BITS-1:0] mul_wid_out;
+    wire [`NUM_THREADS-1:0] mul_tmask_out;
+    wire [31:0] mul_PC_out;
+    wire [`NR_BITS-1:0] mul_rd_out;
+    wire mul_wb_out;
+
+    wire stall_out;
+
+    wire mul_valid_out;
+    wire mul_valid_in = valid_in && !is_div_op;    
+    wire mul_ready_in = ~stall_out || ~mul_valid_out;
+
+    wire is_mulh_in      = (alu_op != `INST_MUL_MUL);
+    wire is_signed_mul_a = (alu_op != `INST_MUL_MULHU);
+    wire is_signed_mul_b = (alu_op != `INST_MUL_MULHU && alu_op != `INST_MUL_MULHSU);
+
+`ifdef IMUL_DPI
+
+    wire [`NUM_THREADS-1:0][31:0] mul_result_tmp;  
+
+    wire mul_fire_in = mul_valid_in && mul_ready_in;
+
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin
+        wire [31:0] mul_resultl, mul_resulth;
+        always @(*) begin        
+            dpi_imul (mul_fire_in, alu_in1[i], alu_in2[i], is_signed_mul_a, is_signed_mul_b, mul_resultl, mul_resulth);
+        end
+        assign mul_result_tmp[i] = is_mulh_in ? mul_resulth : mul_resultl;
+    end
+
+    VX_shift_register #(
+        .DATAW  (1 + `UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `NR_BITS + 1 + (`NUM_THREADS * 32)),
+        .DEPTH  (`LATENCY_IMUL),
+        .RESETW (1)
+    ) mul_shift_reg (
+        .clk(clk),
+        .reset    (reset),
+        .enable   (mul_ready_in),
+        .data_in  ({mul_valid_in,  uuid_in,      wid_in,      tmask_in,       PC_in,      rd_in,      wb_in,      mul_result_tmp}),
+        .data_out ({mul_valid_out, mul_uuid_out, mul_wid_out, mul_tmask_out,  mul_PC_out, mul_rd_out, mul_wb_out, mul_result})
+    );
+
+`else      
+    
+    wire is_mulh_out;
+
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin
+        wire [32:0] mul_in1 = {is_signed_mul_a & alu_in1[i][31], alu_in1[i]};
+        wire [32:0] mul_in2 = {is_signed_mul_b & alu_in2[i][31], alu_in2[i]};
+    `IGNORE_UNUSED_BEGIN
+        wire [65:0] mul_result_tmp;
+    `IGNORE_UNUSED_END
+
+        VX_multiplier #(
+            .WIDTHA  (33),
+            .WIDTHB  (33),
+            .WIDTHP  (66),
+            .SIGNED  (1),
+            .LATENCY (`LATENCY_IMUL)
+        ) multiplier (
+            .clk    (clk),
+            .enable (mul_ready_in),
+            .dataa  (mul_in1),
+            .datab  (mul_in2),
+            .result (mul_result_tmp)
+        );
+
+        assign mul_result[i] = is_mulh_out ? mul_result_tmp[63:32] : mul_result_tmp[31:0];
+    end
+
+    VX_shift_register #(
+        .DATAW  (1 + `UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `NR_BITS + 1 + 1),
+        .DEPTH  (`LATENCY_IMUL),
+        .RESETW (1)
+    ) mul_shift_reg (
+        .clk(clk),
+        .reset    (reset),
+        .enable   (mul_ready_in),
+        .data_in  ({mul_valid_in,  uuid_in,      wid_in,      tmask_in,       PC_in,      rd_in,      wb_in,      is_mulh_in}),
+        .data_out ({mul_valid_out, mul_uuid_out, mul_wid_out, mul_tmask_out,  mul_PC_out, mul_rd_out, mul_wb_out, is_mulh_out})
+    );
+
+`endif
+
+    ///////////////////////////////////////////////////////////////////////////
+
+    wire [`NUM_THREADS-1:0][31:0] div_result;
+    wire [`UUID_BITS-1:0] div_uuid_out;
+    wire [`NW_BITS-1:0] div_wid_out;
+    wire [`NUM_THREADS-1:0] div_tmask_out;
+    wire [31:0] div_PC_out;
+    wire [`NR_BITS-1:0] div_rd_out;
+    wire div_wb_out;
+
+    wire is_rem_op_in  = (alu_op == `INST_MUL_REM) || (alu_op == `INST_MUL_REMU);
+    wire is_signed_div = (alu_op == `INST_MUL_DIV) || (alu_op == `INST_MUL_REM);     
+    wire div_valid_in  = valid_in && is_div_op; 
+    wire div_ready_out = ~stall_out && ~mul_valid_out; // arbitration prioritizes MUL  
+    wire div_ready_in;
+    wire div_valid_out;
+
+`ifdef IDIV_DPI    
+
+    wire [`NUM_THREADS-1:0][31:0] div_result_tmp;
+
+    wire div_fire_in = div_valid_in && div_ready_in;
+    
+    for (genvar i = 0; i < `NUM_THREADS; i++) begin
+        wire [31:0] div_quotient, div_remainder;
+        always @(*) begin        
+            dpi_idiv (div_fire_in, alu_in1[i], alu_in2[i], is_signed_div, div_quotient, div_remainder);
+        end
+        assign div_result_tmp[i] = is_rem_op_in ? div_remainder : div_quotient;
+    end
+
+    VX_shift_register #(
+        .DATAW  (1 + `UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `NR_BITS + 1 + (`NUM_THREADS * 32)),
+        .DEPTH  (`LATENCY_IMUL),
+        .RESETW (1)
+    ) div_shift_reg (
+        .clk(clk),
+        .reset    (reset),
+        .enable   (div_ready_in),
+        .data_in  ({div_valid_in,  uuid_in,      wid_in,      tmask_in,       PC_in,      rd_in,      wb_in,      div_result_tmp}),
+        .data_out ({div_valid_out, div_uuid_out, div_wid_out, div_tmask_out,  div_PC_out, div_rd_out, div_wb_out, div_result})
+    );
+
+    assign div_ready_in = div_ready_out || ~div_valid_out;
+
+`else
+
+    wire [`NUM_THREADS-1:0][31:0] div_result_tmp, rem_result_tmp;
+    wire is_rem_op_out;
+    
+    VX_serial_div #(
+        .WIDTHN (32),
+        .WIDTHD (32),
+        .WIDTHQ (32),
+        .WIDTHR (32),
+        .LANES  (`NUM_THREADS),
+        .TAGW   (64 + `NW_BITS + `NUM_THREADS + 32 + `NR_BITS + 1 + 1)
+    ) divide (
+        .clk       (clk),
+        .reset     (reset),        
+        .valid_in  (div_valid_in),
+        .ready_in  (div_ready_in),
+        .signed_mode(is_signed_div),
+        .tag_in    ({uuid_in, wid_in, tmask_in, PC_in, rd_in, wb_in, is_rem_op_in}),
+        .numer     (alu_in1),
+        .denom     (alu_in2),
+        .quotient  (div_result_tmp),
+        .remainder (rem_result_tmp),
+        .ready_out (div_ready_out),
+        .valid_out (div_valid_out),
+        .tag_out   ({div_uuid_out, div_wid_out, div_tmask_out, div_PC_out, div_rd_out, div_wb_out, is_rem_op_out})
+    );
+
+    assign div_result = is_rem_op_out ? rem_result_tmp : div_result_tmp; 
+
+`endif
+
+    ///////////////////////////////////////////////////////////////////////////
+
+    wire                    rsp_valid = mul_valid_out || div_valid_out;  
+    wire [`UUID_BITS-1:0]   rsp_uuid  = mul_valid_out ? mul_uuid_out : div_uuid_out;
+    wire [`NW_BITS-1:0]     rsp_wid   = mul_valid_out ? mul_wid_out : div_wid_out;
+    wire [`NUM_THREADS-1:0] rsp_tmask = mul_valid_out ? mul_tmask_out : div_tmask_out;
+    wire [31:0]             rsp_PC    = mul_valid_out ? mul_PC_out : div_PC_out;
+    wire [`NR_BITS-1:0]     rsp_rd    = mul_valid_out ? mul_rd_out : div_rd_out;
+    wire                    rsp_wb    = mul_valid_out ? mul_wb_out : div_wb_out;
+    wire [`NUM_THREADS-1:0][31:0] rsp_data = mul_valid_out ? mul_result : div_result;
+
+    assign stall_out = ~ready_out && valid_out;
+
+    VX_pipe_register #(
+        .DATAW  (1 + `UUID_BITS + `NW_BITS + `NUM_THREADS + 32 + `NR_BITS + 1 + (`NUM_THREADS * 32)),
+        .RESETW (1)
+    ) pipe_reg (
+        .clk      (clk),
+        .reset    (reset),
+        .enable   (~stall_out),
+        .data_in  ({rsp_valid, rsp_uuid, rsp_wid, rsp_tmask, rsp_PC, rsp_rd, rsp_wb, rsp_data}),
+        .data_out ({valid_out, uuid_out, wid_out, tmask_out, PC_out, rd_out, wb_out, data_out})
+    );
+
+    // can accept new request?
+    assign ready_in = is_div_op ? div_ready_in : mul_ready_in;
+    
+endmodule

hw/rtl/VX_pipeline.sv

@@ -0,0 +1,261 @@
+`include "VX_define.vh"
+
+module VX_pipeline #( 
+    parameter CORE_ID = 0
+) (        
+    `SCOPE_IO_VX_pipeline
+    
+    // Clock
+    input wire                              clk,
+    input wire                              reset,
+
+    // Dcache core request
+    output wire [`NUM_THREADS-1:0]          dcache_req_valid,
+    output wire [`NUM_THREADS-1:0]          dcache_req_rw,
+    output wire [`NUM_THREADS-1:0][3:0]     dcache_req_byteen,
+    output wire [`NUM_THREADS-1:0][29:0]    dcache_req_addr,
+    output wire [`NUM_THREADS-1:0][31:0]    dcache_req_data,
+    output wire [`NUM_THREADS-1:0][`DCACHE_CORE_TAG_WIDTH-1:0] dcache_req_tag,    
+    input wire [`NUM_THREADS-1:0]           dcache_req_ready,
+
+    // Dcache core reponse    
+    input wire                              dcache_rsp_valid,
+    input wire [`NUM_THREADS-1:0]           dcache_rsp_tmask,
+    input wire [`NUM_THREADS-1:0][31:0]     dcache_rsp_data,
+    input wire [`DCACHE_CORE_TAG_WIDTH-1:0] dcache_rsp_tag,    
+    output wire                             dcache_rsp_ready,      
+
+    // Icache core request
+    output wire                             icache_req_valid,
+    output wire [29:0]                      icache_req_addr,
+    output wire [`ICACHE_CORE_TAG_WIDTH-1:0] icache_req_tag,
+    input wire                              icache_req_ready,
+
+    // Icache core response    
+    input wire                              icache_rsp_valid,
+    input wire [31:0]                       icache_rsp_data,
+    input wire [`ICACHE_CORE_TAG_WIDTH-1:0] icache_rsp_tag,    
+    output wire                             icache_rsp_ready,   
+
+`ifdef PERF_ENABLE
+    VX_perf_memsys_if.slave                 perf_memsys_if,
+`endif
+
+    // Status
+    output wire                             busy
+);
+    //
+    // Dcache request
+    //
+
+    VX_dcache_req_if #(
+        .NUM_REQS  (`NUM_THREADS), 
+        .WORD_SIZE (4), 
+        .TAG_WIDTH (`DCACHE_CORE_TAG_WIDTH)
+    ) dcache_req_if();
+
+    assign dcache_req_valid  = dcache_req_if.valid;
+    assign dcache_req_rw     = dcache_req_if.rw;
+    assign dcache_req_byteen = dcache_req_if.byteen;
+    assign dcache_req_addr   = dcache_req_if.addr;
+    assign dcache_req_data   = dcache_req_if.data;
+    assign dcache_req_tag    = dcache_req_if.tag;
+    assign dcache_req_if.ready = dcache_req_ready;
+ 
+    //
+    // Dcache response
+    //
+
+    VX_dcache_rsp_if #(
+        .NUM_REQS  (`NUM_THREADS), 
+        .WORD_SIZE (4), 
+        .TAG_WIDTH (`DCACHE_CORE_TAG_WIDTH)
+    ) dcache_rsp_if();
+
+    assign dcache_rsp_if.valid = dcache_rsp_valid;
+    assign dcache_rsp_if.tmask = dcache_rsp_tmask;
+    assign dcache_rsp_if.data  = dcache_rsp_data;
+    assign dcache_rsp_if.tag   = dcache_rsp_tag;
+    assign dcache_rsp_ready = dcache_rsp_if.ready;
+
+    //
+    // Icache request
+    //
+
+    VX_icache_req_if #(
+        .WORD_SIZE (4), 
+        .TAG_WIDTH (`ICACHE_CORE_TAG_WIDTH)
+    ) icache_req_if();       
+
+    assign icache_req_valid  = icache_req_if.valid;
+    assign icache_req_addr   = icache_req_if.addr;
+    assign icache_req_tag    = icache_req_if.tag;
+    assign icache_req_if.ready = icache_req_ready;
+
+    //
+    // Icache response
+    //
+
+    VX_icache_rsp_if #(
+        .WORD_SIZE (4), 
+        .TAG_WIDTH (`ICACHE_CORE_TAG_WIDTH)
+    ) icache_rsp_if();    
+
+    assign icache_rsp_if.valid = icache_rsp_valid;
+    assign icache_rsp_if.data  = icache_rsp_data;
+    assign icache_rsp_if.tag   = icache_rsp_tag;
+    assign icache_rsp_ready = icache_rsp_if.ready;
+
+    ///////////////////////////////////////////////////////////////////////////
+
+    VX_fetch_to_csr_if  fetch_to_csr_if();
+    VX_cmt_to_csr_if    cmt_to_csr_if();
+    VX_decode_if        decode_if();
+    VX_branch_ctl_if    branch_ctl_if();
+    VX_warp_ctl_if      warp_ctl_if();
+    VX_ifetch_rsp_if    ifetch_rsp_if();
+    VX_alu_req_if       alu_req_if();
+    VX_lsu_req_if       lsu_req_if();
+    VX_csr_req_if       csr_req_if();
+`ifdef EXT_F_ENABLE 
+    VX_fpu_req_if       fpu_req_if(); 
+`endif
+    VX_gpu_req_if       gpu_req_if();
+    VX_writeback_if     writeback_if();     
+    VX_wstall_if        wstall_if();
+    VX_join_if          join_if();
+    VX_commit_if        alu_commit_if();
+    VX_commit_if        ld_commit_if();
+    VX_commit_if        st_commit_if();
+    VX_commit_if        csr_commit_if();  
+`ifdef EXT_F_ENABLE
+    VX_commit_if        fpu_commit_if();     
+`endif
+    VX_commit_if        gpu_commit_if();     
+
+`ifdef PERF_ENABLE
+    VX_perf_pipeline_if perf_pipeline_if();
+`endif
+
+    `RESET_RELAY (fetch_reset);
+    `RESET_RELAY (decode_reset);
+    `RESET_RELAY (issue_reset);
+    `RESET_RELAY (execute_reset);
+    `RESET_RELAY (commit_reset);
+
+    VX_fetch #(
+        .CORE_ID(CORE_ID)
+    ) fetch (
+        `SCOPE_BIND_VX_pipeline_fetch
+        .clk            (clk),
+        .reset          (fetch_reset),
+        .icache_req_if  (icache_req_if),
+        .icache_rsp_if  (icache_rsp_if), 
+        .wstall_if      (wstall_if),
+        .join_if        (join_if),        
+        .warp_ctl_if    (warp_ctl_if),
+        .branch_ctl_if  (branch_ctl_if),
+        .ifetch_rsp_if  (ifetch_rsp_if),
+        .fetch_to_csr_if(fetch_to_csr_if),
+        .busy           (busy)
+    );
+
+    VX_decode #(
+        .CORE_ID(CORE_ID)
+    ) decode (
+        .clk            (clk),
+        .reset          (decode_reset),        
+    `ifdef PERF_ENABLE
+        .perf_decode_if (perf_pipeline_if.decode),
+    `endif
+        .ifetch_rsp_if  (ifetch_rsp_if),
+        .decode_if      (decode_if),
+        .wstall_if      (wstall_if),
+        .join_if        (join_if)
+    );
+
+    VX_issue #(
+        .CORE_ID(CORE_ID)
+    ) issue (
+        `SCOPE_BIND_VX_pipeline_issue
+
+        .clk            (clk),
+        .reset          (issue_reset),
+
+    `ifdef PERF_ENABLE
+        .perf_issue_if  (perf_pipeline_if.issue),
+    `endif
+
+        .decode_if      (decode_if),
+        .writeback_if   (writeback_if),
+
+        .alu_req_if     (alu_req_if),
+        .lsu_req_if     (lsu_req_if),        
+        .csr_req_if     (csr_req_if),
+    `ifdef EXT_F_ENABLE
+        .fpu_req_if     (fpu_req_if),
+    `endif
+        .gpu_req_if     (gpu_req_if)
+    );
+
+    VX_execute #(
+        .CORE_ID(CORE_ID)
+    ) execute (
+        `SCOPE_BIND_VX_pipeline_execute
+        
+        .clk            (clk),
+        .reset          (execute_reset),
+
+    `ifdef PERF_ENABLE
+        .perf_memsys_if (perf_memsys_if),
+        .perf_pipeline_if (perf_pipeline_if),
+    `endif 
+
+        .dcache_req_if  (dcache_req_if),
+        .dcache_rsp_if  (dcache_rsp_if),
+
+        .cmt_to_csr_if  (cmt_to_csr_if),   
+        .fetch_to_csr_if(fetch_to_csr_if),              
+        
+        .alu_req_if     (alu_req_if),
+        .lsu_req_if     (lsu_req_if),        
+        .csr_req_if     (csr_req_if),
+    `ifdef EXT_F_ENABLE
+        .fpu_req_if     (fpu_req_if),
+    `endif
+        .gpu_req_if     (gpu_req_if),
+
+        .warp_ctl_if    (warp_ctl_if),
+        .branch_ctl_if  (branch_ctl_if),        
+        .alu_commit_if  (alu_commit_if),
+        .ld_commit_if   (ld_commit_if),        
+        .st_commit_if   (st_commit_if),       
+        .csr_commit_if  (csr_commit_if),
+    `ifdef EXT_F_ENABLE
+        .fpu_commit_if  (fpu_commit_if),
+    `endif
+        .gpu_commit_if  (gpu_commit_if),        
+        
+        .busy           (busy)
+    );    
+
+    VX_commit #(
+        .CORE_ID(CORE_ID)
+    ) commit (
+        .clk            (clk),
+        .reset          (commit_reset),
+
+        .alu_commit_if  (alu_commit_if),
+        .ld_commit_if   (ld_commit_if),        
+        .st_commit_if   (st_commit_if),
+        .csr_commit_if  (csr_commit_if),
+    `ifdef EXT_F_ENABLE
+        .fpu_commit_if  (fpu_commit_if),
+    `endif
+        .gpu_commit_if  (gpu_commit_if),
+        
+        .writeback_if   (writeback_if),
+        .cmt_to_csr_if  (cmt_to_csr_if)
+    );
+    
+endmodule

hw/rtl/VX_platform.vh

@@ -1,47 +1,7 @@
-// Copyright © 2019-2023
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// 
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
+`ifndef VX_PLATFORM
+`define VX_PLATFORM
 
-`ifndef VX_PLATFORM_VH
-`define VX_PLATFORM_VH
-
-// enable synthesizable build by default if not SIMULATION
-`ifndef SIMULATION
-`define SYNTHESIS
-`define NDEBUG
-`define DPI_DISABLE
-`else  // !SYNTHESIS
-`define SV_DPI
-`endif
-
-// chipyard-specific configs
-`define GPR_RESET
-`define GPR_DUPLICATED
-`define GBAR_ENABLE
-`define GBAR_CLUSTER_ENABLE
-`define ICACHE_DISABLE
-`define DCACHE_DISABLE
-
-`ifdef SYNTHESIS
-`define NUM_BARRIERS 8
-`define NUM_CORES 4
-`define NUM_THREADS 8
-`define NUM_WARPS 8
-
-`define FPU_FPNEW
-// `define FIRESIM
-`endif // SYNTHESIS
-
-`ifdef SV_DPI
+`ifndef SYNTHESIS
 `include "util_dpi.vh"
 `endif
 
@@ -49,221 +9,98 @@
 
 ///////////////////////////////////////////////////////////////////////////////
 
-`ifdef VIVADO
-`define STRING
-`else
-`ifdef SYNTHESIS
-`define STRING
-`else
-`define STRING string
-`endif
-`endif
+`ifndef SYNTHESIS
 
-`ifdef SYNTHESIS
-`define TRACE(level, args) $write args
-`define TRACE_STARTTIME 32'd10
-`define TRACING_ON
-`define TRACING_OFF
 `ifndef NDEBUG
-    `define DEBUG_BLOCK(x) x
+    `define DEBUG_BLOCK(x) /* verilator lint_off UNUSED */ \
+                           x \
+                           /* verilator lint_on UNUSED */
 `else
     `define DEBUG_BLOCK(x)
 `endif
-`define IGNORE_UNOPTFLAT_BEGIN
-`define IGNORE_UNOPTFLAT_END
+
+`define IGNORE_UNUSED_BEGIN   /* verilator lint_off UNUSED */
+
+`define IGNORE_UNUSED_END     /* verilator lint_on UNUSED */
+
+`define IGNORE_WARNINGS_BEGIN /* verilator lint_off UNUSED */ \
+                              /* verilator lint_off PINCONNECTEMPTY */ \
+                              /* verilator lint_off WIDTH */ \
+                              /* verilator lint_off UNOPTFLAT */ \
+                              /* verilator lint_off UNDRIVEN */ \
+                              /* verilator lint_off DECLFILENAME */ \
+                              /* verilator lint_off IMPLICIT */ \
+                              /* verilator lint_off IMPORTSTAR */
+
+`define IGNORE_WARNINGS_END   /* verilator lint_on UNUSED */ \
+                              /* verilator lint_on PINCONNECTEMPTY */ \
+                              /* verilator lint_on WIDTH */ \
+                              /* verilator lint_on UNOPTFLAT */ \
+                              /* verilator lint_on UNDRIVEN */ \
+                              /* verilator lint_on DECLFILENAME */ \
+                              /* verilator lint_on IMPLICIT */ \
+                              /* verilator lint_on IMPORTSTAR */
+
+`define UNUSED_PARAM(x)  /* verilator lint_off UNUSED */ \
+                         localparam  __``x = x; \
+                         /* verilator lint_on UNUSED */
+
+`define UNUSED_VAR(x) always @(x) begin end
+
+`define UNUSED_PIN(x)  /* verilator lint_off PINCONNECTEMPTY */ \
+                       . x () \
+                       /* verilator lint_on PINCONNECTEMPTY */
+
+`define ERROR(msg) \
+    $error msg
+
+`define ASSERT(cond, msg) \
+    assert(cond) else $error msg
+
+`define STATIC_ASSERT(cond, msg) \
+    generate                     \
+        if (!(cond)) $error msg; \
+    endgenerate
+
+`define RUNTIME_ASSERT(cond, msg)     \
+    always @(posedge clk) begin       \
+        assert(cond) else $error msg; \
+    end
+
+`define TRACING_ON  /* verilator tracing_on */
+`define TRACING_OFF /* verilator tracing_off */
+
+`else // SYNTHESIS
+
+`define DEBUG_BLOCK(x)
 `define IGNORE_UNUSED_BEGIN
 `define IGNORE_UNUSED_END
 `define IGNORE_WARNINGS_BEGIN
 `define IGNORE_WARNINGS_END
 `define UNUSED_PARAM(x)
-`define UNUSED_SPARAM(x)
 `define UNUSED_VAR(x)
 `define UNUSED_PIN(x) . x ()
-`define UNUSED_ARG(x) x
-`else // !SYNTHESIS
-`ifdef VERILATOR
-`define SIMULATION
-`define TRACING_ON      /* verilator tracing_on */
-`define TRACING_OFF     /* verilator tracing_off */
-`ifndef NDEBUG
-    `define DEBUG_BLOCK(x) /* verilator lint_off UNUSED */ \
-                           x \
-                           /* verilator lint_on UNUSED */
-`else
-    `define DEBUG_BLOCK(x)
-`endif
+`define ERROR(msg)
+`define ASSERT(cond, msg) if (cond);
+`define STATIC_ASSERT(cond, msg)
+`define RUNTIME_ASSERT(cond, msg)
+`define TRACING_ON
+`define TRACING_OFF
 
-`define IGNORE_UNOPTFLAT_BEGIN /* verilator lint_off UNOPTFLAT */
-
-`define IGNORE_UNOPTFLAT_END  /* verilator lint_off UNOPTFLAT */
-
-`define IGNORE_UNUSED_BEGIN   /* verilator lint_off UNUSED */
-
-`define IGNORE_UNUSED_END     /* verilator lint_on UNUSED */
-
-`define IGNORE_WARNINGS_BEGIN /* verilator lint_off UNUSED */ \
-                              /* verilator lint_off PINCONNECTEMPTY */ \
-                              /* verilator lint_off WIDTH */ \
-                              /* verilator lint_off UNOPTFLAT */ \
-                              /* verilator lint_off UNDRIVEN */ \
-                              /* verilator lint_off DECLFILENAME */ \
-                              /* verilator lint_off IMPLICIT */ \
-                              /* verilator lint_off PINMISSING */ \
-                              /* verilator lint_off IMPORTSTAR */ \
-                              /* verilator lint_off UNSIGNED */
-
-`define IGNORE_WARNINGS_END   /* verilator lint_on UNUSED */ \
-                              /* verilator lint_on PINCONNECTEMPTY */ \
-                              /* verilator lint_on WIDTH */ \
-                              /* verilator lint_on UNOPTFLAT */ \
-                              /* verilator lint_on UNDRIVEN */ \
-                              /* verilator lint_on DECLFILENAME */ \
-                              /* verilator lint_on IMPLICIT */ \
-                              /* verilator lint_off PINMISSING */ \
-                              /* verilator lint_on IMPORTSTAR */ \
-                              /* verilator lint_on UNSIGNED */
-
-`define UNUSED_PARAM(x)  /* verilator lint_off UNUSED */ \
-                         localparam  __``x = x; \
-                         /* verilator lint_on UNUSED */
-
-`define UNUSED_SPARAM(x) /* verilator lint_off UNUSED */ \
-                         localparam `STRING __``x = x; \
-                         /* verilator lint_on UNUSED */
-
-`define UNUSED_VAR(x)   if (1) begin \
-                            /* verilator lint_off UNUSED */ \
-                            wire [$bits(x)-1:0] __x = x; \
-                            /* verilator lint_on UNUSED */ \
-                        end
-
-`define UNUSED_PIN(x)   /* verilator lint_off PINCONNECTEMPTY */ \
-                        . x () \
-                        /* verilator lint_on PINCONNECTEMPTY */
-`define UNUSED_ARG(x)   /* verilator lint_off UNUSED */ \
-                        x \
-                        /* verilator lint_on UNUSED */
-`define TRACE(level, args) dpi_trace(level, $sformatf args)
-// squelch spurrious traces at the very first few cycles caused by to reset
-// delay
-`define TRACE_STARTTIME 32'd10
-`endif
-// NOTE(hansung): define these macros to be the same as VERILATOR under VCS;
-// they will mostly be ignored
-`ifdef VCS
-`define TRACING_ON      /* verilator tracing_on */
-`define TRACING_OFF     /* verilator tracing_off */
-`ifndef NDEBUG
-    `define DEBUG_BLOCK(x) /* verilator lint_off UNUSED */ \
-                           x \
-                           /* verilator lint_on UNUSED */
-`else
-    `define DEBUG_BLOCK(x)
-`endif
-
-`define IGNORE_UNOPTFLAT_BEGIN /* verilator lint_off UNOPTFLAT */
-
-`define IGNORE_UNOPTFLAT_END  /* verilator lint_off UNOPTFLAT */
-
-`define IGNORE_UNUSED_BEGIN   /* verilator lint_off UNUSED */
-
-`define IGNORE_UNUSED_END     /* verilator lint_on UNUSED */
-
-`define IGNORE_WARNINGS_BEGIN /* verilator lint_off UNUSED */ \
-                              /* verilator lint_off PINCONNECTEMPTY */ \
-                              /* verilator lint_off WIDTH */ \
-                              /* verilator lint_off UNOPTFLAT */ \
-                              /* verilator lint_off UNDRIVEN */ \
-                              /* verilator lint_off DECLFILENAME */ \
-                              /* verilator lint_off IMPLICIT */ \
-                              /* verilator lint_off PINMISSING */ \
-                              /* verilator lint_off IMPORTSTAR */ \
-                              /* verilator lint_off UNSIGNED */
-
-`define IGNORE_WARNINGS_END   /* verilator lint_on UNUSED */ \
-                              /* verilator lint_on PINCONNECTEMPTY */ \
-                              /* verilator lint_on WIDTH */ \
-                              /* verilator lint_on UNOPTFLAT */ \
-                              /* verilator lint_on UNDRIVEN */ \
-                              /* verilator lint_on DECLFILENAME */ \
-                              /* verilator lint_on IMPLICIT */ \
-                              /* verilator lint_off PINMISSING */ \
-                              /* verilator lint_on IMPORTSTAR */ \
-                              /* verilator lint_on UNSIGNED */
-
-`define UNUSED_PARAM(x)  /* verilator lint_off UNUSED */ \
-                         localparam  __``x = x; \
-                         /* verilator lint_on UNUSED */
-
-`define UNUSED_SPARAM(x) /* verilator lint_off UNUSED */ \
-                         localparam `STRING __``x = x; \
-                         /* verilator lint_on UNUSED */
-
-`define UNUSED_VAR(x)   if (1) begin \
-                            /* verilator lint_off UNUSED */ \
-                            wire [$bits(x)-1:0] __x = x; \
-                            /* verilator lint_on UNUSED */ \
-                        end
-
-`define UNUSED_PIN(x)   /* verilator lint_off PINCONNECTEMPTY */ \
-                        . x () \
-                        /* verilator lint_on PINCONNECTEMPTY */
-`define UNUSED_ARG(x)   /* verilator lint_off UNUSED */ \
-                        x \
-                        /* verilator lint_on UNUSED */
-`define TRACE(level, args) $write args
-// squelch spurrious traces at the very first few cycles caused by to reset
-// delay
-`define TRACE_STARTTIME 32'd10
-`endif
-`endif
-
-`ifdef SIMULATION
-    `define STATIC_ASSERT(cond, msg) \
-    generate                     \
-        if (!(cond)) $error msg; \
-    endgenerate
-
-    `define ERROR(msg) \
-        $error msg
-
-    `define ASSERT(cond, msg) \
-        assert(cond) else $error msg
-
-    `define RUNTIME_ASSERT(cond, msg)     \
-        always @(posedge clk) begin       \
-            assert(cond) else $error msg; \
-        end
-`else
-    `define STATIC_ASSERT(cond, msg)
-    `define ERROR(msg)                  //
-    `define ASSERT(cond, msg)           //
-    `define RUNTIME_ASSERT(cond, msg)
-`endif
+`endif // SYNTHESIS
 
 ///////////////////////////////////////////////////////////////////////////////
 
 `ifdef QUARTUS
-`define MAX_FANOUT      4
-`define IF_DATA_SIZE(x) $bits(x.data)
 `define USE_FAST_BRAM   (* ramstyle = "MLAB, no_rw_check" *)
 `define NO_RW_RAM_CHECK (* altera_attribute = "-name add_pass_through_logic_to_inferred_rams off" *)
 `define DISABLE_BRAM    (* ramstyle = "logic" *)
-`define PRESERVE_NET    (* preserve *)
-`elsif VIVADO
-`define MAX_FANOUT      4
-`define IF_DATA_SIZE(x) $bits(x.data)
-`define USE_FAST_BRAM   (* ram_style = "distributed" *)
-`define NO_RW_RAM_CHECK (* rw_addr_collision = "no" *)
-`define DISABLE_BRAM    (* ram_style = "registers" *)
-`define PRESERVE_NET    (* keep = "true" *)
+`define PRESERVE_REG    (* preserve *)
 `else
-`define MAX_FANOUT      4
-`define IF_DATA_SIZE(x) x.DATA_WIDTH
 `define USE_FAST_BRAM
 `define NO_RW_RAM_CHECK
 `define DISABLE_BRAM
-`define PRESERVE_NET
+`define PRESERVE_REG
 `endif
 
 ///////////////////////////////////////////////////////////////////////////////
@@ -275,105 +112,52 @@
 `define LOG2UP(x)   (((x) > 1) ? $clog2(x) : 1)
 `define ISPOW2(x)   (((x) != 0) && (0 == ((x) & ((x) - 1))))
 
-`define ABS(x)      (((x) < 0) ? (-(x)) : (x));
+`define ABS(x)      (($signed(x) < 0) ? (-$signed(x)) : (x));
 
-`ifndef MIN
 `define MIN(x, y)   (((x) < (y)) ? (x) : (y))
-`endif
-
-`ifndef MAX
 `define MAX(x, y)   (((x) > (y)) ? (x) : (y))
-`endif
 
-`ifndef CLAMP
-`define CLAMP(x, lo, hi)   (((x) > (hi)) ? (hi) : (((x) < (lo)) ? (lo) : (x)))
-`endif
-
-`ifndef UP
-`define UP(x)       (((x) != 0) ? (x) : 1)
-`endif
+`define UP(x)       (((x) > 0) ? (x) : 1)
 
 `define RTRIM(x, s) x[$bits(x)-1:($bits(x)-s)]
 
 `define LTRIM(x, s) x[s-1:0]
 
-`define TRACE_ARRAY1D(lvl, arr, m)              \
-    `TRACE(lvl, ("{"));                         \
-    for (integer __i = (m-1); __i >= 0; --__i) begin  \
-        if (__i != (m-1)) `TRACE(lvl, (", "));    \
-        `TRACE(lvl, ("0x%0h", arr[__i]));         \
+`define TRACE_ARRAY1D(a, m)                     \
+    dpi_trace("{");                             \
+    for (integer i = (m-1); i >= 0; --i) begin  \
+        if (i != (m-1)) dpi_trace(", ");        \
+        dpi_trace("0x%0h", a[i]);               \
     end                                         \
-    `TRACE(lvl, ("}"));
+    dpi_trace("}");                             \
 
-`define TRACE_ARRAY2D(lvl, arr, m, n)           \
-    `TRACE(lvl, ("{"));                         \
-    for (integer __i = n-1; __i >= 0; --__i) begin    \
-        if (__i != (n-1)) `TRACE(lvl, (", "));    \
-        `TRACE(lvl, ("{"));                     \
-        for (integer __j = (m-1); __j >= 0; --__j) begin \
-            if (__j != (m-1)) `TRACE(lvl, (", "));\
-            `TRACE(lvl, ("0x%0h", arr[__i][__j]));  \
+`define TRACE_ARRAY2D(a, m, n)                  \
+    dpi_trace("{");                             \
+    for (integer i = n-1; i >= 0; --i) begin    \
+        if (i != (n-1)) dpi_trace(", ");        \
+        dpi_trace("{");                         \
+        for (integer j = (m-1); j >= 0; --j) begin \
+            if (j != (m-1)) dpi_trace(", ");    \
+            dpi_trace("0x%0h", a[i][j]);        \
         end                                     \
-        `TRACE(lvl, ("}"));                     \
+        dpi_trace("}");                         \
     end                                         \
-    `TRACE(lvl, ("}"))
+    dpi_trace("}")
 
-`define RESET_RELAY_EX(dst, src, size, fanout)  \
-    wire [size-1:0] dst;                        \
-    VX_reset_relay #(.N(size), .MAX_FANOUT(fanout)) __``dst ( \
-        .clk     (clk),                         \
-        .reset   (src),                         \
-        .reset_o (dst)                          \
+`define RESET_RELAY(signal)     \
+    wire signal;                \
+    VX_reset_relay __``signal ( \
+        .clk     (clk),         \
+        .reset   (reset),       \
+        .reset_o (signal)       \
     )
 
-`define RESET_RELAY_EN(dst, src, enable) \
-    `RESET_RELAY_EX (dst, src, 1, ((enable) ? 0 : -1))
+`define POP_COUNT(out, in)  \
+    VX_popcount #(          \
+        .N ($bits(in))      \
+    ) __``out (             \
+        .in_i  (in),        \
+        .cnt_o (out)        \
+    )
 
-`define RESET_RELAY(dst, src) \
-    `RESET_RELAY_EX (dst, src, 1, 0)
-
-// size(x): 0 -> 0, 1 -> 1, 2 -> 2, 3 -> 2, 4-> 2
-`define OUT_REG_TO_EB_SIZE(out_reg)   `MIN(out_reg, 2)
-
-// reg(x): 0 -> 0, 1 -> 1, 2 -> 0, 3 -> 1, 4 -> 2
-`define OUT_REG_TO_EB_REG(out_reg)    ((out_reg & 1) + ((out_reg >> 2) << 1))
-
-`define REPEAT(n,f,s)   `_REPEAT_``n(f,s)
-`define _REPEAT_0(f,s)
-`define _REPEAT_1(f,s)  `f(0)
-`define _REPEAT_2(f,s)  `f(1)  `s `_REPEAT_1(f,s)
-`define _REPEAT_3(f,s)  `f(2)  `s `_REPEAT_2(f,s)
-`define _REPEAT_4(f,s)  `f(3)  `s `_REPEAT_3(f,s)
-`define _REPEAT_5(f,s)  `f(4)  `s `_REPEAT_4(f,s)
-`define _REPEAT_6(f,s)  `f(5)  `s `_REPEAT_5(f,s)
-`define _REPEAT_7(f,s)  `f(6)  `s `_REPEAT_6(f,s)
-`define _REPEAT_8(f,s)  `f(7)  `s `_REPEAT_7(f,s)
-`define _REPEAT_9(f,s)  `f(8)  `s `_REPEAT_8(f,s)
-`define _REPEAT_10(f,s) `f(9)  `s `_REPEAT_9(f,s)
-`define _REPEAT_11(f,s) `f(10) `s `_REPEAT_10(f,s)
-`define _REPEAT_12(f,s) `f(11) `s `_REPEAT_11(f,s)
-`define _REPEAT_13(f,s) `f(12) `s `_REPEAT_12(f,s)
-`define _REPEAT_14(f,s) `f(13) `s `_REPEAT_13(f,s)
-`define _REPEAT_15(f,s) `f(14) `s `_REPEAT_14(f,s)
-`define _REPEAT_16(f,s) `f(15) `s `_REPEAT_15(f,s)
-`define _REPEAT_17(f,s) `f(16) `s `_REPEAT_16(f,s)
-`define _REPEAT_18(f,s) `f(17) `s `_REPEAT_17(f,s)
-`define _REPEAT_19(f,s) `f(18) `s `_REPEAT_18(f,s)
-`define _REPEAT_20(f,s) `f(19) `s `_REPEAT_19(f,s)
-`define _REPEAT_21(f,s) `f(20) `s `_REPEAT_20(f,s)
-`define _REPEAT_22(f,s) `f(21) `s `_REPEAT_21(f,s)
-`define _REPEAT_23(f,s) `f(22) `s `_REPEAT_22(f,s)
-`define _REPEAT_24(f,s) `f(23) `s `_REPEAT_23(f,s)
-`define _REPEAT_25(f,s) `f(24) `s `_REPEAT_24(f,s)
-`define _REPEAT_26(f,s) `f(25) `s `_REPEAT_25(f,s)
-`define _REPEAT_27(f,s) `f(26) `s `_REPEAT_26(f,s)
-`define _REPEAT_28(f,s) `f(27) `s `_REPEAT_27(f,s)
-`define _REPEAT_29(f,s) `f(28) `s `_REPEAT_28(f,s)
-`define _REPEAT_30(f,s) `f(29) `s `_REPEAT_29(f,s)
-`define _REPEAT_31(f,s) `f(30) `s `_REPEAT_30(f,s)
-`define _REPEAT_32(f,s) `f(31) `s `_REPEAT_31(f,s)
-
-`define REPEAT_COMMA ,
-`define REPEAT_SEMICOLON ;
-
-`endif // VX_PLATFORM_VH
+`endif

hw/rtl/VX_scope.vh

@@ -1,68 +1,89 @@
-// Copyright © 2019-2023
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// 
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-`ifndef VX_SCOPE_VH
-`define VX_SCOPE_VH
+`ifndef VX_SCOPE
+`define VX_SCOPE
 
 `ifdef SCOPE
 
-`define SCOPE_IO_DECL \
-    input wire scope_reset, \
-    input wire scope_bus_in, \
-    output wire scope_bus_out,
+`include "scope-defs.vh"
 
-`define SCOPE_IO_SWITCH(__count) \
-    wire scope_bus_in_w [__count]; \
-    wire scope_bus_out_w [__count]; \
-    `RESET_RELAY_EX(scope_reset_w, scope_reset, __count, 4); \
-    VX_scope_switch #( \
-        .N (__count) \
-    ) scope_switch ( \
-        .clk (clk), \
-        .reset (scope_reset), \
-        .req_in (scope_bus_in), \
-        .rsp_out (scope_bus_out), \
-        .req_out (scope_bus_in_w), \
-        .rsp_in (scope_bus_out_w) \
-    );
+`define SCOPE_ASSIGN(d,s) assign scope_``d = s
 
-`define SCOPE_IO_BIND(__i) \
-    .scope_reset (scope_reset_w[__i]), \
-    .scope_bus_in (scope_bus_in_w[__i]), \
-    .scope_bus_out (scope_bus_out_w[__i]),
-
-`define SCOPE_IO_UNUSED() \
-    `UNUSED_VAR (scope_reset); \
-    `UNUSED_VAR (scope_bus_in); \
-    assign scope_bus_out = 0;
-
-`define SCOPE_IO_UNUSED_W(__i) \
-    `UNUSED_VAR (scope_reset_w[__i]); \
-    `UNUSED_VAR (scope_bus_in_w[__i]); \
-    assign scope_bus_out_w[__i] = 0;
+`define SCOPE_SIZE 1024
 
 `else
 
-`define SCOPE_IO_DECL
+`define SCOPE_IO_VX_icache_stage
 
-`define SCOPE_IO_SWITCH(__count)
+`define SCOPE_IO_VX_fetch
 
-`define SCOPE_IO_BIND(__i)
+`define SCOPE_BIND_VX_fetch_icache_stage
 
-`define SCOPE_IO_UNUSED_W(__i)
+`define SCOPE_BIND_VX_fetch_warp_sched
 
-`define SCOPE_IO_UNUSED(__i)
+`define SCOPE_IO_VX_warp_sched
+
+`define SCOPE_IO_VX_pipeline
+
+`define SCOPE_BIND_VX_pipeline_fetch
+
+`define SCOPE_IO_VX_core
+
+`define SCOPE_BIND_VX_core_pipeline
+
+`define SCOPE_IO_VX_cluster
+
+`define SCOPE_BIND_VX_cluster_core(__i__)
+
+`define SCOPE_IO_Vortex
+
+`define SCOPE_BIND_Vortex_cluster(__i__)
+
+`define SCOPE_BIND_afu_vortex
+
+`define SCOPE_IO_VX_lsu_unit
+
+`define SCOPE_IO_VX_gpu_unit
+
+`define SCOPE_IO_VX_execute
+
+`define SCOPE_BIND_VX_execute_lsu_unit
+
+`define SCOPE_BIND_VX_execute_gpu_unit
+
+`define SCOPE_BIND_VX_pipeline_execute
+
+`define SCOPE_IO_VX_issue
+
+`define SCOPE_BIND_VX_pipeline_issue
+
+`define SCOPE_IO_VX_bank
+
+`define SCOPE_IO_VX_cache
+
+`define SCOPE_BIND_VX_cache_bank(__i__)
+
+`define SCOPE_BIND_Vortex_l3cache
+
+`define SCOPE_BIND_VX_cluster_l2cache
+
+`define SCOPE_IO_VX_mem_unit
+
+`define SCOPE_BIND_VX_mem_unit_dcache
+
+`define SCOPE_BIND_VX_core_mem_unit
+
+`define SCOPE_BIND_VX_mem_unit_icache
+
+`define SCOPE_BIND_VX_mem_unit_smem
+
+`define SCOPE_DECL_SIGNALS
+
+`define SCOPE_DATA_LIST
+
+`define SCOPE_UPDATE_LIST
+
+`define SCOPE_TRIGGER
+
+`define SCOPE_ASSIGN(d,s)
 
 `endif
-
-`endif // VX_SCOPE_VH
+`endif

hw/rtl/VX_scoreboard.sv

@@ -0,0 +1,85 @@
+`include "VX_define.vh"
+
+module VX_scoreboard  #(
+    parameter CORE_ID = 0
+) (
+    input wire clk,
+    input wire reset,
+
+    VX_ibuffer_if.slave ibuffer_if,
+    VX_writeback_if.slave writeback_if
+);
+    reg [`NUM_WARPS-1:0][`NUM_REGS-1:0] inuse_regs, inuse_regs_n;
+
+    wire reserve_reg = ibuffer_if.valid && ibuffer_if.ready && ibuffer_if.wb;
+
+    wire release_reg = writeback_if.valid && writeback_if.ready && writeback_if.eop;
+    
+    always @(*) begin
+        inuse_regs_n = inuse_regs;
+        if (reserve_reg) begin
+            inuse_regs_n[ibuffer_if.wid][ibuffer_if.rd] = 1;                
+        end       
+        if (release_reg) begin
+            inuse_regs_n[writeback_if.wid][writeback_if.rd] = 0;
+        end
+    end
+
+    always @(posedge clk) begin
+        if (reset) begin
+            inuse_regs <= '0;
+        end else begin
+            inuse_regs <= inuse_regs_n;
+        end
+    end
+    
+    reg deq_inuse_rd, deq_inuse_rs1, deq_inuse_rs2, deq_inuse_rs3;
+
+    always @(posedge clk) begin
+        deq_inuse_rd  <= inuse_regs_n[ibuffer_if.wid_n][ibuffer_if.rd_n];
+        deq_inuse_rs1 <= inuse_regs_n[ibuffer_if.wid_n][ibuffer_if.rs1_n];
+        deq_inuse_rs2 <= inuse_regs_n[ibuffer_if.wid_n][ibuffer_if.rs2_n];
+        deq_inuse_rs3 <= inuse_regs_n[ibuffer_if.wid_n][ibuffer_if.rs3_n];
+    end
+
+    assign writeback_if.ready = 1'b1;
+
+    assign ibuffer_if.ready = ~(deq_inuse_rd 
+                             | deq_inuse_rs1 
+                             | deq_inuse_rs2 
+                             | deq_inuse_rs3);
+
+    `UNUSED_VAR (writeback_if.PC)
+
+    reg [31:0] deadlock_ctr;
+    wire [31:0] deadlock_timeout = 10000 * (1 ** (`L2_ENABLE + `L3_ENABLE));
+
+    always @(posedge clk) begin
+        if (reset) begin
+            deadlock_ctr <= 0;
+        end else begin
+        `ifdef DBG_TRACE_CORE_PIPELINE
+            if (ibuffer_if.valid && ~ibuffer_if.ready) begin
+                dpi_trace("%d: *** core%0d-stall: wid=%0d, PC=%0h, rd=%0d, wb=%0d, inuse=%b%b%b%b (#%0d)\n", 
+                    $time, CORE_ID, ibuffer_if.wid, ibuffer_if.PC, ibuffer_if.rd, ibuffer_if.wb, 
+                    deq_inuse_rd, deq_inuse_rs1, deq_inuse_rs2, deq_inuse_rs3, ibuffer_if.uuid);
+            end
+        `endif
+            if (release_reg) begin
+                `ASSERT(inuse_regs[writeback_if.wid][writeback_if.rd] != 0,
+                    ("%t: *** core%0d: invalid writeback register: wid=%0d, PC=%0h, rd=%0d (#%0d)",
+                                $time, CORE_ID, writeback_if.wid, writeback_if.PC, writeback_if.rd,writeback_if.uuid));
+            end
+            if (ibuffer_if.valid && ~ibuffer_if.ready) begin            
+                deadlock_ctr <= deadlock_ctr + 1;
+                `ASSERT(deadlock_ctr < deadlock_timeout,
+                    ("%t: *** core%0d-deadlock: wid=%0d, PC=%0h, rd=%0d, wb=%0d, inuse=%b%b%b%b (#%0d)",
+                        $time, CORE_ID, ibuffer_if.wid, ibuffer_if.PC, ibuffer_if.rd, ibuffer_if.wb, 
+                        deq_inuse_rd, deq_inuse_rs1, deq_inuse_rs2, deq_inuse_rs3, ibuffer_if.uuid));
+            end else if (ibuffer_if.valid && ibuffer_if.ready) begin
+                deadlock_ctr <= 0;
+            end
+        end
+    end
+
+endmodule