diff --git a/.github/scripts/check-commit.sh b/.github/scripts/check-commit.sh
index 570a6ee5..e906ac29 100755
--- a/.github/scripts/check-commit.sh
+++ b/.github/scripts/check-commit.sh
@@ -45,7 +45,8 @@ search () {
     done
 }
 
-submodules=("cva6" "boom" "ibex" "gemmini" "icenet" "nvdla" "rocket-chip" "rocket-chip-blocks" "rocket-chip-inclusive-cache" "testchipip" "riscv-sodor" "mempress" "bar-fetchers" "shuttle" "constellation" "fft-generator" "hardfloat" "caliptra-aes-acc" "rocc-acc-utils" "diplomacy")
+
+submodules=("cva6" "boom" "ibex" "gemmini" "icenet" "nvdla" "rocket-chip" "rocket-chip-blocks" "rocket-chip-inclusive-cache" "testchipip" "riscv-sodor" "mempress" "bar-fetchers" "shuttle" "constellation" "fft-generator" "hardfloat" "caliptra-aes-acc" "rocc-acc-utils" "diplomacy" "rerocc" "compress-acc")
 dir="generators"
 branches=("master" "main" "dev")
 search
diff --git a/.github/scripts/defaults.sh b/.github/scripts/defaults.sh
index dff549b9..ef48615c 100755
--- a/.github/scripts/defaults.sh
+++ b/.github/scripts/defaults.sh
@@ -30,7 +30,7 @@ REMOTE_COURSIER_CACHE=$REMOTE_WORK_DIR/.coursier-cache
 declare -A grouping
 grouping["group-cores"]="chipyard-cva6 chipyard-ibex chipyard-rocket chipyard-hetero chipyard-boomv3 chipyard-boomv4 chipyard-sodor chipyard-digitaltop chipyard-multiclock-rocket chipyard-nomem-scratchpad chipyard-spike chipyard-clone chipyard-prefetchers chipyard-shuttle"
 grouping["group-peripherals"]="chipyard-dmirocket chipyard-dmiboomv3 chipyard-dmiboomv4 chipyard-spiflashwrite chipyard-mmios chipyard-nocores chipyard-manyperipherals chipyard-chiplike chipyard-tethered chipyard-symmetric chipyard-llcchiplet"
-grouping["group-accels"]="chipyard-mempress chipyard-gemmini chipyard-manymmioaccels chipyard-nvdla chipyard-aes256ecb"
+grouping["group-accels"]="chipyard-compressacc chipyard-mempress chipyard-gemmini chipyard-manymmioaccels chipyard-nvdla chipyard-aes256ecb chipyard-rerocc"
 grouping["group-constellation"]="chipyard-constellation"
 grouping["group-gpu"]="chipyard-radiance chipyard-radiance-fuzzer chipyard-coalescer coalescer"
 grouping["group-tracegen"]="tracegen tracegen-boomv3 tracegen-boomv4"
@@ -42,6 +42,7 @@ declare -A mapping
 mapping["chipyard-rocket"]=" CONFIG=QuadChannelRocketConfig"
 mapping["chipyard-dmirocket"]=" CONFIG=dmiRocketConfig"
 mapping["chipyard-mempress"]=" CONFIG=MempressRocketConfig"
+mapping["chipyard-compressacc"]=" CONFIG=ZstdCompressorRocketConfig"
 mapping["chipyard-prefetchers"]=" CONFIG=PrefetchingRocketConfig"
 mapping["chipyard-digitaltop"]=" TOP=DigitalTop"
 mapping["chipyard-manymmioaccels"]=" CONFIG=ManyMMIOAcceleratorRocketConfig"
@@ -73,6 +74,7 @@ mapping["chipyard-nomem-scratchpad"]=" CONFIG=MMIOScratchpadOnlyRocketConfig"
 mapping["chipyard-constellation"]=" CONFIG=SharedNoCConfig"
 mapping["chipyard-clusters"]=" CONFIG=ClusteredRocketConfig verilog"
 mapping["chipyard-aes256ecb"]=" CONFIG=AES256ECBRocketConfig"
+mapping["chipyard-rerocc"]=" CONFIG=ReRoCCTestConfig"
 mapping["chipyard-radiance"]=" CONFIG=RadianceConfig"
 mapping["chipyard-radiance-fuzzer"]=" CONFIG=RadianceFuzzerConfig"
 mapping["chipyard-coalescer"]=" CONFIG=MemtraceCoreConfig"
diff --git a/.github/scripts/run-tests.sh b/.github/scripts/run-tests.sh
index ce4de3e5..d0804ccf 100755
--- a/.github/scripts/run-tests.sh
+++ b/.github/scripts/run-tests.sh
@@ -72,7 +72,7 @@ case $1 in
         run_bmark
         ;;
     chipyard-prefetchers)
-        run_binary BINARY=$RISCV/riscv64-unknown-elf/share/riscv-tests/benchmarks/dhrystone.riscv
+        run_binary BINARY=$RISCV/riscv64-unknown-elf/share/riscv-tests/benchmarks/dhrystone.riscv LOADMEM=1
         ;;
     rocketchip)
         run_bmark
@@ -81,13 +81,17 @@ case $1 in
         GEMMINI_SOFTWARE_DIR=$LOCAL_SIM_DIR/../../generators/gemmini/software/gemmini-rocc-tests
         rm -rf $GEMMINI_SOFTWARE_DIR/riscv-tests
         cd $LOCAL_SIM_DIR
-        run_binary BINARY=$GEMMINI_SOFTWARE_DIR/build/bareMetalC/aligned-baremetal
-        run_binary BINARY=$GEMMINI_SOFTWARE_DIR/build/bareMetalC/raw_hazard-baremetal
-        run_binary BINARY=$GEMMINI_SOFTWARE_DIR/build/bareMetalC/mvin_mvout-baremetal
+        run_binary BINARY=$GEMMINI_SOFTWARE_DIR/build/bareMetalC/aligned-baremetal LOADMEM=1
+        run_binary BINARY=$GEMMINI_SOFTWARE_DIR/build/bareMetalC/raw_hazard-baremetal LOADMEM=1
+        run_binary BINARY=$GEMMINI_SOFTWARE_DIR/build/bareMetalC/mvin_mvout-baremetal LOADMEM=1
         ;;
     chipyard-mempress)
         (cd $LOCAL_CHIPYARD_DIR/generators/mempress/software/src && make)
-        run_binary BINARY=$LOCAL_CHIPYARD_DIR/generators/mempress/software/src/mempress-rocc.riscv
+        run_binary BINARY=$LOCAL_CHIPYARD_DIR/generators/mempress/software/src/mempress-rocc.riscv LOADMEM=1
+        ;;
+    chipyard-compressacc)
+        (cd $LOCAL_CHIPYARD_DIR/generators/compress-acc/software-zstd/compress && ./build-hcb-single-file.sh)
+        run_binary BINARY=$LOCAL_CHIPYARD_DIR/generators/compress-acc/software-zstd/compress/009987_cl0_ws12.riscv LOADMEM=1
         ;;
     chipyard-manymmioaccels)
 	make -C $LOCAL_CHIPYARD_DIR/tests
@@ -127,6 +131,10 @@ case $1 in
         make -C $LOCAL_CHIPYARD_DIR/tests
         run_binary BINARY=$LOCAL_CHIPYARD_DIR/tests/hello.riscv LOADMEM=1
         ;;
+    chipyard-rerocc)
+        make -C $LOCAL_CHIPYARD_DIR/generators/rerocc/tests
+        run_binary BINARY=$LOCAL_CHIPYARD_DIR/generators/rerocc/tests/rerocc.riscv LOADMEM=1
+        ;;
     tracegen)
         run_tracegen
         ;;
diff --git a/.github/workflows/chipyard-run-tests.yml b/.github/workflows/chipyard-run-tests.yml
index 0b644888..d86c75c9 100644
--- a/.github/workflows/chipyard-run-tests.yml
+++ b/.github/workflows/chipyard-run-tests.yml
@@ -777,6 +777,29 @@ jobs:
           group-key: "group-peripherals"
           project-key: "chipyard-llcchiplet"
 
+  chipyard-rerocc-run-tests:
+    name: chipyard-rerocc-run-tests
+    needs: prepare-chipyard-accels
+    runs-on: as4
+    steps:
+      - name: Delete old checkout
+        run: |
+            ls -alh .
+            rm -rf ${{ github.workspace }}/* || true
+            rm -rf ${{ github.workspace }}/.* || true
+            ls -alh .
+      - name: Checkout
+        uses: actions/checkout@v3
+      - name: Git workaround
+        uses: ./.github/actions/git-workaround
+      - name: Create conda env
+        uses: ./.github/actions/create-conda-env
+      - name: Run tests
+        uses: ./.github/actions/run-tests
+        with:
+          group-key: "group-accels"
+          project-key: "chipyard-rerocc"
+
   chipyard-gemmini-run-tests:
     name: chipyard-gemmini-run-tests
     needs: prepare-chipyard-accels
@@ -869,6 +892,29 @@ jobs:
           group-key: "group-accels"
           project-key: "chipyard-mempress"
 
+  chipyard-compressacc-run-tests:
+    name: chipyard-compressacc-run-tests
+    needs: prepare-chipyard-accels
+    runs-on: as4
+    steps:
+      - name: Delete old checkout
+        run: |
+            ls -alh .
+            rm -rf ${{ github.workspace }}/* || true
+            rm -rf ${{ github.workspace }}/.* || true
+            ls -alh .
+      - name: Checkout
+        uses: actions/checkout@v3
+      - name: Git workaround
+        uses: ./.github/actions/git-workaround
+      - name: Create conda env
+        uses: ./.github/actions/create-conda-env
+      - name: Run tests
+        uses: ./.github/actions/run-tests
+        with:
+          group-key: "group-accels"
+          project-key: "chipyard-compressacc"
+
 
   tracegen-boomv3-run-tests:
     name: tracegen-boomv3-run-tests
@@ -1143,10 +1189,12 @@ jobs:
             chipyard-tethered-run-tests,
             chipyard-symmetric-run-tests,
             chipyard-llcchiplet-run-tests,
+            chipyard-rerocc-run-tests,
             chipyard-gemmini-run-tests,
             chipyard-manymmioaccels-run-tests, # chipyard-nvdla-run-tests,
             chipyard-prefetchers-run-tests,
             chipyard-mempress-run-tests,
+            chipyard-compressacc-run-tests,
             chipyard-constellation-run-tests,
             tracegen-boomv3-run-tests,
             tracegen-boomv4-run-tests,
diff --git a/.gitmodules b/.gitmodules
index 709b4865..afb2ca01 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -151,7 +151,12 @@
 [submodule "tools/midas-targetutils"]
 	path = tools/midas-targetutils
 	url = https://github.com/firesim/midas-targetutils.git
-
+[submodule "generators/rerocc"]
+	path = generators/rerocc
+	url = https://github.com/ucb-bar/rerocc.git
+[submodule "generators/compress-acc"]
+	path = generators/compress-acc
+  url = https://github.com/ucb-bar/compress-acc.git
 [submodule "generators/radiance"]
 	path = generators/radiance
 	url = https://github.com/ucb-bar/radiance.git
diff --git a/build.sbt b/build.sbt
index 1bbf8f11..f510fd90 100644
--- a/build.sbt
+++ b/build.sbt
@@ -98,8 +98,8 @@ lazy val chisel6Settings = Seq(
   addCompilerPlugin("org.chipsalliance" % "chisel-plugin" % "6.0.0" cross CrossVersion.full)
 )
 lazy val chisel3Settings = Seq(
-  libraryDependencies ++= Seq("edu.berkeley.cs" %% "chisel3" % "3.6.0"),
-  addCompilerPlugin("edu.berkeley.cs" % "chisel3-plugin" % "3.6.0" cross CrossVersion.full)
+  libraryDependencies ++= Seq("edu.berkeley.cs" %% "chisel3" % "3.6.1"),
+  addCompilerPlugin("edu.berkeley.cs" % "chisel3-plugin" % "3.6.1" cross CrossVersion.full)
 )
 
 lazy val chiselSettings = (if (chisel6) chisel6Settings else chisel3Settings) ++ Seq(
@@ -172,8 +172,9 @@ val stageDir = if (chisel6) "tools/stage/src/main/scala" else "tools/stage-chise
 lazy val chipyard = (project in file("generators/chipyard"))
   .dependsOn(testchipip, rocketchip, boom, rocketchip_blocks, rocketchip_inclusive_cache,
     dsptools, rocket_dsp_utils,
-    radiance, gemmini, icenet, tracegen, cva6, nvdla, radiance, sodor, ibex, fft_generator,
-    constellation, mempress, barf, shuttle, caliptra_aes)
+    radiance, gemmini, icenet, tracegen, cva6, nvdla, sodor, ibex, fft_generator,
+    constellation, mempress, barf, shuttle, caliptra_aes, rerocc,
+    compressacc)
   .settings(libraryDependencies ++= rocketLibDeps.value)
   .settings(
     libraryDependencies ++= Seq(
@@ -183,6 +184,11 @@ lazy val chipyard = (project in file("generators/chipyard"))
   .settings(commonSettings)
   .settings(Compile / unmanagedSourceDirectories += file(stageDir))
 
+lazy val compressacc = (project in file("generators/compress-acc"))
+  .dependsOn(rocketchip)
+  .settings(libraryDependencies ++= rocketLibDeps.value)
+  .settings(commonSettings)
+
 lazy val mempress = (project in file("generators/mempress"))
   .dependsOn(rocketchip)
   .settings(libraryDependencies ++= rocketLibDeps.value)
@@ -264,6 +270,11 @@ lazy val caliptra_aes = (project in file("generators/caliptra-aes-acc"))
   .settings(libraryDependencies ++= rocketLibDeps.value)
   .settings(commonSettings)
 
+lazy val rerocc = (project in file("generators/rerocc"))
+  .dependsOn(rocketchip, constellation, boom)
+  .settings(libraryDependencies ++= rocketLibDeps.value)
+  .settings(commonSettings)
+
 lazy val rocc_acc_utils = (project in file("generators/rocc-acc-utils"))
   .dependsOn(rocketchip)
   .settings(libraryDependencies ++= rocketLibDeps.value)
@@ -342,6 +353,6 @@ lazy val fpga_shells = (project in file("./fpga/fpga-shells"))
   .settings(libraryDependencies ++= rocketLibDeps.value)
   .settings(commonSettings)
 
-lazy val fpga_platforms = (project in file("./fpga"))
+lazy val chipyard_fpga = (project in file("./fpga"))
   .dependsOn(chipyard, fpga_shells)
   .settings(commonSettings)
diff --git a/common.mk b/common.mk
index e620f31a..51583acb 100644
--- a/common.mk
+++ b/common.mk
@@ -96,19 +96,9 @@ TAPEOUT_SOURCE_DIRS = $(addprefix $(base_dir)/,tools/tapeout)
 TAPEOUT_SCALA_SOURCES = $(call lookup_srcs_by_multiple_type,$(TAPEOUT_SOURCE_DIRS),$(SCALA_EXT))
 TAPEOUT_VLOG_SOURCES = $(call lookup_srcs_by_multiple_type,$(TAPEOUT_SOURCE_DIRS),$(VLOG_EXT))
 # This assumes no SBT meta-build sources
-SBT_SOURCE_DIRS = $(addprefix $(base_dir)/,generators sims/firesim/sim tools)
+SBT_SOURCE_DIRS = $(addprefix $(base_dir)/,generators tools)
 SBT_SOURCES = $(call lookup_srcs,$(SBT_SOURCE_DIRS),sbt) $(base_dir)/build.sbt $(base_dir)/project/plugins.sbt $(base_dir)/project/build.properties
 
-#########################################################################################
-# SBT Server Setup (start server / rebuild proj. defs. if SBT_SOURCES change)
-#########################################################################################
-$(SBT_THIN_CLIENT_TIMESTAMP): $(SBT_SOURCES)
-ifneq (,$(wildcard $(SBT_THIN_CLIENT_TIMESTAMP)))
-	cd $(base_dir) && $(SBT) "reload"
-	touch $@
-else
-	cd $(base_dir) && $(SBT) "exit"
-endif
 
 #########################################################################################
 # copy over bootrom files
@@ -122,13 +112,13 @@ $(BOOTROM_TARGETS): $(build_dir)/bootrom.%.img: $(TESTCHIP_RSRCS_DIR)/testchipip
 #########################################################################################
 # compile scala jars
 #########################################################################################
-$(CHIPYARD_CLASSPATH_TARGETS) &: $(CHIPYARD_SCALA_SOURCES) $(SCALA_BUILDTOOL_DEPS) $(CHIPYARD_VLOG_SOURCES)
+$(GENERATOR_CLASSPATH) &: $(CHIPYARD_SCALA_SOURCES) $(SCALA_BUILDTOOL_DEPS) $(CHIPYARD_VLOG_SOURCES)
 	$(CHECK_SUBMODULES_COMMAND)
 	mkdir -p $(dir $@)
-	$(call run_sbt_assembly,$(SBT_PROJECT),$(CHIPYARD_CLASSPATH))
+	$(call run_sbt_assembly,$(SBT_PROJECT),$(GENERATOR_CLASSPATH))
 
 # order only dependency between sbt runs needed to avoid concurrent sbt runs
-$(TAPEOUT_CLASSPATH_TARGETS) &: $(TAPEOUT_SCALA_SOURCES) $(SCALA_BUILDTOOL_DEPS) $(TAPEOUT_VLOG_SOURCES) | $(CHIPYARD_CLASSPATH_TARGETS)
+$(TAPEOUT_CLASSPATH) &: $(TAPEOUT_SCALA_SOURCES) $(SCALA_BUILDTOOL_DEPS) $(TAPEOUT_VLOG_SOURCES) | $(GENERATOR_CLASSPATH)
 	mkdir -p $(dir $@)
 	$(call run_sbt_assembly,tapeout,$(TAPEOUT_CLASSPATH))
 
@@ -136,9 +126,9 @@ $(TAPEOUT_CLASSPATH_TARGETS) &: $(TAPEOUT_SCALA_SOURCES) $(SCALA_BUILDTOOL_DEPS)
 # verilog generation pipeline
 #########################################################################################
 # AG: must re-elaborate if cva6 sources have changed... otherwise just run firrtl compile
-$(FIRRTL_FILE) $(ANNO_FILE) $(CHISEL_LOG_FILE) &: $(CHIPYARD_CLASSPATH_TARGETS) $(EXTRA_GENERATOR_REQS)
+$(FIRRTL_FILE) $(ANNO_FILE) $(CHISEL_LOG_FILE) &: $(GENERATOR_CLASSPATH) $(EXTRA_GENERATOR_REQS)
 	mkdir -p $(build_dir)
-	(set -o pipefail && $(call run_jar_scala_main,$(CHIPYARD_CLASSPATH),$(GENERATOR_PACKAGE).Generator,\
+	(set -o pipefail && $(call run_jar_scala_main,$(GENERATOR_CLASSPATH),$(GENERATOR_PACKAGE).Generator,\
 		--target-dir $(build_dir) \
 		--name $(long_name) \
 		--top-module $(MODEL_PACKAGE).$(MODEL) \
@@ -182,7 +172,8 @@ SFC_MFC_TARGETS = \
 	$(MFC_MODEL_SMEMS_JSON) \
 	$(MFC_FILELIST) \
 	$(MFC_BB_MODS_FILELIST) \
-	$(GEN_COLLATERAL_DIR)
+	$(GEN_COLLATERAL_DIR) \
+	$(FIRTOOL_LOG_FILE)
 
 MFC_BASE_LOWERING_OPTIONS ?= emittedLineLength=2048,noAlwaysComb,disallowLocalVariables,verifLabels,disallowPortDeclSharing,locationInfoStyle=wrapInAtSquareBracket
 
@@ -197,7 +188,7 @@ endif
 
 $(SFC_MFC_TARGETS) &: $(FIRRTL_FILE) $(FINAL_ANNO_FILE) $(MFC_LOWERING_OPTIONS)
 	rm -rf $(GEN_COLLATERAL_DIR)
-	firtool \
+	(set -o pipefail && firtool \
 		--format=fir \
 		--export-module-hierarchy \
 		--verify-each=true \
@@ -211,7 +202,7 @@ $(SFC_MFC_TARGETS) &: $(FIRRTL_FILE) $(FINAL_ANNO_FILE) $(MFC_LOWERING_OPTIONS)
 		--annotation-file=$(FINAL_ANNO_FILE) \
 		--split-verilog \
 		-o $(GEN_COLLATERAL_DIR) \
-		$(FIRRTL_FILE)
+		$(FIRRTL_FILE) |& tee $(FIRTOOL_LOG_FILE))
 	$(SED) -i 's/.*/& /' $(MFC_SMEMS_CONF) # need trailing space for SFC macrocompiler
 	touch $(MFC_BB_MODS_FILELIST) # if there are no BB's then the file might not be generated, instead always generate it
 # DOC include end: FirrtlCompiler
@@ -221,6 +212,7 @@ $(TOP_MODS_FILELIST) $(MODEL_MODS_FILELIST) $(ALL_MODS_FILELIST) $(BB_MODS_FILEL
 		--model-hier-json $(MFC_MODEL_HRCHY_JSON) \
 		--top-hier-json $(MFC_TOP_HRCHY_JSON) \
 		--in-all-filelist $(MFC_FILELIST) \
+		--in-bb-filelist $(MFC_BB_MODS_FILELIST) \
 		--dut $(TOP) \
 		--model $(MODEL) \
 		--target-dir $(GEN_COLLATERAL_DIR) \
@@ -246,12 +238,12 @@ $(TOP_SMEMS_CONF) $(MODEL_SMEMS_CONF) &:  $(MFC_SMEMS_CONF) $(MFC_MODEL_HRCHY_JS
 
 # This file is for simulation only. VLSI flows should replace this file with one containing hard SRAMs
 TOP_MACROCOMPILER_MODE ?= --mode synflops
-$(TOP_SMEMS_FILE) $(TOP_SMEMS_FIR) &: $(TAPEOUT_CLASSPATH_TARGETS) $(TOP_SMEMS_CONF)
+$(TOP_SMEMS_FILE) $(TOP_SMEMS_FIR) &: $(TAPEOUT_CLASSPATH) $(TOP_SMEMS_CONF)
 	$(call run_jar_scala_main,$(TAPEOUT_CLASSPATH),tapeout.macros.MacroCompiler,-n $(TOP_SMEMS_CONF) -v $(TOP_SMEMS_FILE) -f $(TOP_SMEMS_FIR) $(TOP_MACROCOMPILER_MODE))
 	touch $(TOP_SMEMS_FILE) $(TOP_SMEMS_FIR)
 
 MODEL_MACROCOMPILER_MODE = --mode synflops
-$(MODEL_SMEMS_FILE) $(MODEL_SMEMS_FIR) &: $(TAPEOUT_CLASSPATH_TARGETS) $(MODEL_SMEMS_CONF)
+$(MODEL_SMEMS_FILE) $(MODEL_SMEMS_FIR) &: $(TAPEOUT_CLASSPATH) $(MODEL_SMEMS_CONF)
 	$(call run_jar_scala_main,$(TAPEOUT_CLASSPATH),tapeout.macros.MacroCompiler, -n $(MODEL_SMEMS_CONF) -v $(MODEL_SMEMS_FILE) -f $(MODEL_SMEMS_FIR) $(MODEL_MACROCOMPILER_MODE))
 	touch $(MODEL_SMEMS_FILE) $(MODEL_SMEMS_FIR)
 
diff --git a/conda-reqs/chipyard-extended.yaml b/conda-reqs/chipyard-extended.yaml
index 43ab7303..36e7019a 100644
--- a/conda-reqs/chipyard-extended.yaml
+++ b/conda-reqs/chipyard-extended.yaml
@@ -30,7 +30,7 @@ dependencies:
     - bison
     - flex
     - expat
-    - python>=3.9
+    - python>=3.9,<3.11
     - bc
     - unzip
     - patch
diff --git a/conda-reqs/circt.json b/conda-reqs/circt.json
index 1eb54a43..020fe373 100644
--- a/conda-reqs/circt.json
+++ b/conda-reqs/circt.json
@@ -1,3 +1,3 @@
 {
-  "version": "firtool-1.66.0"
+  "version": "firtool-1.75.0"
 }
diff --git a/docs/Generators/CompressAcc.rst b/docs/Generators/CompressAcc.rst
new file mode 100644
index 00000000..b18b4709
--- /dev/null
+++ b/docs/Generators/CompressAcc.rst
@@ -0,0 +1,6 @@
+CompressAcc
+====================================
+
+We have (de)compression accelerators for two major compression algorithms : Snappy(`Snappy Github <https://github.com/google/snappy>`__) and ZStd(`ZStd Github <https://github.com/facebook/zstd>`__).
+
+For more information, check out the CDPU paper: `CDPU: Co-designing Compression and Decompression Processing Units for Hyperscale Systems <https://dl.acm.org/doi/abs/10.1145/3579371.3589074>`__ .
diff --git a/docs/Generators/index.rst b/docs/Generators/index.rst
index 88d9ee2a..d55dce0c 100644
--- a/docs/Generators/index.rst
+++ b/docs/Generators/index.rst
@@ -33,4 +33,5 @@ so changes to the generators themselves will automatically be used when building
    Sodor
    Shuttle
    Mempress
+   CompressAcc
    Prefetchers
diff --git a/docs/Prototyping/General.rst b/docs/Prototyping/General.rst
index 051cbcf8..17ef4173 100644
--- a/docs/Prototyping/General.rst
+++ b/docs/Prototyping/General.rst
@@ -30,7 +30,7 @@ For example:
 
     # converts to
 
-    make SBT_PROJECT=fpga_platforms MODEL=VCU118FPGATestHarness VLOG_MODEL=VCU118FPGATestHarness MODEL_PACKAGE=chipyard.fpga.vcu118 CONFIG=RocketVCU118Config CONFIG_PACKAGE=chipyard.fpga.vcu118 GENERATOR_PACKAGE=chipyard TB=none TOP=ChipTop BOARD=vcu118 FPGA_BRAND=... bitstream
+    make SBT_PROJECT=chipyard_fpga MODEL=VCU118FPGATestHarness VLOG_MODEL=VCU118FPGATestHarness MODEL_PACKAGE=chipyard.fpga.vcu118 CONFIG=RocketVCU118Config CONFIG_PACKAGE=chipyard.fpga.vcu118 GENERATOR_PACKAGE=chipyard TB=none TOP=ChipTop BOARD=vcu118 FPGA_BRAND=... bitstream
 
 Some ``SUB_PROJECT`` defaults are already defined for use, including ``vcu118`` and ``arty``.
 These default ``SUB_PROJECT``'s setup the necessary test harnesses, packages, and more for the Chipyard make system.
diff --git a/fpga/Makefile b/fpga/Makefile
index 1abb7f02..23616710 100644
--- a/fpga/Makefile
+++ b/fpga/Makefile
@@ -17,7 +17,7 @@ sim_name := none
 SUB_PROJECT ?= vcu118
 
 ifeq ($(SUB_PROJECT),vc707)
-	SBT_PROJECT       ?= fpga_platforms
+	SBT_PROJECT       ?= chipyard_fpga
 	MODEL             ?= VC707FPGATestHarness
 	VLOG_MODEL        ?= VC707FPGATestHarness
 	MODEL_PACKAGE     ?= chipyard.fpga.vc707
@@ -31,7 +31,7 @@ ifeq ($(SUB_PROJECT),vc707)
 endif
 
 ifeq ($(SUB_PROJECT),vcu118)
-	SBT_PROJECT       ?= fpga_platforms
+	SBT_PROJECT       ?= chipyard_fpga
 	MODEL             ?= VCU118FPGATestHarness
 	VLOG_MODEL        ?= VCU118FPGATestHarness
 	MODEL_PACKAGE     ?= chipyard.fpga.vcu118
@@ -45,7 +45,7 @@ ifeq ($(SUB_PROJECT),vcu118)
 endif
 
 ifeq ($(SUB_PROJECT),nexysvideo)
-	SBT_PROJECT       ?= fpga_platforms
+	SBT_PROJECT       ?= chipyard_fpga
 	MODEL             ?= NexysVideoHarness
 	VLOG_MODEL        ?= NexysVideoHarness
 	MODEL_PACKAGE     ?= chipyard.fpga.nexysvideo
@@ -60,7 +60,7 @@ endif
 
 ifeq ($(SUB_PROJECT),arty35t)
 	# TODO: Fix with Arty
-	SBT_PROJECT       ?= fpga_platforms
+	SBT_PROJECT       ?= chipyard_fpga
 	MODEL             ?= Arty35THarness
 	VLOG_MODEL        ?= Arty35THarness
 	MODEL_PACKAGE     ?= chipyard.fpga.arty
@@ -74,7 +74,7 @@ ifeq ($(SUB_PROJECT),arty35t)
 endif
 ifeq ($(SUB_PROJECT),arty100t)
 	# TODO: Fix with Arty
-	SBT_PROJECT       ?= fpga_platforms
+	SBT_PROJECT       ?= chipyard_fpga
 	MODEL             ?= Arty100THarness
 	VLOG_MODEL        ?= Arty100THarness
 	MODEL_PACKAGE     ?= chipyard.fpga.arty100t
diff --git a/fpga/fpga-shells b/fpga/fpga-shells
index 474ad191..2d36b0ab 160000
--- a/fpga/fpga-shells
+++ b/fpga/fpga-shells
@@ -1 +1 @@
-Subproject commit 474ad19113b89ed5679695b269acdb011b9b871a
+Subproject commit 2d36b0ab430fe3dc4cf378f378be21b925379110
diff --git a/fpga/src/main/scala/vc707/Configs.scala b/fpga/src/main/scala/vc707/Configs.scala
index d9fc1322..5b766107 100644
--- a/fpga/src/main/scala/vc707/Configs.scala
+++ b/fpga/src/main/scala/vc707/Configs.scala
@@ -6,7 +6,8 @@ import org.chipsalliance.cde.config.{Config, Parameters}
 import freechips.rocketchip.subsystem.{SystemBusKey, PeripheryBusKey, ControlBusKey, ExtMem}
 import freechips.rocketchip.devices.debug.{DebugModuleKey, ExportDebug, JTAG}
 import freechips.rocketchip.devices.tilelink.{DevNullParams, BootROMLocated}
-import freechips.rocketchip.diplomacy.{DTSModel, DTSTimebase, RegionType, AddressSet}
+import freechips.rocketchip.diplomacy.{RegionType, AddressSet}
+import freechips.rocketchip.resources.{DTSModel, DTSTimebase}
 import freechips.rocketchip.tile.{XLen}
 
 import sifive.blocks.devices.spi.{PeripherySPIKey, SPIParams}
diff --git a/fpga/src/main/scala/vcu118/Configs.scala b/fpga/src/main/scala/vcu118/Configs.scala
index d9caa81d..afcbf7d4 100644
--- a/fpga/src/main/scala/vcu118/Configs.scala
+++ b/fpga/src/main/scala/vcu118/Configs.scala
@@ -6,7 +6,8 @@ import org.chipsalliance.cde.config.{Config, Parameters}
 import freechips.rocketchip.subsystem.{SystemBusKey, PeripheryBusKey, ControlBusKey, ExtMem}
 import freechips.rocketchip.devices.debug.{DebugModuleKey, ExportDebug, JTAG}
 import freechips.rocketchip.devices.tilelink.{DevNullParams, BootROMLocated}
-import freechips.rocketchip.diplomacy.{DTSModel, DTSTimebase, RegionType, AddressSet}
+import freechips.rocketchip.diplomacy.{RegionType, AddressSet}
+import freechips.rocketchip.resources.{DTSModel, DTSTimebase}
 import freechips.rocketchip.tile.{XLen}
 
 import sifive.blocks.devices.spi.{PeripherySPIKey, SPIParams}
diff --git a/generators/boom b/generators/boom
index 506c11c9..6a3ad0a1 160000
--- a/generators/boom
+++ b/generators/boom
@@ -1 +1 @@
-Subproject commit 506c11c99ca9f6015d14fdc02c45c1e22983d06f
+Subproject commit 6a3ad0a1d9ae659aba5aea78978ef1a7cb5dd163
diff --git a/generators/chipyard/src/main/scala/DigitalTop.scala b/generators/chipyard/src/main/scala/DigitalTop.scala
index 7a216fa1..00a9cca4 100644
--- a/generators/chipyard/src/main/scala/DigitalTop.scala
+++ b/generators/chipyard/src/main/scala/DigitalTop.scala
@@ -41,6 +41,7 @@ class DigitalTop(implicit p: Parameters) extends ChipyardSystem
   with chipyard.clocking.CanHaveClockTap // Enables optionally adding a clock tap output port
   with fftgenerator.CanHavePeripheryFFT // Enables optionally having an MMIO-based FFT block
   with constellation.soc.CanHaveGlobalNoC // Support instantiating a global NoC interconnect
+  with rerocc.CanHaveReRoCCTiles // Support tiles that instantiate rerocc-attached accelerators
 {
   override lazy val module = new DigitalTopModule(this)
 }
diff --git a/generators/chipyard/src/main/scala/SpikeTile.scala b/generators/chipyard/src/main/scala/SpikeTile.scala
index 97866c50..6811be7a 100644
--- a/generators/chipyard/src/main/scala/SpikeTile.scala
+++ b/generators/chipyard/src/main/scala/SpikeTile.scala
@@ -14,7 +14,7 @@ import freechips.rocketchip.tilelink._
 import freechips.rocketchip.interrupts._
 import freechips.rocketchip.util._
 import freechips.rocketchip.tile._
-import freechips.rocketchip.prci.ClockSinkParameters
+import freechips.rocketchip.prci._
 
 case class SpikeCoreParams() extends CoreParams {
   val useVM = true
diff --git a/generators/chipyard/src/main/scala/clocking/ClockBinders.scala b/generators/chipyard/src/main/scala/clocking/ClockBinders.scala
index d075fcbb..7ef9f922 100644
--- a/generators/chipyard/src/main/scala/clocking/ClockBinders.scala
+++ b/generators/chipyard/src/main/scala/clocking/ClockBinders.scala
@@ -27,9 +27,9 @@ class WithPLLSelectorDividerClockGenerator(enable: Boolean = true) extends Overr
     val clockSelector = system.prci_ctrl_domain { LazyModule(new TLClockSelector(baseAddress + 0x30000, tlbus.beatBytes, enable=enable)) }
     val pllCtrl       = system.prci_ctrl_domain { LazyModule(new FakePLLCtrl    (baseAddress + 0x40000, tlbus.beatBytes)) }
 
-    clockDivider.tlNode  := system.prci_ctrl_domain { TLFragmenter(tlbus.beatBytes, tlbus.blockBytes) := system.prci_ctrl_bus.get }
-    clockSelector.tlNode := system.prci_ctrl_domain { TLFragmenter(tlbus.beatBytes, tlbus.blockBytes) := system.prci_ctrl_bus.get }
-    pllCtrl.tlNode       := system.prci_ctrl_domain { TLFragmenter(tlbus.beatBytes, tlbus.blockBytes) := system.prci_ctrl_bus.get }
+    clockDivider.tlNode  := system.prci_ctrl_domain { TLFragmenter(tlbus, Some("ClockDivider")) := system.prci_ctrl_bus.get }
+    clockSelector.tlNode := system.prci_ctrl_domain { TLFragmenter(tlbus, Some("ClockSelector")) := system.prci_ctrl_bus.get }
+    pllCtrl.tlNode       := system.prci_ctrl_domain { TLFragmenter(tlbus, Some("PLLCtrl")) := system.prci_ctrl_bus.get }
 
     system.chiptopClockGroupsNode := clockDivider.clockNode := clockSelector.clockNode
 
@@ -72,7 +72,7 @@ class WithPLLSelectorDividerClockGenerator(enable: Boolean = true) extends Overr
     }
   }
 })
- 
+
 // This passes all clocks through to the TestHarness
 class WithPassthroughClockGenerator extends OverrideLazyIOBinder({
   (system: HasChipyardPRCI) => {
@@ -102,6 +102,32 @@ class WithPassthroughClockGenerator extends OverrideLazyIOBinder({
   }
 })
 
+// Broadcasts a single clock IO to all clock domains. Ignores all requested frequencies
+class WithSingleClockBroadcastClockGenerator(freqMHz: Int = 100) extends OverrideLazyIOBinder({
+  (system: HasChipyardPRCI) => {
+    implicit val p = GetSystemParameters(system)
+
+    val clockGroupsAggregator = LazyModule(new ClockGroupAggregator("single_clock"))
+    val clockGroupsSourceNode = ClockGroupSourceNode(Seq(ClockGroupSourceParameters()))
+    system.chiptopClockGroupsNode :*= clockGroupsAggregator.node := clockGroupsSourceNode
+
+    InModuleBody {
+      val clock_wire = Wire(Input(Clock()))
+      val reset_wire = Wire(Input(AsyncReset()))
+      val (clock_io, clockIOCell) = IOCell.generateIOFromSignal(clock_wire, "clock", p(IOCellKey))
+      val (reset_io, resetIOCell) = IOCell.generateIOFromSignal(reset_wire, "reset", p(IOCellKey))
+
+      clockGroupsSourceNode.out.foreach { case (bundle, edge) =>
+        bundle.member.data.foreach { b =>
+          b.clock := clock_io
+          b.reset := reset_io
+        }
+      }
+      (Seq(ClockPort(() => clock_io, freqMHz), ResetPort(() => reset_io)), clockIOCell ++ resetIOCell)
+    }
+  }
+})
+
 class WithClockTapIOCells extends OverrideIOBinder({
   (system: CanHaveClockTap) => {
     system.clockTapIO.map { tap =>
diff --git a/generators/chipyard/src/main/scala/clocking/ClockGroupCombiner.scala b/generators/chipyard/src/main/scala/clocking/ClockGroupCombiner.scala
index 9b53cbef..9508e76f 100644
--- a/generators/chipyard/src/main/scala/clocking/ClockGroupCombiner.scala
+++ b/generators/chipyard/src/main/scala/clocking/ClockGroupCombiner.scala
@@ -48,9 +48,10 @@ class ClockGroupCombiner(implicit p: Parameters, v: ValName) extends LazyModule
       val name = combiners(i)._1
       i = i + 1
       require(g.size >= 1)
+      val names = g.map(_.name.getOrElse("unamed"))
       val takes = g.map(_.take).flatten
       require(takes.distinct.size <= 1,
-        s"Clock group $name has non-homogeneous requested ClockParameters $takes")
+        s"Clock group $name has non-homogeneous requested ClockParameters ${names.zip(takes)}")
       require(takes.size > 0,
         s"Clock group $name has no inheritable frequencies")
       (grouped ++ Seq(ClockSinkParameters(take = takes.headOption, name = Some(name))), r)
diff --git a/generators/chipyard/src/main/scala/clocking/ClockGroupNamePrefixer.scala b/generators/chipyard/src/main/scala/clocking/ClockGroupNamePrefixer.scala
index 08d54acf..367386a1 100644
--- a/generators/chipyard/src/main/scala/clocking/ClockGroupNamePrefixer.scala
+++ b/generators/chipyard/src/main/scala/clocking/ClockGroupNamePrefixer.scala
@@ -28,6 +28,7 @@ class ClockGroupParameterModifier(
     sinkFn:   ClockGroupSinkParameters   => ClockGroupSinkParameters   = { s => s })(
     implicit p: Parameters, v: ValName) extends LazyModule {
   val node = ClockGroupAdapterNode(sourceFn, sinkFn)
+  override def shouldBeInlined = true
   lazy val module = new LazyRawModuleImp(this) {
     (node.out zip node.in).map { case ((o, _), (i, _)) =>
       (o.member.data zip i.member.data).foreach { case (oD, iD) => oD := iD }
diff --git a/generators/chipyard/src/main/scala/clocking/HasChipyardPRCI.scala b/generators/chipyard/src/main/scala/clocking/HasChipyardPRCI.scala
index 9b140e20..f1936d48 100644
--- a/generators/chipyard/src/main/scala/clocking/HasChipyardPRCI.scala
+++ b/generators/chipyard/src/main/scala/clocking/HasChipyardPRCI.scala
@@ -37,9 +37,10 @@ trait HasChipyardPRCI { this: BaseSubsystem with InstantiatesHierarchicalElement
 
   // Set up clock domain
   private val tlbus = locateTLBusWrapper(prciParams.slaveWhere)
-  val prci_ctrl_domain = tlbus.generateSynchronousDomain.suggestName("chipyard_prcictrl_domain")
+  val prci_ctrl_domain = tlbus.generateSynchronousDomain("ChipyardPRCICtrl")
+    .suggestName("chipyard_prcictrl_domain")
 
-  val prci_ctrl_bus = Option.when(prciParams.generatePRCIXBar) { prci_ctrl_domain { TLXbar() } }
+  val prci_ctrl_bus = Option.when(prciParams.generatePRCIXBar) { prci_ctrl_domain { TLXbar(nameSuffix = Some("prcibus")) } }
   prci_ctrl_bus.foreach(xbar => tlbus.coupleTo("prci_ctrl") { (xbar
     := TLFIFOFixer(TLFIFOFixer.all)
     := TLBuffer()
@@ -70,13 +71,13 @@ trait HasChipyardPRCI { this: BaseSubsystem with InstantiatesHierarchicalElement
   }
   val tileClockGater     = Option.when(prciParams.enableTileClockGating) { prci_ctrl_domain {
     val clock_gater = LazyModule(new TileClockGater(prciParams.baseAddress + 0x00000, tlbus.beatBytes))
-    clock_gater.tlNode := TLFragmenter(tlbus.beatBytes, tlbus.blockBytes) := prci_ctrl_bus.get
+    clock_gater.tlNode := TLFragmenter(tlbus.beatBytes, tlbus.blockBytes, nameSuffix = Some("TileClockGater")) := prci_ctrl_bus.get
     clock_gater
   } }
   val tileResetSetter    = Option.when(prciParams.enableTileResetSetting) { prci_ctrl_domain {
     val reset_setter = LazyModule(new TileResetSetter(prciParams.baseAddress + 0x10000, tlbus.beatBytes,
       tile_prci_domains.map(_._2.tile_reset_domain.clockNode.portParams(0).name.get).toSeq, Nil))
-    reset_setter.tlNode := TLFragmenter(tlbus.beatBytes, tlbus.blockBytes) := prci_ctrl_bus.get
+    reset_setter.tlNode := TLFragmenter(tlbus.beatBytes, tlbus.blockBytes, nameSuffix = Some("TileResetSetter")) := prci_ctrl_bus.get
     reset_setter
   } }
 
diff --git a/generators/chipyard/src/main/scala/config/AbstractConfig.scala b/generators/chipyard/src/main/scala/config/AbstractConfig.scala
index 77a998e6..ff65ad89 100644
--- a/generators/chipyard/src/main/scala/config/AbstractConfig.scala
+++ b/generators/chipyard/src/main/scala/config/AbstractConfig.scala
@@ -59,6 +59,7 @@ class AbstractConfig extends Config(
   new chipyard.iobinders.WithNICIOPunchthrough ++
   new chipyard.iobinders.WithTraceIOPunchthrough ++
   new chipyard.iobinders.WithUARTTSIPunchthrough ++
+  new chipyard.iobinders.WithGCDBusyPunchthrough ++
   new chipyard.iobinders.WithNMITiedOff ++
 
 
@@ -122,7 +123,6 @@ class AbstractConfig extends Config(
   new freechips.rocketchip.subsystem.WithCoherentBusTopology ++     /** hierarchical buses including sbus/mbus/pbus/fbus/cbus/l2 */
   new chipyard.config.WithSV48IfPossible ++                         /** use sv48 if possible */
 
-
   // ================================================
   //   Set up power, reset and clocking
   // ================================================
diff --git a/generators/chipyard/src/main/scala/config/ClockingConfigs.scala b/generators/chipyard/src/main/scala/config/ClockingConfigs.scala
new file mode 100644
index 00000000..0d3197ea
--- /dev/null
+++ b/generators/chipyard/src/main/scala/config/ClockingConfigs.scala
@@ -0,0 +1,34 @@
+package chipyard
+
+import org.chipsalliance.cde.config.{Config}
+import freechips.rocketchip.diplomacy._
+import freechips.rocketchip.subsystem.{MBUS, SBUS}
+import testchipip.soc.{OBUS}
+
+//==================================================
+// This file contains examples of the different ways
+// clocks can be generated for chiypard designs
+//==================================================
+
+// The default constructs IOs for all requested clocks in the chiptopClockGroupsNode
+// Note: This is what designs inheriting from AbstractConfig do by default
+class DefaultClockingRocketConfig extends Config(
+  new chipyard.clocking.WithPassthroughClockGenerator ++
+  new freechips.rocketchip.subsystem.WithNBigCores(1) ++
+  new chipyard.config.AbstractConfig)
+
+// This is a more physically realistic approach, normally we can't punch out a separate
+// pin for each clock domain. The standard "test chip" approach is to punch a few slow clock
+// inputs, integrate a PLL, and generate an array of selectors/dividers to configure the
+// clocks for each domain. See the source for WithPLLSelectorDividerClockGenerator for more info
+class ChipLikeClockingRocketConfig extends Config(
+  new chipyard.clocking.WithPLLSelectorDividerClockGenerator ++
+  new freechips.rocketchip.subsystem.WithNBigCores(1) ++
+  new chipyard.config.AbstractConfig)
+
+// This merges all the clock domains in chiptopClockGroupsNode into one, then generates a single
+// clock input pin.
+class SingleClockBroadcastRocketConfig extends Config(
+  new chipyard.clocking.WithSingleClockBroadcastClockGenerator ++
+  new freechips.rocketchip.subsystem.WithNBigCores(1) ++
+  new chipyard.config.AbstractConfig)
diff --git a/generators/chipyard/src/main/scala/config/MMIOAcceleratorConfigs.scala b/generators/chipyard/src/main/scala/config/MMIOAcceleratorConfigs.scala
index fc6466ce..c1213d93 100644
--- a/generators/chipyard/src/main/scala/config/MMIOAcceleratorConfigs.scala
+++ b/generators/chipyard/src/main/scala/config/MMIOAcceleratorConfigs.scala
@@ -1,7 +1,6 @@
 package chipyard
 
 import org.chipsalliance.cde.config.{Config}
-import freechips.rocketchip.diplomacy.{AsynchronousCrossing}
 
 // ------------------------------
 // Configs with MMIO accelerators
diff --git a/generators/chipyard/src/main/scala/config/MemorySystemConfigs.scala b/generators/chipyard/src/main/scala/config/MemorySystemConfigs.scala
index 1b351266..fe99d5f1 100644
--- a/generators/chipyard/src/main/scala/config/MemorySystemConfigs.scala
+++ b/generators/chipyard/src/main/scala/config/MemorySystemConfigs.scala
@@ -1,7 +1,6 @@
 package chipyard
 
 import org.chipsalliance.cde.config.{Config}
-import freechips.rocketchip.diplomacy.{AsynchronousCrossing}
 
 // ------------------------------------------------------------
 // Configs which demonstrate modifying the uncore memory system
diff --git a/generators/chipyard/src/main/scala/config/NoCConfigs.scala b/generators/chipyard/src/main/scala/config/NoCConfigs.scala
index a5acc1e0..ebd196c3 100644
--- a/generators/chipyard/src/main/scala/config/NoCConfigs.scala
+++ b/generators/chipyard/src/main/scala/config/NoCConfigs.scala
@@ -1,7 +1,6 @@
 package chipyard
 
 import org.chipsalliance.cde.config.{Config}
-import freechips.rocketchip.diplomacy.{AsynchronousCrossing}
 import freechips.rocketchip.subsystem.{SBUS, MBUS}
 
 import constellation.channel._
@@ -267,3 +266,26 @@ class SbusMeshNoCConfig extends Config(
   new chipyard.config.AbstractConfig
 )
 
+class QuadRocketSbusRingNoCConfig extends Config(
+  new constellation.soc.WithSbusNoC(constellation.protocol.SimpleTLNoCParams(
+    constellation.protocol.DiplomaticNetworkNodeMapping(
+      inNodeMapping = ListMap(
+        "Core 0 " -> 0,
+        "Core 1 " -> 1,
+        "Core 2 " -> 2,
+        "Core 3 " -> 3,
+        "serial_tl" -> 4),
+      outNodeMapping = ListMap(
+        "system[0]" -> 5,
+        "system[1]" -> 6,
+        "system[2]" -> 7,
+        "system[3]" -> 8,
+        "pbus" -> 4)), // TSI is on the pbus, so serial-tl and pbus should be on the same node
+    nocParams = NoCParams(
+      topology        = UnidirectionalTorus1D(9),
+      channelParamGen = (a, b) => UserChannelParams(Seq.fill(10) { UserVirtualChannelParams(4) }),
+      routingRelation = NonblockingVirtualSubnetworksRouting(UnidirectionalTorus1DDatelineRouting(), 5, 2))
+  )) ++
+  new freechips.rocketchip.subsystem.WithNBigCores(4) ++
+  new freechips.rocketchip.subsystem.WithNBanks(4) ++
+  new chipyard.config.AbstractConfig)
diff --git a/generators/chipyard/src/main/scala/config/NoCoreConfigs.scala b/generators/chipyard/src/main/scala/config/NoCoreConfigs.scala
index 23da9700..5f376eaa 100644
--- a/generators/chipyard/src/main/scala/config/NoCoreConfigs.scala
+++ b/generators/chipyard/src/main/scala/config/NoCoreConfigs.scala
@@ -13,9 +13,9 @@ class NoCoresConfig extends Config(
   new chipyard.config.WithNoUART ++
   new chipyard.config.WithNoTileClockGaters ++
   new chipyard.config.WithNoTileResetSetters ++
-  new chipyard.config.WithNoBusErrorDevices ++
   new chipyard.config.WithNoDebug ++
   new chipyard.config.WithNoPLIC ++
+  new chipyard.config.WithNoBusErrorDevices ++
   new chipyard.config.AbstractConfig)
 
 // A config that uses a empty chiptop module with no rocket-chip soc components
diff --git a/generators/chipyard/src/main/scala/config/PeripheralDeviceConfigs.scala b/generators/chipyard/src/main/scala/config/PeripheralDeviceConfigs.scala
index fa74d9b8..c84427ad 100644
--- a/generators/chipyard/src/main/scala/config/PeripheralDeviceConfigs.scala
+++ b/generators/chipyard/src/main/scala/config/PeripheralDeviceConfigs.scala
@@ -1,7 +1,6 @@
 package chipyard
 
 import org.chipsalliance.cde.config.{Config}
-import freechips.rocketchip.diplomacy.{AsynchronousCrossing}
 import freechips.rocketchip.subsystem.{MBUS}
 
 // ---------------------------------------------------------
diff --git a/generators/chipyard/src/main/scala/config/RoCCAcceleratorConfigs.scala b/generators/chipyard/src/main/scala/config/RoCCAcceleratorConfigs.scala
index 184afaa7..a654d70f 100644
--- a/generators/chipyard/src/main/scala/config/RoCCAcceleratorConfigs.scala
+++ b/generators/chipyard/src/main/scala/config/RoCCAcceleratorConfigs.scala
@@ -1,7 +1,6 @@
 package chipyard
 
 import org.chipsalliance.cde.config.{Config}
-import freechips.rocketchip.diplomacy.{AsynchronousCrossing}
 
 // ------------------------------
 // Configs with RoCC Accelerators
@@ -48,3 +47,27 @@ class AES256ECBRocketConfig extends Config(
   new freechips.rocketchip.subsystem.WithNBigCores(1) ++
   new chipyard.config.WithSystemBusWidth(256) ++
   new chipyard.config.AbstractConfig)
+
+class ReRoCCTestConfig extends Config(
+  new rerocc.WithReRoCC ++
+  new chipyard.config.WithCharacterCountRoCC ++                // rerocc tile4 is charcnt
+  new chipyard.config.WithAccumulatorRoCC ++                   // rerocc tile3 is accum
+  new chipyard.config.WithAccumulatorRoCC ++                   // rerocc tile2 is accum
+  new chipyard.config.WithAccumulatorRoCC ++                   // rerocc tile1 is accum
+  new chipyard.config.WithAccumulatorRoCC ++                   // rerocc tile0 is accum
+  new freechips.rocketchip.subsystem.WithNBigCores(1) ++
+  new chipyard.config.AbstractConfig)
+
+class ReRoCCManyGemminiConfig extends Config(
+  new rerocc.WithReRoCC ++
+  new gemmini.LeanGemminiConfig ++                              // rerocc tile3 is gemmini
+  new gemmini.LeanGemminiConfig ++                              // rerocc tile2 is gemmini
+  new gemmini.LeanGemminiConfig ++                              // rerocc tile1 is gemmini
+  new gemmini.LeanGemminiConfig ++                              // rerocc tile0 is gemmini
+  new freechips.rocketchip.subsystem.WithNBigCores(4) ++        // 4 rocket cores
+  new chipyard.config.AbstractConfig)
+
+class ZstdCompressorRocketConfig extends Config(
+  new compressacc.WithZstdCompressor ++
+  new freechips.rocketchip.subsystem.WithNBigCores(1) ++
+  new chipyard.config.AbstractConfig)
diff --git a/generators/chipyard/src/main/scala/config/RocketConfigs.scala b/generators/chipyard/src/main/scala/config/RocketConfigs.scala
index d3f21584..c0ee4bba 100644
--- a/generators/chipyard/src/main/scala/config/RocketConfigs.scala
+++ b/generators/chipyard/src/main/scala/config/RocketConfigs.scala
@@ -1,7 +1,7 @@
 package chipyard
 
 import org.chipsalliance.cde.config.{Config}
-import freechips.rocketchip.diplomacy.{AsynchronousCrossing}
+import freechips.rocketchip.prci.{AsynchronousCrossing}
 import freechips.rocketchip.subsystem.{InCluster}
 
 // --------------
@@ -12,6 +12,10 @@ class RocketConfig extends Config(
   new freechips.rocketchip.subsystem.WithNBigCores(1) ++         // single rocket-core
   new chipyard.config.AbstractConfig)
 
+class DualRocketConfig extends Config(
+  new freechips.rocketchip.subsystem.WithNBigCores(2) ++
+  new chipyard.config.AbstractConfig)
+
 class TinyRocketConfig extends Config(
   new chipyard.harness.WithDontTouchChipTopPorts(false) ++        // TODO FIX: Don't dontTouch the ports
   new testchipip.soc.WithNoScratchpads ++                         // All memory is the Rocket TCMs
diff --git a/generators/chipyard/src/main/scala/config/fragments/PeripheralFragments.scala b/generators/chipyard/src/main/scala/config/fragments/PeripheralFragments.scala
index 265be720..87229cb4 100644
--- a/generators/chipyard/src/main/scala/config/fragments/PeripheralFragments.scala
+++ b/generators/chipyard/src/main/scala/config/fragments/PeripheralFragments.scala
@@ -7,7 +7,7 @@ import chisel3.util.{log2Up}
 import org.chipsalliance.cde.config.{Config}
 import freechips.rocketchip.devices.tilelink.{BootROMLocated, PLICKey, CLINTKey}
 import freechips.rocketchip.devices.debug.{Debug, ExportDebug, DebugModuleKey, DMI, JtagDTMKey, JtagDTMConfig}
-import freechips.rocketchip.diplomacy.{AsynchronousCrossing}
+import freechips.rocketchip.prci.{AsynchronousCrossing}
 import chipyard.stage.phases.TargetDirKey
 import freechips.rocketchip.subsystem._
 import freechips.rocketchip.tile.{XLen}
diff --git a/generators/chipyard/src/main/scala/config/fragments/RoCCFragments.scala b/generators/chipyard/src/main/scala/config/fragments/RoCCFragments.scala
index 5ef76b75..c29b05d3 100644
--- a/generators/chipyard/src/main/scala/config/fragments/RoCCFragments.scala
+++ b/generators/chipyard/src/main/scala/config/fragments/RoCCFragments.scala
@@ -43,3 +43,17 @@ class WithMultiRoCCGemmini[T <: Data : Arithmetic, U <: Data, V <: Data](
     }))
   }
 })
+
+class WithAccumulatorRoCC(op: OpcodeSet = OpcodeSet.custom1) extends Config((site, here, up) => {
+  case BuildRoCC => up(BuildRoCC) ++ Seq((p: Parameters) => {
+    val accumulator = LazyModule(new AccumulatorExample(op, n = 4)(p))
+    accumulator
+  })
+})
+
+class WithCharacterCountRoCC(op: OpcodeSet = OpcodeSet.custom2) extends Config((site, here, up) => {
+  case BuildRoCC => up(BuildRoCC) ++ Seq((p: Parameters) => {
+    val counter = LazyModule(new CharacterCountExample(op)(p))
+    counter
+  })
+})
diff --git a/generators/chipyard/src/main/scala/config/fragments/SubsystemFragments.scala b/generators/chipyard/src/main/scala/config/fragments/SubsystemFragments.scala
index 8e16b26c..42899286 100644
--- a/generators/chipyard/src/main/scala/config/fragments/SubsystemFragments.scala
+++ b/generators/chipyard/src/main/scala/config/fragments/SubsystemFragments.scala
@@ -2,7 +2,7 @@ package chipyard.config
 
 import org.chipsalliance.cde.config.{Config}
 import freechips.rocketchip.subsystem._
-import freechips.rocketchip.diplomacy.{DTSTimebase}
+import freechips.rocketchip.resources.{DTSTimebase}
 import sifive.blocks.inclusivecache.{InclusiveCachePortParameters}
 
 // Replaces the L2 with a broadcast manager for maintaining coherence
diff --git a/generators/chipyard/src/main/scala/example/TutorialTile.scala b/generators/chipyard/src/main/scala/example/TutorialTile.scala
index 76b17273..f69ec864 100644
--- a/generators/chipyard/src/main/scala/example/TutorialTile.scala
+++ b/generators/chipyard/src/main/scala/example/TutorialTile.scala
@@ -14,7 +14,7 @@ import freechips.rocketchip.interrupts._
 import freechips.rocketchip.util._
 import freechips.rocketchip.tile._
 import freechips.rocketchip.amba.axi4._
-import freechips.rocketchip.prci.ClockSinkParameters
+import freechips.rocketchip.prci._
 
 // Example parameter class copied from CVA6, not included in documentation but for compile check only
 // If you are here for documentation, DO NOT copy MyCoreParams and MyTileParams directly - always figure
diff --git a/generators/chipyard/src/main/scala/iobinders/IOBinders.scala b/generators/chipyard/src/main/scala/iobinders/IOBinders.scala
index a987d1ab..6aaad8db 100644
--- a/generators/chipyard/src/main/scala/iobinders/IOBinders.scala
+++ b/generators/chipyard/src/main/scala/iobinders/IOBinders.scala
@@ -39,6 +39,7 @@ import testchipip.cosim.{CanHaveTraceIO, TraceOutputTop, SpikeCosimConfig}
 import testchipip.tsi.{CanHavePeripheryUARTTSI, UARTTSIIO}
 import icenet.{CanHavePeripheryIceNIC, SimNetwork, NicLoopback, NICKey, NICIOvonly}
 import chipyard.{CanHaveMasterTLMemPort, HasCeaseIO, ChipyardSystem, ChipyardSystemModule}
+import chipyard.example.{CanHavePeripheryGCD}
 
 import scala.reflect.{ClassTag}
 
@@ -548,3 +549,11 @@ class WithNMITiedOff extends ComposeIOBinder({
     (Nil, Nil)
   }
 })
+
+class WithGCDBusyPunchthrough extends OverrideIOBinder({
+  (system: CanHavePeripheryGCD) => system.gcd_busy.map { busy =>
+    val io_gcd_busy = IO(Output(Bool()))
+    io_gcd_busy := busy
+    (Seq(GCDBusyPort(() => io_gcd_busy)), Nil)
+  }.getOrElse((Nil, Nil))
+})
diff --git a/generators/chipyard/src/main/scala/iobinders/Ports.scala b/generators/chipyard/src/main/scala/iobinders/Ports.scala
index a9ea0a2d..dfe9c58f 100644
--- a/generators/chipyard/src/main/scala/iobinders/Ports.scala
+++ b/generators/chipyard/src/main/scala/iobinders/Ports.scala
@@ -109,3 +109,5 @@ case class JTAGResetPort   (val getIO: () => Reset)
 case class TLMemPort       (val getIO: () => HeterogeneousBag[TLBundle])
     extends Port[HeterogeneousBag[TLBundle]]
 
+case class GCDBusyPort     (val getIO: () => Bool)
+    extends Port[Bool]
diff --git a/generators/compress-acc b/generators/compress-acc
new file mode 160000
index 00000000..580fc99a
--- /dev/null
+++ b/generators/compress-acc
@@ -0,0 +1 @@
+Subproject commit 580fc99a8de7e6965c5bc9f08945bc854d7d2a0c
diff --git a/generators/constellation b/generators/constellation
index 6664839b..5c9d2735 160000
--- a/generators/constellation
+++ b/generators/constellation
@@ -1 +1 @@
-Subproject commit 6664839b104a0d34adebfee13a0addc19efbde50
+Subproject commit 5c9d27359dcb2b144c99cc8e3e561979eb2bd905
diff --git a/generators/cva6 b/generators/cva6
index 9d1c1068..de4772f1 160000
--- a/generators/cva6
+++ b/generators/cva6
@@ -1 +1 @@
-Subproject commit 9d1c106834824ddb8052b7f60574b2b544b40395
+Subproject commit de4772f1d66cc6d4ef42e75f34c42eb7b941c8f0
diff --git a/generators/diplomacy b/generators/diplomacy
index 055be698..6b7dc988 160000
--- a/generators/diplomacy
+++ b/generators/diplomacy
@@ -1 +1 @@
-Subproject commit 055be698f4ad55bf4a90b3d5e31d4344be5f788b
+Subproject commit 6b7dc988a771ff21237b102fd02a1f2d2c2e7b72
diff --git a/generators/firechip/src/main/scala/TargetConfigs.scala b/generators/firechip/src/main/scala/TargetConfigs.scala
index 5934f8c2..dbe197db 100644
--- a/generators/firechip/src/main/scala/TargetConfigs.scala
+++ b/generators/firechip/src/main/scala/TargetConfigs.scala
@@ -12,7 +12,8 @@ import freechips.rocketchip.rocket.DCacheParams
 import freechips.rocketchip.subsystem._
 import freechips.rocketchip.devices.tilelink.{BootROMLocated, BootROMParams}
 import freechips.rocketchip.devices.debug.{DebugModuleParams, DebugModuleKey}
-import freechips.rocketchip.diplomacy.{LazyModule, AsynchronousCrossing}
+import freechips.rocketchip.diplomacy.{LazyModule}
+import freechips.rocketchip.prci.{AsynchronousCrossing}
 import testchipip.iceblk.{BlockDeviceKey, BlockDeviceConfig}
 import testchipip.cosim.{TracePortKey, TracePortParams}
 import sifive.blocks.devices.uart.{PeripheryUARTKey, UARTParams}
@@ -368,6 +369,12 @@ class FireSimRadianceClusterSynConfig extends Config(
   new WithFireSimConfigTweaks ++
   new chipyard.RadianceClusterSynConfig)
 
+class FireSimQuadRocketSbusRingNoCConfig extends Config(
+  new chipyard.config.WithNoTraceIO ++
+  new WithDefaultFireSimBridges ++
+  new WithFireSimConfigTweaks++
+  new chipyard.QuadRocketSbusRingNoCConfig)
+
 class FireSimLargeBoomSV39CospikeConfig extends Config(
   new firesim.firesim.WithCospikeBridge ++
   new WithDefaultFireSimBridges ++
diff --git a/generators/hardfloat b/generators/hardfloat
index d93aa570..4225367e 160000
--- a/generators/hardfloat
+++ b/generators/hardfloat
@@ -1 +1 @@
-Subproject commit d93aa570806013dea479a92ba9bb33d1f2d4f69f
+Subproject commit 4225367ed282aadd70f4cbd131f6959f4d00c502
diff --git a/generators/ibex b/generators/ibex
index 89c19c2d..8a43aa70 160000
--- a/generators/ibex
+++ b/generators/ibex
@@ -1 +1 @@
-Subproject commit 89c19c2d7bc2523a0d9fb85e154bcaf2c1cc7665
+Subproject commit 8a43aa70da1c06f4d34ec8710a3d2ee7585948a5
diff --git a/generators/icenet b/generators/icenet
index 969bc8f9..6fd35bf5 160000
--- a/generators/icenet
+++ b/generators/icenet
@@ -1 +1 @@
-Subproject commit 969bc8f9a00e4b11d243656f255761e04e10ccb9
+Subproject commit 6fd35bf5a28310c61671fce7309526b56d91e35d
diff --git a/generators/rerocc b/generators/rerocc
new file mode 160000
index 00000000..a22dce62
--- /dev/null
+++ b/generators/rerocc
@@ -0,0 +1 @@
+Subproject commit a22dce622db350e4a159ed90cf83d00782ed8187
diff --git a/generators/riscv-sodor b/generators/riscv-sodor
index ca043149..732cbe19 160000
--- a/generators/riscv-sodor
+++ b/generators/riscv-sodor
@@ -1 +1 @@
-Subproject commit ca0431493ec35983388cb08bb203d5e12e9a32b2
+Subproject commit 732cbe1990e9ef55ba122664465b372744f2eaab
diff --git a/generators/rocket-chip-blocks b/generators/rocket-chip-blocks
index 2fc961f3..c8c14f7b 160000
--- a/generators/rocket-chip-blocks
+++ b/generators/rocket-chip-blocks
@@ -1 +1 @@
-Subproject commit 2fc961f356b1104a34567adeefa9e3666940643f
+Subproject commit c8c14f7b47c3c790022c293bc7e4309f5c5ed523
diff --git a/generators/rocket-chip-inclusive-cache b/generators/rocket-chip-inclusive-cache
index 45d184f2..4aab5460 160000
--- a/generators/rocket-chip-inclusive-cache
+++ b/generators/rocket-chip-inclusive-cache
@@ -1 +1 @@
-Subproject commit 45d184f2fd1e2189d6302468e208a054b681caab
+Subproject commit 4aab5460bd569b2142eb760e4772ea8db6998dc3
diff --git a/generators/shuttle b/generators/shuttle
index b75fada2..799263c6 160000
--- a/generators/shuttle
+++ b/generators/shuttle
@@ -1 +1 @@
-Subproject commit b75fada257c576de5c9968dbbf5a09302b53b1de
+Subproject commit 799263c6180587670fe42de23f3338f1f218d87e
diff --git a/generators/testchipip b/generators/testchipip
index 731d51ee..9d8e830b 160000
--- a/generators/testchipip
+++ b/generators/testchipip
@@ -1 +1 @@
-Subproject commit 731d51ee15a4ad57cce17cec55ebe549970e4e1e
+Subproject commit 9d8e830be4bc96ff63a95cf3719f6ddb3c734c18
diff --git a/scripts/generate-ckpt.sh b/scripts/generate-ckpt.sh
index 3795b041..fe42d997 100755
--- a/scripts/generate-ckpt.sh
+++ b/scripts/generate-ckpt.sh
@@ -13,6 +13,8 @@ usage() {
     echo "  -i <insns> : Instructions after PC to take checkpoint at [default 0]"
     echo "  -m <isa>   : ISA to pass to spike for checkpoint generation [default rv64gc]"
     echo "  -o <out>   : Output directory to store the checkpoint in. [default <elf>.<pc>.<insns>.loadarch]"
+    echo "  -r <mem>   : Memory regions to pass to spike. Passed to spike's '-m' flag. [default starting at 0x80000000 with 256MiB]"
+    echo "  -v         : Verbose"
     exit "$1"
 }
 
@@ -22,6 +24,8 @@ PC="0x80000000"
 INSNS=0
 ISA="rv64gc"
 OUTPATH=""
+MEMOVERRIDE=""
+VERBOSE=0
 while [ "$1" != "" ];
 do
     case $1 in
@@ -45,6 +49,11 @@ do
         -o )
             shift
             OUTPATH=$1 ;;
+        -r )
+            shift
+            MEMOVERRIDE=$1 ;;
+        -v )
+            VERBOSE=1 ;;
 	* )
 	    error "Invalid option $1"
 	    usage 1 ;;
@@ -52,7 +61,15 @@ do
     shift
 done
 
-BASEMEM="$((0x80000000)):$((0x10000000))"
+if [[ $VERBOSE -eq 1 ]] ; then
+    set -x
+fi
+
+if [ -z "$MEMOVERRIDE" ] ; then
+    BASEMEM="$((0x80000000)):$((0x10000000))"
+else
+    BASEMEM=$MEMOVERRIDE
+fi
 SPIKEFLAGS="-p$NHARTS --pmpregions=0 --isa=$ISA -m$BASEMEM"
 BASENAME=$(basename -- $BINARY)
 
@@ -138,3 +155,5 @@ rm -rf mem.0x80000000.bin
 
 riscv64-unknown-elf-ld -Tdata=0x80000000 -nmagic --defsym tohost=0x$TOHOST --defsym fromhost=0x$FROMHOST -o $LOADMEM_ELF $RAWMEM_ELF
 rm -rf $RAWMEM_ELF
+
+echo "Ensure that at minimum you have memory regions corresponding to $BASEMEM in downstream RTL tooling"
diff --git a/scripts/init-submodules-no-riscv-tools-nolog.sh b/scripts/init-submodules-no-riscv-tools-nolog.sh
index c81c95f1..dec397c3 100755
--- a/scripts/init-submodules-no-riscv-tools-nolog.sh
+++ b/scripts/init-submodules-no-riscv-tools-nolog.sh
@@ -74,6 +74,7 @@ cd "$RDIR"
             toolchains/libgloss \
             generators/gemmini \
             generators/rocket-chip \
+            generators/compress-acc \
             sims/firesim \
             software/nvdla-workload \
             software/coremark \
@@ -121,6 +122,9 @@ cd "$RDIR"
     # Non-recursive clone
     git submodule update --init generators/rocket-chip
 
+    # Non-recursive clone
+    git submodule update --init generators/compress-acc
+
     # Minimal non-recursive clone to initialize sbt dependencies
     git submodule update --init sims/firesim
     git config --local submodule.sims/firesim.update none
diff --git a/scripts/insert-includes.py b/scripts/insert-includes.py
index b939878a..22dd8e9d 100755
--- a/scripts/insert-includes.py
+++ b/scripts/insert-includes.py
@@ -1,80 +1,134 @@
 #!/usr/bin/env python
+"""
+replaces a `include with the full include file.
 
-# replaces a `include with the full include file.
-# recursively replaces `include's until none are left
-#
-# args
-# $1 - file to remove includes from
-# $2 - file to write output to
-# $3 - list of directories to search for includes in (note: NON-RECURSIVE must specify all dirs)
-#      includes are found relative to this path
-#      this is equivalent to something like +incdir+
+args
+$1 - file which has includes to be replaced
+$2 - file in which output will be written
+$3 - list of directories to search for includes
+     (note: NON-RECURSIVE must specify all dirs)
+      includes are found relative to this path
+      this is equivalent to something like +incdir+
+"""
 
-import sys
-import re
 import os
-import tempfile
+import re
 import shutil
+import sys
+import tempfile
 
-inVlog = sys.argv[1]
-outVlog = sys.argv[2]
-print("[INFO] Replaces includes from: " + str(inVlog))
 
-if inVlog == outVlog:
-    sys.exit("[ERROR] The input and output file cannot be the same.")
+def print_info(msg):
+    """
+    Print an info message.
 
-# add directories to search list
-incDirs = sys.argv[3:]
-print("[INFO] Searching following dirs for includes: " + str(incDirs))
+    Args:
+        msg (str): message to print
+    """
+    print(f"[INFO] {msg}")
 
-def process(inF, outF):
-    # open file
-    with open(inF, 'r') as inFile:
-        with open(outF, 'w') as outFile:
-            # for each include found, search through all dirs and replace if found, error if not
-            for num, line in enumerate(inFile, 1):
-                match = re.match(r"^ *`include +\"(.*)\"", line)
-                if match and match.group(1) != "uvm_macros.svh":
-                    print("[INFO] Replacing includes for {}".format(match.group(1)))
-                    # search for include and replace
-                    found = False
-                    for d in incDirs:
-                        potentialIncFileName = d + "/" + match.group(1)
-                        if os.path.exists(potentialIncFileName):
-                            found = True
-                            print("[INFO] Found missing include in {}".format(potentialIncFileName))
-                            with open(potentialIncFileName, 'r') as incFile:
-                                for iline in incFile:
-                                    outFile.write(iline)
-                            break
 
-                    # must find something to include with
-                    if not found:
-                        sys.exit("[ERROR] Couldn't replace include \"" + str(match.group(1)) + "\" found on line " + str(num))
-                else:
-                    outFile.write(line)
+def print_error(msg, critical=True):
+    """
+    Print an error message.
 
-inF = inVlog
+    Args:
+        msg (str): message to print
+        critical (bool): whether to exit after printing the message
+    """
+    if critical:
+        sys.exit(f"[ERROR] {msg}")
+    else:
+        print(f"[ERROR] {msg}")
 
-while True:
-    # create a copy of the input
-    fd, temp_path = tempfile.mkstemp()
-    shutil.copy2(inF, temp_path)
 
-    with open(temp_path, 'r') as inFile:
-        anyIncludes = False
-        for line in inFile:
-            match = re.match(r"^ *`include +\"(.*)\"", line)
-            if match:
-                anyIncludes = True
-                break
+def find_include(file_name, inc_dirs):
+    """
+    Find the include file in the list of directories.
 
-        if anyIncludes:
-            process(temp_path, outVlog)
-            inF = outVlog
-            os.remove(temp_path)
-        else:
-            os.remove(temp_path)
-            break
+    Args:
+        file_name (str): include file name
+        inc_dirs (list): list of directories to search for includes
 
-print("[INFO] Success. Writing output to: " + str(outVlog))
+    Returns:
+        str: full path to the include file
+    """
+    for d in inc_dirs:
+        inc_file_name = d + "/" + file_name
+        if os.path.exists(inc_file_name):
+            return inc_file_name
+    print_error(f"Include file {file_name} not found in {inc_dirs}")
+    return None
+
+
+def process_helper(in_fname, out_f, inc_dirs, replaced_includes):
+    """
+    Helper function to DFS through include files and replace includes.
+    """
+    include_regex = re.compile(r"^ *`include +\"(.*)\"")
+    # slurp the input file.
+    # this avoids having a bunch of fds open during recursion
+    with open(in_fname, "r", encoding="utf-8") as in_file:
+        lines = in_file.readlines()
+
+    for num, line in enumerate(lines, 1):
+        match = re.match(include_regex, line)
+        if not match or match.group(1) == "uvm_macros.svh":
+            # copy the line as is
+            out_f.write(line)
+            continue
+        if match.group(1) in replaced_includes:
+            print_info("Skipping duplicate include")
+            continue
+
+        print_info(
+            f"Replacing includes for {match.group(1)}" f" at line {num}"
+        )
+        # search for include and replace
+        inc_file_name = find_include(match.group(1), inc_dirs)
+        replaced_includes.add(match.group(1))
+        process_helper(inc_file_name, out_f, inc_dirs, replaced_includes)
+
+
+def process(in_fname, out_fname, inc_dirs=None):
+    """
+    Replace include directives in a file with the full include file.
+
+    Args:
+        in_fname (str): input file name
+        out_fname (str): output file name
+        inc_dirs (list): list of directories to search for includes
+    """
+    replaced_includes = set()
+    with open(out_fname, "w", encoding="utf-8") as out_file:
+        process_helper(in_fname, out_file, inc_dirs, replaced_includes)
+
+
+def main():
+    """
+    Entry point for the script.
+
+    Args:
+    <input file> <output file> <list of directories to search for includes>
+    """
+    in_vlog = sys.argv[1]
+    out_vlog = sys.argv[2]
+
+    if in_vlog == out_vlog:
+        sys.exit("[ERROR] The input and output file cannot be the same.")
+
+    # add directories to search list
+    inc_dirs = sys.argv[3:]
+    print("[INFO] Replaces includes from: " + str(in_vlog))
+    print("[INFO] Searching following dirs for includes: " + str(inc_dirs))
+
+    # make a copy of the input file
+    _, temp_path = tempfile.mkstemp()
+    shutil.copy2(in_vlog, temp_path)
+    process(temp_path, out_vlog, inc_dirs)
+
+    print("[INFO] Success. Output written to: " + str(out_vlog))
+
+
+if __name__ == "__main__":
+    main()
diff --git a/scripts/uniquify-module-names.py b/scripts/uniquify-module-names.py
index d2131c33..5c6a9613 100755
--- a/scripts/uniquify-module-names.py
+++ b/scripts/uniquify-module-names.py
@@ -11,6 +11,7 @@ parser = argparse.ArgumentParser(description="")
 parser.add_argument("--model-hier-json", type=str, required=True, help="Path to hierarchy JSON emitted by firtool. Must include DUT as a module.")
 parser.add_argument("--top-hier-json", type=str, required=True, help="Path to hierarchy JSON emitted by firtool. Must include DUT as a module.")
 parser.add_argument('--in-all-filelist', type=str, required=True, help='Path to input filelist that has all modules (relative paths).')
+parser.add_argument('--in-bb-filelist', type=str, required=True, help='Path to input blackbox filelist')
 parser.add_argument("--dut", type=str, required=True, help="Name of the DUT module.")
 parser.add_argument("--model", type=str, required=True, help="Name of the Model module.")
 parser.add_argument('--out-dut-filelist', type=str, required=True, help='Path to output filelist including all modules under the DUT.')
@@ -57,8 +58,8 @@ def get_modules_in_verilog_file(file):
         module_names.append(words[1].replace("(", "").replace(")", "").replace(";", ""))
   return module_names
 
-def get_modules_in_filelist(verilog_module_filename, cc_filelist):
-  with open(args.in_all_filelist) as fl:
+def get_modules_in_filelist(filelist, verilog_module_filename, cc_filelist):
+  with open(filelist) as fl:
     lines = fl.readlines()
     for line in lines:
       path = line.strip()
@@ -186,7 +187,8 @@ def uniquify_modules_under_model(modules_under_model, common_modules, verilog_mo
 def main():
   verilog_module_filename = dict()
   cc_filelist = list()
-  get_modules_in_filelist(verilog_module_filename, cc_filelist)
+  get_modules_in_filelist(args.in_all_filelist, verilog_module_filename, cc_filelist)
+  get_modules_in_filelist(args.in_bb_filelist , verilog_module_filename, cc_filelist)
 
   modules_under_model = get_modules_under_hier(args.model_hier_json, args.dut)
   modules_under_top   = get_modules_under_hier(args.top_hier_json)
diff --git a/software/firemarshal b/software/firemarshal
index b014183a..0c0cdfeb 160000
--- a/software/firemarshal
+++ b/software/firemarshal
@@ -1 +1 @@
-Subproject commit b014183ac6afed7fe1aca89fa010be8d13ad4079
+Subproject commit 0c0cdfeb269b37c3d10e5f21b84cd60c21457487
diff --git a/toolchains/riscv-tools/riscv-pk b/toolchains/riscv-tools/riscv-pk
index e8e6b3aa..4f3debe4 160000
--- a/toolchains/riscv-tools/riscv-pk
+++ b/toolchains/riscv-tools/riscv-pk
@@ -1 +1 @@
-Subproject commit e8e6b3aaee44d43b48164fbd377864c3a682dbd3
+Subproject commit 4f3debe4d04f56d31089c1c716a27e2d5245e9a1
diff --git a/tools/circt b/tools/circt
index 9e0c1696..481cb60a 160000
--- a/tools/circt
+++ b/tools/circt
@@ -1 +1 @@
-Subproject commit 9e0c1696f3caef4059c65774ad6b8efee91d9d9e
+Subproject commit 481cb60add7358934414a3c6b396f5d29ad934fe
diff --git a/tools/rocket-dsp-utils b/tools/rocket-dsp-utils
index c4c638da..6ee2309f 160000
--- a/tools/rocket-dsp-utils
+++ b/tools/rocket-dsp-utils
@@ -1 +1 @@
-Subproject commit c4c638da02351c701ff31e73aadd987baa7a2acc
+Subproject commit 6ee2309f80a54b404795f227793d2a54e1dfadf8
diff --git a/variables.mk b/variables.mk
index 7f6f4523..0a62d2fa 100644
--- a/variables.mk
+++ b/variables.mk
@@ -157,16 +157,16 @@ long_name = $(MODEL_PACKAGE).$(MODEL).$(CONFIG)
 
 # classpaths
 CLASSPATH_CACHE ?= $(base_dir)/.classpath_cache
-CHIPYARD_CLASSPATH ?= $(CLASSPATH_CACHE)/chipyard.jar
+# The generator classpath must contain the Generator main
+GENERATOR_CLASSPATH ?= $(CLASSPATH_CACHE)/$(SBT_PROJECT).jar
+# The tapeout classpath must contain MacroCompiler
 TAPEOUT_CLASSPATH ?= $(CLASSPATH_CACHE)/tapeout.jar
-# if *_CLASSPATH is a true java classpath, it can be colon-delimited list of paths (on *nix)
-CHIPYARD_CLASSPATH_TARGETS ?= $(subst :, ,$(CHIPYARD_CLASSPATH))
-TAPEOUT_CLASSPATH_TARGETS ?= $(subst :, ,$(TAPEOUT_CLASSPATH))
 
 # chisel generated outputs
 FIRRTL_FILE ?= $(build_dir)/$(long_name).fir
 ANNO_FILE   ?= $(build_dir)/$(long_name).anno.json
 CHISEL_LOG_FILE ?= $(build_dir)/$(long_name).chisel.log
+FIRTOOL_LOG_FILE ?= $(build_dir)/$(long_name).firtool.log
 
 # chisel anno modification output
 MFC_EXTRA_ANNO_FILE ?= $(build_dir)/$(long_name).extrafirtool.anno.json
diff --git a/vlsi/example-asap7.yml b/vlsi/example-asap7.yml
index 6fc819aa..ab0d6c8e 100644
--- a/vlsi/example-asap7.yml
+++ b/vlsi/example-asap7.yml
@@ -17,7 +17,7 @@ vlsi.inputs.power_spec_type: "cpf"
 
 # Specify clock signals
 vlsi.inputs.clocks: [
-  {name: "clock_uncore_clock", period: "1ns", uncertainty: "0.1ns"}
+  {name: "clock_uncore", period: "1ns", uncertainty: "0.1ns"}
 ]
 
 # Generate Make include to aid in flow
diff --git a/vlsi/example-design.yml b/vlsi/example-design.yml
index 3f46a443..80f338a9 100644
--- a/vlsi/example-design.yml
+++ b/vlsi/example-design.yml
@@ -10,7 +10,7 @@ vlsi.inputs.power_spec_type: "cpf"
 
 # Specify clock signals
 vlsi.inputs.clocks: [
-  {name: "clock_uncore_clock", period: "2ns", uncertainty: "0.1ns"}
+  {name: "clock_uncore", period: "2ns", uncertainty: "0.1ns"}
 ]
 
 # Specify pin properties
diff --git a/vlsi/example-designs/sky130-commercial.yml b/vlsi/example-designs/sky130-commercial.yml
index fb1f8407..1ec4a08f 100644
--- a/vlsi/example-designs/sky130-commercial.yml
+++ b/vlsi/example-designs/sky130-commercial.yml
@@ -2,7 +2,7 @@
 
 # Specify clock signals
 vlsi.inputs.clocks: [
-  {name: "clock_uncore_clock", period: "30ns", uncertainty: "2ns"}
+  {name: "clock_uncore", period: "30ns", uncertainty: "2ns"}
 ]
 
 # Placement Constraints
diff --git a/vlsi/example-designs/sky130-openroad-rockettile.yml b/vlsi/example-designs/sky130-openroad-rockettile.yml
index 4a6b664c..cfc4c5bb 100644
--- a/vlsi/example-designs/sky130-openroad-rockettile.yml
+++ b/vlsi/example-designs/sky130-openroad-rockettile.yml
@@ -1,7 +1,7 @@
 # Override configurations in ../example-sky130.yml and example-designs
 
 # Specify clock signals
-# Rocket/RocketTile names clock signal "clock" instead of "clock_uncore_clock"
+# Rocket/RocketTile names clock signal "clock" instead of "clock_uncore"
 vlsi.inputs.clocks: [
   {name: "clock", period: "30ns", uncertainty: "3ns"}
 ]
@@ -22,7 +22,7 @@ vlsi.inputs.placement_constraints:
       bottom: 10
 
   # Place SRAM memory instances
-    # data cache
+  # data cache
   - path: "RocketTile/dcache/data/data_arrays_0/data_arrays_0_ext/mem_0_0"
     type: hardmacro
     x: 50
@@ -47,3 +47,4 @@ vlsi.inputs.placement_constraints:
     x: 50
     y: 2100
     orientation: r90
+
diff --git a/vlsi/example-designs/sky130-openroad.yml b/vlsi/example-designs/sky130-openroad.yml
index 3b95d982..a686fbba 100644
--- a/vlsi/example-designs/sky130-openroad.yml
+++ b/vlsi/example-designs/sky130-openroad.yml
@@ -3,7 +3,7 @@
 # Specify clock signals
 # Relax the clock period for OpenROAD to meet timing
 vlsi.inputs.clocks: [
-  {name: "clock_uncore_clock", period: "50ns", uncertainty: "2ns"}
+  {name: "clock_uncore", period: "50ns", uncertainty: "2ns"}
 ]
 
 # Flow parameters that yield a routable design with reasonable timing
@@ -54,36 +54,27 @@ vlsi.inputs.placement_constraints:
       bottom: 10
 
   # Place SRAM memory instances
-  - path: "ChipTop/system/tile_prci_domain/tile_reset_domain_tile/dcache/data/data_arrays_0_0/data_arrays_0_0_ext/mem_0_0"
+  # data cache
+  - path: "ChipTop/system/tile_prci_domain/element_reset_domain_rockettile/dcache/data/data_arrays_0/data_arrays_0_ext/mem_0_0"
     type: hardmacro
     x: 50
     y: 50
     orientation: r90
-  - path: "ChipTop/system/tile_prci_domain/tile_reset_domain_tile/dcache/data/data_arrays_0_1/data_arrays_0_0_ext/mem_0_0"
+  - path: "ChipTop/system/tile_prci_domain/element_reset_domain_rockettile/dcache/data/data_arrays_0/data_arrays_0_ext/mem_1_0"
     type: hardmacro
     x: 50
-    y: 450
-    orientation: r90
-  - path: "ChipTop/system/tile_prci_domain/tile_reset_domain_tile/dcache/data/data_arrays_0_2/data_arrays_0_0_ext/mem_0_0"
-    type: hardmacro
-    x: 50
-    y: 850
-    orientation: r90
-  - path: "ChipTop/system/tile_prci_domain/tile_reset_domain_tile/dcache/data/data_arrays_0_3/data_arrays_0_0_ext/mem_0_0"
-    type: hardmacro
-    x: 50
-    y: 1250
+    y: 800
     orientation: r90
 
   # tag array
-  - path: "ChipTop/system/tile_prci_domain/tile_reset_domain_tile/frontend/icache/tag_array_0/tag_array_0_ext/mem_0_0"
+  - path: "ChipTop/system/tile_prci_domain/element_reset_domain_rockettile/frontend/icache/tag_array_0/tag_array_0_ext/mem_0_0"
     type: hardmacro
     x: 50
     y: 1600
     orientation: r90
 
   # instruction cache
-  - path: "ChipTop/system/tile_prci_domain/tile_reset_domain_tile/frontend/icache/data_arrays_0_0/data_arrays_0_0_0_ext/mem_0_0"
+  - path: "ChipTop/system/tile_prci_domain/element_reset_domain_rockettile/frontend/icache/data_arrays_0_0/data_arrays_0_0_ext/mem_0_0"
     type: hardmacro
     x: 50
     y: 2100
diff --git a/vlsi/example-designs/sky130-rocket.yml b/vlsi/example-designs/sky130-rocket.yml
index 2e7c7a5b..7e95e331 100644
--- a/vlsi/example-designs/sky130-rocket.yml
+++ b/vlsi/example-designs/sky130-rocket.yml
@@ -1,7 +1,7 @@
 # Override configurations in ../example-sky130.yml and example-designs
 
 # Specify clock signals
-# Rocket/RocketTile names clock signal "clock" instead of "clock_uncore_clock"
+# Rocket/RocketTile names clock signal "clock" instead of "clock_uncore"
 vlsi.inputs.clocks: [
   {name: "clock", period: "5ns", uncertainty: "1ns"}
 ]
diff --git a/vlsi/example-sky130.yml b/vlsi/example-sky130.yml
index 1e3faf1c..3f45bd3d 100644
--- a/vlsi/example-sky130.yml
+++ b/vlsi/example-sky130.yml
@@ -20,7 +20,7 @@ vlsi.inputs.power_spec_type: "cpf"
 
 # Specify clock signals
 vlsi.inputs.clocks: [
-  {name: "clock_uncore_clock", period: "20ns", uncertainty: "1ns"}
+  {name: "clock_uncore", period: "20ns", uncertainty: "1ns"}
 ]
 
 # Generate Make include to aid in flow
@@ -42,27 +42,27 @@ vlsi.inputs.placement_constraints:
       bottom: 10
 
   # Place SRAM memory instances
-    # data cache
-  - path: "ChipTop/system/tile_prci_domain/tile_reset_domain_tile/dcache/data/data_arrays_0/data_arrays_0_ext/mem_0_0"
+  # data cache
+  - path: "ChipTop/system/tile_prci_domain/element_reset_domain_rockettile/dcache/data/data_arrays_0/data_arrays_0_ext/mem_0_0"
     type: hardmacro
     x: 50
     y: 50
     orientation: r90
-  - path: "ChipTop/system/tile_prci_domain/tile_reset_domain_tile/dcache/data/data_arrays_0/data_arrays_0_ext/mem_1_0"
+  - path: "ChipTop/system/tile_prci_domain/element_reset_domain_rockettile/dcache/data/data_arrays_0/data_arrays_0_ext/mem_1_0"
     type: hardmacro
     x: 50
     y: 800
     orientation: r90
 
   # tag array
-  - path: "ChipTop/system/tile_prci_domain/tile_reset_domain_tile/frontend/icache/tag_array_0/tag_array_0_ext/mem_0_0"
+  - path: "ChipTop/system/tile_prci_domain/element_reset_domain_rockettile/frontend/icache/tag_array_0/tag_array_0_ext/mem_0_0"
     type: hardmacro
     x: 50
     y: 1600
     orientation: r90
 
   # instruction cache
-  - path: "ChipTop/system/tile_prci_domain/tile_reset_domain_tile/frontend/icache/data_arrays_0_0/data_arrays_0_0_ext/mem_0_0"
+  - path: "ChipTop/system/tile_prci_domain/element_reset_domain_rockettile/frontend/icache/data_arrays_0_0/data_arrays_0_0_ext/mem_0_0"
     type: hardmacro
     x: 50
     y: 2100