wu-arch/chipyard
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Compare commits

base: wu-arch:ec349a854f75017fcf57db00c7bbcd6de5a135e9
Branches Tags
wu-arch:wu-architecture wu-arch:pre-wu-blackwell wu-arch:wu-blackwell-fp8-e4m3 wu-arch:wu-tmem-bank-scalar-softmax wu-arch:nvidia-style-arch wu-arch:feature/scalar-tmem-softmax-vector wu-arch:wu-blackwell-4lane-tc wu-arch:asplos-ae
..
compare: wu-arch:wu-tmem-bank-scalar-softmax
Branches Tags
wu-arch:pre-wu-blackwell wu-arch:wu-blackwell-fp8-e4m3 wu-arch:wu-tmem-bank-scalar-softmax wu-arch:nvidia-style-arch wu-arch:feature/scalar-tmem-softmax-vector wu-arch:wu-blackwell-4lane-tc wu-arch:wu-architecture wu-arch:asplos-ae

12 Commits
ec349a854f ... wu-tmem-ba

Author SHA1 Message Date
Zhongdi LUO
3c9ced31d5 config: reduce Radiance shared memory to 32 KiB 2026-07-13 09:05:36 +00:00
Zhongdi LUO
02d92f6f88 feat: add blackwell synthesis configuration 2026-07-13 00:04:54 +00:00
Zhongdi LUO
fc32e7f59c chore: update radiance pipelined softmax 2026-07-12 02:00:36 +00:00
Zhongdi LUO
71c84f63a5 chore: update radiance fp16 scalar tmem softmax 2026-07-10 13:03:54 +00:00
Zhongdi LUO
55b7fa2007 chore: update radiance fexp support 2026-07-02 07:26:14 +00:00
Zhongdi LUO
bdca34b2cc Update Radiance for scalar TMEM support 2026-06-24 06:25:39 +00:00
Zhongdi LUO
b0a79c26ee Update radiance pre-TMEM-bank baseline 2026-06-21 08:20:52 +00:00
Zhongdi LUO
d764de5115 Update radiance Blackwell SMEM alignment fix 2026-05-27 08:44:49 +00:00
Zhongdi LUO
3cb58af336 Update radiance for 4-lane Blackwell tensor core 2026-05-27 05:55:17 +00:00
abnerhexu
87783dc9f6 Track radiance WU architecture updates
Some checks failed
update-circt / update-circt (push) Has been cancelled
Details
2026-05-27 09:08:16 +08:00
Zhongdi LUO
9206c4452f Update submodules and gitmodules configuration
Some checks failed
update-circt / update-circt (push) Has been cancelled
Details
2026-05-26 12:58:22 +00:00
abnerhexu
5ce98e1491 Update WU architecture integration 2026-05-25 19:26:23 +08:00
4 changed files with 30 additions and 24 deletions
Expand all files Collapse all files

14
.gitmodules vendored
View File

@@ -87,22 +87,22 @@
url = https://github.com/ucb-bar/libgloss-htif.git url = https://github.com/ucb-bar/libgloss-htif.git
[submodule "toolchains/riscv-tools/riscv-isa-sim"] [submodule "toolchains/riscv-tools/riscv-isa-sim"]
path = toolchains/riscv-tools/riscv-isa-sim path = toolchains/riscv-tools/riscv-isa-sim
url = https://github.com/richardyrh/riscv-isa-sim url = https://gh-proxy.org/https://github.com/richardyrh/riscv-isa-sim
[submodule "toolchains/riscv-tools/riscv-openocd"] [submodule "toolchains/riscv-tools/riscv-openocd"]
path = toolchains/riscv-tools/riscv-openocd path = toolchains/riscv-tools/riscv-openocd
url = https://github.com/riscv/riscv-openocd.git url = https://gh-proxy.org/https://github.com/riscv/riscv-openocd.git
[submodule "toolchains/riscv-tools/riscv-pk"] [submodule "toolchains/riscv-tools/riscv-pk"]
path = toolchains/riscv-tools/riscv-pk path = toolchains/riscv-tools/riscv-pk
url = https://github.com/riscv-software-src/riscv-pk.git url = https://gh-proxy.org/https://github.com/riscv-software-src/riscv-pk.git
[submodule "toolchains/riscv-tools/riscv-spike-devices"] [submodule "toolchains/riscv-tools/riscv-spike-devices"]
path = toolchains/riscv-tools/riscv-spike-devices path = toolchains/riscv-tools/riscv-spike-devices
url = https://github.com/ucb-bar/spike-devices.git url = https://gh-proxy.org/https://github.com/ucb-bar/spike-devices.git
[submodule "toolchains/riscv-tools/riscv-tests"] [submodule "toolchains/riscv-tools/riscv-tests"]
path = toolchains/riscv-tools/riscv-tests path = toolchains/riscv-tools/riscv-tests
url = https://github.com/riscv-software-src/riscv-tests.git url = https://gh-proxy.org/https://github.com/riscv-software-src/riscv-tests.git
[submodule "toolchains/riscv-tools/riscv-tools-feedstock"] [submodule "toolchains/riscv-tools/riscv-tools-feedstock"]
path = toolchains/riscv-tools/riscv-tools-feedstock path = toolchains/riscv-tools/riscv-tools-feedstock
url = https://github.com/ucb-bar/riscv-tools-feedstock.git url = https://gh-proxy.org/https://github.com/ucb-bar/riscv-tools-feedstock.git
[submodule "tools/DRAMSim2"] [submodule "tools/DRAMSim2"]
path = tools/DRAMSim2 path = tools/DRAMSim2
url = https://github.com/firesim/DRAMSim2.git url = https://github.com/firesim/DRAMSim2.git
@@ -159,4 +159,4 @@
url = https://github.com/ucb-bar/compress-acc.git url = https://github.com/ucb-bar/compress-acc.git
[submodule "generators/radiance"] [submodule "generators/radiance"]
path = generators/radiance path = generators/radiance
url = https://github.com/ucb-bar/radiance.git url = https://git.nudt.space/wu-arch/radiance.git

22
generators/chipyard/src/main/scala/config/RadianceConfigs.scala
View File

@@ -27,6 +27,7 @@ class VirgoConfig extends RadianceClusterConfig
class VirgoFP16Config extends RadianceFP16ClusterConfig class VirgoFP16Config extends RadianceFP16ClusterConfig
class VirgoHopperConfig extends Radiance4CFP16ClusterConfig class VirgoHopperConfig extends Radiance4CFP16ClusterConfig
class VirgoBlackwellConfig extends RadianceBlackwellClusterConfig class VirgoBlackwellConfig extends RadianceBlackwellClusterConfig
class VirgoBlackwellSynConfig extends RadianceBlackwellClusterSynConfig
class VirgoFlashConfig extends RadianceClusterConfig class VirgoFlashConfig extends RadianceClusterConfig
class VirgoSynConfig extends RadianceClusterSynConfig class VirgoSynConfig extends RadianceClusterSynConfig
class VirgoFP16SynConfig extends RadianceFP16ClusterSynConfig class VirgoFP16SynConfig extends RadianceFP16ClusterSynConfig
@@ -57,7 +58,7 @@ class RadianceBaseConfig extends Config(
class RadianceFP16ClusterConfig extends Config( class RadianceFP16ClusterConfig extends Config(
new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 16, accSizeInKB = 32, tileSize = (8, 4, 8), dataType = RadianceGemminiDataType.FP16) ++ new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 16, accSizeInKB = 32, tileSize = (8, 4, 8), dataType = RadianceGemminiDataType.FP16) ++
new radiance.subsystem.WithRadianceCores(8, location = InCluster(0), tensorCoreFP16 = true, tensorCoreDecoupled = false, useVxCache = false) ++ new radiance.subsystem.WithRadianceCores(8, location = InCluster(0), tensorCoreFP16 = true, tensorCoreDecoupled = false, useVxCache = false) ++
new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 128 << 10, numBanks = 4, numWords = 16) ++ new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 32 << 10, numBanks = 4, numWords = 16) ++
new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++ new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++
new radiance.subsystem.WithVortexL1Banks(nBanks = 8) ++ new radiance.subsystem.WithVortexL1Banks(nBanks = 8) ++
new radiance.subsystem.WithRadianceCluster(0) ++ new radiance.subsystem.WithRadianceCluster(0) ++
@@ -66,7 +67,7 @@ class RadianceFP16ClusterConfig extends Config(
class Radiance8B8WFP16ClusterConfig extends Config( class Radiance8B8WFP16ClusterConfig extends Config(
new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 16, accSizeInKB = 32, tileSize = (8, 4, 8), dataType = RadianceGemminiDataType.FP16) ++ new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 16, accSizeInKB = 32, tileSize = (8, 4, 8), dataType = RadianceGemminiDataType.FP16) ++
new radiance.subsystem.WithRadianceCores(8, location = InCluster(0), tensorCoreFP16 = true, tensorCoreDecoupled = false, useVxCache = false) ++ new radiance.subsystem.WithRadianceCores(8, location = InCluster(0), tensorCoreFP16 = true, tensorCoreDecoupled = false, useVxCache = false) ++
new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 128 << 10, numBanks = 8, numWords = 8) ++ new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 32 << 10, numBanks = 8, numWords = 8) ++
new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++ new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++
new radiance.subsystem.WithVortexL1Banks(nBanks = 8) ++ new radiance.subsystem.WithVortexL1Banks(nBanks = 8) ++
new radiance.subsystem.WithRadianceCluster(0) ++ new radiance.subsystem.WithRadianceCluster(0) ++
@@ -75,12 +76,12 @@ class Radiance8B8WFP16ClusterConfig extends Config(
class Radiance4CFP16ClusterConfig extends Config( class Radiance4CFP16ClusterConfig extends Config(
new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 16, accSizeInKB = 32, tileSize = (8, 4, 8), dataType = RadianceGemminiDataType.FP16) ++ new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 16, accSizeInKB = 32, tileSize = (8, 4, 8), dataType = RadianceGemminiDataType.FP16) ++
new radiance.subsystem.WithRadianceCores(4, location = InCluster(0), tensorCoreFP16 = true, tensorCoreDecoupled = true, useVxCache = false) ++ new radiance.subsystem.WithRadianceCores(4, location = InCluster(0), tensorCoreFP16 = true, tensorCoreDecoupled = true, useVxCache = false) ++
// new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 128 << 10, numBanks = 4, numWords = 16, // new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 32 << 10, numBanks = 4, numWords = 16,
// memType = radiance.subsystem.TwoReadOneWrite, // memType = radiance.subsystem.TwoReadOneWrite,
// serializeUnaligned = radiance.subsystem.CoreSerialized) ++ // serializeUnaligned = radiance.subsystem.CoreSerialized) ++
// NOTE: Hopper Tensor Core does not work with 16-word config due to the // NOTE: Hopper Tensor Core does not work with 16-word config due to the
// address alignment requirement // address alignment requirement
new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 128 << 10, numBanks = 4, numWords = 8) ++ new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 32 << 10, numBanks = 4, numWords = 8) ++
new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++ new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++
new radiance.subsystem.WithVortexL1Banks(nBanks = 8) ++ new radiance.subsystem.WithVortexL1Banks(nBanks = 8) ++
new radiance.subsystem.WithRadianceCluster(0) ++ new radiance.subsystem.WithRadianceCluster(0) ++
@@ -88,11 +89,12 @@ class Radiance4CFP16ClusterConfig extends Config(
class RadianceBlackwellClusterConfig extends Config( class RadianceBlackwellClusterConfig extends Config(
new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 16, accSizeInKB = 32, tileSize = (8, 4, 8), dataType = RadianceGemminiDataType.FP16) ++ new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 16, accSizeInKB = 32, tileSize = (8, 4, 8), dataType = RadianceGemminiDataType.FP16) ++
new radiance.subsystem.WithRadianceCores(4, location = InCluster(0), tensorCoreFP16 = true, tensorCoreDecoupled = false, tensorCoreBlackwell = true, startupAddress = BigInt("80000000", 16), useVxCache = false) ++ new radiance.subsystem.WithRadianceCores(1, location = InCluster(0), tensorCoreFP16 = true, tensorCoreDecoupled = false, tensorCoreBlackwell = true, numTensorWarps = 2, startupAddress = BigInt("80000000", 16), useVxCache = false) ++
new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 128 << 10, numBanks = 4, numWords = 8) ++ new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 32 << 10, numBanks = 4, numWords = 8) ++
new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++ new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++
new radiance.subsystem.WithVortexL1Banks(nBanks = 8) ++ new radiance.subsystem.WithVortexL1Banks(nBanks = 8) ++
new radiance.subsystem.WithRadianceCluster(0) ++ new radiance.subsystem.WithRadianceCluster(0) ++
new radiance.subsystem.WithSimtConfig(nWarps = 4, nCoreLanes = 4, nMemLanes = 4, nSrcIds = 32) ++
new RadianceBaseConfig) new RadianceBaseConfig)
class RadianceClusterConfig extends Config( class RadianceClusterConfig extends Config(
@@ -100,7 +102,7 @@ class RadianceClusterConfig extends Config(
new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 8, accSizeInKB = 16, tileSize = 8) ++ new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 8, accSizeInKB = 16, tileSize = 8) ++
new radiance.subsystem.WithRadianceCores(4, location = InCluster(0), tensorCoreFP16 = false, tensorCoreDecoupled = false, useVxCache = false) ++ new radiance.subsystem.WithRadianceCores(4, location = InCluster(0), tensorCoreFP16 = false, tensorCoreDecoupled = false, useVxCache = false) ++
// new radiance.subsystem.WithRadianceFrameBuffer(x"ff018000", 16, 0x8000, x"ff011000", "fb0") ++ // new radiance.subsystem.WithRadianceFrameBuffer(x"ff018000", 16, 0x8000, x"ff011000", "fb0") ++
new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 256 << 10/*KBytes*/, numBanks = 8, numWords = 8, new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 32 << 10/*KBytes*/, numBanks = 8, numWords = 8,
// memType = radiance.subsystem.TwoReadOneWrite, // memType = radiance.subsystem.TwoReadOneWrite,
serializeUnaligned = radiance.subsystem.CoreSerialized) ++ serializeUnaligned = radiance.subsystem.CoreSerialized) ++
new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++ new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++
@@ -133,7 +135,7 @@ class RadianceBigLittleClusterConfig extends Config(
new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 4, accSizeInKB = 16, tileSize = 16) ++ new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 4, accSizeInKB = 16, tileSize = 16) ++
new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 8, accSizeInKB = 16, tileSize = 8) ++ new radiance.subsystem.WithRadianceGemmini(location = InCluster(0), dim = 8, accSizeInKB = 16, tileSize = 8) ++
new radiance.subsystem.WithRadianceCores(2, location = InCluster(0), useVxCache = false) ++ new radiance.subsystem.WithRadianceCores(2, location = InCluster(0), useVxCache = false) ++
new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 64 << 10, numBanks = 4, numWords = 8) ++ new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 32 << 10, numBanks = 4, numWords = 8) ++
new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++ new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++
new radiance.subsystem.WithVortexL1Banks(nBanks = 8)++ new radiance.subsystem.WithVortexL1Banks(nBanks = 8)++
new radiance.subsystem.WithRadianceCluster(0) ++ new radiance.subsystem.WithRadianceCluster(0) ++
@@ -151,6 +153,10 @@ class Radiance4CFP16ClusterSynConfig extends Config(
new radiance.subsystem.WithRadianceSimParams(false) ++ new radiance.subsystem.WithRadianceSimParams(false) ++
new Radiance4CFP16ClusterConfig) new Radiance4CFP16ClusterConfig)
class RadianceBlackwellClusterSynConfig extends Config(
new radiance.subsystem.WithRadianceSimParams(false) ++
new RadianceBlackwellClusterConfig)
class RadianceBigLittleClusterSynConfig extends Config( class RadianceBigLittleClusterSynConfig extends Config(
new radiance.subsystem.WithRadianceSimParams(false) ++ new radiance.subsystem.WithRadianceSimParams(false) ++
new RadianceBigLittleClusterConfig) new RadianceBigLittleClusterConfig)

2
generators/radiance

Submodule generators/radiance updated: 5112f3665a...035534a486

16
sims/verilator/Makefile
View File

@@ -104,8 +104,9 @@ get_waveform_flag = +vcdfile=$(1).$(if $(filter $(USE_FST),0),vcd,fst)
#---------------------------------------------------------------------------------------- #----------------------------------------------------------------------------------------
# we initially had --noassert for performance, but several modules use # we initially had --noassert for performance, but several modules use
# assertions, including dramsim, so we enable --assert by default # assertions, including dramsim, so we enable --assert by default
VERILATOR_OUTPUT_SPLIT ?= 10000 VERILATOR_OUTPUT_SPLIT ?= 2000
VERILATOR_OUTPUT_SPLIT_CFUNCS ?= 100 VERILATOR_OUTPUT_SPLIT_CFUNCS ?= 20
VERILATOR_OUTPUT_GROUPS ?= 0
VERILATOR_OPT_FLAGS ?= \ VERILATOR_OPT_FLAGS ?= \
-O0 \ -O0 \
@@ -169,6 +170,7 @@ VERILATOR_NONCC_OPTS = \
VERILATOR_CXXFLAGS = \ VERILATOR_CXXFLAGS = \
$(SIM_CXXFLAGS) \ $(SIM_CXXFLAGS) \
$(RUNTIME_PROFILING_CFLAGS) \ $(RUNTIME_PROFILING_CFLAGS) \
-O0 \
-fno-inline \ -fno-inline \
-DVERILATOR -DVERILATOR
@@ -199,25 +201,23 @@ model_mk_debug = $(model_dir_debug)/V$(TB).mk
# build makefile fragment that builds the verilator sim rules # build makefile fragment that builds the verilator sim rules
######################################################################################### #########################################################################################
$(model_mk): $(sim_common_files) $(EXTRA_SIM_REQS) $(model_mk): $(sim_common_files) $(EXTRA_SIM_REQS)
rm -rf $(model_dir)
mkdir -p $(model_dir) mkdir -p $(model_dir)
$(VERILATOR) -j 36 $(VERILATOR_OPTS) $(EXTRA_SIM_SOURCES) -o $(sim) -Mdir $(model_dir) $(VERILATOR) -j $(VERILATOR_MAKE_JOBS) $(VERILATOR_OPTS) $(EXTRA_SIM_SOURCES) -o $(sim) -Mdir $(model_dir)
touch $@ touch $@
$(model_mk_debug): $(sim_common_files) $(EXTRA_SIM_REQS) $(model_mk_debug): $(sim_common_files) $(EXTRA_SIM_REQS)
rm -rf $(model_dir_debug)
mkdir -p $(model_dir_debug) mkdir -p $(model_dir_debug)
$(VERILATOR) -j 36 $(VERILATOR_OPTS) +define+DEBUG $(EXTRA_SIM_SOURCES) -o $(sim_debug) $(TRACING_OPTS) -Mdir $(model_dir_debug) $(VERILATOR) -j $(VERILATOR_MAKE_JOBS) $(VERILATOR_OPTS) +define+DEBUG $(EXTRA_SIM_SOURCES) -o $(sim_debug) $(TRACING_OPTS) -Mdir $(model_dir_debug)
touch $@ touch $@
######################################################################################### #########################################################################################
# invoke make to make verilator sim rules # invoke make to make verilator sim rules
######################################################################################### #########################################################################################
$(sim): $(model_mk) $(dramsim_lib) $(sim): $(model_mk) $(dramsim_lib)
$(MAKE) -j $(VERILATOR_MAKE_JOBS) VM_PARALLEL_BUILDS=1 -C $(model_dir) -f V$(TB).mk $(MAKE) -j $(VERILATOR_MAKE_JOBS) VM_PARALLEL_BUILDS=1 OPT_FAST=-O0 OPT_SLOW=-O0 OPT_GLOBAL=-O0 -C $(model_dir) -f V$(TB).mk
$(sim_debug): $(model_mk_debug) $(dramsim_lib) $(sim_debug): $(model_mk_debug) $(dramsim_lib)
$(MAKE) -j $(VERILATOR_MAKE_JOBS) VM_PARALLEL_BUILDS=1 -C $(model_dir_debug) -f V$(TB).mk $(MAKE) -j $(VERILATOR_MAKE_JOBS) VM_PARALLEL_BUILDS=1 OPT_FAST=-O0 OPT_SLOW=-O0 OPT_GLOBAL=-O0 -C $(model_dir_debug) -f V$(TB).mk
######################################################################################### #########################################################################################
# create a verilator vpd rule # create a verilator vpd rule