Add Blackwell tensor core support to Chipyard
Some checks failed
update-circt / update-circt (push) Has been cancelled

- Update RadianceConfigs.scala with Blackwell configurations
- Update Verilator Makefile with optimized build flags
- Update submodules: radiance (Blackwell implementation), gemmini (params update)
- Update build flags and gitignore
This commit is contained in:
2026-05-06 14:52:08 +08:00
parent f8c98496f5
commit ec349a854f
8 changed files with 24 additions and 10 deletions

View File

@@ -88,7 +88,7 @@ class Radiance4CFP16ClusterConfig 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.WithRadianceCores(4, location = InCluster(0), tensorCoreFP16 = true, tensorCoreDecoupled = false, tensorCoreBlackwell = true, useVxCache = false) ++
new radiance.subsystem.WithRadianceCores(4, location = InCluster(0), tensorCoreFP16 = true, tensorCoreDecoupled = false, tensorCoreBlackwell = true, startupAddress = BigInt("80000000", 16), useVxCache = false) ++
new radiance.subsystem.WithRadianceSharedMem(address = x"ff000000", size = 128 << 10, numBanks = 4, numWords = 8) ++
new radiance.subsystem.WithCoalescer(nNewSrcIds = 16) ++
new radiance.subsystem.WithVortexL1Banks(nBanks = 8) ++