Add GPU-initiated and Host-initiated SDMA executors via Anvil - #343

Open

nileshnegi wants to merge 41 commits into

candidate-1.70from

users/nileshnegi/feat/rad-sdma-anvil

Commits on Aug 13, 2026

feat: add anvil KFD/SDMA queue library headers

nileshnegi
and
claude
committed
0a435c8
feat: add anvil.cpp — xio-free adaptation of anvil.hip

nileshnegi
and
claude
committed
cb8a576
build: add ENABLE_ANVIL_EXEC cmake option for GPU-initiated SDMA executor

nileshnegi
and
claude
committed
1b1ec70
feat: add EXE_GPU_INITIATED_DMA enum value (char code 'S')

nileshnegi
and
claude
committed
324d0d9
build: fix hsakmt IMPORTED target — use STATIC for .a, add drm transitive deps

nileshnegi
and
claude
committed
6e4b475
feat: add anvilQueues field to ExeInfo for SDMA queue handles

nileshnegi
and
claude
committed
6fd0022
feat: add TransfersHaveErrors validation for EXE_GPU_INITIATED_DMA

nileshnegi
and
claude
committed
fd0502a
feat: implement PrepareAnvilExecutor and extend Teardown for anvil

nileshnegi
and
claude
committed
e8232c7
feat: implement AnvilTransferKernel and RunAnvilExecutor

nileshnegi
and
claude
committed
26d5483
fix: add gfx1250 s_waitcnt compat in sdma-ep.h; remove kernel no-op guard

nileshnegi
and
claude
committed
8218dcd
fix: make IsGpuExeType unconditional to prevent segfault on gfx950

nileshnegi
and
claude
committed
cfc5fc3
build: add ENABLE_ANVIL_EXEC flag to Makefile

nileshnegi
and
claude
committed
03f4638
build: pass libhsakmt.a via -Wl, to avoid -x hip parsing it as source

nileshnegi
and
claude
committed
1419b3c
feat: add TB_VERBOSE=1 logging for anvil alloc/dealloc lifecycle

nileshnegi
and
claude
committed
dad5aa7
fix: add EXE_GPU_INITIATED_DMA to ExeTypeToStr as "GISDMA"

nileshnegi
and
claude
committed
716c1a2
feat: extend USE_DMA_EXEC in AllToAll preset to support BMA and GISDMA

nileshnegi
and
claude
committed
5d10c46
fix: remove ANVIL_EXEC_ENABLED guard from RecursiveWildcardTransferExpansion switch

nileshnegi
and
claude
committed
062d1e4
fix: remove ANVIL_EXEC_ENABLED guard from GetHsaAgent switch case

nileshnegi
and
claude
committed
90d287b
anvil: suppress nodiscard warnings on hipFree in SdmaQueue dtor

nileshnegi
and
claude
committed
a5fffe3
anvil: add HSA preferred-engine selection + SDMA affinity topology table

nileshnegi
and
claude
committed
44068b9
topology: fix SDMA affinity table column alignment

nileshnegi
and
claude
committed
98fd410
fix: honor byteOffset in DMA and Anvil writes

nileshnegi
and
cursoragent
committed
f174c05
anvil: run GISDMA transfers concurrently with per-transfer timing

nileshnegi
and
cursoragent
committed
9d4fc48
anvil: add opt-in round-robin SDMA engine selection

nileshnegi
and
cursoragent
committed
d542d0b
anvil: use hipExtLaunchKernelGGL for GISDMA kernel timing

nileshnegi
and
cursoragent
committed
151769f
anvil: use engine-ordered signal for GISDMA completion

nileshnegi
and
cursoragent
committed
ebb9301
anvil: add runtime toggles for GISDMA doorbell placement and completion wait

nileshnegi
and
cursoragent
committed
09c87b2
anvil: add ANVIL_LEGACY toggle for legacy put/quiet GISDMA completion

nileshnegi
and
cursoragent
committed
0ce43f7
anvil: split GISDMA kernel into four non-branching variants

nileshnegi
and
cursoragent
committed
b61b3e1
anvil: chunk GISDMA copies to lift the 1 GiB per-transfer limit

nileshnegi
and
cursoragent
committed
6fa3874
anvil: add host-initiated (CPU-driven) SDMA queue path

nileshnegi
and
cursoragent
committed
1aa6b4e
anvil: add OSS7 fused copy+signal SDMA packet (gfx1250)

nileshnegi
and
cursoragent
committed
e533b6c
build: enable XIO_SDMA_OSS7 fused packets by default (all-arch safe)

nileshnegi
and
cursoragent
committed
45c36c1
anvil: rename GISDMA executor to GMA and split host path into HMA

nileshnegi
and
cursoragent
committed
3c7f986
docs: document GMA/HMA anvil SDMA executors in CHANGELOG

nileshnegi
and
cursoragent
committed
8a36010
anvil: address PR #343 review feedback in vendored SDMA layer

nileshnegi
and
cursoragent
committed
49dae56
a2a: batch each source GPU's BMA copies into one launch

nileshnegi
and
cursoragent
committed
7e82b52
core: scale executor byte total by destination count

nileshnegi
and
cursoragent
committed
9fb2b4b
poda2a: add BMA/GMA/HMA executor support

nileshnegi
and
cursoragent
committed
4587bd0

Commits on Aug 15, 2026

anvil: reuse GMA SDMA queues across re-prepares to fix sweep queue exhaustion

nileshnegi
and
cursoragent
committed
e4a4222
anvil: tighten comments across the SDMA executor changes

nileshnegi
and
cursoragent
committed
dcac4d8