leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

v4.5 dev update. (#3153)

Browse Source
This commit is contained in:
Junkai-Wu
2026-04-08 00:16:05 +08:00
committed by GitHub
parent 418d38a5de
commit a221da7ccf
265 changed files with 4913 additions and 1478 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
examples/python/CuTeDSL/experimental/blackwell/dense_gemm_cute_pipeline.py
View File

@@ -581,7 +581,7 @@ class PersistentDenseGemmKernel:
acc_full_mbar_ptr: cute.struct.MemRange[
cutlass.Int64, self.num_acc_stage * 2
]
tmem_dealloc_mbar_ptr: cutlass.Int64
tmem_dealloc_mbar: cutlass.Int64
tmem_holding_buf: cutlass.Int32
smem = utils.SmemAllocator()
@@ -632,11 +632,11 @@ class PersistentDenseGemmKernel:
)
# Tensor memory dealloc barrier init
tmem = utils.TmemAllocator(
storage.tmem_holding_buf,
storage.tmem_holding_buf.ptr,
barrier_for_retrieve=tmem_alloc_barrier,
allocator_warp_id=self.epilogue_warp_id[0],
is_two_cta=use_2cta_instrs,
two_cta_tmem_dealloc_mbar_ptr=storage.tmem_dealloc_mbar_ptr,
two_cta_tmem_dealloc_mbar_ptr=storage.tmem_dealloc_mbar.ptr,
)
# Cluster arrive after barrier init