Skip to content

feat: bump nvidia-cutlass-dsl to >=4.4.2 #2833

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
limin2021 wants to merge 5 commits into
base: main
Choose a base branch
from
+79 −61
Open

feat: bump nvidia-cutlass-dsl to >=4.4.2 #2833

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
9c8902f
feat: bump nvidia-cutlass-dsl minimum version to 4.4.2
limin2021 Mar 20, 2026
aa17f54
fix: use nvidia-cutlass-dsl[cu13] extra for CUDA 13 libs
limin2021 Mar 20, 2026
ae215a6
fix: auto-detect CUDA version for nvidia-cutlass-dsl extra and clean …
limin2021 Mar 20, 2026
bc3b93c
fix: CUDA 12 needs no extra, only CUDA 13 needs [cu13] for cutlass-dsl
limin2021 Mar 20, 2026
0c966f1
fix cute dsl failure after 4.4.2
nv-yunzheq Mar 21, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions ci/setup_python.env
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,3 +15,6 @@

# Uncomment to override TVM-FFI version:
# TVM_FFI_REF=

# Uncomment to override nvidia-cutlass-dsl version:
CUTLASS_DSL_VERSION=4.4.2
Copy link
Collaborator

@yzh119 yzh119 Mar 20, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please remove it after the CI passed.

11 changes: 4 additions & 7 deletions flashinfer/cute_dsl/gemm_allreduce_two_shot.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,6 @@ def spin_lock_multimem_arrive(lock_ptr: Pointer, loc=None, ip=None) -> None:

# HACK https://github.com/NVIDIA/cutlass/issues/2845
from cutlass._mlir.dialects import nvvm
from cutlass.cutlass_dsl import T
from cutlass._mlir.dialects.nvvm import (
MemOrderKind,
MemScopeKind,
Expand All @@ -56,7 +55,6 @@ def spin_lock_atom_cas_acquire_wait(
result = 0
while result != expected_val:
result = nvvm.atomicrmw(
T.i32(),
AtomicOpKind.CAS,
lock_ptr.llvm_ptr,
Int32(reset_val).ir_value(loc=loc, ip=ip),
Expand All @@ -70,7 +68,6 @@ def spin_lock_atom_cas_acquire_wait(
result = 0
while result != expected_val:
result = nvvm.atomicrmw(
T.i32(),
AtomicOpKind.CAS,
lock_ptr.llvm_ptr,
Int32(reset_val).ir_value(loc=loc, ip=ip),
Expand All @@ -92,7 +89,7 @@ def sm_wise_inter_gpu_multimem_barrier(
bdimx, bdimy, _ = cute.arch.grid_dim()
pid = bidx + bidy * bdimx + bidz * bdimx * bdimy
distributed.multimem_red_release_sys_add1(barrier_mc + pid, loc=loc, ip=ip)
cute.arch.fence_proxy(cute.arch.ProxyKind.alias)
cute.arch.fence_proxy("alias")

# v4.3.1 does not have mem_order="acquire" variant in `distributed` module
# filed issue https://github.com/NVIDIA/cutlass/issues/2845
Expand Down Expand Up @@ -1251,8 +1248,8 @@ def kernel(
)
# Fence and barrier to make sure shared memory store is visible to TMA store
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
epilog_threads = 32 * len(self.epilog_warp_id)
cute.arch.barrier(
Expand Down Expand Up @@ -1312,7 +1309,7 @@ def kernel(
flag = barrier_flag_mc.iterator + tile_id
cute.arch.fence_acq_rel_gpu()
spin_lock_multimem_arrive(flag)
cute.arch.fence_proxy(cute.arch.ProxyKind.alias)
cute.arch.fence_proxy("alias")

#
# Advance to next tile
Expand Down
56 changes: 28 additions & 28 deletions .../fused_moe/cute_dsl/blackwell/blockscaled_contiguous_gather_grouped_gemm_swiglu_fusion.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1512,8 +1512,8 @@ def kernel(
sInfo[(4, tile_info_producer_state.index)] = mn_limit
# fence view async shared
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)

self.sched_sync_barrier.arrive_and_wait()
Expand Down Expand Up @@ -1548,8 +1548,8 @@ def kernel(
sInfo[(4, tile_info_producer_state.index)] = mn_limit
# fence view async shared
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)

self.sched_sync_barrier.arrive_and_wait()
Expand All @@ -1569,8 +1569,8 @@ def kernel(
sInfo[(3, tile_info_producer_state.index)] = cutlass.Int32(0)
sInfo[(4, tile_info_producer_state.index)] = -1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
self.sched_sync_barrier.arrive_and_wait()
tile_info_pipeline.producer_commit(tile_info_producer_state)
Expand Down Expand Up @@ -1669,8 +1669,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -1844,8 +1844,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -1886,8 +1886,8 @@ def kernel(
valid_tile_info[0] = sInfo[(3, tile_info_consumer_state.index)]
is_valid_tile = valid_tile_info[0] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -1927,8 +1927,8 @@ def kernel(
valid_tile_info[0] = sInfo[(3, tile_info_consumer_state.index)]
is_valid_tile = valid_tile_info[0] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -1968,8 +1968,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -2051,8 +2051,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -2152,8 +2152,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -2368,8 +2368,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -2480,8 +2480,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -2811,8 +2811,8 @@ def kernel(
)
# Fence and barrier to make sure shared memory store is visible to TMA store
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
self.epilog_sync_barrier.arrive_and_wait()
#
Expand Down Expand Up @@ -2845,8 +2845,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down
44 changes: 22 additions & 22 deletions ...infer/fused_moe/cute_dsl/blackwell/blockscaled_contiguous_grouped_gemm_finalize_fusion.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1380,8 +1380,8 @@ def kernel(
sInfo[(4, tile_info_producer_state.index)] = mn_limit
# fence view async shared
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)

self.sched_sync_barrier.arrive_and_wait()
Expand Down Expand Up @@ -1416,8 +1416,8 @@ def kernel(
sInfo[(4, tile_info_producer_state.index)] = mn_limit
# fence view async shared
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)

self.sched_sync_barrier.arrive_and_wait()
Expand All @@ -1438,8 +1438,8 @@ def kernel(
sInfo[(3, tile_info_producer_state.index)] = cutlass.Int32(0)
sInfo[(4, tile_info_producer_state.index)] = cutlass.Int32(0)
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
self.sched_sync_barrier.arrive_and_wait()
tile_info_pipeline.producer_commit(tile_info_producer_state)
Expand Down Expand Up @@ -1467,8 +1467,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -1573,8 +1573,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -1659,8 +1659,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -1818,8 +1818,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -1886,8 +1886,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down Expand Up @@ -2023,8 +2023,8 @@ def kernel(

if cutlass.const_expr(self.use_blkred):
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
#
# Async arrive accumulator buffer empty
Expand All @@ -2037,8 +2037,8 @@ def kernel(

if cutlass.const_expr(self.use_blkred):
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
if is_valid_row:
coord_n = mma_tile_coord_mnl[1] * self.cta_tile_shape_mnk[1]
Expand Down Expand Up @@ -2073,8 +2073,8 @@ def kernel(
tile_info[idx] = sInfo[(idx, tile_info_consumer_state.index)]
is_valid_tile = tile_info[3] == 1
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
tile_info_pipeline.consumer_release(tile_info_consumer_state)
tile_info_consumer_state.advance()
Expand Down
1 change: 0 additions & 1 deletion flashinfer/fused_moe/cute_dsl/blackwell/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -208,7 +208,6 @@ def fmin(
) -> cutlass.Float32:
return cutlass.Float32(
nvvm.fmin(
T.f32(),
cutlass.Float32(a).ir_value(loc=loc, ip=ip),
cutlass.Float32(b).ir_value(loc=loc, ip=ip),
nan=nan,
Expand Down
4 changes: 2 additions & 2 deletions flashinfer/gemm/kernels/dense_blockscaled_gemm_sm100.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1469,8 +1469,8 @@ def kernel(
)
# Fence and barrier to make sure shared memory store is visible to TMA store
cute.arch.fence_proxy(
cute.arch.ProxyKind.async_shared,
space=cute.arch.SharedSpace.shared_cta,
"async.shared",
space="cta",
)
epilog_threads = 32 * len(self.epilog_warp_id)
cute.arch.barrier(
Expand Down
2 changes: 1 addition & 1 deletion requirements.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ einops
ninja
numpy
nvidia-cudnn-frontend>=1.13.0
nvidia-cutlass-dsl>=4.3.4
nvidia-cutlass-dsl>=4.4.2
nvidia-ml-py
packaging>=24.2
requests
Expand Down
19 changes: 19 additions & 0 deletions scripts/setup_test_env.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,3 +23,22 @@ if [ -n "${TVM_FFI_REF:-}" ]; then
echo "TVM-FFI override complete."
echo ""
fi

# Override nvidia-cutlass-dsl if specified
if [ -n "${CUTLASS_DSL_VERSION:-}" ]; then
# Detect CUDA major version: only CUDA 13+ needs [cu13] extra
CUDA_MAJOR=$(python -c "import torch; print(torch.version.cuda.split('.')[0])" 2>/dev/null || echo "12")
if [ "$CUDA_MAJOR" = "13" ]; then
CUTLASS_DSL_PKG="nvidia-cutlass-dsl[cu13]==${CUTLASS_DSL_VERSION}"
else
CUTLASS_DSL_PKG="nvidia-cutlass-dsl==${CUTLASS_DSL_VERSION}"
fi
echo "========================================"
echo "Overriding nvidia-cutlass-dsl with: ${CUTLASS_DSL_PKG}"
echo "========================================"
# Clean uninstall old packages first (recommended by NVIDIA docs)
pip uninstall nvidia-cutlass-dsl nvidia-cutlass-dsl-libs-base nvidia-cutlass-dsl-libs-cu12 nvidia-cutlass-dsl-libs-cu13 -y 2>/dev/null || true
pip install "${CUTLASS_DSL_PKG}"
echo "nvidia-cutlass-dsl override complete."
echo ""
fi
Loading