Skip to content
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.

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

Add support for small page sizes #824

Open
wants to merge 19 commits into
base: main
Choose a base branch
from
Open
Changes from 1 commit
Commits
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
Prev Previous commit
Next Next commit
only test trivial block size
skrider committed Mar 26, 2024

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
commit 409431b812bf5ef85096232552735a8ae5b2f87c
2 changes: 1 addition & 1 deletion csrc/cutlass
Submodule cutlass updated 61 files
+2 −7 CHANGELOG.md
+3 −24 CMakeLists.txt
+0 −7 PUBLICATIONS.md
+4 −9 README.md
+38 −0 cmake/version.h.in
+0 −34 cmake/version_extended.h.in
+0 −1 examples/02_dump_reg_shmem/CMakeLists.txt
+2 −2 examples/08_turing_tensorop_gemm/turing_tensorop_gemm.cu
+7 −7 examples/56_hopper_ptr_array_batched_gemm/56_hopper_ptr_array_batched_gemm.cu
+8 −10 examples/56_hopper_ptr_array_batched_gemm/CMakeLists.txt
+49 −96 examples/57_hopper_grouped_gemm/57_hopper_grouped_gemm.cu
+0 −10 examples/57_hopper_grouped_gemm/CMakeLists.txt
+1 −1 include/cute/arch/copy_sm90_desc.hpp
+0 −2 include/cute/atom/mma_atom.hpp
+2 −2 include/cute/util/print.hpp
+0 −3 include/cute/util/type_traits.hpp
+0 −4 include/cutlass/arch/mma_sm90.h
+0 −1 include/cutlass/bfloat16.h
+1 −35 include/cutlass/detail/layout.hpp
+7 −12 include/cutlass/epilogue/collective/builders/sm90_builder.inl
+0 −1 include/cutlass/epilogue/collective/default_epilogue.hpp
+18 −32 include/cutlass/epilogue/collective/default_epilogue_array.hpp
+38 −76 include/cutlass/epilogue/collective/sm90_epilogue_tma_warpspecialized.hpp
+2 −1 include/cutlass/epilogue/dispatch_policy.hpp
+0 −28 include/cutlass/epilogue/fusion/sm90_callbacks_tma_warpspecialized.hpp
+0 −1 include/cutlass/epilogue/fusion/sm90_visitor_store_tma_warpspecialized.hpp
+12 −57 include/cutlass/epilogue/thread/linear_combination.h
+183 −0 include/cutlass/epilogue/threadblock/default_epilogue_tensor_op_row_broadcast.h
+519 −0 include/cutlass/epilogue/threadblock/predicated_tile_iterator_row_broadcast.h
+8 −4 include/cutlass/gemm/collective/builders/sm90_gmma_builder.inl
+29 −45 include/cutlass/gemm/collective/sm90_mma_array_tma_gmma_ss_warpspecialized.hpp
+514 −0 include/cutlass/gemm/device/gemm_sparse_row_broadcast.h
+7 −4 include/cutlass/gemm/dispatch_policy.hpp
+0 −12 include/cutlass/gemm/group_array_problem_shape.hpp
+191 −0 include/cutlass/gemm/kernel/default_gemm_sparse_row_broadcast.h
+35 −30 include/cutlass/gemm/kernel/sm90_gemm_array_tma_warpspecialized_cooperative.hpp
+7 −5 include/cutlass/gemm/kernel/sm90_gemm_tma.hpp
+7 −5 include/cutlass/gemm/kernel/sm90_gemm_tma_warpspecialized.hpp
+7 −5 include/cutlass/gemm/kernel/sm90_gemm_tma_warpspecialized_cooperative.hpp
+7 −5 include/cutlass/gemm/kernel/sm90_gemm_tma_warpspecialized_pingpong.hpp
+7 −5 include/cutlass/gemm/kernel/sm90_gemm_warpspecialized.hpp
+7 −5 include/cutlass/gemm/kernel/sm90_gemm_warpspecialized_cooperative.hpp
+7 −5 include/cutlass/gemm/kernel/sm90_gemm_warpspecialized_pingpong.hpp
+86 −140 include/cutlass/gemm/kernel/sm90_tile_scheduler_group.hpp
+400 −0 include/cutlass/gemm/kernel/sparse_gemm_row_broadcast.h
+6 −14 include/cutlass/gemm/kernel/tile_scheduler_params.h
+0 −80 include/cutlass/version.h
+2 −2 pyproject.toml
+3 −3 python/cutlass/__init__.py
+2 −6 python/cutlass/backend/c_types.py
+1 −23 python/cutlass/backend/epilogue.py
+2 −2 python/cutlass/backend/evt/frontend/frontend_base.py
+16 −0 python/cutlass/backend/evt/passes/graph_drawer.py
+18 −28 python/cutlass/backend/gemm_operation.py
+1 −1 python/setup_library.py
+1 −1 python/setup_pycute.py
+0 −1 test/unit/gemm/device/CMakeLists.txt
+19 −0 test/unit/gemm/device/gemm_f16n_f16n_f16t_tensor_op_f32_sparse_sm80.cu
+0 −685 test/unit/gemm/device/sm90_gemm_f16_f16_f16_tensor_op_f32_cluster_warpspecialized_cooperative_aux_store.cu
+20 −7 test/unit/gemm/device/testbed_sparse.h
+1 −1 tools/util/include/cutlass/util/packed_stride.hpp
9 changes: 4 additions & 5 deletions tests/test_flash_attn.py
Original file line number Diff line number Diff line change
@@ -1832,11 +1832,10 @@ def test_flash_attn_splitkv(
# @pytest.mark.parametrize("rotary_interleaved", [False])
@pytest.mark.parametrize("rotary_fraction", [0.0, 0.5, 1.0])
# @pytest.mark.parametrize("rotary_fraction", [0.0])
@pytest.mark.parametrize("paged_kv_block_size", [None, 256])
# @pytest.mark.parametrize("paged_kv_block_size", [256, 512])
# @pytest.mark.parametrize("paged_kv_block_size", [256])
@pytest.mark.parametrize("has_batch_idx", [False, True])
# @pytest.mark.parametrize("has_batch_idx", [False])
# @pytest.mark.parametrize("paged_kv_block_size", [None, 256, 512])
@pytest.mark.parametrize("paged_kv_block_size", [256])
# @pytest.mark.parametrize("has_batch_idx", [False, True])
@pytest.mark.parametrize("has_batch_idx", [False])
@pytest.mark.parametrize("d", [32, 59, 64, 80, 128, 256])
# @pytest.mark.parametrize("d", [32, 64, 96, 128, 160, 192, 224, 256])
# @pytest.mark.parametrize('d', [32, 40, 64, 80, 96, 128, 160, 192])