Fix issues for Cutlass 3.6

--------- Co-authored-by: Joe Todd <[email protected]>
codeplaysoftware · Dec 5, 2024 · cbea514 · cbea514
1 parent b0e09d7
commit cbea514
Show file tree

Hide file tree

Showing 38 changed files with 1,352 additions and 201 deletions.
diff --git a/examples/CMakeLists.txt b/examples/CMakeLists.txt
@@ -48,13 +48,17 @@ function(cutlass_example_add_executable NAME)
 
   add_dependencies(cutlass_examples ${NAME})
 
+  if (NOT CUTLASS_ENABLE_SYCL)
+    SET(ADD_CUDA ON)
+  endif()
+
   target_link_libraries(
     ${NAME}
     PRIVATE
     CUTLASS
     cutlass_tools_util_includes
     $<$<BOOL:${CUTLASS_ENABLE_CUBLAS}>:nvidia::cublas>
-    cuda
+    $<$<BOOL:${ADD_CUDA}>:cuda>
     )
 
   target_include_directories(
@@ -64,6 +68,13 @@ function(cutlass_example_add_executable NAME)
     ${CUTLASS_EXAMPLES_UTILS_DIR}
     )
 
+  if (CUTLASS_ENABLE_SYCL)
+    add_dependencies(${NAME} onemkl_project)
+    target_include_directories(${NAME} PRIVATE ${ONEMKL_INCLUDE_DIR})
+    target_link_libraries(${NAME} PUBLIC ${ONEMKL_LIB})
+    add_sycl_to_target(TARGET ${NAME})
+  endif()
+
   install(
     TARGETS ${NAME}
     RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
@@ -80,71 +91,79 @@ function(cutlass_example_add_executable NAME)
 
 endfunction()
 
-foreach(EXAMPLE
-  00_basic_gemm
-  01_cutlass_utilities
-  02_dump_reg_shmem
-  03_visualize_layout
-  04_tile_iterator
-  05_batched_gemm
-  06_splitK_gemm
-  07_volta_tensorop_gemm
-  08_turing_tensorop_gemm
-  09_turing_tensorop_conv2dfprop
-  10_planar_complex
-  11_planar_complex_array
-  12_gemm_bias_relu
-  13_two_tensor_op_fusion
-  14_ampere_tf32_tensorop_gemm
-  15_ampere_sparse_tensorop_gemm
-  16_ampere_tensorop_conv2dfprop
-  17_fprop_per_channel_bias
-  18_ampere_fp64_tensorop_affine2_gemm
-  19_tensorop_canonical
-  20_simt_canonical
-  21_quaternion_gemm
-  22_quaternion_conv
-  23_ampere_gemm_operand_reduction_fusion
-  24_gemm_grouped
-  25_ampere_fprop_mainloop_fusion
-  26_ampere_wgrad_mainloop_fusion
-  27_ampere_3xtf32_fast_accurate_tensorop_gemm
-  28_ampere_3xtf32_fast_accurate_tensorop_fprop
-  29_ampere_3xtf32_fast_accurate_tensorop_complex_gemm
-  30_wgrad_split_k
-  31_basic_syrk
-  32_basic_trmm
-  33_ampere_3xtf32_tensorop_symm
-  34_transposed_conv2d
-  35_gemm_softmax
-  36_gather_scatter_fusion
-  37_gemm_layernorm_gemm_fusion
-  38_syr2k_grouped
-  cute
-  39_gemm_permute
-  41_fused_multi_head_attention
-  42_ampere_tensorop_group_conv
-  43_ell_block_sparse_gemm
-  45_dual_gemm
-  46_depthwise_simt_conv2dfprop
-  47_ampere_gemm_universal_streamk
-  48_hopper_warp_specialized_gemm
-  49_hopper_gemm_with_collective_builder
-  50_hopper_gemm_with_epilogue_swizzle
-  51_hopper_gett
-  52_hopper_gather_scatter_fusion
-  53_hopper_gemm_permute
-  54_hopper_fp8_warp_specialized_gemm
-  55_hopper_mixed_dtype_gemm
-  56_hopper_ptr_array_batched_gemm
-  57_hopper_grouped_gemm
-  58_ada_fp8_gemm
-  59_ampere_gather_scatter_conv
-  61_hopper_gemm_with_topk_and_softmax
-  62_hopper_sparse_gemm
-  63_hopper_gemm_with_weight_prefetch
-  )
-
-  add_subdirectory(${EXAMPLE})
-
-endforeach()
+if (NOT CUTLASS_ENABLE_SYCL)
+  foreach(EXAMPLE
+    00_basic_gemm
+    01_cutlass_utilities
+    02_dump_reg_shmem
+    03_visualize_layout
+    04_tile_iterator
+    05_batched_gemm
+    06_splitK_gemm
+    07_volta_tensorop_gemm
+    08_turing_tensorop_gemm
+    09_turing_tensorop_conv2dfprop
+    10_planar_complex
+    11_planar_complex_array
+    12_gemm_bias_relu
+    13_two_tensor_op_fusion
+    14_ampere_tf32_tensorop_gemm
+    15_ampere_sparse_tensorop_gemm
+    16_ampere_tensorop_conv2dfprop
+    17_fprop_per_channel_bias
+    18_ampere_fp64_tensorop_affine2_gemm
+    19_tensorop_canonical
+    20_simt_canonical
+    21_quaternion_gemm
+    22_quaternion_conv
+    23_ampere_gemm_operand_reduction_fusion
+    24_gemm_grouped
+    25_ampere_fprop_mainloop_fusion
+    26_ampere_wgrad_mainloop_fusion
+    27_ampere_3xtf32_fast_accurate_tensorop_gemm
+    28_ampere_3xtf32_fast_accurate_tensorop_fprop
+    29_ampere_3xtf32_fast_accurate_tensorop_complex_gemm
+    30_wgrad_split_k
+    31_basic_syrk
+    32_basic_trmm
+    33_ampere_3xtf32_tensorop_symm
+    34_transposed_conv2d
+    35_gemm_softmax
+    36_gather_scatter_fusion
+    37_gemm_layernorm_gemm_fusion
+    38_syr2k_grouped
+    cute
+    39_gemm_permute
+    41_fused_multi_head_attention
+    42_ampere_tensorop_group_conv
+    43_ell_block_sparse_gemm
+    45_dual_gemm
+    46_depthwise_simt_conv2dfprop
+    47_ampere_gemm_universal_streamk
+    48_hopper_warp_specialized_gemm
+    49_hopper_gemm_with_collective_builder
+    50_hopper_gemm_with_epilogue_swizzle
+    51_hopper_gett
+    52_hopper_gather_scatter_fusion
+    53_hopper_gemm_permute
+    54_hopper_fp8_warp_specialized_gemm
+    55_hopper_mixed_dtype_gemm
+    56_hopper_ptr_array_batched_gemm
+    57_hopper_grouped_gemm
+    58_ada_fp8_gemm
+    59_ampere_gather_scatter_conv
+    61_hopper_gemm_with_topk_and_softmax
+    62_hopper_sparse_gemm
+    63_hopper_gemm_with_weight_prefetch
+    )
+    add_subdirectory(${EXAMPLE})
+  endforeach()
+else()
+  foreach(EXAMPLE
+    14_ampere_tf32_tensorop_gemm
+    cute
+    sycl
+    )
+    add_subdirectory(${EXAMPLE})
+  endforeach()
+endif()