Run sycl ampere benchmarks on cuda (#68)

* Add generic example runner * Init d and ref_d with different values * Move runner to benchmark folder * Add generic example runner * Add Ampere half-float example * Add generic example runner * Add Ampere half-float example * Add Ampere half-float example * Add Ampere bfloat-float example * Add generic example runner * Add Ampere half-float example * Add Ampere bfloat-float example * Run sycl ampere example on cuda
codeplaysoftware · May 31, 2024 · ec5ea79 · ec5ea79
1 parent f3144b2
commit ec5ea79
Show file tree

Hide file tree

Showing 5 changed files with 15 additions and 16 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -159,10 +159,7 @@ set(CUTLASS_ENABLE_LIBRARY ${CUTLASS_ENABLE_LIBRARY_INIT} CACHE BOOL "Enable CUT
 set(CUTLASS_ENABLE_PROFILER ${CUTLASS_ENABLE_LIBRARY} CACHE BOOL "Enable CUTLASS Profiler")
 set(CUTLASS_ENABLE_PERFORMANCE ${CUTLASS_ENABLE_PROFILER} CACHE BOOL "Enable CUTLASS Performance")
 option(CUTLASS_ENABLE_DEBUG_PRINTS "Whether or not to enable debug prints in CUTLASS kernels" OFF)
-
-if (CUTLASS_ENABLE_SYCL)
-  set(CUTLASS_ENABLE_BENCHMARKS ON CACHE BOOL "Enable CUTLASS Benchmarks")
-endif()
+set(CUTLASS_ENABLE_BENCHMARKS ON CACHE BOOL "Enable CUTLASS Benchmarks")
 
 set(CUTLASS_ENABLE_TESTS ${CUTLASS_ENABLE_TESTS_INIT} CACHE BOOL "Enable CUTLASS Tests")
 set(CUTLASS_ENABLE_GTEST_UNIT_TESTS ${CUTLASS_ENABLE_TESTS} CACHE BOOL "Enable CUTLASS GTest-based Unit Tests")

diff --git a/benchmarks/CMakeLists.txt b/benchmarks/CMakeLists.txt
@@ -43,30 +43,32 @@ function(cutlass_benchmark_add_executable NAME)
 
   add_dependencies(cutlass_benchmarks ${NAME})
 
+  if (NOT CUTLASS_ENABLE_SYCL)
+    SET(ADD_CUDA ON)
+  endif()
+
   target_link_libraries(
     ${NAME}
     PRIVATE
     CUTLASS
     cutlass_tools_util_includes
-    )
-
-  target_include_directories(
-    ${NAME}
-    PRIVATE
-    ${CUTLASS_BENCHMARKS_COMMON_SOURCE_DIR}
-    )
+    $<$<BOOL:${CUTLASS_ENABLE_CUBLAS}>:nvidia::cublas>
+    $<$<BOOL:${ADD_CUDA}>:cuda>
+  )
 
-  add_sycl_to_target(TARGET ${NAME})
+  if (CUTLASS_ENABLE_SYCL)
+    add_sycl_to_target(TARGET ${NAME})
+  endif()
 
   install(
     TARGETS ${NAME}
     RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
-    )
+  )
 endfunction()
 
 if(SYCL_INTEL_TARGET)
   add_subdirectory(pvc)
 endif()
-if (SYCL_NVIDIA_TARGET)
+if(SYCL_NVIDIA_TARGET OR NOT CUTLASS_ENABLE_SYCL)
   add_subdirectory(ampere)
 endif()
diff --git a/benchmarks/ampere/CMakeLists.txt b/benchmarks/ampere/CMakeLists.txt
@@ -29,10 +29,10 @@
 
 cutlass_benchmark_add_executable(
   bench_ampere_gemm_fp16_fp16_fp32_tensor_op_fp32
-  bench_ampere_gemm_fp16_fp16_fp32_tensor_op_fp32.cpp
+  bench_ampere_gemm_fp16_fp16_fp32_tensor_op_fp32.cu
 )
 
 cutlass_benchmark_add_executable(
   bench_ampere_gemm_bf16_bf16_fp32_tensor_op_fp32
-  bench_ampere_gemm_bf16_bf16_fp32_tensor_op_fp32.cpp
+  bench_ampere_gemm_bf16_bf16_fp32_tensor_op_fp32.cu
 )
diff --git a/...re_gemm_bf16_bf16_fp32_tensor_op_fp32.cpp → ...ere_gemm_bf16_bf16_fp32_tensor_op_fp32.cu b/...re_gemm_bf16_bf16_fp32_tensor_op_fp32.cpp → ...ere_gemm_bf16_bf16_fp32_tensor_op_fp32.cu
diff --git a/...re_gemm_fp16_fp16_fp32_tensor_op_fp32.cpp → ...ere_gemm_fp16_fp16_fp32_tensor_op_fp32.cu b/...re_gemm_fp16_fp16_fp32_tensor_op_fp32.cpp → ...ere_gemm_fp16_fp16_fp32_tensor_op_fp32.cu