Skip to content

GitLab

Commit 8f9c0243 authored by Alan Turner's avatar Alan Turner
Browse files

Merge branch 'develop' into migx-jit-lib

parents 181ea79a c8a8385f
Expand all Show whitespace changes
Inline Side-by-side
Showing with 777 additions and 116 deletions
example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
View file @ 8f9c0243
...@@ -72,9 +72,12 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config, ...@@ -72,9 +72,12 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
// init to 0 // init to 0
wei_device_buf.SetZero(); wei_device_buf.SetZero();
std::array<ck::index_t, NDimSpatial> input_spatial_lengths{}; std::array<ck::index_t, NDimSpatial + 3> input_lengths{};
std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{}; std::array<ck::index_t, NDimSpatial + 3> input_strides{};
std::array<ck::index_t, NDimSpatial> output_spatial_lengths{}; std::array<ck::index_t, NDimSpatial + 3> filter_lengths{};
std::array<ck::index_t, NDimSpatial + 3> weights_strides{};
std::array<ck::index_t, NDimSpatial + 3> output_lengths{};
std::array<ck::index_t, NDimSpatial + 3> output_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_strides{}; std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_dilations{}; std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
std::array<ck::index_t, NDimSpatial> input_left_pads{}; std::array<ck::index_t, NDimSpatial> input_left_pads{};
...@@ -82,9 +85,12 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config, ...@@ -82,9 +85,12 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
auto range_copy = [](const auto& from, auto to) { std::copy(begin(from), end(from), to); }; auto range_copy = [](const auto& from, auto to) { std::copy(begin(from), end(from), to); };
range_copy(conv_param.input_spatial_lengths_, begin(input_spatial_lengths)); range_copy(in_g_n_c_wis_desc.GetLengths(), begin(input_lengths));
range_copy(conv_param.filter_spatial_lengths_, begin(filter_spatial_lengths)); range_copy(in_g_n_c_wis_desc.GetStrides(), begin(input_strides));
range_copy(conv_param.output_spatial_lengths_, begin(output_spatial_lengths)); range_copy(wei_g_k_c_xs_desc.GetLengths(), begin(filter_lengths));
range_copy(wei_g_k_c_xs_desc.GetStrides(), begin(weights_strides));
range_copy(out_g_n_k_wos_desc.GetLengths(), begin(output_lengths));
range_copy(out_g_n_k_wos_desc.GetStrides(), begin(output_strides));
range_copy(conv_param.conv_filter_strides_, begin(conv_filter_strides)); range_copy(conv_param.conv_filter_strides_, begin(conv_filter_strides));
range_copy(conv_param.conv_filter_dilations_, begin(conv_filter_dilations)); range_copy(conv_param.conv_filter_dilations_, begin(conv_filter_dilations));
range_copy(conv_param.input_left_pads_, begin(input_left_pads)); range_copy(conv_param.input_left_pads_, begin(input_left_pads));
...@@ -96,13 +102,12 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config, ...@@ -96,13 +102,12 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
auto argument = conv.MakeArgument(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()), auto argument = conv.MakeArgument(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()), static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()), static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
conv_param.G_, input_lengths,
conv_param.N_, input_strides,
conv_param.K_, filter_lengths,
conv_param.C_, weights_strides,
input_spatial_lengths, output_lengths,
filter_spatial_lengths, output_strides,
output_spatial_lengths,
conv_filter_strides, conv_filter_strides,
conv_filter_dilations, conv_filter_dilations,
input_left_pads, input_left_pads,
......
example/21_gemm_layernorm/CMakeLists.txt
View file @ 8f9c0243
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0) set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
...@@ -9,3 +10,4 @@ foreach(gpu IN LISTS GPU_TARGETS) ...@@ -9,3 +10,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
endif()
example/22_cgemm/CMakeLists.txt
View file @ 8f9c0243
add_custom_target(example_cgemm_xdl) add_custom_target(example_cgemm_xdl)
add_example_executable(example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp) if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_example_executable(example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp) add_example_executable(example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_bf16)
endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp16)
endif()
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_cgemm_xdl_fp32 cgemm_xdl_fp32.cpp) add_example_executable(example_cgemm_xdl_fp32 cgemm_xdl_fp32.cpp)
add_example_executable(example_cgemm_xdl_int8 cgemm_xdl_int8.cpp) add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp32)
endif()
add_dependencies(example_cgemm_xdl if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
example_cgemm_xdl_bf16 add_example_executable(example_cgemm_xdl_int8 cgemm_xdl_int8.cpp)
example_cgemm_xdl_fp16 add_dependencies(example_cgemm_xdl example_cgemm_xdl_int8)
example_cgemm_xdl_fp32 endif()
example_cgemm_xdl_int8)
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_cgemm_xdl_int4 cgemm_xdl_int4.cpp) add_example_executable(example_cgemm_xdl_int4 cgemm_xdl_int4.cpp)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_int4) add_dependencies(example_cgemm_xdl example_cgemm_xdl_int4)
......
example/24_batched_gemm/CMakeLists.txt
View file @ 8f9c0243
add_custom_target(example_batched_gemm_xdl) add_custom_target(example_batched_gemm_xdl)
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_xdl_fp32 batched_gemm_xdl_fp32.cpp) add_example_executable(example_batched_gemm_xdl_fp32 batched_gemm_xdl_fp32.cpp)
add_example_executable(example_batched_gemm_xdl_fp16 batched_gemm_xdl_fp16.cpp) add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp32)
add_example_executable(example_batched_gemm_xdl_bfp16 batched_gemm_xdl_bfp16.cpp) endif()
add_example_executable(example_batched_gemm_xdl_int8 batched_gemm_xdl_int8.cpp) if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_xdl_fp16 batched_gemm_xdl_fp16.cpp)
add_dependencies(example_batched_gemm_xdl add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp16)
example_batched_gemm_xdl_fp32 endif()
example_batched_gemm_xdl_fp16 if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
example_batched_gemm_xdl_bfp16 add_example_executable(example_batched_gemm_xdl_bfp16 batched_gemm_xdl_bfp16.cpp)
example_batched_gemm_xdl_int8) add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_bfp16)
endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_xdl_int8 batched_gemm_xdl_int8.cpp)
add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int8)
endif()
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_batched_gemm_xdl_int4 batched_gemm_xdl_int4.cpp) add_example_executable(example_batched_gemm_xdl_int4 batched_gemm_xdl_int4.cpp)
add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int4) add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int4)
......
example/25_gemm_bias_e_permute/CMakeLists.txt
View file @ 8f9c0243
add_example_executable(example_gemm_bias_e_permute_g1m3n2k1_xdl_fp16 gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp) if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_gemm_bias_e_permute_g1m2n3k1_xdl_fp16 gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp) add_example_executable(example_gemm_bias_e_permute_g1m3n2k1_xdl_fp16 gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp)
add_example_executable(example_gemm_bias_e_permute_g1m2n3k1_xdl_fp16 gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp)
endif()
example/26_contraction/CMakeLists.txt
View file @ 8f9c0243
add_example_executable(example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp) if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp) add_example_executable(example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp)
add_example_executable(example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp)
add_example_executable(example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp) endif()
add_example_executable(example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp) if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
add_example_executable(example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp)
add_example_executable(example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp)
endif()
example/27_layernorm/CMakeLists.txt
View file @ 8f9c0243
add_example_executable(example_layernorm_fp16 layernorm_fp16.cpp) if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_layernorm_splitk_fp16 layernorm_splitk_fp16.cpp) add_example_executable(example_layernorm_fp16 layernorm_fp16.cpp)
add_example_executable(example_layernorm_splitk_fp16 layernorm_splitk_fp16.cpp)
endif()
example/28_grouped_gemm_bias_e_permute/CMakeLists.txt
View file @ 8f9c0243
add_example_executable(example_grouped_gemm_bias_e_permute_xdl_fp16 grouped_gemm_bias_e_permute_xdl_fp16.cpp) if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_gemm_bias_e_permute_xdl_fp16 grouped_gemm_bias_e_permute_xdl_fp16.cpp)
endif()
example/29_batched_gemm_bias_e_permute/CMakeLists.txt
View file @ 8f9c0243
add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp) if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp)
if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102") if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp) add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp)
endif()
endif() endif()
example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
View file @ 8f9c0243
...@@ -5,23 +5,29 @@ set(target 0) ...@@ -5,23 +5,29 @@ set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list1 AND target EQUAL 0) if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
add_custom_target(example_grouped_conv_fwd_multiple_d) add_custom_target(example_grouped_conv_fwd_multiple_d)
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp) add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16) add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
endif()
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32) add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16) add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8) add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
endif()
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp) add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4) add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
endif() # USE_BITINT_EXTENSION_INT4 endif() # USE_BITINT_EXTENSION_INT4
add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
...@@ -29,8 +35,12 @@ endforeach() ...@@ -29,8 +35,12 @@ endforeach()
set(target 0) set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list2 AND target EQUAL 0) if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp) add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp) add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
endif()
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
example/31_batched_gemm_gemm/CMakeLists.txt
View file @ 8f9c0243
...@@ -3,10 +3,15 @@ list(APPEND gpu_list2 gfx908 gfx90a) ...@@ -3,10 +3,15 @@ list(APPEND gpu_list2 gfx908 gfx90a)
set(target 0) set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list1 AND target EQUAL 0) if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp) add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp) add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp) add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
endif()
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp) add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
endif(USE_BITINT_EXTENSION_INT4) endif(USE_BITINT_EXTENSION_INT4)
...@@ -14,10 +19,8 @@ foreach(gpu IN LISTS GPU_TARGETS) ...@@ -14,10 +19,8 @@ foreach(gpu IN LISTS GPU_TARGETS)
endif() endif()
endforeach() endforeach()
set(target 0) if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
foreach(gpu IN LISTS GPU_TARGETS) if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp) add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
set(target 1)
endif() endif()
endforeach() endif()
\ No newline at end of file
example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
View file @ 8f9c0243
add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_fp16 batched_gemm_scale_softmax_gemm_xdl_fp16.cpp) if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp) add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_fp16 batched_gemm_scale_softmax_gemm_xdl_fp16.cpp)
add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp) add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp) add_example_executable(example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
add_example_executable(example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp) add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp) add_example_executable(example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
add_example_executable(example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp) endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp)
add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp)
endif()
add_custom_target(example_gemm_scale_softmax_gemm) add_custom_target(example_gemm_scale_softmax_gemm)
add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_fp16) if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_bf16) add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_fp16)
add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16) add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16)
add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16) add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16)
add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16) add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16) add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16) endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_bf16)
add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16)
endif()
example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
View file @ 8f9c0243
...@@ -17,7 +17,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g ...@@ -17,7 +17,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/tensor_specialization.hpp" #include "ck/tensor_operation/gpu/device/tensor_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp" #include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
......
example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
View file @ 8f9c0243
...@@ -17,7 +17,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g ...@@ -17,7 +17,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/tensor_specialization.hpp" #include "ck/tensor_operation/gpu/device/tensor_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp" #include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
......
example/34_batchnorm/CMakeLists.txt
View file @ 8f9c0243
add_example_executable(example_batchnorm_forward_training batchnorm_forward_training_nhwc.cpp) add_example_executable(example_batchnorm_forward_training batchnorm_forward_training_nhwc.cpp)
add_example_executable(example_batchnorm_forward_training_obsolete batchnorm_forward_training_nhwc_obsolete.cpp)
add_example_executable(example_batchnorm_forward_inferring batchnorm_forward_inferring_nhwc.cpp) add_example_executable(example_batchnorm_forward_inferring batchnorm_forward_inferring_nhwc.cpp)
add_example_executable(example_batchnorm_backward batchnorm_backward_nhwc.cpp) add_example_executable(example_batchnorm_backward batchnorm_backward_nhwc.cpp)
example/34_batchnorm/batchnorm_forward_training_nhwc.cpp
View file @ 8f9c0243
...@@ -414,7 +414,7 @@ bool bnorm_fwd_nhwc_test(bool do_verification, ...@@ -414,7 +414,7 @@ bool bnorm_fwd_nhwc_test(bool do_verification,
(void)invoker_ptr_ref->Run(argument_ptr_ref.get()); (void)invoker_ptr_ref->Run(argument_ptr_ref.get());
y_dev.FromDevice(y.mData.data()); y_dev.FromDevice(y.mData.data());
pass = pass && ck::utils::check_err(y, y_ref); pass = pass && ck::utils::check_err(y, y_ref, "Incorrect normalized output values");
if(updateMovingAverage) if(updateMovingAverage)
{ {
...@@ -424,8 +424,12 @@ bool bnorm_fwd_nhwc_test(bool do_verification, ...@@ -424,8 +424,12 @@ bool bnorm_fwd_nhwc_test(bool do_verification,
resultRunningMean_dev.FromDevice(resultRunningMean.mData.data()); resultRunningMean_dev.FromDevice(resultRunningMean.mData.data());
resultRunningVariance_dev.FromDevice(resultRunningVariance.mData.data()); resultRunningVariance_dev.FromDevice(resultRunningVariance.mData.data());
pass = pass && ck::utils::check_err(resultRunningMean, resultRunningMean_ref); pass = pass && ck::utils::check_err(resultRunningMean,
pass = pass && ck::utils::check_err(resultRunningVariance, resultRunningVariance_ref); resultRunningMean_ref,
"Incorrect running mean values");
pass = pass && ck::utils::check_err(resultRunningVariance,
resultRunningVariance_ref,
"Incorrect running variance values");
}; };
if(saveMeanAndInvVariance) if(saveMeanAndInvVariance)
...@@ -438,8 +442,11 @@ bool bnorm_fwd_nhwc_test(bool do_verification, ...@@ -438,8 +442,11 @@ bool bnorm_fwd_nhwc_test(bool do_verification,
resultSaveMean_dev.FromDevice(resultSaveMean.mData.data()); resultSaveMean_dev.FromDevice(resultSaveMean.mData.data());
resultSaveInvVariance_dev.FromDevice(resultSaveInvVariance.mData.data()); resultSaveInvVariance_dev.FromDevice(resultSaveInvVariance.mData.data());
pass = pass && ck::utils::check_err(resultSaveMean, resultSaveMean_ref); pass = pass && ck::utils::check_err(
pass = pass && ck::utils::check_err(resultSaveInvVariance, resultSaveInvVariance_ref); resultSaveMean, resultSaveMean_ref, "Incorrect saved mean values");
pass = pass && ck::utils::check_err(resultSaveInvVariance,
resultSaveInvVariance_ref,
"Incorrect saved invvariance values");
}; };
}; };
......
example/35_splitK_gemm/CMakeLists.txt
View file @ 8f9c0243
...@@ -3,17 +3,22 @@ set(target 0) ...@@ -3,17 +3,22 @@ set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0) if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_splitK_gemm_xdl) add_custom_target(example_splitK_gemm_xdl)
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp) add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp) add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16)
endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_example_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp) add_example_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bfp16)
endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp) add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int8)
add_dependencies(example_splitK_gemm_xdl endif()
example_splitK_gemm_xdl_fp32
example_splitK_gemm_xdl_fp16
example_splitK_gemm_xdl_bfp16
example_splitK_gemm_xdl_int8)
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp) add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4) add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
......
example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
View file @ 8f9c0243
...@@ -33,6 +33,7 @@ using ADataType = BF16; ...@@ -33,6 +33,7 @@ using ADataType = BF16;
using BDataType = BF16; using BDataType = BF16;
using AccDataType = F32; using AccDataType = F32;
using CDataType = F32; using CDataType = F32;
using ComputeType = BF16;
using ALayout = Row; using ALayout = Row;
using BLayout = Col; using BLayout = Col;
...@@ -46,11 +47,11 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa ...@@ -46,11 +47,11 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdlSplitKCShuffle using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdlSplitKCShuffle
// clang-format off // clang-format off
//######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| //######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| Compute|
//######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| Type|
//######| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //######| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| |
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 4>; < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 4, ComputeType>;
// clang-format on // clang-format on
#include "run_splitK_gemm_example.inc" #include "run_splitK_gemm_example.inc"
......
example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
View file @ 8f9c0243
...@@ -30,6 +30,7 @@ using ADataType = int8_t; ...@@ -30,6 +30,7 @@ using ADataType = int8_t;
using BDataType = int8_t; using BDataType = int8_t;
using AccDataType = int32_t; using AccDataType = int32_t;
using CDataType = int32_t; using CDataType = int32_t;
using ComputeType = int8_t;
using ALayout = Row; using ALayout = Row;
using BLayout = Col; using BLayout = Col;
...@@ -43,11 +44,11 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa ...@@ -43,11 +44,11 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdlSplitKCShuffle using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdlSplitKCShuffle
// clang-format off // clang-format off
//######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| //######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| Compute|
//######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| Type|
//######| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //######| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| |
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 16, 16, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 16, 16, true, 1, 1, S<1, 32, 1, 8>, 4>; < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 16, 16, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 16, 16, true, 1, 1, S<1, 32, 1, 8>, 4, ComputeType>;
// clang-format on // clang-format on
#include "run_splitK_gemm_example.inc" #include "run_splitK_gemm_example.inc"
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment