Skip to content

GitLab

Commit 6b9a4bd5 authored by Jun Liu's avatar Jun Liu
Browse files

Merge branch 'amd-develop-staging-0423' into amd-master

parents 56de337f c5f1cdf7
Hide whitespace changes
Inline Side-by-side
Showing with 149 additions and 145 deletions
test/batched_gemm_softmax_gemm/CMakeLists.txt
View file @ 6b9a4bd5
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(test_batched_gemm_softmax_gemm)
add_gtest_executable(test_batched_gemm_softmax_gemm_fp16 test_batched_gemm_softmax_gemm_fp16.cpp)
if(result EQUAL 0)
target_link_libraries(test_batched_gemm_softmax_gemm_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_instance)
add_dependencies(test_batched_gemm_softmax_gemm test_batched_gemm_softmax_gemm_fp16)
set(target 1)
endif()
endif()
endforeach()
\ No newline at end of file
add_gtest_executable(test_batched_gemm_softmax_gemm_fp16 test_batched_gemm_softmax_gemm_fp16_xdl.cpp)
if(result EQUAL 0)
add_custom_target(test_batched_gemm_softmax_gemm)
target_link_libraries(test_batched_gemm_softmax_gemm_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_instance)
add_dependencies(test_batched_gemm_softmax_gemm test_batched_gemm_softmax_gemm_fp16)
endif()
test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
View file @ 6b9a4bd5
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(test_batched_gemm_softmax_gemm_permute)
add_gtest_executable(test_batched_gemm_softmax_gemm_permute_fp16 test_batched_gemm_softmax_gemm_permute_fp16.cpp)
if(result EQUAL 0)
target_link_libraries(test_batched_gemm_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_fp16)
endif()
add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_fp16 test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp)
if(result EQUAL 0)
target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_fp16)
endif()
add_gtest_executable(test_batched_gemm_softmax_gemm_permute_bf16 test_batched_gemm_softmax_gemm_permute_bf16.cpp)
if(result EQUAL 0)
target_link_libraries(test_batched_gemm_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_bf16)
endif()
add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_bf16 test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp)
if(result EQUAL 0)
target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_bf16)
endif()
set(target 1)
endif()
endforeach()
\ No newline at end of file
add_custom_target(test_batched_gemm_softmax_gemm_permute)
add_gtest_executable(test_batched_gemm_softmax_gemm_permute_fp16 test_batched_gemm_softmax_gemm_permute_fp16_xdl.cpp)
if(result EQUAL 0)
target_link_libraries(test_batched_gemm_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_fp16)
endif()
add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_fp16 test_batched_gemm_bias_softmax_gemm_permute_fp16_xdl.cpp)
if(result EQUAL 0)
target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_fp16)
endif()
add_gtest_executable(test_batched_gemm_softmax_gemm_permute_bf16 test_batched_gemm_softmax_gemm_permute_bf16_xdl.cpp)
if(result EQUAL 0)
target_link_libraries(test_batched_gemm_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_bf16)
endif()
add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_bf16 test_batched_gemm_bias_softmax_gemm_permute_bf16_xdl.cpp)
if(result EQUAL 0)
target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_bf16)
endif()
test/contraction/CMakeLists.txt
View file @ 6b9a4bd5
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
if((DTYPES MATCHES "fp32" OR DTYPES MATCHES "fp64") OR NOT DEFINED DTYPES)
add_gtest_executable(test_contraction test_contraction.cpp)
target_link_libraries(test_contraction PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
add_gtest_executable(test_contraction_interface test_contraction_interface.cpp)
target_link_libraries(test_contraction_interface PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
set(target 1)
endif()
if((DTYPES MATCHES "fp32" OR DTYPES MATCHES "fp64") OR NOT DEFINED DTYPES)
add_gtest_executable(test_contraction test_contraction_xdl.cpp)
if(result EQUAL 0)
target_link_libraries(test_contraction PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
endif()
endforeach()
add_gtest_executable(test_contraction_interface test_contraction_interface_xdl.cpp)
if(result EQUAL 0)
target_link_libraries(test_contraction_interface PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
endif()
endif()
test/contraction/test_contraction_interface.cpp → test/contraction/test_contraction_interface_xdl.cpp
View file @ 6b9a4bd5
// SPDX-License-Identifier: MIT
// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
#include <stdexcept>
#include <vector>
......@@ -125,18 +125,6 @@ class ContractionDeviceOpWrapper
}
};
TEST(TestContractionInterface, IncorrectNumDims)
{
std::vector<std::vector<ck::index_t>> Dims = {{4, 4}, {4, 4, 4, 4}, {4, 4, 4, 4, 4, 4}};
std::vector<std::vector<ck::index_t>> Strides = {{1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1, 1, 1}};
ContractionDeviceOpWrapper<F32, F32, F32, F32, 1> wrapper_1d;
ContractionDeviceOpWrapper<F32, F32, F32, F32, 2> wrapper_2d;
ContractionDeviceOpWrapper<F32, F32, F32, F32, 3> wrapper_3d;
EXPECT_FALSE(wrapper_1d.IsSupportedInstance(Dims[0], Strides[0]));
EXPECT_TRUE(wrapper_2d.IsSupportedInstance(Dims[1], Strides[1]));
EXPECT_FALSE(wrapper_3d.IsSupportedInstance(Dims[2], Strides[2]));
}
TEST(TestContractionInterface, IncorrectDataTypes)
{
std::vector<ck::index_t> Dims = {4, 4, 4, 4};
......
test/contraction/test_contraction.cpp → test/contraction/test_contraction_xdl.cpp
View file @ 6b9a4bd5
// SPDX-License-Identifier: MIT
// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <iostream>
......@@ -23,8 +23,11 @@ using Col = ck::tensor_layout::gemm::ColumnMajor;
using Bilinear = ck::tensor_operation::element_wise::Bilinear;
using Scale = ck::tensor_operation::element_wise::Scale;
template <ck::index_t NDims>
struct Dimensions
{
constexpr static ck::index_t NumDimMNK = NDims;
std::vector<ck::index_t> M;
std::vector<ck::index_t> N;
std::vector<ck::index_t> K;
......@@ -42,53 +45,58 @@ class TestContraction : public ::testing::Test
using ComputeDataType = std::tuple_element_t<5, Tuple>;
using CDElementOp = std::tuple_element_t<6, Tuple>;
std::vector<Dimensions> dimension_list = {{{32, 32}, {32, 32}, {32, 32}},
{{16, 16}, {32, 32}, {16, 16}}};
std::vector<ck::index_t> init_methods = {1, 2};
std::unique_ptr<CDElementOp> p_cd_element_op;
void Run()
template <ck::index_t NumDim>
void Run(Dimensions<NumDim> dimension_params)
{
for(auto& dimension_params : dimension_list)
constexpr ck::index_t NumDimMNK = ck::remove_cvref_t<decltype(dimension_params)>::NumDimMNK;
std::vector<ck::index_t> StridesA(2 * NumDim);
std::vector<ck::index_t> StridesB(2 * NumDim);
std::vector<ck::index_t> StridesC(2 * NumDim);
std::vector<ck::index_t> StridesD(2 * NumDim);
const auto& M = dimension_params.M;
const auto& N = dimension_params.N;
const auto& K = dimension_params.K;
auto merge_dims = [](const std::vector<ck::index_t>& dims01,
const std::vector<ck::index_t>& dims23) {
std::vector<ck::index_t> dims_szt(dims01.begin(), dims01.end());
dims_szt.insert(dims_szt.end(), dims23.begin(), dims23.end());
return dims_szt;
};
assign_default_strides(ALayout{}, StridesA, merge_dims(M, K));
assign_default_strides(BLayout{}, StridesB, merge_dims(N, K));
assign_default_strides(CDLayout{}, StridesC, merge_dims(M, N));
assign_default_strides(CDLayout{}, StridesD, merge_dims(M, N));
for(const ck::index_t init_method : init_methods)
{
std::vector<ck::index_t> StridesA;
std::vector<ck::index_t> StridesB;
std::vector<ck::index_t> StridesC;
std::vector<ck::index_t> StridesD;
const auto& M = dimension_params.M;
const auto& N = dimension_params.N;
const auto& K = dimension_params.K;
assign_default_strides(ALayout{}, StridesA, {M[0], M[1], K[0], K[1]});
assign_default_strides(BLayout{}, StridesB, {N[0], N[1], K[0], K[1]});
assign_default_strides(CDLayout{}, StridesC, {M[0], M[1], N[0], N[1]});
assign_default_strides(CDLayout{}, StridesD, {M[0], M[1], N[0], N[1]});
for(const ck::index_t init_method : init_methods)
{
bool pass =
ck::profiler::profile_contraction_impl<ALayout,
BLayout,
CDLayout,
DataType,
ComputeDataType,
DTupleDataType,
CDElementOp>(true /*do_verification*/,
init_method,
false /*do_logs*/,
false /*time_kernel*/,
*p_cd_element_op,
dimension_params.M,
dimension_params.N,
dimension_params.K,
StridesA,
StridesB,
StridesC,
StridesD);
EXPECT_TRUE(pass);
}
bool pass =
ck::profiler::profile_contraction_impl<NumDimMNK,
ALayout,
BLayout,
CDLayout,
DataType,
ComputeDataType,
DTupleDataType,
CDElementOp>(true /*do_verification*/,
init_method,
false /*do_logs*/,
false /*time_kernel*/,
*p_cd_element_op,
dimension_params.M,
dimension_params.N,
dimension_params.K,
StridesA,
StridesB,
StridesC,
StridesD);
EXPECT_TRUE(pass);
}
}
};
......@@ -122,17 +130,31 @@ TYPED_TEST_SUITE(TestContractionScale, ScaleKernelTypes);
TYPED_TEST(TestContractionBilinear, bilinear)
{
this->p_cd_element_op = std::make_unique<Bilinear>(1.f, 1.f);
this->Run();
this->template Run<6>({{2, 3, 2, 3, 2, 3}, {2, 3, 2, 3, 2, 3}, {2, 2, 2, 2, 2, 4}});
this->template Run<6>({{1, 1, 1, 3, 2, 3}, {1, 1, 1, 3, 2, 3}, {1, 1, 1, 2, 2, 4}});
this->template Run<2>({{16, 8}, {16, 8}, {16, 8}});
this->template Run<2>({{8, 16}, {16, 8}, {8, 16}});
this->p_cd_element_op = std::make_unique<Bilinear>(-0.5f, 0.5f);
this->Run();
this->template Run<6>({{2, 3, 2, 3, 2, 3}, {2, 3, 2, 3, 2, 3}, {2, 2, 2, 2, 2, 4}});
this->template Run<6>({{1, 1, 1, 3, 2, 3}, {1, 1, 1, 3, 2, 3}, {1, 1, 1, 2, 2, 4}});
this->template Run<2>({{16, 8}, {16, 8}, {16, 8}});
this->template Run<2>({{8, 16}, {16, 8}, {8, 16}});
}
TYPED_TEST(TestContractionScale, scale)
{
this->p_cd_element_op = std::make_unique<Scale>(1.f);
this->Run();
this->template Run<6>({{2, 3, 2, 3, 2, 3}, {2, 3, 2, 3, 2, 3}, {2, 2, 2, 2, 2, 4}});
this->template Run<6>({{1, 1, 1, 3, 2, 3}, {1, 1, 1, 3, 2, 3}, {1, 1, 1, 2, 2, 4}});
this->template Run<2>({{16, 8}, {16, 8}, {16, 8}});
this->template Run<2>({{8, 16}, {16, 8}, {8, 16}});
this->p_cd_element_op = std::make_unique<Scale>(0.5f);
this->Run();
this->template Run<6>({{2, 3, 2, 3, 2, 3}, {2, 3, 2, 3, 2, 3}, {2, 2, 2, 2, 2, 4}});
this->template Run<6>({{1, 1, 1, 3, 2, 3}, {1, 1, 1, 3, 2, 3}, {1, 1, 1, 2, 2, 4}});
this->template Run<2>({{16, 8}, {16, 8}, {16, 8}});
this->template Run<2>({{8, 16}, {16, 8}, {8, 16}});
}
template <typename Tuple>
......@@ -165,15 +187,29 @@ TYPED_TEST_SUITE(TestContractionScaleMixedPrecision, ScaleKernelTypesMixedPrecis
TYPED_TEST(TestContractionBilinearMixedPrecision, bilinear)
{
this->p_cd_element_op = std::make_unique<Bilinear>(1.f, 1.f);
this->Run();
this->template Run<6>({{2, 3, 2, 3, 2, 3}, {2, 3, 2, 3, 2, 3}, {2, 2, 2, 2, 2, 4}});
this->template Run<6>({{1, 1, 1, 3, 2, 3}, {1, 1, 1, 3, 2, 3}, {1, 1, 1, 2, 2, 4}});
this->template Run<2>({{16, 8}, {16, 8}, {16, 8}});
this->template Run<2>({{8, 16}, {16, 8}, {8, 16}});
this->p_cd_element_op = std::make_unique<Bilinear>(-0.5f, 0.5f);
this->Run();
this->template Run<6>({{2, 3, 2, 3, 2, 3}, {2, 3, 2, 3, 2, 3}, {2, 2, 2, 2, 2, 4}});
this->template Run<6>({{1, 1, 1, 3, 2, 3}, {1, 1, 1, 3, 2, 3}, {1, 1, 1, 2, 2, 4}});
this->template Run<2>({{16, 8}, {16, 8}, {16, 8}});
this->template Run<2>({{8, 16}, {16, 8}, {8, 16}});
}
TYPED_TEST(TestContractionScaleMixedPrecision, scale)
{
this->p_cd_element_op = std::make_unique<Scale>(1.f);
this->Run();
this->template Run<6>({{2, 3, 2, 3, 2, 3}, {2, 3, 2, 3, 2, 3}, {2, 2, 2, 2, 2, 4}});
this->template Run<6>({{1, 1, 1, 3, 2, 3}, {1, 1, 1, 3, 2, 3}, {1, 1, 1, 2, 2, 4}});
this->template Run<2>({{16, 8}, {16, 8}, {16, 8}});
this->template Run<2>({{8, 16}, {16, 8}, {8, 16}});
this->p_cd_element_op = std::make_unique<Scale>(0.5f);
this->Run();
this->template Run<6>({{2, 3, 2, 3, 2, 3}, {2, 3, 2, 3, 2, 3}, {2, 2, 2, 2, 2, 4}});
this->template Run<6>({{1, 1, 1, 3, 2, 3}, {1, 1, 1, 3, 2, 3}, {1, 1, 1, 2, 2, 4}});
this->template Run<2>({{16, 8}, {16, 8}, {16, 8}});
this->template Run<2>({{8, 16}, {16, 8}, {8, 16}});
}
test/convnd_bwd_data/CMakeLists.txt
View file @ 6b9a4bd5
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_gtest_executable(test_convnd_bwd_data convnd_bwd_data.cpp)
add_gtest_executable(test_convnd_bwd_data convnd_bwd_data_xdl.cpp)
if(result EQUAL 0)
target_link_libraries(test_convnd_bwd_data PRIVATE utility device_conv1d_bwd_data_instance device_conv2d_bwd_data_instance device_conv3d_bwd_data_instance)
set(target 1)
endif()
endforeach()
\ No newline at end of file
endif()
test/convnd_fwd/CMakeLists.txt
View file @ 6b9a4bd5
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_gtest_executable(test_convnd_fwd convnd_fwd.cpp)
add_gtest_executable(test_convnd_fwd convnd_fwd_xdl.cpp)
if(result EQUAL 0)
target_link_libraries(test_convnd_fwd PRIVATE utility device_conv2d_fwd_instance)
set(target 1)
endif()
endforeach()
endif()
test/gemm_add/CMakeLists.txt
View file @ 6b9a4bd5
add_gtest_executable(test_gemm_add test_gemm_add.hpp)
target_link_libraries(test_gemm_add PRIVATE utility device_gemm_add_instance)
add_gtest_executable(test_gemm_add test_gemm_add_xdl.hpp)
if(result EQUAL 0)
target_link_libraries(test_gemm_add PRIVATE utility device_gemm_add_instance)
endif()
add_gtest_executable(test_gemm_add_relu test_gemm_add_relu.cpp)
target_link_libraries(test_gemm_add_relu PRIVATE utility device_gemm_add_instance device_gemm_add_relu_instance)
add_gtest_executable(test_gemm_add_relu test_gemm_add_relu_xdl.cpp)
if(result EQUAL 0)
target_link_libraries(test_gemm_add_relu PRIVATE utility device_gemm_add_instance device_gemm_add_relu_instance)
endif()
add_gtest_executable(test_gemm_add_silu test_gemm_add_silu.cpp)
target_link_libraries(test_gemm_add_silu PRIVATE utility device_gemm_add_instance device_gemm_add_silu_instance)
add_gtest_executable(test_gemm_add_silu test_gemm_add_silu_xdl.cpp)
if(result EQUAL 0)
target_link_libraries(test_gemm_add_silu PRIVATE utility device_gemm_add_instance device_gemm_add_silu_instance)
endif()
add_gtest_executable(test_gemm_add_fastgelu test_gemm_add_fastgelu.cpp)
target_link_libraries(test_gemm_add_fastgelu PRIVATE utility device_gemm_add_instance device_gemm_add_fastgelu_instance)
add_gtest_executable(test_gemm_add_fastgelu test_gemm_add_fastgelu_xdl.cpp)
if(result EQUAL 0)
target_link_libraries(test_gemm_add_fastgelu PRIVATE utility device_gemm_add_instance device_gemm_add_fastgelu_instance)
endif()
test/gemm_add/test_gemm_add_fastgelu.cpp → test/gemm_add/test_gemm_add_fastgelu_xdl.cpp
View file @ 6b9a4bd5
......@@ -4,7 +4,7 @@
#include "gtest/gtest.h"
#include "ck/ck.hpp"
#include "profiler/profile_gemm_add_fastgelu_impl.hpp"
#include "test_gemm_add.hpp"
#include "test_gemm_add_xdl.hpp"
template <typename Tuple>
class TestGemmAddFastgelu : public TestGemmAdd<Tuple>
......
test/gemm_add/test_gemm_add_relu.cpp → test/gemm_add/test_gemm_add_relu_xdl.cpp
View file @ 6b9a4bd5
......@@ -4,7 +4,7 @@
#include "gtest/gtest.h"
#include "ck/ck.hpp"
#include "profiler/profile_gemm_add_relu_impl.hpp"
#include "test_gemm_add.hpp"
#include "test_gemm_add_xdl.hpp"
template <typename Tuple>
class TestGemmAddRelu : public TestGemmAdd<Tuple>
......
test/gemm_add/test_gemm_add_silu.cpp → test/gemm_add/test_gemm_add_silu_xdl.cpp
View file @ 6b9a4bd5
......@@ -4,7 +4,7 @@
#include "gtest/gtest.h"
#include "ck/ck.hpp"
#include "profiler/profile_gemm_add_silu_impl.hpp"
#include "test_gemm_add.hpp"
#include "test_gemm_add_xdl.hpp"
template <typename Tuple>
class TestGemmAddSilu : public TestGemmAdd<Tuple>
......
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