Skip to content

GitLab

Unverified Commit a8629a98 authored by zjing14's avatar zjing14 Committed by GitHub
Browse files

Merge branch 'develop' into gemm_v2r3_kpad_fix

parents 8dc713ea 94bfa502
Hide whitespace changes
Inline Side-by-side
Showing with 547 additions and 348 deletions
test/elementwise_normalization/CMakeLists.txt
View file @ a8629a98
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_custom_target(test_elementwise_normalization)
add_custom_target(test_elementwise_normalization) add_gtest_executable(test_elementwise_layernorm_fp16 test_elementwise_layernorm_fp16.cpp)
add_gtest_executable(test_elementwise_layernorm_fp16 test_elementwise_layernorm_fp16.cpp) if(result EQUAL 0)
target_link_libraries(test_elementwise_layernorm_fp16 PRIVATE utility device_elementwise_normalization_instance) target_link_libraries(test_elementwise_layernorm_fp16 PRIVATE utility device_elementwise_normalization_instance)
add_dependencies(test_elementwise_normalization test_elementwise_layernorm_fp16) add_dependencies(test_elementwise_normalization test_elementwise_layernorm_fp16)
endif() endif()
\ No newline at end of file
test/gemm/CMakeLists.txt
View file @ a8629a98
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_test_executable(test_gemm_fp32 gemm_fp32.cpp) add_test_executable(test_gemm_fp32 gemm_fp32.cpp)
target_link_libraries(test_gemm_fp32 PRIVATE utility) if(result EQUAL 0)
target_link_libraries(test_gemm_fp32 PRIVATE device_gemm_instance) target_link_libraries(test_gemm_fp32 PRIVATE utility device_gemm_instance)
endif() endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_test_executable(test_gemm_fp16 gemm_fp16.cpp) add_test_executable(test_gemm_fp16 gemm_fp16.cpp)
target_link_libraries(test_gemm_fp16 PRIVATE utility) if(result EQUAL 0)
target_link_libraries(test_gemm_fp16 PRIVATE device_gemm_instance) target_link_libraries(test_gemm_fp16 PRIVATE utility device_gemm_instance)
add_library(gemm_standalone_xdl_fp16_instances STATIC add_library(gemm_standalone_xdl_fp16_instances STATIC
instance/gemm_f16_nn_instance.cpp instance/gemm_f16_nn_instance.cpp
instance/gemm_f16_nt_instance.cpp instance/gemm_f16_nt_instance.cpp
instance/gemm_f16_tn_instance.cpp instance/gemm_f16_tn_instance.cpp
instance/gemm_wavelet_f16_tn_instance.cpp instance/gemm_wavelet_f16_tn_instance.cpp
instance/gemm_f16_tt_instance.cpp instance/gemm_f16_tt_instance.cpp
) )
endif()
add_test_executable(test_gemm_standalone_xdl_fp16 gemm_standalone_xdl_fp16.cpp) add_test_executable(test_gemm_standalone_xdl_fp16 gemm_standalone_xdl_fp16.cpp)
target_link_libraries(test_gemm_standalone_xdl_fp16 PRIVATE gemm_standalone_xdl_fp16_instances utility) if(result EQUAL 0)
target_include_directories(test_gemm_standalone_xdl_fp16 PRIVATE instance/) target_link_libraries(test_gemm_standalone_xdl_fp16 PRIVATE gemm_standalone_xdl_fp16_instances utility)
target_include_directories(test_gemm_standalone_xdl_fp16 PRIVATE instance/)
endif() endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_test_executable(test_gemm_bf16 gemm_bf16.cpp) add_test_executable(test_gemm_bf16 gemm_bf16.cpp)
target_link_libraries(test_gemm_bf16 PRIVATE utility) if(result EQUAL 0)
target_link_libraries(test_gemm_bf16 PRIVATE device_gemm_instance) target_link_libraries(test_gemm_bf16 PRIVATE utility device_gemm_instance)
endif() endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
add_test_executable(test_gemm_int8 gemm_int8.cpp) add_test_executable(test_gemm_int8 gemm_int8.cpp)
target_link_libraries(test_gemm_int8 PRIVATE utility) if(result EQUAL 0)
target_link_libraries(test_gemm_int8 PRIVATE device_gemm_instance) target_link_libraries(test_gemm_int8 PRIVATE utility device_gemm_instance)
endif() endif()
\ No newline at end of file
test/gemm_layernorm/CMakeLists.txt
View file @ a8629a98
...@@ -2,12 +2,12 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) ...@@ -2,12 +2,12 @@ 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)
if(gpu IN_LIST gpu_list AND target EQUAL 0) if(gpu IN_LIST gpu_list AND target EQUAL 0)
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_custom_target(test_gemm_layernorm) add_custom_target(test_gemm_layernorm)
add_gtest_executable(test_gemm_add_relu_add_layernorm_fp16 test_gemm_add_relu_add_layernorm_fp16.cpp) add_gtest_executable(test_gemm_add_relu_add_layernorm_fp16 test_gemm_add_relu_add_layernorm_fp16.cpp)
target_link_libraries(test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance) if(result EQUAL 0)
add_dependencies(test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16) target_link_libraries(test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance)
set(target 1) add_dependencies(test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16)
endif() set(target 1)
endif()
endif() endif()
endforeach() endforeach()
test/gemm_reduce/CMakeLists.txt
View file @ a8629a98
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_test_executable(test_gemm_reduce_fp16 gemm_reduce_fp16.cpp)
add_test_executable(test_gemm_reduce_fp16 gemm_reduce_fp16.cpp) if(result EQUAL 0)
target_link_libraries(test_gemm_reduce_fp16 PRIVATE utility) target_link_libraries(test_gemm_reduce_fp16 PRIVATE utility device_gemm_reduce_instance)
target_link_libraries(test_gemm_reduce_fp16 PRIVATE device_gemm_reduce_instance)
endif() endif()
\ No newline at end of file
test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
View file @ a8629a98
...@@ -14,6 +14,8 @@ ...@@ -14,6 +14,8 @@
#include "profiler/profile_grouped_conv_bwd_weight_impl.hpp" #include "profiler/profile_grouped_conv_bwd_weight_impl.hpp"
using namespace ck::tensor_layout::convolution;
template <typename Tuple> template <typename Tuple>
class TestGroupedConvndBwdWeight : public ::testing::Test class TestGroupedConvndBwdWeight : public ::testing::Test
{ {
...@@ -27,28 +29,59 @@ class TestGroupedConvndBwdWeight : public ::testing::Test ...@@ -27,28 +29,59 @@ class TestGroupedConvndBwdWeight : public ::testing::Test
using NDimSpatial = std::tuple_element_t<6, Tuple>; using NDimSpatial = std::tuple_element_t<6, Tuple>;
std::vector<ck::utils::conv::ConvParam> conv_params; std::vector<ck::utils::conv::ConvParam> conv_params;
ck::index_t split_k{2}; std::vector<ck::index_t> split_ks{1, 2};
bool skip_case(const ck::utils::conv::ConvParam& params, const ck::index_t split_k)
{
// Odd K or C values are supported only by DL kernel (only applies to fp16)
// DL kernel currently supports only `split_k=1`
if constexpr(std::is_same_v<InDataType, ck::half_t>)
{
if(split_k != 1 && (params.K_ % 2 != 0 || params.C_ % 2 != 0))
{
return true;
}
}
// 1d NWGC is only supported by DL kernel
// DL kernel is only supported for split_k=1
if constexpr(std::is_same_v<InLayout, NWGC> && std::is_same_v<OutLayout, NWGK>)
{
if(split_k != 1)
{
return true;
}
}
return false;
}
void Run() void Run()
{ {
EXPECT_FALSE(conv_params.empty()); EXPECT_FALSE(conv_params.empty());
bool pass = true; bool pass = true;
for(auto& param : conv_params) for(auto split_k : split_ks)
{ {
pass = pass && ck::profiler::profile_grouped_conv_bwd_weight_impl<NDimSpatial{}, for(auto& param : conv_params)
InLayout, {
WeiLayout, if(!skip_case(param, split_k))
OutLayout, {
InDataType, pass = pass && ck::profiler::profile_grouped_conv_bwd_weight_impl<NDimSpatial{},
WeiDataType, InLayout,
OutDataType>( WeiLayout,
true, // do_verification OutLayout,
1, // init_method: integer value InDataType,
false, // do_log WeiDataType,
false, // time_kernel OutDataType>(
param, true, // do_verification
split_k); 1, // init_method: integer value
false, // do_log
false, // time_kernel
param,
split_k);
}
}
} }
EXPECT_TRUE(pass); EXPECT_TRUE(pass);
} }
...@@ -69,12 +102,13 @@ class TestGroupedConvndBwdWeight3d : public TestGroupedConvndBwdWeight<Tuple> ...@@ -69,12 +102,13 @@ class TestGroupedConvndBwdWeight3d : public TestGroupedConvndBwdWeight<Tuple>
{ {
}; };
using namespace ck::tensor_layout::convolution;
using KernelTypes1d = ::testing::Types< using KernelTypes1d = ::testing::Types<
std::tuple<float, float, float, GNWC, GKXC, GNWK, ck::Number<1>>, std::tuple<float, float, float, GNWC, GKXC, GNWK, ck::Number<1>>,
std::tuple<ck::half_t, ck::half_t, ck::half_t, GNWC, GKXC, GNWK, ck::Number<1>>, std::tuple<ck::half_t, ck::half_t, ck::half_t, GNWC, GKXC, GNWK, ck::Number<1>>,
std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNWC, GKXC, GNWK, ck::Number<1>>>; std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNWC, GKXC, GNWK, ck::Number<1>>,
std::tuple<float, float, float, NWGC, GKXC, NWGK, ck::Number<1>>,
std::tuple<ck::half_t, ck::half_t, ck::half_t, NWGC, GKXC, NWGK, ck::Number<1>>,
std::tuple<ck::bhalf_t, float, ck::bhalf_t, NWGC, GKXC, NWGK, ck::Number<1>>>;
using KernelTypes2d = ::testing::Types< using KernelTypes2d = ::testing::Types<
std::tuple<float, float, float, GNHWC, GKYXC, GNHWK, ck::Number<2>>, std::tuple<float, float, float, GNHWC, GKYXC, GNHWK, ck::Number<2>>,
std::tuple<ck::half_t, ck::half_t, ck::half_t, GNHWC, GKYXC, GNHWK, ck::Number<2>>, std::tuple<ck::half_t, ck::half_t, ck::half_t, GNHWC, GKYXC, GNHWK, ck::Number<2>>,
......
test/grouped_convnd_fwd/CMakeLists.txt
View file @ a8629a98
add_gtest_executable(test_grouped_convnd_fwd grouped_convnd_fwd.cpp) add_gtest_executable(test_grouped_convnd_fwd test_grouped_convnd_fwd.cpp)
target_link_libraries(test_grouped_convnd_fwd PRIVATE utility device_grouped_conv1d_fwd_instance device_grouped_conv2d_fwd_instance device_grouped_conv3d_fwd_instance) target_link_libraries(test_grouped_convnd_fwd PRIVATE utility device_grouped_conv1d_fwd_instance device_grouped_conv2d_fwd_instance device_grouped_conv3d_fwd_instance)
test/grouped_convnd_fwd/grouped_convnd_fwd.cpp deleted 100644 → 0
View file @ 8dc713ea
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <iostream>
#include <initializer_list>
#include <vector>
#include <gtest/gtest.h>
#include "profiler/profile_grouped_conv_fwd_impl.hpp"
class TestGroupedConvNdFwd : public ::testing::Test
{
protected:
std::vector<ck::utils::conv::ConvParam> conv_params;
};
// 1d GNWC/GKXC/GNWK
TEST_F(TestGroupedConvNdFwd, GroupedConv1dFwdGNWC)
{
conv_params.clear();
conv_params.push_back({1, 2, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
conv_params.push_back({1, 2, 128, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
conv_params.push_back({1, 2, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
conv_params.push_back({1, 1, 1, 1, 32, {3}, {32}, {1}, {1}, {1}, {1}});
conv_params.push_back({1, 1, 1, 64, 3, {3}, {32}, {1}, {1}, {1}, {1}});
for(auto& param : conv_params)
{
bool pass;
// fp32
pass = ck::profiler::profile_grouped_conv_fwd_impl<1,
ck::tensor_layout::convolution::GNWC,
ck::tensor_layout::convolution::GKXC,
ck::tensor_layout::convolution::GNWK,
float,
float,
float>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// fp16
pass = ck::profiler::profile_grouped_conv_fwd_impl<1,
ck::tensor_layout::convolution::GNWC,
ck::tensor_layout::convolution::GKXC,
ck::tensor_layout::convolution::GNWK,
ck::half_t,
ck::half_t,
ck::half_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// bf16
pass = ck::profiler::profile_grouped_conv_fwd_impl<1,
ck::tensor_layout::convolution::GNWC,
ck::tensor_layout::convolution::GKXC,
ck::tensor_layout::convolution::GNWK,
ck::bhalf_t,
ck::bhalf_t,
ck::bhalf_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// int8
pass = ck::profiler::profile_grouped_conv_fwd_impl<1,
ck::tensor_layout::convolution::GNWC,
ck::tensor_layout::convolution::GKXC,
ck::tensor_layout::convolution::GNWK,
int8_t,
int8_t,
int8_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
}
}
// 2d GNHWC/GKYXC/GNHWK
TEST_F(TestGroupedConvNdFwd, GroupedConv2dFwdGNHWC)
{
conv_params.clear();
conv_params.push_back({2, 2, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
conv_params.push_back({2, 2, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
conv_params.push_back({2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
conv_params.push_back({2, 1, 1, 1, 32, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
conv_params.push_back({2, 1, 1, 64, 3, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
conv_params.push_back({2, 1, 1, 1, 1, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
for(auto& param : conv_params)
{
bool pass;
// fp32
pass = ck::profiler::profile_grouped_conv_fwd_impl<2,
ck::tensor_layout::convolution::GNHWC,
ck::tensor_layout::convolution::GKYXC,
ck::tensor_layout::convolution::GNHWK,
float,
float,
float>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// fp16
pass = ck::profiler::profile_grouped_conv_fwd_impl<2,
ck::tensor_layout::convolution::GNHWC,
ck::tensor_layout::convolution::GKYXC,
ck::tensor_layout::convolution::GNHWK,
ck::half_t,
ck::half_t,
ck::half_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// bf16
pass = ck::profiler::profile_grouped_conv_fwd_impl<2,
ck::tensor_layout::convolution::GNHWC,
ck::tensor_layout::convolution::GKYXC,
ck::tensor_layout::convolution::GNHWK,
ck::bhalf_t,
ck::bhalf_t,
ck::bhalf_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// int8
pass = ck::profiler::profile_grouped_conv_fwd_impl<2,
ck::tensor_layout::convolution::GNHWC,
ck::tensor_layout::convolution::GKYXC,
ck::tensor_layout::convolution::GNHWK,
int8_t,
int8_t,
int8_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
}
}
// 3d GNDHWC/GKZYXC/GNDHWK
TEST_F(TestGroupedConvNdFwd, GroupedConv3dFwdGNDHWC)
{
conv_params.clear();
conv_params.push_back(
{3, 2, 128, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
conv_params.push_back(
{3, 2, 128, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
conv_params.push_back(
{3, 2, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
conv_params.push_back(
{3, 1, 1, 1, 32, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
this->conv_params.push_back(
{3, 1, 1, 64, 3, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
conv_params.push_back(
{3, 1, 1, 1, 1, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
for(auto& param : conv_params)
{
bool pass;
// fp32
pass = ck::profiler::profile_grouped_conv_fwd_impl<3,
ck::tensor_layout::convolution::GNDHWC,
ck::tensor_layout::convolution::GKZYXC,
ck::tensor_layout::convolution::GNDHWK,
float,
float,
float>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// fp16
pass = ck::profiler::profile_grouped_conv_fwd_impl<3,
ck::tensor_layout::convolution::GNDHWC,
ck::tensor_layout::convolution::GKZYXC,
ck::tensor_layout::convolution::GNDHWK,
ck::half_t,
ck::half_t,
ck::half_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// bf16
pass = ck::profiler::profile_grouped_conv_fwd_impl<3,
ck::tensor_layout::convolution::GNDHWC,
ck::tensor_layout::convolution::GKZYXC,
ck::tensor_layout::convolution::GNDHWK,
ck::bhalf_t,
ck::bhalf_t,
ck::bhalf_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// int8
pass = ck::profiler::profile_grouped_conv_fwd_impl<3,
ck::tensor_layout::convolution::GNDHWC,
ck::tensor_layout::convolution::GKZYXC,
ck::tensor_layout::convolution::GNDHWK,
int8_t,
int8_t,
int8_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
}
}
// 2d NHWGC/KYXGC/NHWGK
TEST_F(TestGroupedConvNdFwd, GroupedConv2dFwdNHWGC)
{
conv_params.clear();
conv_params.push_back({2, 2, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
conv_params.push_back({2, 2, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
conv_params.push_back({2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
conv_params.push_back({2, 1, 1, 1, 32, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
conv_params.push_back({2, 1, 1, 64, 3, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
conv_params.push_back({2, 1, 1, 1, 1, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
for(auto& param : conv_params)
{
bool pass;
// fp16
pass = ck::profiler::profile_grouped_conv_fwd_impl<2,
ck::tensor_layout::convolution::NHWGC,
ck::tensor_layout::convolution::GKYXC,
ck::tensor_layout::convolution::NHWGK,
ck::half_t,
ck::half_t,
ck::half_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
}
}
test/grouped_convnd_fwd/test_grouped_convnd_fwd.cpp 0 → 100644
View file @ a8629a98
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <iostream>
#include <initializer_list>
#include <vector>
#include <gtest/gtest.h>
#include "profiler/profile_grouped_conv_fwd_impl.hpp"
template <typename Tuple>
class TestGroupedConvndFwd : public ::testing::Test
{
protected:
using DataType = std::tuple_element_t<0, Tuple>;
using InLayout = std::tuple_element_t<1, Tuple>;
using WeiLayout = std::tuple_element_t<2, Tuple>;
using OutLayout = std::tuple_element_t<3, Tuple>;
std::vector<ck::utils::conv::ConvParam> conv_params;
template <ck::index_t NDimSpatial>
void Run()
{
EXPECT_FALSE(conv_params.empty());
bool pass = true;
for(auto& param : conv_params)
{
pass = pass && ck::profiler::profile_grouped_conv_fwd_impl<NDimSpatial,
InLayout,
WeiLayout,
OutLayout,
DataType,
DataType,
DataType>(
true, // do_verification
1, // init_method: integer value
false, // do_log
false, // time_kernel
param);
}
EXPECT_TRUE(pass);
}
};
using namespace ck::tensor_layout::convolution;
using KernelTypes1d = ::testing::Types<std::tuple<float, GNWC, GKXC, GNWK>,
std::tuple<ck::half_t, GNWC, GKXC, GNWK>,
std::tuple<ck::bhalf_t, GNWC, GKXC, GNWK>,
std::tuple<int8_t, GNWC, GKXC, GNWK>>;
using KernelTypes2d = ::testing::Types<std::tuple<float, GNHWC, GKYXC, GNHWK>,
std::tuple<ck::half_t, GNHWC, GKYXC, GNHWK>,
std::tuple<ck::bhalf_t, GNHWC, GKYXC, GNHWK>,
std::tuple<int8_t, GNHWC, GKYXC, GNHWK>,
std::tuple<float, NHWGC, GKYXC, NHWGK>,
std::tuple<ck::half_t, NHWGC, GKYXC, NHWGK>,
std::tuple<ck::bhalf_t, NHWGC, GKYXC, NHWGK>,
std::tuple<int8_t, NHWGC, GKYXC, NHWGK>>;
using KernelTypes3d = ::testing::Types<std::tuple<float, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<ck::half_t, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<ck::bhalf_t, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<int8_t, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<float, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<ck::half_t, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<ck::bhalf_t, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<int8_t, NDHWGC, GKZYXC, NDHWGK>>;
template <typename Tuple>
class TestGroupedConvndFwd1d : public TestGroupedConvndFwd<Tuple>
{
};
template <typename Tuple>
class TestGroupedConvndFwd2d : public TestGroupedConvndFwd<Tuple>
{
};
template <typename Tuple>
class TestGroupedConvndFwd3d : public TestGroupedConvndFwd<Tuple>
{
};
TYPED_TEST_SUITE(TestGroupedConvndFwd1d, KernelTypes1d);
TYPED_TEST_SUITE(TestGroupedConvndFwd2d, KernelTypes2d);
TYPED_TEST_SUITE(TestGroupedConvndFwd3d, KernelTypes3d);
TYPED_TEST(TestGroupedConvndFwd1d, Test1D)
{
this->conv_params.clear();
this->conv_params.push_back({1, 2, 32, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
this->conv_params.push_back({1, 2, 32, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
this->conv_params.push_back({1, 2, 32, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
this->conv_params.push_back({1, 1, 1, 1, 32, {3}, {32}, {1}, {1}, {1}, {1}});
this->conv_params.push_back({1, 1, 1, 64, 3, {3}, {32}, {1}, {1}, {1}, {1}});
this->template Run<1>();
}
TYPED_TEST(TestGroupedConvndFwd2d, Test2D)
{
this->conv_params.clear();
this->conv_params.push_back(
{2, 2, 32, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
this->conv_params.push_back(
{2, 2, 32, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
this->conv_params.push_back(
{2, 2, 32, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
this->conv_params.push_back({2, 1, 1, 1, 32, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
this->conv_params.push_back({2, 1, 1, 64, 3, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
this->conv_params.push_back({2, 1, 1, 1, 1, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
this->template Run<2>();
}
TYPED_TEST(TestGroupedConvndFwd3d, Test3D)
{
this->conv_params.clear();
this->conv_params.push_back(
{3, 2, 32, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
this->conv_params.push_back(
{3, 2, 32, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
this->conv_params.push_back(
{3, 2, 32, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
this->conv_params.push_back(
{3, 1, 1, 1, 32, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
this->conv_params.push_back(
{3, 1, 1, 64, 3, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
this->conv_params.push_back(
{3, 1, 1, 1, 1, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
this->template Run<3>();
}
test/grouped_gemm/CMakeLists.txt
View file @ a8629a98
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)
...@@ -13,4 +12,3 @@ foreach(gpu IN LISTS GPU_TARGETS) ...@@ -13,4 +12,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
endif()
test/image_to_column/CMakeLists.txt 0 → 100644
View file @ a8629a98
add_gtest_executable(test_image_to_column test_image_to_column.cpp)
target_link_libraries(test_image_to_column PRIVATE utility device_image_to_column_instance)
add_gtest_executable(test_image_to_column_interface test_image_to_column_interface.cpp)
target_link_libraries(test_image_to_column_interface PRIVATE utility)
test/image_to_column/test_image_to_column.cpp 0 → 100644
View file @ a8629a98
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <iostream>
#include <initializer_list>
#include <tuple>
#include <vector>
#include <gtest/gtest.h>
#include "profiler/profile_image_to_column_impl.hpp"
template <typename Tuple>
class TestImageToColumn : public ::testing::Test
{
protected:
using InDataType = std::tuple_element_t<0, Tuple>;
using OutDataType = std::tuple_element_t<1, Tuple>;
using InLayout = std::tuple_element_t<2, Tuple>;
std::vector<ck::utils::conv::ConvParam> conv_params;
template <ck::index_t NDimSpatial>
void Run()
{
EXPECT_FALSE(conv_params.empty());
bool pass = true;
for(auto& param : conv_params)
{
pass = pass && ck::profiler::profile_image_to_column_impl<NDimSpatial,
InLayout,
InDataType,
OutDataType>(
true, // do_verification
1, // init_method: integer value
false, // do_log
false, // time_kernel
param);
}
EXPECT_TRUE(pass);
}
};
using namespace ck::tensor_layout::convolution;
using KernelTypes1d = ::testing::Types<std::tuple<float, float, GNWC>,
std::tuple<ck::bhalf_t, ck::bhalf_t, GNWC>,
std::tuple<ck::half_t, ck::half_t, GNWC>,
std::tuple<int8_t, int8_t, GNWC>>;
using KernelTypes2d = ::testing::Types<std::tuple<float, float, GNHWC>,
std::tuple<ck::bhalf_t, ck::bhalf_t, GNHWC>,
std::tuple<ck::half_t, ck::half_t, GNHWC>,
std::tuple<int8_t, int8_t, GNHWC>>;
using KernelTypes3d = ::testing::Types<std::tuple<float, float, GNDHWC>,
std::tuple<ck::bhalf_t, ck::bhalf_t, GNDHWC>,
std::tuple<ck::half_t, ck::half_t, GNDHWC>,
std::tuple<int8_t, int8_t, GNDHWC>>;
template <typename Tuple>
class TestImageToColumn1d : public TestImageToColumn<Tuple>
{
};
template <typename Tuple>
class TestImageToColumn2d : public TestImageToColumn<Tuple>
{
};
template <typename Tuple>
class TestImageToColumn3d : public TestImageToColumn<Tuple>
{
};
TYPED_TEST_SUITE(TestImageToColumn1d, KernelTypes1d);
TYPED_TEST_SUITE(TestImageToColumn2d, KernelTypes2d);
TYPED_TEST_SUITE(TestImageToColumn3d, KernelTypes3d);
TYPED_TEST(TestImageToColumn1d, Test1D)
{
this->conv_params.clear();
this->conv_params.push_back({1, 1, 4, 1, 192, {3}, {28}, {1}, {1}, {1}, {1}});
this->conv_params.push_back({1, 1, 64, 1, 64, {3}, {14}, {1}, {1}, {1}, {1}});
this->conv_params.push_back({1, 1, 64, 1, 64, {1}, {7}, {2}, {1}, {0}, {0}});
this->conv_params.push_back({1, 1, 64, 1, 64, {1}, {3}, {1}, {1}, {0}, {0}});
// ScalarPerVector should be 1
this->conv_params.push_back({1, 1, 4, 1, 1, {3}, {28}, {1}, {1}, {1}, {1}});
// stride != 1
this->conv_params.push_back({1, 1, 1, 1, 4, {3}, {28}, {2}, {1}, {1}, {1}});
// dilation != 1
this->conv_params.push_back({1, 1, 1, 1, 4, {3}, {28}, {1}, {2}, {1}, {1}});
this->template Run<1>();
}
TYPED_TEST(TestImageToColumn2d, Test2D)
{
this->conv_params.clear();
this->conv_params.push_back(
{2, 1, 4, 1, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
this->conv_params.push_back(
{2, 1, 64, 1, 64, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
this->template Run<2>();
}
TYPED_TEST(TestImageToColumn3d, Test3D)
{
this->conv_params.clear();
this->conv_params.push_back(
{3, 1, 16, 1, 64, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
this->conv_params.push_back(
{3, 1, 2, 1, 64, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
this->conv_params.push_back(
{3, 1, 32, 1, 64, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
this->template Run<3>();
}
test/image_to_column/test_image_to_column_interface.cpp 0 → 100644
View file @ a8629a98
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <iostream>
#include <initializer_list>
#include <tuple>
#include <vector>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp"
#include "ck/library/utility/convolution_parameter.hpp"
#include "ck/library/utility/algorithm.hpp"
#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
#include <gtest/gtest.h>
using DataType = float;
using InLayout = ck::tensor_layout::convolution::GNWC;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
template <ck::index_t ScalarPerVector, bool IsCPacked>
class TestImageToColumnInterface : public ::testing::Test
{
protected:
static constexpr ck::index_t NDimSpatial = 1;
// clang-format off
using DeviceImgToColInstance = ck::tensor_operation::device::DeviceImageToColumnImpl
//#####################| Num| InLayout| InDataType| OutDataType| Block| MPer| KPer| Thread| Scalar|
//#####################| Dim| | | | Size| Block| Block| Cluster| Per|
//#####################| Spatial| | | | | | | Lengths| Vector|
//#####################| | | | | | | | | |
< NDimSpatial, InLayout, DataType, DataType, 256, 128, 128, S<16, 16>,ScalarPerVector>;
// clang-format on
ck::utils::conv::ConvParam conv_param;
bool Run()
{
const auto N = conv_param.N_;
const auto C = conv_param.C_;
const auto FakeC =
conv_param.C_ / 2; // Fake C to simulate the behavior that C is not packed
const ck::index_t NDoHoWo =
N *
ck::accumulate_n<ck::index_t>(
conv_param.output_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
const ck::index_t CZYX =
C *
ck::accumulate_n<ck::index_t>(
conv_param.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
const auto in_desc =
ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
conv_param);
const auto out_desc = HostTensorDescriptor({NDoHoWo, CZYX});
std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
std::array<ck::index_t, NDimSpatial + 3> input_g_n_c_wis_strides{};
std::array<ck::index_t, 2> output_m_k_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> input_left_pads{};
std::array<ck::index_t, NDimSpatial> input_right_pads{};
auto copy = [](const auto& x, auto& y) { std::copy(x.begin(), x.end(), y.begin()); };
copy(conv_param.input_spatial_lengths_, input_spatial_lengths);
copy(conv_param.filter_spatial_lengths_, filter_spatial_lengths);
copy(conv_param.output_spatial_lengths_, output_spatial_lengths);
copy(in_desc.GetStrides(), input_g_n_c_wis_strides);
copy(out_desc.GetStrides(), output_m_k_strides);
copy(conv_param.conv_filter_strides_, conv_filter_strides);
copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
copy(conv_param.input_left_pads_, input_left_pads);
copy(conv_param.input_right_pads_, input_right_pads);
auto img2col = DeviceImgToColInstance{};
auto argument = img2col.MakeArgument(nullptr,
nullptr,
N,
IsCPacked ? C : FakeC,
input_spatial_lengths,
filter_spatial_lengths,
output_spatial_lengths,
input_g_n_c_wis_strides,
output_m_k_strides,
conv_filter_strides,
conv_filter_dilations,
input_left_pads,
input_right_pads);
return img2col.IsSupportedArgument(argument);
}
};
class TestImageToColumnInterface1ScalarPerVector : public TestImageToColumnInterface<1, true>
{
};
class TestImageToColumnInterface4ScalarPerVector : public TestImageToColumnInterface<4, true>
{
};
class TestImageToColumnInterface4ScalarPerVectorFakeC : public TestImageToColumnInterface<4, false>
{
};
TEST_F(TestImageToColumnInterface1ScalarPerVector, X1ScalarPerVector)
{
// vector load C * X % ScalarPerVector
this->conv_param = {1, 1, 1, 1, 1, {3}, {3}, {1}, {1}, {0}, {0}};
bool is_supported = this->Run();
EXPECT_TRUE(is_supported);
// vector load C * left_pad_x % ScalarPerVector
this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {3}, {0}};
is_supported = this->Run();
EXPECT_TRUE(is_supported);
// vector load C * right_pad_x % ScalarPerVector
this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {0}, {3}};
is_supported = this->Run();
EXPECT_TRUE(is_supported);
// vector load C % ScalarPerVector, right_pad and stride
this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {0}, {3}};
is_supported = this->Run();
EXPECT_TRUE(is_supported);
// vector load C % ScalarPerVector, left_pad and stride
this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {3}, {0}};
is_supported = this->Run();
EXPECT_TRUE(is_supported);
// vector load C % ScalarPerVector, dilation
this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {2}, {0}, {0}};
is_supported = this->Run();
EXPECT_TRUE(is_supported);
// C = 4
this->conv_param = {1, 1, 1, 1, 4, {3}, {3}, {1}, {1}, {3}, {3}};
is_supported = this->Run();
EXPECT_TRUE(is_supported);
}
TEST_F(TestImageToColumnInterface4ScalarPerVector, X4ScalarPerVector)
{
// vector load C * X % ScalarPerVector
this->conv_param = {1, 1, 1, 1, 1, {3}, {3}, {1}, {1}, {0}, {0}};
bool is_supported = this->Run();
EXPECT_FALSE(is_supported);
// vector load C * left_pad_x % ScalarPerVector
this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {3}, {0}};
is_supported = this->Run();
EXPECT_FALSE(is_supported);
// vector load C * right_pad_x % ScalarPerVector
this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {0}, {3}};
is_supported = this->Run();
EXPECT_FALSE(is_supported);
// vector load C % ScalarPerVector, right_pad and stride
this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {0}, {3}};
is_supported = this->Run();
EXPECT_FALSE(is_supported);
// vector load C % ScalarPerVector, left_pad and stride
this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {3}, {0}};
is_supported = this->Run();
EXPECT_FALSE(is_supported);
// vector load C % ScalarPerVector, dilation
this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {2}, {0}, {0}};
is_supported = this->Run();
EXPECT_FALSE(is_supported);
// C = 4
this->conv_param = {1, 1, 1, 1, 4, {3}, {3}, {1}, {1}, {3}, {3}};
is_supported = this->Run();
EXPECT_TRUE(is_supported);
}
TEST_F(TestImageToColumnInterface4ScalarPerVectorFakeC, X4ScalarPerVectorFakeC)
{
// C = 3
this->conv_param = {1, 1, 1, 1, 3, {4}, {3}, {1}, {1}, {0}, {0}};
bool is_supported = this->Run();
EXPECT_FALSE(is_supported);
// C = 4
this->conv_param = {1, 1, 1, 1, 8, {4}, {3}, {1}, {1}, {0}, {0}};
is_supported = this->Run();
EXPECT_TRUE(is_supported);
}
test/normalization/CMakeLists.txt
View file @ a8629a98
if(DTYPES MATCHES "fp16" OR DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES) add_custom_target(test_normalization)
add_custom_target(test_normalization) add_gtest_executable(test_layernorm2d_fp32 test_layernorm2d_fp32.cpp)
endif() if(result EQUAL 0)
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_gtest_executable(test_layernorm2d_fp32 test_layernorm2d_fp32.cpp)
add_gtest_executable(test_groupnorm_fp32 test_groupnorm_fp32.cpp)
target_link_libraries(test_layernorm2d_fp32 PRIVATE utility device_normalization_instance) target_link_libraries(test_layernorm2d_fp32 PRIVATE utility device_normalization_instance)
target_link_libraries(test_groupnorm_fp32 PRIVATE utility device_normalization_instance)
add_dependencies(test_normalization test_layernorm2d_fp32) add_dependencies(test_normalization test_layernorm2d_fp32)
endif()
add_gtest_executable(test_groupnorm_fp32 test_groupnorm_fp32.cpp)
if(result EQUAL 0)
target_link_libraries(test_groupnorm_fp32 PRIVATE utility device_normalization_instance)
add_dependencies(test_normalization test_groupnorm_fp32) add_dependencies(test_normalization test_groupnorm_fp32)
endif() endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_gtest_executable(test_layernorm2d_fp16 test_layernorm2d_fp16.cpp)
add_gtest_executable(test_layernorm2d_fp16 test_layernorm2d_fp16.cpp) if(result EQUAL 0)
add_gtest_executable(test_groupnorm_fp16 test_groupnorm_fp16.cpp)
target_link_libraries(test_layernorm2d_fp16 PRIVATE utility device_normalization_instance) target_link_libraries(test_layernorm2d_fp16 PRIVATE utility device_normalization_instance)
target_link_libraries(test_groupnorm_fp16 PRIVATE utility device_normalization_instance)
add_dependencies(test_normalization test_layernorm2d_fp16) add_dependencies(test_normalization test_layernorm2d_fp16)
endif()
add_gtest_executable(test_groupnorm_fp16 test_groupnorm_fp16.cpp)
if(result EQUAL 0)
target_link_libraries(test_groupnorm_fp16 PRIVATE utility device_normalization_instance)
add_dependencies(test_normalization test_groupnorm_fp16) add_dependencies(test_normalization test_groupnorm_fp16)
endif() endif()
test/reduce/CMakeLists.txt
View file @ a8629a98
add_test_executable(test_reduce_no_index reduce_no_index.cpp) add_test_executable(test_reduce_no_index reduce_no_index.cpp)
add_test_executable(test_reduce_with_index reduce_with_index.cpp) add_test_executable(test_reduce_with_index reduce_with_index.cpp)
target_link_libraries(test_reduce_no_index PRIVATE utility) target_link_libraries(test_reduce_no_index PRIVATE utility device_reduce_instance)
target_link_libraries(test_reduce_no_index PRIVATE device_reduce_instance) target_link_libraries(test_reduce_with_index PRIVATE utility device_reduce_instance)
target_link_libraries(test_reduce_with_index PRIVATE utility)
target_link_libraries(test_reduce_with_index PRIVATE device_reduce_instance)
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