Skip to content

GitLab

"test/vscode:/vscode.git/clone" did not exist on "f35ad4618bbba732f4aa02c3b7ab7182a5df9e70"
Unverified Commit 30480288 authored by Adam Osewski's avatar Adam Osewski Committed by GitHub
Browse files

Refactor device op implementations into `impl` subdirectory. (#420) 



* Move kernel implementation files under impl directory.

* Update examples paths.

* Update device kernel impl include paths.

* Update tensor operation instances include paths.

* Update profiler and tests include paths.

* Clang-format

* Update include paths for batched gemm reduce

* Refactor UnitTest ConvNDBwdWeight.

* Refactor fwd and bwd data convND UT.

* Fix used test macro.

* Fix include path.

* Fix include paths.

* Fix include paths in profiler and tests.

* Fix include paths.
Co-authored-by: default avatarAdam Osewski <aosewski@amd.com>
parent 1b62bfaa
Hide whitespace changes
Inline Side-by-side
Showing with 187 additions and 599 deletions
test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_util.hpp
View file @ 30480288
...@@ -5,7 +5,7 @@ ...@@ -5,7 +5,7 @@
#include <vector> #include <vector>
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp" #include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp"
#include "profiler/include/profile_batched_gemm_softmax_gemm_impl.hpp" #include "profiler/include/profile_batched_gemm_softmax_gemm_impl.hpp"
using ck::tensor_operation::device::GemmSpecialization; using ck::tensor_operation::device::GemmSpecialization;
......
test/convnd_bwd_data/convnd_bwd_data.cpp
View file @ 30480288
...@@ -5,237 +5,89 @@ ...@@ -5,237 +5,89 @@
#include <iostream> #include <iostream>
#include <initializer_list> #include <initializer_list>
#include <vector> #include <vector>
#include <tuple>
#include <gtest/gtest.h> #include <gtest/gtest.h>
#include "profiler/include/profile_conv_bwd_data_impl.hpp" #include "profiler/include/profile_conv_bwd_data_impl.hpp"
template <typename Tuple>
class TestConvndBwdData : public ::testing::Test class TestConvndBwdData : public ::testing::Test
{ {
protected: protected:
using DataType = std::tuple_element_t<0, Tuple>;
std::vector<ck::utils::conv::ConvParam> conv_params; std::vector<ck::utils::conv::ConvParam> conv_params;
};
// 1d template <ck::index_t NDimSpatial>
TEST_F(TestConvndBwdData, Conv1dBwdData) void Run()
{
conv_params.clear();
conv_params.push_back({1, 1, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
conv_params.push_back({1, 1, 128, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
conv_params.push_back({1, 1, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
for(auto& param : conv_params)
{ {
bool pass; for(auto& param : conv_params)
{
// fp32 bool pass;
pass = ck::profiler::profile_conv_bwd_data_impl<1, EXPECT_FALSE(conv_params.empty());
ck::tensor_layout::convolution::NWC, pass = ck::profiler::profile_conv_bwd_data_impl<
ck::tensor_layout::convolution::KXC, NDimSpatial,
ck::tensor_layout::convolution::NWK, ck::tuple_element_t<NDimSpatial - 1,
float, ck::Tuple<ck::tensor_layout::convolution::NWC,
float, ck::tensor_layout::convolution::NHWC,
float>(true, // do_verification ck::tensor_layout::convolution::NDHWC>>,
1, // init_method ck::tuple_element_t<NDimSpatial - 1,
false, // do_log ck::Tuple<ck::tensor_layout::convolution::KXC,
false, // time_kernel ck::tensor_layout::convolution::KYXC,
param); ck::tensor_layout::convolution::KZYXC>>,
ck::tuple_element_t<NDimSpatial - 1,
EXPECT_TRUE(pass); ck::Tuple<ck::tensor_layout::convolution::NWK,
ck::tensor_layout::convolution::NHWK,
// fp16 ck::tensor_layout::convolution::NDHWK>>,
pass = ck::profiler::profile_conv_bwd_data_impl<1, DataType,
ck::tensor_layout::convolution::NWC, DataType,
ck::tensor_layout::convolution::KXC, DataType>(true, // do_verification
ck::tensor_layout::convolution::NWK, 1, // init_method integer value
ck::half_t, false, // do_log
ck::half_t, false, // time_kernel
ck::half_t>(true, // do_verification param);
1, // init_method EXPECT_TRUE(pass);
false, // do_log }
false, // time_kernel }
param); };
EXPECT_TRUE(pass);
// bf16
pass = ck::profiler::profile_conv_bwd_data_impl<1,
ck::tensor_layout::convolution::NWC,
ck::tensor_layout::convolution::KXC,
ck::tensor_layout::convolution::NWK,
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 using KernelTypes = ::testing::Types<std::tuple<float>,
pass = ck::profiler::profile_conv_bwd_data_impl<1, std::tuple<ck::half_t>,
ck::tensor_layout::convolution::NWC, std::tuple<ck::bhalf_t>,
ck::tensor_layout::convolution::KXC, std::tuple<std::int8_t>>;
ck::tensor_layout::convolution::NWK, TYPED_TEST_SUITE(TestConvndBwdData, KernelTypes);
int8_t,
int8_t,
int8_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass); // 1d
} TYPED_TEST(TestConvndBwdData, Conv1dBwdData)
{
this->conv_params.clear();
this->conv_params.push_back({1, 1, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
this->conv_params.push_back({1, 1, 128, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
this->conv_params.push_back({1, 1, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
this->template Run<1>();
} }
// 2d // 2d
TEST_F(TestConvndBwdData, Conv2dBwdData) TYPED_TEST(TestConvndBwdData, Conv2dBwdData)
{ {
conv_params.clear(); this->conv_params.clear();
conv_params.push_back({2, 1, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}}); this->conv_params.push_back(
conv_params.push_back({2, 1, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}}); {2, 1, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
conv_params.push_back({2, 1, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}}); this->conv_params.push_back(
{2, 1, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
for(auto& param : conv_params) this->conv_params.push_back(
{ {2, 1, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
bool pass; this->template Run<2>();
// fp32
pass = ck::profiler::profile_conv_bwd_data_impl<2,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK,
float,
float,
float>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// fp16
pass = ck::profiler::profile_conv_bwd_data_impl<2,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK,
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_conv_bwd_data_impl<2,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK,
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_conv_bwd_data_impl<2,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK,
int8_t,
int8_t,
int8_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
}
} }
// 3d // 3d
TEST_F(TestConvndBwdData, Conv3dBwdData) TYPED_TEST(TestConvndBwdData, Conv3dBwdData)
{ {
conv_params.clear(); this->conv_params.clear();
conv_params.push_back( this->conv_params.push_back(
{3, 1, 128, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}}); {3, 1, 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( this->conv_params.push_back(
{3, 1, 128, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}}); {3, 1, 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( this->conv_params.push_back(
{3, 1, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}}); {3, 1, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
this->template Run<3>();
for(auto& param : conv_params)
{
bool pass;
// fp32
pass = ck::profiler::profile_conv_bwd_data_impl<3,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::KZYXC,
ck::tensor_layout::convolution::NDHWK,
float,
float,
float>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// fp16
pass = ck::profiler::profile_conv_bwd_data_impl<3,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::KZYXC,
ck::tensor_layout::convolution::NDHWK,
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_conv_bwd_data_impl<3,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::KZYXC,
ck::tensor_layout::convolution::NDHWK,
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_conv_bwd_data_impl<3,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::KZYXC,
ck::tensor_layout::convolution::NDHWK,
int8_t,
int8_t,
int8_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
}
} }
test/convnd_bwd_weight/convnd_bwd_weight.cpp
View file @ 30480288
...@@ -5,201 +5,86 @@ ...@@ -5,201 +5,86 @@
#include <iostream> #include <iostream>
#include <initializer_list> #include <initializer_list>
#include <vector> #include <vector>
#include <tuple>
#include <gtest/gtest.h> #include <gtest/gtest.h>
#include "profiler/include/profile_conv_bwd_weight_impl.hpp" #include "profiler/include/profile_conv_bwd_weight_impl.hpp"
template <typename Tuple>
class TestConvndBwdWeight : public ::testing::Test class TestConvndBwdWeight : public ::testing::Test
{ {
protected: protected:
using DataType = std::tuple_element_t<0, Tuple>;
std::vector<ck::utils::conv::ConvParam> conv_params; std::vector<ck::utils::conv::ConvParam> conv_params;
}; ck::index_t split_k{2};
// 1d
TEST_F(TestConvndBwdWeight, Conv1dBwdWeight)
{
conv_params.clear();
conv_params.push_back({1, 1, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
conv_params.push_back({1, 1, 128, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
conv_params.push_back({1, 1, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
for(auto& param : conv_params) template <ck::index_t NDimSpatial>
void Run()
{ {
bool pass; for(auto& param : conv_params)
{
// fp32 bool pass;
pass = ck::profiler::profile_conv_bwd_weight_impl<1, EXPECT_FALSE(conv_params.empty());
ck::tensor_layout::convolution::NWC, pass = ck::profiler::profile_conv_bwd_weight_impl<
ck::tensor_layout::convolution::KXC, NDimSpatial,
ck::tensor_layout::convolution::NWK, ck::tuple_element_t<NDimSpatial - 1,
float, ck::Tuple<ck::tensor_layout::convolution::NWC,
float, ck::tensor_layout::convolution::NHWC,
float>(true, // do_verification ck::tensor_layout::convolution::NDHWC>>,
1, // init_method ck::tuple_element_t<NDimSpatial - 1,
false, // do_log ck::Tuple<ck::tensor_layout::convolution::KXC,
false, // time_kernel ck::tensor_layout::convolution::KYXC,
param, ck::tensor_layout::convolution::KZYXC>>,
2); ck::tuple_element_t<NDimSpatial - 1,
ck::Tuple<ck::tensor_layout::convolution::NWK,
EXPECT_TRUE(pass); ck::tensor_layout::convolution::NHWK,
ck::tensor_layout::convolution::NDHWK>>,
// fp16 DataType,
pass = ck::profiler::profile_conv_bwd_weight_impl<1, DataType,
ck::tensor_layout::convolution::NWC, DataType>(true, // do_verification
ck::tensor_layout::convolution::KXC, 1, // init_method integer value
ck::tensor_layout::convolution::NWK, false, // do_log
ck::half_t, false, // time_kernel
ck::half_t, param,
ck::half_t>(true, // do_verification split_k);
1, // init_method EXPECT_TRUE(pass);
false, // do_log }
false, // time_kernel }
param, };
2);
EXPECT_TRUE(pass);
// bf16 using KernelTypes =
pass = ck::profiler::profile_conv_bwd_weight_impl<1, ::testing::Types<std::tuple<float>, std::tuple<ck::half_t>, std::tuple<ck::bhalf_t>>;
ck::tensor_layout::convolution::NWC, TYPED_TEST_SUITE(TestConvndBwdWeight, KernelTypes);
ck::tensor_layout::convolution::KXC,
ck::tensor_layout::convolution::NWK,
ck::bhalf_t,
ck::bhalf_t,
ck::bhalf_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param,
2);
EXPECT_TRUE(pass); TYPED_TEST(TestConvndBwdWeight, Test1D)
} {
this->conv_params.clear();
this->conv_params.push_back({1, 1, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
this->conv_params.push_back({1, 1, 128, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
this->conv_params.push_back({1, 1, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
this->template Run<1>();
} }
// 2d TYPED_TEST(TestConvndBwdWeight, Test2D)
TEST_F(TestConvndBwdWeight, Conv2dBwdWeight)
{ {
conv_params.clear(); this->conv_params.clear();
conv_params.push_back({2, 1, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}}); this->conv_params.push_back(
conv_params.push_back({2, 1, 32, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}}); {2, 1, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
conv_params.push_back({2, 1, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}}); this->conv_params.push_back(
{2, 1, 32, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
for(auto& param : conv_params) this->conv_params.push_back(
{ {2, 1, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
bool pass; this->template Run<2>();
// fp32
pass = ck::profiler::profile_conv_bwd_weight_impl<2,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK,
float,
float,
float>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param,
2);
EXPECT_TRUE(pass);
// fp16
pass = ck::profiler::profile_conv_bwd_weight_impl<2,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK,
ck::half_t,
ck::half_t,
ck::half_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param,
2);
EXPECT_TRUE(pass);
// bf16
pass = ck::profiler::profile_conv_bwd_weight_impl<2,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK,
ck::bhalf_t,
ck::bhalf_t,
ck::bhalf_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param,
2);
EXPECT_TRUE(pass);
}
} }
// 3d TYPED_TEST(TestConvndBwdWeight, Test3D)
TEST_F(TestConvndBwdWeight, Conv3dBwdWeight)
{ {
conv_params.clear(); this->conv_params.clear();
conv_params.push_back( this->conv_params.push_back(
{3, 1, 128, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}}); {3, 1, 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( this->conv_params.push_back(
{3, 1, 32, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}}); {3, 1, 32, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
conv_params.push_back( this->conv_params.push_back(
{3, 1, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}}); {3, 1, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
this->template Run<3>();
for(auto& param : conv_params)
{
bool pass;
// fp32
pass = ck::profiler::profile_conv_bwd_weight_impl<3,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::KZYXC,
ck::tensor_layout::convolution::NDHWK,
float,
float,
float>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param,
2);
EXPECT_TRUE(pass);
// fp16
pass = ck::profiler::profile_conv_bwd_weight_impl<3,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::KZYXC,
ck::tensor_layout::convolution::NDHWK,
ck::half_t,
ck::half_t,
ck::half_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param,
2);
EXPECT_TRUE(pass);
// bf16
pass = ck::profiler::profile_conv_bwd_weight_impl<3,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::KZYXC,
ck::tensor_layout::convolution::NDHWK,
ck::bhalf_t,
ck::bhalf_t,
ck::bhalf_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param,
2);
EXPECT_TRUE(pass);
}
} }
test/convnd_fwd/convnd_fwd.cpp
View file @ 30480288
...@@ -5,237 +5,88 @@ ...@@ -5,237 +5,88 @@
#include <iostream> #include <iostream>
#include <initializer_list> #include <initializer_list>
#include <vector> #include <vector>
#include <tuple>
#include <gtest/gtest.h> #include <gtest/gtest.h>
#include "profiler/include/profile_conv_fwd_impl.hpp" #include "profiler/include/profile_conv_fwd_impl.hpp"
template <typename Tuple>
class TestConvndFwd : public ::testing::Test class TestConvndFwd : public ::testing::Test
{ {
protected: protected:
using DataType = std::tuple_element_t<0, Tuple>;
std::vector<ck::utils::conv::ConvParam> conv_params; std::vector<ck::utils::conv::ConvParam> conv_params;
};
// 1d template <ck::index_t NDimSpatial>
TEST_F(TestConvndFwd, Conv1dFwd) void Run()
{
conv_params.clear();
conv_params.push_back({1, 1, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
conv_params.push_back({1, 1, 128, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
conv_params.push_back({1, 1, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
for(auto& param : conv_params)
{ {
bool pass; for(auto& param : conv_params)
{
// fp32 bool pass;
pass = ck::profiler::profile_conv_fwd_impl<1, EXPECT_FALSE(conv_params.empty());
ck::tensor_layout::convolution::NWC, pass = ck::profiler::profile_conv_fwd_impl<
ck::tensor_layout::convolution::KXC, NDimSpatial,
ck::tensor_layout::convolution::NWK, ck::tuple_element_t<NDimSpatial - 1,
float, ck::Tuple<ck::tensor_layout::convolution::NWC,
float, ck::tensor_layout::convolution::NHWC,
float>(true, // do_verification ck::tensor_layout::convolution::NDHWC>>,
1, // init_method ck::tuple_element_t<NDimSpatial - 1,
false, // do_log ck::Tuple<ck::tensor_layout::convolution::KXC,
false, // time_kernel ck::tensor_layout::convolution::KYXC,
param); ck::tensor_layout::convolution::KZYXC>>,
ck::tuple_element_t<NDimSpatial - 1,
EXPECT_TRUE(pass); ck::Tuple<ck::tensor_layout::convolution::NWK,
ck::tensor_layout::convolution::NHWK,
// fp16 ck::tensor_layout::convolution::NDHWK>>,
pass = ck::profiler::profile_conv_fwd_impl<1, DataType,
ck::tensor_layout::convolution::NWC, DataType,
ck::tensor_layout::convolution::KXC, DataType>(true, // do_verification
ck::tensor_layout::convolution::NWK, 1, // init_method integer value
ck::half_t, false, // do_log
ck::half_t, false, // time_kernel
ck::half_t>(true, // do_verification param);
1, // init_method EXPECT_TRUE(pass);
false, // do_log }
false, // time_kernel }
param); };
EXPECT_TRUE(pass);
// bf16
pass = ck::profiler::profile_conv_fwd_impl<1,
ck::tensor_layout::convolution::NWC,
ck::tensor_layout::convolution::KXC,
ck::tensor_layout::convolution::NWK,
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 using KernelTypes = ::testing::Types<std::tuple<float>,
pass = ck::profiler::profile_conv_fwd_impl<1, std::tuple<ck::half_t>,
ck::tensor_layout::convolution::NWC, std::tuple<ck::bhalf_t>,
ck::tensor_layout::convolution::KXC, std::tuple<std::int8_t>>;
ck::tensor_layout::convolution::NWK, TYPED_TEST_SUITE(TestConvndFwd, KernelTypes);
int8_t,
int8_t,
int8_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass); // 1d
} TYPED_TEST(TestConvndFwd, Conv1dFwd)
{
this->conv_params.clear();
this->conv_params.push_back({1, 1, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
this->conv_params.push_back({1, 1, 128, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
this->conv_params.push_back({1, 1, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
this->template Run<1>();
} }
// 2d // 2d
TEST_F(TestConvndFwd, Conv2dFwd) TYPED_TEST(TestConvndFwd, Conv2dFwd)
{ {
conv_params.clear(); this->conv_params.clear();
conv_params.push_back({2, 1, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}}); this->conv_params.push_back(
conv_params.push_back({2, 1, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}}); {2, 1, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
conv_params.push_back({2, 1, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}}); this->conv_params.push_back(
{2, 1, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
for(auto& param : conv_params) this->conv_params.push_back(
{ {2, 1, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
bool pass; this->template Run<2>();
// fp32
pass = ck::profiler::profile_conv_fwd_impl<2,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK,
float,
float,
float>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// fp16
pass = ck::profiler::profile_conv_fwd_impl<2,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK,
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_conv_fwd_impl<2,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK,
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_conv_fwd_impl<2,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK,
int8_t,
int8_t,
int8_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
}
} }
// 3d // 3d
TEST_F(TestConvndFwd, Conv3dFwd) TYPED_TEST(TestConvndFwd, Conv3dFwd)
{ {
conv_params.clear(); this->conv_params.clear();
conv_params.push_back( this->conv_params.push_back(
{3, 1, 128, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}}); {3, 1, 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( this->conv_params.push_back(
{3, 1, 128, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}}); {3, 1, 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( this->conv_params.push_back(
{3, 1, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}}); {3, 1, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
this->template Run<3>();
for(auto& param : conv_params)
{
bool pass;
// fp32
pass = ck::profiler::profile_conv_fwd_impl<3,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::KZYXC,
ck::tensor_layout::convolution::NDHWK,
float,
float,
float>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
// fp16
pass = ck::profiler::profile_conv_fwd_impl<3,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::KZYXC,
ck::tensor_layout::convolution::NDHWK,
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_conv_fwd_impl<3,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::KZYXC,
ck::tensor_layout::convolution::NDHWK,
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_conv_fwd_impl<3,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::KZYXC,
ck::tensor_layout::convolution::NDHWK,
int8_t,
int8_t,
int8_t>(true, // do_verification
1, // init_method
false, // do_log
false, // time_kernel
param);
EXPECT_TRUE(pass);
}
} }
test/normalization/test_layernorm2d_util.hpp
View file @ 30480288
...@@ -9,7 +9,7 @@ ...@@ -9,7 +9,7 @@
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/utility/number.hpp" #include "ck/utility/number.hpp"
#include "ck/tensor_operation/gpu/device/device_normalization_impl.hpp" #include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
......
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