Skip to content

GitLab

Commit 4100d1d8 authored by Alan Turner's avatar Alan Turner
Browse files

Merge remote-tracking branch 'origin/develop' into migx-flash-attn

parents 48717006 c8a8385f
Show whitespace changes
Inline Side-by-side
Showing with 129 additions and 160 deletions
test/pool_fwd/CMakeLists.txt
View file @ 4100d1d8
add_custom_target(test_pool_fwd)
add_gtest_executable(test_avg_pool2d_fwd test_avg_pool2d_fwd.cpp)
add_gtest_executable(test_avg_pool3d_fwd test_avg_pool3d_fwd.cpp)
add_gtest_executable(test_max_pool2d_fwd test_max_pool2d_fwd.cpp)
add_gtest_executable(test_max_pool3d_fwd test_max_pool3d_fwd.cpp)
target_link_libraries(test_avg_pool2d_fwd PRIVATE utility device_pool_fwd_instance)
target_link_libraries(test_avg_pool3d_fwd PRIVATE utility device_pool_fwd_instance)
target_link_libraries(test_max_pool2d_fwd PRIVATE utility device_pool_fwd_instance)
target_link_libraries(test_max_pool3d_fwd PRIVATE utility device_pool_fwd_instance)
target_link_libraries(test_avg_pool3d_fwd PRIVATE utility device_pool3d_fwd_instance)
target_link_libraries(test_max_pool3d_fwd PRIVATE utility device_pool3d_fwd_instance)
add_dependencies(test_pool_fwd test_avg_pool2d_fwd)
add_dependencies(test_pool_fwd test_avg_pool3d_fwd)
add_dependencies(test_pool_fwd test_max_pool2d_fwd)
add_dependencies(test_pool_fwd test_max_pool3d_fwd)
test/pool_fwd/test_avg_pool2d_fwd.cpp deleted 100644 → 0
View file @ 48717006
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "gtest/gtest.h"
#include "profiler/profile_pool2d_fwd_impl.hpp"
#include "test_pool_fwd_common.hpp"
template <typename Tuple>
class TestAvgPool2dFwd : public ::testing::Test
{
protected:
using InDataType = std::tuple_element_t<0, Tuple>;
using OutDataType = std::tuple_element_t<1, Tuple>;
using ComputeDataType = std::tuple_element_t<2, Tuple>;
using IndexDataType = std::tuple_element_t<3, Tuple>;
std::vector<PoolingParam> params;
void Run()
{
for(auto param : params)
{
bool success =
ck::profiler::profile_pool2d_fwd_impl<InDataType,
OutDataType,
ComputeDataType,
IndexDataType,
ck::ReduceTensorOp::AVG,
false,
false>(true,
2,
false,
false,
param.length_,
param.window_spatial_lengths_,
param.window_strides_,
param.input_left_pads_,
param.input_right_pads_);
EXPECT_TRUE(success);
}
}
};
using KernelTypes =
::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
TYPED_TEST_SUITE(TestAvgPool2dFwd, KernelTypes);
TYPED_TEST(TestAvgPool2dFwd, Test_Pool)
{
// length, window_length, window_stride, left_pad, right_pad
this->params = {{{1, 1, 1, 1}, {1, 1}, {1, 1}, {0, 0}, {0, 0}},
{{2, 16, 64, 64}, {64, 64}, {1, 1}, {0, 0}, {0, 0}},
{{2, 32, 30, 30}, {2, 2}, {2, 2}, {1, 1}, {1, 1}}};
this->Run();
}
test/pool_fwd/test_avg_pool3d_fwd.cpp
View file @ 4100d1d8
......@@ -25,6 +25,8 @@ class TestAvgPool3dFwd : public ::testing::Test
OutDataType,
ComputeDataType,
IndexDataType,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::NDHWC,
ck::ReduceTensorOp::AVG,
false,
false>(true,
......@@ -34,23 +36,27 @@ class TestAvgPool3dFwd : public ::testing::Test
param.length_,
param.window_spatial_lengths_,
param.window_strides_,
param.window_dilations_,
param.input_left_pads_,
param.input_right_pads_);
EXPECT_TRUE(success);
}
}
};
#ifdef CK_ENABLE_FP16
using KernelTypes =
::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
#else
using KernelTypes = ::testing::Types<std::tuple<F32, F32, F32, I32>>;
#endif
TYPED_TEST_SUITE(TestAvgPool3dFwd, KernelTypes);
TYPED_TEST(TestAvgPool3dFwd, Test_Pool)
{
// length, window_length, window_stride, left_pad, right_pad
this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
{{2, 16, 64, 64, 64}, {64, 64, 64}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
{{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}}};
// length, window_length, window_stride, window_dilation, left_pad, right_pad
this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
{{2, 16, 64, 64, 64}, {64, 64, 64}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
{{2, 16, 64, 64, 64}, {4, 4, 4}, {4, 4, 4}, {2, 2, 2}, {0, 0, 0}, {0, 0, 0}},
{{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}}};
this->Run();
}
test/pool_fwd/test_max_pool2d_fwd.cpp deleted 100644 → 0
View file @ 48717006
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "gtest/gtest.h"
#include "profiler/profile_pool2d_fwd_impl.hpp"
#include "test_pool_fwd_common.hpp"
template <typename Tuple>
class TestMaxPool2dFwd : public ::testing::Test
{
protected:
using InDataType = std::tuple_element_t<0, Tuple>;
using OutDataType = std::tuple_element_t<1, Tuple>;
using ComputeDataType = std::tuple_element_t<2, Tuple>;
using IndexDataType = std::tuple_element_t<3, Tuple>;
std::vector<PoolingParam> params;
void Run()
{
for(auto param : params)
{
// max pool
bool success =
ck::profiler::profile_pool2d_fwd_impl<InDataType,
OutDataType,
ComputeDataType,
IndexDataType,
ck::ReduceTensorOp::MAX,
false,
false>(true,
2,
false,
false,
param.length_,
param.window_spatial_lengths_,
param.window_strides_,
param.input_left_pads_,
param.input_right_pads_);
EXPECT_TRUE(success);
// max pool + index
success = ck::profiler::profile_pool2d_fwd_impl<InDataType,
OutDataType,
ComputeDataType,
IndexDataType,
ck::ReduceTensorOp::MAX,
false,
true>(true,
2,
false,
false,
param.length_,
param.window_spatial_lengths_,
param.window_strides_,
param.input_left_pads_,
param.input_right_pads_);
EXPECT_TRUE(success);
}
}
};
using KernelTypes =
::testing::Types<std::tuple<F16, F16, F16, I32>, std::tuple<F32, F32, F32, I32>>;
TYPED_TEST_SUITE(TestMaxPool2dFwd, KernelTypes);
TYPED_TEST(TestMaxPool2dFwd, Test_Pool)
{
// length, window_length, window_stride, left_pad, right_pad
this->params = {{{1, 1, 1, 1}, {1, 1}, {1, 1}, {0, 0}, {0, 0}},
{{2, 16, 64, 64}, {64, 64}, {1, 1}, {0, 0}, {0, 0}},
{{2, 32, 30, 30}, {2, 2}, {2, 2}, {1, 1}, {1, 1}}};
this->Run();
}
test/pool_fwd/test_max_pool3d_fwd.cpp
View file @ 4100d1d8
......@@ -26,6 +26,8 @@ class TestMaxPool3dFwd : public ::testing::Test
OutDataType,
ComputeDataType,
IndexDataType,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::NDHWC,
ck::ReduceTensorOp::MAX,
false,
false>(true,
......@@ -35,6 +37,7 @@ class TestMaxPool3dFwd : public ::testing::Test
param.length_,
param.window_spatial_lengths_,
param.window_strides_,
param.window_dilations_,
param.input_left_pads_,
param.input_right_pads_);
EXPECT_TRUE(success);
......@@ -44,6 +47,8 @@ class TestMaxPool3dFwd : public ::testing::Test
OutDataType,
ComputeDataType,
IndexDataType,
ck::tensor_layout::convolution::NDHWC,
ck::tensor_layout::convolution::NDHWC,
ck::ReduceTensorOp::MAX,
false,
true>(true,
......@@ -53,6 +58,7 @@ class TestMaxPool3dFwd : public ::testing::Test
param.length_,
param.window_spatial_lengths_,
param.window_strides_,
param.window_dilations_,
param.input_left_pads_,
param.input_right_pads_);
EXPECT_TRUE(success);
......@@ -60,16 +66,21 @@ class TestMaxPool3dFwd : public ::testing::Test
}
};
#ifdef CK_ENABLE_FP16
using KernelTypes =
::testing::Types<std::tuple<F16, F16, F16, I32>, std::tuple<F32, F32, F32, I32>>;
::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
#else
using KernelTypes = ::testing::Types<std::tuple<F32, F32, F32, I32>>;
#endif
TYPED_TEST_SUITE(TestMaxPool3dFwd, KernelTypes);
TYPED_TEST(TestMaxPool3dFwd, Test_Pool)
{
// length, window_length, window_stride, left_pad, right_pad
this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
{{2, 16, 64, 64, 64}, {64, 64, 64}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
{{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}}};
// length, window_length, window_stride, window_dilation, left_pad, right_pad
this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
{{2, 16, 64, 64, 64}, {64, 64, 64}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
{{2, 16, 64, 64, 64}, {4, 4, 4}, {4, 4, 4}, {2, 2, 2}, {0, 0, 0}, {0, 0, 0}},
{{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}}};
this->Run();
}
test/pool_fwd/test_pool_fwd_common.hpp
View file @ 4100d1d8
......@@ -14,11 +14,13 @@ struct PoolingParam
PoolingParam(const std::vector<index_t>& length,
const std::vector<index_t>& window_spatial_lengths,
const std::vector<index_t>& window_strides,
const std::vector<index_t>& window_dilations,
const std::vector<index_t>& input_left_pads,
const std::vector<index_t>& input_right_pads)
: length_(length),
window_spatial_lengths_(window_spatial_lengths),
window_strides_(window_strides),
window_dilations_(window_dilations),
input_left_pads_(input_left_pads),
input_right_pads_(input_right_pads)
{
......@@ -26,6 +28,7 @@ struct PoolingParam
std::vector<index_t> length_;
std::vector<index_t> window_spatial_lengths_;
std::vector<index_t> window_strides_;
std::vector<index_t> window_dilations_;
std::vector<index_t> input_left_pads_;
std::vector<index_t> input_right_pads_;
};
test/softmax/test_softmax_rank3.cpp
View file @ 4100d1d8
......@@ -10,10 +10,10 @@
template <ck::index_t N>
using I = ck::Number<N>;
#ifdef CK_ENABLE_FP16
using F16 = ck::half_t;
#endif
using F32 = float;
using I8 = int8_t;
template <typename Tuple>
class TestSoftmax : public ck::TestSoftmax<Tuple>
......@@ -23,9 +23,10 @@ class TestSoftmax : public ck::TestSoftmax<Tuple>
// clang-format off
using KernelTypes = ::testing::Types<
// InDataType, AccDataType, OutDataType, Rank
#ifdef CK_ENABLE_FP16
std::tuple< F16, F32, F16, I<3>>,
std::tuple< F32, F32, F32, I<3>>,
std::tuple< I8, F32, I8, I<3>>
#endif
std::tuple< F32, F32, F32, I<3>>
>;
// clang-format on
......
test/softmax/test_softmax_rank4.cpp
View file @ 4100d1d8
......@@ -10,10 +10,10 @@
template <ck::index_t N>
using I = ck::Number<N>;
#ifdef CK_ENABLE_FP16
using F16 = ck::half_t;
#endif
using F32 = float;
using I8 = int8_t;
template <typename Tuple>
class TestSoftmax : public ck::TestSoftmax<Tuple>
......@@ -23,9 +23,10 @@ class TestSoftmax : public ck::TestSoftmax<Tuple>
// clang-format off
using KernelTypes = ::testing::Types<
// InDataType, AccDataType, OutDataType, Rank
#ifdef CK_ENABLE_FP16
std::tuple< F16, F32, F16, I<4>>,
std::tuple< F32, F32, F32, I<4>>,
std::tuple< I8, F32, I8, I<4>>
#endif
std::tuple< F32, F32, F32, I<4>>
>;
// clang-format on
......
test/softmax/test_softmax_util.hpp
View file @ 4100d1d8
......@@ -61,8 +61,92 @@ class TestSoftmax : public ::testing::Test
int init_method = 1; // integer value initialization
bool log = false;
std::vector<ck::index_t> strides; // intenionally empty, to get packed layout.
bool pass = ck::profiler::profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank>(
verify_, init_method, log, bench_, in_length, strides, reduce_dims, alpha, beta);
bool pass = false;
if constexpr(Rank == 3)
{
if(reduce_dims.size() == 1)
pass = ck::profiler::
profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 1>(verify_,
init_method,
log,
bench_,
in_length,
strides,
reduce_dims,
alpha,
beta);
else if(reduce_dims.size() == 2)
pass = ck::profiler::
profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 2>(verify_,
init_method,
log,
bench_,
in_length,
strides,
reduce_dims,
alpha,
beta);
else if(reduce_dims.size() == 3)
pass = ck::profiler::
profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 3>(verify_,
init_method,
log,
bench_,
in_length,
strides,
reduce_dims,
alpha,
beta);
}
else if constexpr(Rank == 4)
{
if(reduce_dims.size() == 1)
pass = ck::profiler::
profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 1>(verify_,
init_method,
log,
bench_,
in_length,
strides,
reduce_dims,
alpha,
beta);
else if(reduce_dims.size() == 2)
pass = ck::profiler::
profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 2>(verify_,
init_method,
log,
bench_,
in_length,
strides,
reduce_dims,
alpha,
beta);
else if(reduce_dims.size() == 3)
pass = ck::profiler::
profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 3>(verify_,
init_method,
log,
bench_,
in_length,
strides,
reduce_dims,
alpha,
beta);
else if(reduce_dims.size() == 4)
pass = ck::profiler::
profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 4>(verify_,
init_method,
log,
bench_,
in_length,
strides,
reduce_dims,
alpha,
beta);
};
EXPECT_TRUE(pass);
}
......
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