Skip to content

GitLab

Commit e70a4d19 authored by Jun Liu's avatar Jun Liu
Browse files

Merge branch 'amd-develop' into amd-master

parents ce72f286 0dacd895
Hide whitespace changes
Inline Side-by-side
Showing with 165 additions and 89 deletions
client_example/09_quantization/conv2d_fwd_bias_tanh_perchannel_quantization.cpp
View file @ e70a4d19
...@@ -83,7 +83,7 @@ int main(int argc, char* argv[]) ...@@ -83,7 +83,7 @@ int main(int argc, char* argv[])
SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * G * K); SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * G * K);
SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K); SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD< using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
NumDimSpatial, NumDimSpatial,
InLayout, InLayout,
WeiLayout, WeiLayout,
......
client_example/09_quantization/conv2d_fwd_bias_tanh_perlayer_quantization.cpp
View file @ e70a4d19
...@@ -79,18 +79,18 @@ int main(int argc, char* argv[]) ...@@ -79,18 +79,18 @@ int main(int argc, char* argv[])
SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K); SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
using DeviceOp = using DeviceOp =
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial, ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
InLayout, InLayout,
WeiLayout, WeiLayout,
ck::Tuple<BiasLayout>, ck::Tuple<BiasLayout>,
OutLayout, OutLayout,
InDataType, InDataType,
WeiDataType, WeiDataType,
ck::Tuple<BiasDataType>, ck::Tuple<BiasDataType>,
OutDataType, OutDataType,
PassThrough, PassThrough,
PassThrough, PassThrough,
OutElementOp>; OutElementOp>;
// get device op instances // get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory< const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances(); DeviceOp>::GetInstances();
......
client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp
View file @ e70a4d19
...@@ -76,19 +76,19 @@ int main(int argc, char* argv[]) ...@@ -76,19 +76,19 @@ int main(int argc, char* argv[])
SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * G * K); SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * G * K);
SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K); SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
using DeviceOp = using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial, NumDimSpatial,
InLayout, InLayout,
WeiLayout, WeiLayout,
ck::Tuple<RequantScaleLayout>, ck::Tuple<RequantScaleLayout>,
OutLayout, OutLayout,
InDataType, InDataType,
WeiDataType, WeiDataType,
ck::Tuple<RequantScaleDataType>, ck::Tuple<RequantScaleDataType>,
OutDataType, OutDataType,
PassThrough, PassThrough,
PassThrough, PassThrough,
OutElementOp>; OutElementOp>;
// get device op instances // get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory< const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances(); DeviceOp>::GetInstances();
......
client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp
View file @ e70a4d19
...@@ -72,18 +72,18 @@ int main(int argc, char* argv[]) ...@@ -72,18 +72,18 @@ int main(int argc, char* argv[])
SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C); SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K); SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial, using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
InLayout, InLayout,
WeiLayout, WeiLayout,
ck::Tuple<>, ck::Tuple<>,
OutLayout, OutLayout,
InDataType, InDataType,
WeiDataType, WeiDataType,
ck::Tuple<>, ck::Tuple<>,
OutDataType, OutDataType,
PassThrough, PassThrough,
PassThrough, PassThrough,
OutElementOp>; OutElementOp>;
// get device op instances // get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory< const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances(); DeviceOp>::GetInstances();
......
client_example/10_grouped_convnd_bwd_data/CMakeLists.txt
View file @ e70a4d19
add_executable(client_grouped_conv2d_bwd_data grouped_conv2d_bwd_data.cpp) add_executable(client_grouped_conv2d_bwd_data grouped_conv2d_bwd_data.cpp)
target_link_libraries(client_grouped_conv2d_bwd_data PRIVATE composable_kernel::device_operations) target_link_libraries(client_grouped_conv2d_bwd_data PRIVATE composable_kernel::device_conv_operations)
add_executable(client_grouped_conv3d_bwd_data grouped_conv3d_bwd_data.cpp) add_executable(client_grouped_conv3d_bwd_data grouped_conv3d_bwd_data.cpp)
target_link_libraries(client_grouped_conv3d_bwd_data PRIVATE composable_kernel::device_operations) target_link_libraries(client_grouped_conv3d_bwd_data PRIVATE composable_kernel::device_conv_operations)
add_executable(client_grouped_conv3d_bwd_data_input_fp16_comp_bf8f8 grouped_conv3d_bwd_data_input_fp16_comp_bf8f8.cpp) add_executable(client_grouped_conv3d_bwd_data_input_fp16_comp_bf8f8 grouped_conv3d_bwd_data_input_fp16_comp_bf8f8.cpp)
target_link_libraries(client_grouped_conv3d_bwd_data_input_fp16_comp_bf8f8 PRIVATE composable_kernel::device_operations) target_link_libraries(client_grouped_conv3d_bwd_data_input_fp16_comp_bf8f8 PRIVATE composable_kernel::device_conv_operations)
client_example/11_grouped_conv_bwd_weight/CMakeLists.txt
View file @ e70a4d19
...@@ -2,8 +2,10 @@ add_executable(client_grouped_conv1d_bwd_weight_fp16 grouped_conv1d_bwd_weight_f ...@@ -2,8 +2,10 @@ add_executable(client_grouped_conv1d_bwd_weight_fp16 grouped_conv1d_bwd_weight_f
add_executable(client_grouped_conv2d_bwd_weight_fp16 grouped_conv2d_bwd_weight_fp16.cpp) add_executable(client_grouped_conv2d_bwd_weight_fp16 grouped_conv2d_bwd_weight_fp16.cpp)
add_executable(client_grouped_conv3d_bwd_weight_fp16 grouped_conv3d_bwd_weight_fp16.cpp) add_executable(client_grouped_conv3d_bwd_weight_fp16 grouped_conv3d_bwd_weight_fp16.cpp)
add_executable(client_grouped_conv3d_bwd_weight_fp32 grouped_conv3d_bwd_weight_fp32.cpp) add_executable(client_grouped_conv3d_bwd_weight_fp32 grouped_conv3d_bwd_weight_fp32.cpp)
add_executable(client_grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8 grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8.cpp)
target_link_libraries(client_grouped_conv1d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations) target_link_libraries(client_grouped_conv1d_bwd_weight_fp16 PRIVATE composable_kernel::device_conv_operations)
target_link_libraries(client_grouped_conv2d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations) target_link_libraries(client_grouped_conv2d_bwd_weight_fp16 PRIVATE composable_kernel::device_conv_operations)
target_link_libraries(client_grouped_conv3d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations) target_link_libraries(client_grouped_conv3d_bwd_weight_fp16 PRIVATE composable_kernel::device_conv_operations)
target_link_libraries(client_grouped_conv3d_bwd_weight_fp32 PRIVATE composable_kernel::device_operations) target_link_libraries(client_grouped_conv3d_bwd_weight_fp32 PRIVATE composable_kernel::device_conv_operations)
target_link_libraries(client_grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8 PRIVATE composable_kernel::device_conv_operations)
client_example/11_grouped_conv_bwd_weight/common.hpp
View file @ e70a4d19
...@@ -85,7 +85,9 @@ template <ck::index_t NumDimSpatial, ...@@ -85,7 +85,9 @@ template <ck::index_t NumDimSpatial,
typename OutDataType, typename OutDataType,
typename InLayout, typename InLayout,
typename WeiLayout, typename WeiLayout,
typename OutLayout> typename OutLayout,
typename AComputeType = InDataType,
typename BComputeType = AComputeType>
bool run_grouped_conv_bwd_weight( bool run_grouped_conv_bwd_weight(
const std::array<ck::index_t, NumDimSpatial + 3>& input_lengths, const std::array<ck::index_t, NumDimSpatial + 3>& input_lengths,
const std::array<ck::index_t, NumDimSpatial + 3>& input_strides, const std::array<ck::index_t, NumDimSpatial + 3>& input_strides,
...@@ -113,7 +115,9 @@ bool run_grouped_conv_bwd_weight( ...@@ -113,7 +115,9 @@ bool run_grouped_conv_bwd_weight(
OutDataType, OutDataType,
PassThrough, PassThrough,
PassThrough, PassThrough,
PassThrough>; PassThrough,
AComputeType,
BComputeType>;
// get device op instances // get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory< const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances(); DeviceOp>::GetInstances();
......
client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8.cpp 0 → 100644
View file @ e70a4d19
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
using InDataType = ck::half_t;
using WeiDataType = ck::half_t;
using OutDataType = ck::half_t;
using InLayout = ck::tensor_layout::convolution::NDHWGC;
using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
using OutLayout = ck::tensor_layout::convolution::NDHWGK;
using AComputeType = ck::bf8_t;
using BComputeType = ck::f8_t;
static constexpr ck::index_t NumDimSpatial = 3;
static constexpr ck::index_t G = 8;
static constexpr ck::index_t N = 64;
static constexpr ck::index_t K = 128;
static constexpr ck::index_t C = 128;
static constexpr ck::index_t Z = 3;
static constexpr ck::index_t Y = 3;
static constexpr ck::index_t X = 3;
static constexpr ck::index_t Di = 28;
static constexpr ck::index_t Hi = 28;
static constexpr ck::index_t Wi = 3;
static constexpr ck::index_t Do = 28;
static constexpr ck::index_t Ho = 28;
static constexpr ck::index_t Wo = 3;
static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_lengths{G, N, C, Di, Hi, Wi};
static constexpr std::array<ck::index_t, NumDimSpatial + 3> filter_lengths{G, K, C, Z, Y, X};
static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_lengths{G, N, K, Do, Ho, Wo};
static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{
N * Di * Hi * Wi * C, Di* Hi* Wi* C, 1, Hi* Wi* C, Wi* C, C};
static constexpr std::array<ck::index_t, NumDimSpatial + 3> weights_strides{
K * Z * Y * X * C, Z* Y* X* C, 1, Y* X* C, X* C, C};
static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{
N * Do * Ho * Wo * K, Do* Ho* Wo* K, 1, Ho* Wo* K, Wo* K, K};
static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1, 1, 1};
static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1, 1, 1};
static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
static constexpr std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1, 1};
int main()
{
return run_grouped_conv_bwd_weight<NumDimSpatial,
InDataType,
WeiDataType,
OutDataType,
InLayout,
WeiLayout,
OutLayout,
AComputeType,
BComputeType>(input_lengths,
input_strides,
filter_lengths,
weights_strides,
output_lengths,
output_strides,
conv_filter_strides,
conv_filter_dilations,
input_left_pads,
input_right_pads)
? EXIT_SUCCESS
: EXIT_FAILURE;
}
client_example/12_elementwise_normalization/CMakeLists.txt
View file @ e70a4d19
add_executable(client_elementwise_layernorm2d elementwise_layernorm2d.cpp) add_executable(client_elementwise_layernorm2d elementwise_layernorm2d.cpp)
target_link_libraries(client_elementwise_layernorm2d PRIVATE composable_kernel::device_operations) target_link_libraries(client_elementwise_layernorm2d PRIVATE composable_kernel::device_other_operations)
client_example/13_batchnorm/CMakeLists.txt
View file @ e70a4d19
add_executable(client_batchnorm_fwd_nhwc batchnorm_fwd_nhwc.cpp) add_executable(client_batchnorm_fwd_nhwc batchnorm_fwd_nhwc.cpp)
add_executable(client_batchnorm_bwd_nhwc batchnorm_bwd_nhwc.cpp) add_executable(client_batchnorm_bwd_nhwc batchnorm_bwd_nhwc.cpp)
add_executable(client_batchnorm_infer_nhwc batchnorm_infer_nhwc.cpp) add_executable(client_batchnorm_infer_nhwc batchnorm_infer_nhwc.cpp)
target_link_libraries(client_batchnorm_fwd_nhwc PRIVATE composable_kernel::device_operations) target_link_libraries(client_batchnorm_fwd_nhwc PRIVATE composable_kernel::device_other_operations)
target_link_libraries(client_batchnorm_bwd_nhwc PRIVATE composable_kernel::device_operations) target_link_libraries(client_batchnorm_bwd_nhwc PRIVATE composable_kernel::device_other_operations)
target_link_libraries(client_batchnorm_infer_nhwc PRIVATE composable_kernel::device_operations) target_link_libraries(client_batchnorm_infer_nhwc PRIVATE composable_kernel::device_other_operations)
client_example/14_instance_id/CMakeLists.txt
View file @ e70a4d19
add_executable(client_batchnorm_fwd_instance_id batchnorm_fwd_instance_id.cpp) add_executable(client_batchnorm_fwd_instance_id batchnorm_fwd_instance_id.cpp)
target_link_libraries(client_batchnorm_fwd_instance_id PRIVATE composable_kernel::device_operations) target_link_libraries(client_batchnorm_fwd_instance_id PRIVATE composable_kernel::device_other_operations)
client_example/15_convnd_bwd_data/CMakeLists.txt
View file @ e70a4d19
add_executable(client_conv3d_bwd_data_fp16 conv3d_bwd_data_fp16.cpp) add_executable(client_conv3d_bwd_data_fp16 conv3d_bwd_data_fp16.cpp)
add_executable(client_conv3d_bwd_data_fp32 conv3d_bwd_data_fp32.cpp) add_executable(client_conv3d_bwd_data_fp32 conv3d_bwd_data_fp32.cpp)
target_link_libraries(client_conv3d_bwd_data_fp16 PRIVATE composable_kernel::device_operations) target_link_libraries(client_conv3d_bwd_data_fp16 PRIVATE composable_kernel::device_conv_operations)
target_link_libraries(client_conv3d_bwd_data_fp32 PRIVATE composable_kernel::device_operations) target_link_libraries(client_conv3d_bwd_data_fp32 PRIVATE composable_kernel::device_conv_operations)
client_example/15_gemm_add_multiply/CMakeLists.txt
View file @ e70a4d19
add_executable(client_gemm_add_multiply gemm_add_multiply.cpp) add_executable(client_gemm_add_multiply gemm_add_multiply.cpp)
target_link_libraries(client_gemm_add_multiply PRIVATE composable_kernel::device_operations) target_link_libraries(client_gemm_add_multiply PRIVATE composable_kernel::device_gemm_operations)
\ No newline at end of file \ No newline at end of file
client_example/15_reduce/CMakeLists.txt
View file @ e70a4d19
add_executable(client_reduce_nhwc_c reduce_nhwc_c.cpp) add_executable(client_reduce_nhwc_c reduce_nhwc_c.cpp)
target_link_libraries(client_reduce_nhwc_c PRIVATE composable_kernel::device_operations) target_link_libraries(client_reduce_nhwc_c PRIVATE composable_kernel::device_reduction_operations)
client_example/16_convnd_fwd/CMakeLists.txt
View file @ e70a4d19
if((DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES) if((DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
add_executable(client_conv3d_fwd_fp16 conv3d_fwd_fp16.cpp) add_executable(client_conv3d_fwd_fp16 conv3d_fwd_fp16.cpp)
target_link_libraries(client_conv3d_fwd_fp16 PRIVATE composable_kernel::device_operations) target_link_libraries(client_conv3d_fwd_fp16 PRIVATE composable_kernel::device_conv_operations)
endif() endif()
if((DTYPES MATCHES "fp8") OR NOT DEFINED DTYPES) if((DTYPES MATCHES "fp8") OR NOT DEFINED DTYPES)
add_executable(client_conv3d_fwd_fp16_comp_fp8 conv3d_fwd_fp16_comp_fp8.cpp) add_executable(client_conv3d_fwd_fp16_comp_fp8 conv3d_fwd_fp16_comp_fp8.cpp)
target_link_libraries(client_conv3d_fwd_fp16_comp_fp8 PRIVATE composable_kernel::device_operations) target_link_libraries(client_conv3d_fwd_fp16_comp_fp8 PRIVATE composable_kernel::device_conv_operations)
endif() endif()
if((DTYPES MATCHES "fp32") OR NOT DEFINED DTYPES) if((DTYPES MATCHES "fp32") OR NOT DEFINED DTYPES)
add_executable(client_conv3d_fwd_fp32 conv3d_fwd_fp32.cpp) add_executable(client_conv3d_fwd_fp32 conv3d_fwd_fp32.cpp)
target_link_libraries(client_conv3d_fwd_fp32 PRIVATE composable_kernel::device_operations) target_link_libraries(client_conv3d_fwd_fp32 PRIVATE composable_kernel::device_conv_operations)
endif() endif()
client_example/16_convnd_fwd/common.hpp
View file @ e70a4d19
...@@ -11,7 +11,7 @@ ...@@ -11,7 +11,7 @@
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp" #include "ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp"
#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp" #include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_abd.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
using PassThrough = ck::tensor_operation::element_wise::PassThrough; using PassThrough = ck::tensor_operation::element_wise::PassThrough;
...@@ -174,19 +174,19 @@ bool run_grouped_conv_fwd(std::array<ck::index_t, NumDimSpatial + NumNonSpatialD ...@@ -174,19 +174,19 @@ bool run_grouped_conv_fwd(std::array<ck::index_t, NumDimSpatial + NumNonSpatialD
std::size_t flop = GetFlops<NumDimSpatial>(out_lengths, wei_lengths); std::size_t flop = GetFlops<NumDimSpatial>(out_lengths, wei_lengths);
std::size_t num_bytes = in_mem_size + wei_mem_size + out_mem_size; std::size_t num_bytes = in_mem_size + wei_mem_size + out_mem_size;
using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial, using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
InLayout, InLayout,
WeiLayout, WeiLayout,
ck::Tuple<>, ck::Tuple<>,
OutLayout, OutLayout,
InDataType, InDataType,
WeiDataType, WeiDataType,
ck::Tuple<>, ck::Tuple<>,
OutDataType, OutDataType,
PassThrough, PassThrough,
PassThrough, PassThrough,
PassThrough, PassThrough,
ComputeType>; ComputeType>;
// get device op instances // get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory< const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances(); DeviceOp>::GetInstances();
......
client_example/17_grouped_gemm_fastgelu/CMakeLists.txt
View file @ e70a4d19
add_executable(client_grouped_gemm_fastgelu grouped_gemm_fastgelu.cpp) add_executable(client_grouped_gemm_fastgelu grouped_gemm_fastgelu.cpp)
target_link_libraries(client_grouped_gemm_fastgelu PRIVATE composable_kernel::device_operations) target_link_libraries(client_grouped_gemm_fastgelu PRIVATE composable_kernel::device_gemm_operations)
\ No newline at end of file \ No newline at end of file
client_example/18_groupnorm/CMakeLists.txt
View file @ e70a4d19
add_executable(client_groupnorm_swish groupnorm_swish.cpp) add_executable(client_groupnorm_swish groupnorm_swish.cpp)
target_link_libraries(client_groupnorm_swish PRIVATE composable_kernel::device_operations) target_link_libraries(client_groupnorm_swish PRIVATE composable_kernel::device_other_operations)
client_example/18_groupnorm/groupnorm_swish.cpp
View file @ e70a4d19
...@@ -7,10 +7,10 @@ ...@@ -7,10 +7,10 @@
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp" #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_normalization.hpp" #include "ck/tensor_operation/gpu/device/device_normalization_fwd.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/gpu/normalization_swish.hpp" #include "ck/library/tensor_operation_instance/gpu/normalization_fwd_swish.hpp"
using XDataType = ck::half_t; using XDataType = ck::half_t;
using GammaDataType = float; using GammaDataType = float;
...@@ -64,14 +64,14 @@ int main(int argc, char* argv[]) ...@@ -64,14 +64,14 @@ int main(int argc, char* argv[])
SimpleDeviceMem save_inv_std_device_buf(sizeof(SaveMeanInvStdDataType) * N * G); SimpleDeviceMem save_inv_std_device_buf(sizeof(SaveMeanInvStdDataType) * N * G);
#endif #endif
using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType, using DeviceOp = ck::tensor_operation::device::DeviceNormalizationFwd<XDataType,
GammaDataType, GammaDataType,
BetaDataType, BetaDataType,
YDataType, YDataType,
SaveMeanInvStdDataType, SaveMeanInvStdDataType,
Swish, Swish,
Rank, Rank,
NumReduceDim>; NumReduceDim>;
// get device op instances // get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory< const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
......
client_example/19_pool/CMakeLists.txt
View file @ e70a4d19
add_executable(client_max_pool2d_fwd max_pool2d_fwd.cpp) add_executable(client_max_pool2d_fwd max_pool2d_fwd.cpp)
target_link_libraries(client_max_pool2d_fwd PRIVATE composable_kernel::device_operations) target_link_libraries(client_max_pool2d_fwd PRIVATE composable_kernel::device_other_operations)
add_executable(client_max_pool2d_bwd max_pool2d_bwd.cpp) add_executable(client_max_pool2d_bwd max_pool2d_bwd.cpp)
target_link_libraries(client_max_pool2d_bwd PRIVATE composable_kernel::device_operations) target_link_libraries(client_max_pool2d_bwd PRIVATE composable_kernel::device_other_operations)
add_executable(client_avg_pool3d_fwd avg_pool3d_fwd.cpp) add_executable(client_avg_pool3d_fwd avg_pool3d_fwd.cpp)
target_link_libraries(client_avg_pool3d_fwd PRIVATE composable_kernel::device_operations) target_link_libraries(client_avg_pool3d_fwd PRIVATE composable_kernel::device_other_operations)
add_executable(client_avg_pool3d_bwd avg_pool3d_bwd.cpp) add_executable(client_avg_pool3d_bwd avg_pool3d_bwd.cpp)
target_link_libraries(client_avg_pool3d_bwd PRIVATE composable_kernel::device_operations) target_link_libraries(client_avg_pool3d_bwd PRIVATE composable_kernel::device_other_operations)
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