Skip to content

GitLab

"docs/vscode:/vscode.git/clone" did not exist on "d5cb0be2cd16e6c5eefd4d266a38357fde83a660"
Unverified Commit c254e5ab authored by ltqin's avatar ltqin Committed by GitHub
Browse files

NHWC conv 2d: bwd fp32/fp16/bfp16/int8, Device level tuning and host API (#92) 



* start conv2d bwd api

* kernel running

* add bwd reference

* change to no shuffle

* fix bwd reference

* pass verification

* add Filter1x1Stride1Pad0 and start testing

* change some tuning parameter

* fix test error

* add fp16 tuning parameter

* add bf16 tuning parameter

* add int8 tuning parameters

* change fp32 tuning parameter

* add bwd to profiler

* fix bug for bwd profiler

* fix ckProfiler bug

* change conv2d_bwd_xdl to fp16

* fix bug in comments

* fix precompile id

* fix enum conv name

* chage _bwd_ to _bwd_data_

* change conv2d_bwd example id

* bwd to bwd data

* fix prehead

* fix MakeDefaultBlock2CTileMap ,import form merge develop

* format bwd instance

* bwd to bwd data

* change name bwd to bwd data

* change name bwd to bwd data in example

* formate code

* change conv2d bwd data id in example

* rewrite readme for example

* fix CalculateMagicNumbers about div zero

* add workaround CK_WORKAROUND_SWDEV_325164

* change test_conf2d_bwd_data show info

* format

* fix bug for workaround:CK_WORKAROUND_SWDEV_325164

* formate tuning parameters

* formate tuning parameters again

* formate tuning parameters 3

* formate tuning parameters 4

* remove add function template

* format

* update comment
Co-authored-by: default avatarltqin <letaoqin@amd.com>
Co-authored-by: default avatarChao Liu <chao.liu2@amd.com>
parent 992f71e3
Expand all Show whitespace changes
Inline Side-by-side
Showing with 2140 additions and 93 deletions
composable_kernel/include/config.hpp
View file @ c254e5ab
...@@ -151,6 +151,12 @@ ...@@ -151,6 +151,12 @@
#define CK_WORKAROUND_SWDEV_XXXXXX_THREAD_WISE_COPY_V1R5_TYPE_CONVERT_ISSUE 1 #define CK_WORKAROUND_SWDEV_XXXXXX_THREAD_WISE_COPY_V1R5_TYPE_CONVERT_ISSUE 1
#endif #endif
// workaround for verifaction failure, due to compiler regression, for conv bwd-data fp16 using some
// tuning parameter
#ifndef CK_WORKAROUND_SWDEV_325164
#define CK_WORKAROUND_SWDEV_325164 1
#endif
namespace ck { namespace ck {
enum InMemoryDataOperationEnum_t enum InMemoryDataOperationEnum_t
......
composable_kernel/include/tensor_operation/element_wise_operation.hpp
View file @ c254e5ab
#ifndef CK_ELEMENT_WISE_OPERATION_HPP #ifndef CK_ELEMENT_WISE_OPERATION_HPP
#define CK_ELEMENT_WISE_OPERATION_HPP #define CK_ELEMENT_WISE_OPERATION_HPP
#include "data_type.hpp"
#include "data_type.hpp" #include "data_type.hpp"
......
composable_kernel/include/utility/magic_division.hpp
View file @ c254e5ab
...@@ -25,7 +25,11 @@ struct MagicDivision ...@@ -25,7 +25,11 @@ struct MagicDivision
// uint32_t // uint32_t
__host__ __device__ static constexpr auto CalculateMagicNumbers(uint32_t divisor) __host__ __device__ static constexpr auto CalculateMagicNumbers(uint32_t divisor)
{ {
// assert(divisior >= 1 && divisior <= INT32_MAX); // WARNING: magic division is only applicable for division inside this range.
// You should use the return value of CalculateMagicNumbers, if division is not inside this
// range. The "else" logic below is to quiet down run-time error.
if(divisor >= 1 && divisor <= INT32_MAX)
{
uint32_t shift = 0; uint32_t shift = 0;
for(shift = 0; shift < 32; ++shift) for(shift = 0; shift < 32; ++shift)
{ {
...@@ -41,6 +45,11 @@ struct MagicDivision ...@@ -41,6 +45,11 @@ struct MagicDivision
return make_tuple(uint32_t(multiplier), shift); return make_tuple(uint32_t(multiplier), shift);
} }
else
{
return make_tuple(uint32_t(0), uint32_t(0));
}
}
__host__ __device__ static constexpr uint32_t CalculateMagicMultiplier(uint32_t divisor) __host__ __device__ static constexpr uint32_t CalculateMagicMultiplier(uint32_t divisor)
{ {
......
device_operation/CMakeLists.txt
View file @ c254e5ab
...@@ -101,8 +101,15 @@ set(DEVICE_CONV2D_FWD_BIAS_RELU_ATOMIC_ADD_INSTANCE_SOURCE ...@@ -101,8 +101,15 @@ set(DEVICE_CONV2D_FWD_BIAS_RELU_ATOMIC_ADD_INSTANCE_SOURCE
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_c_shuffle_bias_relu_atomic_add_nhwc_kyxc_nhwk_f16_instance.cpp; ${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_c_shuffle_bias_relu_atomic_add_nhwc_kyxc_nhwk_f16_instance.cpp;
) )
# device_conv2d_bwd_data_instance
set(DEVICE_CONV2D_BWD_DATA_INSTANCE_SOURCE
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp;
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp;
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp;
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp;
)
add_library(device_gemm_instance SHARED ${DEVICE_GEMM_INSTANCE_SOURCE}) add_library(device_gemm_instance SHARED ${DEVICE_GEMM_INSTANCE_SOURCE})
add_library(device_gemm_bias_2d_instance SHARED ${DEVICE_GEMM_BIAS_2D_INSTANCE_SOURCE})
add_library(device_gemm_bias_relu_instance SHARED ${DEVICE_GEMM_BIAS_RELU_INSTANCE_SOURCE}) add_library(device_gemm_bias_relu_instance SHARED ${DEVICE_GEMM_BIAS_RELU_INSTANCE_SOURCE})
add_library(device_gemm_bias_relu_add_instance SHARED ${DEVICE_GEMM_BIAS_RELU_ADD_INSTANCE_SOURCE}) add_library(device_gemm_bias_relu_add_instance SHARED ${DEVICE_GEMM_BIAS_RELU_ADD_INSTANCE_SOURCE})
add_library(device_batched_gemm_instance SHARED ${DEVICE_BATCHED_GEMM_INSTANCE_SOURCE}) add_library(device_batched_gemm_instance SHARED ${DEVICE_BATCHED_GEMM_INSTANCE_SOURCE})
...@@ -111,6 +118,8 @@ add_library(device_conv2d_fwd_instance SHARED ${DEVICE_CONV2D_FWD_INSTANCE_SOURC ...@@ -111,6 +118,8 @@ add_library(device_conv2d_fwd_instance SHARED ${DEVICE_CONV2D_FWD_INSTANCE_SOURC
add_library(device_conv2d_fwd_bias_relu_instance SHARED ${DEVICE_CONV2D_FWD_BIAS_RELU_INSTANCE_SOURCE}) add_library(device_conv2d_fwd_bias_relu_instance SHARED ${DEVICE_CONV2D_FWD_BIAS_RELU_INSTANCE_SOURCE})
add_library(device_conv2d_fwd_bias_relu_add_instance SHARED ${DEVICE_CONV2D_FWD_BIAS_RELU_ADD_INSTANCE_SOURCE}) add_library(device_conv2d_fwd_bias_relu_add_instance SHARED ${DEVICE_CONV2D_FWD_BIAS_RELU_ADD_INSTANCE_SOURCE})
add_library(device_conv2d_fwd_bias_relu_atomic_add_instance SHARED ${DEVICE_CONV2D_FWD_BIAS_RELU_ATOMIC_ADD_INSTANCE_SOURCE}) add_library(device_conv2d_fwd_bias_relu_atomic_add_instance SHARED ${DEVICE_CONV2D_FWD_BIAS_RELU_ATOMIC_ADD_INSTANCE_SOURCE})
add_library(device_gemm_bias_2d_instance SHARED ${DEVICE_GEMM_BIAS_2D_INSTANCE_SOURCE})
add_library(device_conv2d_bwd_data_instance SHARED ${DEVICE_CONV2D_BWD_DATA_INSTANCE_SOURCE})
target_include_directories(device_gemm_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>) target_include_directories(device_gemm_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
target_include_directories(device_gemm_bias_2d_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>) target_include_directories(device_gemm_bias_2d_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
...@@ -122,6 +131,7 @@ target_include_directories(device_conv2d_fwd_instance SYSTEM PUBLIC $<BUILD_INTE ...@@ -122,6 +131,7 @@ target_include_directories(device_conv2d_fwd_instance SYSTEM PUBLIC $<BUILD_INTE
target_include_directories(device_conv2d_fwd_bias_relu_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>) target_include_directories(device_conv2d_fwd_bias_relu_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
target_include_directories(device_conv2d_fwd_bias_relu_add_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>) target_include_directories(device_conv2d_fwd_bias_relu_add_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
target_include_directories(device_conv2d_fwd_bias_relu_atomic_add_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>) target_include_directories(device_conv2d_fwd_bias_relu_atomic_add_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
target_include_directories(device_conv2d_bwd_data_instance SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
target_compile_features(device_gemm_instance PUBLIC) target_compile_features(device_gemm_instance PUBLIC)
target_compile_features(device_gemm_bias_2d_instance PUBLIC) target_compile_features(device_gemm_bias_2d_instance PUBLIC)
...@@ -133,6 +143,7 @@ target_compile_features(device_conv2d_fwd_instance PUBLIC) ...@@ -133,6 +143,7 @@ target_compile_features(device_conv2d_fwd_instance PUBLIC)
target_compile_features(device_conv2d_fwd_bias_relu_instance PUBLIC) target_compile_features(device_conv2d_fwd_bias_relu_instance PUBLIC)
target_compile_features(device_conv2d_fwd_bias_relu_add_instance PUBLIC) target_compile_features(device_conv2d_fwd_bias_relu_add_instance PUBLIC)
target_compile_features(device_conv2d_fwd_bias_relu_atomic_add_instance PUBLIC) target_compile_features(device_conv2d_fwd_bias_relu_atomic_add_instance PUBLIC)
target_compile_features(device_conv2d_bwd_data_instance PUBLIC)
set_target_properties(device_gemm_instance PROPERTIES POSITION_INDEPENDENT_CODE ON) set_target_properties(device_gemm_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
set_target_properties(device_gemm_bias_2d_instance PROPERTIES POSITION_INDEPENDENT_CODE ON) set_target_properties(device_gemm_bias_2d_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
...@@ -144,6 +155,7 @@ set_target_properties(device_conv2d_fwd_instance PROPERTIES POSITION_INDEPENDENT ...@@ -144,6 +155,7 @@ set_target_properties(device_conv2d_fwd_instance PROPERTIES POSITION_INDEPENDENT
set_target_properties(device_conv2d_fwd_bias_relu_instance PROPERTIES POSITION_INDEPENDENT_CODE ON) set_target_properties(device_conv2d_fwd_bias_relu_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
set_target_properties(device_conv2d_fwd_bias_relu_add_instance PROPERTIES POSITION_INDEPENDENT_CODE ON) set_target_properties(device_conv2d_fwd_bias_relu_add_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
set_target_properties(device_conv2d_fwd_bias_relu_atomic_add_instance PROPERTIES POSITION_INDEPENDENT_CODE ON) set_target_properties(device_conv2d_fwd_bias_relu_atomic_add_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
set_target_properties(device_conv2d_bwd_data_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
install(TARGETS device_gemm_instance LIBRARY DESTINATION lib) install(TARGETS device_gemm_instance LIBRARY DESTINATION lib)
install(TARGETS device_gemm_bias_2d_instance LIBRARY DESTINATION lib) install(TARGETS device_gemm_bias_2d_instance LIBRARY DESTINATION lib)
...@@ -155,3 +167,4 @@ install(TARGETS device_conv2d_fwd_instance LIBRARY DESTINATION lib) ...@@ -155,3 +167,4 @@ install(TARGETS device_conv2d_fwd_instance LIBRARY DESTINATION lib)
install(TARGETS device_conv2d_fwd_bias_relu_instance LIBRARY DESTINATION lib) install(TARGETS device_conv2d_fwd_bias_relu_instance LIBRARY DESTINATION lib)
install(TARGETS device_conv2d_fwd_bias_relu_add_instance LIBRARY DESTINATION lib) install(TARGETS device_conv2d_fwd_bias_relu_add_instance LIBRARY DESTINATION lib)
install(TARGETS device_conv2d_fwd_bias_relu_atomic_add_instance LIBRARY DESTINATION lib) install(TARGETS device_conv2d_fwd_bias_relu_atomic_add_instance LIBRARY DESTINATION lib)
install(TARGETS device_conv2d_bwd_data_instance LIBRARY DESTINATION lib)
device_operation/include/convolution_backward_data_specialization.hpp 0 → 100644
View file @ c254e5ab
#ifndef CONVOLUTION_BACKWARD_DATA_SPECIALIZATION
#define CONVOLUTION_BACKWARD_DATA_SPECIALIZATION
namespace ck {
namespace tensor_operation {
namespace device {
enum ConvolutionBackwardDataSpecialization_t
{
Default,
Filter1x1Stride1Pad0,
};
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
device_operation/include/device_conv_bwd_data.hpp 0 → 100644
View file @ c254e5ab
#ifndef DEVICE_CONV_BWD_DATA_HPP
#define DEVICE_CONV_BWD_DATA_HPP
#include <iostream>
#include "device_base.hpp"
#include "element_wise_operation.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
template <typename InElementwiseOperation,
typename WeiElementwiseOperation,
typename OutElementwiseOperation>
struct DeviceConvBwdData : public BaseOperator
{
virtual std::unique_ptr<BaseArgument>
MakeArgumentPointer(void* p_in,
const void* p_wei,
const void* p_out,
ck::index_t N,
ck::index_t K,
ck::index_t C,
std::vector<ck::index_t> input_spatial_lengths,
std::vector<ck::index_t> filter_spatial_lengths,
std::vector<ck::index_t> output_spatial_lengths,
std::vector<ck::index_t> conv_filter_strides,
std::vector<ck::index_t> conv_filter_dilations,
std::vector<ck::index_t> input_left_pads,
std::vector<ck::index_t> input_right_pads,
InElementwiseOperation in_element_op,
WeiElementwiseOperation wei_element_op,
OutElementwiseOperation out_element_op) = 0;
virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
};
template <typename InElementwiseOperation,
typename WeiElementwiseOperation,
typename OutElementwiseOperation>
using DeviceConvBwdDataPtr = std::unique_ptr<
DeviceConvBwdData<InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation>>;
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
device_operation/src/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp 0 → 100644
View file @ c254e5ab
#include <stdlib.h>
#include "config.hpp"
#include "device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_instance {
using BF16 = ushort;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances = std::tuple<
// clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//####################################################################| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//####################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>
// clang-format on
>;
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances =
std::tuple<
// clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//####################################################################| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//####################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>
// clang-format on
>;
void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances{});
}
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
device_operation/src/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp 0 → 100644
View file @ c254e5ab
#include <stdlib.h>
#include "config.hpp"
#include "device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_instance {
using F16 = ck::half_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances = std::tuple<
// clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//####################################################################| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//####################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
#if !CK_WORKAROUND_SWDEV_325164
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
#endif
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>
// clang-format on
>;
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances =
std::tuple<
// clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//####################################################################| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//####################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1>
// clang-format on
>;
void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances{});
}
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
device_operation/src/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp 0 → 100644
View file @ c254e5ab
#include <stdlib.h>
#include "config.hpp"
#include "device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_instance {
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances = std::tuple<
// clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//####################################################################| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//####################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>
// clang-format on
>;
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances =
std::tuple<
// clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//####################################################################| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//####################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 4, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 4, true, 7, 1>
// clang-format on
>;
void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances{});
}
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
device_operation/src/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp 0 → 100644
View file @ c254e5ab
#include <stdlib.h>
#include "config.hpp"
#include "device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_instance {
using DataType = int8_t;
using AccType = int32_t;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances = std::tuple<
// clang-format off
//####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//####################################################################| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//####################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataDefault, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>
// clang-format on
>;
using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances =
std::tuple<
// clang-format off
//#####################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//#####################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#####################################################################| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#####################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 1, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 2, 16, true, 7, 1>,
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DataType, DataType, DataType, AccType, PassThrough, PassThrough, PassThrough, ConvBwdDataFilter1x1Stride1Pad0, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<2, 0, 1>, S<0, 2, 1>, 1, 4, 16, true, 7, 1>
// clang-format on
>;
void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
std::vector<DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances{});
}
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
example/12_conv2d_bwd_data_xdl/README.md 0 → 100644
View file @ c254e5ab
# Instructions for ```conv2d_bwd_data_xdl``` Example
## Docker script
```bash
docker run \
-it \
--rm \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build ```conv2d_bwd_data_xdl```
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
```bash
make -j conv2d_bwd_data_xdl
```
## Run ```conv2d_bwd_data_xdl```
```bash
#arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1)
#arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx
./bin/conv2d_bwd_data_xdl 0 1 5
```
Result
```
in_n_c_hi_wi: dim 4, lengths {128, 256, 71, 71}, strides {1290496, 1, 18176, 256}
wei_k_c_y_x: dim 4, lengths {256, 256, 3, 3}, strides {2304, 1, 768, 256}
out_n_k_ho_wo: dim 4, lengths {128, 256, 36, 36}, strides {331776, 1, 9216, 256}
arg.a_grid_desc_k0_m_k1_container_{128, 175232, 8}
arg.b_grid_desc_k0_n_k1_container_{128, 256, 8}
arg.c_grid_desc_m_n_container_{ 175232, 256}
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( 2738, 4, 2, 2, 4, 2 )
launch_and_time_kernel: grid_dim {2738, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 1 times...
arg.a_grid_desc_k0_m_k1_container_{64, 175232, 8}
arg.b_grid_desc_k0_n_k1_container_{64, 256, 8}
arg.c_grid_desc_m_n_container_{ 175232, 256}
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( 2738, 4, 2, 2, 4, 2 )
launch_and_time_kernel: grid_dim {2738, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 1 times...
arg.a_grid_desc_k0_m_k1_container_{64, 175232, 8}
arg.b_grid_desc_k0_n_k1_container_{64, 256, 8}
arg.c_grid_desc_m_n_container_{ 175232, 256}
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( 2738, 4, 2, 2, 4, 2 )
launch_and_time_kernel: grid_dim {2738, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 1 times...
arg.a_grid_desc_k0_m_k1_container_{32, 175232, 8}
arg.b_grid_desc_k0_n_k1_container_{32, 256, 8}
arg.c_grid_desc_m_n_container_{ 175232, 256}
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( 2738, 4, 2, 2, 4, 2 )
launch_and_time_kernel: grid_dim {2738, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 1 times...
Perf: 2.45966 ms, 79.5597 TFlops, 169.325 GB/s
```
example/12_conv2d_bwd_data_xdl/conv2d_bwd_data_xdl.cpp 0 → 100644
View file @ c254e5ab
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include <stdlib.h>
#include <half.hpp>
#include "config.hpp"
#include "print.hpp"
#include "device.hpp"
#include "host_tensor.hpp"
#include "host_tensor_generator.hpp"
#include "device_tensor.hpp"
#include "tensor_layout.hpp"
#include "element_wise_operation.hpp"
#include "device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp"
#include "reference_conv_bwd_data.hpp"
using InDataType = ck::half_t;
using WeiDataType = ck::half_t;
using OutDataType = ck::half_t;
using AccDataType = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
using DeviceConvBwdDataInstance = ck::tensor_operation::device::
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
InDataType, // InDataType
WeiDataType, // WeiDataType
OutDataType, // OutDataType
AccDataType, // AccDataType
InElementOp, // InElementwiseOperation
WeiElementOp, // WeiElementwiseOperation
OutElementOp, // OutElementwiseOperation
ConvBwdDefault, // ConvolutionBackwardDataSpecialization_t
256, // BlockSize
128, // MPerBlock
128, // NPerBlock
4, // K0PerBlock
8, // K1
32, // MPerXdl
32, // NPerXdl
2, // MXdlPerWave
2, // NXdlPerWave
S<4, 64, 1>, // ABlockTransferThreadClusterLengths_K0_M_K1
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // ABlockTransferSrcAccessOrder
2, // ABlockTransferSrcVectorDim
8, // ABlockTransferSrcScalarPerVector
8, // ABlockTransferDstScalarPerVector_K1
true, // ABlockLdsAddExtraM
S<4, 64, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1
S<2, 0, 1>, // BBlockTransferThreadClusterArrangeOrder
S<0, 2, 1>, // BBlockTransferSrcAccessOrder
1, // BBlockTransferSrcVectorDim
2, // BBlockTransferSrcScalarPerVector
8, // BBlockTransferDstScalarPerVector_K1
true, // BBlockLdsAddExtraN
7,
1>; // GemmCThreadTransferDstScalarPerVector
using ReferenceConvBwdInstance = ck::tensor_operation::host::ReferenceConvBwdData<InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp>;
int main(int argc, char* argv[])
{
bool do_verification = 0;
int init_method = 0;
int nrepeat = 5;
// Conv shape
ck::index_t N = 128;
ck::index_t K = 256;
ck::index_t C = 256;
ck::index_t Y = 3;
ck::index_t X = 3;
ck::index_t Hi = 71;
ck::index_t Wi = 71;
ck::index_t conv_stride_h = 2;
ck::index_t conv_stride_w = 2;
ck::index_t conv_dilation_h = 1;
ck::index_t conv_dilation_w = 1;
ck::index_t in_left_pad_h = 1;
ck::index_t in_left_pad_w = 1;
ck::index_t in_right_pad_h = 1;
ck::index_t in_right_pad_w = 1;
if(argc == 4)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
}
else if(argc == 19)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
N = std::stoi(argv[4]);
K = std::stoi(argv[5]);
C = std::stoi(argv[6]);
Y = std::stoi(argv[7]);
X = std::stoi(argv[8]);
Hi = std::stoi(argv[9]);
Wi = std::stoi(argv[10]);
conv_stride_h = std::stoi(argv[11]);
conv_stride_w = std::stoi(argv[12]);
conv_dilation_h = std::stoi(argv[13]);
conv_dilation_w = std::stoi(argv[14]);
in_left_pad_h = std::stoi(argv[15]);
in_left_pad_w = std::stoi(argv[16]);
in_right_pad_h = std::stoi(argv[17]);
in_right_pad_w = std::stoi(argv[18]);
}
else
{
printf("arg1: verification (0=no, 1=yes)\n");
printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
printf("arg3: run kernel # of times (>1)\n");
printf("arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
"RightPx\n");
exit(0);
}
const ck::index_t YEff = (Y - 1) * conv_dilation_h + 1;
const ck::index_t XEff = (X - 1) * conv_dilation_w + 1;
const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - YEff) / conv_stride_h + 1;
const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1;
const std::vector<ck::index_t> conv_filter_strides{{conv_stride_h, conv_stride_w}};
const std::vector<ck::index_t> conv_filter_dilations{{conv_dilation_h, conv_dilation_w}};
const std::vector<ck::index_t> input_left_pads{{in_left_pad_h, in_left_pad_w}};
const std::vector<ck::index_t> input_right_pads{{in_right_pad_h, in_right_pad_w}};
// tensor layout
auto f_host_tensor_descriptor =
[](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W) {
return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
std::vector<std::size_t>({C_ * H * W, 1, W * C_, C_}));
};
Tensor<OutDataType> out_n_k_ho_wo(f_host_tensor_descriptor(N, K, Ho, Wo));
Tensor<WeiDataType> wei_k_c_y_x(f_host_tensor_descriptor(K, C, Y, X));
Tensor<InDataType> in_n_c_hi_wi_host_result(f_host_tensor_descriptor(N, C, Hi, Wi));
Tensor<InDataType> in_n_c_hi_wi_device_result(f_host_tensor_descriptor(N, C, Hi, Wi));
std::cout << "in_n_c_hi_wi: " << in_n_c_hi_wi_host_result.mDesc << std::endl;
std::cout << "wei_k_c_y_x: " << wei_k_c_y_x.mDesc << std::endl;
std::cout << "out_n_k_ho_wo: " << out_n_k_ho_wo.mDesc << std::endl;
switch(init_method)
{
case 0: break;
case 1:
out_n_k_ho_wo.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
wei_k_c_y_x.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
break;
default:
out_n_k_ho_wo.GenerateTensorValue(GeneratorTensor_1<OutDataType>{1});
wei_k_c_y_x.GenerateTensorValue(GeneratorTensor_1<WeiDataType>{1});
}
DeviceMem in_device_buf(sizeof(InDataType) *
in_n_c_hi_wi_device_result.mDesc.GetElementSpace());
DeviceMem wei_device_buf(sizeof(WeiDataType) * wei_k_c_y_x.mDesc.GetElementSpace());
DeviceMem out_device_buf(sizeof(OutDataType) * out_n_k_ho_wo.mDesc.GetElementSpace());
out_device_buf.ToDevice(out_n_k_ho_wo.mData.data());
wei_device_buf.ToDevice(wei_k_c_y_x.mData.data());
// do GEMM
auto conv = DeviceConvBwdDataInstance{};
auto invoker = conv.MakeInvoker();
auto argument = conv.MakeArgument(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
N,
K,
C,
std::vector<ck::index_t>{{Hi, Wi}},
std::vector<ck::index_t>{{Y, X}},
std::vector<ck::index_t>{{Ho, Wo}},
conv_filter_strides,
conv_filter_dilations,
input_left_pads,
input_right_pads,
InElementOp{},
WeiElementOp{},
OutElementOp{});
if(!conv.IsSupportedArgument(argument))
{
throw std::runtime_error(
"wrong! device_conv with the specified compilation parameters does "
"not support this Conv problem");
}
float ave_time = invoker.Run(argument, nrepeat);
std::size_t flop = std::size_t(2) * N * K * Ho * Wo * C * Y * X;
std::size_t num_btype = sizeof(InDataType) * (N * C * Hi * Wi) +
sizeof(WeiDataType) * (K * C * Y * X) +
sizeof(OutDataType) * (N * K * Ho * Wo);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
float gb_per_sec = num_btype / 1.E6 / ave_time;
std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
<< std::endl;
if(do_verification)
{
auto ref_conv = ReferenceConvBwdInstance{};
auto ref_invoker = ref_conv.MakeInvoker();
auto ref_argument = ref_conv.MakeArgument(in_n_c_hi_wi_host_result,
wei_k_c_y_x,
out_n_k_ho_wo,
conv_filter_strides,
conv_filter_dilations,
input_left_pads,
input_right_pads,
InElementOp{},
WeiElementOp{},
OutElementOp{});
ref_invoker.Run(ref_argument);
in_device_buf.FromDevice(in_n_c_hi_wi_device_result.mData.data());
check_error(in_n_c_hi_wi_host_result, in_n_c_hi_wi_device_result);
}
}
example/CMakeLists.txt
View file @ c254e5ab
...@@ -24,6 +24,7 @@ set(GEMM_XDL_ALPHA_BETA_SOURCE 8_gemm_xdl_alpha_beta/gemm_xdl_alpha_beta.cpp) ...@@ -24,6 +24,7 @@ set(GEMM_XDL_ALPHA_BETA_SOURCE 8_gemm_xdl_alpha_beta/gemm_xdl_alpha_beta.cpp)
set(CONV2D_FWD_XDL_INT8_SOURCE 9_conv2d_fwd_xdl_int8/conv2d_fwd_xdl_int8.cpp) set(CONV2D_FWD_XDL_INT8_SOURCE 9_conv2d_fwd_xdl_int8/conv2d_fwd_xdl_int8.cpp)
set(CONV3D_FWD_XDL_SOURCE 10_conv3d_fwd_xdl/conv3d_fwd_xdl.cpp) set(CONV3D_FWD_XDL_SOURCE 10_conv3d_fwd_xdl/conv3d_fwd_xdl.cpp)
set(CONVND_FWD_XDL_SOURCE 11_convnd_fwd_xdl/convnd_fwd_xdl.cpp) set(CONVND_FWD_XDL_SOURCE 11_convnd_fwd_xdl/convnd_fwd_xdl.cpp)
set(CONV2D_BWD_DATA_XDL_SOURCE 12_conv2d_bwd_data_xdl/conv2d_bwd_data_xdl.cpp)
add_executable(gemm_xdl ${GEMM_XDL_SOURCE}) add_executable(gemm_xdl ${GEMM_XDL_SOURCE})
add_executable(gemm_xdl_bias_relu ${GEMM_XDL_BIAS_RELU_SOURCE}) add_executable(gemm_xdl_bias_relu ${GEMM_XDL_BIAS_RELU_SOURCE})
...@@ -36,6 +37,7 @@ add_executable(gemm_xdl_alpha_beta ${GEMM_XDL_ALPHA_BETA_SOURCE}) ...@@ -36,6 +37,7 @@ add_executable(gemm_xdl_alpha_beta ${GEMM_XDL_ALPHA_BETA_SOURCE})
add_executable(conv2d_fwd_xdl_int8 ${CONV2D_FWD_XDL_INT8_SOURCE}) add_executable(conv2d_fwd_xdl_int8 ${CONV2D_FWD_XDL_INT8_SOURCE})
add_executable(conv3d_fwd_xdl ${CONV3D_FWD_XDL_SOURCE}) add_executable(conv3d_fwd_xdl ${CONV3D_FWD_XDL_SOURCE})
add_executable(convnd_fwd_xdl ${CONVND_FWD_XDL_SOURCE}) add_executable(convnd_fwd_xdl ${CONVND_FWD_XDL_SOURCE})
add_executable(conv2d_bwd_data_xdl ${CONV2D_BWD_DATA_XDL_SOURCE})
target_link_libraries(gemm_xdl PRIVATE host_tensor) target_link_libraries(gemm_xdl PRIVATE host_tensor)
target_link_libraries(gemm_xdl_bias_relu PRIVATE host_tensor) target_link_libraries(gemm_xdl_bias_relu PRIVATE host_tensor)
...@@ -48,3 +50,5 @@ target_link_libraries(gemm_xdl_alpha_beta PRIVATE host_tensor) ...@@ -48,3 +50,5 @@ target_link_libraries(gemm_xdl_alpha_beta PRIVATE host_tensor)
target_link_libraries(conv2d_fwd_xdl_int8 PRIVATE host_tensor) target_link_libraries(conv2d_fwd_xdl_int8 PRIVATE host_tensor)
target_link_libraries(conv3d_fwd_xdl PRIVATE host_tensor) target_link_libraries(conv3d_fwd_xdl PRIVATE host_tensor)
target_link_libraries(convnd_fwd_xdl PRIVATE host_tensor) target_link_libraries(convnd_fwd_xdl PRIVATE host_tensor)
target_link_libraries(conv2d_bwd_data_xdl PRIVATE host_tensor)
profiler/CMakeLists.txt
View file @ c254e5ab
...@@ -25,6 +25,7 @@ set(PROFILER_SOURCE ...@@ -25,6 +25,7 @@ set(PROFILER_SOURCE
src/profile_conv_fwd_bias_relu_add.cpp src/profile_conv_fwd_bias_relu_add.cpp
src/profile_conv_fwd_bias_relu_atomic_add.cpp src/profile_conv_fwd_bias_relu_atomic_add.cpp
src/profile_batched_gemm.cpp src/profile_batched_gemm.cpp
src/profile_conv_bwd_data.cpp
) )
add_executable(ckProfiler ${PROFILER_SOURCE}) add_executable(ckProfiler ${PROFILER_SOURCE})
...@@ -39,3 +40,4 @@ target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_instance) ...@@ -39,3 +40,4 @@ target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_instance)
target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_add_instance) target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_add_instance)
target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_atomic_add_instance) target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_atomic_add_instance)
target_link_libraries(ckProfiler PRIVATE device_batched_gemm_instance) target_link_libraries(ckProfiler PRIVATE device_batched_gemm_instance)
target_link_libraries(ckProfiler PRIVATE device_conv2d_bwd_data_instance)
profiler/include/profile_conv_bwd_data_impl.hpp 0 → 100644
View file @ c254e5ab
#pragma once
#include "config.hpp"
#include "device.hpp"
#include "host_tensor.hpp"
#include "host_tensor_generator.hpp"
#include "tensor_layout.hpp"
#include "device_tensor.hpp"
#include "device_conv_bwd_data.hpp"
#include "element_wise_operation.hpp"
#include "reference_conv_bwd_data.hpp"
using F16 = ck::half_t;
using F32 = float;
using BF16 = ushort;
using INT8 = int8_t;
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_conv2d_bwd_data_instance {
using DeviceConvBwdDataNoOpPtr =
DeviceConvBwdDataPtr<ck::tensor_operation::element_wise::PassThrough,
ck::tensor_operation::element_wise::PassThrough,
ck::tensor_operation::element_wise::PassThrough>;
void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
std::vector<DeviceConvBwdDataNoOpPtr>&);
void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
std::vector<DeviceConvBwdDataNoOpPtr>&);
void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
std::vector<DeviceConvBwdDataNoOpPtr>&);
void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
std::vector<DeviceConvBwdDataNoOpPtr>&);
} // namespace device_conv2d_bwd_data_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
namespace ck {
namespace profiler {
template <int NDimSpatial,
typename InDataType,
typename WeiDataType,
typename OutDataType,
typename InLayout,
typename WeiLayout,
typename OutLayout>
void profile_conv_bwd_data_impl(int do_verification,
int init_method,
bool do_log,
int nrepeat,
ck::index_t N,
ck::index_t K,
ck::index_t C,
std::vector<ck::index_t> input_spatial_lengths,
std::vector<ck::index_t> filter_spatial_lengths,
std::vector<ck::index_t> output_spatial_lengths,
std::vector<ck::index_t> conv_filter_strides,
std::vector<ck::index_t> conv_filter_dilations,
std::vector<ck::index_t> input_left_pads,
std::vector<ck::index_t> input_right_pads)
{
const ck::index_t Y = filter_spatial_lengths[0];
const ck::index_t X = filter_spatial_lengths[1];
const ck::index_t Hi = input_spatial_lengths[0];
const ck::index_t Wi = input_spatial_lengths[1];
const ck::index_t Ho = output_spatial_lengths[0];
const ck::index_t Wo = output_spatial_lengths[1];
auto f_host_tensor_descriptor =
[](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W, auto layout) {
if constexpr(is_same<decltype(layout), ck::tensor_layout::convolution::NCHW>::value ||
is_same<decltype(layout), ck::tensor_layout::convolution::KCYX>::value ||
is_same<decltype(layout), ck::tensor_layout::convolution::NKHW>::value)
{
return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
std::vector<std::size_t>({C_ * H * W, H * W, W, 1}));
}
else if constexpr(is_same<decltype(layout), tensor_layout::convolution::NHWC>::value ||
is_same<decltype(layout), tensor_layout::convolution::KYXC>::value ||
is_same<decltype(layout), tensor_layout::convolution::NHWK>::value)
{
return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
std::vector<std::size_t>({C_ * H * W, 1, W * C_, C_}));
}
};
Tensor<InDataType> in_n_c_hi_wi_host_result(f_host_tensor_descriptor(N, C, Hi, Wi, InLayout{}));
Tensor<InDataType> in_n_c_hi_wi_device_result(
f_host_tensor_descriptor(N, C, Hi, Wi, InLayout{}));
Tensor<WeiDataType> wei_k_c_y_x(f_host_tensor_descriptor(K, C, Y, X, WeiLayout{}));
Tensor<OutDataType> out_n_k_ho_wo(f_host_tensor_descriptor(N, K, Ho, Wo, OutLayout{}));
std::cout << "in_n_c_hi_wi: " << in_n_c_hi_wi_host_result.mDesc << std::endl;
std::cout << "wei_k_c_y_x: " << wei_k_c_y_x.mDesc << std::endl;
std::cout << "out_n_k_ho_wo: " << out_n_k_ho_wo.mDesc << std::endl;
switch(init_method)
{
case 0: break;
case 1:
out_n_k_ho_wo.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5});
wei_k_c_y_x.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
break;
default:
out_n_k_ho_wo.GenerateTensorValue(GeneratorTensor_3<InDataType>{0.0, 1.0});
wei_k_c_y_x.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.5, 0.5});
}
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
const auto in_element_op = InElementOp{};
const auto wei_element_op = WeiElementOp{};
const auto out_element_op = OutElementOp{};
if(do_verification)
{
using ReferenceConvBwdDataInstance =
ck::tensor_operation::host::ReferenceConvBwdData<InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp>;
auto ref_conv = ReferenceConvBwdDataInstance{};
auto ref_invoker = ref_conv.MakeInvoker();
auto ref_argument = ref_conv.MakeArgument(in_n_c_hi_wi_host_result,
wei_k_c_y_x,
out_n_k_ho_wo,
conv_filter_strides,
conv_filter_dilations,
input_left_pads,
input_right_pads,
in_element_op,
wei_element_op,
out_element_op);
ref_invoker.Run(ref_argument);
}
DeviceMem in_device_buf(sizeof(InDataType) *
in_n_c_hi_wi_device_result.mDesc.GetElementSpace());
DeviceMem wei_device_buf(sizeof(WeiDataType) * wei_k_c_y_x.mDesc.GetElementSpace());
DeviceMem out_device_buf(sizeof(OutDataType) * out_n_k_ho_wo.mDesc.GetElementSpace());
out_device_buf.ToDevice(out_n_k_ho_wo.mData.data());
wei_device_buf.ToDevice(wei_k_c_y_x.mData.data());
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using DeviceConvBwdDataNoOpPtr =
ck::tensor_operation::device::DeviceConvBwdDataPtr<PassThrough, PassThrough, PassThrough>;
// add device Conv instances
std::vector<DeviceConvBwdDataNoOpPtr> conv_ptrs;
if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, float> &&
ck::is_same_v<ck::remove_cv_t<WeiDataType>, float> &&
ck::is_same_v<ck::remove_cv_t<OutDataType>, float>)
{
ck::tensor_operation::device::device_conv2d_bwd_data_instance::
add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(conv_ptrs);
}
else if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, ck::half_t> &&
ck::is_same_v<ck::remove_cv_t<WeiDataType>, ck::half_t> &&
ck::is_same_v<ck::remove_cv_t<OutDataType>, ck::half_t>)
{
ck::tensor_operation::device::device_conv2d_bwd_data_instance::
add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
}
else if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, ushort> &&
ck::is_same_v<ck::remove_cv_t<WeiDataType>, ushort> &&
ck::is_same_v<ck::remove_cv_t<OutDataType>, ushort>)
{
ck::tensor_operation::device::device_conv2d_bwd_data_instance::
add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(conv_ptrs);
}
else if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, int8_t> &&
ck::is_same_v<ck::remove_cv_t<WeiDataType>, int8_t> &&
ck::is_same_v<ck::remove_cv_t<OutDataType>, int8_t>)
{
ck::tensor_operation::device::device_conv2d_bwd_data_instance::
add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(conv_ptrs);
}
if(conv_ptrs.size() <= 0)
{
throw std::runtime_error("wrong! no device Conv instance found");
}
std::string best_conv_name;
float best_ave_time = 0;
float best_tflops = 0;
float best_gb_per_sec = 0;
// profile device Conv instances
for(auto& conv_ptr : conv_ptrs)
{
auto argument_ptr = conv_ptr->MakeArgumentPointer(
static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
N,
K,
C,
input_spatial_lengths,
filter_spatial_lengths,
output_spatial_lengths,
conv_filter_strides,
conv_filter_dilations,
input_left_pads,
input_right_pads,
in_element_op,
wei_element_op,
out_element_op);
auto invoker_ptr = conv_ptr->MakeInvokerPointer();
if(conv_ptr->IsSupportedArgument(argument_ptr.get()))
{
std::string conv_name = conv_ptr->GetTypeString();
float ave_time = invoker_ptr->Run(argument_ptr.get(), nrepeat);
std::size_t flop = std::size_t(2) * N * K * Ho * Wo * C * Y * X;
std::size_t num_btype = sizeof(InDataType) * (N * C * Hi * Wi) +
sizeof(WeiDataType) * (K * C * Y * X) +
sizeof(OutDataType) * (N * K * Ho * Wo);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
float gb_per_sec = num_btype / 1.E6 / ave_time;
std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
<< " GB/s, " << conv_name << std::endl;
if(tflops > best_tflops)
{
best_conv_name = conv_name;
best_tflops = tflops;
best_ave_time = ave_time;
best_gb_per_sec = gb_per_sec;
}
if(do_verification)
{
in_device_buf.FromDevice(in_n_c_hi_wi_device_result.mData.data());
check_error(in_n_c_hi_wi_host_result, in_n_c_hi_wi_device_result);
if(do_log)
{
LogRangeAsType<float>(std::cout << "in : ", out_n_k_ho_wo.mData, ",")
<< std::endl;
LogRangeAsType<float>(std::cout << "wei: ", wei_k_c_y_x.mData, ",")
<< std::endl;
LogRangeAsType<float>(
std::cout << "out_host : ", in_n_c_hi_wi_host_result.mData, ",")
<< std::endl;
LogRangeAsType<float>(
std::cout << "out_device: ", in_n_c_hi_wi_device_result.mData, ",")
<< std::endl;
}
}
}
}
std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
<< best_gb_per_sec << " GB/s, " << best_conv_name << std::endl;
}
} // namespace profiler
} // namespace ck
profiler/src/profile_conv_bwd_data.cpp 0 → 100644
View file @ c254e5ab
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include <stdlib.h>
#include <half.hpp>
#include "profile_conv_bwd_data_impl.hpp"
enum ConvDataType
{
F32_F32_F32, // 0
F16_F16_F16, // 1
BF16_BF16_BF16, // 2
INT8_INT8_INT8, // 3
};
enum ConvInputLayout
{
NCHW, // 0
NHWC, // 1
};
enum ConvWeightLayout
{
KCYX, // 0
KYXC, // 1
};
enum ConvOutputLayout
{
NKHW, // 0
NHWK, // 1
};
int profile_conv_bwd_data(int argc, char* argv[])
{
if(argc != 25)
{
printf("arg1: tensor operation (conv_bwd: BackwardConvolution)\n");
printf("arg2: data type (0: fp32; 1: fp16)\n");
printf("arg3: input tensor layout (0: NCHW; 1: NHWC)\n");
printf("arg4: weight tensor layout (0: KCYX; 1: KYXC)\n");
printf("arg5: output tensor layout (0: NKHW; 1: NHWK)\n");
printf("arg6: verification (0: no; 1: yes)\n");
printf("arg7: initialization (0: no init; 1: integer value; 2: decimal value)\n");
printf("arg8: print tensor value (0: no; 1: yes)\n");
printf("arg9: run kernel # of times (>1)\n");
printf("arg10 to 24: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
"RightPx\n");
exit(1);
}
const int data_type = static_cast<ConvDataType>(std::stoi(argv[2]));
const int in_layout = static_cast<ConvInputLayout>(std::stoi(argv[3]));
const int wei_layout = static_cast<ConvWeightLayout>(std::stoi(argv[4]));
const int out_layout = static_cast<ConvOutputLayout>(std::stoi(argv[5]));
const bool do_verification = std::stoi(argv[6]);
const int init_method = std::stoi(argv[7]);
const bool do_log = std::stoi(argv[8]);
const int nrepeat = std::stoi(argv[9]);
const ck::index_t N = std::stoi(argv[10]);
const ck::index_t K = std::stoi(argv[11]);
const ck::index_t C = std::stoi(argv[12]);
const ck::index_t Y = std::stoi(argv[13]);
const ck::index_t X = std::stoi(argv[14]);
const ck::index_t Hi = std::stoi(argv[15]);
const ck::index_t Wi = std::stoi(argv[16]);
const ck::index_t conv_stride_h = std::stoi(argv[17]);
const ck::index_t conv_stride_w = std::stoi(argv[18]);
const ck::index_t conv_dilation_h = std::stoi(argv[19]);
const ck::index_t conv_dilation_w = std::stoi(argv[20]);
const ck::index_t in_left_pad_h = std::stoi(argv[21]);
const ck::index_t in_left_pad_w = std::stoi(argv[22]);
const ck::index_t in_right_pad_h = std::stoi(argv[23]);
const ck::index_t in_right_pad_w = std::stoi(argv[24]);
const ck::index_t YEff = (Y - 1) * conv_dilation_h + 1;
const ck::index_t XEff = (X - 1) * conv_dilation_w + 1;
const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - YEff) / conv_stride_h + 1;
const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1;
if(data_type == ConvDataType::F32_F32_F32 && in_layout == ConvInputLayout::NHWC &&
wei_layout == ConvWeightLayout::KYXC && out_layout == ConvOutputLayout::NHWK)
{
ck::profiler::profile_conv_bwd_data_impl<2,
float,
float,
float,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK>(
do_verification,
init_method,
do_log,
nrepeat,
N,
K,
C,
std::vector<ck::index_t>{Hi, Wi},
std::vector<ck::index_t>{Y, X},
std::vector<ck::index_t>{Ho, Wo},
std::vector<ck::index_t>{conv_stride_h, conv_stride_w},
std::vector<ck::index_t>{conv_dilation_h, conv_dilation_w},
std::vector<ck::index_t>{in_left_pad_h, in_left_pad_w},
std::vector<ck::index_t>{in_right_pad_h, in_right_pad_w});
}
else if(data_type == ConvDataType::F16_F16_F16 && in_layout == ConvInputLayout::NHWC &&
wei_layout == ConvWeightLayout::KYXC && out_layout == ConvOutputLayout::NHWK)
{
ck::profiler::profile_conv_bwd_data_impl<2,
ck::half_t,
ck::half_t,
ck::half_t,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK>(
do_verification,
init_method,
do_log,
nrepeat,
N,
K,
C,
std::vector<ck::index_t>{Hi, Wi},
std::vector<ck::index_t>{Y, X},
std::vector<ck::index_t>{Ho, Wo},
std::vector<ck::index_t>{conv_stride_h, conv_stride_w},
std::vector<ck::index_t>{conv_dilation_h, conv_dilation_w},
std::vector<ck::index_t>{in_left_pad_h, in_left_pad_w},
std::vector<ck::index_t>{in_right_pad_h, in_right_pad_w});
}
else if(data_type == ConvDataType::BF16_BF16_BF16 && in_layout == ConvInputLayout::NHWC &&
wei_layout == ConvWeightLayout::KYXC && out_layout == ConvOutputLayout::NHWK)
{
ck::profiler::profile_conv_bwd_data_impl<2,
uint16_t,
uint16_t,
uint16_t,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK>(
do_verification,
init_method,
do_log,
nrepeat,
N,
K,
C,
std::vector<ck::index_t>{Hi, Wi},
std::vector<ck::index_t>{Y, X},
std::vector<ck::index_t>{Ho, Wo},
std::vector<ck::index_t>{conv_stride_h, conv_stride_w},
std::vector<ck::index_t>{conv_dilation_h, conv_dilation_w},
std::vector<ck::index_t>{in_left_pad_h, in_left_pad_w},
std::vector<ck::index_t>{in_right_pad_h, in_right_pad_w});
}
else if(data_type == ConvDataType::INT8_INT8_INT8 && in_layout == ConvInputLayout::NHWC &&
wei_layout == ConvWeightLayout::KYXC && out_layout == ConvOutputLayout::NHWK)
{
ck::profiler::profile_conv_bwd_data_impl<2,
int8_t,
int8_t,
int8_t,
ck::tensor_layout::convolution::NHWC,
ck::tensor_layout::convolution::KYXC,
ck::tensor_layout::convolution::NHWK>(
do_verification,
init_method,
do_log,
nrepeat,
N,
K,
C,
std::vector<ck::index_t>{Hi, Wi},
std::vector<ck::index_t>{Y, X},
std::vector<ck::index_t>{Ho, Wo},
std::vector<ck::index_t>{conv_stride_h, conv_stride_w},
std::vector<ck::index_t>{conv_dilation_h, conv_dilation_w},
std::vector<ck::index_t>{in_left_pad_h, in_left_pad_w},
std::vector<ck::index_t>{in_right_pad_h, in_right_pad_w});
}
else
{
throw std::runtime_error("wrong! this Conv data_type & layout is not implemented");
}
return 1;
}
profiler/src/profiler.cpp
View file @ c254e5ab
...@@ -14,6 +14,7 @@ int profile_conv_fwd(int, char*[]); ...@@ -14,6 +14,7 @@ int profile_conv_fwd(int, char*[]);
int profile_conv_fwd_bias_relu(int, char*[]); int profile_conv_fwd_bias_relu(int, char*[]);
int profile_conv_fwd_bias_relu_add(int, char*[]); int profile_conv_fwd_bias_relu_add(int, char*[]);
int profile_conv_fwd_bias_relu_atomic_add(int, char*[]); int profile_conv_fwd_bias_relu_atomic_add(int, char*[]);
int profile_conv_bwd_data(int, char*[]);
int main(int argc, char* argv[]) int main(int argc, char* argv[])
{ {
...@@ -53,6 +54,10 @@ int main(int argc, char* argv[]) ...@@ -53,6 +54,10 @@ int main(int argc, char* argv[])
{ {
return profile_conv_fwd_bias_relu_atomic_add(argc, argv); return profile_conv_fwd_bias_relu_atomic_add(argc, argv);
} }
else if(strcmp(argv[1], "conv_bwd") == 0)
{
return profile_conv_bwd_data(argc, argv);
}
else else
{ {
// clang-format off // clang-format off
...@@ -63,7 +68,8 @@ int main(int argc, char* argv[]) ...@@ -63,7 +68,8 @@ int main(int argc, char* argv[])
" conv_fwd: ForwardConvolution\n" " conv_fwd: ForwardConvolution\n"
" conv_fwd_bias_relu: ForwardConvolution+Bias+ReLU\n" " conv_fwd_bias_relu: ForwardConvolution+Bias+ReLU\n"
" conv_fwd_bias_relu_add: ForwardConvolution+Bias+ReLU+Add\n" " conv_fwd_bias_relu_add: ForwardConvolution+Bias+ReLU+Add\n"
" conv_fwd_bias_relu_atomic_add: ForwardConvolution+Bias+ReLU+AtomicAdd\n"); " conv_fwd_bias_relu_atomic_add: ForwardConvolution+Bias+ReLU+AtomicAdd\n"
" conv_bwd: BackwardConvolution\n");
// clang-format on // clang-format on
return 0; return 0;
......
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