mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-24 06:44:36 +00:00
Fusion Conv+Bias+ReLU(+Add) (#62)
* fix relu
* clean up
* clean up
* adding 1x1 conv
* adding 1x1 conv
* added 1x1 conv
* refactor
* refactor
* refactor
* added profiler for conv+bias+relu+add
* clean up
* adding conv+bias+relu
* adding conv+bias+relu
* added conv+bias+relu
* Update README.md
* update cpu verification
* adding c shuffle
* update static_tensor for dealing with invalid element
* adding c shuffle
* debugging
* fix bug
* convert to fp16 before shuffle
* shuffle more than one M/NRepeat
* clean up
* remove coordinate step hack from GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r1
* clean up
* remove coordinate step hack from all gridwise gemm xdl
* clean up coordinate step hack
* clean up coordinate step hack
* ThreadwiseTensorSliceTransfer_v3r2 support pointwise op on both src and dst
* adding output shuffle in conv+bias+relu+add
* update
* added conv+bias+relu+add with c shuffle
* added conv+bias+relu+add with c shuffle
* fix forward_sweep bugs in threadwise copy
* clean up
* refactor
* clean up
* clean up
* added conv_c_shuffle+bias_relu
* clean up
* added conv+bias+relu+atomic_add
* clean up
* clean up
* clean up
* clean up
* clean up
* clean up
* misc fixes; add 1x1 specialization
* clean up
* delete unused device op
* clean up
* add support for odd C value
[ROCm/composable_kernel commit: acbd7bd7c5]
This commit is contained in:
Chao Liu
227
profiler/profile_gemm.cpp
Normal file
227
profiler/profile_gemm.cpp
Normal file
@@ -0,0 +1,227 @@
|
||||
#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 "host_gemm.hpp"
|
||||
#include "device_tensor.hpp"
|
||||
#include "device_base.hpp"
|
||||
#include "device_gemm_xdl.hpp"
|
||||
#include "profile_gemm_impl.hpp"
|
||||
|
||||
enum GemmMatrixLayout
|
||||
{
|
||||
MK_KN_MN, // 0
|
||||
MK_NK_MN, // 1
|
||||
KM_KN_MN, // 2
|
||||
KM_NK_MN, // 3
|
||||
MK_KN_NM, // 4
|
||||
MK_NK_NM, // 5
|
||||
KM_KN_NM, // 6
|
||||
KM_NK_NM, // 7
|
||||
};
|
||||
|
||||
enum GemmDataType
|
||||
{
|
||||
F32_F32_F32, // 0
|
||||
F16_F16_F16, // 1
|
||||
};
|
||||
|
||||
int profile_gemm(int argc, char* argv[])
|
||||
{
|
||||
if(argc != 14)
|
||||
{
|
||||
printf("arg1: tensor operation (gemm: GEMM)\n");
|
||||
printf("arg2: data type (0: fp32; 1: fp16)\n");
|
||||
printf("arg3: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n");
|
||||
printf(" 1: A[m, k] * B[n, k] = C[m, n];\n");
|
||||
printf(" 2: A[k, n] * B[k, n] = C[m, n];\n");
|
||||
printf(" 3: A[k, n] * B[n, k] = C[m, n])\n");
|
||||
printf("arg4: verification (0: no; 1: yes)\n");
|
||||
printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
|
||||
printf("arg8: print tensor value (0: no; 1: yes)\n");
|
||||
printf("arg7: run kernel # of times (>1)\n");
|
||||
printf("arg8 to 13: M, N, K, StrideA, StrideB, StrideC\n");
|
||||
exit(1);
|
||||
}
|
||||
|
||||
const int data_type = static_cast<GemmDataType>(std::stoi(argv[2]));
|
||||
const int layout = static_cast<GemmMatrixLayout>(std::stoi(argv[3]));
|
||||
const bool do_verification = std::stoi(argv[4]);
|
||||
const int init_method = std::stoi(argv[5]);
|
||||
const bool do_log = std::stoi(argv[6]);
|
||||
const int nrepeat = std::stoi(argv[7]);
|
||||
|
||||
const int M = std::stoi(argv[8]);
|
||||
const int N = std::stoi(argv[9]);
|
||||
const int K = std::stoi(argv[10]);
|
||||
|
||||
const int StrideA = std::stoi(argv[11]);
|
||||
const int StrideB = std::stoi(argv[12]);
|
||||
const int StrideC = std::stoi(argv[13]);
|
||||
|
||||
if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
|
||||
{
|
||||
ck::profiler::profile_gemm_impl<ck::half_t,
|
||||
ck::half_t,
|
||||
ck::half_t,
|
||||
ck::tensor_layout::gemm::RowMajor,
|
||||
ck::tensor_layout::gemm::RowMajor,
|
||||
ck::tensor_layout::gemm::RowMajor>(
|
||||
do_verification,
|
||||
init_method,
|
||||
do_log,
|
||||
nrepeat,
|
||||
M,
|
||||
N,
|
||||
K,
|
||||
(StrideA < 0) ? K : StrideA,
|
||||
(StrideB < 0) ? N : StrideB,
|
||||
(StrideC < 0) ? N : StrideC);
|
||||
}
|
||||
else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN)
|
||||
{
|
||||
ck::profiler::profile_gemm_impl<ck::half_t,
|
||||
ck::half_t,
|
||||
ck::half_t,
|
||||
ck::tensor_layout::gemm::RowMajor,
|
||||
ck::tensor_layout::gemm::ColumnMajor,
|
||||
ck::tensor_layout::gemm::RowMajor>(
|
||||
do_verification,
|
||||
init_method,
|
||||
do_log,
|
||||
nrepeat,
|
||||
M,
|
||||
N,
|
||||
K,
|
||||
(StrideA < 0) ? K : StrideA,
|
||||
(StrideB < 0) ? K : StrideB,
|
||||
(StrideC < 0) ? N : StrideC);
|
||||
}
|
||||
else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN)
|
||||
{
|
||||
ck::profiler::profile_gemm_impl<ck::half_t,
|
||||
ck::half_t,
|
||||
ck::half_t,
|
||||
ck::tensor_layout::gemm::ColumnMajor,
|
||||
ck::tensor_layout::gemm::RowMajor,
|
||||
ck::tensor_layout::gemm::RowMajor>(
|
||||
do_verification,
|
||||
init_method,
|
||||
do_log,
|
||||
nrepeat,
|
||||
M,
|
||||
N,
|
||||
K,
|
||||
(StrideA < 0) ? M : StrideA,
|
||||
(StrideB < 0) ? N : StrideB,
|
||||
(StrideC < 0) ? N : StrideC);
|
||||
}
|
||||
else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN)
|
||||
{
|
||||
ck::profiler::profile_gemm_impl<ck::half_t,
|
||||
ck::half_t,
|
||||
ck::half_t,
|
||||
ck::tensor_layout::gemm::ColumnMajor,
|
||||
ck::tensor_layout::gemm::ColumnMajor,
|
||||
ck::tensor_layout::gemm::RowMajor>(
|
||||
do_verification,
|
||||
init_method,
|
||||
do_log,
|
||||
nrepeat,
|
||||
M,
|
||||
N,
|
||||
K,
|
||||
(StrideA < 0) ? M : StrideA,
|
||||
(StrideB < 0) ? K : StrideB,
|
||||
(StrideC < 0) ? N : StrideC);
|
||||
}
|
||||
else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_KN_MN)
|
||||
{
|
||||
ck::profiler::profile_gemm_impl<float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::gemm::RowMajor,
|
||||
ck::tensor_layout::gemm::RowMajor,
|
||||
ck::tensor_layout::gemm::RowMajor>(
|
||||
do_verification,
|
||||
init_method,
|
||||
do_log,
|
||||
nrepeat,
|
||||
M,
|
||||
N,
|
||||
K,
|
||||
(StrideA < 0) ? K : StrideA,
|
||||
(StrideB < 0) ? N : StrideB,
|
||||
(StrideC < 0) ? N : StrideC);
|
||||
}
|
||||
else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_NK_MN)
|
||||
{
|
||||
ck::profiler::profile_gemm_impl<float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::gemm::RowMajor,
|
||||
ck::tensor_layout::gemm::ColumnMajor,
|
||||
ck::tensor_layout::gemm::RowMajor>(
|
||||
do_verification,
|
||||
init_method,
|
||||
do_log,
|
||||
nrepeat,
|
||||
M,
|
||||
N,
|
||||
K,
|
||||
(StrideA < 0) ? K : StrideA,
|
||||
(StrideB < 0) ? K : StrideB,
|
||||
(StrideC < 0) ? N : StrideC);
|
||||
}
|
||||
else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::KM_KN_MN)
|
||||
{
|
||||
ck::profiler::profile_gemm_impl<float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::gemm::ColumnMajor,
|
||||
ck::tensor_layout::gemm::RowMajor,
|
||||
ck::tensor_layout::gemm::RowMajor>(
|
||||
do_verification,
|
||||
init_method,
|
||||
do_log,
|
||||
nrepeat,
|
||||
M,
|
||||
N,
|
||||
K,
|
||||
(StrideA < 0) ? M : StrideA,
|
||||
(StrideB < 0) ? N : StrideB,
|
||||
(StrideC < 0) ? N : StrideC);
|
||||
}
|
||||
else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::KM_NK_MN)
|
||||
{
|
||||
ck::profiler::profile_gemm_impl<float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::gemm::ColumnMajor,
|
||||
ck::tensor_layout::gemm::ColumnMajor,
|
||||
ck::tensor_layout::gemm::RowMajor>(
|
||||
do_verification,
|
||||
init_method,
|
||||
do_log,
|
||||
nrepeat,
|
||||
M,
|
||||
N,
|
||||
K,
|
||||
(StrideA < 0) ? M : StrideA,
|
||||
(StrideB < 0) ? K : StrideB,
|
||||
(StrideC < 0) ? N : StrideC);
|
||||
}
|
||||
else
|
||||
{
|
||||
throw std::runtime_error("wrong! this GEMM data_type & layout is not implemented");
|
||||
}
|
||||
|
||||
return 1;
|
||||
}
|
||||
Reference in New Issue
Block a user