ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-16 19:09:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

Multi-kernel CGEMM (#230)

Browse Source 
* Reference CGEMM + test stub

* Format.

* Incomplete simple implementation

* Library instances

* Sketch of tests

* Test fixes.

* Example added

* Cosmetics

* Add elementwise operation kernel and example

* Add comment

* Add template argument of dim . Prepare to support multiple dimension

* Rename example

* Support 1 dimension

* Add static assert

* Add comment

* Second auxiliary buffer added

* Extract pad

* Remove redundant argument

* Support any dimension for elementwise operation

* Remove line

* Let it be the multiple number of CU

* Move thread per block to the parameter of constructor

* Consuming binary ops to do A+B / A-B

* Fix + cosmetics + bf16 test commented out temporarily

* Format

* Enabling bf16 test

* Revert "Enabling bf16 test"

This reverts commit f497e2ba44.

* Fix + test reenabled

* fix build

* Revert "fix build"

This reverts commit d73102384b.

* post PR #235 merge fix

* amend

* Single workspace for cgemm + helper

* Perf calc fix

* Review remarks: static_cast

* Review remarks: binary ops templated

* Cleaning

* Removal of instances and their tests

* Review remarks from aosew addressed

* Review remark: unnecessary attribute

* Post-merge fixes

* Restrict 4gemm to PassThrough + bug fix

* Review remarks

* update licence

* change cgemm example to fp16

Co-authored-by: rocking <chunylai@amd.com>
Co-authored-by: Chao Liu <chao.liu2@amd.com>
Co-authored-by: Anthony Chang <ac.chang@outlook.com>

[ROCm/composable_kernel commit: 7b1e2c379e]
This commit is contained in:
myamlak
2022-05-31 17:20:55 +02:00
committed by GitHub
parent c31ecedf54
commit 441eaf22b7
12 changed files with 1756 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
example/19_binary_elementwise/broadcast_add_2d_amn_bn.cpp
View File

@@ -1,3 +1,28 @@
/*******************************************************************************
*
* MIT License
*
* Copyright (c) 2022 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*
*******************************************************************************/
#include <iostream>
#include <cstdlib>
#include "check_err.hpp"
@@ -17,7 +42,8 @@ using ABDataType = F16;
using CDataType = F16;
using EltwiseComputeDataType = F32;
using Add = ck::tensor_operation::binary_element_wise::Add;
using Add = ck::tensor_operation::binary_element_wise::
Add<EltwiseComputeDataType, EltwiseComputeDataType, EltwiseComputeDataType>;
using DeviceElementwiseAddInstance =
ck::tensor_operation::device::DeviceBinaryElementwise<ABDataType,
@@ -46,19 +72,19 @@ void host_broadcast2D(
{
for(int n = 0; n < N; ++n)
{
ComputeDataType Amn = static_cast<ComputeDataType>(A(m, n));
ComputeDataType Amn = ck::type_convert<ComputeDataType>(A(m, n));
ComputeDataType Cmn = 0;
if constexpr(broadcastDim == 0)
{
ComputeDataType Bn = static_cast<ComputeDataType>(B(n));
ComputeDataType Bn = ck::type_convert<ComputeDataType>(B(n));
functor(Cmn, Amn, Bn);
}
else
{
ComputeDataType Bm = static_cast<ComputeDataType>(B(m));
ComputeDataType Bm = ck::type_convert<ComputeDataType>(B(m));
functor(Cmn, Amn, Bm);
}
C(m, n) = static_cast<ctype>(Cmn);
C(m, n) = ck::type_convert<ctype>(Cmn);
}
}
}

9
example/19_binary_elementwise/broadcast_add_3d_am_bmnk.cpp
View File

@@ -17,7 +17,8 @@ using ABDataType = F16;
using CDataType = F16;
using EltwiseComputeDataType = F32;
using Add = ck::tensor_operation::binary_element_wise::Add;
using Add = ck::tensor_operation::binary_element_wise::
Add<EltwiseComputeDataType, EltwiseComputeDataType, EltwiseComputeDataType>;
using DeviceElementwiseAddInstance =
ck::tensor_operation::device::DeviceBinaryElementwise<ABDataType,
@@ -48,11 +49,11 @@ void host_broadcast3D_am_bmnk(HostTensorC& C,
for(std::size_t n = 0; n < shape[1]; ++n)
for(std::size_t k = 0; k < shape[2]; ++k)
{
ComputeDataType a_val = static_cast<ComputeDataType>(A(m));
ComputeDataType b_val = static_cast<ComputeDataType>(B(m, n, k));
ComputeDataType a_val = ck::type_convert<ComputeDataType>(A(m));
ComputeDataType b_val = ck::type_convert<ComputeDataType>(B(m, n, k));
ComputeDataType c_val = 0;
functor(c_val, a_val, b_val);
C(m, n, k) = static_cast<ctype>(c_val);
C(m, n, k) = ck::type_convert<ctype>(c_val);
}
}

34
example/19_binary_elementwise/elementwise_add_1d.cpp
View File

@@ -1,3 +1,28 @@
/*******************************************************************************
*
* MIT License
*
* Copyright (c) 2022 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*
*******************************************************************************/
#include <iostream>
#include <cstdlib>
#include "check_err.hpp"
@@ -17,7 +42,8 @@ using ABDataType = F16;
using CDataType = F16;
using EltwiseComputeDataType = F32;
using Add = ck::tensor_operation::binary_element_wise::Add;
using Add = ck::tensor_operation::binary_element_wise::
Add<EltwiseComputeDataType, EltwiseComputeDataType, EltwiseComputeDataType>;
using DeviceElementwiseAddInstance =
ck::tensor_operation::device::DeviceBinaryElementwise<ABDataType,
@@ -43,11 +69,11 @@ void host_elementwise1D(
for(int m = 0; m < M; ++m)
{
ComputeDataType Am = static_cast<ComputeDataType>(A(m));
ComputeDataType Bm = static_cast<ComputeDataType>(B(m));
ComputeDataType Am = ck::type_convert<ComputeDataType>(A(m));
ComputeDataType Bm = ck::type_convert<ComputeDataType>(B(m));
ComputeDataType Cm = 0;
functor(Cm, Am, Bm);
C(m) = static_cast<ctype>(Cm);
C(m) = ck::type_convert<ctype>(Cm);
}
}

34
example/19_binary_elementwise/elementwise_add_4d.cpp
View File

@@ -1,3 +1,28 @@
/*******************************************************************************
*
* MIT License
*
* Copyright (c) 2020 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*
*******************************************************************************/
#include <iostream>
#include <cstdlib>
#include "check_err.hpp"
@@ -17,7 +42,8 @@ using ABDataType = F16;
using CDataType = F16;
using EltwiseComputeDataType = F32;
using Add = ck::tensor_operation::binary_element_wise::Add;
using Add = ck::tensor_operation::binary_element_wise::
Add<EltwiseComputeDataType, EltwiseComputeDataType, EltwiseComputeDataType>;
using DeviceElementwiseAddInstance =
ck::tensor_operation::device::DeviceBinaryElementwise<ABDataType,
@@ -49,11 +75,11 @@ void host_elementwise4D(HostTensorC& C,
for(std::size_t h = 0; h < shape[2]; ++h)
for(std::size_t w = 0; w < shape[3]; ++w)
{
ComputeDataType a_val = static_cast<ComputeDataType>(A(n, c, h, w));
ComputeDataType b_val = static_cast<ComputeDataType>(B(n, c, h, w));
ComputeDataType a_val = ck::type_convert<ComputeDataType>(A(n, c, h, w));
ComputeDataType b_val = ck::type_convert<ComputeDataType>(B(n, c, h, w));
ComputeDataType c_val = 0;
functor(c_val, a_val, b_val);
C(n, c, h, w) = static_cast<ctype>(c_val);
C(n, c, h, w) = ck::type_convert<ctype>(c_val);
}
}

1
example/22_cgemm/CMakeLists.txt Normal file
View File

@@ -0,0 +1 @@
add_example_executable(example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp)

302
example/22_cgemm/cgemm_xdl_fp16.cpp Normal file
View File

@@ -0,0 +1,302 @@
/*******************************************************************************
*
* MIT License
*
* Copyright (c) 2022 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*
*******************************************************************************/
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include <stdlib.h>
#include <half.hpp>
#include "check_err.hpp"
#include "config.hpp"
#include "device.hpp"
#include "host_tensor.hpp"
#include "host_tensor_generator.hpp"
#include "device_tensor.hpp"
#include "device_cgemm_4gemm_xdl_cshuffle.hpp"
#include "element_wise_operation.hpp"
#include "reference_cgemm.hpp"
#include "gemm_specialization.hpp"
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using F16 = ck::half_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using ADataType = F16;
using BDataType = F16;
using CDataType = F16;
using AccDataType = F32;
using ALayout = ck::tensor_layout::gemm::RowMajor;
using BLayout = ck::tensor_layout::gemm::ColumnMajor;
using CLayout = ck::tensor_layout::gemm::RowMajor;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// clang-format off
using DeviceCGemmInstance = ck::tensor_operation::device::DeviceCGemm_4Gemm_Xdl_CShuffle
<ALayout, // typename ALayout
BLayout, // typename BLayout
CLayout, // typename CLayout
ADataType, // typename ADataType
BDataType, // typename BDataType
CDataType, // typename CDataType
AccDataType, // typename GemmAccDataType
CDataType, // typename CShuffleDataType
PassThrough, // typename AElementwiseOperation
PassThrough, // typename BElementwiseOperation
PassThrough, // typename CElementwiseOperation
GemmDefault, // GemmSpecialization GemmSpec
1, // index_t NumGemmKPrefetchStage
256, // index_t BlockSize
256, // index_t MPerBlock
128, // index_t NPerBlock
32, // index_t KPerBlock
8, // index_t AK1
8, // index_t BK1
32, // index_t MPerXDL
32, // index_t NPerXDL
4, // index_t MXdlPerWave
2, // index_t NXdlPerWave
S<4, 64, 1>, // typename ABlockTransferThreadClusterLengths_AK0_M_AK1
S<1, 0, 2>, // typename ABlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // typename ABlockTransferSrcAccessOrder
2, // index_t ABlockTransferSrcVectorDim
8, // index_t ABlockTransferSrcScalarPerVector
8, // index_t ABlockTransferDstScalarPerVector_AK1
1, // index_t ABlockLdsExtraM
S<4, 64, 1>, // typename BBlockTransferThreadClusterLengths_BK0_N_BK1
S<1, 0, 2>, // typename BBlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // typename BBlockTransferSrcAccessOrder
2, // index_t BBlockTransferSrcVectorDim
8, // index_t BBlockTransferSrcScalarPerVector
8, // index_t BBlockTransferDstScalarPerVector_BK1
1, // index_t BBlockLdsExtraN
1, // index_t CShuffleMXdlPerWavePerShuffle
1, // index_t CShuffleNXdlPerWavePerShuffle
S<1, 32, 1, 8>, // typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
8>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock
// clang-format on
using ReferenceCGemmInstance = ck::tensor_operation::host::
ReferenceCGemm<ADataType, BDataType, CDataType, PassThrough, PassThrough, PassThrough>;
int main(int argc, char* argv[])
{
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
// CGEMM shape
ck::index_t M = 3840;
ck::index_t N = 4096;
ck::index_t K = 4096;
ck::index_t StrideA = 4096;
ck::index_t StrideB = 4096;
ck::index_t StrideC = 4096;
if(argc == 4)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
}
else if(argc == 10)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
M = std::stoi(argv[4]);
N = std::stoi(argv[5]);
K = std::stoi(argv[6]);
StrideA = std::stoi(argv[7]);
StrideB = std::stoi(argv[8]);
StrideC = std::stoi(argv[9]);
}
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 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
exit(0);
}
auto f_host_tensor_descriptor =
[](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
{
return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
std::vector<std::size_t>({stride, 1}));
}
else
{
return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
std::vector<std::size_t>({1, stride}));
}
};
Tensor<ADataType> a_m_k_real(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
Tensor<ADataType> a_m_k_imag(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
Tensor<BDataType> b_k_n_real(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
Tensor<BDataType> b_k_n_imag(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
Tensor<CDataType> c_m_n_real_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
Tensor<CDataType> c_m_n_imag_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
std::cout << "a_m_k_real: " << a_m_k_real.mDesc << std::endl;
std::cout << "a_m_k_imag: " << a_m_k_imag.mDesc << std::endl;
std::cout << "b_k_n_real: " << b_k_n_real.mDesc << std::endl;
std::cout << "b_k_n_imag: " << b_k_n_imag.mDesc << std::endl;
std::cout << "c_m_n_real: " << c_m_n_real_device_result.mDesc << std::endl;
std::cout << "c_m_n_imag: " << c_m_n_imag_device_result.mDesc << std::endl;
switch(init_method)
{
case 0: break;
case 1:
a_m_k_real.GenerateTensorValue(GeneratorTensor_2<ADataType>{-2, 2});
a_m_k_imag.GenerateTensorValue(GeneratorTensor_2<ADataType>{-2, 2});
b_k_n_real.GenerateTensorValue(GeneratorTensor_2<BDataType>{-2, 2});
b_k_n_imag.GenerateTensorValue(GeneratorTensor_2<BDataType>{-2, 2});
break;
default:
a_m_k_real.GenerateTensorValue(GeneratorTensor_3<ADataType>{-0.5, 0.5});
a_m_k_imag.GenerateTensorValue(GeneratorTensor_3<ADataType>{-0.5, 0.5});
b_k_n_real.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
b_k_n_imag.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
}
auto cgemm = DeviceCGemmInstance{};
DeviceMem a_m_k_real_device_buf(sizeof(ADataType) * a_m_k_real.mDesc.GetElementSpace());
DeviceMem a_m_k_imag_device_buf(sizeof(ADataType) * a_m_k_imag.mDesc.GetElementSpace());
DeviceMem b_k_n_real_device_buf(sizeof(BDataType) * b_k_n_real.mDesc.GetElementSpace());
DeviceMem b_k_n_imag_device_buf(sizeof(BDataType) * b_k_n_imag.mDesc.GetElementSpace());
DeviceMem c_m_n_real_device_buf(sizeof(CDataType) *
c_m_n_real_device_result.mDesc.GetElementSpace());
DeviceMem c_m_n_imag_device_buf(sizeof(CDataType) *
c_m_n_imag_device_result.mDesc.GetElementSpace());
DeviceMem workspace_device_buf(cgemm.GetWorkspaceSize(M, N, K, StrideA, StrideB, StrideC));
a_m_k_real_device_buf.ToDevice(a_m_k_real.mData.data());
a_m_k_imag_device_buf.ToDevice(a_m_k_imag.mData.data());
b_k_n_real_device_buf.ToDevice(b_k_n_real.mData.data());
b_k_n_imag_device_buf.ToDevice(b_k_n_imag.mData.data());
auto a_element_op = PassThrough{};
auto b_element_op = PassThrough{};
auto c_element_op = PassThrough{};
// do GEMM
auto invoker = cgemm.MakeInvoker();
auto argument =
cgemm.MakeArgument(static_cast<ADataType*>(a_m_k_real_device_buf.GetDeviceBuffer()),
static_cast<ADataType*>(a_m_k_imag_device_buf.GetDeviceBuffer()),
static_cast<BDataType*>(b_k_n_real_device_buf.GetDeviceBuffer()),
static_cast<BDataType*>(b_k_n_imag_device_buf.GetDeviceBuffer()),
static_cast<CDataType*>(c_m_n_real_device_buf.GetDeviceBuffer()),
static_cast<CDataType*>(c_m_n_imag_device_buf.GetDeviceBuffer()),
static_cast<CDataType*>(workspace_device_buf.GetDeviceBuffer()),
M,
N,
K,
StrideA,
StrideB,
StrideC,
a_element_op,
b_element_op,
c_element_op);
if(!cgemm.IsSupportedArgument(argument))
{
throw std::runtime_error(
"wrong! device_cgemm with the specified compilation parameters does "
"not support this CGEMM problem");
}
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(8) * M * N * K;
std::size_t num_btype =
std::size_t(2) *
(sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(CDataType) * M * N);
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, "
<< cgemm.GetTypeString() << std::endl;
c_m_n_real_device_buf.FromDevice(c_m_n_real_device_result.mData.data());
c_m_n_imag_device_buf.FromDevice(c_m_n_imag_device_result.mData.data());
if(do_verification)
{
Tensor<CDataType> c_m_n_real_host_result(
f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
Tensor<CDataType> c_m_n_imag_host_result(
f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
auto ref_cgemm = ReferenceCGemmInstance{};
auto ref_invoker = ref_cgemm.MakeInvoker();
auto ref_argument = ref_cgemm.MakeArgument(a_m_k_real,
a_m_k_imag,
b_k_n_real,
b_k_n_imag,
c_m_n_real_host_result,
c_m_n_imag_host_result,
a_element_op,
b_element_op,
c_element_op);
ref_invoker.Run(ref_argument);
ck::utils::check_err(c_m_n_real_device_result.mData,
c_m_n_real_host_result.mData,
"Verification error: incorrect results in real part!",
1e-2f,
1e-1f);
ck::utils::check_err(c_m_n_imag_device_result.mData,
c_m_n_imag_host_result.mData,
"Verification error: incorrect results in imaginary part!",
1e-2f,
1e-1f);
}
return 0;
}

3
example/CMakeLists.txt
View File

@@ -48,10 +48,11 @@ add_subdirectory(11_conv2d_bwd_weight)
add_subdirectory(12_reduce)
add_subdirectory(13_pool2d_fwd)
add_subdirectory(14_gemm_xdl_requant_relu_requant)
add_subdirectory(17_convnd_bwd_data_xdl)
add_subdirectory(15_grouped_gemm)
add_subdirectory(16_gemm_reduce)
add_subdirectory(17_convnd_bwd_data_xdl)
add_subdirectory(18_batched_gemm_reduce)
add_subdirectory(19_binary_elementwise)
add_subdirectory(20_convnd_bwd_weight_xdl)
add_subdirectory(21_gemm_layernorm)
add_subdirectory(22_cgemm)