Skip to content

GitLab

"cacheflow/vscode:/vscode.git/clone" did not exist on "c128c2ed0309dd1ff684ac0da2f8fc7ea5994b77"
Commit 95a83c6e authored by Adam Osewski's avatar Adam Osewski
Browse files

Merge remote-tracking branch 'origin/develop' into wavelet_model

parents 5b7c2432 892a8d76
Show whitespace changes
Inline Side-by-side
Showing with 173 additions and 241 deletions
example/13_pool2d_fwd/pool2d_fwd_common.hpp
View file @ 95a83c6e
...@@ -16,6 +16,7 @@ ...@@ -16,6 +16,7 @@
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp"
template <typename InDataType, template <typename InDataType,
typename OutDataType, typename OutDataType,
...@@ -172,16 +173,16 @@ bool pool_test(bool do_verification, ...@@ -172,16 +173,16 @@ bool pool_test(bool do_verification,
// tensor layout // tensor layout
auto f_host_tensor_descriptor = auto f_host_tensor_descriptor =
[](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W, auto layout) { [](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W, auto layout) {
using namespace ck::literals;
if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCHW>::value) if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCHW>::value)
{ {
return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}), return HostTensorDescriptor({N_, C_, H, W}, {C_ * H * W, H * W, W, 1_uz});
std::vector<std::size_t>({C_ * H * W, H * W, W, 1}));
} }
else if constexpr(ck::is_same<decltype(layout), else if constexpr(ck::is_same<decltype(layout),
ck::tensor_layout::convolution::NHWC>::value) ck::tensor_layout::convolution::NHWC>::value)
{ {
return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}), return HostTensorDescriptor({N_, C_, H, W}, {C_ * H * W, 1_uz, W * C_, C_});
std::vector<std::size_t>({C_ * H * W, 1, W * C_, C_}));
} }
}; };
...@@ -267,14 +268,14 @@ bool pool_test(bool do_verification, ...@@ -267,14 +268,14 @@ bool pool_test(bool do_verification,
out_device_buf.FromDevice(out_n_c_ho_wo_device.mData.data()); out_device_buf.FromDevice(out_n_c_ho_wo_device.mData.data());
pass = pass && ck::utils::check_err(out_n_c_ho_wo_device.mData, out_n_c_ho_wo_host.mData); pass = pass && ck::utils::check_err(out_n_c_ho_wo_device, out_n_c_ho_wo_host);
if constexpr(OutputIndex) if constexpr(OutputIndex)
{ {
out_indices_device_buf.FromDevice(out_indices_n_c_ho_wo_device.mData.data()); out_indices_device_buf.FromDevice(out_indices_n_c_ho_wo_device.mData.data());
pass = pass && ck::utils::check_err(out_indices_n_c_ho_wo_device.mData, pass = pass &&
out_indices_n_c_ho_wo_host.mData); ck::utils::check_err(out_indices_n_c_ho_wo_device, out_indices_n_c_ho_wo_host);
}; };
} }
......
example/14_gemm_xdl_quantization/CMakeLists.txt 0 → 100644
View file @ 95a83c6e
add_example_executable(example_gemm_xdl_relu_quantization_int8 gemm_xdl_relu_quantization_int8.cpp)
\ No newline at end of file
example/14_gemm_xdl_requant_relu_requant/gemm_xdl_requant_relu_requant_int8.cpp → example/14_gemm_xdl_quantization/gemm_xdl_relu_quantization_int8.cpp
View file @ 95a83c6e
...@@ -15,33 +15,16 @@ ...@@ -15,33 +15,16 @@
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
struct RequantReluRequant
{
// FIXME: We just need one scale for Relu / Leaky Relu / PRelu
RequantReluRequant(float scaleGemm, float scaleRelu)
: scaleGemm_(scaleGemm), scaleRelu_(scaleRelu)
{
}
__host__ __device__ constexpr void operator()(float& y, const float& x) const
{
float gemm_requant = scaleGemm_ * x;
float relu = gemm_requant > 0 ? gemm_requant : 0;
float relu_requant = scaleRelu_ * relu;
y = relu_requant > 127 ? 127 : relu_requant < -128 ? -128 : relu_requant;
}
float scaleGemm_;
float scaleRelu_;
};
template <ck::index_t... Is> template <ck::index_t... Is>
using S = ck::Sequence<Is...>; using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough; using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using ActivationOp = ck::tensor_operation::element_wise::Relu;
using CElementOp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<ActivationOp>;
using ADataType = int8_t; using ADataType = int8_t;
using BDataType = int8_t; using BDataType = int8_t;
...@@ -67,7 +50,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle ...@@ -67,7 +50,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
CShuffleDataType, // typename CShuffleDataType, CShuffleDataType, // typename CShuffleDataType,
PassThrough, // typename AElementwiseOperation, PassThrough, // typename AElementwiseOperation,
PassThrough, // typename BElementwiseOperation, PassThrough, // typename BElementwiseOperation,
RequantReluRequant, // typename CElementwiseOperation, CElementOp, // typename CElementwiseOperation,
GemmDefault, // GemmSpecialization GemmSpec, GemmDefault, // GemmSpecialization GemmSpec,
1, // index_t NumGemmKPrefetchStage, 1, // index_t NumGemmKPrefetchStage,
256, // index_t BlockSize, 256, // index_t BlockSize,
...@@ -100,13 +83,8 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle ...@@ -100,13 +83,8 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
16>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock> 16>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock>
// clang-format on // clang-format on
using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType, using ReferenceGemmInstance = ck::tensor_operation::host::
BDataType, ReferenceGemm<ADataType, BDataType, CDataType, float, PassThrough, PassThrough, CElementOp>;
CDataType,
float,
PassThrough,
PassThrough,
RequantReluRequant>;
int main(int argc, char* argv[]) int main(int argc, char* argv[])
{ {
...@@ -123,8 +101,7 @@ int main(int argc, char* argv[]) ...@@ -123,8 +101,7 @@ int main(int argc, char* argv[])
ck::index_t StrideB = 4096; ck::index_t StrideB = 4096;
ck::index_t StrideC = 4096; ck::index_t StrideC = 4096;
float scale_gemm = 0.03; float quant_multiplier = 0.03;
float scale_relu = 1;
if(argc == 4) if(argc == 4)
{ {
...@@ -157,15 +134,15 @@ int main(int argc, char* argv[]) ...@@ -157,15 +134,15 @@ int main(int argc, char* argv[])
auto f_host_tensor_descriptor = auto f_host_tensor_descriptor =
[](std::size_t row, std::size_t col, std::size_t stride, auto layout) { [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
using namespace ck::literals;
if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value) if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
{ {
return HostTensorDescriptor(std::vector<std::size_t>({row, col}), return HostTensorDescriptor({row, col}, {stride, 1_uz});
std::vector<std::size_t>({stride, 1}));
} }
else else
{ {
return HostTensorDescriptor(std::vector<std::size_t>({row, col}), return HostTensorDescriptor({row, col}, {1_uz, stride});
std::vector<std::size_t>({1, stride}));
} }
}; };
...@@ -199,7 +176,7 @@ int main(int argc, char* argv[]) ...@@ -199,7 +176,7 @@ int main(int argc, char* argv[])
auto a_element_op = PassThrough{}; auto a_element_op = PassThrough{};
auto b_element_op = PassThrough{}; auto b_element_op = PassThrough{};
auto c_element_op = RequantReluRequant{scale_gemm, scale_relu}; auto c_element_op = CElementOp{quant_multiplier, ActivationOp{}};
// do GEMM // do GEMM
auto gemm = DeviceGemmInstance{}; auto gemm = DeviceGemmInstance{};
...@@ -249,7 +226,7 @@ int main(int argc, char* argv[]) ...@@ -249,7 +226,7 @@ int main(int argc, char* argv[])
ref_invoker.Run(ref_argument); ref_invoker.Run(ref_argument);
return ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData) ? 0 : 1; return ck::utils::check_err(c_m_n_device_result, c_m_n_host_result) ? 0 : 1;
} }
return 0; return 0;
......
example/14_gemm_xdl_requant_relu_requant/CMakeLists.txt deleted 100644 → 0
View file @ 5b7c2432
add_example_executable(example_gemm_xdl_requant_relu_requant_int8 gemm_xdl_requant_relu_requant_int8.cpp)
\ No newline at end of file
example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp
View file @ 95a83c6e
...@@ -16,6 +16,7 @@ ...@@ -16,6 +16,7 @@
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
......
example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp
View file @ 95a83c6e
...@@ -16,6 +16,7 @@ ...@@ -16,6 +16,7 @@
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
......
example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp
View file @ 95a83c6e
...@@ -16,6 +16,7 @@ ...@@ -16,6 +16,7 @@
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
......
example/15_grouped_gemm/grouped_gemm_xdl_int4.cpp
View file @ 95a83c6e
...@@ -16,6 +16,7 @@ ...@@ -16,6 +16,7 @@
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
......
example/15_grouped_gemm/grouped_gemm_xdl_int8.cpp
View file @ 95a83c6e
...@@ -16,6 +16,7 @@ ...@@ -16,6 +16,7 @@
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
......
example/15_grouped_gemm/run_grouped_gemm_example.inc
View file @ 95a83c6e
...@@ -52,15 +52,15 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co ...@@ -52,15 +52,15 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
auto f_host_tensor_descriptor = auto f_host_tensor_descriptor =
[](std::size_t row, std::size_t col, std::size_t stride, auto layout) { [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
using namespace ck::literals;
if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value) if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
{ {
return HostTensorDescriptor(std::vector<std::size_t>({row, col}), return HostTensorDescriptor({row, col}, {stride, 1_uz});
std::vector<std::size_t>({stride, 1}));
} }
else else
{ {
return HostTensorDescriptor(std::vector<std::size_t>({row, col}), return HostTensorDescriptor({row, col}, {1_uz, stride});
std::vector<std::size_t>({1, stride}));
} }
}; };
...@@ -208,10 +208,10 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co ...@@ -208,10 +208,10 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
#ifdef BUILD_INT4_EXAMPLE #ifdef BUILD_INT4_EXAMPLE
const Tensor<EDataType> c_device_result_converted(c_device_tensors[i]); const Tensor<EDataType> c_device_result_converted(c_device_tensors[i]);
pass &= ck::utils::check_err(c_device_result_converted.mData, c_host_tensors[i].mData); pass &= ck::utils::check_err(c_device_result_converted, c_host_tensors[i]);
#else #else
pass &= ck::utils::check_err(c_device_tensors[i].mData, c_host_tensors[i].mData); pass &= ck::utils::check_err(c_device_tensors[i], c_host_tensors[i]);
#endif #endif
} }
} }
......
example/16_gemm_multi_d_multi_reduces/gemm_add_add_mean_meansquare_xdl_fp16.cpp
View file @ 95a83c6e
...@@ -15,6 +15,7 @@ ...@@ -15,6 +15,7 @@
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
...@@ -109,21 +110,20 @@ void DumpPerf(float ave_time, int M, int N, int K) ...@@ -109,21 +110,20 @@ void DumpPerf(float ave_time, int M, int N, int K)
} }
auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) { auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) {
return HostTensorDescriptor(std::vector<std::size_t>({len}), return HostTensorDescriptor({len}, {stride});
std::vector<std::size_t>({stride}));
}; };
auto f_host_tensor_descriptor2d = auto f_host_tensor_descriptor2d =
[](std::size_t row, std::size_t col, std::size_t stride, auto layout) { [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
using namespace ck::literals;
if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value) if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
{ {
return HostTensorDescriptor(std::vector<std::size_t>({row, col}), return HostTensorDescriptor({row, col}, {stride, 1_uz});
std::vector<std::size_t>({stride, 1}));
} }
else else
{ {
return HostTensorDescriptor(std::vector<std::size_t>({row, col}), return HostTensorDescriptor({row, col}, {1_uz, stride});
std::vector<std::size_t>({1, stride}));
} }
}; };
...@@ -259,12 +259,9 @@ int main() ...@@ -259,12 +259,9 @@ int main()
r0_device_buf.FromDevice(r0_m.mData.data()); r0_device_buf.FromDevice(r0_m.mData.data());
r1_device_buf.FromDevice(r1_m.mData.data()); r1_device_buf.FromDevice(r1_m.mData.data());
pass = ck::utils::check_err( pass = ck::utils::check_err(e_m_n, e_m_n_host, "Error: Incorrect results c", 1e-2, 1e-2);
e_m_n.mData, e_m_n_host.mData, "Error: Incorrect results c", 1e-2, 1e-2); pass &= ck::utils::check_err(r0_m, r0_m_host, "Error: Incorrect results d0", 1e-2, 1e-2);
pass &= ck::utils::check_err( pass &= ck::utils::check_err(r1_m, r1_m_host, "Error: Incorrect results d1", 1e-2, 1e-2);
r0_m.mData, r0_m_host.mData, "Error: Incorrect results d0", 1e-2, 1e-2);
pass &= ck::utils::check_err(
r1_m.mData, r1_m_host.mData, "Error: Incorrect results d1", 1e-2, 1e-2);
} }
bool time_kernel = true; bool time_kernel = true;
......
example/16_gemm_multi_d_multi_reduces/gemm_add_addsquare_xdl_int8.cpp
View file @ 95a83c6e
...@@ -160,14 +160,12 @@ bool run_gemm_reduce_add_addsquare_xdl(ck::index_t M, ...@@ -160,14 +160,12 @@ bool run_gemm_reduce_add_addsquare_xdl(ck::index_t M,
{ {
case 0: break; case 0: break;
case 1: case 1:
ck::utils::FillUniformDistributionIntegerValue<ADataType>{-5.f, 5.f}(a_m_k.begin(), ck::utils::FillUniformDistributionIntegerValue<ADataType>{-5.f, 5.f}(a_m_k);
a_m_k.end()); ck::utils::FillUniformDistributionIntegerValue<BDataType>{-5.f, 5.f}(b_k_n);
ck::utils::FillUniformDistributionIntegerValue<BDataType>{-5.f, 5.f}(b_k_n.begin(),
b_k_n.end());
break; break;
default: default:
ck::utils::FillUniformDistribution<ADataType>{-1.f, 1.f}(a_m_k.begin(), a_m_k.end()); ck::utils::FillUniformDistribution<ADataType>{-1.f, 1.f}(a_m_k);
ck::utils::FillUniformDistribution<BDataType>{-1.f, 1.f}(b_k_n.begin(), b_k_n.end()); ck::utils::FillUniformDistribution<BDataType>{-1.f, 1.f}(b_k_n);
break; break;
} }
...@@ -264,15 +262,13 @@ bool run_gemm_reduce_add_addsquare_xdl(ck::index_t M, ...@@ -264,15 +262,13 @@ bool run_gemm_reduce_add_addsquare_xdl(ck::index_t M,
Tensor<EDataType> e_m_n_host_converted(e_m_n_host); Tensor<EDataType> e_m_n_host_converted(e_m_n_host);
pass = ck::utils::check_err( pass = ck::utils::check_err(
e_m_n.mData, e_m_n_host_converted.mData, "Error: Incorrect results c", 1e-2, 1e-2); e_m_n, e_m_n_host_converted, "Error: Incorrect results c", 1e-2, 1e-2);
r0_device_buf.FromDevice(r0_m.mData.data()); r0_device_buf.FromDevice(r0_m.mData.data());
r1_device_buf.FromDevice(r1_m.mData.data()); r1_device_buf.FromDevice(r1_m.mData.data());
pass &= ck::utils::check_err( pass &= ck::utils::check_err(r0_m, r0_m_host, "Error: Incorrect results d0", 1e-2, 1e-2);
r0_m.mData, r0_m_host.mData, "Error: Incorrect results d0", 1e-2, 1e-2); pass &= ck::utils::check_err(r1_m, r1_m_host, "Error: Incorrect results d1", 1e-2, 1e-2);
pass &= ck::utils::check_err(
r1_m.mData, r1_m_host.mData, "Error: Incorrect results d1", 1e-2, 1e-2);
if(pass) if(pass)
{ {
......
example/16_gemm_multi_d_multi_reduces/gemm_reduce_xdl_common.hpp
View file @ 95a83c6e
...@@ -134,14 +134,12 @@ auto run_gemm_reduce_max_xdl(ck::index_t M, ...@@ -134,14 +134,12 @@ auto run_gemm_reduce_max_xdl(ck::index_t M,
{ {
case 0: break; case 0: break;
case 1: case 1:
ck::utils::FillUniformDistributionIntegerValue<ADataType>{-5.f, 5.f}(a_m_k.begin(), ck::utils::FillUniformDistributionIntegerValue<ADataType>{-5.f, 5.f}(a_m_k);
a_m_k.end()); ck::utils::FillUniformDistributionIntegerValue<BDataType>{-5.f, 5.f}(b_k_n);
ck::utils::FillUniformDistributionIntegerValue<BDataType>{-5.f, 5.f}(b_k_n.begin(),
b_k_n.end());
break; break;
default: default:
ck::utils::FillUniformDistribution<ADataType>{-1.f, 1.f}(a_m_k.begin(), a_m_k.end()); ck::utils::FillUniformDistribution<ADataType>{-1.f, 1.f}(a_m_k);
ck::utils::FillUniformDistribution<BDataType>{-1.f, 1.f}(b_k_n.begin(), b_k_n.end()); ck::utils::FillUniformDistribution<BDataType>{-1.f, 1.f}(b_k_n);
break; break;
} }
...@@ -243,8 +241,8 @@ auto run_gemm_reduce_max_xdl(ck::index_t M, ...@@ -243,8 +241,8 @@ auto run_gemm_reduce_max_xdl(ck::index_t M,
if constexpr(std::is_same_v<ADataType, ck::int4_t>) if constexpr(std::is_same_v<ADataType, ck::int4_t>)
{ {
Tensor<EDataType> e_m_n_device_converted(e_m_n); Tensor<EDataType> e_m_n_device_converted(e_m_n);
pass = ck::utils::check_err(e_m_n_device_converted.mData, pass = ck::utils::check_err(e_m_n_device_converted,
e_m_n_host_converted.mData, e_m_n_host_converted,
"Error: Incorrect results c", "Error: Incorrect results c",
1e-2, 1e-2,
1e-2); 1e-2);
...@@ -253,12 +251,11 @@ auto run_gemm_reduce_max_xdl(ck::index_t M, ...@@ -253,12 +251,11 @@ auto run_gemm_reduce_max_xdl(ck::index_t M,
#endif // CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4 #endif // CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
{ {
pass = ck::utils::check_err( pass = ck::utils::check_err(
e_m_n.mData, e_m_n_host_converted.mData, "Error: Incorrect results c", 1e-2, 1e-2); e_m_n, e_m_n_host_converted, "Error: Incorrect results c", 1e-2, 1e-2);
} }
r0_device_buf.FromDevice(r0_m.mData.data()); r0_device_buf.FromDevice(r0_m.mData.data());
pass &= ck::utils::check_err( pass &= ck::utils::check_err(r0_m, r0_m_host, "Error: Incorrect results d0", 1e-2, 1e-2);
r0_m.mData, r0_m_host.mData, "Error: Incorrect results d0", 1e-2, 1e-2);
if(pass) if(pass)
{ {
...@@ -339,14 +336,12 @@ bool run_gemm_reduce_mean_meansquare_xdl(ck::index_t M, ...@@ -339,14 +336,12 @@ bool run_gemm_reduce_mean_meansquare_xdl(ck::index_t M,
{ {
case 0: break; case 0: break;
case 1: case 1:
ck::utils::FillUniformDistributionIntegerValue<ADataType>{-5.f, 5.f}(a_m_k.begin(), ck::utils::FillUniformDistributionIntegerValue<ADataType>{-5.f, 5.f}(a_m_k);
a_m_k.end()); ck::utils::FillUniformDistributionIntegerValue<BDataType>{-5.f, 5.f}(b_k_n);
ck::utils::FillUniformDistributionIntegerValue<BDataType>{-5.f, 5.f}(b_k_n.begin(),
b_k_n.end());
break; break;
default: default:
ck::utils::FillUniformDistribution<ADataType>{-1.f, 1.f}(a_m_k.begin(), a_m_k.end()); ck::utils::FillUniformDistribution<ADataType>{-1.f, 1.f}(a_m_k);
ck::utils::FillUniformDistribution<BDataType>{-1.f, 1.f}(b_k_n.begin(), b_k_n.end()); ck::utils::FillUniformDistribution<BDataType>{-1.f, 1.f}(b_k_n);
break; break;
} }
...@@ -460,8 +455,8 @@ bool run_gemm_reduce_mean_meansquare_xdl(ck::index_t M, ...@@ -460,8 +455,8 @@ bool run_gemm_reduce_mean_meansquare_xdl(ck::index_t M,
if constexpr(std::is_same_v<ADataType, ck::int4_t>) if constexpr(std::is_same_v<ADataType, ck::int4_t>)
{ {
Tensor<EDataType> e_m_n_device_converted(e_m_n); Tensor<EDataType> e_m_n_device_converted(e_m_n);
pass = ck::utils::check_err(e_m_n_device_converted.mData, pass = ck::utils::check_err(e_m_n_device_converted,
e_m_n_host_converted.mData, e_m_n_host_converted,
"Error: Incorrect results c", "Error: Incorrect results c",
1e-2, 1e-2,
1e-2); 1e-2);
...@@ -470,16 +465,14 @@ bool run_gemm_reduce_mean_meansquare_xdl(ck::index_t M, ...@@ -470,16 +465,14 @@ bool run_gemm_reduce_mean_meansquare_xdl(ck::index_t M,
#endif // CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4 #endif // CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
{ {
pass = ck::utils::check_err( pass = ck::utils::check_err(
e_m_n.mData, e_m_n_host_converted.mData, "Error: Incorrect results c", 1e-2, 1e-2); e_m_n, e_m_n_host_converted, "Error: Incorrect results c", 1e-2, 1e-2);
} }
r0_device_buf.FromDevice(r0_m.mData.data()); r0_device_buf.FromDevice(r0_m.mData.data());
r1_device_buf.FromDevice(r1_m.mData.data()); r1_device_buf.FromDevice(r1_m.mData.data());
pass &= ck::utils::check_err( pass &= ck::utils::check_err(r0_m, r0_m_host, "Error: Incorrect results d0", 1e-2, 1e-2);
r0_m.mData, r0_m_host.mData, "Error: Incorrect results d0", 1e-2, 1e-2); pass &= ck::utils::check_err(r1_m, r1_m_host, "Error: Incorrect results d1", 1e-2, 1e-2);
pass &= ck::utils::check_err(
r1_m.mData, r1_m_host.mData, "Error: Incorrect results d1", 1e-2, 1e-2);
if(pass) if(pass)
{ {
......
example/17_convnd_bwd_data/CMakeLists.txt
View file @ 95a83c6e
add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp) add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility) target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
add_example_executable(example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp)
target_link_libraries(example_convnd_bwd_data_dl_fp16 PRIVATE utility)
example/17_convnd_bwd_data/convnd_bwd_data_common.hpp
View file @ 95a83c6e
...@@ -61,9 +61,13 @@ int run_conv_bwd_data(bool do_verification, ...@@ -61,9 +61,13 @@ int run_conv_bwd_data(bool do_verification,
out.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5}); out.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5}); wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
break; break;
default: case 2:
out.GenerateTensorValue(GeneratorTensor_3<OutDataType>{0.0, 1.0}); out.GenerateTensorValue(GeneratorTensor_3<OutDataType>{0.0, 1.0});
wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.5, 0.5}); wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.5, 0.5});
break;
default:
out.GenerateTensorValue(GeneratorTensor_1<OutDataType>{1});
wei.GenerateTensorValue(GeneratorTensor_1<WeiDataType>{1});
} }
DeviceMem in_device_buf(sizeof(InDataType) * in_device.mDesc.GetElementSpaceSize()); DeviceMem in_device_buf(sizeof(InDataType) * in_device.mDesc.GetElementSpaceSize());
...@@ -98,9 +102,8 @@ int run_conv_bwd_data(bool do_verification, ...@@ -98,9 +102,8 @@ int run_conv_bwd_data(bool do_verification,
if(!conv.IsSupportedArgument(argument)) if(!conv.IsSupportedArgument(argument))
{ {
throw std::runtime_error( std::cout << "Not support,please check parameters or device";
"wrong! device_conv with the specified compilation parameters does " return 0;
"not support this Conv problem");
} }
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel}); float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
...@@ -142,7 +145,7 @@ int run_conv_bwd_data(bool do_verification, ...@@ -142,7 +145,7 @@ int run_conv_bwd_data(bool do_verification,
in_device_buf.FromDevice(in_device.mData.data()); in_device_buf.FromDevice(in_device.mData.data());
return ck::utils::check_err(in_device.mData, in_host.mData) ? 0 : 1; return ck::utils::check_err(in_device, in_host) ? 0 : 1;
} }
return 0; return 0;
......
example/20_convnd_bwd_weight/convnd_bwd_weight_xdl_fp16.cpp → example/17_convnd_bwd_data/convnd_bwd_data_dl_fp16.cpp
View file @ 95a83c6e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "convnd_bwd_weight_common.hpp" #include "convnd_bwd_data_common.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.hpp" #include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp"
using InDataType = ck::half_t; using InDataType = ck::half_t;
using WeiDataType = ck::half_t; using WeiDataType = ck::half_t;
...@@ -17,61 +17,31 @@ using InElementOp = ck::tensor_operation::element_wise::PassThrough; ...@@ -17,61 +17,31 @@ using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough; using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::PassThrough; using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault = static constexpr auto ConvBwdDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default; ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
template <ck::index_t NDimSpatial> template <ck::index_t NDimSpatial>
using DeviceConvndBwdWeightInstance = // clang-format off
ck::tensor_operation::device::DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< using DeviceConvNdBwdDataInstance = ck::tensor_operation::device::DeviceConvNdBwdDataNwcKxcNwk_Dl<
NDimSpatial, // NDimSpatial // ######| NDim| InData| WeiData| OutData| AccData| In| Wei| Out| Convolution| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
InDataType, // InDataType // ######| Spatial| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Forward| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
WeiDataType, // WeiDataType // ######| | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
OutDataType, // OutDataType // ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
AccDataType, // AccDataType NDimSpatial, InDataType, WeiDataType, OutDataType, AccDataType, InElementOp, WeiElementOp, OutElementOp, ConvBwdDefault, 256, 128, 128, 16, 2, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 1, 8, 2>, S<16, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>;
InElementOp, // InElementwiseOperation // clang-format on
WeiElementOp, // WeiElementwiseOperation
OutElementOp, // OutElementwiseOperation
ConvBwdWeightDefault, // ConvolutionBackwardWeightSpecialization
256, // BlockSize
128, // MPerBlock
128, // NPerBlock
4, // K0PerBlock
8, // K1
32, // MPerXdl
32, // NPerXdl
2, // MXdlPerWave
2, // NXdlPerWave
S<1, 4, 16, 4>, // ABlockTransferThreadClusterLengths_K0_M_K1
S<0, 3, 1, 2>, // ABlockTransferThreadClusterArrangeOrder
S<0, 2, 1, 3>, // ABlockTransferSrcAccessOrder
2, // ABlockTransferSrcVectorDim
8, // ABlockTransferSrcScalarPerVector
2, // ABlockTransferDstScalarPerVector_K1
true, // ABlockLdsAddExtraM
S<1, 4, 16, 4>, // BBlockTransferThreadClusterLengths_K0_N_K1
S<0, 3, 1, 2>, // BBlockTransferThreadClusterArrangeOrder
S<0, 2, 1, 3>, // BBlockTransferSrcAccessOrder
2, // BBlockTransferSrcVectorDim
8, // BBlockTransferSrcScalarPerVector
2, // BBlockTransferDstScalarPerVector_K1
true, // BBlockLdsAddExtraN
1, // CShuffleMXdlPerWavePerShuffle
1, // CShuffleNXdlPerWavePerShuffle
S<1, 32, 1, 4>, // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
8>; // CBlockTransferScalarPerVector_NWaveNPerXdl
int main(int argc, char* argv[]) int main(int argc, char* argv[])
{ {
namespace ctc = ck::tensor_layout::convolution; namespace ctc = ck::tensor_layout::convolution;
print_helper_msg();
bool do_verification = true; bool do_verification = true;
int init_method = 1; int init_method = 1;
bool time_kernel = false; bool time_kernel = false;
ck::utils::conv::ConvParam conv_param{ ck::utils::conv::ConvParam conv_param{
2, 1, 32, 256, 1024, {3, 3}, {14, 14}, {2, 2}, {1, 1}, {1, 1}, {1, 1}}; 2, 1, 128, 256, 256, {3, 3}, {71, 71}, {2, 2}, {1, 1}, {1, 1}, {1, 1}};
ck::index_t split_k = 4;
if(argc == 1) if(argc == 1)
{ {
...@@ -91,9 +61,6 @@ int main(int argc, char* argv[]) ...@@ -91,9 +61,6 @@ int main(int argc, char* argv[])
const ck::index_t num_dim_spatial = std::stoi(argv[4]); const ck::index_t num_dim_spatial = std::stoi(argv[4]);
conv_param = ck::utils::conv::parse_conv_param(num_dim_spatial, 5, argv); conv_param = ck::utils::conv::parse_conv_param(num_dim_spatial, 5, argv);
split_k = std::stoi(argv[5 + 3 + 6 * num_dim_spatial - 1]);
split_k = std::max(1, split_k);
} }
const auto in_element_op = InElementOp{}; const auto in_element_op = InElementOp{};
...@@ -118,14 +85,14 @@ int main(int argc, char* argv[]) ...@@ -118,14 +85,14 @@ int main(int argc, char* argv[])
ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>( ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
conv_param); conv_param);
return run_conv_bwd_weight<1, return run_conv_bwd_data<1,
InDataType, InDataType,
WeiDataType, WeiDataType,
OutDataType, OutDataType,
InElementOp, InElementOp,
WeiElementOp, WeiElementOp,
OutElementOp, OutElementOp,
DeviceConvndBwdWeightInstance<1>>(do_verification, DeviceConvNdBwdDataInstance<1>>(do_verification,
init_method, init_method,
time_kernel, time_kernel,
conv_param, conv_param,
...@@ -134,8 +101,7 @@ int main(int argc, char* argv[]) ...@@ -134,8 +101,7 @@ int main(int argc, char* argv[])
out_g_n_k_wos_desc, out_g_n_k_wos_desc,
in_element_op, in_element_op,
wei_element_op, wei_element_op,
out_element_op, out_element_op);
split_k);
} }
else if(conv_param.num_dim_spatial_ == 2) else if(conv_param.num_dim_spatial_ == 2)
{ {
...@@ -155,14 +121,14 @@ int main(int argc, char* argv[]) ...@@ -155,14 +121,14 @@ int main(int argc, char* argv[])
ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>( ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
conv_param); conv_param);
return run_conv_bwd_weight<2, return run_conv_bwd_data<2,
InDataType, InDataType,
WeiDataType, WeiDataType,
OutDataType, OutDataType,
InElementOp, InElementOp,
WeiElementOp, WeiElementOp,
OutElementOp, OutElementOp,
DeviceConvndBwdWeightInstance<2>>(do_verification, DeviceConvNdBwdDataInstance<2>>(do_verification,
init_method, init_method,
time_kernel, time_kernel,
conv_param, conv_param,
...@@ -171,8 +137,7 @@ int main(int argc, char* argv[]) ...@@ -171,8 +137,7 @@ int main(int argc, char* argv[])
out_g_n_k_wos_desc, out_g_n_k_wos_desc,
in_element_op, in_element_op,
wei_element_op, wei_element_op,
out_element_op, out_element_op);
split_k);
} }
else if(conv_param.num_dim_spatial_ == 3) else if(conv_param.num_dim_spatial_ == 3)
{ {
...@@ -192,14 +157,14 @@ int main(int argc, char* argv[]) ...@@ -192,14 +157,14 @@ int main(int argc, char* argv[])
ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>( ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
conv_param); conv_param);
return run_conv_bwd_weight<3, return run_conv_bwd_data<3,
InDataType, InDataType,
WeiDataType, WeiDataType,
OutDataType, OutDataType,
InElementOp, InElementOp,
WeiElementOp, WeiElementOp,
OutElementOp, OutElementOp,
DeviceConvndBwdWeightInstance<3>>(do_verification, DeviceConvNdBwdDataInstance<3>>(do_verification,
init_method, init_method,
time_kernel, time_kernel,
conv_param, conv_param,
...@@ -208,8 +173,7 @@ int main(int argc, char* argv[]) ...@@ -208,8 +173,7 @@ int main(int argc, char* argv[])
out_g_n_k_wos_desc, out_g_n_k_wos_desc,
in_element_op, in_element_op,
wei_element_op, wei_element_op,
out_element_op, out_element_op);
split_k);
} }
return 0; return 0;
......
example/18_batched_gemm_reduce/batched_gemm_reduce_xdl_fp16.cpp
View file @ 95a83c6e
...@@ -16,6 +16,7 @@ ...@@ -16,6 +16,7 @@
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
...@@ -132,15 +133,15 @@ int main(int argc, char* argv[]) ...@@ -132,15 +133,15 @@ int main(int argc, char* argv[])
std::size_t col, std::size_t col,
std::size_t stride, std::size_t stride,
auto layout) { auto layout) {
using namespace ck::literals;
if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value) if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
{ {
return HostTensorDescriptor(std::vector<std::size_t>({batch_count, row, col}), return HostTensorDescriptor({batch_count, row, col}, {row * stride, stride, 1_uz});
std::vector<std::size_t>({row * stride, stride, 1}));
} }
else else
{ {
return HostTensorDescriptor(std::vector<std::size_t>({batch_count, row, col}), return HostTensorDescriptor({batch_count, row, col}, {col * stride, 1_uz, stride});
std::vector<std::size_t>({col * stride, 1, stride}));
} }
}; };
...@@ -149,17 +150,13 @@ int main(int argc, char* argv[]) ...@@ -149,17 +150,13 @@ int main(int argc, char* argv[])
Tensor<CDataType> c_g_m_n_host_result( Tensor<CDataType> c_g_m_n_host_result(
f_host_tensor_descriptor(BatchCount, M, N, StrideC, CLayout{})); f_host_tensor_descriptor(BatchCount, M, N, StrideC, CLayout{}));
Tensor<ReduceDataType> d0_g_m_host_result(HostTensorDescriptor(std::vector<std::size_t>( Tensor<ReduceDataType> d0_g_m_host_result({BatchCount, M});
{static_cast<std::size_t>(BatchCount), static_cast<std::size_t>(M)}))); Tensor<ReduceDataType> d1_g_m_host_result({BatchCount, M});
Tensor<ReduceDataType> d1_g_m_host_result(HostTensorDescriptor(std::vector<std::size_t>(
{static_cast<std::size_t>(BatchCount), static_cast<std::size_t>(M)})));
Tensor<CDataType> c_g_m_n_device_result( Tensor<CDataType> c_g_m_n_device_result(
f_host_tensor_descriptor(BatchCount, M, N, StrideC, CLayout{})); f_host_tensor_descriptor(BatchCount, M, N, StrideC, CLayout{}));
Tensor<ReduceDataType> d0_g_m_device_result(HostTensorDescriptor(std::vector<std::size_t>( Tensor<ReduceDataType> d0_g_m_device_result({BatchCount, M});
{static_cast<std::size_t>(BatchCount), static_cast<std::size_t>(M)}))); Tensor<ReduceDataType> d1_g_m_device_result({BatchCount, M});
Tensor<ReduceDataType> d1_g_m_device_result(HostTensorDescriptor(std::vector<std::size_t>(
{static_cast<std::size_t>(BatchCount), static_cast<std::size_t>(M)})));
std::cout << "a_g_m_k: " << a_g_m_k.mDesc << std::endl; std::cout << "a_g_m_k: " << a_g_m_k.mDesc << std::endl;
std::cout << "b_g_k_n: " << b_g_k_n.mDesc << std::endl; std::cout << "b_g_k_n: " << b_g_k_n.mDesc << std::endl;
...@@ -296,16 +293,15 @@ int main(int argc, char* argv[]) ...@@ -296,16 +293,15 @@ int main(int argc, char* argv[])
} }
} }
pass = ck::utils::check_err(c_g_m_n_host_result.mData, pass = ck::utils::check_err(
c_g_m_n_device_result.mData, c_g_m_n_host_result, c_g_m_n_device_result, "Error: Incorrect results c") &&
"Error: Incorrect results c") && ck::utils::check_err(d0_g_m_device_result,
ck::utils::check_err(d0_g_m_device_result.mData, d0_g_m_host_result,
d0_g_m_host_result.mData,
"Error: Incorrect results! D0", "Error: Incorrect results! D0",
1e-4, 1e-4,
1e-5) && 1e-5) &&
ck::utils::check_err(d1_g_m_device_result.mData, ck::utils::check_err(d1_g_m_device_result,
d1_g_m_host_result.mData, d1_g_m_host_result,
"Error: Incorrect results! D1", "Error: Incorrect results! D1",
1e-3, 1e-3,
1e-5); 1e-5);
......
example/19_binary_elementwise/broadcast_add_2d_amn_bn.cpp
View file @ 95a83c6e
...@@ -12,6 +12,7 @@ ...@@ -12,6 +12,7 @@
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp"
using F16 = ck::half_t; using F16 = ck::half_t;
using F32 = float; using F32 = float;
...@@ -71,13 +72,13 @@ int main() ...@@ -71,13 +72,13 @@ int main()
ck::index_t Stride = 1024; ck::index_t Stride = 1024;
auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) { auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) {
return HostTensorDescriptor(std::vector<std::size_t>({len}), return HostTensorDescriptor({len}, {stride});
std::vector<std::size_t>({stride}));
}; };
auto f_host_tensor_descriptor2d = [](std::size_t row, std::size_t col, std::size_t stride) { auto f_host_tensor_descriptor2d = [](std::size_t row, std::size_t col, std::size_t stride) {
return HostTensorDescriptor(std::vector<std::size_t>({row, col}), using namespace ck::literals;
std::vector<std::size_t>({stride, 1}));
return HostTensorDescriptor({row, col}, {stride, 1_uz});
}; };
Tensor<ABDataType> a_m_n(f_host_tensor_descriptor2d(M, N, Stride)); Tensor<ABDataType> a_m_n(f_host_tensor_descriptor2d(M, N, Stride));
...@@ -128,8 +129,7 @@ int main() ...@@ -128,8 +129,7 @@ int main()
host_broadcast2D<Tensor<ABDataType>, Tensor<ABDataType>, Tensor<CDataType>, Add, 0>( host_broadcast2D<Tensor<ABDataType>, Tensor<ABDataType>, Tensor<CDataType>, Add, 0>(
host_c_m_n, a_m_n, b_n, M, N, Add{}); host_c_m_n, a_m_n, b_n, M, N, Add{});
pass &= ck::utils::check_err( pass &= ck::utils::check_err(c_m_n, host_c_m_n, "Error: Incorrect results c", 1e-3, 1e-3);
c_m_n.mData, host_c_m_n.mData, "Error: Incorrect results c", 1e-3, 1e-3);
} }
return pass ? 0 : 1; return pass ? 0 : 1;
......
example/19_binary_elementwise/broadcast_add_3d_am_bmnk.cpp
View file @ 95a83c6e
...@@ -8,6 +8,7 @@ ...@@ -8,6 +8,7 @@
#include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp" #include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
#include "ck/library/utility/algorithm.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
...@@ -82,11 +83,9 @@ int main() ...@@ -82,11 +83,9 @@ int main()
std::array<ck::index_t, 3> b_strides; std::array<ck::index_t, 3> b_strides;
std::array<ck::index_t, 3> c_strides; std::array<ck::index_t, 3> c_strides;
std::copy(mnk.begin(), mnk.end(), abc_lengths.begin()); ck::ranges::copy(mnk, abc_lengths.begin());
std::copy( ck::ranges::copy(b_m_n_k.mDesc.GetStrides(), b_strides.begin());
b_m_n_k.mDesc.GetStrides().begin(), b_m_n_k.mDesc.GetStrides().end(), b_strides.begin()); ck::ranges::copy(c_m_n_k.mDesc.GetStrides(), c_strides.begin());
std::copy(
c_m_n_k.mDesc.GetStrides().begin(), c_m_n_k.mDesc.GetStrides().end(), c_strides.begin());
auto broadcastAdd = DeviceElementwiseAddInstance{}; auto broadcastAdd = DeviceElementwiseAddInstance{};
auto argument = broadcastAdd.MakeArgumentPointer( auto argument = broadcastAdd.MakeArgumentPointer(
...@@ -113,8 +112,8 @@ int main() ...@@ -113,8 +112,8 @@ int main()
host_broadcast3D_am_bmnk<Tensor<ABDataType>, Tensor<ABDataType>, Tensor<CDataType>, Add>( host_broadcast3D_am_bmnk<Tensor<ABDataType>, Tensor<ABDataType>, Tensor<CDataType>, Add>(
host_c_m_n_k, a_m, b_m_n_k, mnk, Add{}); host_c_m_n_k, a_m, b_m_n_k, mnk, Add{});
pass &= ck::utils::check_err( pass &=
c_m_n_k.mData, host_c_m_n_k.mData, "Error: Incorrect results c", 1e-3, 1e-3); ck::utils::check_err(c_m_n_k, host_c_m_n_k, "Error: Incorrect results c", 1e-3, 1e-3);
} }
return pass ? 0 : 1; return pass ? 0 : 1;
......
example/19_binary_elementwise/elementwise_add_1d.cpp
View file @ 95a83c6e
...@@ -53,8 +53,7 @@ int main() ...@@ -53,8 +53,7 @@ int main()
ck::index_t M = 1024; ck::index_t M = 1024;
auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) { auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) {
return HostTensorDescriptor(std::vector<std::size_t>({len}), return HostTensorDescriptor({len}, {stride});
std::vector<std::size_t>({stride}));
}; };
Tensor<ABDataType> a_m(f_host_tensor_descriptor1d(M, 1)); Tensor<ABDataType> a_m(f_host_tensor_descriptor1d(M, 1));
...@@ -105,8 +104,7 @@ int main() ...@@ -105,8 +104,7 @@ int main()
host_elementwise1D<Tensor<ABDataType>, Tensor<ABDataType>, Tensor<CDataType>, Add>( host_elementwise1D<Tensor<ABDataType>, Tensor<ABDataType>, Tensor<CDataType>, Add>(
host_c_m, a_m, b_m, M, Add{}); host_c_m, a_m, b_m, M, Add{});
pass &= ck::utils::check_err( pass &= ck::utils::check_err(c_m, host_c_m, "Error: Incorrect results c", 1e-3, 1e-3);
c_m.mData, host_c_m.mData, "Error: Incorrect results c", 1e-3, 1e-3);
} }
return pass ? 0 : 1; return pass ? 0 : 1;
......
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