ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 06:49:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

Change relu to clamp for grouped conv fwd instances (#2249)

Browse Source
This commit is contained in:
Bartłomiej Kocot
2025-05-29 00:51:25 +02:00
committed by GitHub
parent 6df1c56ad6
commit e7906dd644
34 changed files with 291 additions and 195 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
profiler/include/profiler/profile_grouped_conv_fwd_bias_relu_impl.hpp → profiler/include/profiler/profile_grouped_conv_fwd_bias_clamp_impl.hpp
View File

@@ -11,7 +11,7 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_bias_relu.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_bias_clamp.hpp"
#include "ck/library/utility/algorithm.hpp"
#include "ck/library/utility/check_err.hpp"
@@ -35,19 +35,22 @@ template <ck::index_t NDimSpatial,
typename AComputeType = InDataType,
typename BComputeType = AComputeType,
typename IndexType = ck::index_t>
bool profile_grouped_conv_fwd_bias_relu_impl(int do_verification,
int init_method,
bool do_log,
bool time_kernel,
const ck::utils::conv::ConvParam& conv_param)
bool profile_grouped_conv_fwd_bias_clamp_impl(int do_verification,
int init_method,
bool do_log,
bool time_kernel,
const ck::utils::conv::ConvParam& conv_param)
{
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::AddRelu;
using OutElementOp = ck::tensor_operation::element_wise::AddClamp;
const float floor = 0.f;
const float ceil = 256.f;
const auto in_element_op = InElementOp{};
const auto wei_element_op = WeiElementOp{};
const auto out_element_op = OutElementOp{};
const auto out_element_op = OutElementOp{floor, ceil};
const auto in_g_n_c_wis_desc =
ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_param);