ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-18 03:49:41 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit 'b0a2d99d100f2e4212ebbed080acb49a404035ab' into develop

Browse Source
This commit is contained in:
assistant-librarian[bot]
2025-09-26 01:40:00 +00:00
parent e575ac4332
commit 19f49ee63e
123 changed files with 1733 additions and 849 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
example/44_elementwise_permute/elementwise_binary_4D_fp16.cpp
View File

@@ -22,6 +22,8 @@ using F32 = float;
using ADataType = F16;
using BDataType = F16;
using NchwLayout = ck::tensor_layout::convolution::NCHW;
using NhwcLayout = ck::tensor_layout::convolution::NHWC;
using UnaryScale = ck::tensor_operation::element_wise::Scale;
using UnarySquare = ck::tensor_operation::element_wise::UnarySquare;
using UnaryScaleSquare =
@@ -73,11 +75,11 @@ int main(int argc, char* argv[])
1};
ck::ranges::copy(nchw, ab_lengths.begin());
std::array<Tensor<ADataType>, 2> as = {Tensor<ADataType>(ab_lengths, ab_strides),
Tensor<ADataType>(ab_lengths, ab_strides)};
std::array<Tensor<ADataType>, 2> as = {Tensor<ADataType>(ab_lengths, ab_strides, NchwLayout{}),
Tensor<ADataType>(ab_lengths, ab_strides, NchwLayout{})};
Tensor<ADataType>& a0 = as[0];
Tensor<ADataType>& a1 = as[1];
Tensor<BDataType> b(ab_lengths, ab_strides);
Tensor<BDataType> b(ab_lengths, ab_strides, NchwLayout{});
float alpha = 3.f;
float beta = 2.f;
a0.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
@@ -134,7 +136,7 @@ int main(int argc, char* argv[])
if(do_verification)
{
Tensor<BDataType> host_b(ab_lengths, ab_strides);
Tensor<BDataType> host_b(ab_lengths, ab_strides, NchwLayout{});
using ReferenceElementwiseInstance = ck::tensor_operation::host::
ReferenceElementwise<2, ADataType, BDataType, BinaryAddUnaryScaleSquare>;