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

[rocm-libraries] ROCm/rocm-libraries#4406 (commit 61f9f90)

Browse Source 
[CK] CK Tile grouped convolution direct load

## Motivation

CK Tile grouped convolution forward direct load support.

## Technical Details

Basic pipeline for direct load and new instances for forward for v1 and
v4 pipelines.

## Test Plan

test_grouped_convnd_fwd_tile

## Test Result

CI pending

## Submission Checklist

- [x] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
AICK-130
This commit is contained in:
Bartłomiej Kocot
2026-02-09 21:09:42 +00:00
committed by assistant-librarian[bot]
parent 0cafa68b6f
commit 27e0a34e0f
29 changed files with 739 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
profiler/include/profiler/grouped_convolution_forward_tile_algs.hpp
View File

@@ -15,8 +15,7 @@
#include "ck_tile/builder/testing/conv/reference.hpp"
#include "ck_tile/builder/conv_builder.hpp"
// Temporary disable builder validate since we don't have deduced rtol, atol support
#define ENABLE_BUILDER_VALIDATE 0
#define ENABLE_BUILDER_VALIDATE 1
namespace ck_tile::builder::profiling {
@@ -168,11 +167,8 @@ run_grouped_conv_forward_tile_algs(const ckt::Args<SIGNATURE>& args,
#else
HIP_CHECK_ERROR(
hipMemcpy(&out.data()[0], outputs.output, output_bytes_num, hipMemcpyDeviceToHost));
valid = ck_tile::check_err(out, ref, "Error: Incorrect results!", rtol, atol);
valid = ck_tile::check_err(out, ref, "Error: Incorrect results!");
#endif
std::cout << "Relative error threshold: " << rtol
<< " Absolute error threshold: " << atol << std::endl;
}
else
{