mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 05:01:25 +00:00

Files

Bartłomiej Kocot 1a2c0d835a [rocm-libraries] ROCm/rocm-libraries#4791 (commit 6cc17c6)

[CK][CK TILE] Improve oob check

## Motivation

Improve OOB checks. Remove permutes which have been generated by thread
buffer zero clear. at now in assembly there is only condmask instead of
permute + condmask.

Change number of KPack for generated instances

## Technical Details

Remove permute instructions from assembly

## Test Plan

test_grouped_convnd_fwd_tile

## Test Result

passed

## Submission Checklist

- [x] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

2026-02-24 21:41:44 +00:00

configs

[rocm-libraries] ROCm/rocm-libraries#4407 (commit adde219)

2026-02-11 13:43:01 +00:00

instances

[CK_BUILDER] conv bwd weight testing (#3618 )

2026-01-26 23:50:15 +01:00

CMakeLists.txt

[rocm-libraries] ROCm/rocm-libraries#4407 (commit adde219)

2026-02-11 13:43:01 +00:00

generate_instances.py

[rocm-libraries] ROCm/rocm-libraries#4791 (commit 6cc17c6)

2026-02-24 21:41:44 +00:00

README.md

[CK_BUILDER] Add grouped conv fwd ck tile profiler (#3518 )

2026-01-19 22:29:01 -07:00

README.md

Grouped Convolution Tile Instances Generator

CK Tile Convolution instances implemented via builder and generated via python script. It is integrated with tests and ckProfiler This functionality will be refactored and moved under the Tile Engine. At now to speed up development and provide tests for CK Tile Convolution it has been implemented under experimental directory.