Add support for GKCYX grouped conv weight (#2023)

* Grouped conv bwd weight GKCYX support

* fix and changelog

* fix

* fix

* fixes

* comments

* fix
This commit is contained in:
Bartłomiej Kocot
2025-04-02 23:59:49 +02:00
committed by GitHub
parent e5ad48a784
commit 2ccf914888
101 changed files with 1004 additions and 356 deletions

View File

@@ -30,14 +30,14 @@ List of the device operations for grouped convolution forward in CK:
Table of supported cases by instance factory with XDL instruction:
| |NHWGC/GKYXC/NHWGK|NGCHW/GKYXC/NGKHW|GNHWC/GKYXC/GNHWK|
|-------|---|---|---|
|bf16 |2D, 3D|2D|1D, 2D, 3D|
|fp16 |2D, 3D|2D|1D, 2D, 3D|
|fp32 |2D, 3D|2D|1D, 2D, 3D|
|int8 |2D, 3D|2D|1D, 3D|
|fp8 |3D|✗|✗|
|bf8 |3D|✗|✗|
| |NHWGC/GKYXC/NHWGK|NGCHW/GKYXC/NGKHW|NGCHW/GKCYX/NGKHW|GNHWC/GKYXC/GNHWK|
|-------|---|---|---|---|
|bf16 |2D, 3D|2D|2D|1D, 2D, 3D|
|fp16 |2D, 3D|2D|2D|1D, 2D, 3D|
|fp32 |2D, 3D|2D|2D|1D, 2D, 3D|
|int8 |2D, 3D|2D|2D|1D, 3D|
|fp8 |3D|✗|✗|✗|
|bf8 |3D|✗|✗|✗|
Table of supported cases by instance factory with WMMA instruction:

View File

@@ -34,12 +34,12 @@ List of the device operations for grouped convolution backward weight in CK:
Table of supported cases by instance factory with XDL instruction:
| |NHWGC/GKYXC/NHWGK|NGCHW/GKYXC/NGKHW|GNHWC/GKYXC/GNHWK|
|-------|---|---|---|
|bf16|2D, 3D|2D, 3D|✗|
|bf16(fp32 for weight)|2D, 3D|✗|1D, 2D, 3D|
|fp16 |2D, 3D|2D, 3D|1D, 2D, 3D|
|fp32 |2D, 3D|2D, 3D|1D, 2D, 3D|
| |NHWGC/GKYXC/NHWGK|NGCHW/GKYXC/NGKHW|NGCHW/GKCYX/NGKHW|GNHWC/GKYXC/GNHWK|
|-------|---|---|---|---|
|bf16|2D, 3D|2D, 3D|2D, 3D|✗|
|bf16(fp32 for weight)|2D, 3D|✗|✗|1D, 2D, 3D|
|fp16 |2D, 3D|2D, 3D|2D, 3D|1D, 2D, 3D|
|fp32 |2D, 3D|2D, 3D|2D, 3D|1D, 2D, 3D|
Table of supported cases by instance factory with WMMA instruction: