Anthony Chang 4c1d1a8e59 remove atten kernel workarounds as we move over to rocm 5.3 (#496) ...

[ROCm/composable_kernel commit: 451f1e3d65]

2022-11-02 16:56:07 -06:00

..

block

Add pipeline v1/v2 selector, add more instances (#381)

2022-11-02 16:50:48 -06:00

device

Add client example of grouped conv2d backward data (data type: fp16) (#481)

2022-11-02 16:54:41 -06:00

element

Conv perlayer int8 quantization (#471)

2022-11-02 13:56:26 -06:00

grid

remove atten kernel workarounds as we move over to rocm 5.3 (#496)

2022-11-02 16:56:07 -06:00

thread

Batchnorm-forward implemented using welford method to calculate variance (#403)

2022-10-27 18:52:54 -06:00

warp

Input/output permutation for fused attention (#460)

2022-10-27 14:58:20 -06:00