composable_kernel

composable_kernel copied to clipboard

Numerical errors found in convolution backward data kernel using detetctron2 data

3

We found numerical errors in convolution backward data kernel when running test with the detectron2 data. I put the data and the error log file [here.](http://ginger.amd.com/share/detectron2_data/) . The log contains...

iq136boy

Refactor ck_tile fMHA forward example

2

poyenc

Check window lengths in tile_window_with_static_distribution<> ctor

1

poyenc

zjing14

[Don't merge] Refactor grouped convolution

2

- Remove 1d and 2d grouped convolution instances - Move MakeDescriptor.... functions to transform headers

bartekxk

[GEMM] Set enable-post-misched=0 as default compile option

5

This PR set "-mllvm -enable-post-misched=0" as default compile option of CK. This option will improve gemm_universal performance and ensure the correctness.

aska-0096

Refactor f8_t and bf8_t as custom types, enable use of custom types

2

- Refactor f8_t and bf8_t - Update conversion methods - Update load method - Add dynamic buffer custom types support - Update threadwise conversion Right now custom types are supported...

geyyer

zjing14

### Problem Description // clang-format off using DeviceGemmInstance1 = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle < ALayout, // ALayout BLayout, // BLayout CLayout, // CLayout ADataType, // ADataType BDataType, // BDataType CDataType, // CDataType AccDataType,...

xiaobo1025

zjing14