composable_kernel

composable_kernel copied to clipboard

[Issue]: Cannot compile composable_kernel code without -O2 optimization (inline assembly 'n' constraint requires immediate)

2

### Problem Description During compilation with -O2, the following error is observed multiple times in different files: `error: constraint 'n' expects an integer constant expression`. ### Operating System CentOS Stream...

imyixinw

[Issue]: Very slow perf for Gemm BF16

9

To reproduce: **Command:** `./bin/ckProfiler gemm 2 1 1 2 0 1 32 512 7168 -1 -1 -1 3 100` **GPU Type:** MI300x **Searched Perf:** `Best Perf for datatype = bf16...

ghostplant

[Issue]: [CK_TILE] enhancement of slice_tile related API

2

### Problem Description as reported in PR https://github.com/ROCm/composable_kernel/pull/2338 currently our slice tile API (more specifically is the internal API `detail::slice_distribution_from_x()`) has limitation  we need relax the restriction that the...

carlushuang

The documentation in this repository is currently insufficient for both new and existing contributors. Key areas lack clear explanations, usage examples, and detailed descriptions of core concepts. The following improvements...

vidyasagar-amd

Add IO module for profiler

2

## Problem Description ck_profiler outputs its results to std::cout. Handling the output down stream therefore relies on string parsing which can be error prone and difficult to evolve. ## Suggested...

smedegaard

MidLargeM Test Case Failure for StreamK GEMM bf16/fp16/fp8

1

This test case gives validation error for Stream-K GEMM and should be investigated !!! It is removed from previous PRs for this purpose. 1- https://github.com/ROCm/composable_kernel/pull/2101/files#diff-500d7a966171b800317eb1fdf32a618bd52ac6815e59d5fde63447b3de373170 2- https://github.com/ROCm/composable_kernel/commit/7142d8003c6a99f952a62bbd0b90d5f0261fc807

ozturkosu