[BlockScale GEMM] FP8 Blockscale GEMM optimization and ckProfiler#1913

Merged

carlushuang merged 18 commits into

f8blockscale_opt

Feb 25, 2025

Commits on Dec 20, 2024

Added two kernel for M=32 problem
mtgu0705
committed
Comment the first one
mtgu0705
committed

Commits on Dec 23, 2024

Enable multiply_multiply for Scale_Block_M = 1 for deepseek
mtgu0705
committed

Commits on Dec 25, 2024

Modify the a_thread offset since the A data load is different from B.
mtgu0705
committed

Commits on Dec 26, 2024

Commits on Jan 10, 2025

enable blockwise pipelie v1 and v2. v1 is work for small K.
mtgu0705
committed

Commits on Jan 14, 2025

add instance for gemm_ab_scale
mtgu0705
committed
fix cmakelist of ckProfiler
mtgu0705
committed

Commits on Feb 18, 2025

Merge branch 'develop' of https://github.com/ROCm/composable_kernel into f8blockscale_opt
aska-0096
committed

Commits on Feb 19, 2025

optimize blockscale gemm. todo: reduce vgpr usage
aska-0096
committed

Commits on Feb 21, 2025

fix a correctness bug
aska-0096
committed