ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 06:49:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

Added wmma support for gemm quantization: (#2841)

Browse Source 
- profiler for gemm quantization for DL/XDL
- tests for gemm quantization for DL/XDL
- implementation for gemm quantization for WMMA
- profiler/tests for gemm qunatization for WMMA

Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
This commit is contained in:
Wojciech Laskowski
2025-09-17 01:23:29 +02:00
committed by GitHub
parent 2723dbd332
commit f97b2a3f5d
21 changed files with 1167 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma_cshuffle_v3_common.hpp
View File

@@ -3,6 +3,7 @@
#pragma once
#include <cstdint>
#include <iostream>
#include <sstream>
@@ -171,8 +172,8 @@ struct DeviceGemm_Wmma_CShuffleV3_Common
// other hand, Split K for 16-bit outputs uses packed atomics so ScalarPerVectors cannot
// be odd.
constexpr bool AtomicsImplementationExists =
!(std::is_same_v<EDataType, ck::half_t> ||
std::is_same_v<EDataType, ck::bhalf_t>) ||
!(std::is_same_v<EDataType, ck::half_t> || std::is_same_v<EDataType, ck::bhalf_t> ||
std::is_same_v<EDataType, int8_t>) ||
(CDEShuffleBlockTransferScalarPerVectors{}[0] % 2 == 0);
if(has_main_k_block_loop)