ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-18 12:00:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

Extend XDL kernel to Support RDNA3/4 - Part 5 (#2725)

Browse Source 
* Enable xdl in gfx11 & gfx12

* update cmake file

* fix all instance build (cmake)

* fix batched_gemm_gemm(cmake)

* rebase cmake files

* fix cmake build error

* remve CK_ENABLE_DYNAMIC_WARP_SIZE

* update cmake build error2

* fix gfx11 build

CK_USE_XDL is enabled on gfx11 and gfx12

* fix gfx10 build

* fix gfx11 error

---------

Co-authored-by: Lin, Qun <Quentin.Lin+amdeng@amd.com>

[ROCm/composable_kernel commit: f22740df82]
This commit is contained in:
linqunAMD
2025-09-16 01:59:25 +08:00
committed by GitHub
parent 954db22b39
commit 71dc8a9d4d
33 changed files with 243 additions and 397 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
library/include/ck/library/tensor_operation_instance/gpu/gemm_multiply_multiply.hpp
View File

@@ -638,7 +638,7 @@ struct DeviceOperationInstanceFactory<DeviceGemmMultipleDSplitK<ALayout,
std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
#ifdef CK_USE_XDL
#ifdef CK_ENABLE_FP8
#if defined(CK_USE_FP8_ON_UNSUPPORTED_ARCH) || CK_USE_OCP_FP8 || defined(CK_USE_GFX94)
#ifdef CK_ENABLE_BF16
if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, f8_t> &&
is_same_v<CDataType, bhalf_t>)