ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 21:21:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

fixed fp8 issues (#894)

Browse Source 
* fixed fp8 init; and reference gemm

* Update host_tensor_generator.hpp

* fixed convert

* fixed reference gemm

* fixed comments

* fixed comments

* fixed ci

* fixed computeType

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
This commit is contained in:
zjing14
2023-09-12 22:17:56 -05:00
committed by GitHub
parent 74d32f0719
commit a66d14edf2
6 changed files with 45 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
View File

@@ -144,7 +144,8 @@ template <typename ALayout,
typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
index_t CDEBlockTransferScalarPerVector_NPerBlock,
LoopScheduler LoopSched = make_default_loop_scheduler(),
PipelineVersion PipelineVer = PipelineVersion::v1>
PipelineVersion PipelineVer = PipelineVersion::v1,
typename ComputeDataType = EDataType>
struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
BLayout,
DsLayout,
@@ -243,11 +244,9 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}, {}))>;
using EGridDesc_M_N = decltype(MakeEGridDescriptor_M_N<ELayout>(1, 1, 1));
using ComputeDataType = EDataType;
// GridwiseGemm
using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
ADataType, // TODO: distinguish A/B datatype
ADataType,
BDataType,
ComputeDataType,
AccDataType,