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

Fix the group of quantization_int8 kernels on MI300. (#695)

Browse Source 
* replace amd_buffer_atomic_add with hip_atomic_add

* fix grouped_gemm_splitk kernels on mi300

* fix syntax

* revert experimental atomic_add changes

* fix the group of kernels from ticket 723 on MI300

---------

Co-authored-by: Jing Zhang <jizhan@amd.com>
This commit is contained in:
Illia Silin
2023-05-03 16:27:04 -07:00
committed by GitHub
parent 4a51d2da9d
commit b8635a25b2
2 changed files with 6 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -51,7 +51,7 @@ __global__ void
const Block2CTileMap block_2_ctile_map)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) || \
defined(__gfx90a__) || defined(__gfx1030__))
defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx1030__))
constexpr index_t shared_block_size =
GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(ABDataType);
@@ -552,7 +552,8 @@ struct DeviceGemmMultipleD_Dl : public DeviceGemmMultipleD<ALayout,
static bool IsSupportedArgument(const Argument& arg)
{
if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx908" ||
ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx1030")
ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx1030" ||
ck::get_device_name() == "gfx940")
{
return GridwiseGemm::CheckValidity(
arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.e_grid_desc_m_n_);