ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-11 17:00:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

Revert "Add templates for fp16 and unsigned short atomic add to fix FBGEMM bu…" (#2474)

Browse Source 
This reverts commit 112b47e885.
This commit is contained in:
Illia Silin
2025-07-08 19:01:26 -07:00
committed by GitHub
parent 112b47e885
commit 93420ecf89
2 changed files with 22 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
library/include/ck/library/tensor_operation_instance/gpu/gemm_universal_preshuffle.inc
View File

@@ -10,11 +10,27 @@ namespace instance {
#if(defined(CK_ENABLE_BF16) && defined(CK_ENABLE_FP8))
using GemmF8F8BF16InstanceVector = std::vector<std::unique_ptr<
DeviceGemmV2BPreshuffle<Row, Col, Row, F8, F8, BF16, PassThrough, PassThrough, PassThrough>>>&;
using GemmF8F8BF16InstanceVector =
std::vector<std::unique_ptr<DeviceGemmV2BPreshuffle<Row,
Col,
Row,
F8,
F8,
BF16,
PassThrough,
PassThrough,
PassThrough>>>&;
using GemmF8F8F16InstanceVector = std::vector<std::unique_ptr<
DeviceGemmV2BPreshuffle<Row, Col, Row, F8, F8, F16, PassThrough, PassThrough, PassThrough>>>&;
using GemmF8F8F16InstanceVector =
std::vector<std::unique_ptr<DeviceGemmV2BPreshuffle<Row,
Col,
Row,
F8,
F8,
F16,
PassThrough,
PassThrough,
PassThrough>>>&;
void add_device_gemm_xdl_universal_preshuffle_f8_f8_bf16_mk_mfma32x32_mn_instances(
GemmF8F8BF16InstanceVector& instances);
@@ -32,7 +48,7 @@ void add_device_gemm_xdl_universal_preshuffle_f8_f8_bf16_mk_mfma_mn_p3_instances
GemmF8F8BF16InstanceVector& instances);
void add_device_gemm_xdl_universal_preshuffle_f8_f8_bf16_mk_mfma_mn_p4_instances(
GemmF8F8BF16InstanceVector& instances);
GemmF8F8BF16InstanceVector& instances);
void add_device_gemm_xdl_universal_preshuffle_f8_f8_bf16_mk_mfma_mn_p5_instances(
GemmF8F8BF16InstanceVector& instances);
@@ -68,7 +84,7 @@ void add_device_gemm_universal_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_compute_defau
GemmF8F8F16InstanceVector& instances);
void add_device_gemm_universal_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p1_default_instances(
GemmF8F8F16InstanceVector& instances);
GemmF8F8F16InstanceVector& instances);
void add_device_gemm_universal_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p2_default_instances(
GemmF8F8F16InstanceVector& instances);
void add_device_gemm_universal_preshuffle_xdl_f8_f8_f16_mk_mfma_mn_p3_default_instances(