ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-05 12:44:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit '1b66f3f4a32f1e755e8ac70a16e879f4f6523870' into develop

Browse Source
This commit is contained in:
assistant-librarian[bot]
2025-07-10 15:06:52 +00:00
parent 46b216e813
commit 4577a57769
3 changed files with 40 additions and 32 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

@@ -10,27 +10,11 @@ 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);
@@ -48,7 +32,7 @@ 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);
@@ -84,7 +68,7 @@ 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(