ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-19 12:30:16 +00:00
Code Issues Packages Projects Releases Wiki Activity

Wave Tile Transfer supporting global load with transpose (#3027)

Browse Source 
* Initial implementation:

 - add new thread group transfer supporting transpose instruction
 - refactor AB transfer to switch between thread and wave tiles methods

* Add some comments and remove explicit wave and lane calculations

* Remove compiler option for performance

* fp16 example: use tuned instance

* Missing cleanup

* Integrate wave transfer in existing gemm and batched gemm instances

* Add fast instances

* extend implementation for 8 bit datatypes

packed types not supported

* Address review comments

* Optimize pipeline v1 and re-introduce compiler option

* Disable wave tile approach for b scale gemm

* Fix for clang20

* Avoid code duplication of amd_global_load_transpose_to_vgpr function

[ROCm/composable_kernel commit: 440358c168]
This commit is contained in:
Enrico Degregori
2025-10-16 20:33:56 +02:00
committed by GitHub
parent 06d76b160e
commit 1d9320c8f3
15 changed files with 1513 additions and 720 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_wmma_universal_f16_f16_f16/device_gemm_wmma_universal_f16_f16_f16_km_kn_mn.hpp
View File

@@ -44,6 +44,7 @@ using device_gemm_wmma_universal_f16_f16_f16_km_kn_mn_comp_instances = std::tupl
DeviceGemm_Wmma_CShuffleV3< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 32, 64, 64, 8, 8, 16, 16, 2, 2, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 0, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 0, 1, 1, S<1, 16, 1, 4>, 8, Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 32, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Interwave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Interwave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 64, 8, 8, 16, 16, 2, 8, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 64, 1, 4>, 8, Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 32, 64, 64, 8, 8, 16, 16, 2, 2, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 0, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 0, 1, 1, S<1, 16, 1, 4>, 8, Interwave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 32, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Intrawave, BlockGemmPipelineVersion::v3>,
DeviceGemm_Wmma_CShuffleV3< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Intrawave, BlockGemmPipelineVersion::v3>,

1
library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_wmma_universal_f16_f16_f16/device_gemm_wmma_universal_f16_f16_f16_km_nk_mn.hpp
View File

@@ -42,6 +42,7 @@ using device_gemm_wmma_universal_f16_f16_f16_km_nk_mn_comp_instances = std::tupl
DeviceGemm_Wmma_CShuffleV3< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 32, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 64, 8, 8, 16, 16, 4, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 160, 64, 8, 8, 16, 16, 2, 5, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 4>, 8, Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 64, 8, 8, 16, 16, 2, 8, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 4>, 8, Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 80, 64, 8, 8, 16, 16, 1, 5, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 64, 1, 2>, 8, Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 32, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Interwave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Interwave, BlockGemmPipelineVersion::v1>,

1
library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_wmma_universal_f16_f16_f16/device_gemm_wmma_universal_f16_f16_f16_mk_kn_mn.hpp
View File

@@ -49,6 +49,7 @@ using device_gemm_wmma_universal_f16_f16_f16_mk_kn_mn_comp_instances = std::tupl
DeviceGemm_Wmma_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 128, 32, 8, 8, 16, 16, 4, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, Interwave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 32, 64, 64, 8, 8, 16, 16, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 0, 1, 1, S<1, 16, 1, 4>, 8, Interwave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 64, 32, 8, 8, 16, 16, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 0, 1, 1, S<1, 32, 1, 2>, 8, Interwave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 64, 8, 8, 16, 16, 2, 8, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 64, 1, 4>, 8, Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 32, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Intrawave, BlockGemmPipelineVersion::v3>,
DeviceGemm_Wmma_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Intrawave, BlockGemmPipelineVersion::v3>,
DeviceGemm_Wmma_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 128, 32, 8, 8, 16, 16, 4, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, Intrawave, BlockGemmPipelineVersion::v3>,

1
library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_wmma_universal_f16_f16_f16/device_gemm_wmma_universal_f16_f16_f16_mk_nk_mn.hpp
View File

@@ -51,6 +51,7 @@ using device_gemm_wmma_universal_f16_f16_f16_mk_nk_mn_comp_instances = std::tupl
DeviceGemm_Wmma_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 160, 64, 8, 8, 16, 16, 2, 5, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 4>, 8, Interwave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 128, 32, 8, 8, 16, 16, 4, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, Interwave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 32, 64, 64, 8, 8, 16, 16, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 8, Interwave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 64, 8, 8, 16, 16, 2, 8, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 4>, 8, Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGemm_Wmma_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 32, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Intrawave, BlockGemmPipelineVersion::v3>,
DeviceGemm_Wmma_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 64, 64, 8, 8, 16, 16, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 4>, 8, Intrawave, BlockGemmPipelineVersion::v3>,
DeviceGemm_Wmma_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 160, 64, 8, 8, 16, 16, 2, 5, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 4>, 8, Intrawave, BlockGemmPipelineVersion::v3>,