ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-17 11:30:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

Refactor CI performance tests. (#1726)

Browse Source 
* merge the build and performance tests CI stages together

* add gemm performance test on gfx11/gfx12

* add suffices to distinguish gemm performance logs from different archs

* use smaller gemm set in CI for gfx10/gfx11/gfx12

* disable performance tests on gfx1030

* fix the shashing logic

* fix finding python3 for mha instances

[ROCm/composable_kernel commit: 355893cdd8]
This commit is contained in:
Illia Silin
2024-12-06 13:04:25 -08:00
committed by GitHub
parent 7356ee494e
commit a6fe7d5f3f
8 changed files with 176 additions and 209 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
library/src/tensor_operation_instance/gpu/mha/CMakeLists.txt
View File

@@ -6,7 +6,7 @@ set(CK_TILE_SRC_FOLDER ${CMAKE_SOURCE_DIR}/include/ck_tile/)
# CK Codegen requires dataclass which is added in Python 3.7
# Python version 3.8 is required for general good practice as it is default for Ubuntu 20.04
if(NOT CK_USE_ALTERNATIVE_PYTHON)
find_package(PythonInterp 3 REQUIRED)
find_package(Python3 COMPONENTS Interpreter Development)
else()
message("Using alternative python version")
set(EXTRA_PYTHON_PATH)
@@ -33,7 +33,7 @@ set(FMHA_KNOWN_APIS "fwd,fwd_splitkv,fwd_appendkv,bwd")
# Note: The receipt 3 arg filters the generated backwards instances to reduce compilation time.
# With receipt 3 set, we are generating instances for datatype == {fp16 || bfp16}, bias == {no || alibi}, deterministic == off, and dpad == dvpad.
execute_process(
COMMAND ${PYTHON_EXECUTABLE} ${FMHA_SRC_FOLDER}/generate.py
COMMAND ${Python3_EXECUTABLE} ${FMHA_SRC_FOLDER}/generate.py
--list_blobs ${FMHA_CPP_FOLDER}/blob_list.txt
--api ${FMHA_KNOWN_APIS}
--receipt 3
@@ -50,7 +50,7 @@ endif()
# With receipt 3 set, we are generating instances for datatype == {fp16 || bfp16}, bias == {no || alibi}, deterministic == off, and dpad == dvpad.
add_custom_command(
OUTPUT ${FMHA_GEN_BLOBS}
COMMAND ${PYTHON_EXECUTABLE} ${FMHA_SRC_FOLDER}/generate.py
COMMAND ${Python3_EXECUTABLE} ${FMHA_SRC_FOLDER}/generate.py
--output_dir ${FMHA_CPP_FOLDER}
--api ${FMHA_KNOWN_APIS}
--receipt 3