mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-07-16 16:51:26 +00:00
d5c9215064
[CK_TILE][GFX1250] Enable MX GEMM FLATMM with ASYNC ## Motivation Enables MX GEMM FLATMM pipeline on gfx1250. The pipeline uses an async load instruction for tensor A, which complements the existing MX GEMM FLATMM pipeline with TDM load. At this time, only FLATMM MX pipelines are enabled on gfx1250. ## Technical Details The existing gfx950 implementation was extended to support gfx1250 architecture. All three MX FP data types are supported across the two ASICs. It should be noted that while the TDM pipeline uses an emulated 32x32x128 warp-tile instruction, the present submission relies on the built-in 16x16x128 instruction, called 4 times per warp. ## Test Plan Existing `test/ck_tile/flatmm` tests were extended to cover new gfx1250 functionality. To help facilitate the testing in development, `example/ck_tile/18_flatmm/script/smoke_test_mx.sh` script was introduced to verify various combinations of supported data types and pipeline versions. ## Test Result The present submission is expected to work on both gfx950 and gfx1250 hardware for all reasonable sizes and all MX FP8/FP6/FP4 data types. ## Submission Checklist - [x] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests. - [x] Relies on #6978 and should only be merged after the changes are merged to the `develop`.
50 lines
2.3 KiB
CMake
50 lines
2.3 KiB
CMake
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
|
# SPDX-License-Identifier: MIT
|
|
|
|
set(SUPPORTED_GPUS gfx908 gfx90a gfx942 gfx950 gfx1201 gfx1250)
|
|
|
|
set(has_supported_gpu FALSE)
|
|
foreach(gpu IN LISTS GPU_TARGETS)
|
|
if(gpu IN_LIST SUPPORTED_GPUS)
|
|
set(has_supported_gpu TRUE)
|
|
break()
|
|
endif()
|
|
endforeach()
|
|
|
|
if(has_supported_gpu)
|
|
# NOTE: we turn off undefined-func-template to let source compile without explicit declare function specializations
|
|
# ... because they are auto-generated
|
|
set(EXAMPLE_FLATMM_COMPILE_OPTIONS -Wno-undefined-func-template)
|
|
set(EXAMPLE_MOE_FLATMM_COMPILE_OPTIONS)
|
|
|
|
if(CK_USE_OCP_FP8)
|
|
list(APPEND EXAMPLE_FLATMM_COMPILE_OPTIONS -DCK_TILE_USE_OCP_FP8)
|
|
endif()
|
|
list(APPEND EXAMPLE_FLATMM_COMPILE_OPTIONS "SHELL: -mllvm -greedy-reverse-local-assignment=1")
|
|
add_executable(tile_example_flatmm_basic flatmm_basic.cpp)
|
|
target_compile_options(tile_example_flatmm_basic PRIVATE ${EXAMPLE_FLATMM_COMPILE_OPTIONS})
|
|
|
|
add_executable(tile_example_moe_flatmm moe_flatmm.cpp)
|
|
target_compile_options(tile_example_moe_flatmm PRIVATE ${EXAMPLE_FLATMM_COMPILE_OPTIONS})
|
|
|
|
add_executable(tile_example_grouped_flatmm grouped_flatmm.cpp)
|
|
target_compile_options(tile_example_grouped_flatmm PRIVATE ${EXAMPLE_FLATMM_COMPILE_OPTIONS})
|
|
|
|
if(GPU_TARGETS MATCHES "gfx94|gfx95|gfx12")
|
|
add_executable(tile_example_mixed_prec_flatmm mixed_prec/mixed_prec_flatmm.cpp)
|
|
target_compile_options(tile_example_mixed_prec_flatmm PRIVATE ${EXAMPLE_FLATMM_COMPILE_OPTIONS})
|
|
|
|
add_executable(tile_example_a16w4_moe_flatmm mixed_prec/a16w4_moe_flatmm.cpp)
|
|
target_compile_options(tile_example_a16w4_moe_flatmm PRIVATE ${EXAMPLE_FLATMM_COMPILE_OPTIONS})
|
|
endif()
|
|
if (GPU_TARGETS MATCHES "gfx95" OR GPU_TARGETS MATCHES "gfx125")
|
|
include(mxgemm/mx_flatmm_instance.cmake)
|
|
mx_flatmm_instance_generate(EXAMPLE_MX_FLATMM_FILES)
|
|
message(STATUS "Generated MX FlatMM kernel files: ${EXAMPLE_MX_FLATMM_FILES}")
|
|
|
|
add_executable(tile_example_mx_flatmm mxgemm/mx_flatmm.cpp ${EXAMPLE_MX_FLATMM_FILES})
|
|
target_include_directories(tile_example_mx_flatmm PRIVATE mxgemm)
|
|
target_compile_options(tile_example_mx_flatmm PRIVATE ${EXAMPLE_FLATMM_COMPILE_OPTIONS})
|
|
endif()
|
|
endif()
|