Files
composable_kernel/example/67_gemm_microscaling/CMakeLists.txt
Andriy Roshchenko 518dc21ae8 MX GEMM - FP6 Support in GEMM MX v3 Pipeline (#2481)
* Add GEMM MX BF6 example

* Fix BF6 type_convert

* Add type_convert for bf16x6

* Add compare operator to f4x2_pk_t

* Update README for 67_gemm_microscaling

* Fix host tensor initialization with integer values for FP8
2025-07-11 13:07:05 -06:00

69 lines
3.4 KiB
CMake

add_custom_target(example_gemm_mx)
add_example_executable(example_gemm_mx_fp8 gemm_mx_fp8.cpp)
add_example_dependencies(example_gemm_mx example_gemm_mx_fp8)
add_example_executable(example_gemm_mx_bf8 gemm_mx_bf8.cpp)
add_example_dependencies(example_gemm_mx example_gemm_mx_bf8)
# TODO: Fix RRR
# add_example_executable(example_gemm_mx_fp8_bf8 gemm_mx_fp8_bf8.cpp)
# add_example_dependencies(example_gemm_mx example_gemm_mx_fp8_bf8)
add_example_executable(example_gemm_mx_fp6 gemm_mx_fp6.cpp)
add_example_dependencies(example_gemm_mx example_gemm_mx_fp6)
add_example_executable(example_gemm_mx_bf6 gemm_mx_bf6.cpp)
add_example_dependencies(example_gemm_mx example_gemm_mx_bf6)
add_example_executable(example_gemm_mx_fp4 gemm_mx_fp4.cpp)
add_example_dependencies(example_gemm_mx example_gemm_mx_fp4)
add_example_executable(example_gemm_mx_fp4_bpreshuffle gemm_mx_fp4_bpreshuffle.cpp)
add_example_dependencies(example_gemm_mx example_gemm_mx_fp4_bpreshuffle)
add_example_executable(example_moe_gemm1_xdl_mx_fp4_bns moe_gemm1_xdl_mx_fp4_bns.cpp)
add_example_dependencies(example_gemm_mx example_moe_gemm1_xdl_mx_fp4_bns)
add_example_executable(example_moe_gemm2_xdl_mx_fp4_bns moe_gemm2_xdl_mx_fp4_bns.cpp)
add_example_dependencies(example_gemm_mx example_moe_gemm2_xdl_mx_fp4_bns)
add_example_executable(example_moe_gemm1_xdl_mx_fp4 moe_gemm1_xdl_mx_fp4.cpp)
add_example_dependencies(example_gemm_mx example_moe_gemm1_xdl_mx_fp4)
add_example_executable(example_moe_gemm2_xdl_mx_fp4 moe_gemm2_xdl_mx_fp4.cpp)
add_example_dependencies(example_gemm_mx example_moe_gemm2_xdl_mx_fp4)
add_example_executable(example_moe_gemm1_xdl_mx_fp4_bpreshuffle moe_gemm1_xdl_mx_fp4_bpreshuffle.cpp)
add_example_dependencies(example_gemm_mx example_moe_gemm1_xdl_mx_fp4_bpreshuffle)
add_example_executable(example_moe_gemm2_xdl_mx_fp4_bpreshuffle moe_gemm2_xdl_mx_fp4_bpreshuffle.cpp)
add_example_dependencies(example_gemm_mx example_moe_gemm2_xdl_mx_fp4_bpreshuffle)
set(FP4_MXGEMM_OPTIONS)
list(APPEND FP4_MXGEMM_OPTIONS "SHELL: -mllvm -greedy-reverse-local-assignment=1 -mllvm --amdgpu-use-amdgpu-trackers=1")
example_compile_options(example_gemm_mx_fp4 PRIVATE ${FP4_MXGEMM_OPTIONS})
example_compile_options(example_gemm_mx_fp4_bpreshuffle PRIVATE ${FP4_MXGEMM_OPTIONS})
# mx moe B no-shuffling + scale shuffling
example_compile_options(example_moe_gemm1_xdl_mx_fp4_bns PRIVATE ${FP4_MXGEMM_OPTIONS})
example_compile_options(example_moe_gemm2_xdl_mx_fp4_bns PRIVATE ${FP4_MXGEMM_OPTIONS})
# mx moe B no-shuffling + scale shuffling (async loads)
example_compile_options(example_moe_gemm1_xdl_mx_fp4 PRIVATE ${FP4_MXGEMM_OPTIONS})
example_compile_options(example_moe_gemm2_xdl_mx_fp4 PRIVATE ${FP4_MXGEMM_OPTIONS})
# mx moe B shuffling + scale shuffling (async loads)
example_compile_options(example_moe_gemm1_xdl_mx_fp4_bpreshuffle PRIVATE ${FP4_MXGEMM_OPTIONS})
example_compile_options(example_moe_gemm2_xdl_mx_fp4_bpreshuffle PRIVATE ${FP4_MXGEMM_OPTIONS})
set(FP8_MXGEMM_OPTIONS)
list(APPEND FP8_MXGEMM_OPTIONS "SHELL: -mllvm -greedy-reverse-local-assignment=1 -mllvm --slp-threshold=-32")
example_compile_options(example_gemm_mx_fp8 PRIVATE ${FP8_MXGEMM_OPTIONS})
example_compile_options(example_gemm_mx_bf8 PRIVATE ${FP8_MXGEMM_OPTIONS})
set(FP6_MXGEMM_OPTIONS)
list(APPEND FP6_MXGEMM_OPTIONS -mavx512f)
example_compile_options(example_gemm_mx_fp6 PRIVATE ${FP6_MXGEMM_OPTIONS})
example_compile_options(example_gemm_mx_bf6 PRIVATE ${FP6_MXGEMM_OPTIONS})