mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-07-12 18:17:58 +00:00
Add v3 support. Add compile_option
This commit is contained in:
@@ -16,3 +16,7 @@ foreach(gpu IN LISTS GPU_TARGETS)
|
||||
set(target 1)
|
||||
endif()
|
||||
endforeach()
|
||||
set(EXAMPLE_COMPILE_OPTIONS)
|
||||
list(APPEND EXAMPLE_COMPILE_OPTIONS -v --save-temps -Wno-gnu-line-marker)
|
||||
list(APPEND EXAMPLE_COMPILE_OPTIONS "SHELL: -mllvm -greedy-reverse-local-assignment=1 -mllvm --schedmodel=0 -mllvm -misched=gcn-iterative-max-occupancy-experimental")
|
||||
target_compile_options(example_moe_gemm2_xdl_fp8_blockscale PRIVATE ${EXAMPLE_COMPILE_OPTIONS})
|
||||
@@ -158,11 +158,11 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceMoeGemmBlockScale<
|
||||
MPerBlock, 128, 128,
|
||||
16, 16,
|
||||
32, 32,
|
||||
2, 2,
|
||||
4, 1,
|
||||
S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0,
|
||||
S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0,
|
||||
2, 1, S<1, 8, 1, 32>, S<2, 1, 1, 1>,
|
||||
ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v1, false, false, A0DataType>;
|
||||
1, 1, S<1, 32, 1, 8>, S<2, 1, 1, 1>,
|
||||
ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v3, false, false, A0DataType>;
|
||||
#endif
|
||||
// clang-format on
|
||||
|
||||
|
||||
Reference in New Issue
Block a user