ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-01 12:17:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix build

Browse Source
This commit is contained in:
felix
2025-10-25 13:43:51 +00:00
parent 69bbe0480b
commit 37fdb18e06
2 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
example/67_gemm_microscaling/moe_gemm1_xdl_mx_fp4_bpreshuffle.cpp
View File

@@ -178,7 +178,7 @@ static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecializatio
constexpr ck::index_t DataPackedSize = 2; // Packed representation of data
constexpr ck::index_t ScaleBlockSize = 32; // scaling block size
constexpr ck::index_t KPerBlock = 128 / DataPackedSize; // 256 f4 = 128 fp4x2
constexpr ck::index_t KPerBlock = 256 / DataPackedSize; // 256 f4 = 128 fp4x2
static constexpr ck::index_t Nswizzle = false;
static constexpr ck::index_t ActOP = 0; // 0: gelu_and_mul, 1: silu_and_mul
static constexpr ck::index_t MPerBlock = 32;

4
include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_b_preshuffle_mx_moe_gufusion_v3.hpp
View File

@@ -726,8 +726,8 @@ struct BlockwiseGemmXdlops_pipeline_bpreshuffle_mx_moe_gufusion_v3<
});
});
});
HotLoopScheduler();
if constexpr(MPerBlock >= 64)
HotLoopScheduler();
__builtin_amdgcn_sched_barrier(0);
};