ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-05 06:01:23 +00:00
Code Issues Packages Projects Releases Wiki Activity

enable hotloop

Browse Source
This commit is contained in:
root
2025-08-21 09:46:52 -05:00
parent c378e9bdf8
commit 65989e940c
4 changed files with 39 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
include/ck_tile/ops/moe_flatmm/pipeline/moe_flatmm_pipeline_agmem_bgmem_creg.hpp
View File

@@ -78,7 +78,9 @@ struct MoeFlatmmPipelineAGmemBGmemCRegV1
static constexpr index_t MPerBlockPerIter = kMPerBlock / MIterPerWarp;
static constexpr index_t KPerBlockPerIter = kKPerBlock / KIterPerWarp;
static constexpr index_t K1 = Problem::VectorLoadSize / sizeof(ADataType);
static constexpr int MXFP4PackedSize = 2;
static constexpr index_t AK1 = Problem::VectorLoadSize / sizeof(ADataType);
static constexpr index_t BK1 = Problem::VectorLoadSize / sizeof(BDataType) * MXFP4PackedSize;
static constexpr index_t m_preload = (MIterPerWarp * KIterPerWarp >= DsReadPreload)
? DsReadPreload
: MIterPerWarp * KIterPerWarp;
@@ -100,9 +102,14 @@ struct MoeFlatmmPipelineAGmemBGmemCRegV1
static constexpr index_t dswrite_rep = (dswrite_num_perK + MIterPerWarp - 1) / MIterPerWarp;
static constexpr index_t Aload_num_perK = dswrite_num_perK;
static constexpr index_t Aload_rep = dswrite_rep;
static constexpr index_t Bload_num_perK = kNPerBlock * WG::kK / NWarp / K1 / WaveSize;
static constexpr index_t HalfMIter = (MIterPerWarp + 1) / 2;
static constexpr index_t Bload_rep = (Bload_num_perK + HalfMIter - 1) / HalfMIter;
static constexpr index_t Bload_num_perK = kNPerBlock * WG::kK / NWarp / BK1 / WaveSize;
// static constexpr index_t ScaleBload_K1 = ContinuousScaleNPerThread * ContinuousScaleKPerThread;
// static constexpr index_t ScaleBload_num =
// kNPerBlock * kKPerBlock / NWarp / 32 / ScaleBload_K1 /
// WaveSize; // BlockN * BlockK / NWarp / ScalePerK / ScaleB_K1 / wavesize
// static constexpr index_t KPerScaleLoad = KIterPerWarp / ScaleBload_num;
static constexpr index_t HalfMIter = (MIterPerWarp + 1) / 2;
static constexpr index_t Bload_rep = (Bload_num_perK + HalfMIter - 1) / HalfMIter;
static constexpr index_t mfma_perM_perK = NIterPerWarp * mfma_per_wg;
static constexpr index_t dswrite_mIter = (DsWritePreIssue - 1) % MIterPerWarp;
@@ -342,6 +349,10 @@ struct MoeFlatmmPipelineAGmemBGmemCRegV1
? Aload_rep
: 0;
}
// if((kIter % KPerScaleLoad == 0) && (mIter == 0))
// {
// load_perM = load_perM + 1;
// }
SchedulerPerM(dsread_perM, dswrite_perM, load_perM);
}
}