From 1228d112efa5d99d13914aee552fd41553ad3521 Mon Sep 17 00:00:00 2001 From: mtgu0705 Date: Wed, 19 Mar 2025 04:36:59 +0000 Subject: [PATCH] modified the gemm xlop type from 32x32x16 to 16x16x32 --- example/65_gemm_multiply_multiply/moe_gemm1_xdl_pk_i4.cpp | 6 +++--- .../gpu/block/blockwise_gemm_pipeline_xdlops_base.hpp | 4 +++- 2 files changed, 6 insertions(+), 4 deletions(-) diff --git a/example/65_gemm_multiply_multiply/moe_gemm1_xdl_pk_i4.cpp b/example/65_gemm_multiply_multiply/moe_gemm1_xdl_pk_i4.cpp index c06e595c0f..0b67da832a 100644 --- a/example/65_gemm_multiply_multiply/moe_gemm1_xdl_pk_i4.cpp +++ b/example/65_gemm_multiply_multiply/moe_gemm1_xdl_pk_i4.cpp @@ -171,11 +171,11 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceMoeGemm< AElementOp, BElementOp, CDEElementOp, GemmSpec, 256, MPerBlock, 128, 128, 16, 32, - 32, 32, - 4, 1, + 16, 16, + 8, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 32, 32, 0, - 1, 1, S<1, 32, 1, 8>, S<8, 1, 1>, + 2, 2, S<1, 32, 1, 8>, S<8, 1, 1>, ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v1, Nswizzle, true, A0DataType>; // clang-format on #endif diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_base.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_base.hpp index 45ed6845c2..78e4a3dce5 100644 --- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_base.hpp +++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_base.hpp @@ -46,7 +46,9 @@ struct BlockwiseGemmXdlops_pipeline_base static constexpr index_t A_K0 = ATileDesc{}.GetLength(I0); static constexpr index_t B_K0 = BTileDesc{}.GetLength(I0); static constexpr index_t A_K1 = ATileDesc{}.GetLength(I2); - static constexpr index_t B_K1 = BTileDesc{}.GetLength(I2); + // static constexpr index_t B_K1 = BTileDesc{}.GetLength(I2); + static constexpr index_t B_K1 = + BTileDesc{}.GetLength(Number < BTileDesc{}.GetNumOfDimension() == 4 ? 3 : 2 > {}); static constexpr auto xdlops_gemm = XdlopsGemm{};