From 7b15c22e7e5e34da8baeee0d554e4f62830d37ff Mon Sep 17 00:00:00 2001 From: Kevin Abraham Date: Tue, 13 Jan 2026 08:16:04 +0000 Subject: [PATCH] fixed prefetch stage gemm --- ...uped_conv_fwd_multiple_d_wmma_cshuffle.hpp | 32 +++++++++---------- 1 file changed, 16 insertions(+), 16 deletions(-) diff --git a/experimental/builder/include/ck_tile/builder/reflect/conv_traits_device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp b/experimental/builder/include/ck_tile/builder/reflect/conv_traits_device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp index 4ba06d5487..846f3899d4 100644 --- a/experimental/builder/include/ck_tile/builder/reflect/conv_traits_device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp +++ b/experimental/builder/include/ck_tile/builder/reflect/conv_traits_device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp @@ -63,22 +63,22 @@ constexpr ConvTraits instance_to_conv_traits() .dst_scalar_per_vector_k1 = InstTraits::kBBlockTransferDstScalarPerVectorK1, .lds_padding = static_cast(InstTraits::kBBlockLdsExtraN)}}, - .warp_gemm = {.gemm_m = InstTraits::kMPerWmma, - .gemm_n = InstTraits::kNPerWmma, - .m_iter = InstTraits::kMRepeat, - .n_iter = InstTraits::kNRepeat}, - .c_tile_transfer = {.shuffle_params = {.m_gemms_per_shuffle = - InstTraits::kCShuffleMRepeatPerShuffle, - .n_gemms_per_shuffle = - InstTraits::kCShuffleNRepeatPerShuffle}, - .thread_cluster_dims = {InstTraits::kCDEThreadClusterLengths[0], - InstTraits::kCDEThreadClusterLengths[1], - InstTraits::kCDEThreadClusterLengths[2], - InstTraits::kCDEThreadClusterLengths[3]}, - .scalar_per_vector = InstTraits::kCDEBlockTransferScalarPerVector}, - // .num_gemm_prefetch_stage = InstTraits::kNumGemmKPrefetchStage, - .pipeline_version = get_pipeline_version(), - .pipeline_scheduler = get_pipeline_scheduler(), + .warp_gemm = {.gemm_m = InstTraits::kMPerWmma, + .gemm_n = InstTraits::kNPerWmma, + .m_iter = InstTraits::kMRepeat, + .n_iter = InstTraits::kNRepeat}, + .c_tile_transfer = {.shuffle_params = {.m_gemms_per_shuffle = + InstTraits::kCShuffleMRepeatPerShuffle, + .n_gemms_per_shuffle = + InstTraits::kCShuffleNRepeatPerShuffle}, + .thread_cluster_dims = {InstTraits::kCDEThreadClusterLengths[0], + InstTraits::kCDEThreadClusterLengths[1], + InstTraits::kCDEThreadClusterLengths[2], + InstTraits::kCDEThreadClusterLengths[3]}, + .scalar_per_vector = InstTraits::kCDEBlockTransferScalarPerVector}, + .num_gemm_prefetch_stage = InstTraits::kNumGemmKPrefetchStage, + .pipeline_version = get_pipeline_version(), + .pipeline_scheduler = get_pipeline_scheduler(), }; }