From f0e3f93f1d64e4bef99e46f662d1041999a485e2 Mon Sep 17 00:00:00 2001 From: "assistant-librarian[bot]" <210906412+assistant-librarian[bot]@users.noreply.github.com> Date: Tue, 17 Feb 2026 17:58:11 -0700 Subject: [PATCH] [CK] Add new fwd conv fp16/bf16 instances optimized for unit group size. (#4275) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit ## Proposed changes Added new FP16/BF16 instances that are optimized for group size = 1. The new instance use the compute optimized block GEMM pipeline. | CK prof command | Baseline (TFLOPs) | New V3 instances (TFLOPs) | |:-----|:------:|------:| | grouped_conv_fwd 1 1 1 0 1 0 1 2 1 32 2376 256 3 3 100 100 1 1 1 1 1 1 1 1 | 858.818 | 962.293 | | grouped_conv_fwd 1 1 1 0 1 0 1 2 1 32 256 256 3 3 100 100 1 1 1 1 1 1 1 1 | 979.987 | 1121.11 | | grouped_conv_fwd 1 1 1 0 1 0 1 2 1 32 2376 256 3 3 50 50 1 1 1 1 1 1 1 1 | 945.951 | 1091.66 | --- 🔁 Imported from [ROCm/composable_kernel#3670](https://github.com/ROCm/composable_kernel/pull/3670) 🧑‍💻 Originally authored by @vpietila-amd --------- Co-authored-by: Ville Pietilä <> Co-authored-by: Bartłomiej Kocot Co-authored-by: Ville Pietilä <188998872+vpietila-amd@users.noreply.github.com> Co-authored-by: systems-assistant[bot] --- ...device_grouped_conv_fwd_xdl_comp_instance.hpp | 16 ++++++++++++++-- 1 file changed, 14 insertions(+), 2 deletions(-) diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp index 02c2eb1885..641f7dd18e 100644 --- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp +++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp @@ -63,11 +63,17 @@ template , typename OutElementOp = PassThrough> using device_grouped_conv_fwd_xdl_bf16_comp_instances_2x = std::tuple< - // clang-format off +// clang-format off //########################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| //########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| DataType| Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| //########################################| | | | | | | | | | | | Operation| Operation| Operation| | | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| //########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +#if defined(CK_USE_GFX950) + // Instances optimized for G=1 + // The use more shared memory than what is available for non gfx950 architectures. + DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>, + DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>, +#endif DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> // clang-format on >; @@ -131,11 +137,17 @@ template , typename OutElementOp = PassThrough> using device_grouped_conv_fwd_xdl_f16_comp_instances_2x = std::tuple< - // clang-format off +// clang-format off //########################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| //########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| DataType| Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| //########################################| | | | | | | | | | | | Operation| Operation| Operation| | | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| //########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +#if defined(CK_USE_GFX950) + // Instances optimized for G=1 + // The use more shared memory than what is available for non gfx950 architectures. + DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>, + DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>, +#endif DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> // clang-format on >;