From 468a48199028f64d53e2590e9a674da8bbdde900 Mon Sep 17 00:00:00 2001 From: jefyang1 <146495389+jefyang1@users.noreply.github.com> Date: Wed, 12 Feb 2025 09:46:38 -0800 Subject: [PATCH] Fix KPack and enable existing instances on gfx950 (#1871) [ROCm/composable_kernel commit: 7b826807cd2c7fb051e11561d72e8a7f556ec2d4] --- ...iple_d_welford_first_half_xdl_cshuffle.hpp | 11 +++++-- ...wise_batched_gemm_gemm_xdl_cshuffle_v1.hpp | 10 +++++- ...iple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp | 32 +++++++++++++++---- ...ultiple_d_softmax_gemm_xdl_cshuffle_v1.hpp | 24 +++++++++++--- ...ched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp | 10 +++++- ...e_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp | 10 +++++- ...emm_multiple_d_multiple_r_xdl_cshuffle.hpp | 10 +++++- ...se_gemm_multiple_d_xdl_splitk_cshuffle.hpp | 13 ++++++-- .../gridwise_gemm_reduce_xdl_cshuffle_v1.hpp | 10 +++++- ...e_gemm_split_k_multiple_d_xdl_cshuffle.hpp | 22 ++++++++++--- ...emm_split_k_multiple_d_xdl_cshuffle_v2.hpp | 13 ++++++-- ...idwise_gemm_xdl_cshuffle_bwd_weight_v3.hpp | 11 +++++-- .../gridwise_gemm_xdl_cshuffle_streamk_v3.hpp | 11 +++++-- .../grid/gridwise_gemm_xdl_cshuffle_v2.hpp | 13 ++++++-- .../grid/gridwise_gemm_xdl_cshuffle_v3.hpp | 11 +++++-- .../gridwise_gemm_xdl_cshuffle_v3_b_scale.hpp | 17 +++++++--- ...ridwise_gemm_xdl_layernorm_cshuffle_v1.hpp | 10 +++++- ...ridwise_gemm_xdl_waveletmodel_cshuffle.hpp | 13 ++++++-- .../gpu/grid/gridwise_gemm_xdlops_v3r1.hpp | 10 +++++- ...conv_bwd_weight_two_stage_xdl_instance.hpp | 32 +++++++------------ ...m_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp | 5 ++- ...m_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp | 5 ++- ...m_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp | 10 +++--- ...m_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp | 10 +++--- ...6_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp | 12 +++---- ...f16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp | 7 ++-- ...6_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp | 7 ++-- ...f16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp | 7 ++-- ...6_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp | 11 +++---- ..._c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp | 28 +++++++--------- ..._bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp | 28 +++++++--------- ...s_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp | 28 +++++++--------- ..._2_stage_f16_f16_f16_mk_nk_mn_instance.cpp | 5 ++- ..._shuffle_f16_f16_f16_km_kn_mn_instance.cpp | 5 ++- ..._shuffle_f16_f16_f16_km_nk_mn_instance.cpp | 5 ++- ..._shuffle_f16_f16_f16_mk_kn_mn_instance.cpp | 9 +++--- ..._shuffle_f16_f16_f16_mk_nk_mn_instance.cpp | 9 +++--- ...l_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp | 4 +-- ...m_kn_mn_interwave_pipeline_v1_instance.cpp | 3 +- ...regular_interwave_pipeline_v1_instance.cpp | 3 +- ...m_nk_mn_interwave_pipeline_v1_instance.cpp | 3 +- ...regular_interwave_pipeline_v1_instance.cpp | 3 +- ...k_kn_mn_interwave_pipeline_v1_instance.cpp | 3 +- ...regular_interwave_pipeline_v1_instance.cpp | 3 +- ...k_nk_mn_interwave_pipeline_v1_instance.cpp | 3 +- ...regular_interwave_pipeline_v1_instance.cpp | 3 +- ...16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp | 21 ++++++------ ...16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp | 21 ++++++------ ...16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp | 21 ++++++------ ...16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp | 19 +++++------ ...e_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp | 19 +++++------ ...e_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp | 19 +++++------ ...e_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp | 19 +++++------ ...e_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp | 19 +++++------ ..._layernorm_f16_km_kn_mn_mn_mn_instance.cpp | 10 ++---- ..._layernorm_f16_km_nk_mn_mn_mn_instance.cpp | 11 +++---- ..._layernorm_f16_mk_kn_mn_mn_mn_instance.cpp | 11 +++---- ..._layernorm_f16_mk_nk_mn_mn_mn_instance.cpp | 10 ++---- ..._shuffle_f16_f16_f16_km_kn_mn_instance.cpp | 15 ++++----- ..._shuffle_f16_f16_f16_km_nk_mn_instance.cpp | 15 ++++----- ..._shuffle_f16_f16_f16_mk_kn_mn_instance.cpp | 15 ++++----- ..._shuffle_f16_f16_f16_mk_nk_mn_instance.cpp | 15 ++++----- ...f16_f16_mk_kn_mn_v1_interwave_instance.cpp | 3 +- ...f16_f16_mk_kn_mn_v1_irregular_instance.cpp | 3 +- ...f16_f16_mk_nk_mn_v1_interwave_instance.cpp | 3 +- ..._xdl_universal_bf16_bf16_bf16_km_kn_mn.hpp | 6 ++-- ..._xdl_universal_bf16_bf16_bf16_km_nk_mn.hpp | 6 ++-- ..._xdl_universal_bf16_bf16_bf16_mk_kn_mn.hpp | 6 ++-- ..._xdl_universal_bf16_bf16_bf16_mk_nk_mn.hpp | 6 ++-- ...emm_xdl_universal_f16_f16_f16_mk_kn_mn.hpp | 6 ++-- ...emm_xdl_universal_f16_f16_f16_mk_nk_mn.hpp | 6 ++-- ...gemm_xdl_universal_f16_f8_f16_mk_kn_mn.hpp | 6 ++-- ...gemm_xdl_universal_f16_f8_f16_mk_nk_mn.hpp | 6 ++-- ..._xdl_universal_bf16_bf16_bf16_mk_kn_mn.hpp | 6 ++-- ...mm_xdl_universal_bf16_i8_bf16_mk_kn_mn.hpp | 5 ++- ...emm_xdl_universal_f16_f16_f16_mk_kn_mn.hpp | 6 ++-- ...universal_streamk_f16_f16_f16_mk_nk_mn.hpp | 6 ++-- ...f16_f8_f16_mk_kn_mn_irregular_instance.cpp | 7 ++-- ...le_loop_multiply_bf16_i8_bf16_mk_kn_mn.hpp | 15 ++++----- 79 files changed, 453 insertions(+), 421 deletions(-) mode change 100755 => 100644 include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_streamk_v3.hpp diff --git a/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp index 206ea00b9d..f4d0989088 100644 --- a/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp +++ b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp @@ -515,9 +515,16 @@ struct GridwiseGemmMultipleDWelfordFirstHalf_xdl_cshuffle // c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in // register // sanity check + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; constexpr index_t KPack = - math::max(math::lcm(AK1, BK1), - MfmaSelector::selected_mfma.k_per_blk); + math::max(lcm_AK1_BK1, + MfmaSelector:: + selected_mfma.k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector< BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp index 73bac20e43..55e254e015 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp @@ -448,8 +448,16 @@ struct GridwiseBatchedGemmGemm_Xdl_CShuffle // acc1[m][o] += acc[m][n] * B1[n][o] // sanity check + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; constexpr index_t KPack = math::max( - math::lcm(AK1, BK1), MfmaSelector::selected_mfma.k_per_blk); + lcm_AK1_BK1, + MfmaSelector::selected_mfma + .k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_v2< BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp index 355e0130f1..fd16927cc1 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp @@ -361,10 +361,18 @@ struct GridwiseBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle const auto M = d0_grid_desc_m_n.GetLength(I0); const auto N = d0_grid_desc_m_n.GetLength(I1); - constexpr auto mfma = - MfmaSelector::selected_mfma; - constexpr auto N3 = mfma.num_groups_per_blk; - constexpr auto N5 = mfma.group_size; + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + math::lcm(A0K1, B0K1) <= 4) + ? true + : false; + constexpr auto mfma = MfmaSelector::selected_mfma; + constexpr auto N3 = mfma.num_groups_per_blk; + constexpr auto N5 = mfma.group_size; return transform_tensor_descriptor( d0_grid_desc_m_n, make_tuple(make_unmerge_transform(make_tuple( @@ -643,9 +651,19 @@ struct GridwiseBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle // acc1[m][o] += acc[m][n] * B1[n][o] // sanity check - constexpr index_t KPack = math::max( - math::lcm(A0K1, B0K1), - MfmaSelector::selected_mfma.k_per_blk); + constexpr auto lcm_A0K1_B0K1 = math::lcm(A0K1, B0K1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_A0K1_B0K1 <= 4) + ? true + : false; + constexpr index_t KPack = + math::max(lcm_A0K1_B0K1, + MfmaSelector::selected_mfma.k_per_blk); auto blockwise_gemm0 = BlockwiseGemmXdlops_v2< BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp index 44a488c5dd..1f7458e68f 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp @@ -343,10 +343,16 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle const auto M = d0_grid_desc_m_n.GetLength(I0); const auto N = d0_grid_desc_m_n.GetLength(I1); - constexpr auto mfma = MfmaSelector::selected_mfma; - constexpr auto N3 = mfma.num_groups_per_blk; - constexpr auto N4 = mfma.num_input_blks; - constexpr auto N5 = mfma.group_size; + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + math::lcm(AK1, BK1) <= 4) + ? true + : false; + constexpr auto mfma = + MfmaSelector::selected_mfma; + constexpr auto N3 = mfma.num_groups_per_blk; + constexpr auto N4 = mfma.num_input_blks; + constexpr auto N5 = mfma.group_size; return transform_tensor_descriptor( d0_grid_desc_m_n, make_tuple(make_unmerge_transform( @@ -552,8 +558,16 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle // acc1[m][o] += acc[m][n] * B1[n][o] // sanity check + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; constexpr index_t KPack = math::max( - math::lcm(AK1, BK1), MfmaSelector::selected_mfma.k_per_blk); + lcm_AK1_BK1, + MfmaSelector::selected_mfma + .k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_v2< BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp index 7d2dfab15f..f7746b470f 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp @@ -469,8 +469,16 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle // acc1[m][o] += acc[m][n] * B1[n][o] // sanity check + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; constexpr index_t KPack = math::max( - math::lcm(AK1, BK1), MfmaSelector::selected_mfma.k_per_blk); + lcm_AK1_BK1, + MfmaSelector::selected_mfma + .k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_v2< BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp index 35176591c1..8b3f51b9b0 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp @@ -498,8 +498,16 @@ struct GridwiseGemmBiasAddReduce_k0mk1_k0nk1_mn_xdl_cshuffle_v1 // c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in // register // sanity check + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; constexpr index_t KPack = math::max( - math::lcm(AK1, BK1), MfmaSelector::selected_mfma.k_per_blk); + lcm_AK1_BK1, + MfmaSelector::selected_mfma + .k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector< BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp index 5c9f40b51a..60ee78528d 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp @@ -464,8 +464,16 @@ struct GridwiseGemmMultipleDMultipleR_k0mk1_k0nk1_mn_xdl_cshuffle_v1 // c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in // register // sanity check + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; constexpr index_t KPack = math::max( - math::lcm(AK1, BK1), MfmaSelector::selected_mfma.k_per_blk); + lcm_AK1_BK1, + MfmaSelector::selected_mfma + .k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector< BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp index ae93af192e..d1d97da5b0 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp @@ -599,9 +599,16 @@ struct GridwiseGemmMultipleD_xdl_splitk_cshuffle // c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in // register // sanity check - constexpr index_t KPack = - math::max(math::lcm(AK1, BK1), - MfmaSelector::selected_mfma.k_per_blk); + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; + constexpr index_t KPack = math::max( + lcm_AK1_BK1, + MfmaSelector:: + selected_mfma.k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector< BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp index 0e5777e561..7105fa7012 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp @@ -451,8 +451,16 @@ struct GridwiseGemmReduce_k0mk1_k0nk1_mn_xdl_cshuffle_v1 // c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in // register // sanity check + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; constexpr index_t KPack = math::max( - math::lcm(AK1, BK1), MfmaSelector::selected_mfma.k_per_blk); + lcm_AK1_BK1, + MfmaSelector::selected_mfma + .k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector< BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp index 0078660556..3429c20e73 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp @@ -581,9 +581,16 @@ struct GridwiseGemmSplitKMultipleD_xdl_cshuffle // c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in // register // sanity check + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; constexpr index_t KPack = - math::max(math::lcm(AK1, BK1), - MfmaSelector::selected_mfma.k_per_blk); + math::max(lcm_AK1_BK1, + MfmaSelector:: + selected_mfma.k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector< BlockSize, @@ -1006,9 +1013,16 @@ struct GridwiseGemmSplitKMultipleD_xdl_cshuffle // c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in // register // sanity check + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; constexpr index_t KPack = - math::max(math::lcm(AK1, BK1), - MfmaSelector::selected_mfma.k_per_blk); + math::max(lcm_AK1_BK1, + MfmaSelector:: + selected_mfma.k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector< BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle_v2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle_v2.hpp index caf8f040f4..d7c87a170c 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle_v2.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle_v2.hpp @@ -595,9 +595,16 @@ struct GridwiseGemmMultipleD_xdl_splitk_cshuffle // c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in // register // sanity check - constexpr index_t KPack = - math::max(math::lcm(AK1, BK1), - MfmaSelector::selected_mfma.k_per_blk); + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; + constexpr index_t KPack = math::max( + lcm_AK1_BK1, + MfmaSelector:: + selected_mfma.k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector< BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_bwd_weight_v3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_bwd_weight_v3.hpp index d2a06ba9af..08d9386d72 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_bwd_weight_v3.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_bwd_weight_v3.hpp @@ -79,9 +79,16 @@ struct GridwiseGemm_xdl_cshuffle_v3 static constexpr auto AK1Number = Number{}; static constexpr auto BK1Number = Number{}; + static constexpr auto lcm_AK1_BK1 = math::lcm(AK1Number, BK1Number); + static constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; static constexpr index_t KPack = - math::max(math::lcm(AK1Number, BK1Number), - MfmaSelector::selected_mfma.k_per_blk); + math::max(lcm_AK1_BK1, + MfmaSelector:: + selected_mfma.k_per_blk); using ThisThreadBlock = ThisThreadBlock; diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_streamk_v3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_streamk_v3.hpp old mode 100755 new mode 100644 index 6ef35da485..e04f24c989 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_streamk_v3.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_streamk_v3.hpp @@ -139,9 +139,16 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3 static constexpr auto AK1Number = Number{}; static constexpr auto BK1Number = Number{}; + static constexpr auto lcm_AK1_BK1 = math::lcm(AK1Number, BK1Number); + static constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; static constexpr index_t KPack = - math::max(math::lcm(AK1Number, BK1Number), - MfmaSelector::selected_mfma.k_per_blk); + math::max(lcm_AK1_BK1, + MfmaSelector:: + selected_mfma.k_per_blk); using ThisThreadBlock = ThisThreadBlock; __host__ static auto CalculateMPadded(index_t M) diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v2.hpp index db9625c6e6..af91721c8a 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v2.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v2.hpp @@ -869,9 +869,16 @@ struct GridwiseGemm_xdl_cshuffle_v2 // c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in // register // sanity check - constexpr index_t KPack = - math::max(math::lcm(AK1Number, BK1Number), - MfmaSelector::selected_mfma.k_per_blk); + constexpr auto lcm_AK1_BK1 = math::lcm(AK1Number, BK1Number); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; + constexpr index_t KPack = math::max( + lcm_AK1_BK1, + MfmaSelector:: + selected_mfma.k_per_blk); // auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector< // BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3.hpp index a43f0f880a..55360fc0d0 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3.hpp @@ -147,9 +147,16 @@ struct GridwiseGemm_xdl_cshuffle_v3 static constexpr auto AK1Number = Number{}; static constexpr auto BK1Number = Number{}; + static constexpr auto lcm_AK1_BK1 = math::lcm(AK1Number, BK1Number); + static constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; static constexpr index_t KPack = - math::max(math::lcm(AK1Number, BK1Number), - MfmaSelector::selected_mfma.k_per_blk); + math::max(lcm_AK1_BK1, + MfmaSelector:: + selected_mfma.k_per_blk); using ThisThreadBlock = ThisThreadBlock; diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_b_scale.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_b_scale.hpp index 366a6c59c2..2e62110416 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_b_scale.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_b_scale.hpp @@ -155,9 +155,16 @@ struct GridwiseGemm_xdl_cshuffle_v3 static constexpr auto AK1Number = Number{}; static constexpr auto BK1Number = Number{}; + static constexpr auto lcm_AK1_BK1 = math::lcm(AK1Number, BK1Number); + static constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; static constexpr index_t KPack = - math::max(math::lcm(AK1Number, BK1Number), - MfmaSelector::selected_mfma.k_per_blk); + math::max(lcm_AK1_BK1, + MfmaSelector:: + selected_mfma.k_per_blk); using ThisThreadBlock = ThisThreadBlock; @@ -1424,7 +1431,8 @@ struct GridwiseGemm_xdl_cshuffle_v3 // b scale // static_assert(KPerBlock <= ScaleBlockK); - static constexpr auto mfma = MfmaSelector{}; + static constexpr auto mfma = + MfmaSelector{}; static constexpr auto KPerXdlops = mfma.GetKPerXdlops(); static constexpr auto K1PerXdlops = mfma.GetK1PerXdlops(); static constexpr auto K0PerXdlops = KPerXdlops / K1PerXdlops; @@ -1895,7 +1903,8 @@ struct GridwiseGemm_xdl_cshuffle_v3 KPerBlock); // B scale - static constexpr auto mfma = MfmaSelector{}; + static constexpr auto mfma = + MfmaSelector{}; static constexpr auto KPerXdlops = mfma.GetKPerXdlops(); static constexpr auto K1PerXdlops = mfma.GetK1PerXdlops(); static constexpr auto K0PerXdlops = KPerXdlops / K1PerXdlops; diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp index 7f815de1f9..0a62464cc2 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp @@ -489,8 +489,16 @@ struct GridwiseGemmLayernorm_k0mk1_k0nk1_mn_xdl_cshuffle_v1 // c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in // register // sanity check + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; constexpr index_t KPack = math::max( - math::lcm(AK1, BK1), MfmaSelector::selected_mfma.k_per_blk); + lcm_AK1_BK1, + MfmaSelector::selected_mfma + .k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector< BlockSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp index 8675a9242a..6a4b1cc14b 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp @@ -487,9 +487,16 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_waveletmodel_cshuffle else if(TileMathThreadGroup::IsBelong()) { // branch early for math wave - constexpr index_t KPack = - math::max(math::lcm(AK1, BK1), - MfmaSelector::selected_mfma.k_per_blk); + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; + constexpr index_t KPack = math::max( + lcm_AK1_BK1, + MfmaSelector:: + selected_mfma.k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1< TileMathThreadGroupSize, diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp index 15c64f2e47..7db8798695 100644 --- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp +++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp @@ -446,8 +446,16 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r1 // c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in // register // sanity check + constexpr auto lcm_AK1_BK1 = math::lcm(AK1, BK1); + constexpr bool is_single_rate_mfma = + ((is_same::value || is_same::value) && + lcm_AK1_BK1 <= 4) + ? true + : false; constexpr index_t k_pack = math::max( - math::lcm(AK1, BK1), MfmaSelector::selected_mfma.k_per_blk); + lcm_AK1_BK1, + MfmaSelector::selected_mfma + .k_per_blk); auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 1> #endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 16, 16, 32, 8, 16, 16, 1, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 1> + // clang-format on >; template , S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 1>, DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 32, 32, 32, 8, 32, 32, 1, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, false, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 2>, @@ -77,7 +76,6 @@ using device_grouped_conv_bwd_weight_two_stage_nhwgc_xdl_c_shuffle_f16_instances DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 32, 32, 32, 8, 32, 32, 1, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 2>, DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 64, 32, 32, 8, 32, 32, 2, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 4>, DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 128, 32, 32, 8, 32, 32, 4, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, false, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 8> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -118,10 +116,9 @@ using device_grouped_conv_bwd_weight_two_stage_nhwgc_xdl_c_shuffle_bf16_generic_ //#########################################| Spatial| | | | | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl| Scheduler| Version| | //#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else - DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 16, 16, 32, 8, 16, 16, 1, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 1> #endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 16, 16, 32, 8, 16, 16, 1, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 1> + // clang-format on >; template , S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 1>, DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 32, 32, 32, 8, 32, 32, 1, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, false, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 2>, @@ -148,7 +145,6 @@ using device_grouped_conv_bwd_weight_two_stage_nhwgc_xdl_c_shuffle_bf16_instance DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 32, 32, 32, 8, 32, 32, 1, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 2>, DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 64, 32, 32, 8, 32, 32, 2, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 4>, DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 128, 32, 32, 8, 32, 32, 4, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, false, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 8> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -189,10 +185,9 @@ using device_grouped_conv_bwd_weight_two_stage_ngchw_xdl_c_shuffle_f16_generic_i //#########################################| Spatial| | | | | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl| Scheduler| Version| | //#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else - DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 16, 16, 32, 8, 16, 16, 1, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 1, F16, F16, 1, 1> #endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 16, 16, 32, 8, 16, 16, 1, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 1, F16, F16, 1, 1> + // clang-format on >; // NGCHW requires transpose, we use vector loads and stores params for them @@ -210,7 +205,7 @@ using device_grouped_conv_bwd_weight_two_stage_ngchw_xdl_c_shuffle_f16_instances //#########################################| Spatial| | | | | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl| Scheduler| Version| | //#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 16, 16, 32, 8, 16, 16, 1, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 1, F16, F16, 1, 1>, DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 32, 32, 32, 8, 32, 32, 1, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, false, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 2, F16, F16, 2, 2>, @@ -234,7 +229,6 @@ using device_grouped_conv_bwd_weight_two_stage_ngchw_xdl_c_shuffle_f16_instances DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 64, 32, 32, 8, 32, 32, 2, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 4, F16, F16, 4, 1>, DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 128, 32, 32, 8, 32, 32, 4, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, false, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, false, 1, 1, S<1, 8, 1, 4>, 1, Scheduler, PipelineVersion, 8, F16, F16, 8, 1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -253,10 +247,9 @@ using device_grouped_conv_bwd_weight_two_stage_ngchw_xdl_c_shuffle_bf16_generic_ //#########################################| Spatial| | | | | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl| Scheduler| Version| | //#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else - DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 16, 16, 32, 8, 16, 16, 1, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 1, BF16, BF16, 1, 1> #endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 16, 16, 32, 8, 16, 16, 1, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 1, BF16, BF16, 1, 1> + // clang-format on >; template , S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 1, BF16, BF16, 1, 1>, DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 32, 32, 32, 8, 32, 32, 1, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, false, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 2, BF16, BF16, 2, 2>, @@ -297,7 +290,6 @@ using device_grouped_conv_bwd_weight_two_stage_ngchw_xdl_c_shuffle_bf16_instance DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 64, 32, 32, 8, 32, 32, 2, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 4, false, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 4, false, 1, 1, S<1, 8, 1, 8>, 1, Scheduler, PipelineVersion, 4, BF16, BF16, 4, 1>, DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvSpec, 64, 128, 32, 32, 8, 32, 32, 4, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, false, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, false, 1, 1, S<1, 8, 1, 4>, 1, Scheduler, PipelineVersion, 8, BF16, BF16, 8, 1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp index 6f10205d3d..dad67b396f 100644 --- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp @@ -34,8 +34,8 @@ using device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instances = std::tuple< //##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, @@ -68,7 +68,6 @@ using device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instances = std::tuple< DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v2>, DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp index 6060cac0c8..62df9a81b5 100644 --- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp @@ -34,8 +34,8 @@ using device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instances = std::tuple< //##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, @@ -68,7 +68,6 @@ using device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instances = std::tuple< DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v2>, DeviceBatchedGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp index 8f9eb92a7f..23e176f6e8 100644 --- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp @@ -32,10 +32,9 @@ using device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_generic_instances = std::t //#################| | | | | | | | Operation| Operation| Operation| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| Stage | | | //#################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, 64, 16, 16, 4, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, 64, 16, 16, 4, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, 64, 16, 16, 4, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1> -#endif // clang-format on >; @@ -48,8 +47,8 @@ using device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instances = std::tuple< //#################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, @@ -109,7 +108,6 @@ using device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instances = std::tuple< DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, 128, 16, 32, 4, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v2>, DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, 64, 16, 16, 4, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp index 545986f8b0..0216b002ac 100644 --- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp @@ -32,10 +32,9 @@ using device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_generic_instances = std::t //#################| | | | | | | | Operation| Operation| Operation| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| Stage | | | //#################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1> -#endif // clang-format on >; @@ -48,8 +47,8 @@ using device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instances = std::tuple< //#################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>, @@ -97,7 +96,6 @@ using device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instances = std::tuple< DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v2>, DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp index c971a55363..6af3c2654e 100644 --- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp @@ -26,10 +26,8 @@ using S = ck::Sequence; using PassThrough = ck::tensor_operation::element_wise::PassThrough; using Scale = ck::tensor_operation::element_wise::Scale; -#if !defined(CK_USE_AMD_MFMA_GFX950) static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmPadded = ck::tensor_operation::device::GemmSpecialization::MNKOPadding; -#endif // c[g, m, n] = a[g, m, k] * b[g, n, k] template @@ -41,7 +39,7 @@ using device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_ //#######################################| | | | | | | | | | | Operation| Operation| Operation| Operation| Operation| | Stage| | Block| Block| Block| Block| Block| | | | | | Per| Per| Per| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| Triangle| //#######################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | Wave| Wave| Wave| | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle< Row, Col, Row, Row, F16, F16, F16, F16, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 64, 32, 8, 8, 2, 32, 32, 2, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, Masking>, DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle< Row, Col, Row, Row, F16, F16, F16, F16, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 128, 32, 8, 8, 2, 32, 32, 2, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, Masking>, DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle< Row, Col, Row, Row, F16, F16, F16, F16, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 64, 32, 8, 8, 2, 32, 32, 1, 8, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, Masking>, @@ -57,8 +55,7 @@ using device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_ // Padded fallback kernel DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle< Row, Col, Row, Row, F16, F16, F16, F16, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmPadded, 1, 256, 128, 128, 64, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, Masking>, DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle< Row, Col, Row, Row, F16, F16, F16, F16, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmPadded, 1, 256, 128, 64, 32, 128, 32, 8, 8, 2, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, Masking> -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; template @@ -70,15 +67,14 @@ using device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_ //#######################################| | | | | | | | | | | Operation| Operation| Operation| Operation| Operation| | Stage| | Block| Block| Block| Block| Block| | | | | | Per| Per| Per| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| Triangle| //#######################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | Wave| Wave| Wave| | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle< Row, Col, Row, Row, F16, F16, F16, F16, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmPadded, 1, 256, 256, 128, 40, 64, 32, 4, 4, 2, 32, 32, 2, 4, 2, S<2,128, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, S<2,128, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, Masking>, DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle< Row, Col, Row, Row, F16, F16, F16, F16, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmPadded, 1, 256, 256, 128, 40, 128, 32, 4, 4, 2, 32, 32, 2, 4, 4, S<2,128, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, S<2,128, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, Masking>, DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle< Row, Col, Row, Row, F16, F16, F16, F16, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmPadded, 1, 256, 128, 256, 40, 64, 32, 4, 4, 2, 32, 32, 1, 8, 2, S<2,128, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, S<2,128, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, Masking>, DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle< Row, Col, Row, Row, F16, F16, F16, F16, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmPadded, 1, 256, 128, 256, 40, 128, 32, 4, 4, 2, 32, 32, 1, 8, 4, S<2,128, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, S<2,128, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, Masking>, DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle< Row, Col, Row, Row, F16, F16, F16, F16, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmPadded, 1, 256, 128, 128, 40, 64, 32, 4, 4, 2, 32, 32, 1, 4, 2, S<2,128, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, S<2,128, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, Masking>, DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle< Row, Col, Row, Row, F16, F16, F16, F16, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmPadded, 1, 256, 128, 128, 40, 128, 32, 4, 4, 2, 32, 32, 1, 4, 4, S<2,128, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, S<2,128, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, Masking> -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance( diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp index 6abb90a39b..8382f069d7 100644 --- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp @@ -26,12 +26,10 @@ using S = ck::Sequence; using PassThrough = ck::tensor_operation::element_wise::PassThrough; using ScaleAdd = ck::tensor_operation::element_wise::ScaleAdd; -#if !defined(CK_USE_AMD_MFMA_GFX950) static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmPadded = ck::tensor_operation::device::GemmSpecialization::MNKOPadding; static constexpr auto TensorDefault = ck::tensor_operation::device::TensorSpecialization::Default; -#endif // c[g, m, n] = a[g, m, k] * b[g, n, k] template , ck::Tuple<>, F32, BF16, PassThrough, PassThrough, ScaleAdd, PassThrough, PassThrough, GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 64, 32, 128, 32, 8, 8, 2, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec, 1>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, BF16, BF16, BF16, BF16, ck::Tuple, ck::Tuple<>, F32, BF16, PassThrough, PassThrough, ScaleAdd, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 256, 128, 32, 64, 32, 8, 8, 2, 32, 32, 2, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, BF16, BF16, BF16, BF16, ck::Tuple, ck::Tuple<>, F32, BF16, PassThrough, PassThrough, ScaleAdd, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 256, 128, 32, 128, 32, 8, 8, 2, 32, 32, 2, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, @@ -66,8 +64,7 @@ using device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_ DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, BF16, BF16, BF16, BF16, ck::Tuple, ck::Tuple<>, F32, BF16, PassThrough, PassThrough, ScaleAdd, PassThrough, PassThrough, GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 128, 64, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec, 1>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, BF16, BF16, BF16, BF16, ck::Tuple, ck::Tuple<>, F32, BF16, PassThrough, PassThrough, ScaleAdd, PassThrough, PassThrough, GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 128, 64, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, BF16, BF16, BF16, BF16, ck::Tuple, ck::Tuple<>, F32, BF16, PassThrough, PassThrough, ScaleAdd, PassThrough, PassThrough, GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 64, 32, 128, 32, 8, 8, 2, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec> -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_batched_gemm_bias_masking_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances( diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp index 608443cf44..b6c14d69db 100644 --- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp @@ -26,12 +26,10 @@ using S = ck::Sequence; using PassThrough = ck::tensor_operation::element_wise::PassThrough; using ScaleAdd = ck::tensor_operation::element_wise::ScaleAdd; -#if !defined(CK_USE_AMD_MFMA_GFX950) static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmPadded = ck::tensor_operation::device::GemmSpecialization::MNKOPadding; static constexpr auto TensorDefault = ck::tensor_operation::device::TensorSpecialization::Default; -#endif // c[g, m, n] = a[g, m, k] * b[g, n, k] template , ck::Tuple<>, F32, F16, PassThrough, PassThrough, ScaleAdd, PassThrough, PassThrough, GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 64, 32, 128, 32, 8, 8, 2, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec, 1>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple, ck::Tuple<>, F32, F16, PassThrough, PassThrough, ScaleAdd, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 256, 128, 32, 64, 32, 8, 8, 2, 32, 32, 2, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple, ck::Tuple<>, F32, F16, PassThrough, PassThrough, ScaleAdd, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 256, 128, 32, 128, 32, 8, 8, 2, 32, 32, 2, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, @@ -68,8 +66,7 @@ using device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16 DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple, ck::Tuple<>, F32, F16, PassThrough, PassThrough, ScaleAdd, PassThrough, PassThrough, GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 128, 64, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec, 1>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple, ck::Tuple<>, F32, F16, PassThrough, PassThrough, ScaleAdd, PassThrough, PassThrough, GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 128, 64, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple, ck::Tuple<>, F32, F16, PassThrough, PassThrough, ScaleAdd, PassThrough, PassThrough, GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 64, 32, 128, 32, 8, 8, 2, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec> -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_batched_gemm_bias_masking_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances( diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp index 07eb744323..2ce5124706 100644 --- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp @@ -26,12 +26,10 @@ using S = ck::Sequence; using PassThrough = ck::tensor_operation::element_wise::PassThrough; using Scale = ck::tensor_operation::element_wise::Scale; -#if !defined(CK_USE_AMD_MFMA_GFX950) static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmPadded = ck::tensor_operation::device::GemmSpecialization::MNKOPadding; static constexpr auto TensorDefault = ck::tensor_operation::device::TensorSpecialization::Default; -#endif // c[g, m, n] = a[g, m, k] * b[g, n, k] template , ck::Tuple<>, F32, BF16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 64, 32, 128, 32, 8, 8, 2, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, BF16, BF16, BF16, BF16, ck::Tuple<>, ck::Tuple<>, F32, BF16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 256, 128, 32, 64, 32, 8, 8, 2, 32, 32, 2, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, BF16, BF16, BF16, BF16, ck::Tuple<>, ck::Tuple<>, F32, BF16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 256, 128, 32, 128, 32, 8, 8, 2, 32, 32, 2, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, @@ -64,8 +62,7 @@ using device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_ DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, BF16, BF16, BF16, BF16, ck::Tuple<>, ck::Tuple<>, F32, BF16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 64, 256, 64, 64, 32, 8, 8, 2, 16, 16, 1, 16, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 4, S<1, 32, 1, 8>, 8, MaskingSpec>, // Padded fallback kernel DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, BF16, BF16, BF16, BF16, ck::Tuple<>, ck::Tuple<>, F32, BF16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 128, 64, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec> -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_batched_gemm_masking_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances( diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp index 09055ea188..4e8adceb1c 100644 --- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp @@ -26,12 +26,10 @@ using S = ck::Sequence; using PassThrough = ck::tensor_operation::element_wise::PassThrough; using Scale = ck::tensor_operation::element_wise::Scale; -#if !defined(CK_USE_AMD_MFMA_GFX950) static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmPadded = ck::tensor_operation::device::GemmSpecialization::MNKOPadding; static constexpr auto TensorDefault = ck::tensor_operation::device::TensorSpecialization::Default; -#endif // c[g, m, n] = a[g, m, k] * b[g, n, k] template using device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances = std::tuple< -// clang-format off + // clang-format off // #############################################| NumDimG| NumDimM| NumDimN| NumDimK| NumDimO| AData| B0Data| B1Data| CData| Acc0BiasData| Acc1BiasData| AccData| CShuffle| A| B0| Acc0| B1| C| GEMM| ATensorSpec| B0TensorSpec| B1TensorSpec| CTensorSpec| NumGemmK| Block| Gemm01| Gemm0| Gemm0| Gemm1| Gemm1| AK1| BK1| B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| MaskingSpec| // #############################################| | | | | | Type| Type| Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| Specialization| | | | | Prefetch| Size| MPer| NPer| KPer| NPer| KPer| | | | XDL| XDL| MXdl| NXdl| NXdl| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| | // #############################################| | | | | | | | | | | | | | Operation| Operation| Operation| Operation| Operation| | | | | | Stage| | Block| Block| Block| Block| Block| | | | | | Per| Per| Per| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | // #############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Wave| Wave| Wave| | | | | | | | | | | | | | | | | | | | | | | | | | | -#if defined(CK_USE_AMD_MFMA_GFX950) -#else DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple<>, ck::Tuple<>, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 64, 32, 128, 32, 8, 8, 2, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple<>, ck::Tuple<>, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 256, 128, 32, 64, 32, 8, 8, 2, 32, 32, 2, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple<>, ck::Tuple<>, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 256, 128, 32, 128, 32, 8, 8, 2, 32, 32, 2, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, @@ -62,12 +58,13 @@ using device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_ DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple<>, ck::Tuple<>, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 128, 32, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple<>, ck::Tuple<>, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 64, 256, 32, 128, 32, 8, 8, 2, 16, 16, 1, 16, 8, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 8, S<1, 16, 1,16>, 8, MaskingSpec>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple<>, ck::Tuple<>, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 64, 256, 32, 64, 32, 8, 8, 2, 16, 16, 1, 16, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 4, S<1, 32, 1, 8>, 8, MaskingSpec>, +#if !defined(CK_USE_AMD_MFMA_GFX950) DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple<>, ck::Tuple<>, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 64, 256, 64, 128, 32, 8, 8, 2, 16, 16, 1, 16, 8, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 8, S<1, 16, 1,16>, 8, MaskingSpec>, DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple<>, ck::Tuple<>, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 64, 256, 64, 64, 32, 8, 8, 2, 16, 16, 1, 16, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 4, S<1, 32, 1, 8>, 8, MaskingSpec>, +#endif // !defined(CK_USE_AMD_MFMA_GFX950) // Padded fallback kernel DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, F16, F16, F16, F16, ck::Tuple<>, ck::Tuple<>, F32, F16, PassThrough, PassThrough, Scale, PassThrough, PassThrough, GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault, 1, 256, 128, 128, 64, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8, MaskingSpec> -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_batched_gemm_masking_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances( diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp index d30b93cf6a..c0c74ff7fb 100644 --- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp @@ -45,7 +45,9 @@ using device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances = std::tuple< //##########################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //##########################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //##########################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -#if !defined(CK_USE_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) + DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, @@ -59,9 +61,6 @@ using device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances = std::tuple< DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8> -#else - DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8> -#endif // clang-format on >; @@ -72,7 +71,9 @@ using device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_1x1_p0_f16_instances = std: //##########################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //##########################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //##########################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -#if !defined(CK_USE_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) + DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, @@ -86,9 +87,6 @@ using device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_1x1_p0_f16_instances = std: DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8> -#else - DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8> -#endif // clang-format on >; @@ -99,7 +97,9 @@ using device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances = s //##########################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //##########################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //##########################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -#if !defined(CK_USE_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) + DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, @@ -113,9 +113,6 @@ using device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances = s DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8> -#else - DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8> -#endif // clang-format on >; @@ -125,7 +122,9 @@ using device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_odd_c_f16_instances = std:: //##########################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //##########################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //##########################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -#if !defined(CK_USE_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) + DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdOddC, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdOddC, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdOddC, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdOddC, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 4, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 4, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, @@ -144,9 +143,6 @@ using device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_odd_c_f16_instances = std:: DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdOddC, 256, 256, 64, 2, 4, 32, 32, 4, 1, S<2, 32, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<2, 32, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdOddC, 128, 128, 64, 2, 4, 32, 32, 2, 2, S<2, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<2, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 4>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdOddC, 128, 64, 64, 2, 4, 32, 32, 1, 2, S<2, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<2, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8> -#else - DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdOddC, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8> -#endif // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp index 8d57edf4d7..715dffb7ff 100644 --- a/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp @@ -44,7 +44,9 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instances = s //##########################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| GlobalMemory| Specialization| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //##########################################################################################| | | | | Operation| Operation| Operation| DataOperation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //##########################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -#if !defined(CK_USE_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) + DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, @@ -58,9 +60,6 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instances = s DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8> -#else - DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8> -#endif // clang-format on >; @@ -71,7 +70,9 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_1x1_p0_f16_instan //##########################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| GlobalMemory| Specialization| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //##########################################################################################| | | | | Operation| Operation| Operation| DataOperation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //##########################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -#if !defined(CK_USE_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) + DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1P0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1P0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1P0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, @@ -85,9 +86,6 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_1x1_p0_f16_instan DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1P0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1P0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1P0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8> -#else - DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8> -#endif // clang-format on >; @@ -98,7 +96,9 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_1x1_s1_p0_f16_ins //##########################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| GlobalMemory| Specialization| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //##########################################################################################| | | | | Operation| Operation| Operation| DataOperation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //##########################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -#if !defined(CK_USE_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) + DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1S1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1S1P0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1S1P0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1S1P0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, @@ -112,9 +112,6 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_1x1_s1_p0_f16_ins DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1S1P0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1S1P0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1S1P0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8> -#else - DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwd1x1S1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8> -#endif // clang-format on >; @@ -125,7 +122,9 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_odd_c_f16_instanc //##########################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| GlobalMemory| Specialization| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //##########################################################################################| | | | | Operation| Operation| Operation| DataOperation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //##########################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -#if !defined(CK_USE_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) + DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdOddC, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdOddC, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdOddC, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdOddC, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 4, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 4, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, @@ -144,9 +143,6 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_odd_c_f16_instanc DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdOddC, 256, 256, 64, 2, 4, 32, 32, 4, 1, S<2, 32, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<2, 32, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdOddC, 128, 128, 64, 2, 4, 32, 32, 2, 2, S<2, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<2, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 4>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdOddC, 128, 64, 64, 2, 4, 32, 32, 1, 2, S<2, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<2, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8> -#else - DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddRelu, MemorySet, ConvFwdOddC, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8> -#endif // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu_add/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu_add/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp index 63ca21ff74..5676d77986 100644 --- a/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu_add/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu_add/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp @@ -42,7 +42,9 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instances //##############################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //##############################################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //##############################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -#if !defined(CK_USE_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) + DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, @@ -56,9 +58,6 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instances DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8> -#else - DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8> -#endif // clang-format on >; @@ -69,7 +68,9 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_1x1_p0_f16_in //##############################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //##############################################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //##############################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -#if !defined(CK_USE_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) + DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1P0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1P0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1P0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, @@ -83,9 +84,6 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_1x1_p0_f16_in DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1P0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1P0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1P0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8> -#else - DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8> -#endif // clang-format on >; @@ -96,7 +94,9 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_1x1_s1_p0_f16 //##############################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //##############################################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //##############################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -#if !defined(CK_USE_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) + DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1S1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1S1P0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1S1P0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1S1P0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, @@ -110,9 +110,6 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_1x1_s1_p0_f16 DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1S1P0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1S1P0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1S1P0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8> -#else - DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwd1x1S1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8> -#endif // clang-format on >; @@ -123,7 +120,9 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_odd_c_f16_ins //##############################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| //##############################################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //##############################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -#if !defined(CK_USE_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) + DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdOddC, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdOddC, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdOddC, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdOddC, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 4, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 4, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 16, 1, 1, 8>, 8>, @@ -142,9 +141,6 @@ using device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_odd_c_f16_ins DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdOddC, 256, 256, 64, 2, 4, 32, 32, 4, 1, S<2, 32, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<2, 32, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdOddC, 128, 128, 64, 2, 4, 32, 32, 2, 2, S<2, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<2, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 4>, 8>, DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdOddC, 128, 64, 64, 2, 4, 32, 32, 1, 2, S<2, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<2, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 16, 1, 1, 4>, 8> -#else - DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Add_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, AddReluAdd, ConvFwdOddC, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 8, 8>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 1, 1, S<1, 1, 32, 1, 1, 8>, 8> -#endif // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp index 2189a10e48..0d143b95ee 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp @@ -36,8 +36,8 @@ using device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances = std::tu //#####################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 2, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 2, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 2, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 2, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 2, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -85,7 +85,6 @@ using device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances = std::tu DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 2, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 2, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // !defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp index 73fcc576a1..3ebd0c5351 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp @@ -38,8 +38,8 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances = std::tuple< //#####################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemm_Xdl_CShuffle< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemm_Xdl_CShuffle< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemm_Xdl_CShuffle< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemm_Xdl_CShuffle< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemm_Xdl_CShuffle< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 2, 2, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -96,7 +96,6 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances = std::tuple< DeviceGemm_Xdl_CShuffle< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 2, 2, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemm_Xdl_CShuffle< Col, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // !defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp index be5751e065..bd0f7346f7 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp @@ -38,8 +38,8 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances = std::tuple< //#####################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemm_Xdl_CShuffle< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 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, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemm_Xdl_CShuffle< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 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, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemm_Xdl_CShuffle< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemm_Xdl_CShuffle< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemm_Xdl_CShuffle< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 2, 8, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -96,7 +96,6 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances = std::tuple< DeviceGemm_Xdl_CShuffle< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 2, 8, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemm_Xdl_CShuffle< Col, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 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, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // !defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp index c9221d194a..a6ac333ea3 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp @@ -40,10 +40,9 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_generic_instances = std::tu //#####################| | | | Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| 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_AMD_MFMA_GFX950) - DeviceGemm_Xdl_CShuffle< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, MNKPadding, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else - DeviceGemm_Xdl_CShuffle< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, MNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, 1, LoopScheduler::Default, PipelineVersion::v1> + DeviceGemm_Xdl_CShuffle< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, MNKPadding, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, 1, LoopScheduler::Default, PipelineVersion::v1>, #endif // defined(CK_USE_AMD_MFMA_GFX950) + DeviceGemm_Xdl_CShuffle< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, MNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, 1, LoopScheduler::Default, PipelineVersion::v1> // clang-format on >; @@ -55,7 +54,8 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances = std::tuple< //#####################| | | | Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| 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_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) +#endif // defined(CK_USE_AMD_MFMA_GFX950) // pipeline v1, 1 wave DeviceGemm_Xdl_CShuffle< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemm_Xdl_CShuffle< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -114,7 +114,6 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances = std::tuple< DeviceGemm_Xdl_CShuffle< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemm_Xdl_CShuffle< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // !defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp index beef06c9f4..852b053527 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp @@ -42,10 +42,9 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_generic_instances = std::tu #if defined(CK_USE_AMD_MFMA_GFX950) //DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, //DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 64, 128, 32, 32, 16, 16, 2, 2, S<4, 64, 1>, 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>, 4, LoopScheduler::Default, PipelineVersion::v1> - DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, MNKPadding, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else - DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, MNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, 1, LoopScheduler::Default, PipelineVersion::v1> + DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, MNKPadding, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, 1, LoopScheduler::Default, PipelineVersion::v1>, #endif // defined(CK_USE_AMD_MFMA_GFX950) + DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, MNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, 1, LoopScheduler::Default, PipelineVersion::v1> // clang-format on >; @@ -57,7 +56,8 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::tuple< //#####################| | | | Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| 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_AMD_MFMA_GFX950) +#if defined(CK_USE_AMD_MFMA_GFX950) +#endif // defined(CK_USE_AMD_MFMA_GFX950) // pipeline v1, 1 wave DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, @@ -107,7 +107,6 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::tuple< DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // !defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp index 3fdf6e23ce..41f6ec2bf7 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp @@ -51,8 +51,8 @@ using device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances = , DeviceGemm_Xdl_CShuffle< Row, Col, Row, int8_t, int8_t, int8_t, int32_t, int32_t, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 128, 32, 32, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 4>, 16, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemm_Xdl_CShuffle< Row, Col, Row, int8_t, int8_t, int8_t, int32_t, int32_t, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 64, 256, 64, 64, 16, 16, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 8>, 4, LoopScheduler::Default, PipelineVersion::v1> -#endif - // clang-format on +#endif // defined(CK_USE_AMD_MFMA_GFX950) + // clang-format on >; void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp index 84e6410bfb..74cf837500 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp @@ -18,7 +18,7 @@ using Instances = std::tuple< //##########| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| | | | //##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, @@ -27,7 +27,6 @@ using Instances = std::tuple< DeviceGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) #endif // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp index f9c9a4b876..f2b28f3b40 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp @@ -18,9 +18,8 @@ using Instances = std::tuple< //###########| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| | | | //###########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else - DeviceGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 16, 16, 4, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1> #endif // defined(CK_USE_AMD_MFMA_GFX950) + DeviceGemmXdl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 16, 16, 4, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1> #endif // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp index 1a6bfef326..da5fefe5da 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp @@ -18,7 +18,7 @@ using Instances = std::tuple< //##########| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| | | | //##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, @@ -27,7 +27,6 @@ using Instances = std::tuple< DeviceGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) #endif // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp index 2aaee9ab26..b6c03b3367 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp @@ -18,9 +18,8 @@ using Instances = std::tuple< //###########| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| | | | //###########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else - DeviceGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 16, 16, 4, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1> #endif // defined(CK_USE_AMD_MFMA_GFX950) + DeviceGemmXdl< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 16, 16, 4, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1> #endif // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp index d14137b56c..bf271cc3c3 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp @@ -18,7 +18,7 @@ using Instances = std::tuple< //##########| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| | | | //##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, @@ -36,7 +36,6 @@ using Instances = std::tuple< DeviceGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 16, 64, 4, 8, 16, 16, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 16, 32, 4, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 16, 16, 4, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) #endif // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp index a0b0ba5018..0df59933c2 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp @@ -18,9 +18,8 @@ using Instances = std::tuple< //###########| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| | | | //###########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else - DeviceGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 16, 16, 4, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1> #endif // defined(CK_USE_AMD_MFMA_GFX950) + DeviceGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 16, 16, 4, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1> #endif // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp index 4437150047..d9260d85ab 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp @@ -18,7 +18,7 @@ using Instances = std::tuple< //###########| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| | | | //###########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, @@ -32,7 +32,6 @@ using Instances = std::tuple< DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) #endif // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp index 694b06f0b3..8b98133ada 100644 --- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp @@ -18,11 +18,10 @@ using Instances = std::tuple< //###########| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| | | | //###########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 144, 8, 8, 16, 16, 2, 9, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 8, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1>, DeviceGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 144, 4, 8, 16, 16, 2, 9, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 4>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1, 1, LoopScheduler::Interwave, PipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) #endif // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp index 77bdb78401..9ea79b1467 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp @@ -45,11 +45,10 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> -#else - DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, #endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> + // clang-format on >; using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instances = std::tuple< @@ -60,8 +59,8 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 2, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -118,8 +117,7 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 2, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; // irregular tile size @@ -132,8 +130,8 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // pipeline v1, 2 waves @@ -145,8 +143,7 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn , DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp index 1c93239af6..f6a959a447 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp @@ -45,11 +45,10 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> -#else - DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, #endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> + // clang-format on >; using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instances = std::tuple< @@ -60,8 +59,8 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 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, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 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, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 8, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -118,8 +117,7 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 8, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 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, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; // irregular tile size @@ -132,8 +130,8 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // pipeline v1, 2 waves @@ -145,8 +143,7 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn , DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp index 162b5a0838..8a68a8bb76 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp @@ -45,11 +45,10 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 64, 16, 2, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> -#else - DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 64, 16, 2, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, #endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> + // clang-format on >; using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances = std::tuple< @@ -60,8 +59,8 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -118,8 +117,7 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; // irregular tile size @@ -132,8 +130,8 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // pipeline v1, 2 waves @@ -145,8 +143,7 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn , DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp index 7e3373e370..42f7a10b9f 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp @@ -45,10 +45,9 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> -#endif // clang-format on >; using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances = @@ -60,8 +59,8 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -109,8 +108,7 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; // irregular tile size @@ -123,8 +121,8 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // pipeline v1, 2 waves @@ -136,8 +134,7 @@ using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn , DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp index 245aa28fc6..6d1c16776a 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp @@ -30,11 +30,10 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_generic //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else - DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, #endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> + // clang-format on >; using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances = std::tuple< // clang-format off @@ -44,8 +43,8 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instanc //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 2, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -102,7 +101,6 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instanc DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 2, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -116,8 +114,8 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_irregul //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // pipeline v1, 2 waves @@ -129,8 +127,7 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_irregul , DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp index 5ac4a8c10d..8b746781ef 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp @@ -30,11 +30,10 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_generic //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else - DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, #endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> + // clang-format on >; using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instances = std::tuple< // clang-format off @@ -44,8 +43,8 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instanc //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 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, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 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, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 8, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -102,7 +101,6 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instanc DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 8, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 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, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -116,8 +114,8 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_irregul //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // pipeline v1, 2 waves @@ -129,8 +127,7 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_irregul , DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp index 22c23c1bfe..414ab674d3 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp @@ -30,11 +30,10 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_generic //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else - DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, #endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> + // clang-format on >; using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instances = std::tuple< // clang-format off @@ -44,8 +43,8 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instanc //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -102,7 +101,6 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instanc DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -116,8 +114,8 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_irregul //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // pipeline v1, 2 waves @@ -129,8 +127,7 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_irregul , DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp index d4849138d0..cf6a64eb26 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp @@ -30,11 +30,10 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_generic //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else - DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, #endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> + // clang-format on >; using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instances = std::tuple< // clang-format off @@ -44,8 +43,8 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instanc //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -93,7 +92,6 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instanc DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -107,8 +105,8 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_irregul //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // pipeline v1, 2 waves @@ -120,8 +118,7 @@ using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_irregul , DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp index 1707c34c0e..184d7974d8 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp @@ -28,10 +28,8 @@ using S = ck::Sequence; using PassThrough = ck::tensor_operation::element_wise::PassThrough; using AddReluAdd = ck::tensor_operation::element_wise::AddReluAdd; -#if !defined(CK_USE_AMD_MFMA_GFX950) static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding; -#endif // e = elementwise((a * b), d0, d1) // h = layernorm(e, gamma, beta) @@ -45,7 +43,7 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instan //#######################################| | | | | | | | | | | | | | Operation| Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _M_N| _NWaveNPerXdl| _M_N| _M| | | //#######################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 2, 2, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, @@ -62,7 +60,6 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instan DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 2, 2, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -75,7 +72,7 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_irregu //#######################################| | | | | | | | | | | | | | Operation| Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _M_N| _NWaveNPerXdl| _M_N| _M| | | //#######################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // pipeline v1, 1 wave DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<16, 4>, 1, S<16, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES @@ -88,8 +85,7 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_irregu , DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<16, 4>, 1, S<16, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp index ed10df8a65..ef37c82c7f 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp @@ -28,10 +28,9 @@ using S = ck::Sequence; using PassThrough = ck::tensor_operation::element_wise::PassThrough; using AddReluAdd = ck::tensor_operation::element_wise::AddReluAdd; -#if !defined(CK_USE_AMD_MFMA_GFX950) static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding; -#endif + // e = elementwise((a * b), d0, d1) // h = layernorm(e, gamma, beta) // outout: h[m, n] @@ -44,7 +43,7 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instan //#######################################| | | | | | | | | | | | | | Operation| Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _M_N| _NWaveNPerXdl| _M_N| _M| | | //#######################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 2, 8, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, @@ -61,7 +60,6 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instan DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 2, 8, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -74,7 +72,7 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_irregu //#######################################| | | | | | | | | | | | | | Operation| Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _M_N| _NWaveNPerXdl| _M_N| _M| | | //#######################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // pipeline v1, 1 wave DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<16, 4>, 1, S<16, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES @@ -87,8 +85,7 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_irregu , DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Col, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<16, 4>, 1, S<16, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp index 5b57e3b249..40fbc85be0 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp @@ -28,10 +28,9 @@ using S = ck::Sequence; using PassThrough = ck::tensor_operation::element_wise::PassThrough; using AddReluAdd = ck::tensor_operation::element_wise::AddReluAdd; -#if !defined(CK_USE_AMD_MFMA_GFX950) static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding; -#endif + // e = elementwise((a * b), d0, d1) // h = layernorm(e, gamma, beta) // outout: h[m, n] @@ -44,7 +43,7 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instan //#######################################| | | | | | | | | | | | | | Operation| Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _M_N| _NWaveNPerXdl| _M_N| _M| | | //#######################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, @@ -61,7 +60,6 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instan DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -74,7 +72,7 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_irregu //#######################################| | | | | | | | | | | | | | Operation| Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _M_N| _NWaveNPerXdl| _M_N| _M| | | //#######################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // pipeline v1, 1 wave DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<16, 4>, 1, S<16, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES @@ -87,8 +85,7 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_irregu , DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Row, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<16, 4>, 1, S<16, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp index fe372e5513..464279c376 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp @@ -28,10 +28,8 @@ using S = ck::Sequence; using PassThrough = ck::tensor_operation::element_wise::PassThrough; using AddReluAdd = ck::tensor_operation::element_wise::AddReluAdd; -#if !defined(CK_USE_AMD_MFMA_GFX950) static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding; -#endif // e = elementwise((a * b), d0, d1) // h = layernorm(e, gamma, beta) @@ -45,7 +43,7 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instan //#######################################| | | | | | | | | | | | | | Operation| Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _M_N| _NWaveNPerXdl| _M_N| _M| | | //#######################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, 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<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, 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<32, 8>, 8, S<32, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<16, 8>, 8, S<16, 8>, 1, GemmLoopScheduler, GemmPipeline>, @@ -59,7 +57,6 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instan DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<16, 8>, 8, S<16, 8>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<16, 4>, 8, S<16, 4>, 1, GemmLoopScheduler, GemmPipeline>, DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<16, 4>, 8, S<16, 4>, 1, GemmLoopScheduler, GemmPipeline> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -72,7 +69,7 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_irregu //#######################################| | | | | | | | | | | | | | Operation| Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _M_N| _NWaveNPerXdl| _M_N| _M| | | //#######################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // pipeline v1, 1 wave DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<16, 4>, 1, S<16, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES @@ -85,8 +82,7 @@ using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_irregu , DeviceGemmMultipleDLayernorm_Xdl_CShuffle< Row, Col, Row_Row_Tuple, Row, F16, F16, F32, F32, F16_F16_Tuple, F16, F16, F16, F16, PassThrough, PassThrough, AddReluAdd, PassThrough, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<16, 4>, 1, S<16, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on + // clang-format on >; void add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instances( diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp index 16c64cd276..e2bf62ca94 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp @@ -29,10 +29,9 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_generic_instance = //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else - DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, #endif // defined(CK_USE_AMD_MFMA_GFX950) + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> // clang-format on >; using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances = std::tuple< @@ -43,8 +42,8 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances = std::t //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 2, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -101,7 +100,6 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances = std::t DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 2, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -114,8 +112,8 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_irregular_tile_ins //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // pipeline v1, 2 waves @@ -127,7 +125,6 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_irregular_tile_ins , DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp index 1bdfebe977..bbf2836cf8 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp @@ -29,10 +29,9 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_generic_instance = //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else - DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, #endif // defined(CK_USE_AMD_MFMA_GFX950) + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> // clang-format on >; using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances = std::tuple< @@ -43,8 +42,8 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances = std::t //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 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, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 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, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 8, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -101,7 +100,6 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances = std::t DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 8, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 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, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -114,8 +112,8 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_irregular_tile_ins //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // pipeline v1, 2 waves @@ -127,7 +125,6 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_irregular_tile_ins , DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp index 86f7eac1e2..f22739e1a6 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp @@ -29,10 +29,9 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_generic_instance = //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#endif // clang-format on >; using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances = std::tuple< @@ -43,8 +42,8 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances = std::t //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -101,7 +100,6 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances = std::t DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -114,8 +112,8 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_irregular_tile_ins //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // pipeline v1, 2 waves @@ -127,7 +125,6 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_irregular_tile_ins , DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp index 891fafb3ce..b049a5bac1 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp @@ -29,10 +29,9 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_generic_instance = //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#endif // clang-format on >; using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::tuple< @@ -43,8 +42,8 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::t //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, 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, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, @@ -92,7 +91,6 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::t DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -105,8 +103,8 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_irregular_tile_ins //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // pipeline v1, 1 wave #if defined(CK_USE_AMD_MFMA_GFX950) - DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> -#else + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 64, 16, 16, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v1> #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // pipeline v1, 2 waves @@ -118,7 +116,6 @@ using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_irregular_tile_ins , DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmMNKPadding, 1, 64, 16, 16, 32, 8, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, LoopScheduler::Default, PipelineVersion::v2> #endif -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_v1_interwave_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_v1_interwave_instance.cpp index ae1b7293f5..385f474ecc 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_v1_interwave_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_v1_interwave_instance.cpp @@ -39,7 +39,7 @@ using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_v1_iw_instances = std::tuple< //#########################| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) //PipelineVersion::v1; interwave DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16, PipelineVersion::v1, LoopScheduler::Interwave>, DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16, PipelineVersion::v1, LoopScheduler::Interwave>, @@ -59,7 +59,6 @@ using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_v1_iw_instances = std::tuple< DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, true, 1, 1, S<1, 32, 1, 4>, 8, F16, PipelineVersion::v1, LoopScheduler::Interwave>, DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 32, 32, 4, 8, 32, 32, 1, 1, S<1, 2, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, true, 1, 1, S<1, 16, 1, 4>, 8, F16, PipelineVersion::v1, LoopScheduler::Interwave>, DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 16, 32, 4, 8, 16, 16, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, true, 1, 1, S<1, 16, 1, 4>, 4, F16, PipelineVersion::v1, LoopScheduler::Interwave> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_v1_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_v1_irregular_instance.cpp index 04d16d8a06..0b48bbf606 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_v1_irregular_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_v1_irregular_instance.cpp @@ -40,7 +40,7 @@ using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances = std::tup //#########################| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 16, 128, 4, 8, 16, 16, 1, 4, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, true, 1, 1, S<1, 16, 1, 8>, 4, F16, PipVer, LoopSche>, DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 16, 256, 4, 8, 16, 16, 1, 8, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, true, 1, 1, S<1, 16, 1, 8>, 4, F16, PipVer, LoopSche>, DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 4, 8, 16, 16, 1, 4, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, true, 1, 1, S<1, 16, 1, 16>, 4, F16, PipVer, LoopSche>, @@ -65,7 +65,6 @@ using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances = std::tup DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 16, 8, 8, 16, 16, 4, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 8, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, true, 1, 1, S<1, 32, 1, 4>, 4, F16, PipVer, LoopSche>, DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 256, 16, 8, 8, 16, 16, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 8, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, true, 1, 1, S<1, 32, 1, 4>, 4, F16, PipVer, LoopSche>, DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 16, 8, 8, 16, 16, 4, 1, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 8, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, true, 1, 1, S<1, 64, 1, 4>, 4, F16, PipVer, LoopSche> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_v1_interwave_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_v1_interwave_instance.cpp index e7697d6b07..422db05b35 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_v1_interwave_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_v1_interwave_instance.cpp @@ -40,7 +40,7 @@ using device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_v1_iw_instances = std::tuple< //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //PipelineVersion::v1; interwave #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16, PipelineVersion::v1, LoopScheduler::Interwave>, DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16, PipelineVersion::v1, LoopScheduler::Interwave>, DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 8>, 8, F16, PipelineVersion::v1, LoopScheduler::Interwave>, @@ -54,7 +54,6 @@ using device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_v1_iw_instances = std::tuple< DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 8>, 8, F16, PipelineVersion::v1, LoopScheduler::Interwave>, DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 4>, 8, F16, PipelineVersion::v1, LoopScheduler::Interwave>, DeviceGemmXdlSplitKCShuffle< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 4>, 8, F16, PipelineVersion::v1, LoopScheduler::Interwave> -#endif // !defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_km_kn_mn.hpp b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_km_kn_mn.hpp index 01c4a06dcb..5540d2d884 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_km_kn_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_km_kn_mn.hpp @@ -42,7 +42,7 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_km_kn_mn_comp_instances = std::tu //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemm_Xdl_CShuffleV3< Col, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 4, 4, 32, 32, 4, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3< Col, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3< Col, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 2, 2, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, @@ -55,7 +55,6 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_km_kn_mn_comp_instances = std::tu DeviceGemm_Xdl_CShuffleV3< Col, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>, DeviceGemm_Xdl_CShuffleV3< Col, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>, DeviceGemm_Xdl_CShuffleV3< Col, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -67,7 +66,7 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_km_kn_mn_mem_instances = std::tup //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Latency friendly DeviceGemm_Xdl_CShuffleV3< Col, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 4, 4, 16, 16, 1, 1, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Col, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 2, 2, 16, 16, 1, 1, S<32, 4, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, 0, S<32, 4, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, @@ -86,7 +85,6 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_km_kn_mn_mem_instances = std::tup DeviceGemm_Xdl_CShuffleV3< Col, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 16, 128, 64, 4, 4, 16, 16, 1, 4, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 0, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Col, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 64, 2, 4, 16, 16, 1, 4, S<32, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Col, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 64, 2, 2, 16, 16, 1, 4, S<32, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, 0, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; } // namespace instance diff --git a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_km_nk_mn.hpp b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_km_nk_mn.hpp index aa058b2d69..a97953de35 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_km_nk_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_km_nk_mn.hpp @@ -42,7 +42,7 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_km_nk_mn_comp_instances = std::tu //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Compute friendly DeviceGemm_Xdl_CShuffleV3< Col, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 4, 8, 32, 32, 4, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3< Col, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 4, 4, 32, 32, 4, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, @@ -60,7 +60,6 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_km_nk_mn_comp_instances = std::tu DeviceGemm_Xdl_CShuffleV3< Col, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Col, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<16,16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Col, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 2, 2, 32, 32, 2, 2, S<32, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, 0, S<32, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -72,7 +71,7 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_km_nk_mn_mem_instances = std::tup //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Latency friendly DeviceGemm_Xdl_CShuffleV3< Col, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 4, 8, 16, 16, 1, 1, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Col, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 4, 4, 16, 16, 1, 1, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<16, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, @@ -92,7 +91,6 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_km_nk_mn_mem_instances = std::tup DeviceGemm_Xdl_CShuffleV3< Col, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 16, 128, 64, 4, 8, 16, 16, 1, 4, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Col, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 64, 2, 8, 16, 16, 1, 4, S<32, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Col, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 64, 2, 2, 16, 16, 1, 4, S<32, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, 0, S<32, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; } // namespace instance diff --git a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn.hpp b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn.hpp index 1db09e13cb..e9a6fe313d 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn.hpp @@ -42,7 +42,7 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn_comp_instances = std::tu //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 8, 4, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 4, 4, 32, 32, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 2, 2, 32, 32, 4, 4, S<16,16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, 0, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, @@ -53,7 +53,6 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn_comp_instances = std::tu DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -65,7 +64,7 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn_mem_instances = std::tup //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Latency friendly DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 8, 4, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 4, 4, 16, 16, 1, 1, S<16, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, @@ -84,7 +83,6 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn_mem_instances = std::tup DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 16, 64, 64, 8, 4, 16, 16, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 16, 128, 64, 8, 4, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 64, 8, 4, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; } // namespace instance diff --git a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn.hpp b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn.hpp index b015da393f..918ef57a11 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn.hpp @@ -42,7 +42,7 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn_comp_instances = std::tu //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Compute friendly DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 8, 8, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 4, 4, 32, 32, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, @@ -57,7 +57,6 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn_comp_instances = std::tu DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 8, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 8, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 8, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 4, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -69,7 +68,7 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn_mem_instances = std::tup //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Latency friendly DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 16, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, @@ -88,7 +87,6 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn_mem_instances = std::tup DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 16, 64, 64, 8, 8, 16, 16, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 16, 128, 64, 8, 8, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, BF16, BF16, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 64, 8, 8, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; } // namespace instance diff --git a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn.hpp b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn.hpp index a145b938dd..bffa5db2d4 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn.hpp @@ -40,7 +40,7 @@ using device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn_comp_instances = std::tuple //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 8, 4, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 4, 4, 32, 32, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 2, 2, 32, 32, 4, 4, S<16,16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, 0, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, @@ -55,7 +55,6 @@ using device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn_comp_instances = std::tuple DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<16,16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 2, 2, 32, 32, 2, 2, S<32, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, 0, S<32, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -67,7 +66,7 @@ using device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn_mem_instances = std::tuple< //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Latency friendly DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 8, 4, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 4, 4, 16, 16, 1, 1, S<16, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, @@ -98,7 +97,6 @@ using device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn_mem_instances = std::tuple< DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 64, 8, 4, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 64, 4, 4, 16, 16, 1, 4, S<16, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 32, 256, 64, 8, 4, 32, 32, 1, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; } // namespace instance diff --git a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_nk_mn.hpp b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_nk_mn.hpp index be1dde69db..6c21aeb573 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_nk_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_nk_mn.hpp @@ -40,7 +40,7 @@ using device_gemm_xdl_universal_f16_f16_f16_mk_nk_mn_comp_instances = std::tuple //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Compute friendly DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 8, 8, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 4, 4, 32, 32, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, @@ -67,7 +67,6 @@ using device_gemm_xdl_universal_f16_f16_f16_mk_nk_mn_comp_instances = std::tuple DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 8, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -79,7 +78,7 @@ using device_gemm_xdl_universal_f16_f16_f16_mk_nk_mn_mem_instances = std::tuple< //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Latency friendly DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 4, 4, 16, 16, 1, 1, S<16, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<16, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, @@ -114,7 +113,6 @@ using device_gemm_xdl_universal_f16_f16_f16_mk_nk_mn_mem_instances = std::tuple< DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 32, 256, 64, 8, 8, 32, 32, 1, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 32, 256, 64, 4, 4, 32, 32, 1, 2, S<16,16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<16,16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 32, 256, 64, 2, 2, 32, 32, 1, 2, S<32, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, 0, S<32, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; } // namespace instance diff --git a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f8_f16/device_gemm_xdl_universal_f16_f8_f16_mk_kn_mn.hpp b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f8_f16/device_gemm_xdl_universal_f16_f8_f16_mk_kn_mn.hpp index 1cef1f49f2..2650202a64 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f8_f16/device_gemm_xdl_universal_f16_f8_f16_mk_kn_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f8_f16/device_gemm_xdl_universal_f16_f8_f16_mk_kn_mn.hpp @@ -41,7 +41,7 @@ using device_gemm_xdl_universal_f16_f8_f16_mk_kn_mn_comp_instances = std::tuple< //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 8, 4, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, // Disable due to test failure @@ -51,7 +51,6 @@ using device_gemm_xdl_universal_f16_f8_f16_mk_kn_mn_comp_instances = std::tuple< DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 32, 8, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -63,7 +62,7 @@ using device_gemm_xdl_universal_f16_f8_f16_mk_kn_mn_mem_instances = std::tuple< //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Latency friendly DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 16, 16, 256, 8, 4, 16, 16, 1, 1, S<32, 2, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<64, 1, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 16, 32, 256, 8, 4, 16, 16, 1, 1, S<32, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<64, 2, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, @@ -76,7 +75,6 @@ using device_gemm_xdl_universal_f16_f8_f16_mk_kn_mn_mem_instances = std::tuple< DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 64, 8, 4, 32, 32, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 16, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 64, 8, 4, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Row, Row, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 32, 256, 64, 8, 4, 32, 32, 1, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; } // namespace instance diff --git a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f8_f16/device_gemm_xdl_universal_f16_f8_f16_mk_nk_mn.hpp b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f8_f16/device_gemm_xdl_universal_f16_f8_f16_mk_nk_mn.hpp index 26e630eb5d..5e278de812 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f8_f16/device_gemm_xdl_universal_f16_f8_f16_mk_nk_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f8_f16/device_gemm_xdl_universal_f16_f8_f16_mk_nk_mn.hpp @@ -41,14 +41,13 @@ using device_gemm_xdl_universal_f16_f8_f16_mk_nk_mn_comp_instances = std::tuple< //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Compute friendly DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 16, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 224, 256, 64, 8, 16, 16, 16, 7, 8, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 2, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 16, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 16, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 16, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -60,7 +59,7 @@ using device_gemm_xdl_universal_f16_f8_f16_mk_nk_mn_mem_instances = std::tuple< //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Latency friendly DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 128, 8, 16, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 16, 16, 128, 8, 16, 16, 16, 1, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, @@ -81,7 +80,6 @@ using device_gemm_xdl_universal_f16_f8_f16_mk_nk_mn_mem_instances = std::tuple< DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 128, 8, 16, 32, 32, 1, 2, S<16, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 16, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 128, 8, 16, 16, 16, 1, 4, S<16, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3< Row, Col, Row, F16, F8, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 32, 256, 128, 8, 16, 32, 32, 1, 2, S<16, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; } // namespace instance diff --git a/library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn.hpp b/library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn.hpp index 7dc1b701fd..e36b7f3093 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn.hpp @@ -45,7 +45,7 @@ using device_gemm_xdl_universal_reduce_bf16_bf16_bf16_mk_kn_mn_comp_instances = //#########################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, BF16, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 8, 4, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, BF16, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, BF16, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 8, 4, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>, @@ -56,7 +56,6 @@ using device_gemm_xdl_universal_reduce_bf16_bf16_bf16_mk_kn_mn_comp_instances = DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, BF16, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 32, 8, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, BF16, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 128, 32, 8, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, BF16, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -71,7 +70,7 @@ using device_gemm_xdl_universal_reduce_bf16_bf16_bf16_mk_kn_mn_mem_instances = s //#########################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Latency friendly DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, BF16, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 8, 4, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, BF16, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 16, 16, 128, 8, 4, 16, 16, 1, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 4, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, @@ -94,7 +93,6 @@ using device_gemm_xdl_universal_reduce_bf16_bf16_bf16_mk_kn_mn_mem_instances = s DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, BF16, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 64, 8, 4, 32, 32, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, BF16, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 64, 8, 4, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, BF16, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 32, 256, 64, 8, 4, 32, 32, 1, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; } // namespace instance diff --git a/library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_bf16_i8_bf16/device_gemm_xdl_universal_bf16_i8_bf16_mk_kn_mn.hpp b/library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_bf16_i8_bf16/device_gemm_xdl_universal_bf16_i8_bf16_mk_kn_mn.hpp index 617d5f49a5..4e1e5567d5 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_bf16_i8_bf16/device_gemm_xdl_universal_bf16_i8_bf16_mk_kn_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_bf16_i8_bf16/device_gemm_xdl_universal_bf16_i8_bf16_mk_kn_mn.hpp @@ -47,8 +47,8 @@ using device_gemm_xdl_universal_reduce_bf16_i8_bf16_mk_kn_mn_comp_instances = st //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) //DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, I8, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 128, 16, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> - DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, I8, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 128, 16, 4, 32, 32, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v2> -#else + DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, I8, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 128, 16, 4, 32, 32, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v2>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, I8, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 8, 4, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, I8, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, I8, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 8, 4, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>, @@ -57,7 +57,6 @@ using device_gemm_xdl_universal_reduce_bf16_i8_bf16_mk_kn_mn_comp_instances = st DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, I8, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, I8, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 32, 8, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, BF16, I8, DsDataType, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn.hpp b/library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn.hpp index 7ef8c01729..ef1808c551 100644 --- a/library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn.hpp @@ -45,7 +45,7 @@ using device_gemm_xdl_universal_reduce_f16_f16_f16_mk_kn_mn_comp_instances = std //#########################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, F16, F16, DsDataType, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 8, 4, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, F16, F16, DsDataType, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, F16, F16, DsDataType, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 8, 4, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>, @@ -56,7 +56,6 @@ using device_gemm_xdl_universal_reduce_f16_f16_f16_mk_kn_mn_comp_instances = std DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, F16, F16, DsDataType, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 32, 8, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, F16, F16, DsDataType, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 128, 32, 8, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, F16, F16, DsDataType, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -71,7 +70,7 @@ using device_gemm_xdl_universal_reduce_f16_f16_f16_mk_kn_mn_mem_instances = std: //#########################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Latency friendly DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, F16, F16, DsDataType, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 8, 4, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, F16, F16, DsDataType, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 16, 16, 128, 8, 4, 16, 16, 1, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 4, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, @@ -94,7 +93,6 @@ using device_gemm_xdl_universal_reduce_f16_f16_f16_mk_kn_mn_mem_instances = std: DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, F16, F16, DsDataType, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 64, 8, 4, 32, 32, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, F16, F16, DsDataType, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 16, 256, 64, 8, 4, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffleV3R1< Row, Row, DsLayout, Row, F16, F16, DsDataType, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 32, 256, 64, 8, 4, 32, 32, 1, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; } // namespace instance diff --git a/library/src/tensor_operation_instance/gpu/gemm_universal_streamk/device_gemm_xdl_universal_streamk_f16_f16_f16/device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn.hpp b/library/src/tensor_operation_instance/gpu/gemm_universal_streamk/device_gemm_xdl_universal_streamk_f16_f16_f16/device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn.hpp index 938eea8ef5..9e22c8f992 100755 --- a/library/src/tensor_operation_instance/gpu/gemm_universal_streamk/device_gemm_xdl_universal_streamk_f16_f16_f16/device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/gemm_universal_streamk/device_gemm_xdl_universal_streamk_f16_f16_f16/device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn.hpp @@ -40,7 +40,7 @@ using device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn_comp_instances = st //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Compute friendly DeviceGemm_Xdl_CShuffle_Streamk_V3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 8, 8, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, DeviceGemm_Xdl_CShuffle_Streamk_V3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 256, 32, 4, 4, 32, 32, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, @@ -65,7 +65,6 @@ using device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn_comp_instances = st DeviceGemm_Xdl_CShuffle_Streamk_V3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffle_Streamk_V3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffle_Streamk_V3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 64, 8, 8, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -77,7 +76,7 @@ using device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn_mem_instances = std //#########################| | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision| //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if defined(CK_USE_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) // Latency friendly DeviceGemm_Xdl_CShuffle_Streamk_V3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, DeviceGemm_Xdl_CShuffle_Streamk_V3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 16, 64, 4, 4, 16, 16, 1, 1, S<16, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<16, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, @@ -112,7 +111,6 @@ using device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn_mem_instances = std DeviceGemm_Xdl_CShuffle_Streamk_V3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 32, 256, 64, 8, 8, 32, 32, 1, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffle_Streamk_V3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 32, 256, 64, 4, 4, 32, 32, 1, 2, S<16,16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, S<16,16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, DeviceGemm_Xdl_CShuffle_Streamk_V3< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 32, 256, 64, 2, 2, 32, 32, 1, 2, S<32, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, 0, S<32, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; } // namespace instance diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instance.cpp index eeaf6c048d..0bd53706be 100644 --- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instance.cpp @@ -27,10 +27,8 @@ using S = ck::Sequence; using Empty_Tuple = ck::Tuple<>; -using PassThrough = ck::tensor_operation::element_wise::PassThrough; -#if !defined(CK_USE_AMD_MFMA_GFX950) +using PassThrough = ck::tensor_operation::element_wise::PassThrough; static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding; -#endif using device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_tile_instances = std::tuple< // clang-format off //################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| 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| @@ -38,7 +36,7 @@ using device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_tile_instance //################################| | | | | | | | | | | 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_AMD_MFMA_GFX950) -#else +#endif // defined(CK_USE_AMD_MFMA_GFX950) DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, PipelineVersion::v1>, DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, PipelineVersion::v1>, DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, PipelineVersion::v1>, @@ -101,7 +99,6 @@ using device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_tile_instance DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, PipelineVersion::v2>, DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, PipelineVersion::v2>, DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, PipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_tile_loop/device_grouped_gemm_xdl_tile_loop_multiply_bf16_i8_bf16_mk_kn_mn.hpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_tile_loop/device_grouped_gemm_xdl_tile_loop_multiply_bf16_i8_bf16_mk_kn_mn.hpp index e7c02805af..baf04cf12e 100644 --- a/library/src/tensor_operation_instance/gpu/grouped_gemm_tile_loop/device_grouped_gemm_xdl_tile_loop_multiply_bf16_i8_bf16_mk_kn_mn.hpp +++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_tile_loop/device_grouped_gemm_xdl_tile_loop_multiply_bf16_i8_bf16_mk_kn_mn.hpp @@ -50,8 +50,8 @@ using device_grouped_gemm_xdl_tile_loop_bf16_i8_bf16_mk_kn_mn_comp_instances = s //###########################################| | | | | | | | | | | 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| //###########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | S, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, S<8,8,8>, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1> -#else + DeviceGroupedGemmMultipleDXdlCShuffleTileLoop< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, CDEElementwiseOp, GemmSpec, 1, 256, 128, 128, 128, 16, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, S<8,8,8>, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>, +#endif // defined(CK_USE_AMD_MFMA_GFX950) // DeviceGroupedGemmMultipleDXdlCShuffleTileLoop< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, CDEElementwiseOp, GemmSpec, 1, 256, 256, 256, 32, 8, 4, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, S<8,8,1>, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, // DeviceGroupedGemmMultipleDXdlCShuffleTileLoop< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, CDEElementwiseOp, GemmSpec, 1, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, S<8,8,1>, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, // DeviceGroupedGemmMultipleDXdlCShuffleTileLoop< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, CDEElementwiseOp, GemmSpec, 1, 256, 256, 256, 32, 8, 4, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, S<8,8,1>, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>, @@ -60,7 +60,6 @@ using device_grouped_gemm_xdl_tile_loop_bf16_i8_bf16_mk_kn_mn_comp_instances = s // DeviceGroupedGemmMultipleDXdlCShuffleTileLoop< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, CDEElementwiseOp, GemmSpec, 1, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, S<8,8,1>, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>, // DeviceGroupedGemmMultipleDXdlCShuffleTileLoop< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, CDEElementwiseOp, GemmSpec, 1, 256, 128, 256, 32, 8, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, S<8,8,1>, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>, DeviceGroupedGemmMultipleDXdlCShuffleTileLoop< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, CDEElementwiseOp, GemmSpec, 1, 256, 128, 128, 64, 8, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, S<8,8,8>, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1> -#endif // defined(CK_USE_AMD_MFMA_GFX950) // clang-format on >; @@ -69,14 +68,15 @@ template -using device_grouped_gemm_xdl_tile_loop_bf16_i8_bf16_mk_kn_mn_mem_instances = std::tuple< +using device_grouped_gemm_xdl_tile_loop_bf16_i8_bf16_mk_kn_mn_mem_instances = + std::tuple< // clang-format off //###########################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| 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| //###########################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| 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| //###########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | S, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<64, 1, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 16, 1, 4>, S<4,4,1>, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, // DeviceGroupedGemmMultipleDXdlCShuffleTileLoop< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, CDEElementwiseOp, GemmSpec, 1, 128, 16, 32, 256, 8, 4, 16, 16, 1, 1, S<32, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<64, 2, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 16, 1, 8>, S<4,4,1>, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>, @@ -89,9 +89,8 @@ using device_grouped_gemm_xdl_tile_loop_bf16_i8_bf16_mk_kn_mn_mem_instances = st // DeviceGroupedGemmMultipleDXdlCShuffleTileLoop< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, CDEElementwiseOp, GemmSpec, 1, 128, 32, 128, 64, 8, 4, 32, 32, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 16, 1, 8>, S<8,8,1>, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, // DeviceGroupedGemmMultipleDXdlCShuffleTileLoop< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, CDEElementwiseOp, GemmSpec, 1, 256, 16, 256, 64, 8, 4, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 16, 1, 16>, S<4,4,1>, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>, // DeviceGroupedGemmMultipleDXdlCShuffleTileLoop< Row, Row, DsLayout, Row, BF16, I8, F32, F32, DsDataType, BF16, PassThrough, PassThrough, CDEElementwiseOp, GemmSpec, 1, 256, 32, 256, 64, 8, 4, 32, 32, 1, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 16, 1, 16>, S<8,8,1>, BlkGemmPipeSched, BlockGemmPipelineVersion::v2> -#endif // defined(CK_USE_AMD_MFMA_GFX950) - // clang-format on - >; + // clang-format on + >; } // namespace instance } // namespace device