diff --git a/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_dnnl_emitters.cpp b/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_dnnl_emitters.cpp index 0b315cdd309715..4f5f9317cd20e1 100644 --- a/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_dnnl_emitters.cpp +++ b/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_dnnl_emitters.cpp @@ -37,13 +37,13 @@ jit_dnnl_emitter::jit_dnnl_emitter(jit_generator *host, cpu_isa_t host_isa, void jit_dnnl_emitter::set_injector() { if (host_isa_ == cpu::x64::sse41) { - eltwise_injector_sse42 = std::make_shared>( + eltwise_injector_sse42 = std::make_shared>( h, kind, alpha, beta, 1.f); } else if (host_isa_ == cpu::x64::avx2) { - eltwise_injector_avx2 = std::make_shared>( + eltwise_injector_avx2 = std::make_shared>( h, kind, alpha, beta, 1.f); } else if (host_isa_ == cpu::x64::avx512_core) { - eltwise_injector_avx512_core = std::make_shared>( + eltwise_injector_avx512_core = std::make_shared>( h, kind, alpha, beta, 1.f); } else { OV_CPU_JIT_EMITTER_THROW("Unsupported ISA ", host_isa_); diff --git a/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_dnnl_emitters.hpp b/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_dnnl_emitters.hpp index bdf04108370ed5..36499cd08bbdf5 100644 --- a/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_dnnl_emitters.hpp +++ b/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_dnnl_emitters.hpp @@ -34,9 +34,9 @@ class jit_dnnl_emitter : public jit_emitter { float alpha {0.f}; float beta {0.f}; - std::shared_ptr> eltwise_injector_sse42; - std::shared_ptr> eltwise_injector_avx2; - std::shared_ptr> eltwise_injector_avx512_core; + std::shared_ptr> eltwise_injector_sse42; + std::shared_ptr> eltwise_injector_avx2; + std::shared_ptr> eltwise_injector_avx512_core; private: size_t get_inputs_num() const override; diff --git a/src/plugins/intel_cpu/src/nodes/bin_conv.cpp b/src/plugins/intel_cpu/src/nodes/bin_conv.cpp index d1e82235ba9bb1..c045f961f142c4 100644 --- a/src/plugins/intel_cpu/src/nodes/bin_conv.cpp +++ b/src/plugins/intel_cpu/src/nodes/bin_conv.cpp @@ -62,8 +62,8 @@ struct jit_uni_bin_conv_kernel_f32 : public jit_uni_bin_conv_kernel, public jit_ for (int i = 0; i < end_idx; i++) { auto &post_op = p.entry_[i]; if (post_op.is_eltwise()) { - eltwise_injectors.push_back(std::make_shared>( - this, post_op.eltwise, true, eltwise_reserved, mask_post_op_reserved)); + eltwise_injectors.push_back(std::make_shared>( + this, post_op.eltwise, data_type::f32, true, eltwise_reserved, mask_post_op_reserved)); } else if (post_op.is_depthwise()) { depthwise_injectors.push_back(std::make_shared>( this, post_op, mask_post_op_reserved)); @@ -209,7 +209,7 @@ struct jit_uni_bin_conv_kernel_f32 : public jit_uni_bin_conv_kernel, public jit_ Xbyak::Label l_table; - nstl::vector>> eltwise_injectors; + nstl::vector>> eltwise_injectors; nstl::vector>> depthwise_injectors; void cvt2ps(dnnl::memory::data_type type_in, Vmm vmm_in, const Xbyak::Operand &op, bool scalar_load) { diff --git a/src/plugins/intel_cpu/src/nodes/common/softmax.cpp b/src/plugins/intel_cpu/src/nodes/common/softmax.cpp index 66a6ca9c1b6f53..75efc9454a1eaf 100644 --- a/src/plugins/intel_cpu/src/nodes/common/softmax.cpp +++ b/src/plugins/intel_cpu/src/nodes/common/softmax.cpp @@ -62,7 +62,7 @@ struct jit_uni_softmax_kernel_f32 : public jit_uni_softmax_kernel, public jit_ge } void generate() override { - exp_injector.reset(new jit_uni_eltwise_injector_f32(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.0f)); + exp_injector.reset(new jit_uni_eltwise_injector(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.0f)); if (mayiuse(avx512_core)) uni_vcvtneps2bf16.reset(new jit_uni_vcvtneps2bf16(this, isa)); @@ -192,7 +192,7 @@ struct jit_uni_softmax_kernel_f32 : public jit_uni_softmax_kernel, public jit_ge std::unique_ptr uni_vcvtneps2bf16; - std::shared_ptr> exp_injector; + std::shared_ptr> exp_injector; jit_softmax_config_params jcp_; diff --git a/src/plugins/intel_cpu/src/nodes/interpolate.cpp b/src/plugins/intel_cpu/src/nodes/interpolate.cpp index 37008ee17a9603..a81204d865f36f 100644 --- a/src/plugins/intel_cpu/src/nodes/interpolate.cpp +++ b/src/plugins/intel_cpu/src/nodes/interpolate.cpp @@ -73,7 +73,7 @@ struct jit_uni_interpolate_kernel_f32 : public jit_uni_interpolate_kernel, publi for (int i = 0; i < p.len(); i++) { auto &post_op = p.entry_[i]; if (post_op.is_eltwise()) { - eltwise_injectors.push_back(std::make_shared>( + eltwise_injectors.push_back(std::make_shared>( this, post_op.eltwise.alg, post_op.eltwise.alpha, @@ -273,7 +273,7 @@ struct jit_uni_interpolate_kernel_f32 : public jit_uni_interpolate_kernel, publi std::vector store_pool_vec_idxs; std::vector load_pool_gpr_idxs; - std::vector>> eltwise_injectors; + std::vector>> eltwise_injectors; std::vector>> depthwise_injectors; std::vector>> quantization_injectors; diff --git a/src/plugins/intel_cpu/src/nodes/kernels/x64/mlp_kernel.cpp b/src/plugins/intel_cpu/src/nodes/kernels/x64/mlp_kernel.cpp index fd00fca8431ff0..fd050d4526136d 100644 --- a/src/plugins/intel_cpu/src/nodes/kernels/x64/mlp_kernel.cpp +++ b/src/plugins/intel_cpu/src/nodes/kernels/x64/mlp_kernel.cpp @@ -548,12 +548,13 @@ void GateUpCombine::generate() { const auto zmm_up = zmm0; const auto ymm_dst = ymm5; - auto injector = std::make_shared>( + auto injector = std::make_shared>( this, m_act_alg, 1.f, 1.0f, 1.f, + data_type::f32, true, // save_state, true due to additional r15 is used. Xbyak::Reg64(Xbyak::Operand::R10), // p_table Xbyak::Opmask(1), // k_mask diff --git a/src/plugins/intel_cpu/src/nodes/kernels/x64/non_max_suppression.cpp b/src/plugins/intel_cpu/src/nodes/kernels/x64/non_max_suppression.cpp index 6ec25067cec279..17b414987ee155 100644 --- a/src/plugins/intel_cpu/src/nodes/kernels/x64/non_max_suppression.cpp +++ b/src/plugins/intel_cpu/src/nodes/kernels/x64/non_max_suppression.cpp @@ -18,7 +18,7 @@ void NonMaxSuppression::generate() { load_vector_emitter.reset(new jit_load_emitter(this, isa, ov::element::f32, ov::element::f32, vector_step)); load_scalar_emitter.reset(new jit_load_emitter(this, isa, ov::element::f32, ov::element::f32, scalar_step)); - exp_injector.reset(new x64::jit_uni_eltwise_injector_f32(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.f)); + exp_injector.reset(new x64::jit_uni_eltwise_injector(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.f)); this->preamble(); diff --git a/src/plugins/intel_cpu/src/nodes/kernels/x64/non_max_suppression.hpp b/src/plugins/intel_cpu/src/nodes/kernels/x64/non_max_suppression.hpp index 2b5cf495c426e2..65f320b46e870a 100644 --- a/src/plugins/intel_cpu/src/nodes/kernels/x64/non_max_suppression.hpp +++ b/src/plugins/intel_cpu/src/nodes/kernels/x64/non_max_suppression.hpp @@ -112,7 +112,7 @@ class NonMaxSuppression : public JitKernel { Xbyak::Opmask k_mask = Xbyak::Opmask(7); Xbyak::Opmask k_mask_one = Xbyak::Opmask(6); - std::shared_ptr> exp_injector; + std::shared_ptr> exp_injector; inline void hard_nms(); diff --git a/src/plugins/intel_cpu/src/nodes/mvn.cpp b/src/plugins/intel_cpu/src/nodes/mvn.cpp index 6405a3f1511286..67dc2cd425dd95 100644 --- a/src/plugins/intel_cpu/src/nodes/mvn.cpp +++ b/src/plugins/intel_cpu/src/nodes/mvn.cpp @@ -914,7 +914,7 @@ struct jit_uni_mvn_kernel_f32 : public jit_uni_mvn_kernel, public jit_generator for (int i = 0; i < p.len(); i++) { auto &post_op = p.entry_[i]; if (post_op.is_eltwise()) { - eltwise_injectors.push_back(std::make_shared>( + eltwise_injectors.push_back(std::make_shared>( this, post_op.eltwise.alg, post_op.eltwise.alpha, post_op.eltwise.beta, post_op.eltwise.scale)); } else if (post_op.is_depthwise()) { depthwise_injectors.push_back(std::make_shared>( @@ -1040,7 +1040,7 @@ struct jit_uni_mvn_kernel_f32 : public jit_uni_mvn_kernel, public jit_generator const int tile_size[kTileNum] = {8, 4, 2, 1}; - std::vector>> eltwise_injectors; + std::vector>> eltwise_injectors; std::vector>> depthwise_injectors; std::vector>> quantization_injectors; diff --git a/src/plugins/intel_cpu/src/nodes/normalize.cpp b/src/plugins/intel_cpu/src/nodes/normalize.cpp index ca52e572b73ea8..4b463456347af1 100644 --- a/src/plugins/intel_cpu/src/nodes/normalize.cpp +++ b/src/plugins/intel_cpu/src/nodes/normalize.cpp @@ -222,7 +222,7 @@ struct jit_uni_normalize_kernel_f32 : public jit_uni_normalize_kernel, public ji for (int i = 0; i < p.len(); i++) { auto &post_op = p.entry_[i]; if (post_op.is_eltwise()) { - eltwise_injectors.push_back(std::make_shared>( + eltwise_injectors.push_back(std::make_shared>( this, post_op.eltwise.alg, post_op.eltwise.alpha, post_op.eltwise.beta, post_op.eltwise.scale)); } else if (post_op.is_depthwise()) { depthwise_injectors.push_back(std::make_shared>( @@ -302,7 +302,7 @@ struct jit_uni_normalize_kernel_f32 : public jit_uni_normalize_kernel, public ji std::unique_ptr uni_vcvtneps2bf16 = nullptr; - std::vector>> eltwise_injectors; + std::vector>> eltwise_injectors; std::vector>> depthwise_injectors; std::vector>> quantization_injectors; diff --git a/src/plugins/intel_cpu/src/nodes/reduce.cpp b/src/plugins/intel_cpu/src/nodes/reduce.cpp index 1bc0209e0d9c69..26ad3f94d87ed9 100644 --- a/src/plugins/intel_cpu/src/nodes/reduce.cpp +++ b/src/plugins/intel_cpu/src/nodes/reduce.cpp @@ -128,7 +128,7 @@ struct jit_uni_reduce_kernel_f32 : public jit_uni_reduce_kernel, public jit_gene void generate() override { if (jcp_.reduce_mode == Algorithm::ReduceLogSumExp) { - exp_injector = std::make_shared>(this, alg_kind::eltwise_exp, 0.f, 0.f, 1.f); + exp_injector = std::make_shared>(this, alg_kind::eltwise_exp, 0.f, 0.f, 1.f); } if (mayiuse(avx512_core)) @@ -223,7 +223,7 @@ struct jit_uni_reduce_kernel_f32 : public jit_uni_reduce_kernel, public jit_gene Xbyak::Label l_table; std::shared_ptr uni_vcvtneps2bf16; - std::shared_ptr> exp_injector; + std::shared_ptr> exp_injector; inline void reduce_main() { // ================================================================ @@ -1183,7 +1183,7 @@ struct jit_uni_reduce_post_kernel_f32 : public jit_uni_reduce_post_kernel, publi for (int i = 0; i < p.len(); i++) { auto &post_op = p.entry_[i]; if (post_op.is_eltwise()) { - eltwise_injectors.push_back(std::make_shared>( + eltwise_injectors.push_back(std::make_shared>( this, post_op.eltwise.alg, post_op.eltwise.alpha, post_op.eltwise.beta, post_op.eltwise.scale)); } else if (post_op.is_depthwise()) { depthwise_injectors.push_back(std::make_shared>( @@ -1195,7 +1195,7 @@ struct jit_uni_reduce_post_kernel_f32 : public jit_uni_reduce_post_kernel, publi } if (jcp_.reduce_mode == Algorithm::ReduceLogSum || jcp_.reduce_mode == Algorithm::ReduceLogSumExp) { - log_injector = std::make_shared>(this, alg_kind::eltwise_log, 0.f, 0.f, 1.f); + log_injector = std::make_shared>(this, alg_kind::eltwise_log, 0.f, 0.f, 1.f); } if (mayiuse(avx512_core)) @@ -1306,9 +1306,9 @@ struct jit_uni_reduce_post_kernel_f32 : public jit_uni_reduce_post_kernel, publi Vmm vmm_d_bias = Vmm(8); std::shared_ptr uni_vcvtneps2bf16; - std::shared_ptr> log_injector; + std::shared_ptr> log_injector; - std::vector>> eltwise_injectors; + std::vector>> eltwise_injectors; std::vector>> depthwise_injectors; std::vector>> quantization_injectors; diff --git a/src/plugins/intel_cpu/src/nodes/region_yolo.cpp b/src/plugins/intel_cpu/src/nodes/region_yolo.cpp index c575237f48e042..e362c70a433029 100644 --- a/src/plugins/intel_cpu/src/nodes/region_yolo.cpp +++ b/src/plugins/intel_cpu/src/nodes/region_yolo.cpp @@ -40,7 +40,7 @@ struct jit_uni_logistic_kernel_f32 : public jit_uni_logistic_kernel, public jit_ } void generate() override { - exp_injector.reset(new jit_uni_eltwise_injector_f32(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.f)); + exp_injector.reset(new jit_uni_eltwise_injector(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.f)); if (mayiuse(avx512_core)) uni_vcvtneps2bf16.reset(new jit_uni_vcvtneps2bf16(this, isa)); @@ -124,7 +124,7 @@ struct jit_uni_logistic_kernel_f32 : public jit_uni_logistic_kernel, public jit_ Xbyak::Label l_table; - std::shared_ptr> exp_injector; + std::shared_ptr> exp_injector; jit_logistic_config_params jcp_; diff --git a/src/plugins/intel_cpu/thirdparty/onednn b/src/plugins/intel_cpu/thirdparty/onednn index c60a9946aa2386..1e458a29c50a42 160000 --- a/src/plugins/intel_cpu/thirdparty/onednn +++ b/src/plugins/intel_cpu/thirdparty/onednn @@ -1 +1 @@ -Subproject commit c60a9946aa2386890e5c9f5587974facb7624227 +Subproject commit 1e458a29c50a4299d61af7ecba07be1b4bbd91b0