try to migrate onednn3.6

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<jit_uni_eltwise_injector_f32<cpu::x64::sse41>>(
+        eltwise_injector_sse42 = std::make_shared<jit_uni_eltwise_injector<cpu::x64::sse41>>(
                 h, kind, alpha, beta, 1.f);
     } else if (host_isa_ == cpu::x64::avx2) {
-        eltwise_injector_avx2 = std::make_shared<jit_uni_eltwise_injector_f32<cpu::x64::avx2>>(
+        eltwise_injector_avx2 = std::make_shared<jit_uni_eltwise_injector<cpu::x64::avx2>>(
                 h, kind, alpha, beta, 1.f);
     } else if (host_isa_ == cpu::x64::avx512_core) {
-        eltwise_injector_avx512_core = std::make_shared<jit_uni_eltwise_injector_f32<cpu::x64::avx512_core>>(
+        eltwise_injector_avx512_core = std::make_shared<jit_uni_eltwise_injector<cpu::x64::avx512_core>>(
                 h, kind, alpha, beta, 1.f);
     } else {
         OV_CPU_JIT_EMITTER_THROW("Unsupported ISA ", host_isa_);

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<dnnl::impl::cpu::x64::jit_uni_eltwise_injector_f32<dnnl::impl::cpu::x64::sse41>> eltwise_injector_sse42;
-    std::shared_ptr<dnnl::impl::cpu::x64::jit_uni_eltwise_injector_f32<dnnl::impl::cpu::x64::avx2>> eltwise_injector_avx2;
-    std::shared_ptr<dnnl::impl::cpu::x64::jit_uni_eltwise_injector_f32<dnnl::impl::cpu::x64::avx512_core>> eltwise_injector_avx512_core;
+    std::shared_ptr<dnnl::impl::cpu::x64::jit_uni_eltwise_injector<dnnl::impl::cpu::x64::sse41>> eltwise_injector_sse42;
+    std::shared_ptr<dnnl::impl::cpu::x64::jit_uni_eltwise_injector<dnnl::impl::cpu::x64::avx2>> eltwise_injector_avx2;
+    std::shared_ptr<dnnl::impl::cpu::x64::jit_uni_eltwise_injector<dnnl::impl::cpu::x64::avx512_core>> eltwise_injector_avx512_core;
 
 private:
     size_t get_inputs_num() const override;

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<jit_uni_eltwise_injector_f32<isa>>(
-                        this, post_op.eltwise, true, eltwise_reserved, mask_post_op_reserved));
+                eltwise_injectors.push_back(std::make_shared<jit_uni_eltwise_injector<isa>>(
+                        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<jit_uni_depthwise_injector_f32<isa>>(
                         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<std::shared_ptr<jit_uni_eltwise_injector_f32<isa>>> eltwise_injectors;
+    nstl::vector<std::shared_ptr<jit_uni_eltwise_injector<isa>>> eltwise_injectors;
     nstl::vector<std::shared_ptr<jit_uni_depthwise_injector_f32<isa>>> depthwise_injectors;
 
     void cvt2ps(dnnl::memory::data_type type_in, Vmm vmm_in, const Xbyak::Operand &op, bool scalar_load) {

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<isa>(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.0f));
+        exp_injector.reset(new jit_uni_eltwise_injector<isa>(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<jit_uni_vcvtneps2bf16> uni_vcvtneps2bf16;
 
-    std::shared_ptr<jit_uni_eltwise_injector_f32<isa>> exp_injector;
+    std::shared_ptr<jit_uni_eltwise_injector<isa>> exp_injector;
 
     jit_softmax_config_params jcp_;
 

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<jit_uni_eltwise_injector_f32<isa>>(
+                eltwise_injectors.push_back(std::make_shared<jit_uni_eltwise_injector<isa>>(
                         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<size_t> store_pool_vec_idxs;
     std::vector<size_t> load_pool_gpr_idxs;
 
-    std::vector<std::shared_ptr<jit_uni_eltwise_injector_f32<isa>>> eltwise_injectors;
+    std::vector<std::shared_ptr<jit_uni_eltwise_injector<isa>>> eltwise_injectors;
     std::vector<std::shared_ptr<jit_uni_depthwise_injector_f32<isa>>> depthwise_injectors;
     std::vector<std::shared_ptr<jit_uni_quantization_injector_f32<isa>>> quantization_injectors;
 

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<jit_uni_eltwise_injector_f32<dnnl::impl::cpu::x64::avx512_core>>(
+    auto injector = std::make_shared<jit_uni_eltwise_injector<dnnl::impl::cpu::x64::avx512_core>>(
         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

src/plugins/intel_cpu/src/nodes/kernels/x64/non_max_suppression.cpp

@@ -18,7 +18,7 @@ void NonMaxSuppression<isa>::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<isa>(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.f));
+    exp_injector.reset(new x64::jit_uni_eltwise_injector<isa>(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.f));
 
     this->preamble();
 

src/plugins/intel_cpu/src/nodes/kernels/x64/non_max_suppression.hpp

@@ -112,7 +112,7 @@ class NonMaxSuppression : public JitKernel<NmsCompileParams, NmsCallArgs> {
     Xbyak::Opmask k_mask = Xbyak::Opmask(7);
     Xbyak::Opmask k_mask_one = Xbyak::Opmask(6);
 
-    std::shared_ptr<dnnl::impl::cpu::x64::jit_uni_eltwise_injector_f32<isa>> exp_injector;
+    std::shared_ptr<dnnl::impl::cpu::x64::jit_uni_eltwise_injector<isa>> exp_injector;
 
     inline void hard_nms();
 

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<jit_uni_eltwise_injector_f32<isa>>(
+                eltwise_injectors.push_back(std::make_shared<jit_uni_eltwise_injector<isa>>(
                         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<jit_uni_depthwise_injector_f32<isa>>(
@@ -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<std::shared_ptr<jit_uni_eltwise_injector_f32<isa>>> eltwise_injectors;
+    std::vector<std::shared_ptr<jit_uni_eltwise_injector<isa>>> eltwise_injectors;
     std::vector<std::shared_ptr<jit_uni_depthwise_injector_f32<isa>>> depthwise_injectors;
     std::vector<std::shared_ptr<jit_uni_quantization_injector_f32<isa>>> quantization_injectors;
 

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<jit_uni_eltwise_injector_f32<isa>>(
+                eltwise_injectors.push_back(std::make_shared<jit_uni_eltwise_injector<isa>>(
                         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<jit_uni_depthwise_injector_f32<isa>>(
@@ -302,7 +302,7 @@ struct jit_uni_normalize_kernel_f32 : public jit_uni_normalize_kernel, public ji
 
     std::unique_ptr<jit_uni_vcvtneps2bf16> uni_vcvtneps2bf16 = nullptr;
 
-    std::vector<std::shared_ptr<jit_uni_eltwise_injector_f32<isa>>> eltwise_injectors;
+    std::vector<std::shared_ptr<jit_uni_eltwise_injector<isa>>> eltwise_injectors;
     std::vector<std::shared_ptr<jit_uni_depthwise_injector_f32<isa>>> depthwise_injectors;
     std::vector<std::shared_ptr<jit_uni_quantization_injector_f32<isa>>> quantization_injectors;
 

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<jit_uni_eltwise_injector_f32<isa>>(this, alg_kind::eltwise_exp, 0.f, 0.f, 1.f);
+            exp_injector = std::make_shared<jit_uni_eltwise_injector<isa>>(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<jit_uni_vcvtneps2bf16> uni_vcvtneps2bf16;
-    std::shared_ptr<jit_uni_eltwise_injector_f32<isa>> exp_injector;
+    std::shared_ptr<jit_uni_eltwise_injector<isa>> 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<jit_uni_eltwise_injector_f32<isa>>(
+                eltwise_injectors.push_back(std::make_shared<jit_uni_eltwise_injector<isa>>(
                         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<jit_uni_depthwise_injector_f32<isa>>(
@@ -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<jit_uni_eltwise_injector_f32<isa>>(this, alg_kind::eltwise_log, 0.f, 0.f, 1.f);
+            log_injector = std::make_shared<jit_uni_eltwise_injector<isa>>(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<jit_uni_vcvtneps2bf16> uni_vcvtneps2bf16;
-    std::shared_ptr<jit_uni_eltwise_injector_f32<isa>> log_injector;
+    std::shared_ptr<jit_uni_eltwise_injector<isa>> log_injector;
 
-    std::vector<std::shared_ptr<jit_uni_eltwise_injector_f32<isa>>> eltwise_injectors;
+    std::vector<std::shared_ptr<jit_uni_eltwise_injector<isa>>> eltwise_injectors;
     std::vector<std::shared_ptr<jit_uni_depthwise_injector_f32<isa>>> depthwise_injectors;
     std::vector<std::shared_ptr<jit_uni_quantization_injector_f32<isa>>> quantization_injectors;
 

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<isa>(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.f));
+        exp_injector.reset(new jit_uni_eltwise_injector<isa>(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<jit_uni_eltwise_injector_f32<isa>> exp_injector;
+    std::shared_ptr<jit_uni_eltwise_injector<isa>> exp_injector;
 
     jit_logistic_config_params jcp_;