[CPU][ARM]: Implemented CPU plugin just-in-time emitter for Negative …

…operation (openvinotoolkit#28258) ### Details: - Added JIT emitter for Eltwise Negation operation on ARM64 SIMD - Implemented fp32 optimization replacing C++ Math implementation - Modified ARM64 executor to support new JIT emitter - Updated kernel files to include Negation in Eltwise operations - Added test coverage for JIT implementation verification - Transitioned operation type from Math to Eltwise for better performance ### Tickets: - openvinotoolkit#27500 ![image](https://github.com/user-attachments/assets/2dd781da-94c7-4edc-abbb-e7a048d00944)
MirceaDan99 · Jan 22, 2025 · 2295ee2 · 2295ee2
1 parent c421a52
commit 2295ee2
Show file tree

Hide file tree

Showing 10 changed files with 88 additions and 2 deletions.
diff --git a/src/plugins/intel_cpu/src/cpu_types.cpp b/src/plugins/intel_cpu/src/cpu_types.cpp
@@ -197,12 +197,12 @@ static const TypeToNameMap& get_type_to_name_tbl() {
         {"Atanh", Type::Math},
         {"Ceil", Type::Math},
         {"Ceiling", Type::Eltwise},
+        {"Negative", Type::Eltwise},
         {"Cos", Type::Math},
         {"Cosh", Type::Math},
         {"Floor", Type::Eltwise},
         {"HardSigmoid", Type::Math},
         {"If", Type::If},
-        {"Neg", Type::Math},
         {"Reciprocal", Type::Math},
         {"Selu", Type::Math},
         {"Sign", Type::Math},
@@ -425,6 +425,7 @@ std::string algToString(const Algorithm alg) {
         CASE(EltwiseFloor);
         CASE(EltwiseCeiling);
         CASE(EltwiseFloorMod);
+        CASE(EltwiseNegative);
         CASE(EltwiseMod);
         CASE(EltwiseMaximum);
         CASE(EltwiseMinimum);

diff --git a/src/plugins/intel_cpu/src/cpu_types.h b/src/plugins/intel_cpu/src/cpu_types.h
@@ -167,6 +167,7 @@ enum class Algorithm {
     EltwiseFloor,
     EltwiseCeiling,
     EltwiseFloorMod,
+    EltwiseNegative,
     EltwiseMod,
     EltwiseMaximum,
     EltwiseMinimum,

diff --git a/src/plugins/intel_cpu/src/emitters/plugin/aarch64/jit_eltwise_emitters.cpp b/src/plugins/intel_cpu/src/emitters/plugin/aarch64/jit_eltwise_emitters.cpp
@@ -653,6 +653,46 @@ std::set<std::vector<element::Type>> jit_ceiling_emitter::get_supported_precisio
     return {{element::f32}};
 }
 
+/// NEGATIVE ///
+jit_negative_emitter::jit_negative_emitter(dnnl::impl::cpu::aarch64::jit_generator* host,
+                                           dnnl::impl::cpu::aarch64::cpu_isa_t host_isa,
+                                           const std::shared_ptr<ov::Node>& node)
+    : jit_emitter(host, host_isa, node, get_arithmetic_binary_exec_precision(node)) {}
+
+jit_negative_emitter::jit_negative_emitter(dnnl::impl::cpu::aarch64::jit_generator* host,
+                                           dnnl::impl::cpu::aarch64::cpu_isa_t host_isa,
+                                           const ov::element::Type exec_prc)
+    : jit_emitter(host, host_isa, exec_prc) {}
+
+size_t jit_negative_emitter::get_inputs_count() const {
+    return 1;
+}
+
+void jit_negative_emitter::emit_impl(const std::vector<size_t>& in_vec_idxs,
+                                     const std::vector<size_t>& out_vec_idxs) const {
+    if (host_isa_ == dnnl::impl::cpu::aarch64::asimd) {
+        emit_isa<dnnl::impl::cpu::aarch64::asimd>(in_vec_idxs, out_vec_idxs);
+    } else {
+        OV_CPU_JIT_EMITTER_THROW("Can't create jit eltwise kernel");
+    }
+}
+
+template <dnnl::impl::cpu::aarch64::cpu_isa_t isa>
+void jit_negative_emitter::emit_isa(const std::vector<size_t>& in_vec_idxs,
+                                    const std::vector<size_t>& out_vec_idxs) const {
+    OV_CPU_JIT_EMITTER_ASSERT(exec_prc_ == ov::element::f32, "unsupported precision: " + exec_prc_.to_string());
+
+    using TReg = typename dnnl::impl::cpu::aarch64::cpu_isa_traits<isa>::TReg;
+    TReg src = TReg(in_vec_idxs[0]);
+    TReg dst = TReg(out_vec_idxs[0]);
+    h->fneg(dst.s, src.s);
+}
+
+std::set<std::vector<element::Type>> jit_negative_emitter::get_supported_precisions(
+    const std::shared_ptr<ov::Node>& node) {
+    return {{element::f32}};
+}
+
 /// GELU_ERF ///
 jit_gelu_erf_emitter::jit_gelu_erf_emitter(dnnl::impl::cpu::aarch64::jit_generator* host,
                                            dnnl::impl::cpu::aarch64::cpu_isa_t host_isa,

diff --git a/src/plugins/intel_cpu/src/emitters/plugin/aarch64/jit_eltwise_emitters.hpp b/src/plugins/intel_cpu/src/emitters/plugin/aarch64/jit_eltwise_emitters.hpp
@@ -271,6 +271,28 @@ class jit_ceiling_emitter : public jit_emitter {
     void emit_isa(const std::vector<size_t>& in_vec_idxs, const std::vector<size_t>& out_vec_idxs) const;
 };
 
+class jit_negative_emitter : public jit_emitter {
+public:
+    jit_negative_emitter(dnnl::impl::cpu::aarch64::jit_generator* host,
+                         dnnl::impl::cpu::aarch64::cpu_isa_t host_isa,
+                         const ov::element::Type exec_prc = ov::element::f32);
+
+    jit_negative_emitter(dnnl::impl::cpu::aarch64::jit_generator* host,
+                         dnnl::impl::cpu::aarch64::cpu_isa_t host_isa,
+                         const std::shared_ptr<ov::Node>& node);
+
+    size_t get_inputs_count() const override;
+
+    static std::set<std::vector<element::Type>> get_supported_precisions(
+        const std::shared_ptr<ov::Node>& node = nullptr);
+
+private:
+    void emit_impl(const std::vector<size_t>& in_vec_idxs, const std::vector<size_t>& out_vec_idxs) const override;
+
+    template <dnnl::impl::cpu::aarch64::cpu_isa_t isa>
+    void emit_isa(const std::vector<size_t>& in_vec_idxs, const std::vector<size_t>& out_vec_idxs) const;
+};
+
 class jit_gelu_erf_emitter : public jit_emitter {
 public:
     jit_gelu_erf_emitter(dnnl::impl::cpu::aarch64::jit_generator* host,

diff --git a/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_eltwise_emitters.cpp b/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_eltwise_emitters.cpp
@@ -1999,6 +1999,10 @@ jit_negative_emitter::jit_negative_emitter(x64::jit_generator* host,
                                            const std::shared_ptr<ov::Node>& node,
                                            ov::element::Type exec_prc)
     : jit_emitter(host, host_isa, exec_prc) {}
+jit_negative_emitter::jit_negative_emitter(x64::jit_generator* host,
+                                           x64::cpu_isa_t host_isa,
+                                           ov::element::Type exec_prc)
+    : jit_emitter(host, host_isa, exec_prc) {}
 
 size_t jit_negative_emitter::get_inputs_num() const {
     return 1;

diff --git a/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_eltwise_emitters.hpp b/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_eltwise_emitters.hpp
@@ -599,6 +599,9 @@ class jit_sqrt_emitter : public jit_emitter {
 
 class jit_negative_emitter : public jit_emitter {
 public:
+    jit_negative_emitter(dnnl::impl::cpu::x64::jit_generator* host,
+                         dnnl::impl::cpu::x64::cpu_isa_t host_isa,
+                         ov::element::Type exec_prc = ov::element::f32);
     jit_negative_emitter(dnnl::impl::cpu::x64::jit_generator* host,
                          dnnl::impl::cpu::x64::cpu_isa_t host_isa,
                          const std::shared_ptr<ov::Node>& n,
@@ -609,7 +612,7 @@ class jit_negative_emitter : public jit_emitter {
         const std::shared_ptr<ov::Node>& node = nullptr);
 
 private:
-    void emit_impl(const std::vector<size_t>& in, const std::vector<size_t>& out) const override;
+    void emit_impl(const std::vector<size_t>& in_vec_idxs, const std::vector<size_t>& out_vec_idxs) const override;
 
     template <dnnl::impl::cpu::x64::cpu_isa_t isa>
     void emit_isa(const std::vector<size_t>& in_vec_idxs, const std::vector<size_t>& out_vec_idxs) const;

diff --git a/src/plugins/intel_cpu/src/nodes/eltwise.cpp b/src/plugins/intel_cpu/src/nodes/eltwise.cpp
@@ -265,6 +265,7 @@ std::set<std::vector<element::Type>> eltwise_precision_helper::get_supported_pre
               OV_CASE(Algorithm::EltwiseDivide, jit_divide_emitter),
               OV_CASE(Algorithm::EltwiseFloor, jit_floor_emitter),
               OV_CASE(Algorithm::EltwiseCeiling, jit_ceiling_emitter),
+              OV_CASE(Algorithm::EltwiseNegative, jit_negative_emitter),
               OV_CASE(Algorithm::EltwiseFloorMod, jit_floor_mod_emitter),
               OV_CASE(Algorithm::EltwiseMod, jit_mod_emitter),
               OV_CASE(Algorithm::EltwiseMaximum, jit_maximum_emitter),
@@ -664,6 +665,7 @@ struct jit_uni_eltwise_generic : public jit_uni_eltwise_kernel, public jit_gener
                   OV_CASE(Algorithm::EltwiseDivide, jit_divide_emitter),
                   OV_CASE(Algorithm::EltwiseFloor, jit_floor_emitter),
                   OV_CASE(Algorithm::EltwiseCeiling, jit_ceiling_emitter),
+                  OV_CASE(Algorithm::EltwiseNegative, jit_negative_emitter),
                   OV_CASE(Algorithm::EltwiseFloorMod, jit_floor_mod_emitter),
                   OV_CASE(Algorithm::EltwiseMod, jit_mod_emitter),
                   OV_CASE(Algorithm::EltwiseMaximum, jit_maximum_emitter),
@@ -1149,6 +1151,9 @@ const std::map<const ov::DiscreteTypeInfo, Eltwise::Initializer>& Eltwise::getIn
         {ov::op::v0::Ceiling::get_type_info_static(), [](const std::shared_ptr<ov::Node>& op, Eltwise& node) {
             node.algorithm = Algorithm::EltwiseCeiling;
         }},
+        {ov::op::v0::Negative::get_type_info_static(), [](const std::shared_ptr<ov::Node>& op, Eltwise& node) {
+            node.algorithm = Algorithm::EltwiseNegative;
+        }},
         {ov::op::v0::Floor::get_type_info_static(), [](const std::shared_ptr<ov::Node>& op, Eltwise& node) {
             node.algorithm = Algorithm::EltwiseFloor;
         }},
@@ -1979,6 +1984,9 @@ class EltwiseRefExecutor : public EltwiseRefBaseExecutor<T> {
                 case Algorithm::EltwiseFloor:
                     *dst_ptr_f = floorf(src_f[0]);
                     break;
+                case Algorithm::EltwiseNegative:
+                    *dst_ptr_f = -src_f[0];
+                    break;
                 case Algorithm::EltwiseFloorMod:
                     *dst_ptr_f = src_f[0] - floorf(src_f[0] / src_f[1]) * src_f[1];
                     break;
@@ -2230,6 +2238,7 @@ size_t Eltwise::getOpInputsNum() const {
     case Algorithm::EltwiseGeluErf:
     case Algorithm::EltwiseGeluTanh:
     case Algorithm::EltwiseCeiling:
+    case Algorithm::EltwiseNegative:
     case Algorithm::EltwiseFloor:
     case Algorithm::EltwiseElu:
     case Algorithm::EltwiseTanh:

diff --git a/src/plugins/intel_cpu/src/nodes/executors/aarch64/jit_eltwise.cpp b/src/plugins/intel_cpu/src/nodes/executors/aarch64/jit_eltwise.cpp
@@ -28,6 +28,7 @@ bool JitEltwiseExecutor::isSupported(const Algorithm& algorithm,
                                      Algorithm::EltwiseFloor,
                                      Algorithm::EltwiseFloorMod,
                                      Algorithm::EltwiseCeiling,
+                                     Algorithm::EltwiseNegative,
                                      Algorithm::EltwiseGeluErf,
                                      Algorithm::EltwiseGeluTanh,
                                      Algorithm::EltwiseGreater,

diff --git a/src/plugins/intel_cpu/src/nodes/kernels/aarch64/jit_uni_eltwise_generic.cpp b/src/plugins/intel_cpu/src/nodes/kernels/aarch64/jit_uni_eltwise_generic.cpp
@@ -664,6 +664,7 @@ std::shared_ptr<jit_emitter> jit_uni_eltwise_generic<isa>::create_eltwise_emitte
         OV_CASE(Algorithm::EltwiseFloor, ov::intel_cpu::aarch64::jit_floor_emitter),
         OV_CASE(Algorithm::EltwiseFloorMod, ov::intel_cpu::aarch64::jit_floor_mod_emitter),
         OV_CASE(Algorithm::EltwiseCeiling, ov::intel_cpu::aarch64::jit_ceiling_emitter),
+        OV_CASE(Algorithm::EltwiseNegative, ov::intel_cpu::aarch64::jit_negative_emitter),
         OV_CASE(Algorithm::EltwiseHswish, ov::intel_cpu::aarch64::jit_hswish_emitter),
         OV_CASE(Algorithm::EltwiseIsFinite, ov::intel_cpu::aarch64::jit_is_finite_emitter),
         OV_CASE(Algorithm::EltwiseIsInf, ov::intel_cpu::aarch64::jit_is_inf_emitter),
@@ -851,6 +852,7 @@ std::set<std::vector<element::Type>> eltwise_precision_helper::get_supported_pre
               OV_CASE(Algorithm::EltwiseFloor, jit_floor_emitter),
               OV_CASE(Algorithm::EltwiseFloorMod, jit_floor_mod_emitter),
               OV_CASE(Algorithm::EltwiseCeiling, jit_ceiling_emitter),
+              OV_CASE(Algorithm::EltwiseNegative, jit_negative_emitter),
               OV_CASE(Algorithm::EltwiseGeluErf, jit_gelu_erf_emitter),
               OV_CASE(Algorithm::EltwiseGeluTanh, jit_gelu_tanh_emitter),
               OV_CASE(Algorithm::EltwiseGreater, jit_greater_emitter),

diff --git a/src/plugins/intel_cpu/tests/functional/custom/single_layer_tests/classes/activation.cpp b/src/plugins/intel_cpu/tests/functional/custom/single_layer_tests/classes/activation.cpp
@@ -183,6 +183,7 @@ std::string ActivationLayerCPUTest::getPrimitiveType(const utils::ActivationType
         (activation_type == utils::ActivationTypes::Exp) ||
         (activation_type == utils::ActivationTypes::Floor) ||
         (activation_type == utils::ActivationTypes::Ceiling) ||
+        (activation_type == utils::ActivationTypes::Negative) ||
         (activation_type == utils::ActivationTypes::HSwish) ||
         (activation_type == utils::ActivationTypes::IsInf) ||
         (activation_type == utils::ActivationTypes::HardSigmoid) ||
@@ -210,6 +211,7 @@ std::string ActivationLayerCPUTest::getPrimitiveType(const utils::ActivationType
 #endif
     if ((activation_type == utils::ActivationTypes::Floor) ||
        (activation_type == utils::ActivationTypes::Ceiling) ||
+       (activation_type == utils::ActivationTypes::Negative) ||
        (activation_type == utils::ActivationTypes::IsNaN) ||
        (activation_type == utils::ActivationTypes::IsFinite) ||
        (activation_type == utils::ActivationTypes::RoundHalfAwayFromZero) ||
@@ -253,6 +255,7 @@ const std::map<utils::ActivationTypes, std::vector<std::vector<float>>>& activat
         {Elu,         {{0.1f}}},
         {Floor,       {{}}},
         {Ceiling,     {{}}},
+        {Negative,    {{}}},
         {Swish,       {{0.1f}}},
         {HSwish,      {{}}},
         {PReLu,       {{-0.01f}}},