Add FP32 Neon™ swish activation

Change-Id: Id37b59adbc8c4cbe218d1652aeb02a0b4ce42c66
Signed-off-by: Jonathan Deakin <jonathan.deakin@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8256
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Marquez Tello <pablo.tello@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

8 files changed, 26 insertions, 3 deletions

diff --git a/arm_compute/core/Types.h b/arm_compute/core/Types.h
index 36dba2cc00..fc6d46c53b 100644
--- a/arm_compute/core/Types.h
+++ b/arm_compute/core/Types.h
@@ -1641,7 +1641,8 @@ public:
         SQRT,            /**< Square root ( \f$ f(x) = \sqrt{x} \f$ )*/
         LINEAR,          /**< Linear ( \f$ f(x)= ax + b \f$ ) */
         IDENTITY,        /**< Identity ( \f$ f(x)= x \f$ ) */
-        HARD_SWISH,      /**< Hard-swish ( \f$ f(x) = (x * relu6(x+3))/6 \f$ ) */
+        HARD_SWISH,      /**< Hard-swish ( \f$ f(x) = (x \text{ReLU6}(x+3))/6 = x \min(\max(0,x+3),6)/6 \f$ ) */
+        SWISH,            /**< Swish ( \f$ f(x) = \frac{x}{1 + e^{-ax}} = x \text{logistic}(ax) \f$ ) */
         GELU             /**< GELU ( \f$ f(x) = x * 1/2 * 1 + erf(x / \sqrt{2}) \f$ ) */
     };
 
diff --git a/src/core/Utils.cpp b/src/core/Utils.cpp
index 48eb8b9bf5..d05a044fc8 100644
--- a/src/core/Utils.cpp
+++ b/src/core/Utils.cpp
@@ -178,6 +178,7 @@ const std::string &string_from_activation_func(ActivationLayerInfo::ActivationFu
         { ActivationLayerInfo::ActivationFunction::TANH, "TANH" },
         { ActivationLayerInfo::ActivationFunction::IDENTITY, "IDENTITY" },
         { ActivationLayerInfo::ActivationFunction::HARD_SWISH, "HARD_SWISH" },
+        { ActivationLayerInfo::ActivationFunction::SWISH, "SWISH" },
         { ActivationLayerInfo::ActivationFunction::GELU, "GELU" }
 
     };
diff --git a/src/cpu/kernels/activation/generic/neon/impl.h b/src/cpu/kernels/activation/generic/neon/impl.h
index 4d4aa8d212..4f392866b5 100644
--- a/src/cpu/kernels/activation/generic/neon/impl.h
+++ b/src/cpu/kernels/activation/generic/neon/impl.h
@@ -149,6 +149,9 @@ void fp_neon_activation_impl(const ITensor *src, ITensor *dst, const ActivationL
                 case ActivationLayerInfo::ActivationFunction::HARD_SWISH:
                     tmp = wrapper::vmul(vin, wrapper::vmul(const_inv_6, wrapper::vmin(const_6, wrapper::vmax(const_0, wrapper::vadd(vin, const_3)))));
                     break;
+                case ActivationLayerInfo::ActivationFunction::SWISH:
+                    tmp = wrapper::vmul(vin, wrapper::vinv(wrapper::vadd(const_1, wrapper::vexpq(wrapper::vneg(wrapper::vmul(va, vin))))));
+                    break;
 #ifdef __aarch64__
                 case ActivationLayerInfo::ActivationFunction::GELU:
                     tmp = wrapper::vmul(vin, wrapper::vmul(const_inv_2, wrapper::vadd(const_1, wrapper::verf(wrapper::vmul(vin, const_inv_sqrt_2)))));
@@ -208,6 +211,9 @@ void fp_neon_activation_impl(const ITensor *src, ITensor *dst, const ActivationL
                 case ActivationLayerInfo::ActivationFunction::HARD_SWISH:
                     tmp = in * ((std::min(std::max((in + 3), 0.0f), 6.0f)) * 0.166666667f);
                     break;
+                case ActivationLayerInfo::ActivationFunction::SWISH:
+                    tmp = in / (static_cast<T>(1) + std::exp(-a*in));
+                    break;
                 case ActivationLayerInfo::ActivationFunction::GELU:
                     tmp = in * static_cast<T>(0.5f * (1.0f + erff(static_cast<float>(in) / 1.41421356237f)));
                     break;
diff --git a/src/cpu/kernels/activation/generic/sve/fp16.cpp b/src/cpu/kernels/activation/generic/sve/fp16.cpp
index 5730a361d9..6e9648b5bf 100644
--- a/src/cpu/kernels/activation/generic/sve/fp16.cpp
+++ b/src/cpu/kernels/activation/generic/sve/fp16.cpp
@@ -114,6 +114,9 @@ void sve_fp16_activation(const ITensor *src, ITensor *dst, const ActivationLayer
                 case ActivationLayerInfo::ActivationFunction::HARD_SWISH:
                     tmp = svmul_f16_z(pg, vin, svmul_f16_z(pg, const_inv_6, svmin_f16_z(pg, const_6, svmax_f16_z(pg, const_0, svadd_f16_z(pg, vin, const_3)))));
                     break;
+                case ActivationLayerInfo::ActivationFunction::SWISH:
+                    tmp = svmul_f16_z(pg, vin, svinv_f16_z(pg, svadd_f16_z(pg, const_1, svexp_f16_z(pg, svneg_f16_z(pg, svmul_f16_z(pg, va, vin))))));
+                    break;
                 default:
                     ARM_COMPUTE_ERROR("Unsupported activation function");
             }
@@ -129,4 +132,4 @@ void sve_fp16_activation(const ITensor *src, ITensor *dst, const ActivationLayer
 }
 } // namespace cpu
 } // namespace arm_compute
-#endif /* defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(ENABLE_FP16_KERNELS) */
\ No newline at end of file
+#endif /* defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(ENABLE_FP16_KERNELS) */
diff --git a/src/cpu/kernels/activation/generic/sve/fp32.cpp b/src/cpu/kernels/activation/generic/sve/fp32.cpp
index 7ce2046730..b5490063cf 100644
--- a/src/cpu/kernels/activation/generic/sve/fp32.cpp
+++ b/src/cpu/kernels/activation/generic/sve/fp32.cpp
@@ -113,6 +113,9 @@ void sve_fp32_activation(const ITensor *src, ITensor *dst, const ActivationLayer
                 case ActivationLayerInfo::ActivationFunction::HARD_SWISH:
                     tmp = svmul_f32_z(pg, vin, svmul_f32_z(pg, const_inv_6, svmin_f32_z(pg, const_6, svmax_f32_z(pg, const_0, svadd_f32_z(pg, vin, const_3)))));
                     break;
+                case ActivationLayerInfo::ActivationFunction::SWISH:
+                    tmp = svmul_f32_z(pg, vin, svinv_f32_z(pg, svadd_f32_z(pg, const_1, svexp_f32_z(pg, svneg_f32_z(pg, svmul_f32_z(pg, va, vin))))));
+                    break;
                 default:
                     ARM_COMPUTE_ERROR("Unsupported activation function");
             }
diff --git a/tests/validation/NEON/ActivationLayer.cpp b/tests/validation/NEON/ActivationLayer.cpp
index a2971f28ba..d580a1c2ca 100644
--- a/tests/validation/NEON/ActivationLayer.cpp
+++ b/tests/validation/NEON/ActivationLayer.cpp
@@ -68,6 +68,7 @@ RelativeTolerance<float> relative_tolerance(DataType data_type, ActivationLayerI
         case ActivationLayerInfo::ActivationFunction::SQRT:
         case ActivationLayerInfo::ActivationFunction::TANH:
         case ActivationLayerInfo::ActivationFunction::HARD_SWISH:
+        case ActivationLayerInfo::ActivationFunction::SWISH:
         case ActivationLayerInfo::ActivationFunction::GELU:
             switch(data_type)
             {
@@ -111,6 +112,7 @@ AbsoluteTolerance<float> absolute_tolerance(DataType data_type, ActivationLayerI
         case ActivationLayerInfo::ActivationFunction::LOGISTIC:
         case ActivationLayerInfo::ActivationFunction::SQRT:
         case ActivationLayerInfo::ActivationFunction::TANH:
+        case ActivationLayerInfo::ActivationFunction::SWISH:
         case ActivationLayerInfo::ActivationFunction::HARD_SWISH:
             switch(data_type)
             {
@@ -173,7 +175,8 @@ const auto CNNDataTypes = framework::dataset::make("DataType",
     DataType::F32,
 });
 
-const auto NeonActivationFunctionsDataset = concat(datasets::ActivationFunctions(), framework::dataset::make("ActivationFunction", ActivationLayerInfo::ActivationFunction::HARD_SWISH));
+const auto NeonActivationFunctionsDataset = concat(datasets::ActivationFunctions(),
+    framework::dataset::make("ActivationFunction", {ActivationLayerInfo::ActivationFunction::HARD_SWISH, ActivationLayerInfo::ActivationFunction::SWISH}));
 
 /** Input data sets. */
 const auto ActivationDataset = combine(combine(framework::dataset::make("InPlace", { false, true }), NeonActivationFunctionsDataset), framework::dataset::make("AlphaBeta", { 0.5f, 1.f }));
diff --git a/tests/validation/reference/ActivationLayer.h b/tests/validation/reference/ActivationLayer.h
index 97733238ef..2bf96831a6 100644
--- a/tests/validation/reference/ActivationLayer.h
+++ b/tests/validation/reference/ActivationLayer.h
@@ -84,6 +84,9 @@ inline T activate_float(T x, T a, T b, ActivationLayerInfo::ActivationFunction a
         case ActivationLayerInfo::ActivationFunction::HARD_SWISH:
             ret = x * ((std::min(std::max(static_cast<T>(x + 3), static_cast<T>(0.0f)), static_cast<T>(6.0f))) * 0.166666667f);
             break;
+        case ActivationLayerInfo::ActivationFunction::SWISH:
+            ret = static_cast<T>(x) / (static_cast<T>(1) + std::exp(-a*x));
+            break;
         case ActivationLayerInfo::ActivationFunction::GELU:
             ret = x * 0.5f * (1 + erf(x / std::sqrt(2.0f)));
             break;
diff --git a/utils/TypePrinter.h b/utils/TypePrinter.h
index 21f9c0849c..f55f72a4b8 100644
--- a/utils/TypePrinter.h
+++ b/utils/TypePrinter.h
@@ -628,6 +628,9 @@ inline ::std::ostream &operator<<(::std::ostream &os, const ActivationLayerInfo:
         case ActivationLayerInfo::ActivationFunction::HARD_SWISH:
             os << "HARD_SWISH";
             break;
+        case ActivationLayerInfo::ActivationFunction::SWISH:
+            os << "SWISH";
+            break;
         case ActivationLayerInfo::ActivationFunction::GELU:
             os << "GELU";
             break;