COMPMID-3101 Fuse activation with floating point elementwise operation layers in CL

Signed-off-by: Giorgio Arena <giorgio.arena@arm.com>
Change-Id: I1693f8664ba7c0dc8c076bbe7365cef1e667bd25
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2718
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

4 files changed, 38 insertions, 18 deletions

diff --git a/src/runtime/NEON/functions/NEArithmeticAddition.cpp b/src/runtime/NEON/functions/NEArithmeticAddition.cpp
index 6d0b207cf1..06c71db1bd 100644
--- a/src/runtime/NEON/functions/NEArithmeticAddition.cpp
+++ b/src/runtime/NEON/functions/NEArithmeticAddition.cpp
@@ -31,14 +31,16 @@
 
 namespace arm_compute
 {
-void NEArithmeticAddition::configure(ITensor *input1, ITensor *input2, ITensor *output, ConvertPolicy policy)
+void NEArithmeticAddition::configure(ITensor *input1, ITensor *input2, ITensor *output, ConvertPolicy policy, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_UNUSED(act_info);
     auto k = arm_compute::support::cpp14::make_unique<NEArithmeticAdditionKernel>();
     k->configure(input1, input2, output, policy);
     _kernel = std::move(k);
 }
-Status NEArithmeticAddition::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy)
+Status NEArithmeticAddition::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled());
     return NEArithmeticAdditionKernel::validate(input1, input2, output, policy);
 }
 } // namespace arm_compute
diff --git a/src/runtime/NEON/functions/NEArithmeticSubtraction.cpp b/src/runtime/NEON/functions/NEArithmeticSubtraction.cpp
index 0ad87383ce..454adc336b 100644
--- a/src/runtime/NEON/functions/NEArithmeticSubtraction.cpp
+++ b/src/runtime/NEON/functions/NEArithmeticSubtraction.cpp
@@ -31,8 +31,9 @@
 
 namespace arm_compute
 {
-void NEArithmeticSubtraction::configure(ITensor *input1, ITensor *input2, ITensor *output, ConvertPolicy policy)
+void NEArithmeticSubtraction::configure(ITensor *input1, ITensor *input2, ITensor *output, ConvertPolicy policy, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_UNUSED(act_info);
     auto k = arm_compute::support::cpp14::make_unique<NEArithmeticSubtractionKernel>();
     k->configure(input1, input2, output, policy);
     _kernel = std::move(k);
@@ -48,8 +49,9 @@ void NEArithmeticSubtraction::configure(ITensor *input1, ITensor *input2, ITenso
     }
 }
 
-Status NEArithmeticSubtraction::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy)
+Status NEArithmeticSubtraction::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled());
     return NEArithmeticSubtractionKernel::validate(input1, input2, output, policy);
 }
 } // namespace arm_compute
diff --git a/src/runtime/NEON/functions/NEElementwiseOperators.cpp b/src/runtime/NEON/functions/NEElementwiseOperators.cpp
index 0ba0ddbe3a..7451c6ff2b 100644
--- a/src/runtime/NEON/functions/NEElementwiseOperators.cpp
+++ b/src/runtime/NEON/functions/NEElementwiseOperators.cpp
@@ -32,15 +32,17 @@
 
 namespace arm_compute
 {
-void NEElementwiseMax::configure(ITensor *input1, ITensor *input2, ITensor *output)
+void NEElementwiseMax::configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_UNUSED(act_info);
     auto k = arm_compute::support::cpp14::make_unique<NEArithmeticOperationKernel>();
     k->configure(ArithmeticOperation::MAX, input1, input2, output);
     _kernel = std::move(k);
 }
 
-Status NEElementwiseMax::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
+Status NEElementwiseMax::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled());
     if(input1->data_type() == DataType::QASYMM8_SIGNED)
     {
         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input1, output);
@@ -49,15 +51,17 @@ Status NEElementwiseMax::validate(const ITensorInfo *input1, const ITensorInfo *
     return NEArithmeticOperationKernel::validate(ArithmeticOperation::MAX, input1, input2, output);
 }
 
-void NEElementwiseMin::configure(ITensor *input1, ITensor *input2, ITensor *output)
+void NEElementwiseMin::configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_UNUSED(act_info);
     auto k = arm_compute::support::cpp14::make_unique<NEArithmeticOperationKernel>();
     k->configure(ArithmeticOperation::MIN, input1, input2, output);
     _kernel = std::move(k);
 }
 
-Status NEElementwiseMin::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
+Status NEElementwiseMin::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled());
     if(input1->data_type() == DataType::QASYMM8_SIGNED)
     {
         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input1, output);
@@ -66,39 +70,45 @@ Status NEElementwiseMin::validate(const ITensorInfo *input1, const ITensorInfo *
     return NEArithmeticOperationKernel::validate(ArithmeticOperation::MIN, input1, input2, output);
 }
 
-void NEElementwiseSquaredDiff::configure(ITensor *input1, ITensor *input2, ITensor *output)
+void NEElementwiseSquaredDiff::configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_UNUSED(act_info);
     auto k = arm_compute::support::cpp14::make_unique<NEArithmeticOperationKernel>();
     k->configure(ArithmeticOperation::SQUARED_DIFF, input1, input2, output);
     _kernel = std::move(k);
 }
 
-Status NEElementwiseSquaredDiff::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
+Status NEElementwiseSquaredDiff::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled());
     return NEArithmeticOperationKernel::validate(ArithmeticOperation::SQUARED_DIFF, input1, input2, output);
 }
 
-void NEElementwiseDivision::configure(ITensor *input1, ITensor *input2, ITensor *output)
+void NEElementwiseDivision::configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_UNUSED(act_info);
     auto k = arm_compute::support::cpp14::make_unique<NEDivisionOperationKernel>();
     k->configure(input1, input2, output);
     _kernel = std::move(k);
 }
 
-Status NEElementwiseDivision::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
+Status NEElementwiseDivision::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled());
     return NEDivisionOperationKernel::validate(input1, input2, output);
 }
 
-void NEElementwisePower::configure(ITensor *input1, ITensor *input2, ITensor *output)
+void NEElementwisePower::configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_UNUSED(act_info);
     auto k = arm_compute::support::cpp14::make_unique<NEPowerOperationKernel>();
     k->configure(input1, input2, output);
     _kernel = std::move(k);
 }
 
-Status NEElementwisePower::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
+Status NEElementwisePower::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled());
     return NEPowerOperationKernel::validate(input1, input2, output);
 }
 
diff --git a/src/runtime/NEON/functions/NEPixelWiseMultiplication.cpp b/src/runtime/NEON/functions/NEPixelWiseMultiplication.cpp
index e2516e420c..eaf233b9ed 100644
--- a/src/runtime/NEON/functions/NEPixelWiseMultiplication.cpp
+++ b/src/runtime/NEON/functions/NEPixelWiseMultiplication.cpp
@@ -31,8 +31,10 @@
 
 namespace arm_compute
 {
-void NEPixelWiseMultiplication::configure(ITensor *input1, ITensor *input2, ITensor *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy)
+void NEPixelWiseMultiplication::configure(ITensor *input1, ITensor *input2, ITensor *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy,
+                                          const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_UNUSED(act_info);
     auto k = arm_compute::support::cpp14::make_unique<NEPixelWiseMultiplicationKernel>();
     k->configure(input1, input2, output, scale, overflow_policy, rounding_policy);
     _kernel = std::move(k);
@@ -47,13 +49,16 @@ void NEPixelWiseMultiplication::configure(ITensor *input1, ITensor *input2, ITen
         }
     }
 }
-Status NEPixelWiseMultiplication::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy)
+Status NEPixelWiseMultiplication::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy,
+                                           const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled());
     return NEPixelWiseMultiplicationKernel::validate(input1, input2, output, scale, overflow_policy, rounding_policy);
 }
 
-void NEComplexPixelWiseMultiplication::configure(ITensor *input1, ITensor *input2, ITensor *output)
+void NEComplexPixelWiseMultiplication::configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_UNUSED(act_info);
     auto k = arm_compute::support::cpp14::make_unique<NEComplexPixelWiseMultiplicationKernel>();
     k->configure(input1, input2, output);
     _kernel = std::move(k);
@@ -69,8 +74,9 @@ void NEComplexPixelWiseMultiplication::configure(ITensor *input1, ITensor *input
     }
 }
 
-Status NEComplexPixelWiseMultiplication::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
+Status NEComplexPixelWiseMultiplication::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled());
     return NEComplexPixelWiseMultiplicationKernel::validate(input1, input2, output);
 }