Add in-place calculation support for CL elementwise arithmetic kernels

- Add in-place calculation support in ClArithmeticKernel,  ClSaturatedArithmeticKernel and ClMulKernel
- Add in-place test cases

Resolves: COMPMID-4431

Signed-off-by: Sheri Zhang <sheri.zhang@arm.com>
Change-Id: Id484bdb76b74478a33fedb471ae0c7f799c599f6
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5885
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 32 insertions, 20 deletions

diff --git a/tests/validation/CL/ArithmeticDivision.cpp b/tests/validation/CL/ArithmeticDivision.cpp
index 9dcdfb83e1..94bacba7e5 100644
--- a/tests/validation/CL/ArithmeticDivision.cpp
+++ b/tests/validation/CL/ArithmeticDivision.cpp
@@ -59,6 +59,8 @@ const auto ActivationFunctionsDataset      = framework::dataset::make("Activatio
     ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::BOUNDED_RELU, 0.75f, 0.25f),
     ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC, 0.75f, 0.25f)
 });
+const auto InPlaceDataSet    = framework::dataset::make("InPlace", { false, true });
+const auto OutOfPlaceDataSet = framework::dataset::make("InPlace", { false });
 } // namespace
 
 TEST_SUITE(CL)
@@ -95,14 +97,16 @@ using CLArithmeticDivisionIntegerFixture = ArithmeticDivisionValidationIntegerFi
 TEST_SUITE(Integer)
 TEST_SUITE(S32)
 
-FIXTURE_DATA_TEST_CASE(RunSmallInteger, CLArithmeticDivisionIntegerFixture, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(), ArithmeticDivisionS32Dataset),
-                                                                                                                       EmptyActivationFunctionsDataset))
+FIXTURE_DATA_TEST_CASE(RunSmallInteger, CLArithmeticDivisionIntegerFixture, framework::DatasetMode::PRECOMMIT, combine(combine(combine(datasets::SmallShapes(), ArithmeticDivisionS32Dataset),
+                                                                                                                       EmptyActivationFunctionsDataset),
+                                                                                                                       InPlaceDataSet))
 {
     // Validate output
     validate(CLAccessor(_target), _reference);
 }
-FIXTURE_DATA_TEST_CASE(RunIntegerWithActivation, CLArithmeticDivisionIntegerFixture, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), ArithmeticDivisionS32Dataset),
-                       ActivationFunctionsDataset))
+FIXTURE_DATA_TEST_CASE(RunIntegerWithActivation, CLArithmeticDivisionIntegerFixture, framework::DatasetMode::ALL, combine(combine(combine(datasets::SmallShapes(), ArithmeticDivisionS32Dataset),
+                       ActivationFunctionsDataset),
+                       InPlaceDataSet))
 {
     // Validate output
     validate(CLAccessor(_target), _reference);
@@ -116,14 +120,16 @@ using CLArithmeticDivisionFloatFixture = ArithmeticDivisionValidationFloatFixtur
 
 TEST_SUITE(Float)
 TEST_SUITE(FP16)
-FIXTURE_DATA_TEST_CASE(RunSmall, CLArithmeticDivisionFloatFixture<half>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), ArithmeticDivisionFP16Dataset),
-                                                                                                              EmptyActivationFunctionsDataset))
+FIXTURE_DATA_TEST_CASE(RunSmall, CLArithmeticDivisionFloatFixture<half>, framework::DatasetMode::ALL, combine(combine(combine(datasets::SmallShapes(), ArithmeticDivisionFP16Dataset),
+                                                                                                                      EmptyActivationFunctionsDataset),
+                                                                                                              InPlaceDataSet))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_fp16, 0.01);
 }
-FIXTURE_DATA_TEST_CASE(RunWithActivation, CLArithmeticDivisionFloatFixture<half>, framework::DatasetMode::ALL, combine(combine(datasets::TinyShapes(), ArithmeticDivisionFP16Dataset),
-                                                                                                                       ActivationFunctionsDataset))
+FIXTURE_DATA_TEST_CASE(RunWithActivation, CLArithmeticDivisionFloatFixture<half>, framework::DatasetMode::ALL, combine(combine(combine(datasets::TinyShapes(), ArithmeticDivisionFP16Dataset),
+                                                                                                                       ActivationFunctionsDataset),
+                                                                                                                       InPlaceDataSet))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_fp16, 0.01);
@@ -131,21 +137,24 @@ FIXTURE_DATA_TEST_CASE(RunWithActivation, CLArithmeticDivisionFloatFixture<half>
 TEST_SUITE_END()
 
 TEST_SUITE(FP32)
-FIXTURE_DATA_TEST_CASE(RunSmall, CLArithmeticDivisionFloatFixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(), ArithmeticDivisionFP32Dataset),
-                                                                                                                     EmptyActivationFunctionsDataset))
+FIXTURE_DATA_TEST_CASE(RunSmall, CLArithmeticDivisionFloatFixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(datasets::SmallShapes(), ArithmeticDivisionFP32Dataset),
+                                                                                                                     EmptyActivationFunctionsDataset),
+                                                                                                                     InPlaceDataSet))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_fp32);
 }
-FIXTURE_DATA_TEST_CASE(RunWithActivation, CLArithmeticDivisionFloatFixture<float>, framework::DatasetMode::ALL, combine(combine(datasets::TinyShapes(), ArithmeticDivisionFP32Dataset),
-                                                                                                                        ActivationFunctionsDataset))
+FIXTURE_DATA_TEST_CASE(RunWithActivation, CLArithmeticDivisionFloatFixture<float>, framework::DatasetMode::ALL, combine(combine(combine(datasets::TinyShapes(), ArithmeticDivisionFP32Dataset),
+                                                                                                                        ActivationFunctionsDataset),
+                                                                                                                        InPlaceDataSet))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_fp32);
 }
 
-FIXTURE_DATA_TEST_CASE(RunLarge, CLArithmeticDivisionFloatFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(), ArithmeticDivisionFP32Dataset),
-                                                                                                                   EmptyActivationFunctionsDataset))
+FIXTURE_DATA_TEST_CASE(RunLarge, CLArithmeticDivisionFloatFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(combine(datasets::LargeShapes(), ArithmeticDivisionFP32Dataset),
+                                                                                                                   EmptyActivationFunctionsDataset),
+                                                                                                                   InPlaceDataSet))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_fp32);
@@ -154,24 +163,27 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLArithmeticDivisionFloatFixture<float>, framew
 template <typename T>
 using CLArithmeticDivisionBroadcastFloatFixture = ArithmeticDivisionBroadcastValidationFloatFixture<CLTensor, CLAccessor, CLArithmeticDivision, T>;
 
-FIXTURE_DATA_TEST_CASE(RunSmallBroadcast, CLArithmeticDivisionBroadcastFloatFixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapesBroadcast(),
+FIXTURE_DATA_TEST_CASE(RunSmallBroadcast, CLArithmeticDivisionBroadcastFloatFixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(datasets::SmallShapesBroadcast(),
                        ArithmeticDivisionFP32Dataset),
-                       EmptyActivationFunctionsDataset))
+                       EmptyActivationFunctionsDataset),
+                       OutOfPlaceDataSet))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_fp32);
 }
-FIXTURE_DATA_TEST_CASE(RunWithActivationBroadcast, CLArithmeticDivisionBroadcastFloatFixture<float>, framework::DatasetMode::ALL, combine(combine(datasets::TinyShapesBroadcast(),
+FIXTURE_DATA_TEST_CASE(RunWithActivationBroadcast, CLArithmeticDivisionBroadcastFloatFixture<float>, framework::DatasetMode::ALL, combine(combine(combine(datasets::TinyShapesBroadcast(),
                        ArithmeticDivisionFP32Dataset),
-                       ActivationFunctionsDataset))
+                       ActivationFunctionsDataset),
+                       OutOfPlaceDataSet))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_fp32);
 }
 
-FIXTURE_DATA_TEST_CASE(RunLargeBroadcast, CLArithmeticDivisionBroadcastFloatFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapesBroadcast(),
+FIXTURE_DATA_TEST_CASE(RunLargeBroadcast, CLArithmeticDivisionBroadcastFloatFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(combine(datasets::LargeShapesBroadcast(),
                        ArithmeticDivisionFP32Dataset),
-                       EmptyActivationFunctionsDataset))
+                       EmptyActivationFunctionsDataset),
+                       OutOfPlaceDataSet))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_fp32);