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>
diff --git a/tests/validation/NEON/ElementwiseMax.cpp b/tests/validation/NEON/ElementwiseMax.cpp
index 4bc2631..61421ab 100644
--- a/tests/validation/NEON/ElementwiseMax.cpp
+++ b/tests/validation/NEON/ElementwiseMax.cpp
@@ -62,6 +62,8 @@
#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
const auto ElementwiseMaxFP32Dataset = combine(combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::F32)),
framework::dataset::make("DataType", DataType::F32));
+const auto InPlaceDataSet = framework::dataset::make("InPlace", { false, true });
+const auto OutOfPlaceDataSet = framework::dataset::make("InPlace", { false });
} // namespace
TEST_SUITE(NEON)
@@ -111,7 +113,8 @@
// *INDENT-ON*
TEST_SUITE(S32)
-FIXTURE_DATA_TEST_CASE(RunSmall, NEElementwiseMaxFixture<int32_t>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallShapes(), ElementwiseMaxS32Dataset))
+FIXTURE_DATA_TEST_CASE(RunSmall, NEElementwiseMaxFixture<int32_t>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(), ElementwiseMaxS32Dataset),
+ InPlaceDataSet))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -119,7 +122,8 @@
TEST_SUITE_END() // S32
TEST_SUITE(S16)
-FIXTURE_DATA_TEST_CASE(RunSmall, NEElementwiseMaxFixture<int16_t>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), ElementwiseMaxS16Dataset))
+FIXTURE_DATA_TEST_CASE(RunSmall, NEElementwiseMaxFixture<int16_t>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), ElementwiseMaxS16Dataset),
+ InPlaceDataSet))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -131,11 +135,12 @@
TEST_SUITE(Quantized)
TEST_SUITE(QASYMM8)
-FIXTURE_DATA_TEST_CASE(RunSmall, NEElementwiseMaxQuantizedFixture<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(combine(datasets::SmallShapes(),
+FIXTURE_DATA_TEST_CASE(RunSmall, NEElementwiseMaxQuantizedFixture<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(combine(combine(datasets::SmallShapes(),
ElementwiseMaxQASYMM8Dataset),
framework::dataset::make("QuantizationInfo", { QuantizationInfo(5.f / 255.f, 20) })),
framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f / 255.f, 10) })),
- framework::dataset::make("QuantizationInfo", { QuantizationInfo(1.f / 255.f, 5) })))
+ framework::dataset::make("QuantizationInfo", { QuantizationInfo(1.f / 255.f, 5) })),
+ OutOfPlaceDataSet))
{
// Validate output
validate(Accessor(_target), _reference, tolerance_fp32, 0.01);
@@ -144,11 +149,13 @@
template <typename T>
using NEElementwiseMaxQuantizedBroadcastFixture = ElementwiseMaxQuantizedBroadcastValidationFixture<Tensor, Accessor, NEElementwiseMax, T>;
-FIXTURE_DATA_TEST_CASE(RunSmallBroadcast, NEElementwiseMaxQuantizedBroadcastFixture<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(combine(datasets::SmallShapesBroadcast(),
- ElementwiseMaxQASYMM8Dataset),
- framework::dataset::make("QuantizationInfo", { QuantizationInfo(5.f / 255.f, 20) })),
- framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f / 255.f, 10) })),
- framework::dataset::make("QuantizationInfo", { QuantizationInfo(1.f / 255.f, 5) })))
+FIXTURE_DATA_TEST_CASE(RunSmallBroadcast, NEElementwiseMaxQuantizedBroadcastFixture<uint8_t>, framework::DatasetMode::PRECOMMIT,
+ combine(combine(combine(combine(combine(datasets::SmallShapesBroadcast(),
+ ElementwiseMaxQASYMM8Dataset),
+ framework::dataset::make("QuantizationInfo", { QuantizationInfo(5.f / 255.f, 20) })),
+ framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f / 255.f, 10) })),
+ framework::dataset::make("QuantizationInfo", { QuantizationInfo(1.f / 255.f, 5) })),
+ OutOfPlaceDataSet))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -156,16 +163,26 @@
TEST_SUITE_END()
TEST_SUITE(QASYMM8_SIGNED)
-FIXTURE_DATA_TEST_CASE(RunSmall, NEElementwiseMaxQuantizedFixture<int8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(combine(datasets::SmallShapes(),
+FIXTURE_DATA_TEST_CASE(RunSmall, NEElementwiseMaxQuantizedFixture<int8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(combine(combine(datasets::SmallShapes(),
ElementwiseMaxQASYMM8SignedDataset),
framework::dataset::make("QuantizationInfo", { QuantizationInfo(10.f, 20) })),
framework::dataset::make("QuantizationInfo", { QuantizationInfo(1.f, 0) })),
- framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f, -27) })))
+ framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f, -27) })),
+ OutOfPlaceDataSet))
{
// Validate output
validate(Accessor(_target), _reference, tolerance_qasymm8_signed);
}
-
+FIXTURE_DATA_TEST_CASE(RunSmallInPlace, NEElementwiseMaxQuantizedFixture<int8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(combine(combine(datasets::SmallShapes(),
+ ElementwiseMaxQASYMM8SignedDataset),
+ framework::dataset::make("QuantizationInfo", { QuantizationInfo(10.f, -20) })),
+ framework::dataset::make("QuantizationInfo", { QuantizationInfo(10.f, -20) })),
+ framework::dataset::make("QuantizationInfo", { QuantizationInfo(10.f, -20) })),
+ InPlaceDataSet))
+{
+ // Validate output
+ validate(Accessor(_target), _reference, tolerance_qasymm8_signed);
+}
TEST_SUITE_END()
TEST_SUITE_END()
@@ -173,7 +190,8 @@
TEST_SUITE(Float)
#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
TEST_SUITE(F16)
-FIXTURE_DATA_TEST_CASE(RunSmall, NEElementwiseMaxFixture<half>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), ElementwiseMaxFP16Dataset))
+FIXTURE_DATA_TEST_CASE(RunSmall, NEElementwiseMaxFixture<half>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), ElementwiseMaxFP16Dataset),
+ InPlaceDataSet))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -182,7 +200,8 @@
#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
TEST_SUITE(F32)
-FIXTURE_DATA_TEST_CASE(RunSmall, NEElementwiseMaxFixture<float>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), ElementwiseMaxFP32Dataset))
+FIXTURE_DATA_TEST_CASE(RunSmall, NEElementwiseMaxFixture<float>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), ElementwiseMaxFP32Dataset),
+ InPlaceDataSet))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -190,8 +209,16 @@
template <typename T>
using NEElementwiseMaxBroadcastFixture = ElementwiseMaxBroadcastValidationFixture<Tensor, Accessor, NEElementwiseMax, T>;
-FIXTURE_DATA_TEST_CASE(RunSmallBroadcast, NEElementwiseMaxBroadcastFixture<float>, framework::DatasetMode::ALL, combine(datasets::SmallShapesBroadcast(),
- ElementwiseMaxFP32Dataset))
+FIXTURE_DATA_TEST_CASE(RunSmallBroadcast, NEElementwiseMaxBroadcastFixture<float>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapesBroadcast(),
+ ElementwiseMaxFP32Dataset),
+ OutOfPlaceDataSet))
+{
+ // Validate output
+ validate(Accessor(_target), _reference);
+}
+FIXTURE_DATA_TEST_CASE(RunTinyBroadcastInPlace, NEElementwiseMaxBroadcastFixture<float>, framework::DatasetMode::ALL, combine(combine(datasets::TinyShapesBroadcastInplace(),
+ ElementwiseMaxFP32Dataset),
+ InPlaceDataSet))
{
// Validate output
validate(Accessor(_target), _reference);