COMPMID-661: softmax-fp32 optimisation (#14)

Change-Id: I2007af1ed9dcf68065cf412aa50f73a2025b31a6
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/94605
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>

1 files changed, 20 insertions, 13 deletions

diff --git a/tests/validation/CL/SoftmaxLayer.cpp b/tests/validation/CL/SoftmaxLayer.cpp
index c469b8acd8..7842c5c83b 100644
--- a/tests/validation/CL/SoftmaxLayer.cpp
+++ b/tests/validation/CL/SoftmaxLayer.cpp
@@ -21,6 +21,7 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
+#include "arm_compute/core/CL/kernels/CLSoftmaxLayerKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/CLTensorAllocator.h"
@@ -62,7 +63,7 @@ const auto CNNDataTypes = framework::dataset::make("DataType",
 TEST_SUITE(CL)
 TEST_SUITE(SoftmaxLayer)
 
-DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(concat(datasets::SmallShapes(), datasets::LargeShapes()), CNNDataTypes), shape, data_type)
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(concat(datasets::SoftmaxLayerSmallShapes(), datasets::SoftmaxLayerLargeShapes()), CNNDataTypes), shape, data_type)
 {
     // Set fixed point position data type allowed
     const int fixed_point_position = is_data_type_fixed_point(data_type) ? 3 : 0;
@@ -83,10 +84,16 @@ DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(concat(datase
     validate(src.info()->valid_region(), valid_region);
     validate(dst.info()->valid_region(), valid_region);
 
-    // Validate padding
-    const PaddingSize padding = PaddingCalculator(shape.x(), 16).required_padding();
-    validate(src.info()->padding(), padding);
-    validate(dst.info()->padding(), padding);
+    // Get reduction kernel info
+    CLLogits1DMaxShiftExpSumKernel::ParallelReductionInfo reduction_info = CLLogits1DMaxShiftExpSumKernel::is_parallel_reduction(shape.x());
+
+    // Validate src padding
+    const PaddingSize padding_src = PaddingCalculator(shape.x(), std::get<1>(reduction_info)).required_padding();
+    validate(src.info()->padding(), padding_src);
+
+    // Validate dst padding
+    const PaddingSize padding_dst = PaddingCalculator(shape.x(), 16).required_padding();
+    validate(dst.info()->padding(), padding_dst);
 }
 
 template <typename T>
@@ -94,12 +101,12 @@ using CLSoftmaxLayerFixture = SoftmaxValidationFixture<CLTensor, CLAccessor, CLS
 
 TEST_SUITE(Float)
 TEST_SUITE(FP16)
-FIXTURE_DATA_TEST_CASE(RunSmall, CLSoftmaxLayerFixture<half>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallShapes(), framework::dataset::make("DataType", DataType::F16)))
+FIXTURE_DATA_TEST_CASE(RunSmall, CLSoftmaxLayerFixture<half>, framework::DatasetMode::PRECOMMIT, combine(datasets::SoftmaxLayerSmallShapes(), framework::dataset::make("DataType", DataType::F16)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f16);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, CLSoftmaxLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), framework::dataset::make("DataType", DataType::F16)))
+FIXTURE_DATA_TEST_CASE(RunLarge, CLSoftmaxLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(datasets::SoftmaxLayerLargeShapes(), framework::dataset::make("DataType", DataType::F16)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f16);
@@ -107,12 +114,12 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLSoftmaxLayerFixture<half>, framework::Dataset
 TEST_SUITE_END()
 
 TEST_SUITE(FP32)
-FIXTURE_DATA_TEST_CASE(RunSmall, CLSoftmaxLayerFixture<float>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallShapes(), framework::dataset::make("DataType", DataType::F32)))
+FIXTURE_DATA_TEST_CASE(RunSmall, CLSoftmaxLayerFixture<float>, framework::DatasetMode::PRECOMMIT, combine(datasets::SoftmaxLayerSmallShapes(), framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, CLSoftmaxLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), framework::dataset::make("DataType", DataType::F32)))
+FIXTURE_DATA_TEST_CASE(RunLarge, CLSoftmaxLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(datasets::SoftmaxLayerLargeShapes(), framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32);
@@ -126,14 +133,14 @@ using CLSoftmaxLayerFixedPointFixture = SoftmaxValidationFixedPointFixture<CLTen
 TEST_SUITE(Quantized)
 TEST_SUITE(QS8)
 // Testing for fixed point position [1,6) as reciprocal limits the maximum fixed point position to 5
-FIXTURE_DATA_TEST_CASE(RunSmall, CLSoftmaxLayerFixedPointFixture<int8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(), framework::dataset::make("DataType",
+FIXTURE_DATA_TEST_CASE(RunSmall, CLSoftmaxLayerFixedPointFixture<int8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SoftmaxLayerSmallShapes(), framework::dataset::make("DataType",
                                                                                                                      DataType::QS8)),
                                                                                                                      framework::dataset::make("FractionalBits", 1, 6)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_fixed_point);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, CLSoftmaxLayerFixedPointFixture<int8_t>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(), framework::dataset::make("DataType",
+FIXTURE_DATA_TEST_CASE(RunLarge, CLSoftmaxLayerFixedPointFixture<int8_t>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::SoftmaxLayerLargeShapes(), framework::dataset::make("DataType",
                                                                                                                    DataType::QS8)),
                                                                                                                    framework::dataset::make("FractionalBits", 1, 6)))
 {
@@ -144,7 +151,7 @@ TEST_SUITE_END()
 
 TEST_SUITE(QS16)
 // Testing for fixed point position [1,14) as reciprocal limits the maximum fixed point position to 14
-FIXTURE_DATA_TEST_CASE(RunSmall, CLSoftmaxLayerFixedPointFixture<int16_t>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(),
+FIXTURE_DATA_TEST_CASE(RunSmall, CLSoftmaxLayerFixedPointFixture<int16_t>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SoftmaxLayerSmallShapes(),
                                                                                                                       framework::dataset::make("DataType",
                                                                                                                               DataType::QS16)),
                                                                                                                       framework::dataset::make("FractionalBits", 1, 14)))
@@ -152,7 +159,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLSoftmaxLayerFixedPointFixture<int16_t>, frame
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_fixed_point);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, CLSoftmaxLayerFixedPointFixture<int16_t>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(),
+FIXTURE_DATA_TEST_CASE(RunLarge, CLSoftmaxLayerFixedPointFixture<int16_t>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::SoftmaxLayerLargeShapes(),
                                                                                                                     framework::dataset::make("DataType",
                                                                                                                             DataType::QS16)),
                                                                                                                     framework::dataset::make("FractionalBits", 1, 14)))