COMPMID-616 - Optimizing GEMMLowp on NEON intrinsics

Change-Id: Ibbeff5d37249b6e8fc34ad496035a1511c9da5a3
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/94072
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Pablo Tello <pablo.tello@arm.com>

3 files changed, 8 insertions, 12 deletions

diff --git a/tests/datasets/LargeGEMMLowpDataset.h b/tests/datasets/LargeGEMMLowpDataset.h
index 610a999dd8..10f79e423d 100644
--- a/tests/datasets/LargeGEMMLowpDataset.h
+++ b/tests/datasets/LargeGEMMLowpDataset.h
@@ -43,6 +43,8 @@ public:
     LargeGEMMLowpDataset()
     {
         add_config(TensorShape(923U, 429U), TensorShape(871U, 923U), TensorShape(871U, 429U), 0, 0, 0, 1, 0);
+        add_config(TensorShape(873U, 513U), TensorShape(784U, 873U), TensorShape(784U, 513U), 0, 4, 3, 2, 0);
+        add_config(TensorShape(697U, 872U), TensorShape(563U, 697U), TensorShape(563U, 872U), -2, 0, 1, 1, 0);
         add_config(TensorShape(1021U, 973U), TensorShape(783U, 1021U), TensorShape(783U, 973U), 5, 13, -6, 2, 2);
         add_config(TensorShape(681U, 1023U), TensorShape(213U, 681U), TensorShape(213U, 1023U), -3, -2, 8, 4, 3);
         add_config(TensorShape(941U, 1011U), TensorShape(623U, 941U), TensorShape(623U, 1011U), -9, 1, -3, 3, 1);
diff --git a/tests/datasets/SmallGEMMLowpDataset.h b/tests/datasets/SmallGEMMLowpDataset.h
index 2ed7a759ea..b7fe3907ad 100644
--- a/tests/datasets/SmallGEMMLowpDataset.h
+++ b/tests/datasets/SmallGEMMLowpDataset.h
@@ -43,6 +43,8 @@ public:
     SmallGEMMLowpDataset()
     {
         add_config(TensorShape(21U, 13U), TensorShape(33U, 21U), TensorShape(33U, 13U), 0, 0, 0, 1, 0);
+        add_config(TensorShape(52U, 13U), TensorShape(33U, 52U), TensorShape(33U, 13U), 0, 4, 3, 2, 0);
+        add_config(TensorShape(52U, 26U), TensorShape(33U, 52U), TensorShape(33U, 26U), -2, 0, 1, 1, 0);
         add_config(TensorShape(31U, 27U), TensorShape(23U, 31U), TensorShape(23U, 27U), 5, 13, -6, 2, 2);
         add_config(TensorShape(38U, 12U), TensorShape(21U, 38U), TensorShape(21U, 12U), -3, -2, 8, 4, 3);
         add_config(TensorShape(32U, 72U), TensorShape(17U, 32U), TensorShape(17U, 72U), -9, 1, -3, 3, 1);
diff --git a/tests/validation/NEON/GEMMLowp.cpp b/tests/validation/NEON/GEMMLowp.cpp
index dede3aef29..de7495bb6a 100644
--- a/tests/validation/NEON/GEMMLowp.cpp
+++ b/tests/validation/NEON/GEMMLowp.cpp
@@ -21,11 +21,13 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
+#include "arm_compute/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/NEON/functions/NEGEMMLowp.h"
 #include "arm_compute/runtime/Tensor.h"
 #include "arm_compute/runtime/TensorAllocator.h"
 #include "tests/NEON/Accessor.h"
+#include "tests/NEON/Helper.h"
 #include "tests/datasets/LargeGEMMLowpDataset.h"
 #include "tests/datasets/SmallGEMMLowpDataset.h"
 #include "tests/framework/Asserts.h"
@@ -35,9 +37,6 @@
 #include "tests/validation/fixtures/GEMMInterleaveBlockedFixture.h"
 #include "tests/validation/fixtures/GEMMLowpFixture.h"
 
-#include "arm_compute/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.h"
-#include "tests/NEON/Helper.h"
-
 namespace arm_compute
 {
 namespace test
@@ -46,15 +45,8 @@ namespace validation
 {
 namespace
 {
-const auto data_mnk     = framework::dataset::make("M", 12, 20) * framework::dataset::make("N", 12, 20) * framework::dataset::make("K", 12, 15);
-const auto data_offsets = framework::dataset::make("a", -3, 3) * framework::dataset::make("b", -1, 2) * framework::dataset::make("c", 1, 3) * framework::dataset::make("cm", 0,
-                          3)
-                          * framework::dataset::make("shift", 0, 4);
-
-const auto data_int_blk = framework::dataset::make("M", 8, 12) * framework::dataset::make("N", 8, 12) * framework::dataset::make("by", 8, 13) * framework::dataset::make("block", 4, 9);
-
+const auto data_int_blk    = framework::dataset::make("M", 8, 12) * framework::dataset::make("N", 8, 12) * framework::dataset::make("by", 8, 13) * framework::dataset::make("block", 4, 9);
 const auto data_int_blk_tr = framework::dataset::make("M", 8, 17) * framework::dataset::make("N", 8, 14) * framework::dataset::make("by", 12) * framework::dataset::make("block", 4);
-
 } // namespace
 
 TEST_SUITE(NEON)
@@ -115,7 +107,6 @@ FIXTURE_DATA_TEST_CASE(RunLarge, NEGEMMLowpOffsetFixture, framework::DatasetMode
     // Validate output
     validate(Accessor(_target), _reference);
 }
-TEST_SUITE_END()
 
 //FIXME: This is in the process of being updated, for more info please refer to COMPMID-624.
 #if 0  // defined(__aarch64__)
@@ -131,6 +122,7 @@ TEST_SUITE_END()
 #endif // defined(__aarch64__)
 TEST_SUITE_END()
 TEST_SUITE_END()
+TEST_SUITE_END()
 } // namespace validation
 } // namespace test
 } // namespace arm_compute