Review all shapes in datasets to account for padding removal Part 3

* Add the following configurations for stressing padding removal:
    * size = 1
    * size = multiple of processing size
    * size = non-multiple of processing size

Resolves COMPMID-3865

Signed-off-by: SiCong Li <sicong.li@arm.com>
Change-Id: I2e0e6d4da129f64ba23cf7b9e0fa1fa1ad93efc3
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4440
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

6 files changed, 60 insertions, 35 deletions

diff --git a/tests/datasets/GEMMLowpFusedOffsetOutputDataset.h b/tests/datasets/GEMMLowpFusedOffsetOutputDataset.h
index fe3ca255f7..7ab068c211 100644
--- a/tests/datasets/GEMMLowpFusedOffsetOutputDataset.h
+++ b/tests/datasets/GEMMLowpFusedOffsetOutputDataset.h
@@ -167,10 +167,10 @@ class SmallGEMMLowpFusedOffsetOutputUint8Dataset final : public GEMMLowpFusedOff
 public:
     SmallGEMMLowpFusedOffsetOutputUint8Dataset()
     {
-        add_config(TensorShape(21U, 13U), TensorShape(33U, 21U), TensorShape(33U, 13U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, -100, 2, 13, 10, 210));
+        add_config(TensorShape(21U, 13U), TensorShape(1U, 21U), TensorShape(1U, 13U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, -100, 2, 13, 10, 210));
         add_config(TensorShape(52U, 13U), TensorShape(33U, 52U), TensorShape(33U, 13U), 0, 4, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, 100, 2, 13, 10, 210));
         add_config(TensorShape(31U, 27U), TensorShape(23U, 31U), TensorShape(23U, 27U), 18, 23, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, 200, 2, 13, 10, 210));
-        add_config(TensorShape(32U, 72U), TensorShape(17U, 32U), TensorShape(17U, 72U), -9, 1, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, -100, 2, 13, 10, 210));
+        add_config(TensorShape(32U, 72U), TensorShape(16U, 32U), TensorShape(16U, 72U), -9, 1, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, -100, 2, 13, 10, 210));
 
         add_config(TensorShape(21U, 1U), TensorShape(43U, 21U), TensorShape(43U, 1U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -2, 254601600, 10, 10, 210));
         add_config(TensorShape(31U, 3U), TensorShape(72U, 31U), TensorShape(72U, 3U), -2, 13, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 0, 254601600, 10, 10, 210));
@@ -184,7 +184,7 @@ class SmallGEMMLowpFusedOffsetOutputInt8Dataset final : public GEMMLowpFusedOffs
 public:
     SmallGEMMLowpFusedOffsetOutputInt8Dataset()
     {
-        add_config(TensorShape(21U, 1U), TensorShape(43U, 21U), TensorShape(43U, 1U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, -50, 2, 13, -10, 110));
+        add_config(TensorShape(21U, 1U), TensorShape(1U, 21U), TensorShape(1U, 1U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, -50, 2, 13, -10, 110));
         add_config(TensorShape(31U, 3U), TensorShape(72U, 31U), TensorShape(72U, 3U), -2, 13, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, 0, 2, 13, -10, 110));
         add_config(TensorShape(52U, 26U), TensorShape(33U, 52U), TensorShape(33U, 26U), -2, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, 0, 2, 13, -10, 110));
         add_config(TensorShape(38U, 43U), TensorShape(21U, 38U), TensorShape(21U, 43U), -3, -2, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, -40, 2, 13, -10, 110));
diff --git a/tests/datasets/SmallGEMMLowpDataset.h b/tests/datasets/SmallGEMMLowpDataset.h
index f16e3fa5f8..1b6c65307b 100644
--- a/tests/datasets/SmallGEMMLowpDataset.h
+++ b/tests/datasets/SmallGEMMLowpDataset.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 Arm Limited.
+ * Copyright (c) 2017-2020 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -42,7 +42,7 @@ class SmallGEMMLowpDataset final : public GEMMLowpDataset
 public:
     SmallGEMMLowpDataset()
     {
-        add_config(TensorShape(21U, 1U), TensorShape(43U, 21U), TensorShape(43U, 1U), 0, 0);
+        add_config(TensorShape(21U, 1U), TensorShape(1U, 21U), TensorShape(1U, 1U), 0, 0);
         add_config(TensorShape(21U, 13U), TensorShape(33U, 21U), TensorShape(33U, 13U), 0, 0);
         add_config(TensorShape(31U, 3U), TensorShape(72U, 31U), TensorShape(72U, 3U), -2, 13);
         add_config(TensorShape(52U, 13U), TensorShape(33U, 52U), TensorShape(33U, 13U), 0, 4);
diff --git a/tests/validation/CL/GEMMLowpMatrixMultiplyNative.cpp b/tests/validation/CL/GEMMLowpMatrixMultiplyNative.cpp
index 4a574d8f1d..1057af95f2 100644
--- a/tests/validation/CL/GEMMLowpMatrixMultiplyNative.cpp
+++ b/tests/validation/CL/GEMMLowpMatrixMultiplyNative.cpp
@@ -53,8 +53,15 @@ namespace
 {
 // *INDENT-OFF*
 // clang-format off
-/** M values to test */
-const auto m_values = framework::dataset::make("M", 37);
+/** M, N combinations to test
+ *  1: Special 1x1 case
+ *  2: Special multples of processor size in both dimensions
+ *  3: Non multiples of processor size in both dimensions
+*/
+const auto m_n_values = zip(
+    framework::dataset::make("M", {1, 16, 37}),
+    framework::dataset::make("N", {1, 16, 51})
+    );
 
 /** M_W values to test */
 const auto m_w_values = framework::dataset::make("M_W", 5);
@@ -93,8 +100,7 @@ const auto k0_values_nightly = framework::dataset::make("K0", { 1, 2, 3, 4, 8, 1
 TEST_SUITE(CL)
 TEST_SUITE(GEMMLowpMatrixMultiplyNative)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMLowpMatrixMultiplyNativeFixture, framework::DatasetMode::ALL,
-                combine(combine(combine(combine(combine(combine(m_values,
-                                                                n_values),
+                combine(combine(combine(combine(combine(m_n_values,
                                                                 k_values),
                                                                 b_values),
                                                                 m0_values_precommit),
@@ -106,8 +112,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMLowpMatrixMultiplyNativeFixture, framewor
 }
 
 FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMLowpMatrixMultiplyNativeFixture, framework::DatasetMode::ALL,
-                combine(combine(combine(combine(combine(combine(m_values,
-                                                                n_values),
+                combine(combine(combine(combine(combine(m_n_values,
                                                                 k_values),
                                                                 b_values),
                                                                 m0_values_nightly),
diff --git a/tests/validation/CL/GEMMLowpMatrixMultiplyReshaped.cpp b/tests/validation/CL/GEMMLowpMatrixMultiplyReshaped.cpp
index 0c651cddc2..4873a291ab 100644
--- a/tests/validation/CL/GEMMLowpMatrixMultiplyReshaped.cpp
+++ b/tests/validation/CL/GEMMLowpMatrixMultiplyReshaped.cpp
@@ -62,8 +62,19 @@ namespace
 {
 // *INDENT-OFF*
 // clang-format off
+
+/** M, N combinations to test
+ *  1: Special 1x1 case
+ *  2: Special multples of processor size in both dimensions
+ *  3: Non multiples of processor size in both dimensions
+*/
+const auto m_n_values = zip(
+    framework::dataset::make("M", {1, 16, 37}),
+    framework::dataset::make("N", {1, 16, 51})
+    );
+
 /** M values to test */
-const auto m_values = framework::dataset::make("M", 37);
+const auto m_values = framework::dataset::make("M", {1, 37});
 
 /** M_W values to test */
 const auto m_w_values = framework::dataset::make("M_W", 5);
@@ -72,7 +83,7 @@ const auto m_w_values = framework::dataset::make("M_W", 5);
 const auto m_h_values = framework::dataset::make("M_H", 7);
 
 /** N values to test */
-const auto n_values = framework::dataset::make("N", 51);
+const auto n_values = framework::dataset::make("N", {1, 51});
 
 /** K values to test */
 const auto k_values = framework::dataset::make("K", 23);
@@ -125,9 +136,8 @@ TEST_SUITE(QUANTIZED)
 
 TEST_SUITE(QASYMM8)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMLowpMatrixMultiplyReshapedFixture, framework::DatasetMode::ALL,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
-                                                                   m_values,
-                                                                   n_values),
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                                                                   m_n_values,
                                                                    k_values),
                                                                    b_values),
                                                                    m0_values_precommit_1),
@@ -205,9 +215,8 @@ TEST_SUITE_END() // QASYMM8
 
 TEST_SUITE(QASYMM8_SIGNED)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMLowpMatrixMultiplyReshapedFixture, framework::DatasetMode::ALL,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
-                                                                   m_values,
-                                                                   n_values),
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                                                                   m_n_values,
                                                                    k_values),
                                                                    b_values),
                                                                    m0_values_precommit_2),
diff --git a/tests/validation/CL/GEMMMatrixMultiply.cpp b/tests/validation/CL/GEMMMatrixMultiply.cpp
index 5d2e211d91..fdf7f503ec 100644
--- a/tests/validation/CL/GEMMMatrixMultiply.cpp
+++ b/tests/validation/CL/GEMMMatrixMultiply.cpp
@@ -72,8 +72,16 @@ const auto alpha_values = framework::dataset::make("alpha", {1.0f, -0.75f} );
 /** Beta values to test */
 const auto beta_values = framework::dataset::make("beta", {-0.35f, 0.0f} );
 
-/** M values to test */
-const auto m_values = framework::dataset::make("M", {37, 1});
+/** M, N combinations to test
+ *  1: Special 1x1 case
+ *  2: Special multples of processor size in both dimensions
+ *  3: Non multiples of processor size in both dimensions
+ *  4: Special 1x1003 case
+*/
+const auto m_n_values = zip(
+    framework::dataset::make("M", {1, 16, 37, 1}),
+    framework::dataset::make("N", {1, 16, 51, 1003})
+    );
 
 /** N values to test */
 const auto n_values = framework::dataset::make("N", {51, 1003});
@@ -247,9 +255,8 @@ TEST_SUITE(Float)
 TEST_SUITE(FP32)
 
 FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyNativeFixture<float>, framework::DatasetMode::ALL,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
-                                                                   m_values,
-                                                                   n_values),
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                                                                   m_n_values,
                                                                    k_values),
                                                                    b_values),
                                                                    alpha_values),
@@ -287,9 +294,8 @@ TEST_SUITE_END() // FP32
 
 TEST_SUITE(FP16)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyNativeFixture<half>, framework::DatasetMode::ALL,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
-                                                                   m_values,
-                                                                   n_values),
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                                                                   m_n_values,
                                                                    k_values),
                                                                    b_values),
                                                                    alpha_values),
diff --git a/tests/validation/CL/GEMMMatrixMultiplyInterleavedTransposed.cpp b/tests/validation/CL/GEMMMatrixMultiplyInterleavedTransposed.cpp
index b2701e7f6c..d6507a06c4 100644
--- a/tests/validation/CL/GEMMMatrixMultiplyInterleavedTransposed.cpp
+++ b/tests/validation/CL/GEMMMatrixMultiplyInterleavedTransposed.cpp
@@ -82,8 +82,15 @@ const auto alpha_values = framework::dataset::make("alpha", {1.0f, -0.75f} );
 /** Beta values to test */
 const auto beta_values = framework::dataset::make("beta", {-0.35f, 0.0f} );
 
-/** M values to test */
-const auto m_values = framework::dataset::make("M", {37, 1});
+/** M, N combinations to test
+ *  1: Special 1x1 case
+ *  2: Special multples of processor size in both dimensions
+ *  3: Non multiples of processor size in both dimensions
+*/
+const auto m_n_values = zip(
+    framework::dataset::make("M", {1, 16, 37}),
+    framework::dataset::make("N", {1, 16, 51})
+    );
 
 /** N values to test */
 const auto n_values = framework::dataset::make("N", 51);
@@ -235,9 +242,8 @@ TEST_CASE(Negative, framework::DatasetMode::ALL)
 TEST_SUITE(Float)
 TEST_SUITE(FP32)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedFixture<float>, framework::DatasetMode::ALL,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
-                                                                   m_values,
-                                                                   n_values),
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                                                                   m_n_values,
                                                                    k_values),
                                                                    b_values),
                                                                    alpha_values),
@@ -279,9 +285,8 @@ TEST_SUITE_END() // FP32
 
 TEST_SUITE(FP16)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedFixture<half>, framework::DatasetMode::ALL,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
-                                                                   m_values,
-                                                                   n_values),
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                                                                   m_n_values,
                                                                    k_values),
                                                                    b_values),
                                                                    alpha_values),