COMPMID-636 - Extending GEMMLowp validation for NEON intrinsics

Change-Id: Id02a0b3bf5af65dd940b46b2f6634b6a479cf388
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/92275
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
diff --git a/tests/datasets/GEMMLowpDataset.h b/tests/datasets/GEMMLowpDataset.h
new file mode 100644
index 0000000..4bf2a98
--- /dev/null
+++ b/tests/datasets/GEMMLowpDataset.h
@@ -0,0 +1,147 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_TEST_GEMMLOWP_DATASET
+#define ARM_COMPUTE_TEST_GEMMLOWP_DATASET
+
+#include "utils/TypePrinter.h"
+
+#include "arm_compute/core/TensorShape.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace datasets
+{
+class GEMMLowpDataset
+{
+public:
+    using type = std::tuple<TensorShape, TensorShape, TensorShape, int32_t, int32_t, int32_t, int32_t, int32_t>;
+
+    struct iterator
+    {
+        iterator(std::vector<TensorShape>::const_iterator a_it,
+                 std::vector<TensorShape>::const_iterator b_it,
+                 std::vector<TensorShape>::const_iterator c_it,
+                 std::vector<int32_t>::const_iterator     a_offset_it,
+                 std::vector<int32_t>::const_iterator     b_offset_it,
+                 std::vector<int32_t>::const_iterator     c_offset_it,
+                 std::vector<int32_t>::const_iterator     c_mult_int_it,
+                 std::vector<int32_t>::const_iterator     out_shift_it)
+            : _a_it{ std::move(a_it) },
+              _b_it{ std::move(b_it) },
+              _c_it{ std::move(c_it) },
+              _a_offset_it{ std::move(a_offset_it) },
+              _b_offset_it{ std::move(b_offset_it) },
+              _c_offset_it{ std::move(c_offset_it) },
+              _c_mult_int_it{ std::move(c_mult_int_it) },
+              _out_shift_it{ std::move(out_shift_it) }
+        {
+        }
+
+        std::string description() const
+        {
+            std::stringstream description;
+            description << "A=" << *_a_it << ":";
+            description << "B=" << *_b_it << ":";
+            description << "C=" << *_c_it << ":";
+            description << "a_offset=" << *_a_offset_it << ":";
+            description << "b_offset=" << *_b_offset_it << ":";
+            description << "c_offset=" << *_c_offset_it << ":";
+            description << "c_mult_int=" << *_c_mult_int_it << ":";
+            description << "out_shift=" << *_out_shift_it << ":";
+            return description.str();
+        }
+
+        GEMMLowpDataset::type operator*() const
+        {
+            return std::make_tuple(*_a_it, *_b_it, *_c_it, *_a_offset_it, *_b_offset_it, *_c_offset_it, *_c_mult_int_it, *_out_shift_it);
+        }
+
+        iterator &operator++()
+        {
+            ++_a_it;
+            ++_b_it;
+            ++_c_it;
+            ++_a_offset_it;
+            ++_b_offset_it;
+            ++_c_offset_it;
+            ++_c_mult_int_it;
+            ++_out_shift_it;
+
+            return *this;
+        }
+
+    private:
+        std::vector<TensorShape>::const_iterator _a_it;
+        std::vector<TensorShape>::const_iterator _b_it;
+        std::vector<TensorShape>::const_iterator _c_it;
+        std::vector<int32_t>::const_iterator     _a_offset_it;
+        std::vector<int32_t>::const_iterator     _b_offset_it;
+        std::vector<int32_t>::const_iterator     _c_offset_it;
+        std::vector<int32_t>::const_iterator     _c_mult_int_it;
+        std::vector<int32_t>::const_iterator     _out_shift_it;
+    };
+
+    iterator begin() const
+    {
+        return iterator(_a_shapes.begin(), _b_shapes.begin(), _c_shapes.begin(), _a_offset.begin(), _b_offset.begin(), _c_offset.begin(), _c_mult_int.begin(), _out_shift.begin());
+    }
+
+    int size() const
+    {
+        return std::min(_a_shapes.size(), std::min(_b_shapes.size(), std::min(_c_shapes.size(), std::min(_a_offset.size(), std::min(_b_offset.size(), std::min(_c_offset.size(), std::min(_c_mult_int.size(),
+                                                                                                         _out_shift.size())))))));
+    }
+
+    void add_config(TensorShape a, TensorShape b, TensorShape c, int32_t a_offset, int32_t b_offset, int32_t c_offset, int32_t c_mult_int, int32_t out_shift)
+    {
+        _a_shapes.emplace_back(std::move(a));
+        _b_shapes.emplace_back(std::move(b));
+        _c_shapes.emplace_back(std::move(c));
+        _a_offset.emplace_back(std::move(a_offset));
+        _b_offset.emplace_back(std::move(b_offset));
+        _c_offset.emplace_back(std::move(c_offset));
+        _c_mult_int.emplace_back(std::move(c_mult_int));
+        _out_shift.emplace_back(std::move(out_shift));
+    }
+
+protected:
+    GEMMLowpDataset()                   = default;
+    GEMMLowpDataset(GEMMLowpDataset &&) = default;
+
+private:
+    std::vector<TensorShape> _a_shapes{};
+    std::vector<TensorShape> _b_shapes{};
+    std::vector<TensorShape> _c_shapes{};
+    std::vector<int32_t>     _a_offset{};
+    std::vector<int32_t>     _b_offset{};
+    std::vector<int32_t>     _c_offset{};
+    std::vector<int32_t>     _c_mult_int{};
+    std::vector<int32_t>     _out_shift{};
+};
+} // namespace datasets
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_GEMMLOWP_DATASET */
diff --git a/tests/datasets/LargeGEMMLowpDataset.h b/tests/datasets/LargeGEMMLowpDataset.h
new file mode 100644
index 0000000..610a999
--- /dev/null
+++ b/tests/datasets/LargeGEMMLowpDataset.h
@@ -0,0 +1,54 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_TEST_LARGE_GEMMLOWP_DATASET
+#define ARM_COMPUTE_TEST_LARGE_GEMMLOWP_DATASET
+
+#include "tests/datasets/GEMMLowpDataset.h"
+
+#include "utils/TypePrinter.h"
+
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace datasets
+{
+class LargeGEMMLowpDataset final : public GEMMLowpDataset
+{
+public:
+    LargeGEMMLowpDataset()
+    {
+        add_config(TensorShape(923U, 429U), TensorShape(871U, 923U), TensorShape(871U, 429U), 0, 0, 0, 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);
+    }
+};
+} // namespace datasets
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_LARGE_GEMMLOWP_DATASET */
diff --git a/tests/datasets/SmallGEMMLowpDataset.h b/tests/datasets/SmallGEMMLowpDataset.h
new file mode 100644
index 0000000..2ed7a75
--- /dev/null
+++ b/tests/datasets/SmallGEMMLowpDataset.h
@@ -0,0 +1,54 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_TEST_SMALL_GEMMLOWP_DATASET
+#define ARM_COMPUTE_TEST_SMALL_GEMMLOWP_DATASET
+
+#include "tests/datasets/GEMMLowpDataset.h"
+
+#include "utils/TypePrinter.h"
+
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace datasets
+{
+class SmallGEMMLowpDataset final : public GEMMLowpDataset
+{
+public:
+    SmallGEMMLowpDataset()
+    {
+        add_config(TensorShape(21U, 13U), TensorShape(33U, 21U), TensorShape(33U, 13U), 0, 0, 0, 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);
+    }
+};
+} // namespace datasets
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_SMALL_GEMMLOWP_DATASET */
diff --git a/tests/validation/NEON/GEMMLowp.cpp b/tests/validation/NEON/GEMMLowp.cpp
index 6a972eb..dede3ae 100644
--- a/tests/validation/NEON/GEMMLowp.cpp
+++ b/tests/validation/NEON/GEMMLowp.cpp
@@ -26,6 +26,8 @@
 #include "arm_compute/runtime/Tensor.h"
 #include "arm_compute/runtime/TensorAllocator.h"
 #include "tests/NEON/Accessor.h"
+#include "tests/datasets/LargeGEMMLowpDataset.h"
+#include "tests/datasets/SmallGEMMLowpDataset.h"
 #include "tests/framework/Asserts.h"
 #include "tests/framework/Macros.h"
 #include "tests/framework/datasets/Datasets.h"
@@ -44,8 +46,6 @@
 {
 namespace
 {
-constexpr AbsoluteTolerance<float> tolerance_f(0.001f); /**< Tolerance value for comparing reference's output against implementation's output for floating point data types */
-
 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)
@@ -69,7 +69,7 @@
 FIXTURE_DATA_TEST_CASE(RunSmall, NEGEMMInterleaveBlockedFixture, framework::DatasetMode::PRECOMMIT, data_int_blk)
 {
     // Validate output
-    validate(Accessor(_target), _reference, tolerance_f);
+    validate(Accessor(_target), _reference);
 }
 TEST_SUITE_END()
 
@@ -79,16 +79,41 @@
 FIXTURE_DATA_TEST_CASE(RunSmall, NEGEMMInterleaveBlockedTransposedFixture, framework::DatasetMode::PRECOMMIT, data_int_blk_tr)
 {
     // Validate output
-    validate(Accessor(_target), _reference, tolerance_f);
+    validate(Accessor(_target), _reference);
 }
 
 TEST_SUITE_END()
 
 using NEGEMMLowpOffsetFixture = GEMMLowpOffsetValidationFixture<Tensor, Accessor, NEGEMMLowp>;
-FIXTURE_DATA_TEST_CASE(RunSmall, NEGEMMLowpOffsetFixture, framework::DatasetMode::PRECOMMIT, data_mnk *data_offsets)
+
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(framework::dataset::concat(datasets::SmallGEMMLowpDataset(), datasets::LargeGEMMLowpDataset()), framework::dataset::make("DataType",
+                                                                   DataType::U8)),
+               shape_a, shape_b, shape_c, a_offset, b_offset, c_offset, c_mult_int, out_shift, data_type)
+{
+    // Create tensors
+    Tensor a = create_tensor<Tensor>(shape_a, data_type);
+    Tensor b = create_tensor<Tensor>(shape_b, data_type);
+    Tensor c = create_tensor<Tensor>(shape_c, data_type);
+
+    ARM_COMPUTE_EXPECT(a.info()->is_resizable(), framework::LogLevel::ERRORS);
+    ARM_COMPUTE_EXPECT(b.info()->is_resizable(), framework::LogLevel::ERRORS);
+    ARM_COMPUTE_EXPECT(c.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+    // Create and configure function
+    NEGEMMLowp gemmlowp;
+    gemmlowp.configure(&a, &b, &c, a_offset, b_offset, c_offset, c_mult_int, out_shift);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall, NEGEMMLowpOffsetFixture, framework::DatasetMode::ALL, combine(datasets::SmallGEMMLowpDataset(), framework::dataset::make("DataType", DataType::U8)))
 {
     // Validate output
-    validate(Accessor(_target), _reference, tolerance_f);
+    validate(Accessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, NEGEMMLowpOffsetFixture, framework::DatasetMode::NIGHTLY, combine(datasets::LargeGEMMLowpDataset(), framework::dataset::make("DataType", DataType::U8)))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
 }
 TEST_SUITE_END()
 
@@ -100,7 +125,7 @@
                        16))
 {
     // Validate output
-    validate(Accessor(_target), _reference, tolerance_f);
+    validate(Accessor(_target), _reference);
 }
 TEST_SUITE_END()
 #endif // defined(__aarch64__)
diff --git a/tests/validation/fixtures/GEMMLowpFixture.h b/tests/validation/fixtures/GEMMLowpFixture.h
index 556b6c4..1a4084f 100644
--- a/tests/validation/fixtures/GEMMLowpFixture.h
+++ b/tests/validation/fixtures/GEMMLowpFixture.h
@@ -47,13 +47,10 @@
 {
 public:
     template <typename...>
-    void setup(size_t m, size_t n, size_t k, int32_t a_offset, int32_t b_offset, int32_t c_offset, int32_t c_mult_int, int32_t out_shift)
+    void setup(TensorShape shape_a, TensorShape shape_b, TensorShape shape_c, int32_t a_offset, int32_t b_offset, int32_t c_offset, int32_t c_mult_int, int32_t out_shift, DataType data_type)
     {
-        const TensorShape shape_a(k, m);
-        const TensorShape shape_b(n, k);
-        const TensorShape shape_c(n, m);
-        _target    = compute_target(shape_a, shape_b, shape_c, a_offset, b_offset, c_offset, c_mult_int, out_shift);
-        _reference = compute_reference(shape_a, shape_b, shape_c, a_offset, b_offset, c_offset, c_mult_int, out_shift);
+        _target    = compute_target(shape_a, shape_b, shape_c, a_offset, b_offset, c_offset, c_mult_int, out_shift, data_type);
+        _reference = compute_reference(shape_a, shape_b, shape_c, a_offset, b_offset, c_offset, c_mult_int, out_shift, data_type);
     }
 
 protected:
@@ -66,12 +63,12 @@
     }
 
     TensorType compute_target(const TensorShape &shape_a, const TensorShape &shape_b, const TensorShape &shape_c,
-                              int32_t a_offset, int32_t b_offset, int32_t c_offset, int32_t c_mult_int, int32_t out_shift)
+                              int32_t a_offset, int32_t b_offset, int32_t c_offset, int32_t c_mult_int, int32_t out_shift, DataType data_type)
     {
         // Create tensors
-        TensorType a = create_tensor<TensorType>(shape_a, DataType::U8, 1);
-        TensorType b = create_tensor<TensorType>(shape_b, DataType::U8, 1);
-        TensorType c = create_tensor<TensorType>(shape_c, DataType::U8, 1);
+        TensorType a = create_tensor<TensorType>(shape_a, data_type, 1);
+        TensorType b = create_tensor<TensorType>(shape_b, data_type, 1);
+        TensorType c = create_tensor<TensorType>(shape_c, data_type, 1);
 
         // Create and configure function
         FunctionType gemmlowp;
@@ -101,12 +98,12 @@
     }
 
     SimpleTensor<uint8_t> compute_reference(const TensorShape &shape_a, const TensorShape &shape_b, const TensorShape &shape_c,
-                                            int32_t a_offset, int32_t b_offset, int32_t c_offset, int32_t c_mult_int, int32_t out_shift)
+                                            int32_t a_offset, int32_t b_offset, int32_t c_offset, int32_t c_mult_int, int32_t out_shift, DataType data_type)
     {
         // Create reference
-        SimpleTensor<uint8_t> a{ shape_a, DataType::U8, 1 };
-        SimpleTensor<uint8_t> b{ shape_b, DataType::U8, 1 };
-        SimpleTensor<uint8_t> c{ shape_c, DataType::U8, 1 };
+        SimpleTensor<uint8_t> a{ shape_a, data_type, 1 };
+        SimpleTensor<uint8_t> b{ shape_b, data_type, 1 };
+        SimpleTensor<uint8_t> c{ shape_c, data_type, 1 };
 
         // Fill reference
         fill(a, 0);