COMPMID: GEMMTranspose1xW tests.

Change-Id: I62a7a1871b93fafc65eb58fa550bc86179bdffe7
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/112489
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>

6 files changed, 313 insertions, 9 deletions

diff --git a/arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h b/arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h
index e8ee2a7d29..451dce2cfc 100644
--- a/arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h
+++ b/arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2017 ARM Limited.
+ * Copyright (c) 2016, 2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -51,10 +51,10 @@ class ITensor;
  *
  * @f[
  * \left( \begin{array}{cccccccc}
- * a00 & a01 & a02 & a03 & a04 & a05 & a06 & a7 \\
- * a10 & a11 & a12 & a13 & a14 & a15 & a16 & 17 \\
- * a20 & a21 & a22 & a23 & a24 & a25 & a26 & 27 \\
- * a30 & a31 & a32 & a33 & a34 & a35 & a36 & 37 \\
+ * a00 & a01 & a02 & a03 & a04 & a05 & a06 & a07 \\
+ * a10 & a11 & a12 & a13 & a14 & a15 & a16 & a17 \\
+ * a20 & a21 & a22 & a23 & a24 & a25 & a26 & a27 \\
+ * a30 & a31 & a32 & a33 & a34 & a35 & a36 & a37 \\
  * \end{array} \right)
  * \rightarrow
  * \left( \begin{array}{cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc}
diff --git a/tests/CL/Helper.h b/tests/CL/Helper.h
index 78b6d0734d..3f19d61a7d 100644
--- a/tests/CL/Helper.h
+++ b/tests/CL/Helper.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017, 2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -59,7 +59,6 @@ public:
         _border_handler.configure(first, BorderSize(bordersize), BorderMode::CONSTANT, PixelValue(0));
     }
 };
-
 } // namespace test
 } // namespace arm_compute
 #endif /* __ARM_COMPUTE_TEST_CL_HELPER_H__ */
diff --git a/tests/validation/CL/GEMM.cpp b/tests/validation/CL/GEMM.cpp
index 255f5d934a..5114ed6c13 100644
--- a/tests/validation/CL/GEMM.cpp
+++ b/tests/validation/CL/GEMM.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017, 2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -22,6 +22,7 @@
  * SOFTWARE.
  */
 #include "arm_compute/core/CL/kernels/CLGEMMInterleave4x4Kernel.h"
+#include "arm_compute/core/CL/kernels/CLGEMMTranspose1xWKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/CLTensorAllocator.h"
@@ -37,6 +38,7 @@
 #include "tests/validation/Validation.h"
 #include "tests/validation/fixtures/GEMMFixture.h"
 #include "tests/validation/fixtures/GEMMInterleave4x4Fixture.h"
+#include "tests/validation/fixtures/GEMMTranspose1xWFixture.h"
 
 namespace arm_compute
 {
@@ -62,6 +64,8 @@ const auto CNNDataTypes = framework::dataset::make("DataType",
 });
 } // namespace
 
+const auto data_transpose = framework::dataset::make("M", 8, 14) * framework::dataset::make("N", 7, 14);
+
 TEST_SUITE(CL)
 TEST_SUITE(GEMM)
 
@@ -131,6 +135,46 @@ DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(framework::da
 template <typename T>
 using CLGEMMFixture = GEMMValidationFixture<CLTensor, CLAccessor, CLGEMM, T>;
 
+TEST_SUITE(TRANSPOSE_1XW)
+using CLGEMMTranspose1xW        = CLSynthetizeFunctionWithZeroConstantBorder<CLGEMMTranspose1xWKernel, 4>;
+using CLGEMMTranspose1xWFixture = GEMMTranspose1xWValidationFixture<CLTensor, CLAccessor, CLGEMMTranspose1xW, float>;
+TEST_SUITE(FP32)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMTranspose1xWFixture, framework::DatasetMode::PRECOMMIT, data_transpose * framework::dataset::make("DataType", DataType::F32))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // FP32
+
+TEST_SUITE(Quantized)
+TEST_SUITE(QS8)
+using CLGEMMTranspose1xW        = CLSynthetizeFunctionWithZeroConstantBorder<CLGEMMTranspose1xWKernel, 16>;
+using CLGEMMTranspose1xWFixture = GEMMTranspose1xWValidationFixedPointFixture<CLTensor, CLAccessor, CLGEMMTranspose1xW, int8_t>;
+FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMTranspose1xWFixture, framework::DatasetMode::PRECOMMIT, data_transpose *
+                       framework::dataset::make("DataType", DataType::QS8)
+                       * framework::dataset::make("FractionalBits", 1, 7))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END()
+
+TEST_SUITE(QS16)
+using CLGEMMTranspose1xW        = CLSynthetizeFunctionWithZeroConstantBorder<CLGEMMTranspose1xWKernel, 8>;
+using CLGEMMTranspose1xWFixture = GEMMTranspose1xWValidationFixedPointFixture<CLTensor, CLAccessor, CLGEMMTranspose1xW, int16_t>;
+FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMTranspose1xWFixture, framework::DatasetMode::PRECOMMIT, data_transpose *
+                       framework::dataset::make("DataType", DataType::QS16)
+                       * framework::dataset::make("FractionalBits", 1, 14))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END()
+
+TEST_SUITE_END()
+
+TEST_SUITE_END() //TRANSPOSE_1XW
+
 TEST_SUITE(Float)
 TEST_SUITE(FP16)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMFixture<half>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallGEMMDataset(), framework::dataset::make("DataType", DataType::F16)))
diff --git a/tests/validation/NEON/GEMM.cpp b/tests/validation/NEON/GEMM.cpp
index 57e8ce7ea9..1145d0b79e 100644
--- a/tests/validation/NEON/GEMM.cpp
+++ b/tests/validation/NEON/GEMM.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017, 2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -22,6 +22,7 @@
  * SOFTWARE.
  */
 #include "arm_compute/core/NEON/kernels/NEGEMMInterleave4x4Kernel.h"
+#include "arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/NEON/functions/NEGEMM.h"
 #include "arm_compute/runtime/Tensor.h"
@@ -37,6 +38,7 @@
 #include "tests/validation/Validation.h"
 #include "tests/validation/fixtures/GEMMFixture.h"
 #include "tests/validation/fixtures/GEMMInterleave4x4Fixture.h"
+#include "tests/validation/fixtures/GEMMTranspose1xWFixture.h"
 
 namespace arm_compute
 {
@@ -61,11 +63,53 @@ const auto CNNDataTypes = framework::dataset::make("DataType",
 });
 
 const auto data_interleave = framework::dataset::make("M", 8, 12) * framework::dataset::make("N", 8, 12);
+const auto data_transpose  = framework::dataset::make("M", 8, 14) * framework::dataset::make("N", 7, 14);
+
 } // namespace
 
 TEST_SUITE(NEON)
 TEST_SUITE(GEMM)
 
+TEST_SUITE(TRANSPOSE_1XW)
+using NEGEMMTranspose1xW        = NESynthetizeFunctionWithZeroConstantBorder<NEGEMMTranspose1xWKernel, 4>;
+using NEGEMMTranspose1xWFixture = GEMMTranspose1xWValidationFixture<Tensor, Accessor, NEGEMMTranspose1xW, float>;
+TEST_SUITE(FP32)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEGEMMTranspose1xWFixture, framework::DatasetMode::PRECOMMIT, data_transpose * framework::dataset::make("DataType", DataType::F32))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // FP32
+
+TEST_SUITE(Quantized)
+TEST_SUITE(QS8)
+using NEGEMMTranspose1xW        = NESynthetizeFunctionWithZeroConstantBorder<NEGEMMTranspose1xWKernel, 16>;
+using NEGEMMTranspose1xWFixture = GEMMTranspose1xWValidationFixedPointFixture<Tensor, Accessor, NEGEMMTranspose1xW, int8_t>;
+FIXTURE_DATA_TEST_CASE(RunSmall, NEGEMMTranspose1xWFixture, framework::DatasetMode::PRECOMMIT, data_transpose *
+                       framework::dataset::make("DataType", DataType::QS8)
+                       * framework::dataset::make("FractionalBits", 1, 7))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END()
+
+TEST_SUITE(QS16)
+using NEGEMMTranspose1xW        = NESynthetizeFunctionWithZeroConstantBorder<NEGEMMTranspose1xWKernel, 8>;
+using NEGEMMTranspose1xWFixture = GEMMTranspose1xWValidationFixedPointFixture<Tensor, Accessor, NEGEMMTranspose1xW, int16_t>;
+FIXTURE_DATA_TEST_CASE(RunSmall, NEGEMMTranspose1xWFixture, framework::DatasetMode::PRECOMMIT, data_transpose *
+                       framework::dataset::make("DataType", DataType::QS16)
+                       * framework::dataset::make("FractionalBits", 1, 14))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END()
+
+TEST_SUITE_END()
+
+TEST_SUITE_END() // TRANSPOSE_1XW
+
 TEST_SUITE(INTERLEAVE_4X4)
 using NEGEMMInterleave4x4 = NESynthetizeFunctionWithZeroConstantBorder<NEGEMMInterleave4x4Kernel, 4>;
 
diff --git a/tests/validation/fixtures/GEMMTranspose1xWFixture.h b/tests/validation/fixtures/GEMMTranspose1xWFixture.h
new file mode 100644
index 0000000000..d83d5e9c06
--- /dev/null
+++ b/tests/validation/fixtures/GEMMTranspose1xWFixture.h
@@ -0,0 +1,141 @@
+/*
+ * Copyright (c) 2017, 2018 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_GEMM_TRANSPOSE_1XW_FIXTURE
+#define ARM_COMPUTE_TEST_GEMM_TRANSPOSE_1XW_FIXTURE
+
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+#include "tests/AssetsLibrary.h"
+#include "tests/Globals.h"
+#include "tests/IAccessor.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Fixture.h"
+#include "tests/validation/Helpers.h"
+#include "tests/validation/reference/GEMMTranspose1xW.h"
+
+#include <random>
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class GEMMTranspose1xWValidationFixedPointFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(size_t x, size_t y, DataType data_type, int fractional_bits)
+    {
+        _fractional_bits = fractional_bits;
+        _data_type       = data_type;
+        const TensorShape  shape_a(x, y);
+        const unsigned int transpose_w = 16 / data_size_from_type(data_type);
+        const TensorShape  shape_b(static_cast<size_t>(y * transpose_w), static_cast<size_t>(std::ceil(x / static_cast<float>(transpose_w))));
+        _target    = compute_target(shape_a, shape_b, data_type, fractional_bits);
+        _reference = compute_reference(shape_a, shape_b, data_type, fractional_bits);
+    }
+
+protected:
+    template <typename U>
+    void fill(U &&tensor, int i)
+    {
+        switch(tensor.data_type())
+        {
+            case DataType::F16:
+            case DataType::F32:
+            {
+                std::uniform_real_distribution<> distribution(-1.f, 1.f);
+                library->fill(tensor, distribution, i);
+                break;
+            }
+            default:
+                library->fill_tensor_uniform(tensor, i);
+                break;
+        }
+    }
+
+    TensorType compute_target(const TensorShape &shape_a, const TensorShape &shape_b, DataType data_type, int fixed_point_position)
+    {
+        // Create tensors
+        TensorType a = create_tensor<TensorType>(shape_a, data_type, 1, fixed_point_position);
+        TensorType b = create_tensor<TensorType>(shape_b, data_type, 1, fixed_point_position);
+
+        // Create and configure function
+        FunctionType f;
+        f.configure(&a, &b);
+
+        ARM_COMPUTE_EXPECT(a.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(b.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Allocate tensors
+        a.allocator()->allocate();
+        b.allocator()->allocate();
+
+        ARM_COMPUTE_EXPECT(!a.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(!b.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Fill tensors
+        fill(AccessorType(a), 0);
+        fill(AccessorType(b), 1);
+
+        // Compute GEMM function
+        f.run();
+
+        return b;
+    }
+
+    SimpleTensor<T> compute_reference(const TensorShape &shape_a, const TensorShape &shape_b, DataType data_type, int fixed_point_position)
+    {
+        // Create reference
+        SimpleTensor<T> a{ shape_a, data_type, 1, fixed_point_position };
+
+        // Fill reference
+        fill(a, 0);
+
+        return reference::gemm_transpose_1xW<T>(a);
+    }
+
+    TensorType      _target{};
+    SimpleTensor<T> _reference{};
+    int             _fractional_bits{};
+    DataType        _data_type{};
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class GEMMTranspose1xWValidationFixture : public GEMMTranspose1xWValidationFixedPointFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+    template <typename...>
+    void setup(size_t x, size_t y, DataType data_type)
+    {
+        GEMMTranspose1xWValidationFixedPointFixture<TensorType, AccessorType, FunctionType, T>::setup(x, y, data_type, 0);
+    }
+};
+
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_GEMM_TRANSPOSE_1XW_FIXTURE */
diff --git a/tests/validation/reference/GEMMTranspose1xW.h b/tests/validation/reference/GEMMTranspose1xW.h
new file mode 100644
index 0000000000..d6a2e89176
--- /dev/null
+++ b/tests/validation/reference/GEMMTranspose1xW.h
@@ -0,0 +1,76 @@
+/*
+ * Copyright (c) 2017, 2018 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.
+ */
+#include "GEMM.h"
+
+#include "arm_compute/core/Types.h"
+#include "tests/validation/FixedPoint.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> gemm_transpose_1xW(const SimpleTensor<T> &in)
+{
+    const int         W = 16 / sizeof(T);
+    const TensorShape shape_out(static_cast<size_t>(in.shape().y() * W), static_cast<size_t>(std::ceil(in.shape().x() / static_cast<float>(W))));
+    SimpleTensor<T>   out(shape_out, in.data_type());
+    const int32_t     in_height     = in.shape().y();
+    const int32_t     in_width      = in.shape().x();
+    const int32_t     out_width     = out.shape().x();
+    const T          *in_base_addr  = reinterpret_cast<const T *>(in.data());
+    T                *out_base_addr = reinterpret_cast<T *>(out.data());
+    int               x             = 0;
+    for(; x < in_width; x += W)
+    {
+        for(int y = 0; y < in_height; y++)
+        {
+            const T *in_addr  = (in_base_addr + x + y * in_width);
+            T       *out_addr = (out_base_addr + y * W + (x / W) * out_width);
+
+            for(int k = 0; k < W; ++k)
+            {
+                // If the input width is not multiple of W, we fill the reference with 0s
+                if((x + k) >= in_width)
+                {
+                    out_addr[k] = T(0);
+                }
+                else
+                {
+                    out_addr[k] = in_addr[k];
+                }
+            }
+        }
+    }
+    return out;
+}
+
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute