COMPMID-765 - Extended GEMM benchmark

Added new benchmarks GEMM in order to evaluate the performance when the
input matrix B has to be reshaped only once

Change-Id: I1c4790213704ce57ea7b28f6f362c56edccd1eb9
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/118910
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele DiGiorgio <michele.digiorgio@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

6 files changed, 57 insertions, 26 deletions

diff --git a/arm_compute/runtime/GLES_COMPUTE/functions/GCGEMM.h b/arm_compute/runtime/GLES_COMPUTE/functions/GCGEMM.h
index f2484cd801..8ddfae1169 100644
--- a/arm_compute/runtime/GLES_COMPUTE/functions/GCGEMM.h
+++ b/arm_compute/runtime/GLES_COMPUTE/functions/GCGEMM.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -58,14 +58,16 @@ public:
      *
      * @note Whilst the first input tensor can be a vector, the second input tensor must be at least a matrix
      *
-     * @param[in]  a      First input tensor  (Matrix or Vector A). Data types supported: F32
-     * @param[in]  b      Second input tensor (Matrix B). Data type supported: same as @p a.
-     * @param[in]  c      Third input tensor  (Matrix C). It can be a nullptr if just the multiplication between @p a and @p b is needed. Data type supported: same as @p a.
-     * @param[out] output Output tensor. Data type supported: same as @p a
-     * @param[in]  alpha  Weight of the matrix product
-     * @param[in]  beta   Weight of matrix C
+     * @param[in]  a         First input tensor  (Matrix or Vector A). Data types supported: F32
+     * @param[in]  b         Second input tensor (Matrix B). Data type supported: same as @p a.
+     * @param[in]  c         Third input tensor  (Matrix C). It can be a nullptr if just the multiplication between @p a and @p b is needed. Data type supported: same as @p a.
+     * @param[out] output    Output tensor. Data type supported: same as @p a
+     * @param[in]  alpha     Weight of the matrix product
+     * @param[in]  beta      Weight of matrix C
+     * @param[in]  gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and
+     *                       if the reshape of matrix B should happen only for the first run
      */
-    void configure(const IGCTensor *a, const IGCTensor *b, const IGCTensor *c, IGCTensor *output, float alpha, float beta);
+    void configure(const IGCTensor *a, const IGCTensor *b, const IGCTensor *c, IGCTensor *output, float alpha, float beta, const GEMMInfo &gemm_info = GEMMInfo());
 
     // Inherited methods overridden:
     void run() override;
diff --git a/src/runtime/GLES_COMPUTE/functions/GCGEMM.cpp b/src/runtime/GLES_COMPUTE/functions/GCGEMM.cpp
index 7aa2d421eb..5122c20504 100644
--- a/src/runtime/GLES_COMPUTE/functions/GCGEMM.cpp
+++ b/src/runtime/GLES_COMPUTE/functions/GCGEMM.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -45,10 +45,14 @@ GCGEMM::GCGEMM()
 {
 }
 
-void GCGEMM::configure(const IGCTensor *a, const IGCTensor *b, const IGCTensor *c, IGCTensor *output, float alpha, float beta)
+void GCGEMM::configure(const IGCTensor *a, const IGCTensor *b, const IGCTensor *c, IGCTensor *output, float alpha, float beta, const GEMMInfo &gemm_info)
 {
     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(a, 1, DataType::F32);
     ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(a, b, output);
+    ARM_COMPUTE_ERROR_ON_MSG(gemm_info.is_a_reshaped(), "Matrix A already reshaped is not supported");
+    ARM_COMPUTE_ERROR_ON_MSG(gemm_info.is_b_reshaped(), "Matrix B already reshaped is not supported");
+    ARM_COMPUTE_ERROR_ON_MSG(gemm_info.reshape_b_only_on_first_run(), "Reshape matrix B only on first run is not supported");
+    ARM_COMPUTE_UNUSED(gemm_info);
 
     if(c != nullptr)
     {
diff --git a/tests/benchmark/CL/GEMM.cpp b/tests/benchmark/CL/GEMM.cpp
index 615b492ee0..60362b43e1 100644
--- a/tests/benchmark/CL/GEMM.cpp
+++ b/tests/benchmark/CL/GEMM.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, 2018 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -41,17 +41,26 @@ namespace test
 {
 namespace
 {
-const auto data_types = framework::dataset::make("DataType", { DataType::F16, DataType::F32 });
+const auto data_types          = framework::dataset::make("DataType", { DataType::F16, DataType::F32 });
+const auto reshape_b_only_once = framework::dataset::make("ReshapeBOnlyOnce", { false, true });
 } // namespace
 
 using CLGEMMFixture = GEMMFixture<CLTensor, CLGEMM, CLAccessor>;
 
 TEST_SUITE(CL)
 
-REGISTER_FIXTURE_DATA_TEST_CASE(GoogLeNetInceptionV1GEMM, CLGEMMFixture, framework::DatasetMode::ALL, framework::dataset::combine(datasets::GoogLeNetInceptionV1GEMMDataset(), data_types));
-REGISTER_FIXTURE_DATA_TEST_CASE(MatrixMultiplyGEMM, CLGEMMFixture, framework::DatasetMode::ALL, framework::dataset::combine(datasets::MatrixMultiplyGEMMDataset(), data_types));
-REGISTER_FIXTURE_DATA_TEST_CASE(GoogleNetGEMM, CLGEMMFixture, framework::DatasetMode::NIGHTLY, framework::dataset::combine(datasets::GoogleNetGEMMDataset(), data_types));
-REGISTER_FIXTURE_DATA_TEST_CASE(AlexNetGEMM, CLGEMMFixture, framework::DatasetMode::NIGHTLY, framework::dataset::combine(datasets::AlexNetGEMMDataset(), data_types));
+REGISTER_FIXTURE_DATA_TEST_CASE(GoogLeNetInceptionV1GEMM, CLGEMMFixture, framework::DatasetMode::ALL,
+                                framework::dataset::combine(framework::dataset::combine(datasets::GoogLeNetInceptionV1GEMMDataset(),
+                                                                                        data_types),
+                                                            reshape_b_only_once));
+REGISTER_FIXTURE_DATA_TEST_CASE(MatrixMultiplyGEMM, CLGEMMFixture, framework::DatasetMode::ALL, framework::dataset::combine(framework::dataset::combine(datasets::MatrixMultiplyGEMMDataset(),
+                                data_types),
+                                reshape_b_only_once));
+REGISTER_FIXTURE_DATA_TEST_CASE(GoogleNetGEMM, CLGEMMFixture, framework::DatasetMode::NIGHTLY, framework::dataset::combine(framework::dataset::combine(datasets::GoogleNetGEMMDataset(), data_types),
+                                reshape_b_only_once));
+REGISTER_FIXTURE_DATA_TEST_CASE(AlexNetGEMM, CLGEMMFixture, framework::DatasetMode::NIGHTLY, framework::dataset::combine(framework::dataset::combine(datasets::AlexNetGEMMDataset(),
+                                data_types),
+                                reshape_b_only_once));
 
 TEST_SUITE_END()
 } // namespace test
diff --git a/tests/benchmark/GLES_COMPUTE/GEMM.cpp b/tests/benchmark/GLES_COMPUTE/GEMM.cpp
index 69d8e54919..97c92c679b 100644
--- a/tests/benchmark/GLES_COMPUTE/GEMM.cpp
+++ b/tests/benchmark/GLES_COMPUTE/GEMM.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -40,16 +40,24 @@ namespace test
 {
 namespace
 {
-const auto data_types = framework::dataset::make("DataType", { DataType::F32 });
+const auto data_types          = framework::dataset::make("DataType", { DataType::F32 });
+const auto reshape_b_only_once = framework::dataset::make("ReshapeBOnlyOnce", { false });
 } // namespace
 
 using GCGEMMFixture = GEMMFixture<GCTensor, GCGEMM, GCAccessor>;
 
 TEST_SUITE(GC)
 
-REGISTER_FIXTURE_DATA_TEST_CASE(GoogLeNetInceptionV1GEMM, GCGEMMFixture, framework::DatasetMode::ALL, framework::dataset::combine(datasets::GoogLeNetInceptionV1GEMMDataset(), data_types));
-REGISTER_FIXTURE_DATA_TEST_CASE(MatrixMultiplyGEMM, GCGEMMFixture, framework::DatasetMode::ALL, framework::dataset::combine(datasets::MatrixMultiplyGEMMDataset(), data_types));
-REGISTER_FIXTURE_DATA_TEST_CASE(GoogleNetGEMM, GCGEMMFixture, framework::DatasetMode::NIGHTLY, framework::dataset::combine(datasets::GoogleNetGEMMDataset(), data_types));
+REGISTER_FIXTURE_DATA_TEST_CASE(GoogLeNetInceptionV1GEMM, GCGEMMFixture, framework::DatasetMode::ALL,
+                                framework::dataset::combine(framework::dataset::combine(datasets::GoogLeNetInceptionV1GEMMDataset(),
+                                                                                        data_types),
+                                                            reshape_b_only_once));
+REGISTER_FIXTURE_DATA_TEST_CASE(MatrixMultiplyGEMM, GCGEMMFixture, framework::DatasetMode::ALL, framework::dataset::combine(framework::dataset::combine(datasets::MatrixMultiplyGEMMDataset(),
+                                data_types),
+                                reshape_b_only_once));
+REGISTER_FIXTURE_DATA_TEST_CASE(GoogleNetGEMM, GCGEMMFixture, framework::DatasetMode::NIGHTLY, framework::dataset::combine(framework::dataset::combine(datasets::GoogleNetGEMMDataset(),
+                                data_types),
+                                reshape_b_only_once));
 
 TEST_SUITE_END()
 } // namespace test
diff --git a/tests/benchmark/NEON/GEMM.cpp b/tests/benchmark/NEON/GEMM.cpp
index 1d6ea8df7f..473d6429fb 100644
--- a/tests/benchmark/NEON/GEMM.cpp
+++ b/tests/benchmark/NEON/GEMM.cpp
@@ -48,15 +48,23 @@ const auto data_types = framework::dataset::make("DataType",
 #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
     DataType::F32
 });
+const auto reshape_b_only_once = framework::dataset::make("ReshapeBOnlyOnce", { false, true });
 } // namespace
 
 using NEGEMMFixture = GEMMFixture<Tensor, NEGEMM, Accessor>;
 
 TEST_SUITE(NEON)
 
-REGISTER_FIXTURE_DATA_TEST_CASE(GoogLeNetInceptionV1GEMM, NEGEMMFixture, framework::DatasetMode::ALL, framework::dataset::combine(datasets::GoogLeNetInceptionV1GEMMDataset(), data_types));
-REGISTER_FIXTURE_DATA_TEST_CASE(MatrixMultiplyGEMM, NEGEMMFixture, framework::DatasetMode::ALL, framework::dataset::combine(datasets::MatrixMultiplyGEMMDataset(), data_types));
-REGISTER_FIXTURE_DATA_TEST_CASE(GoogleNetGEMM, NEGEMMFixture, framework::DatasetMode::NIGHTLY, framework::dataset::combine(datasets::GoogleNetGEMMDataset(), data_types));
+REGISTER_FIXTURE_DATA_TEST_CASE(GoogLeNetInceptionV1GEMM, NEGEMMFixture, framework::DatasetMode::ALL,
+                                framework::dataset::combine(framework::dataset::combine(datasets::GoogLeNetInceptionV1GEMMDataset(),
+                                                                                        data_types),
+                                                            reshape_b_only_once));
+REGISTER_FIXTURE_DATA_TEST_CASE(MatrixMultiplyGEMM, NEGEMMFixture, framework::DatasetMode::ALL, framework::dataset::combine(framework::dataset::combine(datasets::MatrixMultiplyGEMMDataset(),
+                                data_types),
+                                reshape_b_only_once));
+REGISTER_FIXTURE_DATA_TEST_CASE(GoogleNetGEMM, NEGEMMFixture, framework::DatasetMode::NIGHTLY, framework::dataset::combine(framework::dataset::combine(datasets::GoogleNetGEMMDataset(),
+                                data_types),
+                                reshape_b_only_once));
 
 TEST_SUITE_END()
 } // namespace test
diff --git a/tests/benchmark/fixtures/GEMMFixture.h b/tests/benchmark/fixtures/GEMMFixture.h
index e958d4fdde..3a308d9ada 100644
--- a/tests/benchmark/fixtures/GEMMFixture.h
+++ b/tests/benchmark/fixtures/GEMMFixture.h
@@ -40,7 +40,7 @@ class GEMMFixture : public framework::Fixture
 {
 public:
     template <typename...>
-    void setup(TensorShape shape_a, TensorShape shape_b, TensorShape shape_c, TensorShape shape_dst, float alpha, float beta, DataType data_type)
+    void setup(TensorShape shape_a, TensorShape shape_b, TensorShape shape_c, TensorShape shape_dst, float alpha, float beta, DataType data_type, bool reshape_b_only_on_first_run)
     {
         constexpr int fixed_point_position = 4;
 
@@ -51,7 +51,7 @@ public:
         dst = create_tensor<TensorType>(shape_dst, data_type, 1, fixed_point_position);
 
         // Create and configure function
-        gemm.configure(&a, &b, &c, &dst, alpha, beta);
+        gemm.configure(&a, &b, &c, &dst, alpha, beta, GEMMInfo(false, false, reshape_b_only_on_first_run));
 
         // Allocate tensors
         a.allocator()->allocate();