Optimize NEStackLayer
Optimize the stack operation in Cpu by leveraging block memcpy.
Resolves: COMPMID-6498
Change-Id: I49d79d179f0375a73d654edd59fb33072112569b
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10451
Reviewed-by: SiCong Li <sicong.li@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
diff --git a/tests/validation/NEON/StackLayer.cpp b/tests/validation/NEON/StackLayer.cpp
index d88f713..3828010 100644
--- a/tests/validation/NEON/StackLayer.cpp
+++ b/tests/validation/NEON/StackLayer.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018-2021 Arm Limited.
+ * Copyright (c) 2018-2021, 2023 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -44,69 +44,74 @@
{
namespace validation
{
+
+using framework::dataset::make;
namespace
{
// *INDENT-OFF*
// clang-format off
/** Data types */
-const auto data_types = framework::dataset::make("DataType", { DataType::QASYMM8, DataType::F16, DataType::F32 });
+const auto data_types = make("DataType", { DataType::QASYMM8, DataType::F16, DataType::F32 });
/** Num tensors values to test */
-const auto n_values = framework::dataset::make("NumTensors", { 3, 4 });
+const auto n_values = make("NumTensors", { 3, 4 });
/** Shapes 1D to test */
-const auto shapes_1d_small = combine(datasets::Small1DShapes(), framework::dataset::make("Axis", -1, 2));
+const auto shapes_1d_small = combine(datasets::Small1DShapes(), make("Axis", -1, 2));
/** Shapes 2D to test */
-const auto shapes_2d_small = combine(datasets::Small2DShapes(), framework::dataset::make("Axis", -2, 3));
+const auto shapes_2d_small = combine(datasets::Small2DShapes(), make("Axis", -2, 3));
/** Shapes 3D to test */
-const auto shapes_3d_small = combine(datasets::Small3DShapes(), framework::dataset::make("Axis", -3, 4));
+const auto shapes_3d_small = combine(datasets::Small3DShapes(), make("Axis", -3, 4));
/** Shapes 4D to test */
-const auto shapes_4d_small = combine(datasets::Small4DShapes(), framework::dataset::make("Axis", -4, 5));
+const auto shapes_4d_small = combine(datasets::Small4DShapes(), make("Axis", -4, 5));
/** Shapes 1D to test */
-const auto shapes_1d_large = combine(datasets::Large1DShapes(), framework::dataset::make("Axis", -1, 2));
+const auto shapes_1d_large = combine(datasets::Large1DShapes(), make("Axis", -1, 2));
/** Shapes 2D to test */
-const auto shapes_2d_large = combine(datasets::Medium2DShapes(), framework::dataset::make("Axis", -2, 3));
+const auto shapes_2d_large = combine(datasets::Medium2DShapes(), make("Axis", -2, 3));
/** Shapes 3D to test */
-const auto shapes_3d_large = combine(datasets::Medium3DShapes(), framework::dataset::make("Axis", -3, 4));
+const auto shapes_3d_large = combine(datasets::Medium3DShapes(), make("Axis", -3, 4));
/** Shapes 4D to test */
-const auto shapes_4d_large = combine(datasets::Medium4DShapes(), framework::dataset::make("Axis", -4, 5));
+const auto shapes_4d_large = combine(datasets::Medium4DShapes(), make("Axis", -4, 5));
} // namespace
/** Fixture to use */
template<typename T>
using NEStackLayerFixture = StackLayerValidationFixture<Tensor, ITensor, Accessor, NEStackLayer, T>;
+template<typename T>
+using NEStackLayerWithPaddingFixture = StackLayerWithPaddingValidationFixture<Tensor, ITensor, Accessor, NEStackLayer, T>;
+
using namespace arm_compute::misc::shape_calculator;
TEST_SUITE(NEON)
TEST_SUITE(StackLayer)
-DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
- framework::dataset::make("InputInfo",
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(
+make("InputInfo",
{
std::vector<TensorInfo>{ TensorInfo(TensorShape(9U, 8U), 1, DataType::U8) },
- std::vector<TensorInfo>{ TensorInfo(TensorShape(1U, 2U), 1, DataType::U8) , TensorInfo(TensorShape(1U, 2U), 1, DataType::U8), TensorInfo(TensorShape(1U, 2U), 1, DataType::U8)},
+ std::vector<TensorInfo>{ TensorInfo(TensorShape(1U, 2U), 1, DataType::U8) , TensorInfo(TensorShape(1U, 2U), 1, DataType::U8), TensorInfo(TensorShape(1U, 2U), 1, DataType::U8)},
std::vector<TensorInfo>{ TensorInfo(TensorShape(2U, 3U), 1, DataType::S32) },
- std::vector<TensorInfo>{ TensorInfo(TensorShape(7U, 5U, 3U, 8U, 2U), 1, DataType::S32), TensorInfo(TensorShape(7U, 5U, 3U, 8U, 2U), 1, DataType::S32)},
+ std::vector<TensorInfo>{ TensorInfo(TensorShape(7U, 5U, 3U, 8U, 2U), 1, DataType::S32), TensorInfo(TensorShape(7U, 5U, 3U, 8U, 2U), 1, DataType::S32)},
std::vector<TensorInfo>{ TensorInfo(TensorShape(9U, 8U), 1, DataType::S32) },
}),
-framework::dataset::make("OutputInfo",
+make("OutputInfo",
{
TensorInfo(TensorShape(1U, 9U, 8U), 1, DataType::U8), // Passes, stack 1 tensor on x axis
TensorInfo(TensorShape(1U, 3U, 2U), 1, DataType::U8), // Passes, stack 3 tensors on y axis
TensorInfo(TensorShape(1U, 2U, 3U), 1, DataType::S32), // fails axis < (- input's rank)
TensorInfo(TensorShape(3U, 7U, 5U), 1, DataType::S32), // fails, input dimensions > 4
TensorInfo(TensorShape(1U, 2U, 3U), 1, DataType::U8), // fails mismatching data types
-})),
-framework::dataset::make("Axis", { -3, 1, -4, -3, 1 })),
-framework::dataset::make("Expected", { true, true, false, false, false })),
+}),
+make("Axis", { -3, 1, -4, -3, 1 }),
+make("Expected", { true, true, false, false, false })),
input_info, output_info, axis, expected)
{
std::vector<TensorInfo> ti(input_info);
@@ -121,18 +126,18 @@
TEST_SUITE(Shapes1D)
TEST_SUITE(S32)
FIXTURE_DATA_TEST_CASE(RunSmall, NEStackLayerFixture<int>, framework::DatasetMode::ALL,
- combine(combine(shapes_1d_small,
- framework::dataset::make("DataType", { DataType::S32 })),
- n_values))
+ combine(shapes_1d_small,
+ make("DataType", { DataType::S32 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
FIXTURE_DATA_TEST_CASE(RunLarge, NEStackLayerFixture<int>, framework::DatasetMode::NIGHTLY,
- combine(combine(shapes_1d_large,
- framework::dataset::make("DataType", { DataType::S32 })),
- n_values))
+ combine(shapes_1d_large,
+ make("DataType", { DataType::S32 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -141,18 +146,18 @@
TEST_SUITE(S16)
FIXTURE_DATA_TEST_CASE(RunSmall, NEStackLayerFixture<short>, framework::DatasetMode::ALL,
- combine(combine(shapes_1d_small,
- framework::dataset::make("DataType", { DataType::S16 })),
- n_values))
+ combine(shapes_1d_small,
+ make("DataType", { DataType::S16 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
FIXTURE_DATA_TEST_CASE(RunLarge, NEStackLayerFixture<short>, framework::DatasetMode::NIGHTLY,
- combine(combine(shapes_1d_large,
- framework::dataset::make("DataType", { DataType::S16 })),
- n_values))
+ combine(shapes_1d_large,
+ make("DataType", { DataType::S16 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -161,18 +166,18 @@
TEST_SUITE(S8)
FIXTURE_DATA_TEST_CASE(RunSmall, NEStackLayerFixture<char>, framework::DatasetMode::ALL,
- combine(combine(shapes_1d_small,
- framework::dataset::make("DataType", { DataType::S8 })),
- n_values))
+ combine(shapes_1d_small,
+ make("DataType", { DataType::S8 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
FIXTURE_DATA_TEST_CASE(RunLarge, NEStackLayerFixture<char>, framework::DatasetMode::NIGHTLY,
- combine(combine(shapes_1d_large,
- framework::dataset::make("DataType", { DataType::S8 })),
- n_values))
+ combine(shapes_1d_large,
+ make("DataType", { DataType::S8 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -183,18 +188,18 @@
TEST_SUITE(Shapes2D)
TEST_SUITE(S32)
FIXTURE_DATA_TEST_CASE(RunSmall, NEStackLayerFixture<int>, framework::DatasetMode::ALL,
- combine(combine(shapes_2d_small,
- framework::dataset::make("DataType", { DataType::S32 })),
- n_values))
+ combine(shapes_2d_small,
+ make("DataType", { DataType::S32 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
FIXTURE_DATA_TEST_CASE(RunLarge, NEStackLayerFixture<int>, framework::DatasetMode::NIGHTLY,
- combine(combine(shapes_2d_large,
- framework::dataset::make("DataType", { DataType::S32 })),
- n_values))
+ combine(shapes_2d_large,
+ make("DataType", { DataType::S32 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -203,18 +208,18 @@
TEST_SUITE(S16)
FIXTURE_DATA_TEST_CASE(RunSmall, NEStackLayerFixture<short>, framework::DatasetMode::ALL,
- combine(combine(shapes_2d_small,
- framework::dataset::make("DataType", { DataType::S16 })),
- n_values))
+ combine(shapes_2d_small,
+ make("DataType", { DataType::S16 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
FIXTURE_DATA_TEST_CASE(RunLarge, NEStackLayerFixture<short>, framework::DatasetMode::NIGHTLY,
- combine(combine(shapes_2d_large,
- framework::dataset::make("DataType", { DataType::S16 })),
- n_values))
+ combine(shapes_2d_large,
+ make("DataType", { DataType::S16 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -223,18 +228,18 @@
TEST_SUITE(S8)
FIXTURE_DATA_TEST_CASE(RunSmall, NEStackLayerFixture<char>, framework::DatasetMode::ALL,
- combine(combine(shapes_2d_small,
- framework::dataset::make("DataType", { DataType::S8 })),
- n_values))
+ combine(shapes_2d_small,
+ make("DataType", { DataType::S8 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
FIXTURE_DATA_TEST_CASE(RunLarge, NEStackLayerFixture<char>, framework::DatasetMode::NIGHTLY,
- combine(combine(shapes_2d_large,
- framework::dataset::make("DataType", { DataType::S8 })),
- n_values))
+ combine(shapes_2d_large,
+ make("DataType", { DataType::S8 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -245,18 +250,18 @@
TEST_SUITE(Shapes3D)
TEST_SUITE(S32)
FIXTURE_DATA_TEST_CASE(RunSmall, NEStackLayerFixture<int>, framework::DatasetMode::ALL,
- combine(combine(shapes_3d_small,
- framework::dataset::make("DataType", { DataType::S32 })),
- n_values))
+ combine(shapes_3d_small,
+ make("DataType", { DataType::S32 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
FIXTURE_DATA_TEST_CASE(RunLarge, NEStackLayerFixture<int>, framework::DatasetMode::NIGHTLY,
- combine(combine(shapes_3d_large,
- framework::dataset::make("DataType", { DataType::S32 })),
- n_values))
+ combine(shapes_3d_large,
+ make("DataType", { DataType::S32 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -265,18 +270,18 @@
TEST_SUITE(S16)
FIXTURE_DATA_TEST_CASE(RunSmall, NEStackLayerFixture<short>, framework::DatasetMode::ALL,
- combine(combine(shapes_3d_small,
- framework::dataset::make("DataType", { DataType::S16 })),
- n_values))
+ combine(shapes_3d_small,
+ make("DataType", { DataType::S16 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
FIXTURE_DATA_TEST_CASE(RunLarge, NEStackLayerFixture<short>, framework::DatasetMode::NIGHTLY,
- combine(combine(shapes_3d_large,
- framework::dataset::make("DataType", { DataType::S16 })),
- n_values))
+ combine(shapes_3d_large,
+ make("DataType", { DataType::S16 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -285,18 +290,18 @@
TEST_SUITE(S8)
FIXTURE_DATA_TEST_CASE(RunSmall, NEStackLayerFixture<char>, framework::DatasetMode::ALL,
- combine(combine(shapes_3d_small,
- framework::dataset::make("DataType", { DataType::S8 })),
- n_values))
+ combine(shapes_3d_small,
+ make("DataType", { DataType::S8 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
FIXTURE_DATA_TEST_CASE(RunLarge, NEStackLayerFixture<char>, framework::DatasetMode::NIGHTLY,
- combine(combine(shapes_3d_large,
- framework::dataset::make("DataType", { DataType::S8 })),
- n_values))
+ combine(shapes_3d_large,
+ make("DataType", { DataType::S8 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -307,18 +312,29 @@
TEST_SUITE(Shapes4D)
TEST_SUITE(S32)
FIXTURE_DATA_TEST_CASE(RunSmall, NEStackLayerFixture<int>, framework::DatasetMode::ALL,
- combine(combine(shapes_4d_small,
- framework::dataset::make("DataType", { DataType::S32 })),
- n_values))
+ combine(shapes_4d_small,
+ make("DataType", { DataType::S32 }),
+ n_values))
+{
+ // Validate output
+ validate(Accessor(_target), _reference);
+}
+
+// Testing the case with padding for only 4d shapes and for one data type. This is because the underlying code
+// path depends only on the padding, which isn't affected by the shapes or data types.
+FIXTURE_DATA_TEST_CASE(RunSmallWithPadding, NEStackLayerWithPaddingFixture<int>, framework::DatasetMode::ALL,
+ combine(shapes_4d_small,
+ make("DataType", { DataType::S32 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
FIXTURE_DATA_TEST_CASE(RunLarge, NEStackLayerFixture<int>, framework::DatasetMode::NIGHTLY,
- combine(combine(shapes_4d_large,
- framework::dataset::make("DataType", { DataType::S32 })),
- n_values))
+ combine(shapes_4d_large,
+ make("DataType", { DataType::S32 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -327,18 +343,18 @@
TEST_SUITE(S16)
FIXTURE_DATA_TEST_CASE(RunSmall, NEStackLayerFixture<short>, framework::DatasetMode::ALL,
- combine(combine(shapes_4d_small,
- framework::dataset::make("DataType", { DataType::S16 })),
- n_values))
+ combine(shapes_4d_small,
+ make("DataType", { DataType::S16 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
FIXTURE_DATA_TEST_CASE(RunLarge, NEStackLayerFixture<short>, framework::DatasetMode::NIGHTLY,
- combine(combine(shapes_4d_large,
- framework::dataset::make("DataType", { DataType::S16 })),
- n_values))
+ combine(shapes_4d_large,
+ make("DataType", { DataType::S16 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
@@ -347,24 +363,37 @@
TEST_SUITE(S8)
FIXTURE_DATA_TEST_CASE(RunSmall, NEStackLayerFixture<char>, framework::DatasetMode::ALL,
- combine(combine(shapes_4d_small,
- framework::dataset::make("DataType", { DataType::S8 })),
- n_values))
+ combine(shapes_4d_small,
+ make("DataType", { DataType::S8 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
FIXTURE_DATA_TEST_CASE(RunLarge, NEStackLayerFixture<char>, framework::DatasetMode::NIGHTLY,
- combine(combine(shapes_4d_large,
- framework::dataset::make("DataType", { DataType::S8 })),
- n_values))
+ combine(shapes_4d_large,
+ make("DataType", { DataType::S8 }),
+ n_values))
{
// Validate output
validate(Accessor(_target), _reference);
}
TEST_SUITE_END() // S8
TEST_SUITE_END() // Shapes4D
+
+TEST_SUITE(HighDimensional)
+// The Cpu implementation supports tensors of size 4D+, but reference implementation does not.
+FIXTURE_DATA_TEST_CASE(RunHighDimensional, NEStackLayerFixture<char>, framework::DatasetMode::DISABLED,
+ combine(make("Shape", { TensorShape{2U, 3U, 4U, 5U, 3U} }),
+ make("Axis", { 5, 0, -3, 2 }),
+ make("DataType", { DataType::S8 }),
+ make("NumTensors", { 3 })))
+{
+ // Validate output
+ validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // HighDimensional
TEST_SUITE_END() // StackLayer
TEST_SUITE_END() // Neon
} // namespace validation