Add new operator AddMulAdd for Neon™ backend for Float/Quantized types
This is a fused operator that merges Add + Mul + Add [+ Relu-based-Activation] layers and have an intermediate output after the first Add. It's supported for FP16/32/QASYMM8/QASYMM8_SIGNED data types.
The subsequent Add and Mul are intended for scaling and the coefficients only have one dimension (per channel).
The inputs are
- input1 : nD tensor [X, Y, Z, W, ..]
- input2 : nD tensor [X, Y, Z, W, ..]
- add_coef : 1D tensor [X]
- mul_coef : 1D tensor [X]
The outputs are
- out1 : nD tensor (intermediate output) [X, Y, Z, W, ..]
- out2 : nD tensor (final output) [X, Y, Z, W, ..]
The operation can be summarized as follows:
out1 <- input1 + input2
out2 <- Act(out1 * mul_coef + add_coef)
The activation function can be Identity, Relu, Bounded Relu or Lower/Upper Bounded Relu. The intermediate output can be skipped by providing a nullptr.
The reason of providing this operator is to be able to fuse in case of Residual network patterns and save computations by reducing memory back and forward.
Resolves: COMPMID-5463
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Change-Id: I8ef577aa623b036e9a9f655cc088493fd19a6109
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9055
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
diff --git a/tests/validation/NEON/AddMulAdd.cpp b/tests/validation/NEON/AddMulAdd.cpp
new file mode 100644
index 0000000..f0aba78
--- /dev/null
+++ b/tests/validation/NEON/AddMulAdd.cpp
@@ -0,0 +1,230 @@
+/*
+ * Copyright (c) 2023 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 "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/functions/NEAddMulAdd.h"
+#include "arm_compute/runtime/Tensor.h"
+#include "arm_compute/runtime/TensorAllocator.h"
+
+#include "tests/NEON/Accessor.h"
+#include "tests/datasets/ShapeDatasets.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/AddMulAddFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+constexpr AbsoluteTolerance<float> tolerance_fp32(0.001f); /**< Tolerance for floating point tests */
+const AbsoluteTolerance<half> tolerance_fp16(half(0.1f)); /**< Tolerance for 16-bit floating point tests */
+constexpr AbsoluteTolerance<float> tolerance_quant(1); /**< Tolerance for quantized tests */
+
+const auto ActivationFunctionsDataset = framework::dataset::make("ActivationInfo",
+{
+ ActivationLayerInfo(),
+ ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU),
+
+ // Boundaries are aligned with Quantized Data ranges -- DOUBLE check before changing
+ ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::BOUNDED_RELU, 6.f),
+ ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 8.f, -2.f)
+});
+
+// QASYMM8 test quantizations
+const auto qasymm8_input1_qinfo_set = framework::dataset::make("Input1QInfo", { QuantizationInfo(0.1, 10) }); // Representable Range: [-1, 24.5]
+const auto qasymm8_input2_qinfo_set = framework::dataset::make("Input2QInfo", { QuantizationInfo(0.2, 60) }); // Representable Range: [-12, 39]
+const auto qasymm8_bn_mul_qinfo_set = framework::dataset::make("BnMulInfo", { QuantizationInfo(0.001, 55) }); // Representable Range: [-0.11, 0.2]
+const auto qasymm8_bn_add_qinfo_set = framework::dataset::make("BnAddInfo", { QuantizationInfo(0.02, 20) }); // Representable Range: [-0.4, 4.7]
+
+// Representable Range: [-9.36, 51.84], Expected F32 range: [-13, 63.5], leaving some space for saturation
+const auto qasymm8_add_output_qinfo_set = framework::dataset::make("AddOutputInfo", { QuantizationInfo(0.24, 39) });
+
+// Representable Range: [-4.8, 10.5], Expected FP32 range: [-6.985, 12.7], leaving some space for saturation
+// This range also makes sense with the activation boundaries above, i.e. [-2, 8] for LU_BOUNDED_RELU and [0, 6] for BOUNDED_RELU
+const auto qasymm8_final_output_qinfo_set = framework::dataset::make("FinalOutputInfo", { QuantizationInfo(0.06, 80) });
+
+// QASYMM8_SIGNED test quantizations
+const auto qasymm8_signed_input1_qinfo_set = framework::dataset::make("Input1QInfo", { QuantizationInfo(0.1, 10) }); // Representable Range: [-13.8, 11.7]
+const auto qasymm8_signed_input2_qinfo_set = framework::dataset::make("Input2QInfo", { QuantizationInfo(0.2, -60) }); // Representable Range: [-13.6, 39.4]
+const auto qasymm8_signed_bn_mul_qinfo_set = framework::dataset::make("BnMulInfo", { QuantizationInfo(0.001, 55) }); // Representable Range: [-0.183, 0.072]
+const auto qasymm8_signed_bn_add_qinfo_set = framework::dataset::make("BnAddInfo", { QuantizationInfo(0.4, -120) }); // Representable Range: [-0.32, 9.08]
+
+// Representable Range: [-21.36, 39.84], Expected F32 range: [-27.4, 51.1], leaving some space for saturation
+const auto qasymm8_signed_add_output_qinfo_set = framework::dataset::make("AddOutputInfo", { QuantizationInfo(0.24, -39) });
+
+// Representable Range: [-4.8, 10.5], Expected FP32 range: [-9.6713, 14.0942], leaving some space for saturation
+// This range also makes sense with the activation boundaries above, i.e. [-2, 8] for LU_BOUNDED_RELU and [0, 6] for BOUNDED_RELU
+const auto qasymm8_signed_final_output_qinfo_set = framework::dataset::make("FinalOutputInfo", { QuantizationInfo(0.06, -48) });
+
+} // namespace
+
+TEST_SUITE(NEON)
+TEST_SUITE(AddMulAdd)
+
+template <typename T>
+using NEAddMulAddFloatFixture = AddMulAddFloatValidationFixture<Tensor, Accessor, NEAddMulAdd, T, true>;
+
+template <typename T>
+using NEAddMulAddFloatFixtureWoIntermOut = AddMulAddFloatValidationFixture<Tensor, Accessor, NEAddMulAdd, T, false>;
+
+TEST_SUITE(Float)
+
+TEST_SUITE(F32)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEAddMulAddFloatFixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(),
+ framework::dataset::make("DataType", DataType::F32)),
+ ActivationFunctionsDataset))
+{
+ // Validate outputs
+ validate(Accessor(_interm_target), _interm_reference); // Arithmetic Addition has more strict tolerance
+ validate(Accessor(_target), _reference, tolerance_fp32);
+}
+
+// This test is to stress the case when there is no intermediate output required (i.e. nullptr)
+FIXTURE_DATA_TEST_CASE(RunSmallWithoutIntermOutput, NEAddMulAddFloatFixtureWoIntermOut<float>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(),
+ framework::dataset::make("DataType", DataType::F32)),
+ framework::dataset::make("ActivationInfo", { ActivationLayerInfo() })))
+{
+ // Validate outputs
+ validate(Accessor(_target), _reference, tolerance_fp32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, NEAddMulAddFloatFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(),
+ framework::dataset::make("DataType", DataType::F32)),
+ ActivationFunctionsDataset))
+{
+ // Validate outputs
+ validate(Accessor(_interm_target), _interm_reference); // Arithmetic Addition has more strict tolerance
+ validate(Accessor(_target), _reference, tolerance_fp32);
+}
+
+TEST_SUITE_END() // F32
+
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+TEST_SUITE(F16)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEAddMulAddFloatFixture<half>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(),
+ framework::dataset::make("DataType", DataType::F16)),
+ ActivationFunctionsDataset))
+{
+ // Validate outputs
+ validate(Accessor(_interm_target), _interm_reference); // Arithmetic Addition has more strict tolerance
+ validate(Accessor(_target), _reference, tolerance_fp16);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, NEAddMulAddFloatFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(),
+ framework::dataset::make("DataType", DataType::F16)),
+ ActivationFunctionsDataset))
+{
+ // Validate outputs
+ validate(Accessor(_interm_target), _interm_reference); // Arithmetic Addition has more strict tolerance
+ validate(Accessor(_target), _reference, tolerance_fp16);
+}
+TEST_SUITE_END() // F16
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+TEST_SUITE_END() // Float
+
+template <typename T>
+using NEAddMulQuantizedFixture = AddMulAddQuantizedValidationFixture<Tensor, Accessor, NEAddMulAdd, T, true>;
+
+template <typename T>
+using NEAddMulAddQuantizedFixtureWoIntermOut = AddMulAddQuantizedValidationFixture<Tensor, Accessor, NEAddMulAdd, T, false>;
+
+TEST_SUITE(Quantized)
+
+TEST_SUITE(QASYMM8)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEAddMulQuantizedFixture<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(combine(combine(combine(combine(combine(datasets::SmallShapes(),
+ framework::dataset::make("DataType", DataType::QASYMM8)),
+ ActivationFunctionsDataset),
+ qasymm8_input1_qinfo_set),
+ qasymm8_input2_qinfo_set),
+ qasymm8_bn_mul_qinfo_set),
+ qasymm8_bn_add_qinfo_set),
+ qasymm8_add_output_qinfo_set),
+ qasymm8_final_output_qinfo_set))
+{
+ // Validate outputs
+ validate(Accessor(_interm_target), _interm_reference, tolerance_quant);
+ validate(Accessor(_target), _reference, tolerance_quant);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, NEAddMulQuantizedFixture<uint8_t>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(combine(datasets::LargeShapes(),
+ framework::dataset::make("DataType", DataType::QASYMM8)),
+ ActivationFunctionsDataset),
+ qasymm8_input1_qinfo_set),
+ qasymm8_input2_qinfo_set),
+ qasymm8_bn_mul_qinfo_set),
+ qasymm8_bn_add_qinfo_set),
+ qasymm8_add_output_qinfo_set),
+ qasymm8_final_output_qinfo_set))
+{
+ // Validate outputs
+ validate(Accessor(_interm_target), _interm_reference, tolerance_quant);
+ validate(Accessor(_target), _reference, tolerance_quant);
+}
+TEST_SUITE_END() // QASYMM8
+
+TEST_SUITE(QASYMM8_SIGNED)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEAddMulQuantizedFixture<int8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(combine(combine(combine(combine(combine(datasets::SmallShapes(),
+ framework::dataset::make("DataType", DataType::QASYMM8_SIGNED)),
+ ActivationFunctionsDataset),
+ qasymm8_signed_input1_qinfo_set),
+ qasymm8_signed_input2_qinfo_set),
+ qasymm8_signed_bn_mul_qinfo_set),
+ qasymm8_signed_bn_add_qinfo_set),
+ qasymm8_signed_add_output_qinfo_set),
+ qasymm8_signed_final_output_qinfo_set))
+{
+ // Validate outputs
+ validate(Accessor(_interm_target), _interm_reference, tolerance_quant);
+ validate(Accessor(_target), _reference, tolerance_quant);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, NEAddMulQuantizedFixture<int8_t>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(combine(datasets::LargeShapes(),
+ framework::dataset::make("DataType", DataType::QASYMM8_SIGNED)),
+ ActivationFunctionsDataset),
+ qasymm8_signed_input1_qinfo_set),
+ qasymm8_signed_input2_qinfo_set),
+ qasymm8_signed_bn_mul_qinfo_set),
+ qasymm8_signed_bn_add_qinfo_set),
+ qasymm8_signed_add_output_qinfo_set),
+ qasymm8_signed_final_output_qinfo_set))
+{
+ // Validate outputs
+ validate(Accessor(_interm_target), _interm_reference, tolerance_quant);
+ validate(Accessor(_target), _reference, tolerance_quant);
+}
+TEST_SUITE_END() // QASYMM8_SIGNED
+
+TEST_SUITE_END() // Quantized
+
+TEST_SUITE_END() // AddMulAdd
+TEST_SUITE_END() // NEON
+} // namespace validation
+} // namespace test
+} // namespace arm_compute