aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/assembly
diff options
context:
space:
mode:
authorSang-Hoon Park <sang-hoon.park@arm.com>2020-10-19 16:00:11 +0100
committerGeorgios Pinitas <georgios.pinitas@arm.com>2020-10-20 10:27:40 +0000
commit68dd25fbe6e4d3c3513fa5993863419769aa08fc (patch)
treeb918be923f9e4550c306d7f44d168ab938a71fc8 /src/core/NEON/kernels/assembly
parentf0a4e609d98f111b6a7d4a2b578d1b7cba64b805 (diff)
downloadComputeLibrary-68dd25fbe6e4d3c3513fa5993863419769aa08fc.tar.gz
COMPMID-3637: Move utility headers from arm_compute to src
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com> Change-Id: If9d6fa8c900b68c4b6fd373f2fc1f9abb83ea917 Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4145 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/assembly')
-rw-r--r--src/core/NEON/kernels/assembly/INEGEMMWrapperKernel.cpp4
-rw-r--r--src/core/NEON/kernels/assembly/INEGEMMWrapperKernel.h108
-rw-r--r--src/core/NEON/kernels/assembly/NEDepthwiseConvolutionAssemblyKernelWrapper.h88
-rw-r--r--src/core/NEON/kernels/assembly/arm_gemm_local.hpp34
4 files changed, 232 insertions, 2 deletions
diff --git a/src/core/NEON/kernels/assembly/INEGEMMWrapperKernel.cpp b/src/core/NEON/kernels/assembly/INEGEMMWrapperKernel.cpp
index b071be3749..760274dba1 100644
--- a/src/core/NEON/kernels/assembly/INEGEMMWrapperKernel.cpp
+++ b/src/core/NEON/kernels/assembly/INEGEMMWrapperKernel.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018-2019 Arm Limited.
+ * Copyright (c) 2018-2020 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -22,7 +22,7 @@
* SOFTWARE.
*/
-#include "arm_compute/core/NEON/kernels/assembly/INEGEMMWrapperKernel.h"
+#include "src/core/NEON/kernels/assembly/INEGEMMWrapperKernel.h"
#include "arm_compute/core/Error.h"
#include "arm_compute/core/Helpers.h"
diff --git a/src/core/NEON/kernels/assembly/INEGEMMWrapperKernel.h b/src/core/NEON/kernels/assembly/INEGEMMWrapperKernel.h
new file mode 100644
index 0000000000..030f1aad12
--- /dev/null
+++ b/src/core/NEON/kernels/assembly/INEGEMMWrapperKernel.h
@@ -0,0 +1,108 @@
+/*
+ * Copyright (c) 2018-2020 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 SRC_INEGEMMWRAPPERKERNEL_H
+#define SRC_INEGEMMWRAPPERKERNEL_H
+
+#include "arm_compute/core/NEON/INEKernel.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Common interface for all the arm_gemm Gemms
+ */
+class INEGEMMWrapperKernel : public INEKernel
+{
+public:
+ /** Parameters defining the dimensions of the matrices being multiplied */
+ struct Params
+ {
+ unsigned int M{ 0 }; /**< Rows in output matrix C (and input matrix A). */
+ unsigned int N{ 0 }; /**< Columns in output matrix C (and input matrix B). */
+ unsigned int K{ 0 }; /**< Columns of input matrix A (= rows of input matrix B). */
+ unsigned int batches{ 0 }; /**< Number of "batched" GEMMs (unique A and C, shared B). */
+ unsigned int multis{ 0 }; /**< Number of "multi" GEMMs (unique A, B and C). */
+ };
+
+ static Params extract_parameters(const ITensor *a, const ITensor *b, const ITensor *c, const GEMMInfo &gemm_info);
+
+ /** Constructor */
+ INEGEMMWrapperKernel();
+ /** Prevent instances of this class from being copied */
+ INEGEMMWrapperKernel(const INEGEMMWrapperKernel &) = delete;
+ /** Prevent instances of this class from being copied */
+ INEGEMMWrapperKernel &operator=(const INEGEMMWrapperKernel &) = delete;
+ /** Allow instances of this class to be moved */
+ INEGEMMWrapperKernel(INEGEMMWrapperKernel &&) = default;
+ /** Allow instances of this class to be moved */
+ INEGEMMWrapperKernel &operator=(INEGEMMWrapperKernel &&) = default;
+ /** Initialise the kernel's input and output.
+ *
+ * @note The input and output tensor must have the same dimensions
+ *
+ * @param[in] a Input tensor (Matrix A)
+ * @param[in] b Input tensor (Matrix B)
+ * @param[out] c Output tensor to store the result of matrix multiplication. Data type supported: same as @p input0.
+ * @param[in] alpha Scalar multiplier to apply to AB matrix product.
+ * @param[in] beta Scalar multiplier to apply to input C matrix before adding product.
+ * @param[in] gemm_info GEMM meta-data
+ */
+ void configure(const ITensor *a, const ITensor *b, ITensor *c, float alpha, float beta, const GEMMInfo &gemm_info);
+
+ // Inherited methods overridden:
+ void run(const Window &window, const ThreadInfo &info) override;
+
+protected:
+ /** Called as part of configure() after _a, _b, _c and _params have been set.
+ *
+ * @param[in] alpha Scalar multiplier to apply to AB matrix product.
+ * @param[in] beta Scalar multiplier to apply to input C matrix before adding product.
+ *
+ * @return A 3D execution window.
+ */
+ virtual Window configure_internal(float alpha, float beta) = 0;
+
+ /** Run the kernel from the start to the end offset in window.
+ *
+ * @param[in] window Window to use for the iteration
+ * @param[in] start_offset Where to start iterating from (In Window coordinates)
+ * @param[in] end_offset Where to stop iterating (In Window coordinates).
+ * @param[in] info Info about executing thread and CPU.
+ */
+ virtual void run_internal(const Window &window, const Coordinates &start_offset, const Coordinates &end_offset, const ThreadInfo &info) = 0;
+
+ const ITensor *_a;
+ const ITensor *_b;
+ ITensor *_c;
+ Params _params;
+ GEMMInfo _gemm_info;
+
+private:
+ Window _window3d;
+ TensorShape _window_shape;
+};
+
+} // namespace arm_compute
+
+#endif /* SRC_INEGEMMRAPPERKERNEL_H */
diff --git a/src/core/NEON/kernels/assembly/NEDepthwiseConvolutionAssemblyKernelWrapper.h b/src/core/NEON/kernels/assembly/NEDepthwiseConvolutionAssemblyKernelWrapper.h
new file mode 100644
index 0000000000..a2f7e3bd59
--- /dev/null
+++ b/src/core/NEON/kernels/assembly/NEDepthwiseConvolutionAssemblyKernelWrapper.h
@@ -0,0 +1,88 @@
+/*
+ * Copyright (c) 2019-2020 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 SRC_ASSEMBLY_DEPTHWISE_CONVOLUTION_ASSEMBLY_WRAPPER_KERNEL_H
+#define SRC_ASSEMBLY_DEPTHWISE_CONVOLUTION_ASSEMBLY_WRAPPER_KERNEL_H
+
+#include "arm_compute/core/NEON/INEKernel.h"
+#include "arm_compute/core/Utils.h"
+#include "arm_compute/core/Validate.h"
+
+#include "src/core/NEON/kernels/convolution/depthwise/depthwise.hpp"
+
+namespace arm_compute
+{
+// Forward declarations
+class ITensor;
+
+/** This class is a wrapper for the depthwise convolution assembly kernels. */
+class NEDepthwiseConvolutionAssemblyKernelWrapper final : public INEKernel
+{
+public:
+ const char *name() const override
+ {
+ return "NEDepthwiseConvolutionAssemblyKernelWrapper";
+ }
+
+ /** Default constructor */
+ NEDepthwiseConvolutionAssemblyKernelWrapper()
+ : _kernel(nullptr)
+ {
+ }
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEDepthwiseConvolutionAssemblyKernelWrapper(const NEDepthwiseConvolutionAssemblyKernelWrapper &) = delete;
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEDepthwiseConvolutionAssemblyKernelWrapper &operator=(const NEDepthwiseConvolutionAssemblyKernelWrapper &) = delete;
+ /** Default Move Constructor. */
+ NEDepthwiseConvolutionAssemblyKernelWrapper(NEDepthwiseConvolutionAssemblyKernelWrapper &&) = default;
+ /** Default move assignment operator */
+ NEDepthwiseConvolutionAssemblyKernelWrapper &operator=(NEDepthwiseConvolutionAssemblyKernelWrapper &&) = default;
+
+ /** Initialise the kernel's input and output.
+ *
+ * @param[in] kernel Pointer to an assembly kernel implementation.
+ */
+ void configure(depthwise::IDepthwiseConvolution *kernel)
+ {
+ ARM_COMPUTE_ERROR_ON_NULLPTR((reinterpret_cast<void *>(kernel)));
+ _kernel = kernel;
+ Window win;
+ win.set(Window::DimX, Window::Dimension(0, _kernel->get_window(), 1));
+ INEKernel::configure(win);
+ }
+
+ // Inherited methods overridden:
+ void run(const Window &window, const ThreadInfo &info) override
+ {
+ ARM_COMPUTE_ERROR_ON_NULLPTR((reinterpret_cast<void *>(_kernel)));
+ ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+ auto first = window.x().start();
+ auto last = window.x().end();
+ _kernel->run(first, last, info.thread_id);
+ }
+
+private:
+ depthwise::IDepthwiseConvolution *_kernel;
+};
+} // namespace arm_compute
+#endif /* SRC_ASSEMBLY_DEPTHWISE_CONVOLUTION_ASSEMBLY_WRAPPER_KERNEL_H */
diff --git a/src/core/NEON/kernels/assembly/arm_gemm_local.hpp b/src/core/NEON/kernels/assembly/arm_gemm_local.hpp
new file mode 100644
index 0000000000..4715f2500a
--- /dev/null
+++ b/src/core/NEON/kernels/assembly/arm_gemm_local.hpp
@@ -0,0 +1,34 @@
+/*
+ * Copyright (c) 2018-2020 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.
+ */
+#pragma once
+
+/* This file is used to configure integration-specific aspects of arm_gemm into ACL */
+
+#include "arm_compute/core/CPP/CPPTypes.h"
+
+namespace arm_gemm
+{
+using CPUModel = arm_compute::CPUModel;
+using CPUInfo = arm_compute::CPUInfo;
+} // namespace arm_compute