COMPMID-1961: Implement FFT (1D) on NEON

Change-Id: I0bea3bfbc3b0cd9e8c9a0e0f6f430640573f08d1
Signed-off-by: giuros01 <giuseppe.rossini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/996
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

5 files changed, 263 insertions, 0 deletions

diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h
index 1ce3821e81..b8ae467c6d 100644
--- a/arm_compute/core/NEON/NEKernels.h
+++ b/arm_compute/core/NEON/NEKernels.h
@@ -62,6 +62,8 @@
 #include "arm_compute/core/NEON/kernels/NEElementwiseOperationKernel.h"
 #include "arm_compute/core/NEON/kernels/NEElementwiseUnaryKernel.h"
 #include "arm_compute/core/NEON/kernels/NEErodeKernel.h"
+#include "arm_compute/core/NEON/kernels/NEFFTDigitReverseKernel.h"
+#include "arm_compute/core/NEON/kernels/NEFFTRadixStageKernel.h"
 #include "arm_compute/core/NEON/kernels/NEFastCornersKernel.h"
 #include "arm_compute/core/NEON/kernels/NEFillArrayKernel.h"
 #include "arm_compute/core/NEON/kernels/NEFillBorderKernel.h"
diff --git a/arm_compute/core/NEON/kernels/NEFFTDigitReverseKernel.h b/arm_compute/core/NEON/kernels/NEFFTDigitReverseKernel.h
new file mode 100644
index 0000000000..84d55fd8f4
--- /dev/null
+++ b/arm_compute/core/NEON/kernels/NEFFTDigitReverseKernel.h
@@ -0,0 +1,83 @@
+/*
+ * Copyright (c) 2019 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_NEFFTDIGITREVERSEKERNEL_H__
+#define __ARM_COMPUTE_NEFFTDIGITREVERSEKERNEL_H__
+
+#include "arm_compute/core/NEON/INEKernel.h"
+
+namespace arm_compute
+{
+// Forward declarations
+class ITensor;
+
+/** Interface for the digit reverse operation kernel. */
+class NEFFTDigitReverseKernel : public INEKernel
+{
+public:
+    const char *name() const override
+    {
+        return "NEFFTDigitReverseKernel";
+    }
+    /** Constructor */
+    NEFFTDigitReverseKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEFFTDigitReverseKernel(const NEFFTDigitReverseKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEFFTDigitReverseKernel &operator=(const NEFFTDigitReverseKernel &) = delete;
+    /** Default Move Constructor. */
+    NEFFTDigitReverseKernel(NEFFTDigitReverseKernel &&) = default;
+    /** Default move assignment operator */
+    NEFFTDigitReverseKernel &operator=(NEFFTDigitReverseKernel &&) = default;
+    /** Default destructor */
+    ~NEFFTDigitReverseKernel() = default;
+    /** Set the input and output tensors.
+     *
+     * @param[in]  input  Source tensor. Data types supported: F32.
+     * @param[out] output Destination tensor. Data type supported: same as @p input
+     * @param[in]  idx    Digit reverse index tensor. Data type supported: U32
+     * @param[in]  axis   Axis to perform digit reverse on.
+     */
+    void configure(const ITensor *input, ITensor *output, const ITensor *idx, unsigned int axis);
+    /** Static function to check if given info will lead to a valid configuration of @ref NEFFTDigitReverseKernel
+     *
+     * @param[in] input  Source tensor info. Data types supported: F32.
+     * @param[in] output Destination tensor info. Data type supported: same as @p input
+     * @param[in] idx    Digit reverse index tensor info. Data type supported: U32
+     * @param[in] axis   Axis to perform digit reverse on.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *idx, unsigned int axis);
+
+    // Inherited methods overridden:
+    void run(const Window &window, const ThreadInfo &info) override;
+
+private:
+    const ITensor *_input;
+    ITensor       *_output;
+    const ITensor *_idx;
+    unsigned int   _axis;
+};
+} // namespace arm_compute
+#endif /*__ARM_COMPUTE_NEFFTDIGITREVERSEKERNEL_H__ */
diff --git a/arm_compute/core/NEON/kernels/NEFFTRadixStageKernel.h b/arm_compute/core/NEON/kernels/NEFFTRadixStageKernel.h
new file mode 100644
index 0000000000..a4c4be6f35
--- /dev/null
+++ b/arm_compute/core/NEON/kernels/NEFFTRadixStageKernel.h
@@ -0,0 +1,98 @@
+/*
+ * Copyright (c) 2019 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_NEFFTRADIXSTAGEKERNEL_H__
+#define __ARM_COMPUTE_NEFFTRADIXSTAGEKERNEL_H__
+
+#include "arm_compute/core/NEON/INEKernel.h"
+
+#include "arm_compute/core/KernelDescriptors.h"
+
+#include <set>
+
+namespace arm_compute
+{
+// Forward declarations
+class ITensor;
+
+/** Interface for the FFT kernel. */
+class NEFFTRadixStageKernel : public INEKernel
+{
+public:
+    const char *name() const override
+    {
+        return "NEFFTRadixStageKernel";
+    }
+    /** Constructor */
+    NEFFTRadixStageKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEFFTRadixStageKernel(const NEFFTRadixStageKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEFFTRadixStageKernel &operator=(const NEFFTRadixStageKernel &) = delete;
+    /** Default Move Constructor. */
+    NEFFTRadixStageKernel(NEFFTRadixStageKernel &&) = default;
+    /** Default move assignment operator */
+    NEFFTRadixStageKernel &operator=(NEFFTRadixStageKernel &&) = default;
+    /** Default destructor */
+    ~NEFFTRadixStageKernel() = default;
+    /** Set the input and output tensors.
+     *
+     * @note If the output tensor is nullptr, the FFT will be performed in-place
+     *
+     * @param[in,out] input  Source tensor. Data types supported: F32.
+     * @param[out]    output Destination tensor. Data type supported: same as @p input
+     * @param[in]     config FFT descriptor metadata.
+     */
+    void configure(ITensor *input, ITensor *output, const FFTRadixStageKernelInfo &config);
+    /** Static function to check if given info will lead to a valid configuration of @ref NEFFTRadixStageKernel
+     *
+     * @param[in] input  Source tensor info. Data types supported: F32.
+     * @param[in] output Destination tensor info. Data type supported: same as @p input
+     * @param[in] config FFT descriptor metadata.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const FFTRadixStageKernelInfo &config);
+    /** Returns the radix that are support by the FFT kernel
+     *
+     * @return A set of supported radix
+     */
+    static std::set<unsigned int> supported_radix();
+
+    // Inherited methods overridden:
+    void run(const Window &window, const ThreadInfo &info) override;
+
+private:
+    ITensor     *_input;
+    ITensor     *_output;
+    bool         _run_in_place;
+    unsigned int _Nx;
+
+    template <bool first_stage>
+    void set_radix_stage_fun(unsigned int radix);
+
+    using FFTFunctionPointerInPlace = std::function<void(float *, float *, unsigned int, unsigned int)>;
+    FFTFunctionPointerInPlace _func;
+};
+} // namespace arm_compute
+#endif /*__ARM_COMPUTE_NEFFTKERNEL_H__ */
diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h
index 432c751308..d8f54ea231 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -63,6 +63,7 @@
 #include "arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEEqualizeHistogram.h"
 #include "arm_compute/runtime/NEON/functions/NEErode.h"
+#include "arm_compute/runtime/NEON/functions/NEFFT1D.h"
 #include "arm_compute/runtime/NEON/functions/NEFastCorners.h"
 #include "arm_compute/runtime/NEON/functions/NEFillBorder.h"
 #include "arm_compute/runtime/NEON/functions/NEFlattenLayer.h"
diff --git a/arm_compute/runtime/NEON/functions/NEFFT1D.h b/arm_compute/runtime/NEON/functions/NEFFT1D.h
new file mode 100644
index 0000000000..9b5ada746a
--- /dev/null
+++ b/arm_compute/runtime/NEON/functions/NEFFT1D.h
@@ -0,0 +1,79 @@
+/*
+ * Copyright (c) 2019 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_NEFFT1D_H__
+#define __ARM_COMPUTE_NEFFT1D_H__
+
+#include "arm_compute/core/NEON/kernels/NEFFTDigitReverseKernel.h"
+#include "arm_compute/core/NEON/kernels/NEFFTRadixStageKernel.h"
+#include "arm_compute/runtime/IFunction.h"
+
+#include "arm_compute/runtime/FunctionDescriptors.h"
+#include "arm_compute/runtime/MemoryGroup.h"
+#include "arm_compute/runtime/Tensor.h"
+
+namespace arm_compute
+{
+// Forward declaration
+class ITensor;
+
+/** Basic function to execute one dimensional FFT. This function calls the following OpenCL kernels:
+ *
+ * -# @ref CLFFTDigitReverseKernel Performs digit reverse
+ * -# @ref NEFFTRadixStageKernel   A list of FFT kernels depending on the radix decomposition
+ */
+class NEFFT1D : public IFunction
+{
+public:
+    /** Default Constructor */
+    NEFFT1D(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Initialise the function's source, destinations and border mode.
+     *
+     * @param[in]  input  Source tensor. Data types supported: F32.
+     * @param[out] output Destination tensor. Data types and data layouts supported: Same as @p input.
+     * @param[in]  config FFT related configuration
+     */
+    void configure(const ITensor *input, ITensor *output, const FFT1DInfo &config);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLFFT1D.
+     *
+     * @param[in] input  Source tensor info. Data types supported: F32.
+     * @param[in] output Destination tensor info. Data types and data layouts supported: Same as @p input.
+     * @param[in] config FFT related configuration
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const FFT1DInfo &config);
+
+    // Inherited methods overridden:
+    void run() override;
+
+protected:
+    MemoryGroup                        _memory_group;
+    Tensor                             _digit_reversed_input;
+    Tensor                             _digit_reverse_indices;
+    NEFFTDigitReverseKernel            _digit_reverse_kernel;
+    std::vector<NEFFTRadixStageKernel> _fft_kernels;
+    unsigned int                       _n_ffts;
+};
+} // namespace arm_compute
+#endif /*__ARM_COMPUTE_NEFFT1D_H__ */