MLECO-2682: CMake and source refactoring.
MLECO-2930: logging macros were extracted from hal.h and used separately around the code.
MLECO-2931: arm_math lib introduced, cmsis-dsp removed from top level linkage.
MLECO-2915: platform related post-build steps.
Change-Id: Id718884e22f262a5c070ded3f3f5d4b048820147
Signed-off-by: alexander <alexander.efremov@arm.com>
diff --git a/source/math/CMakeLists.txt b/source/math/CMakeLists.txt
new file mode 100644
index 0000000..eab6622
--- /dev/null
+++ b/source/math/CMakeLists.txt
@@ -0,0 +1,44 @@
+#----------------------------------------------------------------------------
+# Copyright (c) 2022 Arm Limited. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#----------------------------------------------------------------------------
+
+#######################################################
+# Math functions for ML pipelines. #
+#######################################################
+
+project(arm_math
+ DESCRIPTION "Collection of the optimised math functions for ML pipelines"
+ LANGUAGES C CXX)
+
+add_library(arm_math STATIC)
+
+target_sources(arm_math
+ PRIVATE
+ PlatformMath.cc)
+
+target_include_directories(arm_math PUBLIC include)
+
+target_link_libraries(arm_math PRIVATE log)
+
+if (${CMAKE_SYSTEM_PROCESSOR} STREQUAL cortex-m55)
+ include(${CMAKE_SCRIPTS_DIR}/cmsis-dsp.cmake)
+ target_link_libraries(arm_math PUBLIC cmsis-dsp)
+endif ()
+
+message(STATUS "*******************************************************")
+message(STATUS "Library : " arm_math)
+message(STATUS "CMAKE_SYSTEM_PROCESSOR : " ${CMAKE_SYSTEM_PROCESSOR})
+message(STATUS "*******************************************************")
diff --git a/source/math/PlatformMath.cc b/source/math/PlatformMath.cc
new file mode 100644
index 0000000..cc603f3
--- /dev/null
+++ b/source/math/PlatformMath.cc
@@ -0,0 +1,308 @@
+/*
+ * Copyright (c) 2021 Arm Limited. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include "PlatformMath.hpp"
+#include "log_macros.h"
+#include <algorithm>
+
+namespace arm {
+namespace app {
+namespace math {
+
+ float MathUtils::CosineF32(float radians)
+ {
+#if ARM_MATH_DSP
+ return arm_cos_f32(radians);
+#else /* ARM_MATH_DSP */
+ return cosf(radians);
+#endif /* ARM_MATH_DSP */
+ }
+
+ float MathUtils::SineF32(float radians)
+ {
+#if ARM_MATH_DSP
+ return arm_sin_f32(radians);
+#else /* ARM_MATH_DSP */
+ return sinf(radians);
+#endif /* ARM_MATH_DSP */
+ }
+
+ float MathUtils::SqrtF32(float input)
+ {
+#if ARM_MATH_DSP
+ float output = 0.f;
+ arm_sqrt_f32(input, &output);
+ return output;
+#else /* ARM_MATH_DSP */
+ return sqrtf(input);
+#endif /* ARM_MATH_DSP */
+ }
+
+ float MathUtils::MeanF32(float* ptrSrc, const uint32_t srcLen)
+ {
+ if (!srcLen) {
+ return 0.f;
+ }
+
+#if ARM_MATH_DSP
+ float result = 0.f;
+ arm_mean_f32(ptrSrc, srcLen, &result);
+ return result;
+#else /* ARM_MATH_DSP */
+ float acc = std::accumulate(ptrSrc, ptrSrc + srcLen, 0.0);
+ return acc/srcLen;
+#endif /* ARM_MATH_DSP */
+ }
+
+ float MathUtils::StdDevF32(float* ptrSrc, const uint32_t srcLen,
+ const float mean)
+ {
+ if (!srcLen) {
+ return 0.f;
+ }
+#if ARM_MATH_DSP
+ /**
+ * Note Standard deviation calculation can be off
+ * by > 0.01 but less than < 0.1, according to
+ * preliminary findings.
+ **/
+ UNUSED(mean);
+ float stdDev = 0;
+ arm_std_f32(ptrSrc, srcLen, &stdDev);
+ return stdDev;
+#else /* ARM_MATH_DSP */
+ auto VarianceFunction = [=](float acc, const float value) {
+ return acc + (((value - mean) * (value - mean))/ srcLen);
+ };
+
+ float acc = std::accumulate(ptrSrc, ptrSrc + srcLen, 0.0,
+ VarianceFunction);
+
+ return sqrtf(acc);
+#endif /* ARM_MATH_DSP */
+ }
+
+ void MathUtils::FftInitF32(const uint16_t fftLen,
+ FftInstance& fftInstance,
+ const FftType type)
+ {
+ fftInstance.m_fftLen = fftLen;
+ fftInstance.m_initialised = false;
+ fftInstance.m_optimisedOptionAvailable = false;
+ fftInstance.m_type = type;
+
+#if ARM_MATH_DSP
+ arm_status status = ARM_MATH_ARGUMENT_ERROR;
+ switch (fftInstance.m_type) {
+ case FftType::real:
+ status = arm_rfft_fast_init_f32(&fftInstance.m_instanceReal, fftLen);
+ break;
+
+ case FftType::complex:
+ status = arm_cfft_init_f32(&fftInstance.m_instanceComplex, fftLen);
+ break;
+
+ default:
+ printf_err("Invalid FFT type\n");
+ return;
+ }
+
+ if (ARM_MATH_SUCCESS != status) {
+ printf_err("Failed to initialise FFT for len %d\n", fftLen);
+ } else {
+ fftInstance.m_optimisedOptionAvailable = true;
+ }
+#endif /* ARM_MATH_DSP */
+
+ debug("Optimised FFT will be used: %s.\n", fftInstance.m_optimisedOptionAvailable? "yes": "no");
+
+ fftInstance.m_initialised = true;
+ }
+
+ static void FftRealF32(std::vector<float>& input,
+ std::vector<float>& fftOutput)
+ {
+ const size_t inputLength = input.size();
+ const size_t halfLength = input.size() / 2;
+
+ fftOutput[0] = 0;
+ fftOutput[1] = 0;
+ for (size_t t = 0; t < inputLength; t++) {
+ fftOutput[0] += input[t];
+ fftOutput[1] += input[t] *
+ MathUtils::CosineF32(2 * M_PI * halfLength * t / inputLength);
+ }
+
+ for (size_t k = 1, j = 2; k < halfLength; ++k, j += 2) {
+ float sumReal = 0;
+ float sumImag = 0;
+
+ const auto theta = static_cast<float>(2 * M_PI * k / inputLength);
+
+ for (size_t t = 0; t < inputLength; t++) {
+ const auto angle = static_cast<float>(t * theta);
+ sumReal += input[t] * MathUtils::CosineF32(angle);
+ sumImag += -input[t]* MathUtils::SineF32(angle);
+ }
+
+ /* Arrange output to [real0, realN/2, real1, im1, real2, im2, ...] */
+ fftOutput[j] = sumReal;
+ fftOutput[j + 1] = sumImag;
+ }
+ }
+
+ static void FftComplexF32(std::vector<float>& input,
+ std::vector<float>& fftOutput)
+ {
+ const size_t fftLen = input.size() / 2;
+ for (size_t k = 0; k < fftLen; k++) {
+ float sumReal = 0;
+ float sumImag = 0;
+ const auto theta = static_cast<float>(2 * M_PI * k / fftLen);
+ for (size_t t = 0; t < fftLen; t++) {
+ const auto angle = theta * t;
+ const auto cosine = MathUtils::CosineF32(angle);
+ const auto sine = MathUtils::SineF32(angle);
+ sumReal += input[t*2] * cosine + input[t*2 + 1] * sine;
+ sumImag += -input[t*2] * sine + input[t*2 + 1] * cosine;
+ }
+ fftOutput[k*2] = sumReal;
+ fftOutput[k*2 + 1] = sumImag;
+ }
+ }
+
+ void MathUtils::FftF32(std::vector<float>& input,
+ std::vector<float>& fftOutput,
+ arm::app::math::FftInstance& fftInstance)
+ {
+ if (!fftInstance.m_initialised) {
+ printf_err("FFT uninitialised\n");
+ return;
+ } else if (input.size() < fftInstance.m_fftLen) {
+ printf_err("FFT len: %" PRIu16 "; input len: %zu\n",
+ fftInstance.m_fftLen, input.size());
+ return;
+ } else if (fftOutput.size() < input.size()) {
+ printf_err("Output vector len insufficient to hold FFTs\n");
+ return;
+ }
+
+ switch (fftInstance.m_type) {
+ case FftType::real:
+
+#if ARM_MATH_DSP
+ if (fftInstance.m_optimisedOptionAvailable) {
+ arm_rfft_fast_f32(&fftInstance.m_instanceReal, input.data(), fftOutput.data(), 0);
+ return;
+ }
+#endif /* ARM_MATH_DSP */
+ FftRealF32(input, fftOutput);
+ return;
+
+ case FftType::complex:
+ if (input.size() < fftInstance.m_fftLen * 2) {
+ printf_err("Complex FFT instance should have input size >= (FFT len x 2)");
+ return;
+ }
+#if ARM_MATH_DSP
+ if (fftInstance.m_optimisedOptionAvailable) {
+ fftOutput = input; /* Complex function works in-place */
+ arm_cfft_f32(&fftInstance.m_instanceComplex, fftOutput.data(), 0, 1);
+ return;
+ }
+#endif /* ARM_MATH_DSP */
+ FftComplexF32(input, fftOutput);
+ return;
+
+ default:
+ printf_err("Invalid FFT type\n");
+ return;
+ }
+ }
+
+ void MathUtils::VecLogarithmF32(std::vector <float>& input,
+ std::vector <float>& output)
+ {
+#if ARM_MATH_DSP
+ arm_vlog_f32(input.data(), output.data(),
+ output.size());
+#else /* ARM_MATH_DSP */
+ for (auto in = input.begin(), out = output.begin();
+ in != input.end() && out != output.end(); ++in, ++out) {
+ *out = logf(*in);
+ }
+#endif /* ARM_MATH_DSP */
+ }
+
+ float MathUtils::DotProductF32(float* srcPtrA, float* srcPtrB,
+ const uint32_t srcLen)
+ {
+ float output = 0.f;
+
+#if ARM_MATH_DSP
+ arm_dot_prod_f32(srcPtrA, srcPtrB, srcLen, &output);
+#else /* ARM_MATH_DSP */
+ for (uint32_t i = 0; i < srcLen; ++i) {
+ output += *srcPtrA++ * *srcPtrB++;
+ }
+#endif /* ARM_MATH_DSP */
+
+ return output;
+ }
+
+ bool MathUtils::ComplexMagnitudeSquaredF32(float* ptrSrc,
+ const uint32_t srcLen,
+ float* ptrDst,
+ const uint32_t dstLen)
+ {
+ if (dstLen < srcLen/2) {
+ printf_err("dstLen must be greater than srcLen/2");
+ return false;
+ }
+
+#if ARM_MATH_DSP
+ arm_cmplx_mag_squared_f32(ptrSrc, ptrDst, srcLen/2);
+#else /* ARM_MATH_DSP */
+ for (uint32_t j = 0; j < srcLen/2; ++j) {
+ const float real = *ptrSrc++;
+ const float im = *ptrSrc++;
+ *ptrDst++ = real*real + im*im;
+ }
+#endif /* ARM_MATH_DSP */
+ return true;
+ }
+
+ void MathUtils::SoftmaxF32(std::vector<float>& vec)
+ {
+ /* Fix for numerical stability and apply exp. */
+ auto start = vec.begin();
+ auto end = vec.end();
+
+ float maxValue = *std::max_element(start, end);
+ for (auto it = start; it != end; ++it) {
+ *it = std::exp((*it) - maxValue);
+ }
+
+ float sumExp = std::accumulate(start, end, 0.0f);
+
+ for (auto it = start; it != end; ++it) {
+ *it = (*it)/sumExp;
+ }
+ }
+
+} /* namespace math */
+} /* namespace app */
+} /* namespace arm */
diff --git a/source/math/include/PlatformMath.hpp b/source/math/include/PlatformMath.hpp
new file mode 100644
index 0000000..5ac10de
--- /dev/null
+++ b/source/math/include/PlatformMath.hpp
@@ -0,0 +1,174 @@
+/*
+ * Copyright (c) 2021 Arm Limited. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef PLATFORM_MATH_HPP
+#define PLATFORM_MATH_HPP
+
+/* See if ARM DSP functions can be used. */
+#if defined(ARM_MATH_DSP)
+
+ #include "arm_math.h"
+ #define M_PI (PI)
+#else
+ #include <cmath>
+#endif
+
+#include <vector>
+#include <cstdint>
+#include <numeric>
+
+namespace arm {
+namespace app {
+namespace math {
+
+ enum class FftType {
+ real = 0,
+ complex = 1
+ };
+
+ struct FftInstance {
+#if ARM_MATH_DSP
+ arm_rfft_fast_instance_f32 m_instanceReal;
+ arm_cfft_instance_f32 m_instanceComplex;
+#endif
+ uint16_t m_fftLen{0};
+ FftType m_type{FftType::real};
+ bool m_optimisedOptionAvailable{false};
+ bool m_initialised{false};
+ };
+
+ /* Class to provide Math functions like FFT, mean, stddev etc.
+ * This will allow other classes, functions to be independent of
+ * #if definition checks and provide a cleaner API. Also, it will
+ * consolidate all arm math functions used in one place and make
+ * them easier to test. */
+ class MathUtils {
+
+ public:
+ /**
+ * @brief Get the cosine value of the argument in floating point.
+ * @param[in] radians Angle in radians.
+ * @return Cosine value (floating point).
+ */
+ static float CosineF32(float radians);
+
+ /**
+ * @brief Get the sine value of the argument in floating point.
+ * @param[in] radians Angle in radians.
+ * @return Sine value (floating point).
+ */
+ static float SineF32(float radians);
+
+ /**
+ * @brief Get the square root of the argument in floating point.
+ * @param[in] input Value to compute square root of.
+ * @return Square root (floating point) value.
+ */
+ static float SqrtF32(float input);
+
+ /**
+ * @brief Gets the mean of a floating point array of elements.
+ * @param[in] ptrSrc Pointer to the first element.
+ * @param[in] srcLen Number of elements in the array/vector.
+ * @return Average value.
+ */
+ static float MeanF32(float* ptrSrc, uint32_t srcLen);
+
+ /**
+ * @brief Gets the standard deviation of a floating point array
+ * of elements.
+ * @param[in] ptrSrc Pointer to the first element.
+ * @param[in] srcLen Number of elements in the array/vector.
+ * @param[in] mean Pre-computed mean value.
+ * @return Standard deviation value.
+ */
+ static float StdDevF32(float* ptrSrc, uint32_t srcLen,
+ float mean);
+
+ /**
+ * @brief Initialises the internal FFT structures (if available
+ * for the platform). This function should be called
+ * prior to Fft32 function call if built with ARM DSP functions.
+ * @param[in] fftLen Requested length of the FFT.
+ * @param[in] fftInstance FFT instance struct to use.
+ * @param[in] type FFT type (real or complex)
+ */
+ static void FftInitF32(uint16_t fftLen,
+ FftInstance& fftInstance,
+ FftType type = FftType::real);
+
+ /**
+ * @brief Computes the FFT for the input vector.
+ * @param[in] input Floating point vector of input elements
+ * @param[out] fftOutput Output buffer to be populated by computed FFTs.
+ * @param[in] fftInstance FFT instance struct to use.
+ */
+ static void FftF32(std::vector<float>& input,
+ std::vector<float>& fftOutput,
+ FftInstance& fftInstance);
+
+ /**
+ * @brief Computes the natural logarithms of input floating point
+ * vector
+ * @param[in] input Floating point input vector
+ * @param[out] output Pre-allocated buffer to be populated with
+ * natural log values of each input element.
+ */
+ static void VecLogarithmF32(std::vector <float>& input,
+ std::vector <float>& output);
+
+ /**
+ * @brief Computes the dot product of two 1D floating point
+ * vectors.
+ * result = sum(srcA[0]*srcB[0] + srcA[1]*srcB[1] + ..)
+ * @param[in] srcPtrA Pointer to the first element of first
+ * array.
+ * @param[in] srcPtrB Pointer to the first element of second
+ * array.
+ * @param[in] srcLen Number of elements in the array/vector.
+ * @return Dot product.
+ */
+ static float DotProductF32(float* srcPtrA, float* srcPtrB,
+ uint32_t srcLen);
+
+ /**
+ * @brief Computes the squared magnitude of floating point
+ * complex number array.
+ * @param[in] ptrSrc Pointer to the first element of input
+ * array.
+ * @param[in] srcLen Number of elements in the array/vector.
+ * @param[out] ptrDst Output buffer to be populated.
+ * @param[in] dstLen Output buffer len (for sanity check only).
+ * @return true if successful, false otherwise.
+ */
+ static bool ComplexMagnitudeSquaredF32(float* ptrSrc,
+ uint32_t srcLen,
+ float* ptrDst,
+ uint32_t dstLen);
+
+ /**
+ * @brief Scales output scores for an arbitrary number of classes so
+ * that they sum to 1, allowing output to be expressed as a probability.
+ * @param[in] vector Vector of floats modified in-place
+ */
+ static void SoftmaxF32(std::vector<float>& vec);
+ };
+
+} /* namespace math */
+} /* namespace app */
+} /* namespace arm */
+
+#endif /* PLATFORM_MATH_HPP */
\ No newline at end of file