sergiotarxz/nuttx-apps
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
5dd4813b34
nuttx-apps/mlearning/cmsis/cmsis-nn-support_nnabla.patch
Alin Jerpelea 02a50ccb1f mlearning: CMSIS NN: add support for the CHW tensor layout 
- support float version of convolution
 - support the CHW tensor layout

following function prototypes are added:
 - arm_convolve_CHW_f32_basic_nonsquare()
 - arm_convolve_CHW_q15_basic_nonsquare()
 - arm_convolve_CHW_q7_basic_nonsquare()
 - arm_nn_CHW_mat_mult_kernel_q7_q15()

NOTE:this patch will be contributed to SMSIS and reverted later from NuttX
Signed-off-by: Alin Jerpelea <alin.jerpelea@sony.com>
2022-02-19 00:54:49 +08:00

1108 lines
46 KiB
Diff
Raw Blame History

diff --git a/CMSIS_5/CMSIS/NN/Include/arm_nnfunctions_nnabla.h CMSIS_5/CMSIS/NN/Include/arm_nnfunctions_nnabla.h
===CHANGE_NOTICE(1/5)===========================================================
Sony Corporation added this file to 5.4.0
to add the following function prototypes:
- arm_convolve_CHW_f32_basic_nonsquare()
- arm_convolve_CHW_q15_basic_nonsquare()
- arm_convolve_CHW_q7_basic_nonsquare()
- arm_nn_CHW_mat_mult_kernel_q7_q15()
================================================================================
--- /dev/null
+++ CMSIS_5/CMSIS/NN/Include/arm_nnfunctions_nnabla.h
@@ -0,0 +1,217 @@
+/*
+ * Copyright (C) 2010-2018 Arm Limited or its affiliates. All rights reserved.
+ * Copyright 2018 Sony Corporation
+ *
+ * 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
+ *
+ * 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.
+ */
+
+/* ----------------------------------------------------------------------
+ * Title: arm_nnfunctions_nnabla.h
+ * Author: Sony Corporation
+ * Description: Sony Corporation added this file to 5.4.0
+ * to add the following function prototypes:
+ * - arm_convolve_CHW_f32_basic_nonsquare()
+ * - arm_convolve_CHW_q15_basic_nonsquare()
+ * - arm_convolve_CHW_q7_basic_nonsquare()
+ * - arm_nn_CHW_mat_mult_kernel_q7_q15()
+ * $Date: 14. September 2018
+ * -------------------------------------------------------------------- */
+
+#ifndef _ARM_NNFUNCTIONS_CHW_H
+#define _ARM_NNFUNCTIONS_CHW_H
+
+#include "arm_nnsupportfunctions.h"
+#include "arm_nn_tables.h"
+
+#define USE_INTRINSIC
+
+//#define ARM_NN_TRUNCATE /* This config the rounding model to floor or round to the nearest int */
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+
+ /**
+ * @brief Basic float32 convolution function (non-sqaure shape)
+ * @param[in] Im_in pointer to input tensor
+ * @param[in] dim_im_in_x input tensor dimention x
+ * @param[in] dim_im_in_y input tensor dimention y
+ * @param[in] ch_im_in number of input tensor channels
+ * @param[in] wt pointer to kernel weights
+ * @param[in] ch_im_out number of filters, i.e., output tensor channels
+ * @param[in] dim_kernel_x filter kernel size x
+ * @param[in] dim_kernel_y filter kernel size y
+ * @param[in] padding_x padding size x
+ * @param[in] padding_y padding size y
+ * @param[in] stride_x convolution stride x
+ * @param[in] stride_y convolution stride y
+ * @param[in] bias pointer to bias
+ * @param[in,out] Im_out pointer to output tensor
+ * @param[in] dim_im_out_x output tensor dimension x
+ * @param[in] dim_im_out_y output tensor dimension y
+ * @param[in,out] bufferA pointer to buffer space for input
+ * @param[in,out] bufferB pointer to buffer space for output
+ * @return The function returns <code>ARM_MATH_SUCCESS</code>
+ */
+
+ arm_status
+ arm_convolve_CHW_f32_basic_nonsquare(const float * Im_in,
+ const uint16_t dim_im_in_x,
+ const uint16_t dim_im_in_y,
+ const uint16_t ch_im_in,
+ const float * wt,
+ const uint16_t ch_im_out,
+ const uint16_t dim_kernel_x,
+ const uint16_t dim_kernel_y,
+ const uint16_t padding_x,
+ const uint16_t padding_y,
+ const uint16_t stride_x,
+ const uint16_t stride_y,
+ const float * bias,
+ float * Im_out,
+ const uint16_t dim_im_out_x,
+ const uint16_t dim_im_out_y,
+ float * bufferA,
+ float * bufferB);
+
+
+ /**
+ * @brief Basic Q15 version of CHW convolution (non-sqaure shape)
+ * @param[in] Im_in pointer to input tensor
+ * @param[in] dim_im_in_x input tensor dimention x
+ * @param[in] dim_im_in_y input tensor dimention y
+ * @param[in] ch_im_in number of input tensor channels
+ * @param[in] wt pointer to kernel weights
+ * @param[in] ch_im_out number of filters, i.e., output tensor channels
+ * @param[in] dim_kernel_x filter kernel size x
+ * @param[in] dim_kernel_y filter kernel size y
+ * @param[in] padding_x padding sizes x
+ * @param[in] padding_y padding sizes y
+ * @param[in] stride_x convolution stride x
+ * @param[in] stride_y convolution stride y
+ * @param[in] bias pointer to bias
+ * @param[in] bias_shift amount of left-shift for bias
+ * @param[in] out_shift amount of right-shift for output
+ * @param[in,out] Im_out pointer to output tensor
+ * @param[in] dim_im_out_x output tensor dimension x
+ * @param[in] dim_im_out_y output tensor dimension y
+ * @param[in,out] bufferA pointer to buffer space for input
+ * @param[in,out] bufferB pointer to buffer space for output
+ * @return The function returns <code>ARM_MATH_SUCCESS</code>
+ */
+
+ arm_status
+ arm_convolve_CHW_q15_basic_nonsquare(const q15_t * Im_in,
+ const uint16_t dim_im_in_x,
+ const uint16_t dim_im_in_y,
+ const uint16_t ch_im_in,
+ const q15_t * wt,
+ const uint16_t ch_im_out,
+ const uint16_t dim_kernel_x,
+ const uint16_t dim_kernel_y,
+ const uint16_t padding_x,
+ const uint16_t padding_y,
+ const uint16_t stride_x,
+ const uint16_t stride_y,
+ const q15_t * bias,
+ const uint16_t bias_shift,
+ const uint16_t out_shift,
+ q15_t * Im_out,
+ const uint16_t dim_im_out_x,
+ const uint16_t dim_im_out_y,
+ q15_t * bufferA,
+ q7_t * bufferB);
+
+ /**
+ * @brief Basic Q7 version of CHW convolution (non-sqaure shape)
+ * @param[in] Im_in pointer to input tensor
+ * @param[in] dim_im_in_x input tensor dimention x
+ * @param[in] dim_im_in_y input tensor dimention y
+ * @param[in] ch_im_in number of input tensor channels
+ * @param[in] wt pointer to kernel weights
+ * @param[in] ch_im_out number of filters, i.e., output tensor channels
+ * @param[in] dim_kernel_x filter kernel size x
+ * @param[in] dim_kernel_y filter kernel size y
+ * @param[in] padding_x padding size x
+ * @param[in] padding_y padding size y
+ * @param[in] stride_x convolution stride x
+ * @param[in] stride_y convolution stride y
+ * @param[in] bias pointer to bias
+ * @param[in] bias_shift amount of left-shift for bias
+ * @param[in] out_shift amount of right-shift for output
+ * @param[in,out] Im_out pointer to output tensor
+ * @param[in] dim_im_out_x output tensor dimension x
+ * @param[in] dim_im_out_y output tensor dimension y
+ * @param[in,out] bufferA pointer to buffer space for input
+ * @param[in,out] bufferB pointer to buffer space for output
+ * @return The function returns <code>ARM_MATH_SUCCESS</code>
+ */
+
+ arm_status
+ arm_convolve_CHW_q7_basic_nonsquare(const q7_t * Im_in,
+ const uint16_t dim_im_in_x,
+ const uint16_t dim_im_in_y,
+ const uint16_t ch_im_in,
+ const q7_t * wt,
+ const uint16_t ch_im_out,
+ const uint16_t dim_kernel_x,
+ const uint16_t dim_kernel_y,
+ const uint16_t padding_x,
+ const uint16_t padding_y,
+ const uint16_t stride_x,
+ const uint16_t stride_y,
+ const q7_t * bias,
+ const uint16_t bias_shift,
+ const uint16_t out_shift,
+ q7_t * Im_out,
+ const uint16_t dim_im_out_x,
+ const uint16_t dim_im_out_y,
+ q15_t * bufferA,
+ q7_t * bufferB);
+
+ /**
+ * @brief Matrix-multiplication function for convolution with CHW output
+ * @param[in] pA pointer to operand A
+ * @param[in] pInBuffer pointer to operand B, always conssists of 2 vectors
+ * @param[in] ch_im_out numRow of A
+ * @param[in] numCol_A numCol of A
+ * @param[in] out_stride output buffer channel stride
+ * @param[in] bias_shift amount of left-shift for bias
+ * @param[in] out_shift amount of right-shift for output
+ * @param[in] bias the bias
+ * @param[in,out] pOut pointer to output
+ * @return The function returns the incremented output pointer
+ *
+ * @details
+ *
+ * This function does the matrix multiplication with weight matrix
+ * and 2 columns from im2col.
+ */
+
+ q7_t *arm_nn_CHW_mat_mult_kernel_q7_q15(const q7_t * pA,
+ const q15_t * pInBuffer,
+ const uint16_t ch_im_out,
+ const uint16_t numCol_A,
+ const uint16_t out_stride,
+ const uint16_t bias_shift,
+ const uint16_t out_shift,
+ const q7_t * bias,
+ q7_t * pOut);
+
+#ifdef __cplusplus
+}
+#endif
+#endif
diff --git a/CMSIS_5/CMSIS/NN/Source/ConvolutionFunctions/arm_convolve_CHW_f32_basic_nonsquare.c CMSIS_5/CMSIS/NN/Source/ConvolutionFunctions/arm_convolve_CHW_f32_basic_nonsquare.c
===CHANGE_NOTICE(2/5)===========================================================
Sony Corporation added this file to 5.4.0 for these reasons:
- support float version of convolution
- support the CHW tensor layout
================================================================================
--- /dev/null
+++ CMSIS_5/CMSIS/NN/Source/ConvolutionFunctions/arm_convolve_CHW_f32_basic_nonsquare.c
@@ -0,0 +1,207 @@
+/*
+ * Copyright (C) 2010-2018 Arm Limited or its affiliates. All rights reserved.
+ * Copyright 2018 Sony Corporation
+ *
+ * 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
+ *
+ * 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.
+ */
+
+/* ----------------------------------------------------------------------
+ * Title: arm_convolve_CHW_f32_basic_nonsquare.c
+ * Author: Sony Corporation
+ * Description: Sony Corporation added this file to 5.4.0 for these reasons:
+ * - support float version of convolution
+ * - support the CHW tensor layout
+ * $Date: 14. September 2018
+ * -------------------------------------------------------------------- */
+
+#include "arm_math.h"
+#include "arm_nnfunctions_nnabla.h"
+
+/**
+ * @ingroup groupNN
+ */
+
+/**
+ * @addtogroup NNConv
+ * @{
+ */
+
+ /**
+ * @brief Basic float32 version of CHW convolution (non-sqaure shape)
+ * @param[in] Im_in pointer to input tensor
+ * @param[in] dim_im_in_x input tensor dimention x
+ * @param[in] dim_im_in_y input tensor dimention y
+ * @param[in] ch_im_in number of input tensor channels
+ * @param[in] wt pointer to kernel weights
+ * @param[in] ch_im_out number of filters, i.e., output tensor channels
+ * @param[in] dim_kernel_x filter kernel size x
+ * @param[in] dim_kernel_y filter kernel size y
+ * @param[in] padding_x padding sizes x
+ * @param[in] padding_y padding sizes y
+ * @param[in] stride_x convolution stride x
+ * @param[in] stride_y convolution stride y
+ * @param[in] bias pointer to bias
+ * @param[in,out] Im_out pointer to output tensor
+ * @param[in] dim_im_out_x output tensor dimension x
+ * @param[in] dim_im_out_y output tensor dimension y
+ * @param[in,out] bufferA pointer to buffer space for input
+ * @param[in,out] bufferB pointer to buffer space for output
+ * @return The function returns <code>ARM_MATH_SUCCESS</code>
+ *
+ * @details
+ *
+ * <b>Buffer size:</b>
+ *
+ * bufferA size: ch_im_in*dim_kernel_x*dim_kernel_y
+ *
+ * bufferB size: 0
+ *
+ * This basic version is designed to work for any input tensor and weight
+ * dimension.
+ */
+
+arm_status
+arm_convolve_CHW_f32_basic_nonsquare(const float * Im_in,
+ const uint16_t dim_im_in_x,
+ const uint16_t dim_im_in_y,
+ const uint16_t ch_im_in,
+ const float * wt,
+ const uint16_t ch_im_out,
+ const uint16_t dim_kernel_x,
+ const uint16_t dim_kernel_y,
+ const uint16_t padding_x,
+ const uint16_t padding_y,
+ const uint16_t stride_x,
+ const uint16_t stride_y,
+ const float * bias,
+ float * Im_out,
+ const uint16_t dim_im_out_x,
+ const uint16_t dim_im_out_y,
+ float * bufferA,
+ float * bufferB)
+{
+
+ /* Run the following code for Cortex-M4 and Cortex-M7 */
+
+ int16_t i_out_y, i_out_x, i_ker_y, i_ker_x;
+ int16_t i_ker_x_begin, i_ker_y_begin;
+ int16_t i_ker_x_end, i_ker_y_end;
+ int16_t single_in_map_size = dim_im_in_x * dim_im_in_y;
+ int16_t kernel_size_2d = dim_kernel_x * dim_kernel_y;
+
+ uint16_t im2col_out_pixel_index = 0;
+ float *pBuffer = bufferA;
+ float *im_buffer = bufferA;
+ const float *pA;
+ int i;
+
+ /* This part implements the im2col function */
+ for (i_out_y = 0; i_out_y < dim_im_out_y; i_out_y++)
+ {
+ for (i_out_x = 0; i_out_x < dim_im_out_x; i_out_x++)
+ {
+ i_ker_y_begin = i_out_y * stride_y - padding_y;
+ i_ker_y_end = i_ker_y_begin + dim_kernel_y;
+
+ for (i_ker_y = i_ker_y_begin; i_ker_y < i_ker_y_end; i_ker_y++)
+ {
+ i_ker_x_begin = i_out_x * stride_x - padding_x;
+ i_ker_x_end = i_ker_x_begin + dim_kernel_x;
+
+ for (i_ker_x = i_ker_x_begin; i_ker_x < i_ker_x_end; i_ker_x++)
+ {
+ float *pDest = pBuffer + (i_ker_y - i_ker_y_begin) * dim_kernel_x + (i_ker_x - i_ker_x_begin);
+ float *pDestEnd = pDest + ch_im_in * kernel_size_2d;
+
+ if (i_ker_y < 0 || i_ker_y >= dim_im_in_y || i_ker_x < 0 || i_ker_x >= dim_im_in_x)
+ {
+ /* Out of bound zero values */
+ for (; pDest < pDestEnd;)
+ {
+ *pDest = 0;
+ pDest += kernel_size_2d;
+ }
+ } else
+ {
+ const float *pSrc = Im_in + i_ker_y * dim_im_in_x + i_ker_x;
+ for (; pDest < pDestEnd;)
+ {
+ *pDest = *pSrc;
+ pSrc += single_in_map_size;
+ pDest += kernel_size_2d;
+ }
+ }
+ }
+ }
+
+ pA = wt;
+ float *pOut = Im_out++;
+ int16_t map_size_out = dim_im_out_x * dim_im_out_y;
+ for (i = 0; i < ch_im_out; i++)
+ {
+ float sum = 0;
+ float *pB = im_buffer;
+ uint16_t colCnt = (ch_im_in * dim_kernel_x * dim_kernel_y) >> 2;
+
+ if (bias)
+ {
+ sum = bias[i];
+ }
+
+ while (colCnt)
+ {
+ float inA1 = *pA++;
+ float inB1 = *pB++;
+ float inA2 = *pA++;
+ float inB2 = *pB++;
+
+ sum += inA1 * inB1;
+ sum += inA2 * inB2;
+
+ inA1 = *pA++;
+ inB1 = *pB++;
+ inA2 = *pA++;
+ inB2 = *pB++;
+
+ sum += inA1 * inB1;
+ sum += inA2 * inB2;
+
+ colCnt--;
+ }
+ colCnt = (ch_im_in * dim_kernel_x * dim_kernel_y) & 0x3;
+ while (colCnt)
+ {
+ float inA1 = *pA++;
+ float inB1 = *pB++;
+ sum += inA1 * inB1;
+ colCnt--;
+ }
+ *pOut = sum;
+ pOut += map_size_out;
+ }
+
+ /* counter reset */
+ pBuffer = im_buffer;
+ im2col_out_pixel_index++;
+ }
+ }
+
+ /* Return to application */
+ return ARM_MATH_SUCCESS;
+}
+
+/**
+ * @} end of NNConv group
+ */
diff --git a/CMSIS_5/CMSIS/NN/Source/ConvolutionFunctions/arm_convolve_CHW_q15_basic_nonsquare.c CMSIS_5/CMSIS/NN/Source/ConvolutionFunctions/arm_convolve_CHW_q15_basic_nonsquare.c
===CHANGE_NOTICE(3/5)===========================================================
Sony Corporation added this file to 5.4.0 to support the CHW tensor layout
================================================================================
--- /dev/null
+++ CMSIS_5/CMSIS/NN/Source/ConvolutionFunctions/arm_convolve_CHW_q15_basic_nonsquare.c
@@ -0,0 +1,231 @@
+/*
+ * Copyright (C) 2010-2018 Arm Limited or its affiliates. All rights reserved.
+ * Copyright 2018 Sony Corporation
+ *
+ * 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
+ *
+ * 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.
+ */
+
+/* ----------------------------------------------------------------------
+ * Title: arm_convolve_CHW_q15_basic_nonsquare.c
+ * Author: Sony Corporation
+ * Description: Sony Corporation added this file to 5.4.0
+ * to support the CHW tensor layout
+ * $Date: 14. September 2018
+ * -------------------------------------------------------------------- */
+
+#include "arm_math.h"
+#include "arm_nnfunctions_nnabla.h"
+
+/**
+ * @ingroup groupNN
+ */
+
+/**
+ * @addtogroup NNConv
+ * @{
+ */
+
+ /**
+ * @brief Basic Q15 version of CHW convolution (non-sqaure shape)
+ * @param[in] Im_in pointer to input tensor
+ * @param[in] dim_im_in_x input tensor dimention x
+ * @param[in] dim_im_in_y input tensor dimention y
+ * @param[in] ch_im_in number of input tensor channels
+ * @param[in] wt pointer to kernel weights
+ * @param[in] ch_im_out number of filters, i.e., output tensor channels
+ * @param[in] dim_kernel_x filter kernel size x
+ * @param[in] dim_kernel_y filter kernel size y
+ * @param[in] padding_x padding sizes x
+ * @param[in] padding_y padding sizes y
+ * @param[in] stride_x convolution stride x
+ * @param[in] stride_y convolution stride y
+ * @param[in] bias pointer to bias
+ * @param[in] bias_shift amount of left-shift for bias
+ * @param[in] out_shift amount of right-shift for output
+ * @param[in,out] Im_out pointer to output tensor
+ * @param[in] dim_im_out_x output tensor dimension x
+ * @param[in] dim_im_out_y output tensor dimension y
+ * @param[in,out] bufferA pointer to buffer space for input
+ * @param[in,out] bufferB pointer to buffer space for output
+ * @return The function returns <code>ARM_MATH_SUCCESS</code>
+ *
+ * @details
+ *
+ * <b>Buffer size:</b>
+ *
+ * bufferA size: ch_im_in*dim_kernel_x*dim_kernel_y
+ *
+ * bufferB size: 0
+ *
+ * This basic version is designed to work for any input tensor and weight
+ * dimension.
+ */
+
+arm_status
+arm_convolve_CHW_q15_basic_nonsquare(const q15_t * Im_in,
+ const uint16_t dim_im_in_x,
+ const uint16_t dim_im_in_y,
+ const uint16_t ch_im_in,
+ const q15_t * wt,
+ const uint16_t ch_im_out,
+ const uint16_t dim_kernel_x,
+ const uint16_t dim_kernel_y,
+ const uint16_t padding_x,
+ const uint16_t padding_y,
+ const uint16_t stride_x,
+ const uint16_t stride_y,
+ const q15_t * bias,
+ const uint16_t bias_shift,
+ const uint16_t out_shift,
+ q15_t * Im_out,
+ const uint16_t dim_im_out_x,
+ const uint16_t dim_im_out_y,
+ q15_t * bufferA,
+ q7_t * bufferB)
+{
+
+ /* Run the following code for Cortex-M4 and Cortex-M7 */
+
+ int16_t i_out_y, i_out_x, i_ker_y, i_ker_x;
+ int16_t i_ker_x_begin, i_ker_y_begin;
+ int16_t i_ker_x_end, i_ker_y_end;
+ int16_t single_in_map_size = dim_im_in_x * dim_im_in_y;
+ int16_t kernel_size_2d = dim_kernel_x * dim_kernel_y;
+
+ uint16_t im2col_out_pixel_index = 0;
+ q15_t *pBuffer = bufferA;
+ q15_t *im_buffer = bufferA;
+ const q15_t *pA;
+ int i;
+
+ /* This part implements the im2col function */
+ for (i_out_y = 0; i_out_y < dim_im_out_y; i_out_y++)
+ {
+ for (i_out_x = 0; i_out_x < dim_im_out_x; i_out_x++)
+ {
+#define USE_CHW_IN_COL
+#ifdef USE_CHW_IN_COL
+
+ i_ker_y_begin = i_out_y * stride_y - padding_y;
+ i_ker_y_end = i_ker_y_begin + dim_kernel_y;
+
+ for (i_ker_y = i_ker_y_begin; i_ker_y < i_ker_y_end; i_ker_y++)
+ {
+ i_ker_x_begin = i_out_x * stride_x - padding_x;
+ i_ker_x_end = i_ker_x_begin + dim_kernel_x;
+
+ for (i_ker_x = i_ker_x_begin; i_ker_x < i_ker_x_end; i_ker_x++)
+ {
+ q15_t *pDest = pBuffer + (i_ker_y - i_ker_y_begin) * dim_kernel_x + (i_ker_x - i_ker_x_begin);
+ q15_t *pDestEnd = pDest + ch_im_in * kernel_size_2d;
+
+ if (i_ker_y < 0 || i_ker_y >= dim_im_in_y || i_ker_x < 0 || i_ker_x >= dim_im_in_x)
+ {
+ /* Out of bound zero values */
+ for (; pDest < pDestEnd;)
+ {
+ *pDest = 0;
+ pDest += kernel_size_2d;
+ }
+ } else
+ {
+ const q15_t *pSrc = Im_in + i_ker_y * dim_im_in_x + i_ker_x;
+ for (; pDest < pDestEnd;)
+ {
+ *pDest = *pSrc;
+ pSrc += single_in_map_size;
+ pDest += kernel_size_2d;
+ }
+ }
+ }
+ }
+#else
+ // HWC in columns
+ for (i_ker_y = i_out_y * stride_y - padding_y; i_ker_y < i_out_y * stride_y - padding_y + dim_kernel_y; i_ker_y++)
+ {
+ for (i_ker_x = i_out_x * stride_x - padding_x; i_ker_x < i_out_x * stride_x - padding_x + dim_kernel_x; i_ker_x++)
+ {
+ if (i_ker_y < 0 || i_ker_y >= dim_im_in_y || i_ker_x < 0 || i_ker_x >= dim_im_in_y)
+ {
+ /* Filling 0 for out-of-bound paddings */
+ /* arm_fill_q15(0, pBuffer, ch_im_in); */
+ memset(pBuffer, 0, sizeof(q15_t)*ch_im_in);
+ } else
+ {
+ /* load CHW patch to HWC column */
+ const q15_t *pSrc = Im_in + i_ker_y * dim_im_in_x + i_ker_x;
+ for (int16_t ch_idx = 0; ch_idx < ch_im_in; ch_idx++)
+ {
+ pBuffer[ch_idx++] = *pSrc;
+ pSrc += single_in_map_size;
+ }
+ }
+
+ pBuffer += ch_im_in;
+ }
+ }
+#endif
+
+ pA = wt;
+ q15_t *pOut = Im_out++;
+ int16_t map_size_out = dim_im_out_x * dim_im_out_y;
+ for (i = 0; i < ch_im_out; i++)
+ {
+ q31_t sum = 0;
+ q15_t *pB = im_buffer;
+ uint16_t colCnt = (ch_im_in * dim_kernel_x * dim_kernel_y) >> 2;
+
+ if (bias)
+ {
+ sum = ((q31_t)bias[i] << bias_shift) + NN_ROUND(out_shift);
+ }
+
+ while (colCnt)
+ {
+ q31_t inA1 = *__SIMD32(pA)++;
+ q31_t inB1 = *__SIMD32(pB)++;
+ q31_t inA2 = *__SIMD32(pA)++;
+ q31_t inB2 = *__SIMD32(pB)++;
+
+ sum = __SMLAD(inA1, inB1, sum);
+ sum = __SMLAD(inA2, inB2, sum);
+
+ colCnt--;
+ }
+ colCnt = (ch_im_in * dim_kernel_x * dim_kernel_y) & 0x3;
+ while (colCnt)
+ {
+ q15_t inA1 = *pA++;
+ q15_t inB1 = *pB++;
+ sum += inA1 * inB1;
+ colCnt--;
+ }
+ *pOut = (q15_t) __SSAT((sum >> out_shift), 16);
+ pOut += map_size_out;
+ }
+
+ /* counter reset */
+ pBuffer = im_buffer;
+ im2col_out_pixel_index++;
+ }
+ }
+
+ /* Return to application */
+ return ARM_MATH_SUCCESS;
+}
+
+/**
+ * @} end of NNConv group
+ */
diff --git a/CMSIS_5/CMSIS/NN/Source/ConvolutionFunctions/arm_convolve_CHW_q7_basic_nonsquare.c CMSIS_5/CMSIS/NN/Source/ConvolutionFunctions/arm_convolve_CHW_q7_basic_nonsquare.c
===CHANGE_NOTICE(4/5)===========================================================
Sony Corporation added this file to 5.4.0 to support the CHW tensor layout
================================================================================
--- /dev/null
+++ CMSIS_5/CMSIS/NN/Source/ConvolutionFunctions/arm_convolve_CHW_q7_basic_nonsquare.c
@@ -0,0 +1,214 @@
+/*
+ * Copyright (C) 2010-2018 Arm Limited or its affiliates. All rights reserved.
+ * Copyright 2018 Sony Corporation
+ *
+ * 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
+ *
+ * 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.
+ */
+
+/* ----------------------------------------------------------------------
+ * Title: arm_convolve_CHW_q7_basic_nonsquare.c
+ * Author: Sony Corporation
+ * Description: Sony Corporation added this file to 5.4.0
+ * to support the CHW tensor layout
+ * $Date: 14. September 2018
+ * -------------------------------------------------------------------- */
+
+#include "arm_math.h"
+#include "arm_nnfunctions_nnabla.h"
+
+/**
+ * @ingroup groupNN
+ */
+
+/**
+ * @addtogroup NNConv
+ * @{
+ */
+
+ /**
+ * @brief Basic Q7 version of CHW convolution (non-sqaure shape)
+ * @param[in] Im_in pointer to input tensor
+ * @param[in] dim_im_in_x input tensor dimention x
+ * @param[in] dim_im_in_y input tensor dimention y
+ * @param[in] ch_im_in number of input tensor channels
+ * @param[in] wt pointer to kernel weights
+ * @param[in] ch_im_out number of filters, i.e., output tensor channels
+ * @param[in] dim_kernel_x filter kernel size x
+ * @param[in] dim_kernel_y filter kernel size y
+ * @param[in] padding_x padding size x
+ * @param[in] padding_y padding size y
+ * @param[in] stride_x convolution stride x
+ * @param[in] stride_y convolution stride y
+ * @param[in] bias pointer to bias
+ * @param[in] bias_shift amount of left-shift for bias
+ * @param[in] out_shift amount of right-shift for output
+ * @param[in,out] Im_out pointer to output tensor
+ * @param[in] dim_im_out_x output tensor dimension x
+ * @param[in] dim_im_out_y output tensor dimension y
+ * @param[in,out] bufferA pointer to buffer space for input
+ * @param[in,out] bufferB pointer to buffer space for output
+ * @return The function returns <code>ARM_MATH_SUCCESS</code>
+ */
+
+arm_status arm_convolve_CHW_q7_basic_nonsquare(const q7_t * Im_in,
+ const uint16_t dim_im_in_x,
+ const uint16_t dim_im_in_y,
+ const uint16_t ch_im_in,
+ const q7_t * wt,
+ const uint16_t ch_im_out,
+ const uint16_t dim_kernel_x,
+ const uint16_t dim_kernel_y,
+ const uint16_t padding_x,
+ const uint16_t padding_y,
+ const uint16_t stride_x,
+ const uint16_t stride_y,
+ const q7_t * bias,
+ const uint16_t bias_shift,
+ const uint16_t out_shift,
+ q7_t * Im_out,
+ const uint16_t dim_im_out_x,
+ const uint16_t dim_im_out_y,
+ q15_t * bufferA,
+ q7_t * bufferB)
+{
+
+ /* Run the following code for Cortex-M4 and Cortex-M7 */
+
+ int16_t i_out_y, i_out_x, i_ker_y, i_ker_x;
+ int16_t i_ker_x_begin, i_ker_y_begin;
+ int16_t i_ker_x_end, i_ker_y_end;
+ int16_t single_in_map_size = dim_im_in_x * dim_im_in_y;
+ int16_t kernel_size_2d = dim_kernel_x * dim_kernel_y;
+ int16_t kernel_size_3d = ch_im_in * kernel_size_2d;
+
+ /*
+ * Here we use bufferA as q15_t internally as computation are done with q15_t level
+ * im2col are done to output in q15_t format from q7_t input
+ */
+ q15_t *pBuffer = bufferA;
+ q7_t *pOut = Im_out;
+
+ /* This part implements the im2col function */
+ for (i_out_y = 0; i_out_y < dim_im_out_y; i_out_y++)
+ {
+ for (i_out_x = 0; i_out_x < dim_im_out_x; i_out_x++)
+ {
+ i_ker_y_begin = i_out_y * stride_y - padding_y;
+ i_ker_y_end = i_out_y * stride_y - padding_y + dim_kernel_y;
+
+ for (i_ker_y = i_ker_y_begin; i_ker_y < i_ker_y_end; i_ker_y++)
+ {
+
+ i_ker_x_begin = i_out_x * stride_x - padding_x;
+ i_ker_x_end = i_out_x * stride_x - padding_x + dim_kernel_x;
+
+ for (i_ker_x = i_ker_x_begin; i_ker_x < i_ker_x_end; i_ker_x++)
+ {
+
+ q15_t *pDest = pBuffer + (i_ker_y - i_ker_y_begin) * dim_kernel_x + (i_ker_x - i_ker_x_begin);
+ q15_t *pDestEnd = pDest + ch_im_in * kernel_size_2d;
+
+ if (i_ker_y < 0 || i_ker_y >= dim_im_in_y || i_ker_x < 0 || i_ker_x >= dim_im_in_x)
+ {
+ /* Filling 0 for out-of-bound paddings */
+ for (; pDest < pDestEnd;)
+ {
+ *pDest = 0;
+ pDest += kernel_size_2d;
+ }
+ } else
+ {
+ /* Copying the pixel data to column */
+ const q7_t *pSrc = Im_in + i_ker_y * dim_im_in_x + i_ker_x;
+ for (; pDest < pDestEnd;)
+ {
+ *pDest = *pSrc;
+ pSrc += single_in_map_size;
+ pDest += kernel_size_2d;
+ }
+ }
+ }
+ }
+
+ pBuffer += kernel_size_3d;
+
+ /* Computation is filed for every 2 columns */
+ if (pBuffer == bufferA + 2 * kernel_size_3d)
+ {
+ pOut =
+ arm_nn_CHW_mat_mult_kernel_q7_q15(wt, bufferA,
+ ch_im_out,
+ ch_im_in *
+ dim_kernel_y * dim_kernel_x,
+ dim_im_out_y * dim_im_out_x,
+ bias_shift, out_shift, bias, pOut);
+
+ /* counter reset */
+ pBuffer = bufferA;
+ }
+ }
+ }
+
+ /* left-over because odd number of output pixels */
+ if (pBuffer != bufferA)
+ {
+ const q7_t *pA = wt;
+ int i;
+
+ for (i = 0; i < ch_im_out; i++)
+ {
+ /* Load the accumulator with bias first */
+ q31_t sum = ((q31_t)bias[i] << bias_shift) + NN_ROUND(out_shift);
+
+ /* Point to the beging of the im2col buffer */
+ q15_t *pB = bufferA;
+
+ /* Each time it process 4 entries */
+ uint16_t colCnt = kernel_size_3d >> 2;
+
+ while (colCnt)
+ {
+ q31_t inA1, inA2;
+ q31_t inB1, inB2;
+
+ pA = (q7_t *) read_and_pad((void *)pA, &inA1, &inA2);
+
+ inB1 = *__SIMD32(pB)++;
+ sum = __SMLAD(inA1, inB1, sum);
+ inB2 = *__SIMD32(pB)++;
+ sum = __SMLAD(inA2, inB2, sum);
+
+ colCnt--;
+ }
+ colCnt = kernel_size_3d & 0x3;
+ while (colCnt)
+ {
+ q7_t inA1 = *pA++;
+ q15_t inB1 = *pB++;
+ sum += inA1 * inB1;
+ colCnt--;
+ }
+ *pOut = (q7_t) __SSAT((sum >> out_shift), 8);
+ pOut += dim_im_out_y * dim_im_out_x;
+ }
+ }
+
+ /* Return to application */
+ return ARM_MATH_SUCCESS;
+}
+
+/**
+ * @} end of NNConv group
+ */
diff --git a/CMSIS_5/CMSIS/NN/Source/ConvolutionFunctions/arm_nn_CHW_mat_mult_kernel_q7_q15.c CMSIS_5/CMSIS/NN/Source/ConvolutionFunctions/arm_nn_CHW_mat_mult_kernel_q7_q15.c
===CHANGE_NOTICE(5/5)===========================================================
Sony Corporation added this file to 5.4.0 to support the CHW tensor layout
================================================================================
--- /dev/null
+++ CMSIS_5/CMSIS/NN/Source/ConvolutionFunctions/arm_nn_CHW_mat_mult_kernel_q7_q15.c
@@ -0,0 +1,196 @@
+/*
+ * Copyright (C) 2010-2018 Arm Limited or its affiliates. All rights reserved.
+ * Copyright 2018 Sony Corporation
+ *
+ * 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
+ *
+ * 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.
+ */
+
+/* ----------------------------------------------------------------------
+ * Title: arm_nn_CHW_mat_mult_kernel_q7_q15.c
+ * Author: Sony Corporation
+ * Description: Sony Corporation added this file to 5.4.0
+ * to support the CHW tensor layout in convolution
+ * $Date: 14. September 2018
+ * -------------------------------------------------------------------- */
+
+#include "arm_math.h"
+#include "arm_nnfunctions_nnabla.h"
+
+ /**
+ * @brief Matrix-multiplication function for convolution with CHW output
+ * @param[in] pA pointer to operand A
+ * @param[in] pInBuffer pointer to operand B, always conssists of 2 vectors
+ * @param[in] ch_im_out numRow of A
+ * @param[in] numCol_A numCol of A
+ * @param[in] out_stride output buffer channel stride
+ * @param[in] bias_shift amount of left-shift for bias
+ * @param[in] out_shift amount of right-shift for output
+ * @param[in] bias the bias
+ * @param[in,out] pOut pointer to output
+ * @return The function returns the incremented output pointer
+ *
+ * @details
+ *
+ * This function does the matrix multiplication with weight matrix
+ * and 2 columns from im2col.
+ */
+
+q7_t *arm_nn_CHW_mat_mult_kernel_q7_q15(const q7_t * pA,
+ const q15_t * pInBuffer,
+ const uint16_t ch_im_out,
+ const uint16_t numCol_A,
+ const uint16_t out_stride,
+ const uint16_t bias_shift,
+ const uint16_t out_shift,
+ const q7_t * bias,
+ q7_t * pOut)
+{
+ /* set up the second output pointers */
+ q7_t *pOut_base = pOut;
+ q7_t *pOut2;
+ const q7_t *pBias = bias;
+ int16_t i_row;
+
+ uint16_t rowCnt = ch_im_out >> 1;
+ /* this loop over rows in A */
+ for (i_row = 0; i_row < rowCnt; ++i_row)
+ {
+ /* setup output pointers */
+ pOut = pOut_base + 2 * i_row * out_stride;
+ pOut2 = pOut + out_stride;
+
+ /* setup pointers for B */
+ const q15_t *pB = pInBuffer;
+ const q15_t *pB2 = pB + numCol_A;
+
+ /* align the second pointer for A */
+ const q7_t *pA2 = pA + numCol_A;
+
+ /* sum & sum3 belong to same outmap, sum2 & sum4 belong to another outmap
+ *
+ * sum sum3
+ * sum2 sum4
+ *
+ */
+ /* init the sum with bias */
+ q31_t sum = ((q31_t)(*pBias++) << bias_shift) + NN_ROUND(out_shift);
+ q31_t sum3 = sum;
+ q31_t sum2 = ((q31_t)(*pBias++) << bias_shift) + NN_ROUND(out_shift);
+ q31_t sum4 = sum2;
+
+ uint16_t colCnt = numCol_A >> 2;
+ /* accumulate over the vector */
+ while (colCnt)
+ {
+ q31_t inA11, inA12, inA21, inA22;
+ q31_t inB1 = *__SIMD32(pB)++;
+ q31_t inB2 = *__SIMD32(pB2)++;
+
+ /* pA is in CHW -> inA11 & inA12 belong to same out-map weight */
+ pA = (q7_t *) read_and_pad((void *)pA, &inA11, &inA12);
+ pA2 = (q7_t *) read_and_pad((void *)pA2, &inA21, &inA22);
+
+ /* inB1 belongs to the first columns, inB2 is the second column */
+ sum = __SMLAD(inA11, inB1, sum);
+ sum3 = __SMLAD(inA11, inB2, sum3);
+ sum2 = __SMLAD(inA21, inB1, sum2);
+ sum4 = __SMLAD(inA21, inB2, sum4);
+
+ inB1 = *__SIMD32(pB)++;
+ inB2 = *__SIMD32(pB2)++;
+
+ sum = __SMLAD(inA12, inB1, sum);
+ sum3 = __SMLAD(inA12, inB2, sum3);
+ sum2 = __SMLAD(inA22, inB1, sum2);
+ sum4 = __SMLAD(inA22, inB2, sum4);
+
+ colCnt--;
+ } /* while over colCnt */
+ colCnt = numCol_A & 0x3;
+ while (colCnt)
+ {
+ q7_t inA1 = *pA++;
+ q15_t inB1 = *pB++;
+ q7_t inA2 = *pA2++;
+ q15_t inB2 = *pB2++;
+
+ sum += inA1 * inB1;
+ sum3 += inA1 * inB2;
+ sum2 += inA2 * inB1;
+ sum4 += inA2 * inB2;
+ colCnt--;
+ } /* while over colCnt */
+ *pOut++ = (q7_t) __SSAT((sum >> out_shift), 8);
+ *pOut = (q7_t) __SSAT((sum3 >> out_shift), 8);
+ *pOut2++ = (q7_t) __SSAT((sum2 >> out_shift), 8);
+ *pOut2 = (q7_t) __SSAT((sum4 >> out_shift), 8);
+
+ /* skip the row computed with A2 */
+ pA += numCol_A;
+ } /* for over ch_im_out */
+
+ /* compute left-over row if any */
+ if (ch_im_out & 0x1)
+ {
+ /* setup output pointers */
+ pOut = pOut_base + (ch_im_out - 1) * out_stride;
+
+ /* setup pointers for B */
+ const q15_t *pB = pInBuffer;
+ const q15_t *pB2 = pB + numCol_A;
+
+ /* load the bias */
+ q31_t sum = ((q31_t)(*pBias) << bias_shift) + NN_ROUND(out_shift);
+ q31_t sum3 = sum;
+
+ uint16_t colCnt = numCol_A >> 2;
+ while (colCnt)
+ {
+ q31_t inA11, inA12;
+ q31_t inB1 = *__SIMD32(pB)++;
+ q31_t inB2 = *__SIMD32(pB2)++;
+
+ pA = (q7_t *) read_and_pad((void *)pA, &inA11, &inA12);
+
+ sum = __SMLAD(inA11, inB1, sum);
+ sum3 = __SMLAD(inA11, inB2, sum3);
+
+ inB1 = *__SIMD32(pB)++;
+ inB2 = *__SIMD32(pB2)++;
+
+ sum = __SMLAD(inA12, inB1, sum);
+ sum3 = __SMLAD(inA12, inB2, sum3);
+
+ colCnt--;
+ }
+ colCnt = numCol_A & 0x3;
+ while (colCnt)
+ {
+ q7_t inA1 = *pA++;
+ q15_t inB1 = *pB++;
+ q15_t inB2 = *pB2++;
+
+ sum += inA1 * inB1;
+ sum3 += inA1 * inB2;
+ colCnt--;
+ }
+
+ *pOut++ = (q7_t) __SSAT((sum >> out_shift), 8);
+ *pOut = (q7_t) __SSAT((sum3 >> out_shift), 8);
+ }
+
+ /* return the new output pointer with offset */
+ return pOut_base + 2;
+}
Reference in New Issue View Git Blame Copy Permalink