Subversion Repositories AFRtranscoder

/*

 * Copyright (C) 2010-2018 Arm Limited or its affiliates. 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

 *

 * 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.

 */

/* ----------------------------------------------------------------------

 * Project:      CMSIS NN Library

 * Title:        arm_nn_mat_mult_kernel_q7_q15.c

 * Description:  Matrix-multiplication function for convolution

 *

 * $Date:        17. January 2018

 * $Revision:    V.1.0.0

 *

 * Target Processor:  Cortex-M cores

 * -------------------------------------------------------------------- */

#include "arm_math.h"

#include "arm_nnfunctions.h"

  /**

   * @brief Matrix-multiplication function for convolution

   * @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]       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_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 bias_shift,

                                        const uint16_t out_shift,

                                        const q7_t * bias,

                                        q7_t * pOut)

{

#if defined (ARM_MATH_DSP)

    /* set up the second output pointers */

    q7_t     *pOut2 = pOut + ch_im_out;

    const q7_t *pBias = bias;

    uint16_t  rowCnt = ch_im_out >> 1;

    /* this loop over rows in A */

    while (rowCnt)

    {

        /* 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;

        /* init the sum with bias */

        q31_t     sum =  ((q31_t)(*pBias) << bias_shift) + NN_ROUND(out_shift);

        q31_t     sum2 = ((q31_t)(*pBias++) << bias_shift) + NN_ROUND(out_shift);

        q31_t     sum3 = ((q31_t)(*pBias) << bias_shift) + NN_ROUND(out_shift);

        q31_t     sum4 = ((q31_t)(*pBias++) << bias_shift) + NN_ROUND(out_shift);

        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 = (q7_t *) read_and_pad((void *)pA, &inA11, &inA12);

            pA2 = (q7_t *) read_and_pad((void *)pA2, &inA21, &inA22);

            sum = __SMLAD(inA11, inB1, sum);

            sum2 = __SMLAD(inA11, inB2, sum2);

            sum3 = __SMLAD(inA21, inB1, sum3);

            sum4 = __SMLAD(inA21, inB2, sum4);

            inB1 = *__SIMD32(pB)++;

            inB2 = *__SIMD32(pB2)++;

            sum = __SMLAD(inA12, inB1, sum);

            sum2 = __SMLAD(inA12, inB2, sum2);

            sum3 = __SMLAD(inA22, inB1, sum3);

            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;

            sum2 += inA1 * inB2;

            sum3 += 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;

        rowCnt--;

    }                           /* for over ch_im_out */

    /* compute left-over row if any */

    if (ch_im_out & 0x1)

    {

        /* 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     sum2 = ((q31_t)(*pBias++) << bias_shift) + NN_ROUND(out_shift);

        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);

            sum2 = __SMLAD(inA11, inB2, sum2);

            inB1 = *__SIMD32(pB)++;

            inB2 = *__SIMD32(pB2)++;

            sum = __SMLAD(inA12, inB1, sum);

            sum2 = __SMLAD(inA12, inB2, sum2);

            colCnt--;

        }

        colCnt = numCol_A & 0x3;

        while (colCnt)

        {

            q7_t      inA1 = *pA++;

            q15_t     inB1 = *pB++;

            q15_t     inB2 = *pB2++;

            sum += inA1 * inB1;

            sum2 += inA1 * inB2;

            colCnt--;

        }

        *pOut++ = (q7_t) __SSAT((sum >> out_shift), 8);

        *pOut2++ = (q7_t) __SSAT((sum2 >> out_shift), 8);

    }

    pOut += ch_im_out;

    /* return the new output pointer with offset */

    return pOut;

#else

    /* To be completed */

    return NULL;

#endif                          /* ARM_MATH_DSP */

}