Robert Lopez / CMSIS5

Aded CMSIS5 DSP and NN folder. Needs some work

DSP/source/BasicMathFunctions/arm_mult_q7.c@0:eedb7d567a5d, 2018-04-12 (annotated)

Committer:
robert_lp
Date:
Thu Apr 12 01:31:58 2018 +0000
Revision:
0:eedb7d567a5d
CMSIS5 Library

Who changed what in which revision?

UserRevisionLine numberNew contents of line
robert_lp 0:eedb7d567a5d 1 /* ----------------------------------------------------------------------
robert_lp 0:eedb7d567a5d 2 * Project: CMSIS DSP Library
robert_lp 0:eedb7d567a5d 3 * Title: arm_mult_q7.c
robert_lp 0:eedb7d567a5d 4 * Description: Q7 vector multiplication
robert_lp 0:eedb7d567a5d 5 *
robert_lp 0:eedb7d567a5d 6 * $Date: 27. January 2017
robert_lp 0:eedb7d567a5d 7 * $Revision: V.1.5.1
robert_lp 0:eedb7d567a5d 8 *
robert_lp 0:eedb7d567a5d 9 * Target Processor: Cortex-M cores
robert_lp 0:eedb7d567a5d 10 * -------------------------------------------------------------------- */
robert_lp 0:eedb7d567a5d 11 /*
robert_lp 0:eedb7d567a5d 12 * Copyright (C) 2010-2017 ARM Limited or its affiliates. All rights reserved.
robert_lp 0:eedb7d567a5d 13 *
robert_lp 0:eedb7d567a5d 14 * SPDX-License-Identifier: Apache-2.0
robert_lp 0:eedb7d567a5d 15 *
robert_lp 0:eedb7d567a5d 16 * Licensed under the Apache License, Version 2.0 (the License); you may
robert_lp 0:eedb7d567a5d 17 * not use this file except in compliance with the License.
robert_lp 0:eedb7d567a5d 18 * You may obtain a copy of the License at
robert_lp 0:eedb7d567a5d 19 *
robert_lp 0:eedb7d567a5d 20 * www.apache.org/licenses/LICENSE-2.0
robert_lp 0:eedb7d567a5d 21 *
robert_lp 0:eedb7d567a5d 22 * Unless required by applicable law or agreed to in writing, software
robert_lp 0:eedb7d567a5d 23 * distributed under the License is distributed on an AS IS BASIS, WITHOUT
robert_lp 0:eedb7d567a5d 24 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
robert_lp 0:eedb7d567a5d 25 * See the License for the specific language governing permissions and
robert_lp 0:eedb7d567a5d 26 * limitations under the License.
robert_lp 0:eedb7d567a5d 27 */
robert_lp 0:eedb7d567a5d 28
robert_lp 0:eedb7d567a5d 29 #include "arm_math.h"
robert_lp 0:eedb7d567a5d 30
robert_lp 0:eedb7d567a5d 31 /**
robert_lp 0:eedb7d567a5d 32 * @ingroup groupMath
robert_lp 0:eedb7d567a5d 33 */
robert_lp 0:eedb7d567a5d 34
robert_lp 0:eedb7d567a5d 35 /**
robert_lp 0:eedb7d567a5d 36 * @addtogroup BasicMult
robert_lp 0:eedb7d567a5d 37 * @{
robert_lp 0:eedb7d567a5d 38 */
robert_lp 0:eedb7d567a5d 39
robert_lp 0:eedb7d567a5d 40 /**
robert_lp 0:eedb7d567a5d 41 * @brief Q7 vector multiplication
robert_lp 0:eedb7d567a5d 42 * @param[in] *pSrcA points to the first input vector
robert_lp 0:eedb7d567a5d 43 * @param[in] *pSrcB points to the second input vector
robert_lp 0:eedb7d567a5d 44 * @param[out] *pDst points to the output vector
robert_lp 0:eedb7d567a5d 45 * @param[in] blockSize number of samples in each vector
robert_lp 0:eedb7d567a5d 46 * @return none.
robert_lp 0:eedb7d567a5d 47 *
robert_lp 0:eedb7d567a5d 48 * <b>Scaling and Overflow Behavior:</b>
robert_lp 0:eedb7d567a5d 49 * \par
robert_lp 0:eedb7d567a5d 50 * The function uses saturating arithmetic.
robert_lp 0:eedb7d567a5d 51 * Results outside of the allowable Q7 range [0x80 0x7F] will be saturated.
robert_lp 0:eedb7d567a5d 52 */
robert_lp 0:eedb7d567a5d 53
robert_lp 0:eedb7d567a5d 54 void arm_mult_q7(
robert_lp 0:eedb7d567a5d 55 q7_t * pSrcA,
robert_lp 0:eedb7d567a5d 56 q7_t * pSrcB,
robert_lp 0:eedb7d567a5d 57 q7_t * pDst,
robert_lp 0:eedb7d567a5d 58 uint32_t blockSize)
robert_lp 0:eedb7d567a5d 59 {
robert_lp 0:eedb7d567a5d 60 uint32_t blkCnt; /* loop counters */
robert_lp 0:eedb7d567a5d 61
robert_lp 0:eedb7d567a5d 62 #if defined (ARM_MATH_DSP)
robert_lp 0:eedb7d567a5d 63
robert_lp 0:eedb7d567a5d 64 /* Run the below code for Cortex-M4 and Cortex-M3 */
robert_lp 0:eedb7d567a5d 65 q7_t out1, out2, out3, out4; /* Temporary variables to store the product */
robert_lp 0:eedb7d567a5d 66
robert_lp 0:eedb7d567a5d 67 /* loop Unrolling */
robert_lp 0:eedb7d567a5d 68 blkCnt = blockSize >> 2U;
robert_lp 0:eedb7d567a5d 69
robert_lp 0:eedb7d567a5d 70 /* First part of the processing with loop unrolling. Compute 4 outputs at a time.
robert_lp 0:eedb7d567a5d 71 ** a second loop below computes the remaining 1 to 3 samples. */
robert_lp 0:eedb7d567a5d 72 while (blkCnt > 0U)
robert_lp 0:eedb7d567a5d 73 {
robert_lp 0:eedb7d567a5d 74 /* C = A * B */
robert_lp 0:eedb7d567a5d 75 /* Multiply the inputs and store the results in temporary variables */
robert_lp 0:eedb7d567a5d 76 out1 = (q7_t) __SSAT((((q15_t) (*pSrcA++) * (*pSrcB++)) >> 7), 8);
robert_lp 0:eedb7d567a5d 77 out2 = (q7_t) __SSAT((((q15_t) (*pSrcA++) * (*pSrcB++)) >> 7), 8);
robert_lp 0:eedb7d567a5d 78 out3 = (q7_t) __SSAT((((q15_t) (*pSrcA++) * (*pSrcB++)) >> 7), 8);
robert_lp 0:eedb7d567a5d 79 out4 = (q7_t) __SSAT((((q15_t) (*pSrcA++) * (*pSrcB++)) >> 7), 8);
robert_lp 0:eedb7d567a5d 80
robert_lp 0:eedb7d567a5d 81 /* Store the results of 4 inputs in the destination buffer in single cycle by packing */
robert_lp 0:eedb7d567a5d 82 *__SIMD32(pDst)++ = __PACKq7(out1, out2, out3, out4);
robert_lp 0:eedb7d567a5d 83
robert_lp 0:eedb7d567a5d 84 /* Decrement the blockSize loop counter */
robert_lp 0:eedb7d567a5d 85 blkCnt--;
robert_lp 0:eedb7d567a5d 86 }
robert_lp 0:eedb7d567a5d 87
robert_lp 0:eedb7d567a5d 88 /* If the blockSize is not a multiple of 4, compute any remaining output samples here.
robert_lp 0:eedb7d567a5d 89 ** No loop unrolling is used. */
robert_lp 0:eedb7d567a5d 90 blkCnt = blockSize % 0x4U;
robert_lp 0:eedb7d567a5d 91
robert_lp 0:eedb7d567a5d 92 #else
robert_lp 0:eedb7d567a5d 93
robert_lp 0:eedb7d567a5d 94 /* Run the below code for Cortex-M0 */
robert_lp 0:eedb7d567a5d 95
robert_lp 0:eedb7d567a5d 96 /* Initialize blkCnt with number of samples */
robert_lp 0:eedb7d567a5d 97 blkCnt = blockSize;
robert_lp 0:eedb7d567a5d 98
robert_lp 0:eedb7d567a5d 99 #endif /* #if defined (ARM_MATH_DSP) */
robert_lp 0:eedb7d567a5d 100
robert_lp 0:eedb7d567a5d 101
robert_lp 0:eedb7d567a5d 102 while (blkCnt > 0U)
robert_lp 0:eedb7d567a5d 103 {
robert_lp 0:eedb7d567a5d 104 /* C = A * B */
robert_lp 0:eedb7d567a5d 105 /* Multiply the inputs and store the result in the destination buffer */
robert_lp 0:eedb7d567a5d 106 *pDst++ = (q7_t) __SSAT((((q15_t) (*pSrcA++) * (*pSrcB++)) >> 7), 8);
robert_lp 0:eedb7d567a5d 107
robert_lp 0:eedb7d567a5d 108 /* Decrement the blockSize loop counter */
robert_lp 0:eedb7d567a5d 109 blkCnt--;
robert_lp 0:eedb7d567a5d 110 }
robert_lp 0:eedb7d567a5d 111 }
robert_lp 0:eedb7d567a5d 112
robert_lp 0:eedb7d567a5d 113 /**
robert_lp 0:eedb7d567a5d 114 * @} end of BasicMult group
robert_lp 0:eedb7d567a5d 115 */
robert_lp 0:eedb7d567a5d 116