ZYNQ的双核A9移植ARM DSP库也是没问题的，很多函数也做了NEON指令加速，使能宏定义ARM_MATH_NEON即可，爽歪歪
专门为A核信号处理而做，爽爽爽！
以下面函数为例：
/* ----------------------------------------------------------------------
* Project: CMSIS DSP Library
* Title: arm_braycurtis_distance_f32.c
* Description:Bray-Curtis distance between two vectors
*
* $Date: 23 April 2021
* $Revision: V1.9.0
*
* Target Processor: Cortex-M and Cortex-A cores
* -------------------------------------------------------------------- */
/*
* Copyright (C) 2010-2021 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.
*/
#include "dsp/distance_functions.h"
#include <limits.h>
#include <math.h>
/**
@addtogroup braycurtis
@{
*/
/**
* @brief Bray-Curtis distance between two vectors
* @param pA First vector
* @param pB Second vector
* @param blockSizevector length
* @return distance
*
*/
#if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
#include "arm_helium_utils.h"
float32_t arm_braycurtis_distance_f32(const float32_t *pA,const float32_t *pB, uint32_t blockSize)
{
float32_t accumDiff = 0.0f, accumSum = 0.0f;
uint32_t blkCnt;
f32x4_t a, b, c, accumDiffV, accumSumV;
accumDiffV = vdupq_n_f32(0.0f);
accumSumV = vdupq_n_f32(0.0f);
blkCnt = blockSize >> 2;
while (blkCnt > 0) {
a = vld1q(pA);
b = vld1q(pB);
c = vabdq(a, b);
accumDiffV = vaddq(accumDiffV, c);
c = vaddq_f32(a, b);
c = vabsq_f32(c);
accumSumV = vaddq(accumSumV, c);
pA += 4;
pB += 4;
blkCnt--;
}
blkCnt = blockSize & 3;
if (blkCnt > 0U) {
mve_pred16_t p0 = vctp32q(blkCnt);
a = vldrwq_z_f32(pA, p0);
b = vldrwq_z_f32(pB, p0);
c = vabdq(a, b);
accumDiffV = vaddq_m(accumDiffV, accumDiffV, c, p0);
c = vaddq_f32(a, b);
c = vabsq_f32(c);
accumSumV = vaddq_m(accumSumV, accumSumV, c, p0);
}
accumDiff = vecAddAcrossF32Mve(accumDiffV);
accumSum = vecAddAcrossF32Mve(accumSumV);
/*
It is assumed that accumSum is not zero. Since it is the sum of several absolute
values it would imply that all of them are zero. It is very unlikely for long vectors.
*/
return (accumDiff / accumSum);
}
#else
#if defined(ARM_MATH_NEON)
#include "NEMath.h"
float32_t arm_braycurtis_distance_f32(const float32_t *pA,const float32_t *pB, uint32_t blockSize)
{
float32_t accumDiff=0.0f, accumSum=0.0f;
uint32_t blkCnt;
float32x4_t a,b,c,accumDiffV, accumSumV;
float32x2_t accumV2;
accumDiffV = vdupq_n_f32(0.0f);
accumSumV = vdupq_n_f32(0.0f);
blkCnt = blockSize >> 2;
while(blkCnt > 0)
{
a = vld1q_f32(pA);
b = vld1q_f32(pB);
c = vabdq_f32(a,b);
accumDiffV = vaddq_f32(accumDiffV,c);
c = vaddq_f32(a,b);
c = vabsq_f32(c);
accumSumV = vaddq_f32(accumSumV,c);
pA += 4;
pB += 4;
blkCnt --;
}
accumV2 = vpadd_f32(vget_low_f32(accumDiffV),vget_high_f32(accumDiffV));
accumDiff = vget_lane_f32(accumV2, 0) + vget_lane_f32(accumV2, 1);
accumV2 = vpadd_f32(vget_low_f32(accumSumV),vget_high_f32(accumSumV));
accumSum = vget_lane_f32(accumV2, 0) + vget_lane_f32(accumV2, 1);
blkCnt = blockSize & 3;
while(blkCnt > 0)
{
accumDiff += fabsf(*pA - *pB);
accumSum += fabsf(*pA++ + *pB++);
blkCnt --;
}
/*
It is assumed that accumSum is not zero. Since it is the sum of several absolute
values it would imply that all of them are zero. It is very unlikely for long vectors.
*/
return(accumDiff / accumSum);
}
#else
float32_t arm_braycurtis_distance_f32(const float32_t *pA,const float32_t *pB, uint32_t blockSize)
{
float32_t accumDiff=0.0f, accumSum=0.0f, tmpA, tmpB;
while(blockSize > 0)
{
tmpA = *pA++;
tmpB = *pB++;
accumDiff += fabsf(tmpA - tmpB);
accumSum += fabsf(tmpA + tmpB);
blockSize --;
}
/*
It is assumed that accumSum is not zero. Since it is the sum of several absolute
values it would imply that all of them are zero. It is very unlikely for long vectors.
*/
return(accumDiff / accumSum);
}
#endif
#endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
/**
* @} end of braycurtis group
*/
移植成功了分享下，我好白嫖:lol missfox 发表于 2021-11-19 09:43
移植成功了分享下，我好白嫖
这个可以有。 看来以后又可以白嫖zynq了:lol，我靠白嫖都快成为公司ARM专家了{:16:} 它这里面的 fft 算法是支持 neon 的吗，我粗略看了下似乎只有是 mve 的。
是否能移植研究下 Ne10 这个库。 比较下性能。我最近也想研究下算法优化 luguan1997 发表于 2021-11-20 14:30
它这里面的 fft 算法是支持 neon 的吗，我粗略看了下似乎只有是 mve 的。
是否能移植研究下 Ne10 这个库。 ...
FFT没有SIMD加速，带硬件FPU就好使。 硬汉哥，请问下可以出个简单的教学教下咱们怎么移植CMSIS-DSP库到ZYNQ么，很想给这个库和NEON用上，然而搜遍全网没啥资料，就这里有一点{:16:} 请问这个DSP库搞咋样啦？？ 本帖最后由 venus5712 于 2022-9-30 14:01 编辑
请问这个DSP库搞咋样啦？(不好意思多发了一个，DSP库我下载了一个，地址：Release CMSIS 5.9.0 · ARM-software/CMSIS_5 · GitHub)
请问ZYNQ 移植DSP库有工程吗，可以发出来吗
楼主有没有教程资料啊 这个有教程可以看看吗 请问有移植教程了吗？分享一下啊:lol kl_upc 发表于 2023-5-22 13:34
这个有教程可以看看吗
请问你移植成功了吗？分享一下哈 可以移植吗，是否需要上系统，裸机能玩吗
页:
[1]