MCU上三角函数运算速度对比

前言

使用170M主频M4f核芯片(STM32G431),启用浮点加速，启用arm_dsp库。
对比条件：

方式	输入	输出
cordic	q15	q15
arm_dsp_q15	q15	q15
arm_dsp_q31	q31	q31
arm_dsp_f32	float	float
math	float	float
math	double	double
taylor6	float	float

测试代码

硬件cordic

/**
  * @brief  Trigonometrical functions type definition
  */
typedef struct
{
  int16_t hCos;
  int16_t hSin;
} Trig_Components;
/* CORDIC FUNCTION: COSINE q1.15 */
#define CORDIC_CONFIG_COSINE    (LL_CORDIC_FUNCTION_COSINE | LL_CORDIC_PRECISION_6CYCLES | LL_CORDIC_SCALE_0 |\
         LL_CORDIC_NBWRITE_1 | LL_CORDIC_NBREAD_1 |\
         LL_CORDIC_INSIZE_16BITS | LL_CORDIC_OUTSIZE_16BITS)

__weak Trig_Components MCM_Trig_Functions(int16_t hAngle)
{
  union u32toi16x2 {
    uint32_t CordicRdata;
    Trig_Components Components;
  } CosSin;

  WRITE_REG(CORDIC->CSR, CORDIC_CONFIG_COSINE);
  LL_CORDIC_WriteData(CORDIC, ((uint32_t)0x7FFF0000) + ((uint32_t)hAngle));
  CosSin.CordicRdata = LL_CORDIC_ReadData(CORDIC);
  return (CosSin.Components);
 
}

泰勒展开函数

float factorial(int n) {
    if (n == 0) {
        return 1.0f;
    } else {
        return n * factorial(n - 1);
    }
}

float taylor_sin(float x, int n) {
    float result = 0.0f;
    float sign = 1.0f;
    for (int i = 0; i < n; i++) {
        result += (sign * powf(x, 2 * i + 1)) / (factorial(2 * i + 1));
        sign = -sign;
    }
    return result;
}


float taylor_cos(float x, int n) {
    float result = 0.0f;
    float sign = 1.0f;
    for (int i = 0; i < n; i++) {
        result += (sign * powf(x, 2 * i)) / (factorial(2 * i));
        sign = -sign;
  }
    return result;
}

测试流程

#include "arm_math.h"
#include "math.h"

void math_test(void)
{
  usb_printf("\r\n\r\n######################\r\n");
  HAL_Delay(0);
  int16_t angle = rand() % 0xffff - 0x7fff;
  float anglef = (float)angle / 0x7fff * PI;
  usb_printf("[input]:%.6f\r\n",anglef * 360 / 2 / PI);
  HAL_Delay(0);
  
  
  int time_base = 0,time_cost = 0;
  float sindata = 0;
  float cosdata = 0;
  Trig_Components sincos = {0};

  time_base = time.time();
  sincos = MCM_Trig_Functions(angle);
  sindata = (float)sincos.hSin / 0x7fff;
  cosdata = (float)sincos.hCos / 0x7fff;
  time_cost = time.time() - time_base;
  usb_printf("[type]:cordq15\t[time]:%fus [output]:%.6f,%.6f\r\n",(float)time_cost / 170,sindata,cosdata);
  HAL_Delay(10);
  
  time_base = time.time();
  sindata = (float)arm_sin_q15(angle >> 1) / 0x7fff;
  cosdata = (float)arm_cos_q15(angle >> 1) / 0x7fff;
  time_cost = time.time() - time_base;
  usb_printf("[type]:dspq15\t[time]:%fus [output]:%.6f,%.6f\r\n",(float)time_cost / 170,sindata,cosdata);
  HAL_Delay(10);
  
  time_base = time.time();
  sindata = (float)arm_sin_q31(angle << 15) / 0x7fffffff;
  cosdata = (float)arm_cos_q31(angle << 15) / 0x7fffffff;
  time_cost = time.time() - time_base;
  usb_printf("[type]:dspq31\t[time]:%fus [output]:%.6f,%.6f\r\n",(float)time_cost / 170,sindata,cosdata);
  HAL_Delay(10);
  
  time_base = time.time();
  sindata = arm_sin_f32(anglef);
  cosdata = arm_cos_f32(anglef);
  time_cost = time.time() - time_base;
  usb_printf("[type]:dspf32\t[time]:%fus [output]:%.6f,%.6f\r\n",(float)time_cost / 170,sindata,cosdata);
  HAL_Delay(10);
  
  time_base = time.time();
  sindata = sinf(anglef);
  cosdata = cosf(anglef);
  time_cost = time.time() - time_base;
  usb_printf("[type]:mathf\t[time]:%fus [output]:%.6f,%.6f\r\n",(float)time_cost / 170,sindata,cosdata);
  HAL_Delay(10);

  time_base = time.time();
  sindata = sin(anglef);
  cosdata = cos(anglef);
  time_cost = time.time() - time_base;
  usb_printf("[type]:mathd\t[time]:%fus [output]:%.6f,%.6f\r\n",(float)time_cost / 170,sindata,cosdata);
  HAL_Delay(10);
  
  time_base = time.time();
  sindata = taylor_sin(anglef,6);
  cosdata = taylor_cos(anglef,6);
  time_cost = time.time() - time_base;
  usb_printf("[type]:taylor6\t[time]:%fus [output]:%.6f,%.6f\r\n",(float)time_cost / 170,sindata,cosdata);
  HAL_Delay(10);
}

输出结果

######################
[input]:140.343643
[type]:cordq15	[time]:0.923529us [output]:0.638234,-0.769829
[type]:dspq15	[time]:1.441176us [output]:0.638203,-0.769860
[type]:dspq31	[time]:1.447059us [output]:0.638228,-0.769824
[type]:dspf32	[time]:1.682353us [output]:0.638170,-0.769872
[type]:mathf	[time]:2.458823us [output]:0.638182,-0.769886
[type]:mathd	[time]:36.694118us [output]:0.638182,-0.769886
[type]:taylor6	[time]:38.758823us [output]:0.638164,-0.769980


######################
[input]:67.578964
[type]:cordq15	[time]:0.923529us [output]:0.924406,0.381451
[type]:dspq15	[time]:1.382353us [output]:0.924345,0.381420
[type]:dspq31	[time]:1.552941us [output]:0.924386,0.381440
[type]:dspf32	[time]:1.629412us [output]:0.924399,0.381407
[type]:mathf	[time]:2.458823us [output]:0.924406,0.381410
[type]:mathd	[time]:30.770588us [output]:0.924406,0.381410
[type]:taylor6	[time]:39.188236us [output]:0.924406,0.381410


######################
[input]:51.236301
[type]:cordq15	[time]:0.923529us [output]:0.779717,0.626118
[type]:dspq15	[time]:1.382353us [output]:0.779626,0.626118
[type]:dspq31	[time]:1.835294us [output]:0.779711,0.626126
[type]:dspf32	[time]:1.858824us [output]:0.779728,0.626104
[type]:mathf	[time]:2.458823us [output]:0.779735,0.626110
[type]:mathd	[time]:29.305882us [output]:0.779735,0.626110
[type]:taylor6	[time]:39.223530us [output]:0.779735,0.626110


######################
[input]:-95.369728
[type]:cordq15	[time]:0.923529us [output]:-0.995575,-0.093539
[type]:dspq15	[time]:1.388235us [output]:-0.995697,-0.093692
[type]:dspq31	[time]:1.558824us [output]:-0.995599,-0.093530
[type]:dspf32	[time]:1.623529us [output]:-0.995594,-0.093581
[type]:mathf	[time]:2.505882us [output]:-0.995612,-0.093582
[type]:mathd	[time]:27.211765us [output]:-0.995612,-0.093582
[type]:taylor6	[time]:41.788235us [output]:-0.995612,-0.093583