AGC算法中有两个很重要的优化函数,没有具体测试这两个函数到底优化多少
2022年8月18日做了一个MIPS的效率测试,测试函数如下
void TranslateDBCompare1(void)
{
int i = 0;
float white;
for (i = 0; i < 100; i++)
{
white = (float)(i * (1.0 / 100.0)) + 0.5;
white = AMF_TranslateVeryFastAmpToDB_Fast(white);
}
}
void TranslateDBCompare2(void)
{
int i = 0;
float white;
for (i = 0; i < 100; i++)
{
white = (float)(i * (1.0 / 100.0)) + 0.5;
white = AMF_TranslateVeryFastAmpToDB(white);
}
}
void TranslateDBCompare3(void)
{
int i = 0;
float white;
for (i = 0; i < 100; i++)
{
white = (float)(i * (1.0 / 100.0)) + 0.5;
white = AMF_TranslateVeryFastDBToAmp(white);
}
}
void TranslateDBCompare4(void)
{
int i = 0;
float white;
for (i = 0; i < 100; i++)
{
white = (float)(i * (1.0 / 100.0)) + 0.5;
white = AMF_TranslateVeryFastDBToAmp_Fast(white);
}
}
打印结果如下
white: AMF_TranslateVeryFastAmpToDB_Fast AMF_TranslateVeryFastAmpToDB
0.500000: -6.020600 -6.020600
0.510000: -5.900188 -5.848597
0.520000: -5.779776 -5.679934
0.530000: -5.659364 -5.514483
0.540000: -5.538952 -5.352125
0.550000: -5.418540 -5.192746
0.560000: -5.298128 -5.036240
0.570000: -5.177716 -4.882503
0.580000: -5.057304 -4.731440
0.590000: -4.936892 -4.582959
0.600000: -4.816480 -4.436975
0.610000: -4.696068 -4.293403
0.620000: -4.575656 -4.152166
0.630000: -4.455244 -4.013189
0.640000: -4.334832 -3.876400
0.650000: -4.214420 -3.741733
0.660000: -4.094008 -3.609122
0.670000: -3.973596 -3.478504
0.680000: -3.853184 -3.349822
0.690000: -3.732772 -3.223018
0.700000: -3.612360 -3.098039
0.710000: -3.491948 -2.974833
0.720000: -3.371536 -2.853350
0.730000: -3.251124 -2.733543
0.740000: -3.130712 -2.615366
0.750000: -3.010300 -2.498775
0.760000: -2.889888 -2.383728
0.770000: -2.769476 -2.270186
0.780000: -2.649064 -2.158108
0.790000: -2.528652 -2.047459
0.800000: -2.408240 -1.938200
0.810000: -2.287828 -1.830300
0.820000: -2.167416 -1.723723
0.830000: -2.047004 -1.618438
0.840000: -1.926592 -1.514414
0.850000: -1.806180 -1.411621
0.860000: -1.685768 -1.310031
0.870000: -1.565356 -1.209615
0.880000: -1.444944 -1.110347
0.890000: -1.324532 -1.012200
0.900000: -1.204120 -0.915150
0.910000: -1.083708 -0.819173
0.920000: -0.963297 -0.724244
0.930000: -0.842884 -0.630341
0.940000: -0.722472 -0.537443
0.950000: -0.602060 -0.445528
0.960000: -0.481648 -0.354575
0.970000: -0.361236 -0.264565
0.980000: -0.240824 -0.175478
0.990000: -0.120412 -0.087296
1.000000: 0.000000 0.000000
1.010000: 0.060206 0.086427
1.020000: 0.120412 0.172003
1.030000: 0.180618 0.256744
1.040000: 0.240824 0.340666
1.050000: 0.301030 0.423786
1.060000: 0.361236 0.506117
1.070000: 0.421442 0.587675
1.080000: 0.481648 0.668474
1.090000: 0.541853 0.748529
1.100000: 0.602060 0.827854
1.110000: 0.662266 0.906460
1.120000: 0.722472 0.984361
1.130000: 0.782678 1.061569
1.140000: 0.842884 1.138097
1.150000: 0.903090 1.213957
1.160000: 0.963297 1.289160
1.170000: 1.023502 1.363718
1.180000: 1.083708 1.437641
1.190000: 1.143914 1.510940
1.200000: 1.204120 1.583625
1.210000: 1.264326 1.655708
1.220000: 1.324532 1.727197
1.230000: 1.384738 1.798102
1.240000: 1.444944 1.868434
1.250000: 1.505150 1.938200
1.260000: 1.565356 2.007411
1.270000: 1.625562 2.076074
1.280000: 1.685768 2.144199
1.290000: 1.745974 2.211794
1.300000: 1.806180 2.278867
1.310000: 1.866386 2.345426
1.320000: 1.926592 2.411478
1.330000: 1.986798 2.477032
1.340000: 2.047004 2.542095
1.350000: 2.107210 2.606676
1.360000: 2.167416 2.670778
1.370000: 2.227622 2.734411
1.380000: 2.287828 2.797582
1.390000: 2.348034 2.860296
1.400000: 2.408240 2.922561
1.410000: 2.468446 2.984383
1.420000: 2.528652 3.045767
1.430000: 2.588858 3.106721
1.440000: 2.649064 3.167250
1.450000: 2.709270 3.227360
1.460000: 2.769476 3.287057
1.470000: 2.829682 3.346347
1.480000: 2.889888 3.405235
1.490000: 2.950094 3.463726
white: AMF_TranslateVeryFastDBToAmp_Fast AMF_TranslateVeryFastDBToAmp
0.500000: -6.020600 -6.020600
0.510000: -5.900188 -5.848597
0.520000: -5.779776 -5.679934
0.530000: -5.659364 -5.514483
0.540000: -5.538952 -5.352125
0.550000: -5.418540 -5.192746
0.560000: -5.298128 -5.036240
0.570000: -5.177716 -4.882503
0.580000: -5.057304 -4.731440
0.590000: -4.936892 -4.582959
0.600000: -4.816480 -4.436975
0.610000: -4.696068 -4.293403
0.620000: -4.575656 -4.152166
0.630000: -4.455244 -4.013189
0.640000: -4.334832 -3.876400
0.650000: -4.214420 -3.741733
0.660000: -4.094008 -3.609122
0.670000: -3.973596 -3.478504
0.680000: -3.853184 -3.349822
0.690000: -3.732772 -3.223018
0.700000: -3.612360 -3.098039
0.710000: -3.491948 -2.974833
0.720000: -3.371536 -2.853350
0.730000: -3.251124 -2.733543
0.740000: -3.130712 -2.615366
0.750000: -3.010300 -2.498775
0.760000: -2.889888 -2.383728
0.770000: -2.769476 -2.270186
0.780000: -2.649064 -2.158108
0.790000: -2.528652 -2.047459
0.800000: -2.408240 -1.938200
0.810000: -2.287828 -1.830300
0.820000: -2.167416 -1.723723
0.830000: -2.047004 -1.618438
0.840000: -1.926592 -1.514414
0.850000: -1.806180 -1.411621
0.860000: -1.685768 -1.310031
0.870000: -1.565356 -1.209615
0.880000: -1.444944 -1.110347
0.890000: -1.324532 -1.012200
0.900000: -1.204120 -0.915150
0.910000: -1.083708 -0.819173
0.920000: -0.963297 -0.724244
0.930000: -0.842884 -0.630341
0.940000: -0.722472 -0.537443
0.950000: -0.602060 -0.445528
0.960000: -0.481648 -0.354575
0.970000: -0.361236 -0.264565
0.980000: -0.240824 -0.175478
0.990000: -0.120412 -0.087296
1.000000: 0.000000 0.000000
1.010000: 0.060206 0.086427
1.020000: 0.120412 0.172003
1.030000: 0.180618 0.256744
1.040000: 0.240824 0.340666
1.050000: 0.301030 0.423786
1.060000: 0.361236 0.506117
1.070000: 0.421442 0.587675
1.080000: 0.481648 0.668474
1.090000: 0.541853 0.748529
1.100000: 0.602060 0.827854
1.110000: 0.662266 0.906460
1.120000: 0.722472 0.984361
1.130000: 0.782678 1.061569
1.140000: 0.842884 1.138097
1.150000: 0.903090 1.213957
1.160000: 0.963297 1.289160
1.170000: 1.023502 1.363718
1.180000: 1.083708 1.437641
1.190000: 1.143914 1.510940
1.200000: 1.204120 1.583625
1.210000: 1.264326 1.655708
1.220000: 1.324532 1.727197
1.230000: 1.384738 1.798102
1.240000: 1.444944 1.868434
1.250000: 1.505150 1.938200
1.260000: 1.565356 2.007411
1.270000: 1.625562 2.076074
1.280000: 1.685768 2.144199
1.290000: 1.745974 2.211794
1.300000: 1.806180 2.278867
1.310000: 1.866386 2.345426
1.320000: 1.926592 2.411478
1.330000: 1.986798 2.477032
1.340000: 2.047004 2.542095
1.350000: 2.107210 2.606676
1.360000: 2.167416 2.670778
1.370000: 2.227622 2.734411
1.380000: 2.287828 2.797582
1.390000: 2.348034 2.860296
1.400000: 2.408240 2.922561
1.410000: 2.468446 2.984383
1.420000: 2.528652 3.045767
1.430000: 2.588858 3.106721
1.440000: 2.649064 3.167250
1.450000: 2.709270 3.227360
1.460000: 2.769476 3.287057
1.470000: 2.829682 3.346347
1.480000: 2.889888 3.405235
1.490000: 2.950094 3.463726
总结:
1.这是一种近似的算法,可以预见最后的结果和计算结果不一致,可用于dB值不太精确的场合
2.这种近似的算法,效率却还并不一定高
| Name | Address | File | Line | Size | Total Without Children | Total % Without Children | Total | Total % | level | call count | 帧长度 | 采样率 | MCPS(Hz) |
| TranslateDBCompare1 | 0x394 | 79 | 102 | 10785616 | 0.89 | 32802678 | 2.72 | 3 | 5625 | 256 | 48000 | 1093422.6 | |
| TranslateDBCompare2 | 0x395 | 80 | 103 | 10785617 | 1.89 | 32802679 | 3.72 | 4 | 5626 | 256 | 48000 | 1093228.282 | |
| TranslateDBCompare3 | 0x396 | 81 | 104 | 10785618 | 2.89 | 32802680 | 4.72 | 5 | 5627 | 256 | 48000 | 1093034.032 | |
| TranslateDBCompare4 | 0x397 | 82 | 105 | 10785619 | 3.89 | 32802681 | 5.72 | 6 | 5628 | 256 | 48000 | 1092839.852 |
最后是函数原型
float AMF_TranslateVeryFastAmpToDB(float famp)
{
// out[0] = log (in[0]) * 0.43429448190325f;
return 20.0 * log10f(famp);
}
float AMF_TranslateVeryFastDBToAmp(float fdB)
{
return powf(10.0, fdB / 20.0f);
}
float AMF_TranslateVeryFastDBToAmp_Fast(float dB)
{
// uses nonlinear table lookup to approximate
// pow(10.0, dB/20.0)
// Some math tricks that are used:
// pow(10.0, x) = pow(2.0, x*logBase2of10)
// pow(2.0, x) = pow(2.0, x.int) * pow(2.0, x.frac)
int xint;
float xfrac, tmp, powxfrac;
tmp = dB * 0.166096; // tmp = input * 0.05 * log_base2_of10
xint = (int)floor(tmp); // get integer part of input - set mode 1 bit to truncate towards -infinity
xfrac = tmp - (float)xint; // get fractional part of input
powxfrac = xfrac + 1.0; // linear approximation of desired function
return ldexp(powxfrac, xint); // use scalb assembly instruction
}
float AMF_TranslateVeryFastAmpToDB_Fast(float amp)
{
int exponent, ind1;
float mantissa;
mantissa = frexp(amp, &exponent); // get mantissa and exponent of input number
mantissa = (mantissa - 1.0) * 2.0; // linear approximation of log over range (-1,0)
// add exponent to mantissa, and scale by 20.0 * (1/log10(2)
return 6.0206 * (exponent + mantissa);
}
文章最后,不推荐这种算法
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
