Revert, use sin_cos table in fft.h

#define FFT_USE_SIN_COS_TABLE by default
Increase code size on 700 bytes, but huge speedup (x3) transform
This commit is contained in:
DiSlord 2020-03-23 01:16:36 +03:00
parent 0c0f399657
commit 9221ad04ca

71
fft.h
View file

@ -26,6 +26,8 @@
#include <math.h>
#include <stdint.h>
#define FFT_USE_SIN_COS_TABLE
static uint16_t reverse_bits(uint16_t x, int n) {
uint16_t result = 0;
int i;
@ -33,6 +35,41 @@ static uint16_t reverse_bits(uint16_t x, int n) {
result = (result << 1) | (x & 1U);
return result;
}
#ifdef FFT_USE_SIN_COS_TABLE
static const float sin_table[] = {
/*
* float has about 7.2 digits of precision
for (uint8_t i = 0; i < FFT_SIZE - (FFT_SIZE / 4); i++) {
printf("% .8f,%c", sin(2 * M_PI * i / FFT_SIZE), i % 8 == 7 ? '\n' : ' ');
}
*/
0.00000000, 0.02454123, 0.04906767, 0.07356456, 0.09801714, 0.12241068, 0.14673047, 0.17096189,
0.19509032, 0.21910124, 0.24298018, 0.26671276, 0.29028468, 0.31368174, 0.33688985, 0.35989504,
0.38268343, 0.40524131, 0.42755509, 0.44961133, 0.47139674, 0.49289819, 0.51410274, 0.53499762,
0.55557023, 0.57580819, 0.59569930, 0.61523159, 0.63439328, 0.65317284, 0.67155895, 0.68954054,
0.70710678, 0.72424708, 0.74095113, 0.75720885, 0.77301045, 0.78834643, 0.80320753, 0.81758481,
0.83146961, 0.84485357, 0.85772861, 0.87008699, 0.88192126, 0.89322430, 0.90398929, 0.91420976,
0.92387953, 0.93299280, 0.94154407, 0.94952818, 0.95694034, 0.96377607, 0.97003125, 0.97570213,
0.98078528, 0.98527764, 0.98917651, 0.99247953, 0.99518473, 0.99729046, 0.99879546, 0.99969882,
1.00000000, 0.99969882, 0.99879546, 0.99729046, 0.99518473, 0.99247953, 0.98917651, 0.98527764,
0.98078528, 0.97570213, 0.97003125, 0.96377607, 0.95694034, 0.94952818, 0.94154407, 0.93299280,
0.92387953, 0.91420976, 0.90398929, 0.89322430, 0.88192126, 0.87008699, 0.85772861, 0.84485357,
0.83146961, 0.81758481, 0.80320753, 0.78834643, 0.77301045, 0.75720885, 0.74095113, 0.72424708,
0.70710678, 0.68954054, 0.67155895, 0.65317284, 0.63439328, 0.61523159, 0.59569930, 0.57580819,
0.55557023, 0.53499762, 0.51410274, 0.49289819, 0.47139674, 0.44961133, 0.42755509, 0.40524131,
0.38268343, 0.35989504, 0.33688985, 0.31368174, 0.29028468, 0.26671276, 0.24298018, 0.21910124,
0.19509032, 0.17096189, 0.14673047, 0.12241068, 0.09801714, 0.07356456, 0.04906767, 0.02454123,
0.00000000, -0.02454123, -0.04906767, -0.07356456, -0.09801714, -0.12241068, -0.14673047, -0.17096189,
-0.19509032, -0.21910124, -0.24298018, -0.26671276, -0.29028468, -0.31368174, -0.33688985, -0.35989504,
-0.38268343, -0.40524131, -0.42755509, -0.44961133, -0.47139674, -0.49289819, -0.51410274, -0.53499762,
-0.55557023, -0.57580819, -0.59569930, -0.61523159, -0.63439328, -0.65317284, -0.67155895, -0.68954054,
-0.70710678, -0.72424708, -0.74095113, -0.75720885, -0.77301045, -0.78834643, -0.80320753, -0.81758481,
-0.83146961, -0.84485357, -0.85772861, -0.87008699, -0.88192126, -0.89322430, -0.90398929, -0.91420976,
-0.92387953, -0.93299280, -0.94154407, -0.94952818, -0.95694034, -0.96377607, -0.97003125, -0.97570213,
-0.98078528, -0.98527764, -0.98917651, -0.99247953, -0.99518473, -0.99729046, -0.99879546, -0.99969882,
};
const float* const cos_table = &sin_table[64];
#endif
/***
* dir = forward: 0, inverse: 1
@ -58,27 +95,49 @@ static void fft256(float array[][2], const uint8_t dir) {
array[j][imag] = temp;
}
}
#ifdef FFT_USE_SIN_COS_TABLE
// Cooley-Tukey decimation-in-time radix-2 FFT
for (size = 2; size <= n; size *= 2) {
uint16_t halfsize = size / 2;
uint16_t tablestep = n / size;
uint16_t i;
for (i = 0; i < n; i += size) {
uint16_t j, k;
for (j = i, k = 0; j < i + halfsize; j++, k += tablestep) {
uint16_t l = j + halfsize;
float tpre = array[l][real] * cos(2 * VNA_PI * k / 256) + array[l][imag] * sin(2 * VNA_PI * k / 256);
float tpim = -array[l][real] * sin(2 * VNA_PI * k / 256) + array[l][imag] * cos(2 * VNA_PI * k / 256);
float tpre = array[l][real] * cos_table[k] + array[l][imag] * sin_table[k];
float tpim = -array[l][real] * sin_table[k] + array[l][imag] * cos_table[k];
array[l][real] = array[j][real] - tpre;
array[l][imag] = array[j][imag] - tpim;
array[j][real] += tpre;
array[j][imag] += tpim;
}
}
if (size == n) // Prevent overflow in 'size *= 2'
break;
// if (size == n) // Prevent overflow in 'size *= 2'
// break;
}
#else
// Cooley-Tukey decimation-in-time radix-2 FFT
for (size = 2; size <= n; size *= 2) {
uint16_t halfsize = size / 2;
uint16_t tablestep = n / size;
for (i = 0; i < n; i += size) {
uint16_t j, k;
for (j = i, k = 0; j < i + halfsize; j++, k += tablestep) {
uint16_t l = j + halfsize;
float c = cos((2 * VNA_PI / 256) * k);
float s = sin((2 * VNA_PI / 256) * k);
float tpre = array[l][real] * c + array[l][imag] * s;
float tpim = -array[l][real] * s + array[l][imag] * c;
array[l][real] = array[j][real] - tpre;
array[l][imag] = array[j][imag] - tpim;
array[j][real] += tpre;
array[j][imag] += tpim;
}
}
// if (size == n) // Prevent overflow in 'size *= 2'
// break;
}
#endif
}
static inline void fft256_forward(float array[][2]) {