this repo has no description
bench
+94
-2
+94
-2
src/ext/cqt.c
+94
-2
src/ext/cqt.c
···
8
8
#include <string.h>
9
9
#include <stdbool.h>
10
10
#include <stdio.h>
11
+
#include <time.h>
11
12
12
13
#define CQT_DEBUG
13
14
···
16
17
static float* cqtAudioBuffer = NULL;
17
18
static kiss_fft_cpx* cqtFftOutput = NULL;
18
19
20
+
// Benchmark different FFT sizes
21
+
static void CQT_BenchmarkFFT(void)
22
+
{
23
+
printf("\nCQT FFT Benchmark on this CPU:\n");
24
+
printf("================================\n");
25
+
26
+
int sizes[] = {4096, 6144, 8192, 12288, 16384};
27
+
int numSizes = 5;
28
+
29
+
for (int s = 0; s < numSizes; s++)
30
+
{
31
+
int fftSize = sizes[s];
32
+
33
+
// Allocate buffers
34
+
float* testInput = (float*)calloc(fftSize, sizeof(float));
35
+
kiss_fft_cpx* testOutput = (kiss_fft_cpx*)malloc((fftSize/2 + 1) * sizeof(kiss_fft_cpx));
36
+
kiss_fftr_cfg testCfg = kiss_fftr_alloc(fftSize, 0, NULL, NULL);
37
+
38
+
if (!testInput || !testOutput || !testCfg)
39
+
{
40
+
printf("Failed to allocate for size %d\n", fftSize);
41
+
continue;
42
+
}
43
+
44
+
// Fill with test signal
45
+
for (int i = 0; i < fftSize; i++)
46
+
{
47
+
testInput[i] = sin(2.0 * M_PI * 440.0 * i / 44100.0);
48
+
}
49
+
50
+
// Warm up
51
+
for (int i = 0; i < 10; i++)
52
+
{
53
+
kiss_fftr(testCfg, testInput, testOutput);
54
+
}
55
+
56
+
// Time 100 iterations
57
+
clock_t start = clock();
58
+
for (int i = 0; i < 100; i++)
59
+
{
60
+
kiss_fftr(testCfg, testInput, testOutput);
61
+
}
62
+
clock_t end = clock();
63
+
64
+
double avgTime = (double)(end - start) / CLOCKS_PER_SEC * 1000.0 / 100.0;
65
+
printf("%5d-point FFT: %.3f ms\n", fftSize, avgTime);
66
+
67
+
// Cleanup
68
+
free(testInput);
69
+
free(testOutput);
70
+
free(testCfg);
71
+
}
72
+
73
+
printf("================================\n\n");
74
+
}
75
+
19
76
// Initialize CQT processing
20
77
bool CQT_Open(void)
21
78
{
···
56
113
{
57
114
CQT_Close();
58
115
return false;
116
+
}
117
+
118
+
// Run FFT benchmark on first initialization
119
+
static bool benchmarkRun = false;
120
+
if (!benchmarkRun)
121
+
{
122
+
CQT_BenchmarkFFT();
123
+
benchmarkRun = true;
59
124
}
60
125
61
126
// Debug: Print first few center frequencies and expected FFT bins
···
172
237
return;
173
238
}
174
239
175
-
// Perform 4096-point FFT
240
+
// Profiling variables
241
+
static double totalFftTime = 0.0;
242
+
static double totalKernelTime = 0.0;
243
+
static int profileCount = 0;
244
+
245
+
// Perform 6144-point FFT with timing
246
+
clock_t fftStart = clock();
176
247
kiss_fftr(cqtFftCfg, cqtAudioBuffer, cqtFftOutput);
248
+
clock_t fftEnd = clock();
177
249
178
250
// Extract real and imaginary components for kernel application
179
251
float fftReal[CQT_FFT_SIZE/2 + 1];
···
185
257
fftImag[i] = cqtFftOutput[i].i;
186
258
}
187
259
188
-
// Apply CQT kernels
260
+
// Apply CQT kernels with timing
261
+
clock_t kernelStart = clock();
189
262
CQT_ApplyKernels(fftReal, fftImag);
263
+
clock_t kernelEnd = clock();
264
+
265
+
// Calculate times in milliseconds
266
+
double fftTime = (double)(fftEnd - fftStart) / CLOCKS_PER_SEC * 1000.0;
267
+
double kernelTime = (double)(kernelEnd - kernelStart) / CLOCKS_PER_SEC * 1000.0;
268
+
269
+
totalFftTime += fftTime;
270
+
totalKernelTime += kernelTime;
271
+
profileCount++;
272
+
273
+
// Print profiling info every 60 frames (~1 second)
274
+
if (profileCount % 60 == 0)
275
+
{
276
+
printf("CQT Performance (6K FFT):\n");
277
+
printf(" FFT avg: %.3fms\n", totalFftTime / profileCount);
278
+
printf(" Kernels avg: %.3fms\n", totalKernelTime / profileCount);
279
+
printf(" Total avg: %.3fms\n", (totalFftTime + totalKernelTime) / profileCount);
280
+
printf(" Samples: %d\n\n", profileCount);
281
+
}
190
282
191
283
#ifdef CQT_DEBUG
192
284
// Reduced debug output - CQT is working correctly now