[pulseaudio-discuss] [RFC] [PATCH] tests: Add a latency measurement test

[Peter Meerwald]

Hello, 

> This test is intended to measure real latency by playing a sample to a
> sink and capturing that over a loopback interface. The loopback can
> either be physical (cable running from headphone out to line in) or
> virtual (monitor source or module loopback).

nice idea :)

the sine tone in the beginning has chirps, intended?

the program's output could be clearer; it is not obvious what is going on 
('giving up' but the program does not stop?):

Capture signal too weak at 100% volume (0.766978). Giving up.
Capture signal too weak at 100% volume (0.748493). Giving up.
Capture signal too weak at 100% volume (0.797834). Giving up.
Underflow
Capture signal too weak at 100% volume (0.418111). Giving up.
Too much noise on capture (0.606354). Giving up.
Latency 10565
Latency 10678
Latency 10709


+#define TONE_HZ SAMPLE_HZ / 100

maybe #define TONE_HZ (SAMPLE_HZ / 100)

+#define PLAYBACK_LATENCY 25 /* ms */
+#define CAPTURE_LATENCY 5 /* ms */

> +static const char *bname = NULL;

bname is not very clear; what is it for?

> +static float out[N_OUT][CHANNELS];
> +static int ppos = 0;
> +
> +static int n_underflow = 0;
> +static int n_overflow = 0;

unsigned maybe?

> +static inline float rms(const float *s, int n) {
> +    float sq = 0;
> +    int i;

float sq = 0.0f;
unsigned i, unsigned n?

> +
> +    for (i = 0; i < n; i++)
> +        sq += s[i] * s[i];
> +
> +    return sqrt(sq / n);

should be sqrtf() since sq is float

> +#define WINDOW 2 * CHANNELS

#define WINDOW (2 * CHANNELS)

> +static void read_cb(pa_stream *s, size_t nbytes, void *userdata) {
> +    static float last = 0.0;

static float last = 0.0f;

> +    const float *in;
> +    float cur;
> +    int r;
> +    unsigned int i = 0;

here we have an unsigned?

> +        /* We leave the definition of 0 generous since the window might
> +         * straddle the 0->1 transition, raising the average power. We keep the
> +         * definition of 1 tight in this case and detect the transition in the
> +         * next round. */
> +        if (last < 0.5 && cur > 0.8) {

I'd append an f to denote float precision, so 0.5f and 0.8f

> +            pa_gettimeofday(&tv_in);
> +            fprintf(stderr, "Latency %llu\n", (unsigned long long) pa_timeval_diff(&tv_in, &tv_out));

is the cast necessary?

> +enum {
> +    CALIBRATION_ONE,
> +    CALIBRATION_ZERO,
> +    CALIBRATION_DONE,
> +};
> +
> +static int cal_state = CALIBRATION_ONE;
> +
> +static void calibrate_write_cb(pa_stream *s, size_t nbytes, void *userdata) {
> +    int i, r, nsamp = nbytes / fs;
> +    float tmp[nsamp][2];
> +    static int count = 0;
> +
> +    /* Write out a sine tone */
> +    for (i = 0; i < nsamp; i++)
> +        tmp[i][0] = tmp[i][1] = cal_state == CALIBRATION_ONE ? sin(count++ * TONE_HZ * 2 * M_PI / SAMPLE_HZ) : 0.0;

sinf() since float, 0.0f


> +START_TEST (loopback_test) {
> +    pa_mainloop* m = NULL;
> +    int i, ret = 0, pulse_hz = N_OUT / 1000;

division rounds towards zero, intended?

> +
> +    /* Generate a square pulse */
> +    for (i = 0; i < N_OUT; i++)
> +        if (i < pulse_hz)
> +            out[i][0] = out[i][1] = 1.0;
> +        else
> +            out[i][0] = out[i][1] = 0.0;

-- 

Peter Meerwald
+43-664-2444418 (mobile)