sokol: sokol_audio ALSA: buffer size initialization problem on some Linux configs

In function _saudio_backend_init(), snd_pcm_hw_params_set_rate_near() return negative result.

Reproduced on two different computers. Both run Arch current (x86_64):

Card: HDA Intel, Chip: Analog Devices AD1989B
Card: HDA Intel PCH, Chip: Conexant CX20751/2

The following patch fixes sound (saudio-app and modplay-app work), but I’m not sure that it doesn’t hammer something else.

diff --git a/modules/sokol/sokol_audio.h b/modules/sokol/sokol_audio.h
index bf7e4a7..ebafca2 100644
--- a/modules/sokol/sokol_audio.h
+++ b/modules/sokol/sokol_audio.h
@@ -1044,7 +1044,6 @@ _SOKOL_PRIVATE bool _saudio_backend_init(void) {
     snd_pcm_hw_params_any(_saudio.backend.device, params);
     snd_pcm_hw_params_set_access(_saudio.backend.device, params, SND_PCM_ACCESS_RW_INTERLEAVED);
     snd_pcm_hw_params_set_channels(_saudio.backend.device, params, _saudio.num_channels);
-    snd_pcm_hw_params_set_buffer_size(_saudio.backend.device, params, _saudio.buffer_frames);
     if (0 > snd_pcm_hw_params_test_format(_saudio.backend.device, params, SND_PCM_FORMAT_FLOAT_LE)) {
         goto error;
     }
@@ -1056,6 +1055,12 @@ _SOKOL_PRIVATE bool _saudio_backend_init(void) {
     if (0 > snd_pcm_hw_params_set_rate_near(_saudio.backend.device, params, &val, &dir)) {
         goto error;
     }
+    snd_pcm_uframes_t periodsize = _saudio.buffer_frames * 2;
+    rc = snd_pcm_hw_params_set_buffer_size_near(_saudio.backend.device, params, &periodsize);
+    if(rc < 0) {
+        goto error;
+    }
+
     if (0 > snd_pcm_hw_params(_saudio.backend.device, params)) {
         goto error;
     }

I.e. we set rate first, then try to set buffer size. The code was copied from alsa samples. I’m not sure, if we must change anything else (period_size).

P.S. It would be nice, if samples report sound initializion failure somehow. With red screen for example.

About this issue

Original URL
State: open
Created 4 years ago
Comments: 19 (13 by maintainers)

Most upvoted comments

…another more robust solution is to rewrite the ring buffer code so that the audio streaming buffer size doesn’t need to be a multiple of the ring-buffer packet size… I just remembered that I did something similar in the plmpeg.h sokol-sample… This would allow sokol_audio.h to work with odd buffer sizes (like 1881 frames we’ve seen here).

I need to think this through a bit, and it will be a little while until I can take care of this.

Yes, to this. The AAudio backend could benefit from being able to take a lower latency path where the callback can be called with different values (even within one audio session) depending on the characteristics of the underlying audio device.

oviano on Feb 21, 2021

Alsamixer.

Conexant is on Lenovo IdeaPad 100. “Analog Devices AD1989B” on desktop with some ASUS motherboard.

xglabs on Oct 13, 2020

I’ve looked at openal-soft ALSA initialization code and borrowed one function call, that you are not using. With this patch, everything works. But I don’t understand why…

diff --git a/sokol_audio.h b/sokol_audio.h
index 8df5813..897526a 100644
--- a/sokol_audio.h
+++ b/sokol_audio.h
@@ -1083,6 +1083,11 @@ _SOKOL_PRIVATE bool _saudio_backend_init(void) {
         SOKOL_LOG("sokol_audio.h: requested channel count not supported");
         goto error;
     }
+    /* disable resampler */
+    if(snd_pcm_hw_params_set_rate_resample(_saudio.backend.device, params, 0) < 0) {
+        SOKOL_LOG("set_rate_resample failed");
+        goto error;
+    }
     /* let ALSA pick a nearby sampling rate */
     rate = _saudio.sample_rate;
     dir = 0;

xglabs on Oct 12, 2020