Re: [PATCH v7] audio/pwaudio.c: Add Pipewire audio backend for QEMU

[Volker Rümelin]

Am 13.03.23 um 14:11 schrieb Dorinda Bassey:
> Hi Volker,
>
>     To hear this,
>     start QEMU with qemu-system-x86_64 -machine pcspk-audiodev=audio0
>     -device ich9-intel-hda -device hda-duplex,audiodev=audio0 -audiodev
>     pipewire,id=audio0,out.mixing-engine=off ...
>
> I hear the clipped audio stream with these options. IMO, I don't think 
> memset is responsible for that behaviour, I still hear the harsh sound 
> with "-audiodev pa". I also tried using an alternative like:

Hi Dorinda,

when you test -audiodev pa with a pulseaudio server, audio playback is 
fine. Audio playback with -audiodev pa with the pipewire-pulse server is 
clipped. This is a pipewire bug.

With best regards,
Volker

> @@ -117,7 +118,7 @@ playback_on_process(void *data)
>      }
>
>      if (avail == 0) {
> -        memset(p, 0, n_bytes);
> +        p = g_malloc0(sizeof(n_bytes));
>      } else {
>
> The clipped audio issue is still persistent.
>
> Thanks,
> Dorinda.
>
> On Sun, Mar 12, 2023 at 9:01 AM Volker Rümelin <vr_qemu@t-online.de> 
> wrote:
>
>     > +/* output data processing function to read stuffs from the
>     buffer */
>     > +static void
>     > +playback_on_process(void *data)
>     > +{
>     > +    PWVoice *v = (PWVoice *) data;
>     > +    void *p;
>     > +    struct pw_buffer *b;
>     > +    struct spa_buffer *buf;
>     > +    uint32_t n_frames, req, index, n_bytes;
>     > +    int32_t avail;
>     > +
>     > +    if (!v->stream) {
>     > +        return;
>     > +    }
>     > +
>     > +    /* obtain a buffer to read from */
>     > +    b = pw_stream_dequeue_buffer(v->stream);
>     > +    if (b == NULL) {
>     > +        error_report("out of buffers: %s", strerror(errno));
>     > +        return;
>     > +    }
>     > +
>     > +    buf = b->buffer;
>     > +    p = buf->datas[0].data;
>     > +    if (p == NULL) {
>     > +        return;
>     > +    }
>     > +    req = b->requested * v->frame_size;
>     > +    if (req == 0) {
>     > +        req = 4096 * v->frame_size;
>     > +    }
>     > +    n_frames = SPA_MIN(req, buf->datas[0].maxsize);
>     > +    n_bytes = n_frames * v->frame_size;
>     > +
>     > +    /* get no of available bytes to read data from buffer */
>     > +
>     > +    avail = spa_ringbuffer_get_read_index(&v->ring, &index);
>     > +
>     > +    if (!v->enabled) {
>     > +        avail = 0;
>     > +    }
>     > +
>     > +    if (avail == 0) {
>     > +        memset(p, 0, n_bytes);
>
>     memset() doesn't work for unsigned samples. For unsigned samples, a
>     stream of zeros is silence with a DC offset. When Pipewire mixes this
>     stream with another, the result is a clipped audio stream. To hear
>     this,
>     start QEMU with qemu-system-x86_64 -machine pcspk-audiodev=audio0
>     -device ich9-intel-hda -device hda-duplex,audiodev=audio0 -audiodev
>     pipewire,id=audio0,out.mixing-engine=off ... and start playback
>     with the
>     hda device.
>
>     With best regards,
>     Volker
>
>     > +    } else {
>     > +        if (avail < (int32_t) n_bytes) {
>     > +            n_bytes = avail;
>     > +        }
>     > +
>     > +        spa_ringbuffer_read_data(&v->ring,
>     > +                                    v->buffer, RINGBUFFER_SIZE,
>     > +                                    index & RINGBUFFER_MASK, p,
>     n_bytes);
>     > +
>     > +        index += n_bytes;
>     > +        spa_ringbuffer_read_update(&v->ring, index);
>     > +    }
>     > +
>     > +    buf->datas[0].chunk->offset = 0;
>     > +    buf->datas[0].chunk->stride = v->frame_size;
>     > +    buf->datas[0].chunk->size = n_bytes;
>     > +
>     > +    /* queue the buffer for playback */
>     > +    pw_stream_queue_buffer(v->stream, b);
>     > +}
>     > +
>     >