[Bug 14717] resampled sound is horrible

http://bugs.winehq.org/show_bug.cgi?id=14717

--- Comment #153 from Raymond  2011-02-22 19:54:19 CST ---
(In reply to comment #151)
> DSound may be not designed for playing high-quality music, but even the
> highest-quality resampler in the patch easily fits the latency budget that you
> imply by saying "dsound has to use small period size (I guess is 10ms) and try
> to mix the new sound effect and play them out as soon as possible".
>
> As for the acceptable tradeoffs in terms of CPU/latency/quality, I think they
> need to be discussed, otherwise someone will be able to say "mixing must be as
> fast as possible, so let's convert anything to 8 bits for performance reasons"
> (and that's in the same realm of strawman arguments as your suggestion to use a
> zero-order-hold resampler). My viewpoint is that wine either must provide the
> same resampling quality & latency as Windows does for DirectSound apps, or use
> the system's mixer/resampler and thus have the same characteristics as linux
> sound apps.

The main problem is dsound did not correctly caclulated the number of period ,
period size and buffer size

You can see the above examples, some sound cards have 128K byte , 64K bytes or
32 Kbytes Buffer,

when winealsa.drv try to set a buffer time of 0.5 seconds for 44100 S16 stereo,
it will get 371ms buffer for those sound cards with 64K Bytes buffer , but when
you will get a different buffer times when the application select a different
rate or U8 or mono

since oss use a power of two frag size, wineoss.drv is usually better than
winealsa.drv when using emulation

bug#8668

if you search wine-developer mailing list archive, you will find many wine
developer had said they will fix this issue

http://www.intel.com/support/motherboards/desktop/sb/cs-020642.htm#multistre...

you can also use dsound in foobar2000 to play audio to the front panel
headphone if your hda-intel codec have the extra DAC for multistream playback

Take a look at dsound/primary.c

DWORD DSOUND_fraglen(DWORD nSamplesPerSec, DWORD nBlockAlign)
{
    /* Given a timer delay of 10ms, the fragment size is approximately:
     *     fraglen = (nSamplesPerSec * 10 / 1000) * nBlockAlign
     * ==> fraglen = (nSamplesPerSec / 100) * nBlockSize
     *
     * ALSA uses buffers that are powers of 2. Because of this, fraglen
     * is rounded up to the nearest power of 2:
     */

if (nSamplesPerSec <= 12800)
        return 128 * nBlockAlign;

if (nSamplesPerSec <= 25600)
        return 256 * nBlockAlign;

if (nSamplesPerSec <= 51200)
        return 512 * nBlockAlign;

return 1024 * nBlockAlign;
}

static void DSOUND_RecalcPrimary(DirectSoundDevice *device)
{
    TRACE("(%p)\n", device);

device->helfrags = device->buflen / device->fraglen;
    TRACE("fraglen=%d helfrags=%d\n", device->fraglen, device->helfrags);

if (device->hwbuf && device->drvdesc.dwFlags & DSDDESC_DONTNEEDWRITELEAD)
        device->writelead = 0;
    else
        /* calculate the 10ms write lead */
        device->writelead = (device->pwfx->nSamplesPerSec / 100) *
device->pwfx->nBlockAlign;
}