Re: [PATCH] pcm: add new 32-bit DSD sample format

[Jurgen Kramer <gtmkramer@xs4all.nl>]

On Thu, 2014-11-20 at 10:58 +0100, Takashi Iwai wrote:
> At Wed, 19 Nov 2014 21:50:30 +0200,
> Jussi Laako wrote:
> > 
> > Hi,
> > 
> > Sorry for catching up late on this patch...
> > 
> > On 10.09.2014 10:00, Jurgen Kramer wrote:
> > > +	FORMATD(DSD_U32_LE, "Direct Stream Digital, 4-byte (x32), little endian, oldest bits in MSB"),
> > 
> > There is bug in this patch. The spec for these formats is, as above says 
> > "little endian, oldest bits in MSB", however with the iFi iDSD Nano this 
> > doesn't seem to be the case. Oldest bits are in MSB of the byte, but 
> > byte order is big endian.
> 
> Is it a bug of spec, or a bug of device?
> In the latter case, we'd need to introduce DSD_U32_BE format and apply
> a quirk, for example.
DSD is a bit odd wrt LE/BE. Both formats are allowed and used.
The device (XMOS based/Marantz+Denon) want the following order of
samples: 
L1 L2 L3 L4 R1 R2 R3 R4. Where Lx/Rx are 8-bit DSD samples, L1-L4 are
packed to 1 32-bit sample.

I do think real BE will not work.

Jurgen
> 
> Takashi
> 
> > IOW, following alone will produce incorrect 
> > output:
> > 
> > size_t N = 32;
> > uint32_t x = 0;
> > while (N--)
> > {
> > 	x <<= 1;
> > 	x |= nextbit;
> > }
> > 
> > But if you add __builtin_byteswap32(x) then the output is correct.
> > 
> > I guess this has not been caught because of just memcpy()'ing the 
> > oldest-bit-in-MSB byte stream to the sample buffer instead of 
> > constructing the words...
> > 
> > 
> > Best regards,
> > 
> > 	- Jussi
> > _______________________________________________
> > Alsa-devel mailing list
> > Alsa-devel@alsa-project.org
> > http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
> > 
> _______________________________________________
> Alsa-devel mailing list
> Alsa-devel@alsa-project.org
> http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
>