On Wed, Oct 12, 2005 at 01:06:59PM -0400, Paul Davis wrote:
> > The UA-25 uses S24_3LE. See the link for a jack patch that kindof swaps
> > the bytes, but it leaves byte 4 empty. Hence samples loose their sign plus
> > more.
> > Shouldn't the jack buffer convert the samples to a full 32 bit value?
> > How are applications suposed to know how many bits are valid in the
> > jack_port_buffer?
>
> that is a meaningless question. all audio within the JACK graph is 32
> bit float audio. there is no access to h/w-native data formats from
> within the graph: only the backend has access to them, and that is where
> the patch takes effect.
Your answer to my first question makes my second one void indeed. Still
thanks for answering my first one :)
> if the patch does not work correctly, we need to know (as does its
> author).
Well, I'm still working on a final fix. What I have attached corrects the
buffer for the S24_3LE mode. Before this fix -1 from the device (0xffffff)
became 0x00ffffff (large positive number) in the buffer. With the patch
it is 0xffffffff, i.e. -1 as it should be.
I created a very similar patch for 'arecord -vv', which is in alsa cvs now.
I'll post a final jack fix to jackit-devel, and probably append it to the
sf entry. But there's still plenty of testing to do before that.
--
Martin
--- memops.c.patched 2005-10-10 23:33:02.000000000 +0100
+++ memops.c 2005-10-12 18:01:55.000000000 +0100
@@ -30,6 +30,7 @@
#include <memory.h>
#include <stdlib.h>
#include <limits.h>
+#include <endian.h>
#include <jack/memops.h>
@@ -443,16 +444,22 @@
x |= (unsigned char)(src[1]);
x <<= 8;
x |= (unsigned char)(src[2]);
- x <<= 8;
+ /* Correct sign bit and the rest of the top byte */
+ if (src[0] & 0x80) {
+ x |= 0xff << 24;
+ }
#elif __BYTE_ORDER == __BIG_ENDIAN
x = (unsigned char)(src[2]);
x <<= 8;
x |= (unsigned char)(src[1]);
x <<= 8;
x |= (unsigned char)(src[0]);
- x <<= 8;
+ /* Correct sign bit and the rest of the top byte */
+ if (src[2] & 0x80) {
+ x |= 0xff << 24;
+ }
#endif
- *dst = (x >> 8) / SAMPLE_MAX_24BIT;
+ *dst = x / SAMPLE_MAX_24BIT;
dst++;
src += src_skip;
}
