Hi,
2014-02-10 9:03 GMT+01:00 Kostya Shishkov kostya.shish...@gmail.com:
What you suggest is still valid, and would catch things even if they
changed, though.
Maybe it's not worth warning about it at all.
Well, the format allows 8N coefficients with N=1, so a file could
theoretically contain
On Mon, Feb 10, 2014 at 08:39:59AM +0100, Christophe Gisquet wrote:
Hi,
2014-02-10 6:42 GMT+01:00 Kostya Shishkov kostya.shish...@gmail.com:
On Sun, Feb 09, 2014 at 11:21:33PM +0100, Christophe Gisquet wrote:
+if(s-cdlms[c][i].order 8) {
+static int warned;
Hi
since then, it appears it is simpler to extend the buffer, and
whenever the order change, pad the end of the coeff buffer with 0.
This way, no need to handle the tail of the loop.
As the case where an unusual order is encountered (it should still be
handled fine), a sample is requested.
The
On Sun, Feb 09, 2014 at 11:21:33PM +0100, Christophe Gisquet wrote:
Hi
since then, it appears it is simpler to extend the buffer, and
whenever the order change, pad the end of the coeff buffer with 0.
This way, no need to handle the tail of the loop.
As the case where an unusual order is
Hi,
2014-02-10 6:42 GMT+01:00 Kostya Shishkov kostya.shish...@gmail.com:
On Sun, Feb 09, 2014 at 11:21:33PM +0100, Christophe Gisquet wrote:
+if(s-cdlms[c][i].order 8) {
+static int warned;
+if(!warned)
+
2012/12/7 Christophe Gisquet christophe.gisq...@gmail.com:
All samples (stereo, 44100Hz) I have (found on samples.mplayerhq,
provided or self-encoded) seem to only use orders 16 and 64.
It seems any other channel count or samplerate requires 24bits,
currently not supported by the decoder.
On 12/24/2012 10:36 AM, Christophe Gisquet wrote:
But what the samples ever contained (files I generated myself or
provided) are loops of 16 or 64.
So we have multiple courses of action:
1) do nothing
2) use the current dsp code as is, fail on samples with a loop count
multiple of 8, and
hi,
2012/12/24 Luca Barbato lu_z...@gentoo.org:
What about having a fallback and request those samples?
That's possible, either wmalossless.c keeps a local copy of the C
code, or the C dsp function is accessible from it, and then the
function pointer is changed.
Not very clean, but here's a
On 12/24/2012 04:28 PM, Christophe Gisquet wrote:
hi,
2012/12/24 Luca Barbato lu_z...@gentoo.org:
What about having a fallback and request those samples?
That's possible, either wmalossless.c keeps a local copy of the C
code, or the C dsp function is accessible from it, and then the
2012/11/30 Christophe Gisquet christophe.gisq...@gmail.com:
Moved to a macro. This was mimicking apedec codec.
If the patches are declared ok, someone should check the neon code.
All samples (stereo, 44100Hz) I have (found on samples.mplayerhq,
provided or self-encoded) seem to only use orders
2012/11/27 Justin Ruggles justin.rugg...@gmail.com:
Either make the function name lowercase or make it a macro. Also, if you
leave it as an inline function, put the opening brace on a separate line.
Moved to a macro. This was mimicking apedec codec.
If the patches are declared ok, someone
On 11/24/2012 09:58 AM, Christophe Gisquet wrote:
Hi,
2012/3/4 Ronald S. Bultje rsbul...@gmail.com:
Since you're working on this, would you mind adding one (or a few)
fate tests so we can ensure that we don't break this while we're
optimizing it?
I have no idea how to do that, and I
On Sat, Nov 24, 2012 at 03:58:37PM +0100, Christophe Gisquet wrote:
2012/3/4 Ronald S. Bultje rsbul...@gmail.com:
Since you're working on this, would you mind adding one (or a few)
fate tests so we can ensure that we don't break this while we're
optimizing it?
I have no idea how to do
2012/11/25 Diego Biurrun di...@biurrun.de:
It's really not that hard to do. First we need a sample that exercises
the code in question. Do you have one or is one already in the collection?
In fact, checking different fate files, I came upon:
fate-lossless-wma
which should exercise it. Though
On Sun, Nov 25, 2012 at 02:54:45PM +0100, Christophe Gisquet wrote:
2012/11/25 Diego Biurrun di...@biurrun.de:
It's really not that hard to do. First we need a sample that exercises
the code in question. Do you have one or is one already in the collection?
In fact, checking different
Hi,
If, specifically, it is a test for scalarproduct_and_madd_int16 that
is asked, then this means starting to test all of dsputil.
Anyway, I am no longer interested in working on that fate stuff.
Christophe
___
libav-devel mailing list
On Sun, Nov 25, 2012 at 05:56:38PM +0100, Christophe Gisquet wrote:
If, specifically, it is a test for scalarproduct_and_madd_int16 that
is asked, then this means starting to test all of dsputil.
No, that is not what is asked. FATE is not a unit testing framework
that exercises specific
2012/11/25 Diego Biurrun di...@biurrun.de:
No, that is not what is asked. FATE is not a unit testing framework
that exercises specific functions in isolation. It's more of a holistic
approach that exercises specific samples completely.
I was looking at the dct/fft test and didn't feel like
On Sun, Nov 25, 2012 at 06:10:03PM +0100, Christophe Gisquet wrote:
2012/11/25 Diego Biurrun di...@biurrun.de:
No, that is not what is asked. FATE is not a unit testing framework
that exercises specific functions in isolation. It's more of a holistic
approach that exercises specific
2012/11/25 Diego Biurrun di...@biurrun.de:
On Sun, Nov 25, 2012 at 06:10:03PM +0100, Christophe Gisquet wrote:
Then we're fine already. If you want to, you could double-check by
intentionally breaking your optimization and confirming that fate
complains.
Breaking the optimization for the
On Sun, Nov 25, 2012 at 07:22:06PM +0100, Christophe Gisquet wrote:
2012/11/25 Diego Biurrun di...@biurrun.de:
On Sun, Nov 25, 2012 at 06:10:03PM +0100, Christophe Gisquet wrote:
Then we're fine already. If you want to, you could double-check by
intentionally breaking your optimization and
Hi,
2012/3/4 Ronald S. Bultje rsbul...@gmail.com:
Since you're working on this, would you mind adding one (or a few)
fate tests so we can ensure that we don't break this while we're
optimizing it?
I have no idea how to do that, and I would probably need several times
the time it takes for
2012/3/3 Christophe Gisquet christophe.gisq...@gmail.com:
add a tail to handle the remaining 8 elements to the current code.
This is not correctly implemented (for lack of file to evaluate it),
so please hold off any review.
Christophe
___
libav-devel
Hi,
On Sun, Mar 4, 2012 at 9:12 AM, Christophe Gisquet
christophe.gisq...@gmail.com wrote:
2012/3/3 Christophe Gisquet christophe.gisq...@gmail.com:
add a tail to handle the remaining 8 elements to the current code.
This is not correctly implemented (for lack of file to evaluate it),
so
Hi,
after some investigation, it seems wma lossless can almost reuse
scalarproduct_and_madd_int16, also used by apedec.c
The only requirement is that multiple-of-8 numbers of loop must be
handled. After discussing with Kostya, it seems the best option (eg
because of very large number of loops)
25 matches
Mail list logo
