Age | Commit message (Collapse) | Author |
|
CVS patchset: 8770
CVS date: 2007/03/29 18:58:21
|
|
height to uint16_t (they are read as 12-bit values).
CVS patchset: 8769
CVS date: 2007/03/29 18:52:45
|
|
CVS patchset: 8768
CVS date: 2007/03/29 18:41:02
|
|
Some tests were rearranged (AC_TRY_LINK moved out of AC_TRY_RUN
cross-compilation case, with arrangements to maintain the semantics) to
avoid autoupdate errors.
Factored out common (duplicate) code in aa.m4.
Tested with autoconf 2.61; needs testing with 2.60.
CVS patchset: 8767
CVS date: 2007/03/29 18:23:36
|
|
it on instance disposal.
CVS patchset: 8766
CVS date: 2007/03/29 18:00:23
|
|
CVS patchset: 8765
CVS date: 2007/03/29 17:59:35
|
|
reducing the holes.
CVS patchset: 8764
CVS date: 2007/03/29 17:48:34
|
|
inline the ff_video_decoder_t struct.
CVS patchset: 8763
CVS date: 2007/03/29 17:31:43
|
|
CVS patchset: 8762
CVS date: 2007/03/29 17:11:36
|
|
CVS patchset: 8761
CVS date: 2007/03/29 17:03:06
|
|
make seek_flag a 1-bit field. This reduces as much as possible the padding. Note that block_align should remain on the same cacheline as the fifo pointers, as it's used at every chunk, while the rest of the fields are used only when seeking.
CVS patchset: 8760
CVS date: 2007/03/29 17:00:32
|
|
CVS patchset: 8759
CVS date: 2007/03/29 16:52:23
|
|
variables fall into the same cacheline, while the header can simply remain on a different one, as it's needed only at the begin of the encoding.
CVS patchset: 8758
CVS date: 2007/03/29 16:46:23
|
|
CVS patchset: 8757
CVS date: 2007/03/29 16:32:12
|
|
CVS patchset: 8756
CVS date: 2007/03/29 16:21:48
|
|
plugin, only a subset of the information is needed, so hide the rest in an anonymous structure inside the parsing function. This reduces the header structure size to 24 rather than 40 (of the version already reduced by using bitmasks) and allows it to stay in a single cacheline.
Also, don't consider those flags that we don't use; leave the code in place so that if they need to be parsed in the future they can, but under #if 0 so that it's not compiled.
Move the tables for mp3's magic constants inside the parsing function, to hide them from the global namespace, and resize them to use uint16_t rather than int.
Use memcpy rather than copying the bytes one by one for the xing header.
CVS patchset: 8755
CVS date: 2007/03/29 16:01:30
|
|
bit, padding or no padding.
CVS patchset: 8754
CVS date: 2007/03/26 21:52:12
|
|
as boolean, and one int for just two LSBs, use a 1/1/2 bitmask, saves 11 bytes of memory (and probably also cpu instructions as there is almost no arithmetic done with it, but just logic that can be easily used bitwise.
CVS patchset: 8753
CVS date: 2007/03/26 21:18:52
|
|
CVS patchset: 8752
CVS date: 2007/03/26 21:06:32
|
|
CVS patchset: 8751
CVS date: 2007/03/26 14:21:10
|
|
CVS patchset: 8750
CVS date: 2007/03/26 12:52:05
|
|
Nissl. Also bump the libtool version info.
CVS patchset: 8749
CVS date: 2007/03/26 11:48:00
|
|
thus reported as 0×0), to avoid disabling shared memory for all others. Patch by Reinhard Nissl.
CVS patchset: 8748
CVS date: 2007/03/25 23:13:53
|
|
CVS patchset: 8747
CVS date: 2007/03/25 23:09:42
|
|
CVS patchset: 8746
CVS date: 2007/03/25 23:07:23
|
|
get loaded into shared memory for copy-on-write.
CVS patchset: 8745
CVS date: 2007/03/24 03:14:15
|
|
visibility but not protected (as the default is actually kinda like protected), thanks to Matt Messier in bug #1686194 for pointing me at that.
CVS patchset: 8744
CVS date: 2007/03/24 02:36:52
|
|
libvcdinfo 0.7.23 closes the device in this case whereas the built-in version
did not, and this was causing double frees due to my previous fix which
addressed only the built-in version (which needs to be either updated to 0.7.23
or dropped).
CVS patchset: 8743
CVS date: 2007/03/23 21:47:31
|
|
macro ASMALIGN(ZEROBITS) in ffmpeg's code, define it directly at configure, this way it can be used for the planar postplugin, that will then build with Apple's AS.
CVS patchset: 8742
CVS date: 2007/03/22 23:32:05
|
|
CVS patchset: 8741
CVS date: 2007/03/22 22:34:28
|
|
than split in two. Check for GNU as (macro adapted from libtool's check for GNU ld), as for instance Apple's assembler is not going to compile the Win32 wrappers. When GNU as is not found and user hasn't requested win32 codecs support explicitely, skip over it; if user requested win32 codecs and GNU as is not found, bail out.
CVS patchset: 8740
CVS date: 2007/03/22 22:31:48
|
|
using hacksaround. Also replace the whole pthread check with an improved macro originally written for XCB, this way it's not going to try linking the fake -lpthread on Darwin, and it also does not force a -I/usr/local/include on FreeBSD. The new macro respects the same variables set by ports, so that it's even more transparent to FreeBSD users.
CVS patchset: 8739
CVS date: 2007/03/22 20:44:58
|
|
together, and that the programs are checked. This way all the compilers and flags variables are grouped together on the --help output.
CVS patchset: 8738
CVS date: 2007/03/22 18:21:16
|
|
change their class to xineinclude, and set xineincludedir in Makefile.common.
CVS patchset: 8737
CVS date: 2007/03/22 17:37:17
|
|
decide to push back to support 2.59 again, then the AC_PREREQ should be changed too.
Leaving 2.53 is out of question for sure.
CVS patchset: 8736
CVS date: 2007/03/22 14:32:56
|
|
CVS patchset: 8735
CVS date: 2007/03/20 14:10:49
|
|
support seems to still not work there.
CVS patchset: 8734
CVS date: 2007/03/20 01:40:34
|
|
CVS patchset: 8733
CVS date: 2007/03/20 01:34:08
|
|
CVS patchset: 8732
CVS date: 2007/03/20 01:25:03
|
|
Emanuele Giaquinta for pointing me at the right LDFLAG.
CVS patchset: 8731
CVS date: 2007/03/20 01:18:51
|
|
CVS patchset: 8730
CVS date: 2007/03/20 01:13:30
|
|
and macosx plugins. Also pass the VISIBILITY_FLAG to the correct set of OBJCFLAGS variable (even if it's a moot point, OS X has no visibility support as it is). Patch by Martin Aumueller.
CVS patchset: 8729
CVS date: 2007/03/20 00:47:04
|
|
CVS patchset: 8728
CVS date: 2007/03/20 00:46:06
|
|
CVS as needs to be submitted.
CVS patchset: 8727
CVS date: 2007/03/20 00:35:29
|
|
aligned available as 64 even on Darwin where the maximum is 16, as it does not use -Werror, and the wrong alignment is considered only a warning. Rewrite it as CC_ATTRIBUTE_ALIGNED (in attributes.m4, where other similar macros are located). Also reverse the order of the checks, so that the highest version is tried first; this way you can avoid five compile tests on systems where 64 is the highest alignment available.
CVS patchset: 8726
CVS date: 2007/03/20 00:30:55
|
|
CVS patchset: 8725
CVS date: 2007/03/20 00:01:19
|
|
CVS patchset: 8724
CVS date: 2007/03/19 23:55:00
|
|
based on Intel CPUs). Patch by Martin Aumueller.
CVS patchset: 8723
CVS date: 2007/03/19 23:48:58
|
|
used to.
CVS patchset: 8722
CVS date: 2007/03/19 23:41:49
|
|
enabled, as this one caches its tests, so a run using cached values will be faster.
CVS patchset: 8721
CVS date: 2007/03/19 23:41:05
|