1. 25 Oct, 2000 1 commit
    • Michel Lespinasse's avatar
      · 7fadb3b1
      Michel Lespinasse authored
      --disable-shared is now the default
      (I wonder why this was not the case already ?????)
      walken, trying to learn the autotools
  2. 23 Oct, 2000 3 commits
    • Thomas Mirlacher's avatar
      added accel.h (as proposed by walken) · 22b34e55
      Thomas Mirlacher authored
      ... also added some cvs magic, to keep this file in sync (new mpeg2dec version
      is available as mpeg2dec.new - using the proper place in cvs ... and will be
      switched to be the standard once approved by walken)
    • Michel Lespinasse's avatar
      · e98337d5
      Michel Lespinasse authored
      added runtime selection of the SSE idct routine
    • Michel Lespinasse's avatar
      · 3150dbdf
      Michel Lespinasse authored
      Added runtime detection of MMX/3Dnow/SSE instructions.
      dent, you should probably include a similar scheme in oms... probably
      just copy my accel files to oms... maybe I should put them in a subdir
      if it helps with the cvs magic...
      I should also change the init functions so that the acceleration bitmap
      is passed at init time - so that the user can force different accelerations
      if he prefers. I havent done it yet for oms compatibility reasons but I'd
      like to do it soon.
      I'll add runtime execution of the SSE functions soon.
  3. 21 Oct, 2000 1 commit
    • Michel Lespinasse's avatar
      · 0bbda427
      Michel Lespinasse authored
      more scheduling updates :
      * mmx idct : scheduling between consecutive rows
      * sse idct : row routine uses a reordered input vector
      the mmx update does not improve the speed a lot but the sse one works
      very well.
  4. 20 Oct, 2000 1 commit
    • Michel Lespinasse's avatar
      · d30777d2
      Michel Lespinasse authored
      still more idct updates.
      * sse idct : better scheduling between consecutive row transforms
      * mmx idct : faster row transform, using a reordered input vector
  5. 19 Oct, 2000 3 commits
    • Michel Lespinasse's avatar
      · 71a1f72b
      Michel Lespinasse authored
      mmx idct column code : scheduling + precision enhancements
    • Michel Lespinasse's avatar
      · 6b01e55b
      Michel Lespinasse authored
      optimized the saturation/addition code that is done after the idct
    • Michel Lespinasse's avatar
      · 67437964
      Michel Lespinasse authored
      small mmx idct enhancements - playing around with the compiler to prevent
      it from generating suboptimal code
  6. 18 Oct, 2000 1 commit
    • Thomas Mirlacher's avatar
      overal slight cosmetic adjustments ... again - and not for the last time :) · bf5ebc83
      Thomas Mirlacher authored
      	moved fps display to 'the RIGHT place' (tm) ;)
      	won't be loaded if the kernel module isn't loaded (see this
      	for an auto-loading of modules) ... should be done for all
      	hardware dependent plugins - works for the ui also (great work
      	yoann :)
      	x11 is now the default output lpugins (i still hope people can
      	built x11 plugins easier than sdl plugins :) ... and on top of
      	them they're faster (with recent gfx cards and Xservers)
      .. noone expects the spanish inquisition ... (and oms to work :)
  7. 17 Oct, 2000 2 commits
  8. 16 Oct, 2000 1 commit
    • Thomas Mirlacher's avatar
      2000-10-16 Thomas Mirlacher <dent@linuxvideo.org> · c1c2b761
      Thomas Mirlacher authored
      * src/plugin/codec/mpeg2dec/dec_mpeg2.c: commented framedropping output,
      	so no more '+' or '-' is done by default
      * src/plugin/nav/nav_dvd.c: since the decoder is stable above, we don't
      	need to wait for a new gop, but can jump immediately
      * src/plugin/output/video/video_out_x11.c: made direct calls to proper
      	xv or x11 functions, without the need to have a wrapper function
      	added a _check_event function
      	check if window == NULL in the close function - pluginswitching
      	during runtime is now ALMOST possible (problem: setup call - we
      	need to tell the decoder to resetup the new plugin)
  9. 15 Oct, 2000 1 commit
  10. 13 Oct, 2000 4 commits
  11. 12 Oct, 2000 6 commits
    • Thomas Mirlacher's avatar
      added dynamic b-frame dropping (thanks to walken) · 52175c0a
      Thomas Mirlacher authored
      ... this is NOT for syncing - but rather the first level for dropping frames,
      so sound can be smooth - the check function within the audio plugin still has
      to be adjusted, and coded correctly to get really SMOOTH sound, but this is a
      first try.
    • Michel Lespinasse's avatar
      · 79a2d33d
      Michel Lespinasse authored
      two small changes to make things easier for OMS :
      * do an emms() call after using the mmx instructions
      * added a flag to drop b frames
      the drop flag is only meant as a temporary solution before we can implement
      a better logic but it will do okay for now
    • Michel Lespinasse's avatar
      · 98b7b8fd
      Michel Lespinasse authored
      Yet another small alignment improvement for mmx idct
    • Michel Lespinasse's avatar
      · 78689f13
      Michel Lespinasse authored
      Forgot to comment out the SSE #define. oops.
    • Michel Lespinasse's avatar
      · 1beec70c
      Michel Lespinasse authored
      minor mmx idct speedup
    • Michel Lespinasse's avatar
      · 7e02a1fc
      Michel Lespinasse authored
      More precise mmx IDCT. I had this code around for a long time but it was
      a 20% slowdown (with null output). But I found an alignment problem so
      now that I corrected it its only a 6% slowdown (3.5% for people who have
      sse). Also people using slower computers have reported that it was actually
      faster for them, though I'd like to get more feedback in that area.
      This brings our mmx IDCT well within the precision requirements of the
  12. 11 Oct, 2000 2 commits
    • Michel Lespinasse's avatar
      · 5bce7a5c
      Michel Lespinasse authored
      removed D pictures from TODO list
    • Michel Lespinasse's avatar
      · 4a7ef540
      Michel Lespinasse authored
      added support for mpeg1 D pictures
  13. 10 Oct, 2000 2 commits
    • Michel Lespinasse's avatar
      · 370138ed
      Michel Lespinasse authored
      Implemented DMV motion compensation. Now the only missing mpeg2 feature is
      the support for field pictures.
      The DMV code is not super-clean yet. I'll have to make another pass on it
      after we get feature-completeness.
    • Michel Lespinasse's avatar
      · 293ae0d7
      Michel Lespinasse authored
      very minor MV parsing optimization. I figured out we can do with a smaller
      table. The speed increase will be very small but heh :)
  14. 09 Oct, 2000 1 commit
    • Michel Lespinasse's avatar
      · 6539580c
      Michel Lespinasse authored
      added mpeg1 oddification - we now pass most mpeg1 conformance tests.
      I'm still looking for mpeg1 streams that use full-pixel MC to test them...
      we dont support that feature yet.
  15. 08 Oct, 2000 1 commit
    • Michel Lespinasse's avatar
      · 32aa5221
      Michel Lespinasse authored
      regression testing updates
  16. 02 Oct, 2000 6 commits
    • Michel Lespinasse's avatar
      · 3e2a3d2c
      Michel Lespinasse authored
      Fixed a parsing bug that wasnt detected before because it would only show up
      with streams that have intra_dc_precision > 0. Thanks to bruno for providing
      a short test case where the errors were actually visible - it helped me
      narrow down the point where I had introduced the bug. With this fix we
      pass all the gi/* conformance tests... nice :)
    • David I. Lehn's avatar
      - OMS can now be configured and built outside of the source dir. · ccc2e2ba
      David I. Lehn authored
      - Got rid of the use of DEFS in Makefile.am's since it was overriding
        automakes definition (-I. -I$(srcdir) -Iwherever_config.h_is).
        Moved code from DEFS into INCLUDES.  If you know another way around
        this please speak up.
      - acconfig.h generated from autoconf.sh, removed it from CVS.
      - Updated some m4.
      - Other minor build fixes.
    • Michel Lespinasse's avatar
      · 3bedb745
      Michel Lespinasse authored
      The chunk buffer was too small - some compliant MP@ML streams could overflow
      it. tcela/tcela-15-stuffing for example, was crashing because of that. Fixed
      it - but streams that are not MP@ML can still crash the decoder this way
      though. Added a TODO item to fix that eventually.
    • Michel Lespinasse's avatar
      · 47027a8b
      Michel Lespinasse authored
      stupid no-op checkin :)
    • Michel Lespinasse's avatar
      · f03c196c
      Michel Lespinasse authored
      This should fix builds with glibc2.0 (hopefully)
    • Bruno Barreyra's avatar
      fix auto* stuff · e1da8c7c
      Bruno Barreyra authored
      - bruno
  17. 01 Oct, 2000 2 commits
  18. 29 Sep, 2000 2 commits
    • Michel Lespinasse's avatar
      · 27d81e08
      Michel Lespinasse authored
      Fixed the slice routine output value. It broke the two tcela tests about
      short slices. Fixed now :)
      I have classified the streams that we dont decode correctly right now.
      There are a few categories :
      * mpeg1 - we do not do oddification yet, we crash on D pictures, but I think
      	the other things are correct.
      * field pictures - we dont support them yet
      * DMV motion compensation - we dont support that yet
      * tcela/tcela-15-stuffing - I think it overflows the chunk_buffer
      * gi/* - I'm not sure about them. we get off-by-one errors, even in I frames.
      	I suspect we do not support correctly intra_dc_precision>0 yet.
      	I can not see the error though :-/
    • Michel Lespinasse's avatar
      · 45c08c8f
      Michel Lespinasse authored
      added regression tests. see conformance/README to use it.