[MPlayer-users] ssa/ass rendering uses lots of cpu with -vo gl

[Reimar Döffinger]

On Sun, May 01, 2011 at 09:46:32PM +0400, Vladimir Mosgalin wrote:
> In window:
> # Events: 1K cycles
> #
> # Overhead  Command                 Shared Object                                        Symbol
> # ........  .......  ............................  ............................................
> #
>     26.75%  mplayer  mplayer                       [.] deNoise
>     10.80%  mplayer  r300_dri.so                   [.] 0x525df         
>      3.75%  mplayer  libc-2.13.so                  [.] __memcpy_ssse3
>      3.42%  mplayer  mplayer                       [.] ff_h264_decode_mb_cabac
>      2.33%  mplayer  mplayer                       [.] lineNoiseAvg_MMX
>      2.31%  mplayer  mplayer                       [.] decode_cabac_residual_nondc_internal
>      2.28%  mplayer  mplayer                       [.] decode_cabac_mb_mvd
>      2.17%  mplayer  mplayer                       [.] put_h264_qpel8or16_hv_lowpass_ssse3.clone.35
>      1.47%  mplayer  libc-2.13.so                  [.] _int_malloc
>      1.45%  mplayer  [kernel.kallsyms]             [k] __list_del_entry
>      1.44%  mplayer  mplayer                       [.] get_cabac_noinline
> 
> In fullscreen:
>  Events: 4K cycles
> #
> # Overhead  Command                 Shared Object                                          Symbol
> # ........  .......  ............................  ..............................................
> #
>     14.47%  mplayer  libfreetype.so.6.6.0          [.] 0xfae6          
>     14.40%  mplayer  r300_dri.so                   [.] 0x416be         
>     11.17%  mplayer  [radeon]                      [k] radeon_dvi_detect
>      8.92%  mplayer  mplayer                       [.] deNoise
>      3.31%  mplayer  [drm]                         [k] drm_mm_init_scan
>      1.60%  mplayer  [kernel.kallsyms]             [k] kmem_cache_alloc_trace
>      1.32%  mplayer  libc-2.13.so                  [.] __memset_sse2
>      1.22%  mplayer  libdrm_radeon.so.1.0.0        [.] 0x2c9a          
>      1.19%  mplayer  [kernel.kallsyms]             [k] copy_user_generic_string
>      1.12%  mplayer  libc-2.13.so                  [.] __memcpy_ssse3
>      0.88%  mplayer  mplayer                       [.] lineNoiseAvg_MMX

This doesn't look like you ran the windowed test at the fullscreen resolution.
At least that's the only explanation I have for the libfreetype CPU usage change.
You can try -vo gl:scaled-osd to avoid that change to get a clearer look on the rest.
Since we can assume that deNoise used about the same CPU for both, we can conclude
that r300_dri.so used about 4 times as much CPU as before, and the radeon kernel
module came in with about the same amount of CPU usage.
A possibilty is that it is relate to transfers, older GPUs had much, much worse
DMA support (not to mention that to my knowledge no ATI drivers, proprietary or OpenSource,
have managed to actually handle transfers well. Actually it seems to be
kind of an Achillees heel for all GPUs).