[FFmpeg-devel] [PATCH] h264: integrate clear_blocks calls with IDCT.

[Ronald S. Bultje]

Hi,

On Mon, Feb 18, 2013 at 7:55 AM, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Sun, Feb 17, 2013 at 07:47:22PM -0800, Ronald S. Bultje wrote:
>> Hi,
>>
>> On Sun, Feb 17, 2013 at 6:04 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
>> > On Sun, Feb 17, 2013 at 02:52:54PM -0800, Ronald S. Bultje wrote:
>> >> From: "Ronald S. Bultje" <rsbultje at gmail.com>
>> >>
>> >> The non-intra-pcm branch in hl_decode_mb (simple, 8bpp) goes from 700
>> >> to 672 cycles, and the complete loop of decode_mb_cabac and hl_decode_mb
>> >> (in the decode_slice loop) goes from 1759 to 1733 cycles on the clip
>> >> tested (cathedral), i.e. almost 30 cycles per mb faster.
>> >>
>> >
>> >> Arm assembly changes untested.
>> >
>> > fate-h264 (h264-conformance-ba_mw_d in this case but its not the only
>> > one)
>> >
>> > Program received signal SIGSEGV, Segmentation fault.
>> > ff_h264_idct_add8_neon () at ffmpeg/libavcodec/arm/h264idct_neon.S:166
>> > 166         ldrsh           r8,  [r1]
>>
>> diff --git a/libavcodec/arm/h264idct_neon.S b/libavcodec/arm/h264idct_neon.S
>> index a2521b7..8d85227 100644
>> --- a/libavcodec/arm/h264idct_neon.S
>> +++ b/libavcodec/arm/h264idct_neon.S
>> @@ -105,6 +105,7 @@ function ff_h264_idct_add16_neon, export=1
>>          ldr             r0,  [r5], #4
>>          ldrb            r8,  [r6, r8]
>>          subs            r8,  r8,  #1
>> +        mov             r3,  r1
>>          blt             2f
>>          ldrsh           lr,  [r1]
>>          add             r0,  r0,  r4
>> @@ -116,7 +117,7 @@ function ff_h264_idct_add16_neon, export=1
>>          adreq           lr,  ff_h264_idct_add_neon    + CONFIG_THUMB
>>          blx             lr
>>  2:      subs            ip,  ip,  #1
>> -        add             r1,  r1,  #32
>> +        add             r1,  r3,  #32
>>          bne             1b
>>          pop             {r4-r8,pc}
>>  endfunc
>> @@ -136,13 +137,14 @@ function ff_h264_idct_add16intra_neon, export=1
>>          add             r0,  r0,  r4
>>          cmp             r8,  #0
>>          ldrsh           r8,  [r1]
>> +        mov             r3,  r1
>>          iteet           ne
>>          adrne           lr,  ff_h264_idct_add_neon    + CONFIG_THUMB
>>          adreq           lr,  ff_h264_idct_dc_add_neon + CONFIG_THUMB
>>          cmpeq           r8,  #0
>>          blxne           lr
>>          subs            ip,  ip,  #1
>> -        add             r1,  r1,  #32
>> +        add             r1,  r3,  #32
>>          bne             1b
>>          pop             {r4-r8,pc}
>>  endfunc
>>
>> ?
>
> Program received signal SIGSEGV, Segmentation fault.
> ff_h264_idct_add8_neon () at ffmpeg/libavcodec/arm/h264idct_neon.S:168
> 168         ldrsh           r8,  [r1]
>
> btw, dont you have some linux system with qemu-arm available ?

I'll see if I can get that to work sometime this week...

Ronald