[FFmpeg-devel] [PATCH] h264: assembly version of get_cabac for x86_64 with PIC (v4)

Roland Scheidegger rscheidegger_lists at hispeed.ch
Sat Apr 21 17:22:57 CEST 2012

Am 21.04.2012 10:57, schrieb Reimar Döffinger:
> On Sat, Apr 21, 2012 at 10:47:55AM +0200, Reimar Döffinger wrote:
>> On 21 Apr 2012, at 07:25, Roland Scheidegger <rscheidegger_lists at hispeed.ch> wrote:
>>> Am 21.04.2012 02:15, schrieb Michael Niedermayer:
>>>> On Sat, Apr 21, 2012 at 02:10:46AM +0200, Michael Niedermayer wrote:
>>>>> On Sat, Apr 21, 2012 at 01:26:54AM +0200, Michael Niedermayer wrote:
>>>>>> On Fri, Apr 20, 2012 at 02:10:57AM +0200, Roland Scheidegger wrote:
>>>>>>> This adds a hand-optimized assembly version for get_cabac much like the
>>>>>>> existing one, but it works if the table offsets are RIP-relative.
>>>>>>> Compared to the non-RIP-relative version this adds 2 lea instructions
>>>>>>> and it needs one extra register.
>>>>>>> There is a surprisingly large performance improvement over the c version (more
>>>>>>> so than the generated assembly seems to suggest) just in get_cabac, I measured
>>>>>>> roughly 40% faster for get_cabac on a K8. However, overall the difference is
>>>>>>> not that big, I measured roughly 5% on a test clip on a K8 and a Core2.
>>>>>>> Hopefully it still compiles on x86 32bit...
>>>>>>> v2: incorporated feedback from Loren Merritt to avoid rip-relative movs
>>>>>>> for every table, and got rid of unnecessary @GOTPCREL.
>>>>>>> v3: apply similar fixes to the the decode_significance functions, and use
>>>>>>> same macro arguments for non-pic case.
>>>>>>> v4: prettify inline asm arguments, add a non-fast-cmov version (as I expect
>>>>>>> the c code to be faster otherwise since both cmov and sbb suck hard on a
>>>>>>> Prescott, even can't construct the mask with a 64bit shift as that's just as
>>>>>>> terrible - it's quite difficult to find usable instructions on that chip...).
>>>>>>> This is tested to work but not on a P4, in theory it _should_ be fast there.
>>>>>> applied
>>>>>> if someone has more ideas on how to improve it, it can easily be done
>>>>>> lets hope it doesnt fail on any odd platforms ...
>>>>> fails on darwin
>>>>> http://fate.ffmpeg.org/log.cgi?time=20120420235051&log=compile&slot=x86_64-darwin-llvm-gcc-4.2.1
>>>> ill revert it in a moment, dont want to leave compile broken until
>>>> another solution is proposed
>>> Oops. Any ideas what went wrong? Somehow the trick to access the table
>>> via the difference of the table symbol and the label doesn't work? Is
>>> that something just not supposed to work on darwin (PIC code should work
>>> the same there right for x64?) or is this fixable? I think the only
>>> thing I could do about is is just not enable this on darwin but surely
>>> there has to be some way to make it work?
>> The output looks to me like half the symbols are missing the leading _, i.e. EXTERN_PREFIX?
> Just wrapping all the ff_ symbols used in the asm with LABEL_MANGLE()
> should work.

Ahh thanks that makes sense. The original code had MANGLE which couldn't
work but this was EXTERN_PREFIX LOCAL_MANGLE so when I removed this I
not only removed the LOCAL_MANGLE bit (which can't work there) but also
the EXTERN_PREFIX bit. I'll send out a new version.


More information about the ffmpeg-devel mailing list