Hi: In my opinion, we should axe all pure asm implementations in src/libFLAC/ia32 and the relevant configure options. Reasons are simple: - modern compilers plus the use of intrisincs make the code as faster as possible, if you need maximum speed I suggest you to build with profiling enabled. ;) - there is no support for x86_64 (that is.. all modern PC ;-) ) or for arm (most modern phones and media players) - AFAIK, those assembler optimizations may be even slower than the C code or buggy. Thoughts ? Cheers!
No. It is much faster, measurably. Anyone who believes the old adage that compilers outperform humans isn't writing math code or is just a lousy assembly programmer. On Apr 12, 2012, at 9:32 PM, Cristian Rodr?guez <crrodriguez at opensuse.org> wrote:> Hi: > > In my opinion, we should axe all pure asm implementations in > src/libFLAC/ia32 and the relevant configure options. > > Reasons are simple: > > - modern compilers plus the use of intrisincs make the code as faster > as possible, if you need maximum speed I suggest you to build with > profiling enabled. ;) > > - there is no support for x86_64 (that is.. all modern PC ;-) ) or for > arm (most modern phones and media players) > > - AFAIK, those assembler optimizations may be even slower than the C > code or buggy. > > Thoughts ? > > Cheers! > _______________________________________________ > flac-dev mailing list > flac-dev at xiph.org > http://lists.xiph.org/mailman/listinfo/flac-dev
El 12/04/12 22:55, Ben Allison escribi?:> No. It is much faster, measurably. Anyone who believes the old adage that compilers outperform humans isn't writing math code or is just a lousy assembly programmer.In which system/compiler did you tried it ?
Ben Allison wrote:> No. It is much faster, measurably.If thats the case, it would be really cool to have a couple of small benchmark programs in the tests/ directory that show this. It also suggests thats its high time we added an x86_64 version of the asm code. Cheers, Erik -- ---------------------------------------------------------------------- Erik de Castro Lopo http://www.mega-nerd.com/
Possibly Parallel Threads
- [LLVMdev] llvm-gcc4 and setjmp
- [GIT PULL] Assorted bugfixes and improvements (from openSUSE)
- [PATCH] for src/libFLAC/include/private/bitmath.h
- [PATCH] remove check for OS SSE support from src/libFLAC/cpu.c
- [PATCH] src/libFLAC/stream_decoder.c : Rework fix for seeking bug.