flac dev - Sep 2013 - Performance and precompute_partition_info_sums_32bit_asm_ia32_()

Previously I wrote that precompute_partition_info_sums_32bit_asm_ia32_() only
makes encoding slower. Now I managed to compile flac with GCC 4.8.1, with this
function enabled and disabled. NASM was enabled, SSE intrinsics disabled.
Then I added -msse option (so that all C code was compiled with -msse),
then -msse2 and so on.

Input file for test: 44.1kHz/16bit/stereo; best compression mode (flac -8); CPU
= Core i7.

Here are the results (1st column: SSE instruction set, 2nd column: the state of
precompute_partition_info_sums_32bit_asm_ia32_(), 3rd column: encoding time
in seconds, smaller=better):

no SSE   disabled    53.9
no SSE    enabled    55.2

SSE1     disabled    53.9
SSE1      enabled    55.3

SSE2     disabled    51.9
SSE2      enabled    53.1

SSE3     disabled    51.8
SSE3      enabled    53.2

SSSE3    disabled    45.7
SSSE3     enabled    51.4

SSE41    disabled    46.1
SSE41     enabled    51.6

SSE42    disabled    46.1
SSE42     enabled    51.6


Conclusions:

1) flac is always faster when precompute_partition_info_sums_32bit_asm_ia32_()
is disabled.

2) Some C code benefits noticeably from SSSE3 instructions; at least when
compiled with GCC 4.8.1.

lvqcl wrote:
> Conclusions:
> 
> 1) flac is always faster when
precompute_partition_info_sums_32bit_asm_ia32_()
> is disabled.
Maybe its time to either improve or remove that code.

Erik
-- 
----------------------------------------------------------------------
Erik de Castro Lopo
http://www.mega-nerd.com/

Erik de Castro Lopo wrote:
> Maybe its time to either improve or remove that code.
I posted a patch that adds SSE2/SSSE3 versions of
precompute_partition_info_sums_().
I didn't remove the old code; just changed "#if defined ..." to
"#if defined ... && 0".


Currently there are two useless asm files:
* bitreader_asm.nasm (defines
FLAC__bitreader_read_rice_signed_block_asm_ia32_bswap)
* stream_encoder_asm.nasm (defines
precompute_partition_info_sums_32bit_asm_ia32_)
and I doubt that they will become useful again.


BTW, after removing bitreader_asm.nasm from build it will be possible to remove
local_bitreader_read_rice_signed_block from struct FLAC__StreamDecoderPrivate
and call FLAC__bitreader_read_rice_signed_block() directly (IOW, revert the
patch
http://git.xiph.org/?p=flac.git;a=commit;h=ddddff6a5604da5c7223a075e58ca532d7ad375d
).
Or to make bitreader_read_from_client_() function static again (it was static
before
http://git.xiph.org/?p=flac.git;a=commit;h=c63cf41cccba25a268f235e83ed8603adc0ac3ec
).