[XviD-devel] H263 quant. rewrite
jmtest at voila.fr
jmtest at voila.fr
Mon Mar 22 08:42:09 CET 2004
Hello,
I rewrote the H263 quant/dequant assembly functions and I got some speed gain of about 20%:
Here are the results of xvid_bench on a P4 2.4 Ghz
Before:
MMX - quant_h263_intra 3.047 usec crc32=0x2eba9d43
MMX - quant_h263_inter 3.047 usec crc32=0xbd315a7e
MMX - dequant_h263_intra 3.672 usec crc32=0x9841212a
MMX - dequant_h263_inter 3.672 usec crc32=0xe7df8fba
---
MMXEXT - quant_h263_intra 3.047 usec crc32=0x2eba9d43
MMXEXT - quant_h263_inter 3.047 usec crc32=0xbd315a7e
MMXEXT - dequant_h263_intra 3.672 usec crc32=0x9841212a
MMXEXT - dequant_h263_inter 3.672 usec crc32=0xe7df8fba
---
SSE2 - quant_h263_intra 2.461 usec crc32=0x2eba9d43
SSE2 - quant_h263_inter 2.422 usec crc32=0xbd315a7e
SSE2 - dequant_h263_intra 3.672 usec crc32=0x9841212a
SSE2 - dequant_h263_inter 3.047 usec crc32=0xe7df8fba
After:
MMX - quant_h263_intra 2.422 usec crc32=0x2eba9d43
MMX - quant_h263_inter 3.047 usec crc32=0xbd315a7e
MMX - dequant_h263_intra 3.047 usec crc32=0x9841212a
MMX - dequant_h263_inter 3.047 usec crc32=0xe7df8fba
---
MMXEXT - quant_h263_intra 2.422 usec crc32=0x2eba9d43
MMXEXT - quant_h263_inter 3.047 usec crc32=0xbd315a7e
MMXEXT - dequant_h263_intra 3.086 usec crc32=0x9841212a
MMXEXT - dequant_h263_inter 2.422 usec crc32=0xe7df8fba
---
SSE2 - quant_h263_intra 1.836 usec crc32=0x2eba9d43
SSE2 - quant_h263_inter 2.422 usec crc32=0xbd315a7e
SSE2 - dequant_h263_intra 1.836 usec crc32=0x9841212a
SSE2 - dequant_h263_inter 2.422 usec crc32=0xe7df8fba
Note that I didn't rewrite the quant_h263_inter because I couldn't find a better one. :-(
I can send the new source to anyone who knows how to compile and is available for more tests...
Regards,
JM
------------------------------------------
Faites un voeu et puis Voila ! www.voila.fr
More information about the XviD-devel
mailing list