p7zip

1) Please compile again two versions (with/without fastcall) with asm listing (and C code in comments) and upload both versions of LzmaDec.asm (produced from LzmaDec.c) to your server. I'll try to look the difference.

Check also again that there is difference in Decompression speed on Core2 for these versions.

2) Why don't you use asm code of CRC? Is asm a problem for Linux? For example, ASM version of CRC is optimized for speed so it's 4-5 times faster, the algorithm in asm-crc is slighly different from default C code.