Hi all,
Attached is an update to quantize-pvt.c, with full asm implementations of
quantize_xrpow and quantize_xrpow_ISO for gcc/i386. They're basically Acy
Stapp's MSVC code, converted to gcc's arcane inline assembler syntax. On
this PIII/gcc-2.95 system, I get an overall speed increase of up to 10%,
most noticable in "quantization-heavy" modes such as high quality VBR.
In both the MSVC and gcc versions, I've also changed the "loop" instruction
in quantize_xrpow_ISO to a dec/jnz pair, which I believe is faster on all
Pentiums.
Problems? Suggestions for further improvements?
-- Mat.
quantize-pvt_update.c
