Quote:
Keep in mind that x32 does not perform better than x64, it's just smaller. So if you So if you > build and run 64-bit right now you already are seeing max performance, it's just x32 will be a bit smaller.
Actually you are wrong here, benchmark results have been shown by H.J.Lu of Intel (who implemented the x32 abi into gcc/linux) and recently by Micheal Liao (also Intel) who submitted a first x32 patch for LLVM which shows that it beats the equivalent 64-bit code quite readily in some cpu intense benchmarks due to the code footprint being smaller resulting in more code fitting into the cpu caches (this is also the reason why 32-bit can sometimes beat 64-bit code even though it only has half the available registers).
Also worth noting is that the x32 code often ends up smaller even than 32-bit code due to not having to perform as much pushing/pop'ing as a result of having twice the amount of registers.
Here's the link to some benchmark results posted by Micheal Liao as part of his x32 patch announcement at the LLVM mailing list: http://lists.cs.uiuc.edu/pipermail/llvmdev/2012-June/050360.html
You can find the original x32 abi page here: https://sites.google.com/site/x32abi/
|