Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on August 17, 2012 2:41 pm wrote:
> IIRC most 64-bit architectures
> zero-extend a 32-bit load. That means you can use 32-bit pointers to access the
> first 4GB without any penalty at all. There is an effort to create an x32 ABI on
> x64 that does exactly that.

Maybe, but there are two cases:

1) All your data fits in 4GB (or maybe 2GB). In that case
you should just run in 32bit mode, if your architecture has one.

2) Your data might go beyond 4GB. And in that case the pain of
managing separate regions for below-4GB and above-4GB memory,
requiring different instruction sequences to access the different
flavors of pointers, would be simply awful - probably much
more effort than using compressed pointers into a uniform
address space.

So I don't think there's much benefit to this in general - though I guess
it might be useful for the particular case of x86 vs x86_64, where the
32bit architecture has a horrible shortage of registers. But I don't
x86_64 would let you use a 32bit register as a 64bit address, does it ?