Re: PowerPC 64 x 32 bits performance

2005-03-07 Thread Robert Dewar
Alan Modra wrote: These are most likey 64-bit address constant loads. On ppc32, a 32-bit address constant can be calculated in at most 2 instructions. A 64-bit address takes up to 5 instructions to calculate in-line, or a TOC memory load. One general principle at work here is that there is no rul

Re: PowerPC 64 x 32 bits performance

2005-03-06 Thread Alan Modra
On Fri, Mar 04, 2005 at 04:59:25PM -0600, Edmar Wienskoski wrote: > and notice a considerable number of load instructions in the 64 bits one. > > Does anyone have an insight on why this is happening ? These are most likey 64-bit address constant loads. On ppc32, a 32-bit address constant can be

Re: PowerPC 64 x 32 bits performance

2005-03-04 Thread Andrew Pinski
On Mar 4, 2005, at 6:57 PM, David Edelsohn wrote: Edmar Wienskoski writes: Edmar> I checked with several gcc versions. Some of the spec2k Edmar> benchmarks has a considerable performance loss when Edmar> the benchmark is compiled with a 64 bits tool. Edmar> I looked into the assembler code generate

Re: PowerPC 64 x 32 bits performance

2005-03-04 Thread David Edelsohn
> Edmar Wienskoski writes: Edmar> I checked with several gcc versions. Some of the spec2k Edmar> benchmarks has a considerable performance loss when Edmar> the benchmark is compiled with a 64 bits tool. Edmar> I looked into the assembler code generated for try_route (route.c) Edmar> (compiled