RE: [PATCH][RFC] Fix PR49957 - build array index differently

2011-08-25 Thread Greta Yorsh
This patch causes regression in one of the Spec2000 benchmarks on arm-none-linux-gnueabi cortex-a9. The benchmark 173.applu from CFP2000 dropped performance by about 8% between revisions 177367 and 177368. Other benchmarks are not affected. In the assembly generated by the new version, the two m

Re: [PATCH][RFC] Fix PR49957 - build array index differently

2011-08-04 Thread Mikael Morin
On Thursday 04 August 2011 13:12:04 Richard Guenther wrote: > On Thu, 4 Aug 2011, Richard Guenther wrote: > > On Wed, 3 Aug 2011, Mikael Morin wrote: > > > Hello, > > > > > > On Wednesday 03 August 2011 15:47:37 Richard Guenther wrote: > > > > Comments? Any idea why reversing the loop would break

Re: [PATCH][RFC] Fix PR49957 - build array index differently

2011-08-04 Thread Richard Guenther
On Thu, 4 Aug 2011, Richard Guenther wrote: > On Wed, 3 Aug 2011, Mikael Morin wrote: > > > Hello, > > > > On Wednesday 03 August 2011 15:47:37 Richard Guenther wrote: > > > Comments? Any idea why reversing the loop would break? > > > > Yes, the list of scalarized expressions has to be created

Re: [PATCH][RFC] Fix PR49957 - build array index differently

2011-08-04 Thread Richard Guenther
On Wed, 3 Aug 2011, Mikael Morin wrote: > Hello, > > On Wednesday 03 August 2011 15:47:37 Richard Guenther wrote: > > Comments? Any idea why reversing the loop would break? > > Yes, the list of scalarized expressions has to be created in the same order > it > is consumed. Here the scalarized

Re: [PATCH][RFC] Fix PR49957 - build array index differently

2011-08-03 Thread Mikael Morin
Hello, On Wednesday 03 August 2011 15:47:37 Richard Guenther wrote: > Comments? Any idea why reversing the loop would break? Yes, the list of scalarized expressions has to be created in the same order it is consumed. Here the scalarized expressions are array indexes to be precomputed out of th

Re: [PATCH][RFC] Fix PR49957 - build array index differently

2011-08-03 Thread Tobias Burnus
Richard Guenther wrote: On Wed, 3 Aug 2011, Richard Guenther wrote: Bootstrap and regtest is currently running on x86_64-unknown-linux-gnu, the reversed loop one was ok (well, apart from those 2-3 fails). Re-bootstrapping currently. The previous patch tested ok. The patch looks OK. I don't k

Re: [PATCH][RFC] Fix PR49957 - build array index differently

2011-08-03 Thread Richard Guenther
On Wed, 3 Aug 2011, Richard Guenther wrote: > > This fixes PR49957 by keeping the array index into a multi-dimensional > array in optimal associated form which is ((off + outermost) + ...) + > innermost) + constant) so that dependence analysis can properly > handle it. It doesn't work right now