Re: [patch][RFC] 160-bits bitmap_element

2012-08-17 Thread Steven Bosscher
On Fri, Aug 17, 2012 at 2:21 PM, Jakub Jelinek wrote: > On Fri, Aug 17, 2012 at 02:16:12PM +0200, Richard Guenther wrote: >> Or figure out if or why not the vectorizer does catch this (of course we do >> not enable that with -O2 which we eventually should in a very conservative >> mode). > > It mi

Re: [patch][RFC] 160-bits bitmap_element

2012-08-17 Thread Jakub Jelinek
On Fri, Aug 17, 2012 at 02:16:12PM +0200, Richard Guenther wrote: > Or figure out if or why not the vectorizer does catch this (of course we do > not enable that with -O2 which we eventually should in a very conservative > mode). It might be helpful if we for the BITMAP_ELEMENT_WORDS == 5 case reo

Re: [patch][RFC] 160-bits bitmap_element

2012-08-17 Thread Richard Guenther
On Fri, Aug 17, 2012 at 2:12 PM, Richard Guenther wrote: > On Fri, Aug 17, 2012 at 2:04 PM, Steven Bosscher > wrote: >> On Fri, Aug 17, 2012 at 1:54 PM, Richard Guenther >> wrote: >>> Well, another effect of reducing the size of BITMAP_WORD is that >>> operations are not performed in a mode opt

Re: [patch][RFC] 160-bits bitmap_element

2012-08-17 Thread Jakub Jelinek
On Fri, Aug 17, 2012 at 02:04:50PM +0200, Steven Bosscher wrote: > On Fri, Aug 17, 2012 at 1:54 PM, Richard Guenther > wrote: > > Well, another effect of reducing the size of BITMAP_WORD is that > > operations are not performed in a mode optimally using CPU regs > > (did you check code generation

Re: [patch][RFC] 160-bits bitmap_element

2012-08-17 Thread Richard Guenther
On Fri, Aug 17, 2012 at 2:04 PM, Steven Bosscher wrote: > On Fri, Aug 17, 2012 at 1:54 PM, Richard Guenther > wrote: >> Well, another effect of reducing the size of BITMAP_WORD is that >> operations are not performed in a mode optimally using CPU regs >> (did you check code generation differences

Re: [patch][RFC] 160-bits bitmap_element

2012-08-17 Thread Steven Bosscher
On Fri, Aug 17, 2012 at 1:54 PM, Richard Guenther wrote: > Well, another effect of reducing the size of BITMAP_WORD is that > operations are not performed in a mode optimally using CPU regs > (did you check code generation differences on a 64bit host?). I did, on x86_64 and on powerpc64. The effe

Re: [patch][RFC] 160-bits bitmap_element

2012-08-17 Thread Richard Guenther
On Fri, Aug 17, 2012 at 1:46 PM, Steven Bosscher wrote: > Hello, > > On a 64-bits host we leave 32 bits of each bitmap_element unused. And > actually, on a 32-bits host we do that too for GGC-allocated bitmaps > (due to alignment). > > With this patch, we use those 32 bits by making BITMAP_WORD an