Re: PR78634: ifcvt/i386 cost updates

2017-01-18 Thread Uros Bizjak
On Wed, Jan 18, 2017 at 1:10 PM, Bernd Schmidt wrote: > On 12/09/2016 12:49 PM, Bernd Schmidt wrote: >> >> On 12/03/2016 10:49 AM, Uros Bizjak wrote: >> >>> Based on the above explanation, the patch is OK. >> >> >> I'll be treating the ifcvt part of it as obvious. However,

Re: PR78634: ifcvt/i386 cost updates

2017-01-18 Thread Bernd Schmidt
On 12/09/2016 12:49 PM, Bernd Schmidt wrote: On 12/03/2016 10:49 AM, Uros Bizjak wrote: Based on the above explanation, the patch is OK. I'll be treating the ifcvt part of it as obvious. However, testing showed an issue with the i386 funcspec-11 test: /* PR target/36936 */ /* { dg-do

Re: PR78634: ifcvt/i386 cost updates

2016-12-09 Thread Bernd Schmidt
On 12/03/2016 10:49 AM, Uros Bizjak wrote: Based on the above explanation, the patch is OK. I'll be treating the ifcvt part of it as obvious. However, testing showed an issue with the i386 funcspec-11 test: /* PR target/36936 */ /* { dg-do compile } */ /* { dg-require-effective-target ia32

Re: PR78634: ifcvt/i386 cost updates

2016-12-03 Thread Uros Bizjak
On Fri, Dec 2, 2016 at 8:58 PM, James Greenhalgh wrote: > On Fri, Dec 02, 2016 at 05:00:05PM +0100, Bernd Schmidt wrote: >> With the i386 backend no longer double-counting the cost of a SET, >> the default implementation default_max_noce_ifcvt_seq_cost now >> provides

Re: PR78634: ifcvt/i386 cost updates

2016-12-02 Thread James Greenhalgh
On Fri, Dec 02, 2016 at 05:00:05PM +0100, Bernd Schmidt wrote: > With the i386 backend no longer double-counting the cost of a SET, > the default implementation default_max_noce_ifcvt_seq_cost now > provides too high a bound for if conversion, allowing very costly > substitutions. > > The

PR78634: ifcvt/i386 cost updates

2016-12-02 Thread Bernd Schmidt
With the i386 backend no longer double-counting the cost of a SET, the default implementation default_max_noce_ifcvt_seq_cost now provides too high a bound for if conversion, allowing very costly substitutions. The following patch fixes this by making an i386 variant of the hook, but first -