Re: Semantics of SAD_EXPR and usad/ssad optabs

2018-05-10 Thread Richard Biener
On May 10, 2018 10:53:19 AM GMT+02:00, Kyrill Tkachov wrote: >Hi Richard, > >On 09/05/18 19:37, Richard Biener wrote: >> On May 9, 2018 6:19:47 PM GMT+02:00, Kyrill Tkachov > wrote: >>> Hi all, >>> >>> I'm looking into implementing the usad/ssad optabs for aarch64 to >catch >>> code like in PR

Re: Semantics of SAD_EXPR and usad/ssad optabs

2018-05-10 Thread Kyrill Tkachov
Hi Richard, On 09/05/18 19:37, Richard Biener wrote: On May 9, 2018 6:19:47 PM GMT+02:00, Kyrill Tkachov wrote: Hi all, I'm looking into implementing the usad/ssad optabs for aarch64 to catch code like in PR 85693 and I'm a bit lost with what the midend expects the optabs to produce. The do

Re: Semantics of SAD_EXPR and usad/ssad optabs

2018-05-09 Thread Richard Biener
On May 9, 2018 6:19:47 PM GMT+02:00, Kyrill Tkachov wrote: >Hi all, > >I'm looking into implementing the usad/ssad optabs for aarch64 to catch >code like in PR 85693 >and I'm a bit lost with what the midend expects the optabs to produce. >The documentation for them says that the addend operand (

Semantics of SAD_EXPR and usad/ssad optabs

2018-05-09 Thread Kyrill Tkachov
Hi all, I'm looking into implementing the usad/ssad optabs for aarch64 to catch code like in PR 85693 and I'm a bit lost with what the midend expects the optabs to produce. The documentation for them says that the addend operand (op 3) is of mode equal or wider than the mode of the product (and