On Jan 6, 2014, at 11:02 AM, Chandler Carruth <chandlerc at google.com> wrote:> > On Mon, Jan 6, 2014 at 1:42 PM, Andrew Trick <atrick at apple.com> wrote: > On Jan 6, 2014, at 10:10 AM, Chandler Carruth <chandlerc at google.com> wrote: > >> Trying to bubble way back to the top, Andy, do you think there is anything else that needs to be done here before it can go in? I feel like most of our discussion centered around future work, and I'd like to unblock the immediate work to support subtarget-specific code generation. > > Do we have a solution that can work with TTI? I don’t think there was any objection to putting CGContext in TargetMachine, except that we have too many layers of abstraction, which you said will be cleaned up eventually: > > IR Transform > (links with) -> TargetTransformInfo > (dynamic call) -> X86TTI > (links with) -> TargetMachine > (provides) -> CGContext > > This seems fine to me?Ok. Future work: do something simpler for TTI, as opposed to analysis groups. -Andy -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140106/532a2148/attachment.html>
Hi Andy et al, Attached is an updated version of the CGContext patch. This fixes some bugs in the previous version, and moves the CGContext registry to TargetMachine. Is this in line with what you expected? Dan On Mon, Jan 6, 2014 at 11:09 AM, Andrew Trick <atrick at apple.com> wrote:> > On Jan 6, 2014, at 11:02 AM, Chandler Carruth <chandlerc at google.com> > wrote: > > > On Mon, Jan 6, 2014 at 1:42 PM, Andrew Trick <atrick at apple.com> wrote: > >> On Jan 6, 2014, at 10:10 AM, Chandler Carruth <chandlerc at google.com> >> wrote: >> >> Trying to bubble way back to the top, Andy, do you think there is >> anything else that needs to be done here before it can go in? I feel like >> most of our discussion centered around future work, and I'd like to unblock >> the immediate work to support subtarget-specific code generation. >> >> >> Do we have a solution that can work with TTI? I don’t think there was any >> objection to putting CGContext in TargetMachine, except that we have too >> many layers of abstraction, which you said will be cleaned up eventually: >> >> IR Transform >> (links with) -> TargetTransformInfo >> (dynamic call) -> X86TTI >> (links with) -> TargetMachine >> (provides) -> CGContext >> > > This seems fine to me? > > > Ok. Future work: do something simpler for TTI, as opposed to analysis > groups. > -Andy >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140121/d4592a1e/attachment.html> -------------- next part -------------- A non-text attachment was scrubbed... Name: cgcontext.patch Type: text/x-diff Size: 10995 bytes Desc: not available URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140121/d4592a1e/attachment.patch>
On Jan 21, 2014, at 5:17 PM, Dan Gohman <dan433584 at gmail.com> wrote:> Hi Andy et al, > > Attached is an updated version of the CGContext patch. This fixes some bugs in the previous version, and moves the CGContext registry to TargetMachine. Is this in line with what you expected?Sorry, I put this aside for too long when it wasn’t hard to review at all. This is right in line with what I expected. CGContext is not in the patch, but I get the idea. -Andy> On Mon, Jan 6, 2014 at 11:09 AM, Andrew Trick <atrick at apple.com> wrote: > > On Jan 6, 2014, at 11:02 AM, Chandler Carruth <chandlerc at google.com> wrote: > >> >> On Mon, Jan 6, 2014 at 1:42 PM, Andrew Trick <atrick at apple.com> wrote: >> On Jan 6, 2014, at 10:10 AM, Chandler Carruth <chandlerc at google.com> wrote: >> >>> Trying to bubble way back to the top, Andy, do you think there is anything else that needs to be done here before it can go in? I feel like most of our discussion centered around future work, and I'd like to unblock the immediate work to support subtarget-specific code generation. >> >> Do we have a solution that can work with TTI? I don’t think there was any objection to putting CGContext in TargetMachine, except that we have too many layers of abstraction, which you said will be cleaned up eventually: >> >> IR Transform >> (links with) -> TargetTransformInfo >> (dynamic call) -> X86TTI >> (links with) -> TargetMachine >> (provides) -> CGContext >> >> This seems fine to me? > > Ok. Future work: do something simpler for TTI, as opposed to analysis groups. > -Andy > > <cgcontext.patch>-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140204/260e783a/attachment.html>