- User Since
- Mar 27 2015, 6:20 AM (120 w, 6 d)
Tue, Jul 18
Thanks Tim. Will fix the commented code and commit.
Mon, Jul 17
Thanks Tim/others for the suggestions. Updated, using 'Requires' , FeatureBits and factoring out common functions. Is much more compact and unified now, as I set out to do.
Sat, Jul 15
Fri, Jul 14
Ok, I see that now too. Will use MI->operands() as suggested, as that will suffice.
I have re-implemented using SearchableTable which improves the code quality immensely (great suggestion by Oliver). I have also create a Utils sub-directory (and ARMUtils library) that allows utility functions such as generated by SearchableTable to be incorporated into both CodeGen and others. seamlessly.
Thu, Jul 13
Wed, Jul 12
Sure, will make the changes and wait till tomorrow before committing, in case Renato or Diana have further comments.
Ping! Renato, Diana, Florian ?
Tue, Jul 11
Thanks Florian. I will fix the indents where I missed out.
Mon, Jul 10
Fri, Jul 7
Good catch. Thanks. Will fix these and commit.
Thanks Florian. I will wait a bit for Renato/Diana to comment before committing, in case they have more comments/suggestions.
Tue, Jul 4
Wed, Jun 28
scheduling part LGTM now
Tue, Jun 27
Thu, Jun 22
Abandoning as there is a separate pragma clang section implementation now.
Wed, Jun 21
LGTM but I will wait a day before accepting to allow time for others to comment too, just in case.
Thanks for this.
Jun 20 2017
Thanks Matthias for review. Please find the changes as requested.
Jun 19 2017
Jun 18 2017
Jun 17 2017
Jun 14 2017
Jun 13 2017
Thanks Florian for this. Some comments below -
LGTM. @sbaranga are you also ok with this?
Thanks Oliver for fixing it.
Jun 9 2017
Jun 8 2017
Thanks Oliver for adding the missing sched defs. I am surprised there was assertion failures for missing scheds as default behaviour of machine scheduler used to be 'to-silently-ignore'. Anyways, that's a separate issue. Still good to have complete defs.
Jun 7 2017
Jun 6 2017
I have made changes as suggested (unless I got somethings wrong).
- Now, except for android, AAPCS => 64-bit alignment
- The fix relies only on the triple for I am not sure if 'right cc1' flag test is needed.
- Added gnueabi and FreeBSD as suggested.
Please have a look. Thanks.
My two cents. I think this is likely to give gains where pipeline leverages it, and unlikely to cause regressions otherwise.
Joel Jones mentioned checking performance on ThuderX so we should probably wait for him to come back on it.
Jun 5 2017
Jun 2 2017
Vedant, Diana : Does this look ok to you as well?
Jun 1 2017
May 31 2017
Approving as all actions requested are resolved and the overall geomean is positive. Thanks for this work!
Thanks Roger for the review and the suggestion to add documentation in LanguageExtensions.rst. I have done so now.
May 30 2017
No problem. Please have a look at http://llvm.org/docs/Phabricator.html (sec. Requesting a review...). Basically, 'git diff -U9999 your-branch' will create a patch with more context.