- User Since
- Sep 26 2016, 7:58 AM (172 w, 5 d)
Wed, Jan 15
Thanks for looking into this. I did not know about shouldChangeType but that seems like a useful function here.
Tue, Jan 14
Mon, Dec 30
Dec 19 2019
Dec 18 2019
Dec 16 2019
Dec 15 2019
Fix typo in code comments.
Update code comments.
Dec 13 2019
Dec 12 2019
(if removing the unused VarInstPair/InstsToSink)
Don't fully understand. This revision is closed, but latest diff is showing a new diff (not the one closed/reverted, right?). If I understand it correctly revision should be re-opened for review again, right?
There are several other passes between isel and phi elimination. So if we now have a rule that there must not be any dbg instructions between PHI-nodes and labels that should be at the beginning of the MBB, then perhaps we need to add a check in MachineVerfier that the new rule is followed by all those passes.
Dec 11 2019
Dec 8 2019
Dec 3 2019
Is it possible to do the WebAssembly updates in a separate patch, or is those changes dependent on also changing how CodeGenPrepare works?
Dec 2 2019
Nov 29 2019
Some more refactoring, very much inspired by the ideas from Roman.
Realized it was stupid to use a temporary vector when it is possible to simply use append. So a yet another minor refactoring.
Minor refactoring and added some more code comments.
Rebase after update to run cast.ll for both big and little endian.
Nov 28 2019
Nov 26 2019
Nov 24 2019
Nov 21 2019
Thanks for the LGTM, and the earlier review comments. Your idea about using skipDebugInstructionsForward simplified things even more! :-)
Update based on review comments.
Nov 15 2019
Cleaned up. No longer WIP. Ready for review.
Nov 14 2019
Nov 11 2019
Nov 9 2019
Nov 7 2019
Here is an test case that currently fail. The test1, test2 and test3 functions only differ by DBG_VALUE and CFI_INSTRUCTION instructions. So I guess we want to see the same real instructions being generated for all three functions.
Unfortunately I see a lot more test cases failing due to codegen not being debug invariant after having merged this patch, compared to the past.
Nov 3 2019
Nov 1 2019
Oct 31 2019
Oct 30 2019
Addressed review comments.
(Reverse ping!) Any reason why this hasn't landed yet?
Oct 29 2019
Oct 28 2019
Looks like this one was accepted a long time ago. But I can't see that it has been committed yet. So what is the status?
Oct 23 2019
Oct 22 2019
Oct 21 2019
Thanks for the patience with all my remarks, turning this around completely compared to the original patch.
Oct 18 2019
Thanks! I like this solution.
Oct 16 2019
- Fixup mistake regarding FDiv (it was not supposed to be included in the new restriction).
- Added test cases for sdiv/urem/udiv (showing that the new restriction works).
- Added test cases for fdiv/frem (showing that we can eliminate the shufflevector).