Page MenuHomePhabricator

JesperAntonsson (Jesper Antonsson)
User

Projects

User does not belong to any projects.

User Details

User Since
May 25 2016, 6:41 AM (168 w, 2 d)

Recent Activity

May 9 2019

JesperAntonsson created D61725: Non-8-bit addressable units showcase.
May 9 2019, 5:12 AM · Restricted Project
JesperAntonsson added inline comments to D61432: Non-8-bit bytes showcase.
May 9 2019, 5:12 AM · Restricted Project
JesperAntonsson updated the diff for D61432: Non-8-bit bytes showcase.

Added an inBytes() helper method. (We could contemplate putting this in DataLayout instead for more general use.) Adjusted READNONE to READONLY. Improved comments.

May 9 2019, 5:07 AM · Restricted Project

May 2 2019

JesperAntonsson created D61432: Non-8-bit bytes showcase.
May 2 2019, 4:49 AM · Restricted Project

Mar 14 2019

JesperAntonsson created D59358: [CodeGenPrepare] avoid crashing from replacing a phi twice.
Mar 14 2019, 5:29 AM · Restricted Project

Oct 1 2018

JesperAntonsson committed rL343486: [InstCombine] Handle vector compares in foldGEPIcmp(), take 2.
[InstCombine] Handle vector compares in foldGEPIcmp(), take 2
Oct 1 2018, 8:01 AM
JesperAntonsson closed D52494: [InstCombine] Handle vector compares in foldGEPIcmp(), take 2.
Oct 1 2018, 8:01 AM
JesperAntonsson added a comment to D52494: [InstCombine] Handle vector compares in foldGEPIcmp(), take 2.

Thanks.

Oct 1 2018, 7:59 AM
JesperAntonsson updated the diff for D52494: [InstCombine] Handle vector compares in foldGEPIcmp(), take 2.

Thanks, Sanjay. Using makeCmpResultType() in this patch for a smaller and nicer diff.

Oct 1 2018, 5:44 AM

Sep 27 2018

JesperAntonsson updated the diff for D52494: [InstCombine] Handle vector compares in foldGEPIcmp(), take 2.

This patch checks icmp types directly. This might be simpler than checking the operand type.

Sep 27 2018, 1:29 AM
JesperAntonsson added inline comments to D52494: [InstCombine] Handle vector compares in foldGEPIcmp(), take 2.
Sep 27 2018, 1:07 AM

Sep 25 2018

JesperAntonsson created D52494: [InstCombine] Handle vector compares in foldGEPIcmp(), take 2.
Sep 25 2018, 7:24 AM

Sep 20 2018

JesperAntonsson committed rL342647: [InstCombine] Handle vector compares in foldGEPIcmp().
[InstCombine] Handle vector compares in foldGEPIcmp()
Sep 20 2018, 6:38 AM
JesperAntonsson closed D52263: [InstCombine] Handle vector compares in foldGEPIcmp().
Sep 20 2018, 6:38 AM
JesperAntonsson updated the diff for D52263: [InstCombine] Handle vector compares in foldGEPIcmp().

Used an overloaded getter with simpler interface.

Sep 20 2018, 2:13 AM

Sep 19 2018

JesperAntonsson updated the diff for D52263: [InstCombine] Handle vector compares in foldGEPIcmp().

Roman Lebedev commented with an elegant way to create varying integer values, so this patch makes use of that. Changed name of patch to reflect that we now handle rather than ignore vector icmps.

Sep 19 2018, 6:44 AM
JesperAntonsson created D52263: [InstCombine] Handle vector compares in foldGEPIcmp().
Sep 19 2018, 6:06 AM

Jun 28 2018

JesperAntonsson committed rL335832: Comment change to verify commit rights. NFC..
Comment change to verify commit rights. NFC.
Jun 28 2018, 3:59 AM
JesperAntonsson closed D48709: Comment change to verify commit rights. NFC..
Jun 28 2018, 3:59 AM
JesperAntonsson created D48709: Comment change to verify commit rights. NFC..
Jun 28 2018, 2:05 AM

Jun 21 2018

JesperAntonsson added a comment to D48420: [CodeGen} Avoid handling DBG_VALUE in LiveRegUnits::stepBackward.

Thanks Krzysztof. I don't have commit rights yet, so I'd appreciate help to get this merged to trunk.

Jun 21 2018, 6:38 AM · debug-info
JesperAntonsson created D48420: [CodeGen} Avoid handling DBG_VALUE in LiveRegUnits::stepBackward.
Jun 21 2018, 4:42 AM · debug-info

Jun 20 2018

JesperAntonsson updated the diff for D48319: [DebugInfo] Make sure all DBG_VALUEs' reguse operands have IsDebug property.

Thanks Adrian for your feedback. The new diff includes refactorings to use MachineInstrBuilder construction functions where possible.

Jun 20 2018, 5:30 AM

Jun 19 2018

JesperAntonsson created D48319: [DebugInfo] Make sure all DBG_VALUEs' reguse operands have IsDebug property.
Jun 19 2018, 5:45 AM

May 7 2018

JesperAntonsson added a comment to D46404: [DAG] Avoid using deleted node in rebuildSetCC.

This works for me too and looks good (but again, I don't know the framework very well). Thanks!

May 7 2018, 2:34 AM

Dec 8 2017

JesperAntonsson updated the diff for D40648: A few initializations to please Valgrind. NFC.

Updated according to Bjorns suggestions: added TiedTo initializations and made bool args initialize to false instead of 0.

Dec 8 2017, 2:12 AM

Dec 1 2017

JesperAntonsson updated the diff for D40639: Bail out of a SimplifyCFG switch table opt at undef values..

The testcase could fail on other targets, since the label names could become different. Therefore, the checks were relaxed, since the test really only needs to verify that it doesn't crash and emits some code.

Dec 1 2017, 6:19 AM

Nov 30 2017

JesperAntonsson created D40648: A few initializations to please Valgrind. NFC.
Nov 30 2017, 5:16 AM
JesperAntonsson created D40639: Bail out of a SimplifyCFG switch table opt at undef values..
Nov 30 2017, 1:09 AM

Sep 14 2017

JesperAntonsson updated the diff for D37611: [IfConversion] More simple, correct dead/kill liveness handling.

Reverted the changes in the previous patch, except for the name improvements. I guess we can land this, then. I don't have the privileges yet, so I'm grateful for help with that. :-)

Sep 14 2017, 8:31 AM
JesperAntonsson added a comment to D37611: [IfConversion] More simple, correct dead/kill liveness handling.

Ah, thanks. Then the fault is in our backend, that we somehow miss to add that imp-use in some circumstances. My previous patch is probably better, then. I'll reupload that with Matthias suggested name changes. :-)

Sep 14 2017, 8:04 AM
JesperAntonsson updated the diff for D37611: [IfConversion] More simple, correct dead/kill liveness handling.

I found and fixed the reason our randomized tests didn't run cleanly. It was essentially that stepBackwards removed defined regs from the set even if the instruction was predicated. That meant a dominating def could be marked dead, which is wrong. The reason I found this is that I use the new recompute function also in a later stage in our out-of-tree backend, and that went bad when I started to add faulty deads. If I understand it correctly, this is a bug in the current implementation of LivePhysRegs.

Sep 14 2017, 7:43 AM

Sep 13 2017

JesperAntonsson added a comment to D37611: [IfConversion] More simple, correct dead/kill liveness handling.

This new implementation doesn't run cleanly with our extended test suite. I have to look a bit more into it.

Sep 13 2017, 6:40 AM
JesperAntonsson updated the diff for D37611: [IfConversion] More simple, correct dead/kill liveness handling.

This is a new patch that will handle the issue Krzysztof pointed out. Some more Hexagon tests affected. Makes me wonder: Is it ok to say that a predicated instruction's use of a register is killed if it is defined by the same instruction?

Sep 13 2017, 2:29 AM

Sep 11 2017

JesperAntonsson added inline comments to D37611: [IfConversion] More simple, correct dead/kill liveness handling.
Sep 11 2017, 9:28 AM
JesperAntonsson updated the diff for D37611: [IfConversion] More simple, correct dead/kill liveness handling.

Thanks for the reviews, everybody! Here I'm submitting an updated patch based on Matthias' suggestions.

Sep 11 2017, 7:18 AM

Sep 8 2017

JesperAntonsson created D37611: [IfConversion] More simple, correct dead/kill liveness handling.
Sep 8 2017, 1:17 AM

Aug 3 2017

JesperAntonsson updated the diff for D36224: [TwoAddressInstructionPass] Replace subregister uses when processing tied operands.

Hmm, sorry, I just realized that the check I introduced in the second patch made it unnecessary to set the subregister conditionally, it could just as well be set unconditionally to zero. So I'm doing this in this update.

Aug 3 2017, 7:21 AM
JesperAntonsson updated the diff for D36224: [TwoAddressInstructionPass] Replace subregister uses when processing tied operands.

Made sure that we don't replace register if the subregister is different.

Aug 3 2017, 6:39 AM
JesperAntonsson added a comment to D36224: [TwoAddressInstructionPass] Replace subregister uses when processing tied operands.

Krzysztof, thanks, I think it actually would replace it (though it would have before my patch too). I'll upload a patch that'll prevent it.

Aug 3 2017, 5:51 AM

Aug 2 2017

JesperAntonsson created D36224: [TwoAddressInstructionPass] Replace subregister uses when processing tied operands.
Aug 2 2017, 8:39 AM

Jul 3 2017

JesperAntonsson updated the diff for D34394: [MachineVerifier] Add check that tied physregs aren't different..

Added a MIR test for the check. Thanks to Quentin for input.

Jul 3 2017, 7:33 AM

Jun 28 2017

JesperAntonsson added a comment to D34394: [MachineVerifier] Add check that tied physregs aren't different..

Ping. How about it? Could we land this patch?

Jun 28 2017, 3:55 AM

Jun 21 2017

JesperAntonsson updated the diff for D34394: [MachineVerifier] Add check that tied physregs aren't different..

Removed the check that tied physregs and virtregs aren't mixed. Removed changes in a test that had such mixes.

Jun 21 2017, 2:26 AM
JesperAntonsson added inline comments to D34394: [MachineVerifier] Add check that tied physregs aren't different..
Jun 21 2017, 12:59 AM

Jun 20 2017

JesperAntonsson added a comment to D21468: Fixed LegacyPassManager inconsistency.

This fix is still relevant and would be nice to land, but 3 out of 4 related TR have gone into hiding/are non-reproducible after AssertingValueHandle functionality was changed. 27697 is still reproducible.

Jun 20 2017, 8:59 AM
JesperAntonsson updated subscribers of D34394: [MachineVerifier] Add check that tied physregs aren't different..

Missed adding llvm-commits as subscriber initially. Adding this note as a ping to the mailing list.

Jun 20 2017, 7:03 AM
JesperAntonsson created D34394: [MachineVerifier] Add check that tied physregs aren't different..
Jun 20 2017, 4:43 AM

Oct 10 2016

JesperAntonsson added a comment to D21468: Fixed LegacyPassManager inconsistency.

Pinging Mehdi Amini. :-)

Oct 10 2016, 6:49 AM

Oct 3 2016

JesperAntonsson added a comment to D21468: Fixed LegacyPassManager inconsistency.

Good that this isn't forgotten. :-) It's Monday, so I'm sending another friendly ping.

Oct 3 2016, 1:18 AM

Sep 12 2016

JesperAntonsson added a comment to D21468: Fixed LegacyPassManager inconsistency.

Monday, so I'm pinging. :-)

Sep 12 2016, 4:33 AM

Aug 16 2016

JesperAntonsson added a comment to D21468: Fixed LegacyPassManager inconsistency.

So, I'm back in business. :-) Mehdi, what say you?

Aug 16 2016, 2:09 AM

Jul 8 2016

JesperAntonsson added a comment to D21468: Fixed LegacyPassManager inconsistency.

Whether this is the the right fix or band-aid is a fair question. :-) The problematic operation seems to be the transitive update of LastUser, and that's what I'm restraining in a minimal way to retain consistency with preservation info. OTOH, I don't understand/like the transitive update at all, but perhaps you understand its implications and what, if anything, would need to be done if it were to be removed entirely? There are other possibilities as well but I guess those would require more fundamental redesign.

Jul 8 2016, 1:47 AM

Jun 23 2016

JesperAntonsson updated the diff for D21468: Fixed LegacyPassManager inconsistency.

Thank you to Mehdi Amini for good review comments. I've updated the diff accordingly: Improved description, changed to a more appropriate container type for the book-keeping and also I added a C++ unittest.

Jun 23 2016, 1:57 AM

Jun 17 2016

JesperAntonsson retitled D21468: Fixed LegacyPassManager inconsistency from to Fixed LegacyPassManager inconsistency.
Jun 17 2016, 6:18 AM

Jun 16 2016

JesperAntonsson closed D20725: Make sure function passes are run.

rL272770

Jun 16 2016, 4:19 AM

Jun 8 2016

JesperAntonsson added a reviewer for D20725: Make sure function passes are run: patrik.h.hagglund.
Jun 8 2016, 6:09 AM
JesperAntonsson updated the diff for D20685: SLPVectorizer to handle GEP with differing constant index types.

Corrected test case in accordance to the review comments from Michael Zolotukhin. (Thanks Michael!)

Jun 8 2016, 5:48 AM

May 27 2016

JesperAntonsson retitled D20725: Make sure function passes are run from to Make sure function passes are run.
May 27 2016, 2:57 AM
JesperAntonsson added a comment to D20685: SLPVectorizer to handle GEP with differing constant index types.

Thanks! This is my first contribution and I don't have commit rights. Could you help me commit, please?

May 27 2016, 12:37 AM

May 26 2016

JesperAntonsson retitled D20685: SLPVectorizer to handle GEP with differing constant index types from to SLPVectorizer to handle GEP with differing constant index types.
May 26 2016, 9:12 AM