rovka (Diana Picus)
User

Projects

User does not belong to any projects.

User Details

User Since
Apr 21 2016, 3:27 AM (82 w, 3 d)

Recent Activity

Thu, Nov 16

rovka committed rL318398: [ARM GlobalISel] Add tests for BIC. NFC.
[ARM GlobalISel] Add tests for BIC. NFC
Thu, Nov 16, 5:33 AM
rovka committed rL318393: [ARM GlobalISel] Add tests for REVSH patterns. NFC.
[ARM GlobalISel] Add tests for REVSH patterns. NFC
Thu, Nov 16, 4:30 AM
rovka added a comment to D40074: [GISel] Canonicalize constants to RHS for commutative operations.

If this is the first canonicalization (of potentially many to come) in the IRTranslator, would it make sense to try and document the canonical representation(s) somewhere and/or somehow?

Even better would be a pass that verified canonical order, that could be inserted between other passes in debug mode to verify that optimisations preserve canonical-ness

Thu, Nov 16, 3:45 AM

Wed, Nov 15

rovka created D40074: [GISel] Canonicalize constants to RHS for commutative operations.
Wed, Nov 15, 5:51 AM

Tue, Nov 14

rovka committed rL318146: [ARM GlobalISel] Remove C++ code for G_CONSTANT.
[ARM GlobalISel] Remove C++ code for G_CONSTANT
Tue, Nov 14, 3:20 AM

Mon, Nov 13

rovka added a comment to D39267: [GISel]: Change Legalization from top down to bottom up + DCE.

Hi, I've updated the ARM tests, and you'll need to also commit this together with your patch: https://reviews.llvm.org/differential/diff/122657/

Mon, Nov 13, 8:18 AM
rovka committed rL318047: [ARM GlobalISel] Update legalizer test.
[ARM GlobalISel] Update legalizer test
Mon, Nov 13, 8:03 AM

Fri, Nov 10

rovka added inline comments to D39808: [ARM GlobalISel] Fix call opcode for Thumb .
Fri, Nov 10, 10:04 AM

Tue, Nov 7

rovka accepted D39747: [globalisel][tablegen] Generate rule coverage and use it to identify untested rules.

Awesome, I could use something like this. LGTM with a few nits.

Tue, Nov 7, 10:44 PM
rovka added a comment to D39267: [GISel]: Change Legalization from top down to bottom up + DCE.

Sure, will do.

Tue, Nov 7, 8:47 PM
rovka added inline comments to D39592: [ARM|GlobalISel] : Adding legalizer tests for Thumb.
Tue, Nov 7, 4:04 AM

Mon, Nov 6

rovka added inline comments to D39592: [ARM|GlobalISel] : Adding legalizer tests for Thumb.
Mon, Nov 6, 3:05 AM

Fri, Nov 3

rovka added a comment to D39592: [ARM|GlobalISel] : Adding legalizer tests for Thumb.

I think it would be simpler to add a RUN line to the existing test (you can rename it to just legalizer.mir if you think the current name would be misleading).

Fri, Nov 3, 8:52 AM
rovka added a comment to D39592: [ARM|GlobalISel] : Adding legalizer tests for Thumb.

Hi Javed,

Fri, Nov 3, 6:10 AM
rovka committed rL317313: [globalisel][tablegen] Skip src child predicates.
[globalisel][tablegen] Skip src child predicates
Fri, Nov 3, 3:30 AM
rovka closed D39554: [globalisel][tablegen] Skip src child predicates by committing rL317313: [globalisel][tablegen] Skip src child predicates.
Fri, Nov 3, 3:30 AM
rovka committed rL317312: [ARM GlobalISel] Move the check for Thumb higher up.
[ARM GlobalISel] Move the check for Thumb higher up
Fri, Nov 3, 3:30 AM

Thu, Nov 2

rovka added a comment to D39554: [globalisel][tablegen] Skip src child predicates.

Sorry about the indirect test, but it's a bit difficult to test directly since this is restricting functionality rather than adding it. Are we interested in having tests for failed imports? I haven't seen any, but I may have missed them...

Thu, Nov 2, 8:14 AM
rovka created D39554: [globalisel][tablegen] Skip src child predicates.
Thu, Nov 2, 8:09 AM

Wed, Nov 1

rovka added a comment to D39415: [ARMISelLowering] Better handling of NEON load/store for sequential memory regions.

I tried to give this a run [1] on top of r317072 on an NVIDIA TK1 (Cortex-A15) and I'm getting some failures in the test-suite (in 42 of the benchmarks) along the lines of [2]. I think we should fix those before worrying about performance numbers...

Wed, Nov 1, 6:24 AM

Mon, Oct 30

rovka accepted D39261: [GlobalISel|ARM] : Allow legalizing G_FSUB.

LGTM, thanks!

Mon, Oct 30, 5:00 AM
rovka committed rL316897: [ARM GlobalISel] Fixup r316572. NFC.
[ARM GlobalISel] Fixup r316572. NFC
Mon, Oct 30, 4:58 AM

Thu, Oct 26

rovka added inline comments to D39264: [GISel]: Make the combiner check if shifts are legal before creating them.
Thu, Oct 26, 2:58 AM

Wed, Oct 25

rovka committed rL316573: [ARM GlobalISel] Remove redundant testcases. NFC.
[ARM GlobalISel] Remove redundant testcases. NFC
Wed, Oct 25, 5:22 AM
rovka committed rL316572: [ARM GlobalISel] Update test after r316479. NFC.
[ARM GlobalISel] Update test after r316479. NFC
Wed, Oct 25, 5:22 AM
rovka committed rL316570: [ARM GlobalISel] Fix call opcodes.
[ARM GlobalISel] Fix call opcodes
Wed, Oct 25, 4:43 AM
rovka committed rL316569: [ARM GlobalISel] Split test into 3. NFC.
[ARM GlobalISel] Split test into 3. NFC
Wed, Oct 25, 4:21 AM
rovka accepted D39264: [GISel]: Make the combiner check if shifts are legal before creating them.

LGTM with nit.

Wed, Oct 25, 1:16 AM
rovka added a comment to D39261: [GlobalISel|ARM] : Allow legalizing G_FSUB.

Hi Javed,

Wed, Oct 25, 1:03 AM

Tue, Oct 24

rovka committed rL316438: Switch ARM & AArch64 test-suite builders to CMake.
Switch ARM & AArch64 test-suite builders to CMake
Tue, Oct 24, 7:05 AM

Oct 19 2017

rovka added a comment to D37775: Add a verifier test to check the access on both sides of COPY are the same.

Hi, here's the patch for the ARM GlobalISel tests: https://reviews.llvm.org/differential/diff/119563/
Let me know if there's anything else that needs addressing on ARM.

Oct 19 2017, 4:14 AM
rovka committed rL316155: [ARM GlobalISel] Fix liveins in test. NFC.
[ARM GlobalISel] Fix liveins in test. NFC
Oct 19 2017, 2:29 AM
rovka committed rL316154: [ARM GlobalISel] Remove redundant tests.
[ARM GlobalISel] Remove redundant tests
Oct 19 2017, 1:51 AM

Oct 18 2017

rovka added a comment to D38918: Update successor after branch relaxation in ARM.

I only cared about not having the Dest BB as a successor because it might cause scheduling issues and\or labels not being emitted, having extra successor wont.

Oct 18 2017, 2:08 AM

Oct 17 2017

rovka added a comment to D37775: Add a verifier test to check the access on both sides of COPY are the same.

Sorry, I just noticed this. I'll have a look at the ARM side of things.

Oct 17 2017, 1:56 AM
rovka added a comment to D38918: Update successor after branch relaxation in ARM.

This looks correct, but is there anything removing DestBB from the newly created block's successors if necessary?

Oct 17 2017, 1:33 AM

Oct 12 2017

rovka committed rL315579: MachineInstr: Make isEqual agree with getHashValue in….
MachineInstr: Make isEqual agree with getHashValue in…
Oct 12 2017, 7:00 AM
rovka closed D38789: MachineInstr: Make isEqual agree with getHashValue in MachineInstrExpressionTrait by committing rL315579: MachineInstr: Make isEqual agree with getHashValue in….
Oct 12 2017, 7:00 AM
rovka updated the diff for D38789: MachineInstr: Make isEqual agree with getHashValue in MachineInstrExpressionTrait.

Add unit tests and update summary before commit.

Oct 12 2017, 6:49 AM

Oct 11 2017

rovka updated the diff for D38789: MachineInstr: Make isEqual agree with getHashValue in MachineInstrExpressionTrait.

Update to take into account isDef only when ignoring virtual defs, since this is what the getHashValue function assumes.
First of all, this changes from using isPhysicalRegister to using !isVirtualRegisters, which also includes the sentinel values that we're interested in.
Secondly, it changes from checking MO.getReg to checking MO.isIdenticalTo, which takes into account MO.getReg, MO.getSubReg and MO.isDef. These are the same things that the hash function combines when getting the hashes.

Oct 11 2017, 9:11 AM
rovka added a comment to D38789: MachineInstr: Make isEqual agree with getHashValue in MachineInstrExpressionTrait.

Here's a comment from include/llvm/CodeGen/MachineInstr.h (definition of MachineInstrExpressionTrait)

/// Special DenseMapInfo traits to compare MachineInstr* by *value* of the
/// instruction rather than by pointer value.
/// The hashing and equality testing functions ignore definitions so this is
/// useful for CSE, etc.

The isEqual function explicitly requests IgnoreVRegDefs, supposedly on the basis that getHashValue would do the same. Your observation is that this is not the case. Assuming that this comment was true at some point, it may be worthwhile to check how the deviation was introduced.

Oct 11 2017, 7:48 AM
rovka added a comment to D38789: MachineInstr: Make isEqual agree with getHashValue in MachineInstrExpressionTrait.

This is a latent issue that triggers on one of the buildbots if the planets align. I'm holding one of Sanjay's InstCombine patches (r314698) hostage because of this bug, so please have a look. Thanks!

Oct 11 2017, 2:24 AM
rovka created D38789: MachineInstr: Make isEqual agree with getHashValue in MachineInstrExpressionTrait.
Oct 11 2017, 2:22 AM

Oct 6 2017

rovka added a comment to D37640: [GISel]: Fix generation of illegal COPYs (of different sizes) during CallLowering .

Hi Aditya,

Oct 6 2017, 11:06 AM
rovka committed rL315083: [ARM] GlobalISel: Make tests less strict.
[ARM] GlobalISel: Make tests less strict
Oct 6 2017, 10:49 AM
rovka committed rL315071: [ARM] GlobalISel: Select shifts.
[ARM] GlobalISel: Select shifts
Oct 6 2017, 8:41 AM
rovka committed rL315067: [ARM] GlobalISel: Map shift operands to GPRs.
[ARM] GlobalISel: Map shift operands to GPRs
Oct 6 2017, 7:54 AM
rovka committed rL315064: [ARM] GlobalISel: Mark shifts as legal for s32.
[ARM] GlobalISel: Mark shifts as legal for s32
Oct 6 2017, 7:31 AM

Oct 3 2017

rovka added a comment to D37640: [GISel]: Fix generation of illegal COPYs (of different sizes) during CallLowering .

Could someone familiar with ARM codebase implement that?

Oct 3 2017, 3:34 AM

Sep 6 2017

rovka committed rL312647: [zorg] Remove clang-cmake-aarch64-42vma builder.
[zorg] Remove clang-cmake-aarch64-42vma builder
Sep 6 2017, 9:35 AM
rovka closed D37459: [buildbot] Remove clang-cmake-aarch64-42vma builder by committing rL312647: [zorg] Remove clang-cmake-aarch64-42vma builder.
Sep 6 2017, 9:35 AM

Sep 5 2017

rovka committed rL312522: [ARM] GlobalISel: Minor cleanups in inst selector.
[ARM] GlobalISel: Minor cleanups in inst selector
Sep 5 2017, 1:24 AM
rovka committed rL312521: [ARM] GlobalISel: Support global variables for RWPI.
[ARM] GlobalISel: Support global variables for RWPI
Sep 5 2017, 12:59 AM

Sep 1 2017

rovka committed rL312323: [ARM] GlobalISel: Support ROPI global variables.
[ARM] GlobalISel: Support ROPI global variables
Sep 1 2017, 4:15 AM
rovka committed rL312320: Move static helper into ARMTargetLowering. NFC.
Move static helper into ARMTargetLowering. NFC
Sep 1 2017, 3:46 AM
rovka committed rL312319: [ARM] GlobalISel: More tests. NFC..
[ARM] GlobalISel: More tests. NFC.
Sep 1 2017, 3:20 AM

Aug 30 2017

rovka added a comment to D36534: [aarch64] Support APInt and APFloat in ImmLeaf subclasses and make AArch64 use them..

I can see why you need to change the patterns to be more general, but is the distinction between APInt, APFloat and 64-bit values really relevant? I looked a bit at the following patch which uses this for GlobalISel and I'm wondering if we could just get away with checking for each immediate a combination of isCImm, isFPImm and getBitWidth (so we wouldn't need to introduce different states for different kinds of immediates).

Aug 30 2017, 7:44 AM

Aug 29 2017

rovka committed rL311992: [ARM] GlobalISel: Select globals in PIC mode.
[ARM] GlobalISel: Select globals in PIC mode
Aug 29 2017, 2:49 AM
rovka closed D36507: [ARM] GlobalISel: Select globals in PIC mode by committing rL311992: [ARM] GlobalISel: Select globals in PIC mode.
Aug 29 2017, 2:49 AM
rovka committed rL311989: [ARM] GlobalISel: Rename tests. NFC..
[ARM] GlobalISel: Rename tests. NFC.
Aug 29 2017, 2:02 AM

Aug 24 2017

rovka committed rL311638: clang-native-arm-lnt-perf: Move to test-suite producer.
clang-native-arm-lnt-perf: Move to test-suite producer
Aug 24 2017, 1:41 AM

Aug 22 2017

rovka accepted D36086: [globalisel][tablegen] Add support for ImmLeaf without SDNodeXForm.

Thanks for the updates, this looks really good now!

Aug 22 2017, 7:09 PM

Aug 21 2017

rovka added a comment to D36507: [ARM] GlobalISel: Select globals in PIC mode.

@t.p.northover Could you have a quick look at this, please?

Aug 21 2017, 9:47 AM

Aug 18 2017

rovka committed rL311151: Revert "GlobalISel (AArch64): fix ABI at border between GPRs and SP.".
Revert "GlobalISel (AArch64): fix ABI at border between GPRs and SP."
Aug 18 2017, 2:32 AM

Aug 17 2017

rovka added a comment to D36086: [globalisel][tablegen] Add support for ImmLeaf without SDNodeXForm.

This should have a summary.

Aug 17 2017, 9:17 AM

Aug 11 2017

rovka accepted D36085: [globalisel][tablegen] Generate TypeObject table. NFC.

LGTM now.

Aug 11 2017, 2:06 AM
rovka added a comment to D36084: [globalisel][tablegen] Support zero-instruction emission..

I had a testcase for this one but it seems to have gone missing. I'll update with one shortly

Aug 11 2017, 1:55 AM

Aug 10 2017

rovka added a comment to D36576: Fix -fPIC handling on arm64.

Thanks for sending the patch.

Aug 10 2017, 6:14 AM
rovka updated the diff for D36507: [ARM] GlobalISel: Select globals in PIC mode.

Thanks for the comments! I've updated the code to never use MOVT for ELF. I've also reworded the FIXME and mentioned the bug report in it.

Aug 10 2017, 4:23 AM

Aug 9 2017

rovka committed rL310464: Revert "PR19668, PR23034: Fix handling of move constructors and deleted copy….
Revert "PR19668, PR23034: Fix handling of move constructors and deleted copy…
Aug 9 2017, 5:23 AM
rovka created D36507: [ARM] GlobalISel: Select globals in PIC mode.
Aug 9 2017, 2:22 AM

Aug 3 2017

rovka added a comment to D35833: [globalisel][tablegen] Add support for importing 'imm' operands..

Well, considering that Mips prefers SExt and nothing (that we're aware of) prefers ZExt, it might be better to just use SExt from the start. But if you're feeling lazy ZExt is fine too :)

Aug 3 2017, 6:27 AM
rovka committed rL309927: [ARM] GlobalISel: Select simple G_GLOBAL_VALUE instructions.
[ARM] GlobalISel: Select simple G_GLOBAL_VALUE instructions
Aug 3 2017, 2:15 AM
rovka closed D35883: [ARM] GlobalISel: Select simple G_GLOBAL_VALUE instructions by committing rL309927: [ARM] GlobalISel: Select simple G_GLOBAL_VALUE instructions.
Aug 3 2017, 2:15 AM

Aug 2 2017

rovka added inline comments to D36219: [ARM] Tidy up banked registers encoding.
Aug 2 2017, 7:34 AM
rovka added inline comments to D35833: [globalisel][tablegen] Add support for importing 'imm' operands..
Aug 2 2017, 7:22 AM
rovka added a comment to D35883: [ARM] GlobalISel: Select simple G_GLOBAL_VALUE instructions.

Ping

Aug 2 2017, 7:16 AM
rovka added a comment to D36085: [globalisel][tablegen] Generate TypeObject table. NFC.

You should add checks for the GILLT enum in GlobalISelEmitter.td, since now it won't be the same enum for all targets.

Aug 2 2017, 7:05 AM
rovka added a comment to D36084: [globalisel][tablegen] Support zero-instruction emission..

Hi Daniel,

Aug 2 2017, 5:51 AM
rovka committed rL309806: [MIR] Print target-specific constant pools.
[MIR] Print target-specific constant pools
Aug 2 2017, 4:10 AM
rovka closed D36092: [MIR] Print target specific constant pools by committing rL309806: [MIR] Print target-specific constant pools.
Aug 2 2017, 4:10 AM
rovka updated the diff for D36092: [MIR] Print target specific constant pools.

Refresh before commit. Now including a test for the error-path!

Aug 2 2017, 3:04 AM

Aug 1 2017

rovka updated the diff for D36092: [MIR] Print target specific constant pools.

Thanks for the detailed response!

Aug 1 2017, 4:31 AM

Jul 31 2017

rovka created D36092: [MIR] Print target specific constant pools.
Jul 31 2017, 8:07 AM

Jul 27 2017

rovka accepted D35884: Update to use enum classes for various ARM *Kind enums.

Looks entirely mechanical, I don't see any problem with it (just a couple of nits).

Jul 27 2017, 8:00 AM
rovka accepted D35882: [TargetParser] Use enum classes for various ARM kind enums..

LGTM.

Jul 27 2017, 7:45 AM
rovka added a comment to D29840: test-release.sh: Remove workaround for test-suite build.

Hi Tom,

Jul 27 2017, 6:43 AM
rovka added inline comments to D35882: [TargetParser] Use enum classes for various ARM kind enums..
Jul 27 2017, 5:15 AM
rovka accepted D35833: [globalisel][tablegen] Add support for importing 'imm' operands..

Seems fine.

Jul 27 2017, 4:15 AM

Jul 26 2017

rovka committed rL309101: [ARM] GlobalISel: Map G_GLOBAL_VALUE to GPR.
[ARM] GlobalISel: Map G_GLOBAL_VALUE to GPR
Jul 26 2017, 4:01 AM
rovka created D35883: [ARM] GlobalISel: Select simple G_GLOBAL_VALUE instructions.
Jul 26 2017, 2:33 AM
rovka committed rL309090: [ARM] GlobalISel: Mark G_GLOBAL_VALUE as legal.
[ARM] GlobalISel: Mark G_GLOBAL_VALUE as legal
Jul 26 2017, 2:27 AM

Jul 25 2017

rovka accepted D35681: [globalisel][tablegen] Fuse the generated tables together..

Cool, LGTM, thanks.

Jul 25 2017, 6:11 AM
rovka added inline comments to D35681: [globalisel][tablegen] Fuse the generated tables together..
Jul 25 2017, 3:30 AM

Jul 20 2017

rovka committed rL308603: Revert "GlobalISel: select G_EXTRACT and G_INSERT instructions on AArch64.".
Revert "GlobalISel: select G_EXTRACT and G_INSERT instructions on AArch64."
Jul 20 2017, 4:36 AM

Jul 19 2017

rovka accepted D35117: [globalisel][tablegen] Add control-flow to the MatchTable..

LGTM with minor nits.

Jul 19 2017, 3:04 AM

Jul 18 2017

rovka committed rL308285: [ARM] GlobalISel: Support G_(S|U)REM for s8 and s16.
[ARM] GlobalISel: Support G_(S|U)REM for s8 and s16
Jul 18 2017, 3:07 AM
rovka committed rL308278: GlobalISel: Support G_(S|U)REM widening in LegalizerHelper.
GlobalISel: Support G_(S|U)REM widening in LegalizerHelper
Jul 18 2017, 2:09 AM

Jul 17 2017

rovka added a comment to D35117: [globalisel][tablegen] Add control-flow to the MatchTable..

I think the commit message should elaborate a bit on why we want a single table.
Other nits:

Jul 17 2017, 4:46 AM
rovka committed rL308165: Set higher timeouts for Misc-C++ on AArch64.
Set higher timeouts for Misc-C++ on AArch64
Jul 17 2017, 2:04 AM