qcolombet (Quentin Colombet)
User

Projects

User does not belong to any projects.

User Details

User Since
Dec 17 2012, 10:03 AM (231 w, 4 d)

Recent Activity

Tue, May 23

qcolombet accepted D33408: MachineCSE: Respect interblock physreg liveness.
Tue, May 23, 1:16 PM

Fri, May 19

qcolombet added a comment to D31821: Remove redundant copy in recurrences.

Hi Taewook,

Fri, May 19, 2:03 PM

Thu, May 18

qcolombet added a comment to D30527: Replacing float with new class Fraction for LSR alternative way of resolving complex solution.

Please follow @scanon recommendation.

Thu, May 18, 12:18 PM
qcolombet accepted D32464: LivePhysRegs: Fix addLiveOutsNoPristines() for return blocks past PEI.

LGTM

Thu, May 18, 12:03 PM
qcolombet accepted D33317: [GlobalISel] IRTranslator: Translate ConstantStruct.

LGTM with nitpicks

Thu, May 18, 10:16 AM

Wed, May 17

qcolombet committed rL303313: Revert "[globalisel][tablegen] Import rules containing intrinsic_wo_chain.".
Revert "[globalisel][tablegen] Import rules containing intrinsic_wo_chain."
Wed, May 17, 4:31 PM
qcolombet accepted D33294: Only enable LiveRangeShrink for x86..

Thanks Dehao.
LGTM.

Wed, May 17, 1:10 PM
qcolombet added a comment to D31821: Remove redundant copy in recurrences.
The problem doesn't appear while we're in SSA [...]
Wed, May 17, 8:42 AM

Tue, May 16

qcolombet accepted D33181: Don't allow -optimize-regalloc=false with -regalloc given for anything other than 'fast'.
Tue, May 16, 1:39 PM

Mon, May 15

qcolombet requested changes to D33181: Don't allow -optimize-regalloc=false with -regalloc given for anything other than 'fast'.
Mon, May 15, 9:42 AM

Fri, May 12

qcolombet accepted D33084: [GISel]: Fix undefined behavior while accessing DefaultAction map.
Fri, May 12, 2:31 PM
qcolombet accepted D32746: [globalisel][legalizer] G_LOAD/G_STORE NarrowScalar should not emit G_GEP x, 0..

LGTM modulo a small refactoring to avoid code duplication.

Fri, May 12, 2:15 PM
qcolombet accepted D32791: [globalisel][tablegen] Add support for (set $dst, 1) and test X86's OptForSize predicate..

LGTM, nitpick below

Fri, May 12, 2:12 PM
qcolombet accepted D32861: [globalisel][tablegen] Demote OptForSize/OptForMinSize/ForCodeSize to per-function predicates..

LGTM

Fri, May 12, 2:08 PM
qcolombet accepted D32278: [globalisel][tablegen] Import rules containing intrinsic_wo_chain..

LGTM

Fri, May 12, 2:06 PM
qcolombet added inline comments to D32868: [globalisel][tablegen] Fix nullptr dereferences when getVRegDef() is called on a phys reg..
Fri, May 12, 1:59 PM
qcolombet accepted D33144: [RegisterBankInfo] Remove overly-agressive asserts.

LGTM, nitpick below

Fri, May 12, 12:58 PM
qcolombet accepted D33051: [GISel]: Add a getConstantFPVRegVal utility.

LGTM

Fri, May 12, 12:05 PM
qcolombet added a comment to D33084: [GISel]: Fix undefined behavior while accessing DefaultAction map.

Hi Aditya,

Fri, May 12, 12:04 PM
qcolombet requested changes to D33144: [RegisterBankInfo] Remove overly-agressive asserts.
Fri, May 12, 11:59 AM
qcolombet accepted D32715: shrink-wrap: fix shrink-wrapping for no-return paths.

LGTM

Fri, May 12, 11:31 AM

Thu, May 11

qcolombet accepted D32869: [globalisel][tablegen] Require that all registers between instructions of a match are virtual..

Thanks for the clarification, now I get it :).

Thu, May 11, 5:55 PM
qcolombet added a comment to D32869: [globalisel][tablegen] Require that all registers between instructions of a match are virtual..

Hi Daniel,

Thu, May 11, 3:22 PM
qcolombet added a comment to D32868: [globalisel][tablegen] Fix nullptr dereferences when getVRegDef() is called on a phys reg..

Hi Daniel,

Thu, May 11, 2:45 PM
qcolombet accepted D32304: Reduce printing values with default in MIR based codegen testing YAML.

LGTM

Thu, May 11, 10:48 AM

Wed, May 10

qcolombet committed rL302679: [AArch64][RegisterBankInfo] Change the default mapping of fp stores..
[AArch64][RegisterBankInfo] Change the default mapping of fp stores.
Wed, May 10, 8:33 AM

Tue, May 9

qcolombet added a comment to D32744: [globalisel] Improve legalizer DEBUG_ONLY output..

It does not sound to me like something we can ensure. I wouldn't pursue in that direction IMHO.

Tue, May 9, 11:49 AM

Mon, May 8

qcolombet committed rL302453: [AArch64][RegisterBankInfo] Change the default mapping of fp loads..
[AArch64][RegisterBankInfo] Change the default mapping of fp loads.
Mon, May 8, 11:29 AM
qcolombet committed rL302452: [AArch64][RegisterBankInfo] Fix mapping cost for GPR..
[AArch64][RegisterBankInfo] Fix mapping cost for GPR.
Mon, May 8, 11:29 AM
qcolombet added inline comments to D31262: MIParser/MIRPrinter: Compute block successors if not explicitely specified.
Mon, May 8, 8:06 AM

Fri, May 5

qcolombet committed rL302316: [RegisterBankInfo] Uniquely allocate instruction mapping..
[RegisterBankInfo] Uniquely allocate instruction mapping.
Fri, May 5, 4:01 PM

Thu, May 4

qcolombet committed rL302202: [RegisterBankInfo] Fix 80-col introduced in r293506..
[RegisterBankInfo] Fix 80-col introduced in r293506.
Thu, May 4, 3:56 PM
qcolombet committed rL302201: [GlobalISel] Add missing doxygen keyword for doxygen groups..
[GlobalISel] Add missing doxygen keyword for doxygen groups.
Thu, May 4, 3:56 PM
qcolombet added a comment to D32869: [globalisel][tablegen] Require that all registers between instructions of a match are virtual..

If that happens, doesn't that means we need to support it?

Thu, May 4, 3:55 PM
qcolombet added inline comments to D32304: Reduce printing values with default in MIR based codegen testing YAML.
Thu, May 4, 3:50 PM
qcolombet added a comment to D32744: [globalisel] Improve legalizer DEBUG_ONLY output..

Hi Daniel,

Thu, May 4, 2:10 PM
qcolombet accepted D32575: Don't try to legalize Intermediate instructions (with generic types).

LGTM

Thu, May 4, 2:04 PM
qcolombet accepted D32814: [GISel]: Add support to translate ConstantVectors.

LGTM

Thu, May 4, 1:57 PM
qcolombet accepted D31262: MIParser/MIRPrinter: Compute block successors if not explicitely specified.

Thanks Matthias.

Thu, May 4, 10:49 AM

Mon, May 1

qcolombet committed rL301841: [AArch64] Move GISel accessor initialization from TargetMachine to Subtarget..
[AArch64] Move GISel accessor initialization from TargetMachine to Subtarget.
Mon, May 1, 3:06 PM
qcolombet added inline comments to D32058: Add a section about simplifying .mir tests.
Mon, May 1, 10:40 AM
qcolombet added inline comments to D32058: Add a section about simplifying .mir tests.
Mon, May 1, 10:15 AM
qcolombet added a comment to D31262: MIParser/MIRPrinter: Compute block successors if not explicitely specified.

In a nutshell, I am saying that we shouldn't conflict two goals: editing and printing. Printing should be full fledged, editing should be minimal.

Mon, May 1, 10:10 AM
qcolombet added a comment to D31262: MIParser/MIRPrinter: Compute block successors if not explicitely specified.

I also think there is value in reducing the amount of information printed. In this specific case the information omitted is quite obvious IMO.

Mon, May 1, 10:06 AM
qcolombet accepted D31623: [AArch64] Remove AArch64AddressTypePromotion pass.

LGTM

Mon, May 1, 10:01 AM
qcolombet requested changes to D32304: Reduce printing values with default in MIR based codegen testing YAML.

Hi guys,

Mon, May 1, 7:56 AM
qcolombet added a comment to D32570: MachineFrameInfo: Track whether MaxCallFrameSize is computed yet.

Hi Matthias,

Mon, May 1, 7:49 AM

Thu, Apr 27

qcolombet added a comment to D32575: Don't try to legalize Intermediate instructions (with generic types).

Hi Aditya,

Thu, Apr 27, 9:28 AM

Apr 1 2017

qcolombet accepted D28680: [CodeGenPrep] move aarch64-type-promotion to CGP.

LGTM with a nitpick.

Apr 1 2017, 10:42 AM

Mar 31 2017

qcolombet added a comment to D31503: [GlobalISel]: Fix bug where we can report GISelFailure on erased instructions.

Sounds good to me.

Mar 31 2017, 6:44 PM
qcolombet committed rL299288: Revert "Feature generic option to setup start/stop-after/before".
Revert "Feature generic option to setup start/stop-after/before"
Mar 31 2017, 6:39 PM
qcolombet committed rL299287: Revert "Localizer fun".
Revert "Localizer fun"
Mar 31 2017, 6:39 PM
qcolombet committed rL299286: Revert "Instrument SDISel C++ patterns".
Revert "Instrument SDISel C++ patterns"
Mar 31 2017, 6:38 PM
qcolombet committed rL299285: [RegBankSelect] Support REG_SEQUENCE for generic mapping.
[RegBankSelect] Support REG_SEQUENCE for generic mapping
Mar 31 2017, 6:38 PM
qcolombet committed rL299283: Localizer fun.
Localizer fun
Mar 31 2017, 6:34 PM
qcolombet committed rL299284: Instrument SDISel C++ patterns.
Instrument SDISel C++ patterns
Mar 31 2017, 6:34 PM
qcolombet committed rL299282: Feature generic option to setup start/stop-after/before.
Feature generic option to setup start/stop-after/before
Mar 31 2017, 6:33 PM

Mar 30 2017

qcolombet added inline comments to D28680: [CodeGenPrep] move aarch64-type-promotion to CGP.
Mar 30 2017, 5:34 PM
qcolombet added a comment to D31503: [GlobalISel]: Fix bug where we can report GISelFailure on erased instructions.

Here's another idea I considered: have legalizeInstr return the instruction it failed on, in a std::pair<LegalizeResult, MachineInstr*>.

How would that work with a deleted instruction?

Mar 30 2017, 3:11 PM
qcolombet added a comment to D31503: [GlobalISel]: Fix bug where we can report GISelFailure on erased instructions.

I considered and dismissed speculatively printing: it's a pretty expensive operation

Mar 30 2017, 2:48 PM
qcolombet added a comment to D31503: [GlobalISel]: Fix bug where we can report GISelFailure on erased instructions.

Hi Aditya,

Mar 30 2017, 12:01 PM

Mar 29 2017

qcolombet accepted D27264: Refactor code to create getFallThrough method in MachineBasicBlock..

Please next time upload a patch with the full context (diff -U9999999).

Mar 29 2017, 4:36 PM
qcolombet added a comment to D31446: [GlobalISel]: Target specific MIs like REG_SEQUENCE shouldn't be Regbank selected.

*shouldn't

Mar 29 2017, 8:56 AM
qcolombet added a comment to D31446: [GlobalISel]: Target specific MIs like REG_SEQUENCE shouldn't be Regbank selected.

Actually, now, that I think about it, REG_SEQUENCE, if any, should be supported on AArch64.
Thus, we should need anything.

Mar 29 2017, 8:53 AM
qcolombet requested changes to D31446: [GlobalISel]: Target specific MIs like REG_SEQUENCE shouldn't be Regbank selected.

Hi Aditya,

Mar 29 2017, 7:58 AM

Mar 28 2017

qcolombet requested changes to D28680: [CodeGenPrep] move aarch64-type-promotion to CGP.
Mar 28 2017, 10:16 AM
qcolombet added a comment to D30998: [globalisel][tablegen] Move <Target>InstructionSelector declarations to anonymous namespaces.

That sounds good to me. Can we tack it onto the end of the main patch series though? Some of those patches change <Target>InstructionSelector and I'd like to avoid resolving the conflicts if I don't need to.

Mar 28 2017, 9:51 AM

Mar 27 2017

qcolombet added a comment to D31188: [AntiDepBreaker] Use liveins as well in StartBlock.

Oh you mean turn this IR test into a mir test?

Mar 27 2017, 10:34 AM
qcolombet accepted D31188: [AntiDepBreaker] Use liveins as well in StartBlock.

Hi Tim,

Mar 27 2017, 10:01 AM

Mar 25 2017

qcolombet added a comment to D30350: [LSR] Add a cap for reassociation of AllFixupsOutsideLoop type LSRUse to protect compile time.

It would be even better if we had a fix to the SCEVExpander's pathological behavior.

Mar 25 2017, 8:41 PM

Mar 24 2017

qcolombet added a comment to D30350: [LSR] Add a cap for reassociation of AllFixupsOutsideLoop type LSRUse to protect compile time.

Hi Wei,

Mar 24 2017, 11:58 PM
qcolombet added a comment to D31262: MIParser/MIRPrinter: Compute block successors if not explicitely specified.

Hi Matthias,

Mar 24 2017, 9:12 PM
qcolombet accepted D30996: Correct OptionCategoryCompare() in the command line library..

LGTM

Mar 24 2017, 7:00 PM
qcolombet added a comment to D30299: [tablegen] Optionally format tablegen targets with clang-format.

Hi Daniel,

Mar 24 2017, 6:59 PM

Mar 21 2017

qcolombet added inline comments to D31188: [AntiDepBreaker] Use liveins as well in StartBlock.
Mar 21 2017, 11:24 AM

Mar 16 2017

qcolombet added a comment to D30900: [PPC] Eliminate stack frame in non-leaf function based on shrink wrapping.

I haven't looked at the patch proper yet.
Inlined a couple of stylistic comment.

Mar 16 2017, 6:23 PM
qcolombet accepted D27853: [CodeGenPrep]Restructure promoting Ext to form ExtLoad.

Hi Jun,

Mar 16 2017, 6:00 PM
qcolombet accepted D30925: VirtRegMap: Correctly deal with bundles when deleting identity copies..

LGTM

Mar 16 2017, 4:58 PM
qcolombet accepted D30438: SplitKit: Correctly implement partial copies.

LGTM with more checks in the test.

Mar 16 2017, 3:55 PM
qcolombet added a comment to D30751: [MachineCopyPropagation] Extend pass to do COPY source forwarding.

This change is useful primarily to increase scheduling flexibility and reducing the critical path, though it does also make some COPYs unnecessary leading to their removal. Handling just the former in the scheduler is a possibility, but it would have the drawback of not provide a benefit to OoO cores that don't do post-RA scheduling.

Mar 16 2017, 3:51 PM

Mar 15 2017

qcolombet added a comment to D30808: shrink-wrap: implement more advanced algorithm.
I actually think that I should look into that (maybe I should have gathered those numbers from the beginning?).
Mar 15 2017, 2:12 PM
qcolombet added a comment to D30808: shrink-wrap: implement more advanced algorithm.
I couldn't think of a way to produce multiple save / restore points for one register (which seems to be the feature that we're the most interested in?) using dominators (I could look into it if you have any ideas, of course).
Mar 15 2017, 10:23 AM
qcolombet added a comment to D30973: [GlobalISel][X86] Support G_STORE/G_LOAD operation.

Hi Igor,

Mar 15 2017, 7:42 AM
qcolombet added a comment to D30808: shrink-wrap: implement more advanced algorithm.

Hi Francis,

Mar 15 2017, 7:36 AM

Mar 14 2017

qcolombet accepted D30948: [GlobalISel] IRTranslator: Return the scalar for <1 x Ty> constant vectors.

Nice catch.

Mar 14 2017, 4:54 PM
qcolombet added a comment to D30913: [NFC] Feature generic options to setup start/stop-after/before.

Hi Matthias,

Mar 14 2017, 11:21 AM
qcolombet requested changes to D30562: Add Instruction number to LSR cost model (PR23384) part 3 of 3.

I see two different things in that patch:

  1. X86 cost model change
  2. GenerateCrossUseICmpZero
Mar 14 2017, 10:59 AM
qcolombet accepted D30561: Add Instruction number to LSR cost model (PR23384) part 2 of 3.

LGTM.

Mar 14 2017, 10:55 AM

Mar 13 2017

qcolombet created D30913: [NFC] Feature generic options to setup start/stop-after/before.
Mar 13 2017, 2:57 PM

Mar 10 2017

qcolombet added inline comments to D30259: GlobalISel: Translate ConstantAggregateZero vectors.
Mar 10 2017, 4:42 PM
qcolombet committed rL297535: [IRTranslator] Simplify error handling for translating constants. NFC..
[IRTranslator] Simplify error handling for translating constants. NFC.
Mar 10 2017, 4:40 PM
qcolombet added a comment to D27853: [CodeGenPrep]Restructure promoting Ext to form ExtLoad.

Thanks for your patience.

Mar 10 2017, 4:16 PM
qcolombet accepted D30089: [globalisel][tblgen] Add support for ComplexPatterns.

The constructor call comes from the need to call addOperand(). If we reset the operand list for the placeholder instruction on each call to selectImpl() then we would need something like:

PlaceholderInsn.addOperand(MachineOperand::CreatePlaceholder());
PlaceholderInsn.addOperand(MachineOperand::CreatePlaceholder());
if (matches && selectComplexPattern(PlaceholderInsn.getOperand(0), PlaceholderInsn.getOperand(1))) {

BuildMI(...).add(PlaceholderInsn.getOperand(0)).add(PlaceholderInsn.getOperand(1));
...

}
PlaceholderInsn.resetOperands();
Each call to addOperand() involves a placement new operator which uses the copy-constructor and therefore pays the same overhead as the current local variable approach.

Mar 10 2017, 3:19 PM
qcolombet added a comment to D30521: Introduce llc/ExecuteTestCommands pass.

Hi Matthias,

Mar 10 2017, 11:19 AM
qcolombet accepted D30838: [GlobalISel] Make LegalizerInfo accessible in LegalizerHelper.

Make sense to me.
The helper is already bound to a MachineFunction, so having the LegalizerInfo directly accessible is sensible IMHO.

Mar 10 2017, 10:44 AM
qcolombet accepted D30259: GlobalISel: Translate ConstantAggregateZero vectors.

LGTM

Mar 10 2017, 10:00 AM
qcolombet accepted D30216: GlobalISel: Translate ConstantDataVector.

LGTM + test case

Mar 10 2017, 9:57 AM
qcolombet added inline comments to D30786: [GlobalISel] LegalizerHelper: Lower (G_FSUB X, Y) to (G_FADD X, (G_FNEG Y)).
Mar 10 2017, 9:52 AM
qcolombet added inline comments to D30786: [GlobalISel] LegalizerHelper: Lower (G_FSUB X, Y) to (G_FADD X, (G_FNEG Y)).
Mar 10 2017, 9:39 AM

Mar 9 2017

qcolombet accepted D30786: [GlobalISel] LegalizerHelper: Lower (G_FSUB X, Y) to (G_FADD X, (G_FNEG Y)).

LGTM with nitpicks

Mar 9 2017, 6:29 PM