Page MenuHomePhabricator

Gerolf (Gerolf Hoflehner)
User

Projects

User does not belong to any projects.

User Details

User Since
Apr 16 2014, 6:04 PM (287 w, 5 d)

Recent Activity

Aug 21 2019

Gerolf added reviewers for D61259: AArch64: support arm64_32, an ILP32 slice for watchOS.: aemerson, thegameg, paquette.
Aug 21 2019, 2:13 PM · Restricted Project

Jun 26 2019

Gerolf created D63855: [ADCE] Fix for non-determistic uselists after dead block removal.
Jun 26 2019, 5:16 PM · Restricted Project
Gerolf committed rGe311a4d5c4d3: [SCCP] Fix non-deterministic uselists of return values (DenseMap -> MapVector) (authored by Gerolf).
[SCCP] Fix non-deterministic uselists of return values (DenseMap -> MapVector)
Jun 26 2019, 2:48 PM
Gerolf added reviewers for D61259: AArch64: support arm64_32, an ILP32 slice for watchOS.: eli.friedman, kristof.beyls, rovka, fhahn, aadg.

+ reviewers to move this forward hopefully

Jun 26 2019, 1:25 PM · Restricted Project

Apr 16 2019

Gerolf added a comment to D60690: [AArch64] Skip mask checks for masks with an odd number of elements..

Could you separate size one checks are OK vs. checks that require even powers of 2? That smells like an inconsistency worth calling out explicitly.

Apr 16 2019, 3:31 PM · Restricted Project

Mar 18 2019

Gerolf added reviewers for D59507: [DAGCombine] Fix a miscompile when reducing BUILD_VECTORs to a shuffle: aemerson, paquette.

Looks reasonable. Added more reviewers for second opinions.

Mar 18 2019, 5:29 PM · Restricted Project

Dec 17 2018

Gerolf added reviewers for D55775: [Driver] Don't override '-march' when using '-arch x86_64h': ahatanak, qcolombet.

+ Akira, Quentin for their driver / x86_64h experience for a quick double check. Fwiw, LGTM.

Dec 17 2018, 11:23 AM

Dec 14 2018

Gerolf accepted D54743: SROA: preserve alignment tags on loads and stores..

This is straightforward way to pass along alignment tags and fixes (at least one critical) bug. LGTM.

Dec 14 2018, 5:25 PM · Restricted Project

Dec 13 2018

Gerolf added inline comments to D54743: SROA: preserve alignment tags on loads and stores..
Dec 13 2018, 1:37 PM · Restricted Project
Gerolf accepted D55573: [MachO][TLOF] Add support for local symbols in the indirect symbol table.

LGTM

Dec 13 2018, 9:12 AM

Dec 12 2018

Gerolf added inline comments to D55573: [MachO][TLOF] Add support for local symbols in the indirect symbol table.
Dec 12 2018, 7:38 PM

Oct 23 2018

Gerolf added a reviewer for D53535: [MC] Separate masm integer literal lexer support from inline asm: thegameg.
Oct 23 2018, 5:26 PM

Jul 24 2018

Gerolf updated subscribers of D48925: X86: add alias for pushfw/popfw in Intel mode.

+ David, Zia

Jul 24 2018, 8:10 PM

Jun 13 2018

Gerolf added inline comments to D47655: [MachineOutliner] Don't outline sequences where x16/x17/nzcv are live across.
Jun 13 2018, 2:26 PM

May 8 2018

Gerolf updated subscribers of D46600: [MergedLoadStoreMotion] Fix a debug invariant bug in mergeStores.

The test case could possibly be shorter, but the change LGTM.

May 8 2018, 1:48 PM

Feb 22 2018

Gerolf added a comment to D43177: Mark MergedLoadStoreMotion as not preserving MemDep results.

Yes, this LGTM. I could not find a way to perform updates properly in the example. If recomputing the dependencies results in a compile-time issue we will have to dig deeper into MemDep.

Feb 22 2018, 10:34 AM

Feb 21 2018

Gerolf added inline comments to D40602: [X86] Add MC level selection support for SHLD (64-bit only).
Feb 21 2018, 10:57 AM

Jan 31 2018

Gerolf added a comment to D42612: [GlobalOpt] Improve common case efficiency of static global initializer evaluation.

I assume you will take care of the comment. LGTM.

Jan 31 2018, 1:10 PM

Jan 30 2018

Gerolf added a comment to D42612: [GlobalOpt] Improve common case efficiency of static global initializer evaluation.

This is very close now. Could you add an explicit examples (eg show the IR) showing which initialization remain slow (Complex) and which are fast now? This should also address the spirit of Adrian's question I think.

Jan 30 2018, 8:55 PM

Jan 29 2018

Gerolf added a comment to D41766: [MachineCombiner] Add check for optimal pattern order..

I like the spirit of the idea. What made you look into this? Some more questions and suggestions below, but LGTM as is.

Jan 29 2018, 8:47 PM
Gerolf added a comment to D42612: [GlobalOpt] Improve common case efficiency of static global initializer evaluation.

Thank you drilling into this! I have a few questions below. Also, could you comment on the time savings you measured for your implementation?

Jan 29 2018, 6:49 PM

Jan 4 2018

Gerolf added a reviewer for D41672: support phi ranges for machine-level IR: MatzeB.

+ Matthias for review & thoughts about the unit test.

Jan 4 2018, 11:15 AM

Jan 2 2018

Gerolf added a comment to D41362: [AArch64][GlobalISel] Enable GlobalISel at -O0 by default.

Thanks! This LGTM then.

Jan 2 2018, 7:03 AM
Gerolf added inline comments to D41278: [MachineCombiner] Improve debug output (NFC).
Jan 2 2018, 2:22 AM
Gerolf added a comment to D41362: [AArch64][GlobalISel] Enable GlobalISel at -O0 by default.

I see all issues that came up in this thread covered by the last patch. Before this can be committed I still want to check that all paths/issues are tested/covered wrt to fast-isel:
a) how do we guarantee that there is no fallback from GISel to FastISel (when GISel is supported)? This is probably a nit since it is obvious to everyone deeper in the implementation then I am.
b) for all the tests where fast-isel was added shouldn't there equivalent tests for GISel, too? Even if the tests target fast-isel specific issues how do we make sure GISel does not expose similar/same bugs?

Jan 2 2018, 1:41 AM

Dec 5 2017

Gerolf added reviewers for D40864: [Darwin] Add a new -mstack-probe option and enable by default: dexonsmith, ahatanak, MatzeB.

It looks pretty straightforward, but I'd ask (at least) Duncan or Akira, and Matthias to review this more carefully.

Dec 5 2017, 5:07 PM
Gerolf added a comment to D40307: [MachineCombiner] Add up latencies of all instructions in new pattern..

I think your commit makes accounting for both cases - the instructions inserted and the instructions deleted - consistent. From that angle this look OK. I have also some food for thought: in both cases the code makes the assumption that it inserts a dependent instruction chain. Only then it is correct to simply add instruction latencies. I have not checked that this assumption is correct for all cases. So I suggest to add a comment about this. Assuming also that you don't see any notable perf regression you can go ahead and commit from my perspective: LGTM.

Dec 5 2017, 2:01 PM

Oct 11 2017

Gerolf added a comment to D38734: [MachineCombiner] Fix initialisation of LastUpdate for incremental update..

Thanks for working on this. LGTM!

Oct 11 2017, 10:43 AM

Sep 6 2017

Gerolf added a comment to D36696: [MachineTraceMetrics] Add computeDepth function (NFCI)..

LGTM. Just some food for thought for possible follow up commits.

Sep 6 2017, 4:07 PM

Sep 5 2017

Gerolf added a comment to D36619: [MachineCombiner] Update instruction depths incrementally for large BBs..

I added a few minor remarks that you might want to consider before commit. Again, thanks for working on this and your patience!

Sep 5 2017, 7:04 PM

Aug 25 2017

Gerolf added a comment to D36619: [MachineCombiner] Update instruction depths incrementally for large BBs..

Thank you for sharing your numbers. I was hoping you had that show-off case.
I just have a few minor comments about readability and asserts. I still need bit more time to convince myself that all the pieces fit together. I was wondering if you could add a test that dumps the depths w/ and w/o the incremental updates (compile w/ full comination + dumps depths, compile with a threashold of 1 + dump depths, diff the dump)?

Aug 25 2017, 12:23 PM

Aug 21 2017

Gerolf added a comment to D36619: [MachineCombiner] Update instruction depths incrementally for large BBs..

Hi Florian

Aug 21 2017, 2:18 AM

Jul 27 2017

Gerolf added inline comments to D35594: [GISel]: ConstantFold operations when building MIR.
Jul 27 2017, 12:18 AM

Jul 19 2017

Gerolf added a comment to D34781: Introduce a MCReloc class.

It is not clear to me if and how the original questions have been answered by this patch yet. Could you elaborate and add comments, please? Much appreciate!

Jul 19 2017, 5:03 PM

Jul 18 2017

Gerolf added a comment to D35594: [GISel]: ConstantFold operations when building MIR.

Thanks for working on this! Please add comments and show off one or two examples that highlight the benefits of your extensions.

Jul 18 2017, 6:00 PM

Jul 12 2017

Gerolf closed D35105: [SjLj] Replace recursive block marking algorithm with iterative algorithm.
Jul 12 2017, 4:05 PM

Jul 6 2017

Gerolf created D35105: [SjLj] Replace recursive block marking algorithm with iterative algorithm.
Jul 6 2017, 9:34 PM
Gerolf added a comment to D32201: [RALLOC] Increase CSR cost in RegAllocGreedy to favour splitting over CSR first use.

This is interesting. Unfortunately these heuristic changes are always tricky and never satisfying. Is this change motivated entirely by limitations of the current shrink-wrapping algorithm or do you see gains from better allocation also.

Jul 6 2017, 3:28 PM · Restricted Project

Jun 29 2017

Gerolf added reviewers for D34781: Introduce a MCReloc class: ab, Gerolf.

Just a first superficial review. I haven't thought about the underlying concept itself yet.

Jun 29 2017, 5:52 PM

May 11 2017

Gerolf added a comment to D33101: [AArch64][MachineCombine] Fold FNMUL+FSUB -> FNMADD.

LGTM.

May 11 2017, 10:02 AM

May 2 2017

Gerolf added a comment to D31964: CodeGen: BlockPlacement: Clear ComputedEdges between functions..

Thanks, Kyle!

May 2 2017, 11:25 PM

Apr 20 2017

Gerolf added a comment to D32261: [LoopUnroll] Don't try to unroll non-rotated loops.

I don't know if this patch makes the situation better or worse. It seems to touch on the tip of an iceberg. The underlying problem is: What can or should happen when a loop is irreducible? It the answer is "don't unroll", then this is the fix your actually looking for. Obviously this loop is irreducible since it has a retreat edge from sink to body2, but body2 does not dominate sink. I'm also curious if this loop is in the source code already or if some pass in the compiler actually generated. FWIW, unless the compiler has a problem irreducible loops should be rare. And when the user writes an irreducible loop, I think it is OK when compiler optimizations turn conservative.

Apr 20 2017, 3:27 PM

Apr 16 2017

Gerolf added a comment to D29641: [SLP] Fix for PR31847: Assertion failed: (isLoopInvariant(Operands[i], L) && "SCEVAddRecExpr operand is not loop-invariant!").

I would like to understand this better. Could somebody explain what assumptions SCEV makes about its clients? Which assumption(s) is broken by SLP? It seems to me that this issue potentially touches fundamental design decisions/questions and I don't see any verifiers in place.

Apr 16 2017, 2:37 PM · Restricted Project

Apr 12 2017

Gerolf requested changes to D31964: CodeGen: BlockPlacement: Clear ComputedEdges between functions..

Please double check your code one more time. Thank you!

Apr 12 2017, 3:51 PM

Mar 28 2017

Gerolf added a comment to D31443: [LTO] Do not reorder global variables unnecessarily during merging.

I'd love to see some comments an get your thoughts about a verifier.

Mar 28 2017, 3:29 PM

Feb 9 2017

Gerolf added a comment to D29627: Compile time decreasing in the case we're dealing with Machine Combiner.

Thanks for following up on this.

Feb 9 2017, 5:53 PM

Jan 26 2017

Gerolf added a comment to D26855: New unsafe-fp-math implementation for X86 target.

I think the only issue that needs to be addressed is (finally!) sharing perf data. This has been raised at least 3 times. The possible compile-time implication, the speciality of the application (fast-math) etc are well understood.

Jan 26 2017, 9:43 AM

Jan 24 2017

Gerolf added a comment to D27327: [InstCombine] Fix to switch narrowing.

Hans, you were right. The computeKnowBits etc parts should have been moved also. Sanjay committed the fix proper in r289442. I only kept the regression test in narrow-switch.ll (r293018).

Jan 24 2017, 8:48 PM

Jan 18 2017

Gerolf added a comment to D26855: New unsafe-fp-math implementation for X86 target.

I'm leaning towards a LGTM since you addressed basically all my issues, but more people mushroomed and are curious about your performance data. So I think can't dodge that question anymore and need to share some good data for the benchmark(s) you are tuning for before you get the nod.

Jan 18 2017, 6:40 PM

Jan 2 2017

Gerolf added a comment to D26855: New unsafe-fp-math implementation for X86 target.

From my perspective the implementation is close and only requires a few minor changes.

Jan 2 2017, 1:49 PM

Dec 22 2016

Gerolf added a comment to D26855: New unsafe-fp-math implementation for X86 target.

The "automatic" generation of pattern e.g. with TableGen is on my longer term wish list, not a requirement for this patch. Sorry if my wording was confusing.
Do you have performance numbers?

Dec 22 2016, 4:07 PM

Dec 21 2016

Gerolf added a comment to D26855: New unsafe-fp-math implementation for X86 target.

A few general remarks:

  • I'm very much in favor of the MC combiner approach
  • But I'm getting increasingly concerned about the MC code quality. I felt the FMA/Aarch64 support starts looking crabby (I'm the culprit) and this doesn't look much better I'm afraid. The approach is in need of more automation for better maintainability. This is meant as food for thought. Since I"m too blame for the approach I can't to be harsh in reviews :-)
  • I'm also concerned about the compile-time (in particular since we don't track x86 specific issues ( or any other backend than Aarch64- or least I'm not aware that anyone is watching closely). Could you share some specific data about your perf gains and compile-time measurements? However, I think this optimization is for fast math only and compile-time is probably less of an issue in that mode. One way to deal with compile-time issues is to wrap some MC under an option.
  • Perhaps I missed it but I expected the optimization to kick in only under fast math. I saw 'fast' in the test cases, but didn't see a check in the code.
Dec 21 2016, 2:36 PM

Dec 20 2016

Gerolf added a comment to D27645: debug output incorrect ["UNKNOWN"] when a MachineInstr is generated during the MachineCombiner pass and the relevant debug output is enabled.

Thanks! LGTM.

Dec 20 2016, 11:05 AM

Dec 13 2016

Gerolf added inline comments to D27645: debug output incorrect ["UNKNOWN"] when a MachineInstr is generated during the MachineCombiner pass and the relevant debug output is enabled.
Dec 13 2016, 1:02 PM

Dec 12 2016

Gerolf added inline comments to D27645: debug output incorrect ["UNKNOWN"] when a MachineInstr is generated during the MachineCombiner pass and the relevant debug output is enabled.
Dec 12 2016, 11:40 AM

Dec 9 2016

Gerolf added inline comments to D27645: debug output incorrect ["UNKNOWN"] when a MachineInstr is generated during the MachineCombiner pass and the relevant debug output is enabled.
Dec 9 2016, 5:28 PM
Gerolf requested changes to D27466: [SLP] Fix sign-extends for type-shrinking.

Michael, could you also lend this your expert eye? Do you agree with the extra memory (pair vs vector) and compile-time (look at the loop tree) investment for this fix? Shortening the types may not be worth it at least on some architectures.

Dec 9 2016, 2:53 PM
Gerolf added inline comments to D27327: [InstCombine] Fix to switch narrowing.
Dec 9 2016, 3:59 AM

Dec 8 2016

Gerolf added a comment to D27552: machine-instruction combiner uses too-simple algorithm to compute cost of one of the two alternatives, throws away combination results too often.

Good catch. LGTM.

Dec 8 2016, 3:06 PM

Dec 1 2016

Gerolf retitled D27327: [InstCombine] Fix to switch narrowing from to [InstCombine] Fix to switch narrowing.
Dec 1 2016, 6:04 PM

Nov 18 2016

Gerolf added a comment to D26848: [PATCH] Reduce inline thresholds to compensate for cost changes.

The Oz case looks interesting. Can you share more details/insights about the "inaccuracies" w/ specific examples? I'm wondering if that can be fixed in general or be more triggered towards some trunk characteristics. But this is just something we can think about and discuss while moving on and celebrate the ct/cs recoveries :-). So LGTM!

Nov 18 2016, 1:02 PM

Nov 17 2016

Gerolf added a reviewer for D26525: [InstCombine] canonicalize min/max constant to select's false value : Gerolf.
Nov 17 2016, 11:31 PM
Gerolf added reviewers for D26529: [LoopReroll] Make root-finding more aggressive.: anemet, mzolotukhin.
Nov 17 2016, 11:29 PM

Nov 14 2016

Gerolf added a comment to D25966: [AArch64] Lower multiplication by a constant int to shl+add+shl.

Thanks for following up!
LGTM

Nov 14 2016, 12:09 PM

Nov 1 2016

Gerolf added a comment to D25966: [AArch64] Lower multiplication by a constant int to shl+add+shl.

I thought I understand this until about the middle of the review. Now I could use some help perhaps with variable names and comments that reflect more clearly on the expression(s) you simplify. I think this is what Renato is looking for, too.

Nov 1 2016, 2:26 PM

Oct 27 2016

Gerolf added a reviewer for D25963: [LoopUnroll] Implement profile-based loop peeling: anemet.
Oct 27 2016, 7:49 PM
Gerolf added a comment to D25966: [AArch64] Lower multiplication by a constant int to shl+add+shl.

Hi Haicheng,

Oct 27 2016, 7:29 PM

Oct 26 2016

Gerolf added reviewers for D25999: LegalizeDAG: Implement expansion of f16 = FP_TO_FP16 f64: ab, t.p.northover.
Oct 26 2016, 10:25 AM
Gerolf added a comment to D25963: [LoopUnroll] Implement profile-based loop peeling.

Could you provide more background on this idea? What is your motivational use case? When the trip count is low why optimize? If the profile is wrong and it actually is a hot loop for a regular/different input set peeling could hurt. There are also side effects on code size, register pressure etc. that could hurt performance.

Oct 26 2016, 10:08 AM

Oct 14 2016

Gerolf added a comment to D25634: Fix replacedSelectWithOperand in InstCombiner to handle branch having two same successors..

This looks weird - checking for null and then ignoring it?! Could you check why this IR is generated? I suspect there is either a bug or a missing optimization. In that case I suggest getting that fixed and adding an assertion in the select optimization.

Oct 14 2016, 6:18 PM

Sep 15 2016

Gerolf added a comment to D24577: [MachineMemOperand] Move synchronization scope and atomic orderings from SDNode to MachineMemOperand.

My concern is this would increase memory consumption in all backends. Maybe you can hide most of it with an opaque pointer in the machine operand.

Sep 15 2016, 12:45 PM

Sep 8 2016

Gerolf added a comment to D24366: [MachineCombiner] Support for floating-point FMA under the FPOpFusion Fast flag on ARM64.

LGTM

Sep 8 2016, 8:31 PM

Jul 28 2016

Gerolf updated the diff for D22900: Revert r244207 - Mark calls in thunk functions as tail-call optimization.

Reduced test case.

Jul 28 2016, 8:52 PM

Jul 27 2016

Gerolf added a comment to D22900: Revert r244207 - Mark calls in thunk functions as tail-call optimization.

Nope, I don't see the tail call. Anyway, I'll simplify my test case. Don't worry about it.

Jul 27 2016, 9:25 PM
Gerolf added a comment to D22886: [llvm] Loop Vectorization fault in collectLoopUni.

Hm, FWIW this test doesn't reproduce with our compiler that exposes the problem at stage 2. And I still think when there are no uniforms within the loop there is no need to traverser the PHI nodes later.
Since the patch clearly fixes an obvious problem I'm OK with a commit even w/o a test case - LGTM.

Jul 27 2016, 9:05 PM
Gerolf added a comment to D22900: Revert r244207 - Mark calls in thunk functions as tail-call optimization.

Please add the options you used to compile? I can certainly shrink the test case a bit before I commit.

Jul 27 2016, 8:23 PM
Gerolf retitled D22900: Revert r244207 - Mark calls in thunk functions as tail-call optimization from to Revert r244207 - Mark calls in thunk functions as tail-call optimization.
Jul 27 2016, 7:15 PM
Gerolf added inline comments to D22886: [llvm] Loop Vectorization fault in collectLoopUni.
Jul 27 2016, 3:58 PM

Jul 7 2016

Gerolf added a comment to D21948: [DSE] fix - missing store to runtime stack in thunk with tail call bvval arg.

Ahmed pointed out the clang commit causing the issue is r244207 - Mark calls in thunk functions as tail-call optimization candidates

Jul 7 2016, 8:36 PM

Jul 6 2016

Gerolf added inline comments to D8688: Update MergedLoadStoreMotion to use MemorySSA.
Jul 6 2016, 8:11 PM

Jul 5 2016

Gerolf added inline comments to D8688: Update MergedLoadStoreMotion to use MemorySSA.
Jul 5 2016, 8:42 PM

Jul 1 2016

Gerolf retitled D21948: [DSE] fix - missing store to runtime stack in thunk with tail call bvval arg from to [DSE] fix - missing store to runtime stack in thunk with tail call bvval arg.
Jul 1 2016, 5:41 PM

Jun 24 2016

Gerolf added a comment to D8688: Update MergedLoadStoreMotion to use MemorySSA.

I'll take a more in depth look into the load and store merge routines also. At a first glance it seems one could just add a few MSSA hooks rather than copy-paste-modify the code base.
I also take a (late :-() look at the core MSSA design. Could you outline the high-level MSSA design in more detail than http://reviews.llvm.org/D7864?Eg. how does it compare/relate to http://www.airs.com/dnovillo/Papers/mem-ssa.pdf? Thanks!

Jun 24 2016, 5:16 PM

Jun 22 2016

Gerolf added inline comments to D8688: Update MergedLoadStoreMotion to use MemorySSA.
Jun 22 2016, 7:07 PM
Gerolf closed D4890: [Cleanup] Utility function to erase instruction and mark DBG_Values.
Jun 22 2016, 6:04 PM

May 27 2016

Gerolf added inline comments to D19553: Unroll pass restructure..
May 27 2016, 4:44 PM
Gerolf added a comment to D20220: [PEI, AArch64] Use empty spaces in stack area for local stack slot allocation..

Thanks, Geoff. I take back my previous comment. I was worried that when A and B are allocated at stack offset x and this optimization changes x, it would only change it for A but not for B. But looking deeper into the code this is not the case.

May 27 2016, 4:16 PM
Gerolf added a comment to D20220: [PEI, AArch64] Use empty spaces in stack area for local stack slot allocation..

Since I couldn't spot this in the code: in the scenario where there is one empty reallocate both of them to the empty slot or give up after it reallocated the first object?

May 27 2016, 11:21 AM

May 23 2016

Gerolf added a comment to D20319: [InstCombine] Fix assertion when bitcast is converted to gep.

Committed revision 270479

May 23 2016, 12:30 PM

May 16 2016

Gerolf retitled D20319: [InstCombine] Fix assertion when bitcast is converted to gep from to [InstCombine] Fix assertion when bitcast is converted to gep.
May 16 2016, 11:37 PM

May 13 2016

Gerolf added a comment to rL269392: [LoopDist] Only run LAA for loops with the pragma.

LGTM

May 13 2016, 2:14 PM

May 9 2016

Gerolf added a comment to rL268987: [LV] Hint at the new loop distribution pragma in optimization remark.

LGTM.

May 9 2016, 7:41 PM

Apr 27 2016

Gerolf added a comment to D19002: [LazyValueInfo] Fix for a nasty compile-time problem with questions.

Philip,
can you share data about your compile-time improvements? Platform, benchmarks, options? Did you measure performance impact also?

Apr 27 2016, 11:33 AM
Gerolf added a comment to D19550: [InstCombine] Determine the result of a select based on a dominating condition..

I sharpened the test case.

Apr 27 2016, 10:43 AM

Apr 26 2016

Gerolf added inline comments to D19550: [InstCombine] Determine the result of a select based on a dominating condition..
Apr 26 2016, 5:41 PM

Apr 23 2016

Gerolf added a comment to D18751: [MachineCombiner] Support for floating-point FMA on ARM64.

Thanks, Quentin! I should have addressed all your concerns.

Apr 23 2016, 10:26 PM

Apr 21 2016

Gerolf added a comment to D18751: [MachineCombiner] Support for floating-point FMA on ARM64.

Thanks, Quentin! I should have addressed all your concerns.

Apr 21 2016, 7:20 PM

Apr 20 2016

Gerolf updated the diff for D18751: [MachineCombiner] Support for floating-point FMA on ARM64.

Comments and local handling of optlevel as Quentin suggested.

Apr 20 2016, 7:53 PM

Apr 19 2016

Gerolf updated the diff for D19002: [LazyValueInfo] Fix for a nasty compile-time problem with questions.

Mostly removed some of my questions and added a few comments based on Philip's
review.

Apr 19 2016, 7:22 PM
Gerolf added a comment to D19002: [LazyValueInfo] Fix for a nasty compile-time problem with questions.

This should also take care of http://reviews.llvm.org/D18066.

Apr 19 2016, 7:19 PM