philip.pfaffe (Philip Pfaffe)
User

Projects

User does not belong to any projects.

User Details

User Since
Apr 21 2015, 2:43 AM (164 w, 6 d)

Recent Activity

Today

philip.pfaffe added inline comments to D48136: [Polly] Implement an iterator for isl maps, basic_maps, sets, basic_sets.
Mon, Jun 18, 3:24 AM
philip.pfaffe updated the diff for D48136: [Polly] Implement an iterator for isl maps, basic_maps, sets, basic_sets.
Free the context in the unittest.
Mon, Jun 18, 3:21 AM

Thu, Jun 14

philip.pfaffe added reviewers for D48136: [Polly] Implement an iterator for isl maps, basic_maps, sets, basic_sets: Meinersbur, grosser.
Thu, Jun 14, 2:04 AM
philip.pfaffe added a comment to D48136: [Polly] Implement an iterator for isl maps, basic_maps, sets, basic_sets.

Looks like I forgot to add pollydev as a subscriber. Shall I resubmit the patch?

Thu, Jun 14, 2:03 AM
philip.pfaffe updated the diff for D48136: [Polly] Implement an iterator for isl maps, basic_maps, sets, basic_sets.
Add a unittest and address comments
Thu, Jun 14, 2:01 AM

Wed, Jun 13

philip.pfaffe added inline comments to D48136: [Polly] Implement an iterator for isl maps, basic_maps, sets, basic_sets.
Wed, Jun 13, 2:36 PM
philip.pfaffe created D48136: [Polly] Implement an iterator for isl maps, basic_maps, sets, basic_sets.
Wed, Jun 13, 10:36 AM

Tue, Jun 12

philip.pfaffe created D48070: [Polly] Simplify the implementation of getCUDALibDeviceFunction. NFC..
Tue, Jun 12, 3:19 AM

Mon, Jun 11

philip.pfaffe added a comment to D48026: [ScopHelper] Provide support for recognising collective invariant loads.

As I mentioned in person: Instead of detecting this specific pattern, would it be possible to treat this as a fixpoint iteration? I.e., maintain a set of undecided address sources and keep shrinking it until only the non-invariant ones remein.

Mon, Jun 11, 8:40 AM
philip.pfaffe added inline comments to D47930: Make email options of find_interesting_reviews more flexible..
Mon, Jun 11, 2:45 AM

Fri, Jun 8

philip.pfaffe updated the diff for D47890: [Polly][zorg] Enable GPGPU Codegen on the builders only if NVPTX is an LLVM target.

Fix weird iterator comprehension.

Fri, Jun 8, 11:34 AM
philip.pfaffe added a comment to D47890: [Polly][zorg] Enable GPGPU Codegen on the builders only if NVPTX is an LLVM target.

I still think that any(True for ... is unnecessary.

Fri, Jun 8, 1:11 AM

Thu, Jun 7

philip.pfaffe updated the diff for D47890: [Polly][zorg] Enable GPGPU Codegen on the builders only if NVPTX is an LLVM target.

Add a comment

Thu, Jun 7, 2:18 PM
philip.pfaffe added inline comments to D47890: [Polly][zorg] Enable GPGPU Codegen on the builders only if NVPTX is an LLVM target.
Thu, Jun 7, 2:11 PM
philip.pfaffe created D47890: [Polly][zorg] Enable GPGPU Codegen on the builders only if NVPTX is an LLVM target.
Thu, Jun 7, 9:24 AM
philip.pfaffe created D47888: [Polly] Back out of GPU Codegen if NVPTX is not available.
Thu, Jun 7, 8:39 AM

Wed, Jun 6

philip.pfaffe added a comment to D46201: [DA] Use SCEV to conclude DVEntry::EQ in some cases..

This does not look correct. Consider the AA.ll testcase. Your change makes DA claim there is no output dependency on the load in %for.inner, but there is! Carried over the outer loop. So the result should be input[* =] (don't quote me on the syntax though).

Wed, Jun 6, 3:22 AM
philip.pfaffe accepted D47469: [Unittests] Change linker flags of dynamic library tests.

Interesting, I was under the impression that the relevant differences between MODULE and SHARED were only on Mac. In either case, this change LGTM.

Wed, Jun 6, 3:09 AM

Mon, Jun 4

philip.pfaffe added a comment to D47469: [Unittests] Change linker flags of dynamic library tests.

What platform does this occur on? Mac?

Mon, Jun 4, 5:30 AM
philip.pfaffe added a comment to D47604: [Polly] Add isl C++ list iterators.

This might be too much to ask for, but is it possible to report_fatal_error() at the earliest when accessing out of bounds using the iterator? If ISL performs inbounds checks, then disregard this.

Mon, Jun 4, 3:25 AM
philip.pfaffe requested changes to D47604: [Polly] Add isl C++ list iterators.

I love the gist of this, but there are a couple of implementation issues, some of which @Meinersbur pointed out already.

Mon, Jun 4, 2:50 AM
philip.pfaffe added a comment to D45872: [DA] Enable -da-delinearize by default.

Some inline nits. What worries me slightly is that the test oracle changed in GCD.ll. Why is that? Where the tests plainly wrong before?

Mon, Jun 4, 2:11 AM

May 18 2018

philip.pfaffe added a comment to D47082: Fix build warning compiling TestPlugin on Windows and disable Passes plugin stuff on Windows since it fundamentally can't work.

Thanks for the clarification @rnk!

May 18 2018, 2:28 PM
philip.pfaffe added a comment to D47082: Fix build warning compiling TestPlugin on Windows and disable Passes plugin stuff on Windows since it fundamentally can't work.

I agree with @rnk, too, the question is do we want to enable this in the meantime?

May 18 2018, 2:18 PM
philip.pfaffe added a comment to D47082: Fix build warning compiling TestPlugin on Windows and disable Passes plugin stuff on Windows since it fundamentally can't work.

There is this ancient commit here D18826 that introduced auto-export of all symbols. Shouldn't this work here as well?

May 18 2018, 1:56 PM
philip.pfaffe added a comment to D47082: Fix build warning compiling TestPlugin on Windows and disable Passes plugin stuff on Windows since it fundamentally can't work.

I'm likely not the right person to review this, I know next to nothing about windows dev.

May 18 2018, 1:10 PM

May 16 2018

philip.pfaffe accepted D46898: Give shared modules in unittests the platform-native extension, make PipSqueak a MODULE.

Thanks!

May 16 2018, 9:30 AM
philip.pfaffe added a comment to D46898: Give shared modules in unittests the platform-native extension, make PipSqueak a MODULE.

Since this is only for the testcase, sounds perfect!

May 16 2018, 9:23 AM
philip.pfaffe added a comment to D46898: Give shared modules in unittests the platform-native extension, make PipSqueak a MODULE.

Fundamentally looks good, but I'm slightly concerned about using different macros in the Code (LTDL_SHLIB_EXT) vs. in the build (LLVM_PLUGIN_EXT). Right now those are identical, but it might change. Can we export LLVM_PLUGIN_EXT in config.h?

May 16 2018, 9:15 AM
philip.pfaffe accepted D46935: [ScopInfo] Remove usage of isl_set_n_basic_set().

Do you need help committing this?

May 16 2018, 5:04 AM
philip.pfaffe added inline comments to D46935: [ScopInfo] Remove usage of isl_set_n_basic_set().
May 16 2018, 4:43 AM

May 15 2018

philip.pfaffe accepted D46843: Rename three cxx files in unittests to cpp..

Since the diff being broken is mostly a workflow issue, this patch LGTM.

May 15 2018, 7:51 AM
philip.pfaffe added a comment to D46843: Rename three cxx files in unittests to cpp..

After some digging, I believe this is caused by svn diff comparing the new file against the old one, which is of course unsuitable to create patches from. So you likely should create the diff with svn diff --show-copies-as-adds or svn diff --patch-compatible.

May 15 2018, 7:12 AM
philip.pfaffe added a comment to D46843: Rename three cxx files in unittests to cpp..

That's how svn diff looks after svn mv. Maybe you use git (which doesn't have a native "move file" operation) and aren't used to seeing this type of diff?

I'm not sure what caused it, but the patch is not applicable.

May 15 2018, 6:35 AM
philip.pfaffe updated the diff for D46869: [Polly][arc] Remove unittesting from arcconfig.

Actually remove unittest driver as well.

May 15 2018, 4:24 AM
philip.pfaffe created D46869: [Polly][arc] Remove unittesting from arcconfig.
May 15 2018, 4:13 AM
philip.pfaffe created D46868: [Polly] Create Scop name lazily.
May 15 2018, 3:52 AM
philip.pfaffe added a comment to D46843: Rename three cxx files in unittests to cpp..

Something is off with the diff. E.g. the file TestPlugin.cpp doesn't exist, yet the diff contains modifications. Did you miss a local commit?

May 15 2018, 12:06 AM

May 8 2018

philip.pfaffe created D46578: Fix gdb pretty printers to work with Python 3..
May 8 2018, 5:22 AM

May 2 2018

philip.pfaffe added a comment to D46227: [islpp] Do not abuse isl::stat::error as early-abort.

I'm generally fine with this, just one more nit: I don't like for the argument to be called Aborted, for that suggests an error condition. What about BreakIteration instead?

May 2 2018, 8:23 AM
philip.pfaffe added a comment to rL328666: [polly] [ScopInfo] Don't pre-compute the name of the Scop's region..

This is a use-after-free.

May 2 2018, 7:54 AM
philip.pfaffe created D46359: [Polly] Pass compiler arguments in the create_ll.sh script.
May 2 2018, 7:29 AM

Apr 29 2018

philip.pfaffe added a comment to D46192: Script to match open Phabricator reviews with potential reviewers.

Do you plan on actually commiting this to the repository, or would you like any feedback nontheless?

Apr 29 2018, 4:16 AM
philip.pfaffe added inline comments to D46227: [islpp] Do not abuse isl::stat::error as early-abort.
Apr 29 2018, 4:11 AM
philip.pfaffe added a comment to D46231: Move to new C++ binding interface that requires isl::quota annotations [RFC].

I think this diff is too large and should be split into at least two parts:

  1. The implementation of isl::quota, and possibly a small unittest showcasing how it's intended to be used
  2. Update of the APIs
Apr 29 2018, 4:06 AM

Apr 28 2018

philip.pfaffe added a comment to D46227: [islpp] Do not abuse isl::stat::error as early-abort.

I think this makes sense, with some inline nits.

Apr 28 2018, 3:22 PM

Apr 27 2018

philip.pfaffe added a comment to rL329273: [Plugins] Add a slim plugin API to work together with the new PM.

@RKSimon Is this still happening?

Apr 27 2018, 3:19 AM

Apr 25 2018

philip.pfaffe added inline comments to D45743: [Polly] Print executed statement instances at runtime..
Apr 25 2018, 10:42 AM · Restricted Project
philip.pfaffe accepted D46012: Avoid a warning on pointer casting, NFC.

LGTM. Thanks for fixing!

Apr 25 2018, 4:45 AM

Apr 20 2018

philip.pfaffe requested changes to D45743: [Polly] Print executed statement instances at runtime..
Apr 20 2018, 6:19 AM · Restricted Project
philip.pfaffe added inline comments to D45743: [Polly] Print executed statement instances at runtime..
Apr 20 2018, 3:37 AM · Restricted Project

Apr 19 2018

philip.pfaffe added inline comments to D45743: [Polly] Print executed statement instances at runtime..
Apr 19 2018, 9:22 AM · Restricted Project
philip.pfaffe accepted D45811: [Unittests] Fix plugins test.

LGTM, and solves the issue locally.

Apr 19 2018, 6:36 AM

Apr 17 2018

philip.pfaffe added a comment to D45649: [Polly][ScopDetect] Reject loop with multiple exit blocks..

A BasicBlock that, for instance, terminates with a ret or unreachable, Polly cannot recognize as a SCoP even if that block would be considered as an error-block, because RegionInfo does not create a Region for it since it doesn't know about error blocks.

Apr 17 2018, 1:00 AM · Restricted Project

Apr 16 2018

philip.pfaffe added a comment to D45493: [Polly][NewPM] Port IslAst, ScopDetect, and ScopInfo lit testsuites over to the new PM.

Instead of adding -aa-pipeline=basic-aa everywhere, did you consider adding it to %loadPolly, or only add it if the test otherwise fails?

I can add it to %loadPolly. I'm hesitant to make this opt-in, though, since I fear the tests could get very flaky.

Can you elaborate on this?

In a legacy PM pipeline, basic-aa is always the default. I.e., if you do AA queries without specifying any AA passes, you always get basic-aa results. In the new PM, that's not true. So i think it should be present in the testcases. I'm happy to add it to `%loadPolly', though!

Apr 16 2018, 4:04 AM
philip.pfaffe updated the diff for D45484: [Polly][NewPM] Update pass registration for the LLVM plugin interface.

Add a testcase.

Apr 16 2018, 3:55 AM

Apr 11 2018

philip.pfaffe added a comment to D45534: Add isl operator overloads for isl::pw_aff (Try II).

Since you expect users to include the header, the operators should also be declared static.

Apr 11 2018, 1:48 PM
philip.pfaffe added inline comments to D45484: [Polly][NewPM] Update pass registration for the LLVM plugin interface.
Apr 11 2018, 3:24 AM
philip.pfaffe added a comment to D45493: [Polly][NewPM] Port IslAst, ScopDetect, and ScopInfo lit testsuites over to the new PM.

Instead of adding -aa-pipeline=basic-aa everywhere, did you consider adding it to %loadPolly, or only add it if the test otherwise fails?

I can add it to %loadPolly. I'm hesitant to make this opt-in, though, since I fear the tests could get very flaky.

Apr 11 2018, 3:18 AM

Apr 10 2018

philip.pfaffe added a comment to D45484: [Polly][NewPM] Update pass registration for the LLVM plugin interface.

The tests for this I added in a follow-up revision D45493, which, as you can see, is quite large. I wanted to keep the functional change here and the mechanical changes to the tests separate for better review.

Apr 10 2018, 8:46 AM
philip.pfaffe added a comment to D45493: [Polly][NewPM] Port IslAst, ScopDetect, and ScopInfo lit testsuites over to the new PM.

It's obviously hard for reviewers to examine all changes, but I'd be happy if you could take a look at a random sample :)

Apr 10 2018, 8:40 AM
philip.pfaffe created D45493: [Polly][NewPM] Port IslAst, ScopDetect, and ScopInfo lit testsuites over to the new PM.
Apr 10 2018, 8:36 AM
philip.pfaffe added a dependent revision for D45484: [Polly][NewPM] Update pass registration for the LLVM plugin interface: D45493: [Polly][NewPM] Port IslAst, ScopDetect, and ScopInfo lit testsuites over to the new PM.
Apr 10 2018, 8:36 AM
philip.pfaffe created D45484: [Polly][NewPM] Update pass registration for the LLVM plugin interface.
Apr 10 2018, 5:50 AM
philip.pfaffe accepted D45210: [New-PM] Lift Scop Pipeline to CGSCC-level.

LGTM too!

Apr 10 2018, 1:24 AM

Apr 5 2018

philip.pfaffe added a comment to D45210: [New-PM] Lift Scop Pipeline to CGSCC-level.
  1. A function-level transformation pass wants to use Polly analysis. Since function passes can be run in any order by the pass manager, it does not necessarily correspond to the CGSCC (reverse? post-order) expected by the Polly passes. Callees have been already modified by this hypothetical pass after being analyzed by Polly.

This patch only changes the layering for ScopPasses. Note that Scop analyses are still being proxied in and out of function. As a consequence, function layer passes are not affected by this change.

Apr 5 2018, 3:57 AM

Apr 4 2018

philip.pfaffe added inline comments to D45210: [New-PM] Lift Scop Pipeline to CGSCC-level.
Apr 4 2018, 1:36 PM
philip.pfaffe updated the diff for D35258: [Plugins] Add a slim plugin API to work together with the new PM.

Address review comments:

  • Use more StringRef and Twine
  • Hide the PassPluginLibraryInfo object
Apr 4 2018, 5:19 AM

Apr 3 2018

philip.pfaffe added inline comments to D35258: [Plugins] Add a slim plugin API to work together with the new PM.
Apr 3 2018, 3:50 AM
philip.pfaffe accepted D45171: [Polly] abort() on assertion violation..

We might also think about using the standard assert() macro instead.

Apr 3 2018, 2:10 AM · Restricted Project

Mar 28 2018

philip.pfaffe updated the diff for D35258: [Plugins] Add a slim plugin API to work together with the new PM.

Rename, move around, and simplify some things.

Mar 28 2018, 5:51 AM

Mar 26 2018

philip.pfaffe added a comment to D35258: [Plugins] Add a slim plugin API to work together with the new PM.

Ping

Mar 26 2018, 3:32 PM

Mar 14 2018

philip.pfaffe updated the diff for D35258: [Plugins] Add a slim plugin API to work together with the new PM.

Address review comments, mostly adding commentary and renaming things.

Mar 14 2018, 9:14 AM

Mar 13 2018

philip.pfaffe added inline comments to D35258: [Plugins] Add a slim plugin API to work together with the new PM.
Mar 13 2018, 11:33 AM

Mar 4 2018

philip.pfaffe created D44078: [Polly][CMake] Fix lit setup for building the in the mono repo.
Mar 4 2018, 7:43 AM

Mar 3 2018

philip.pfaffe closed D43872: [Polly][Acc] Fix for PR33208.

Commited in rPLO326643, reverted in rPLO326656.

Mar 3 2018, 7:41 AM · Restricted Project

Mar 2 2018

philip.pfaffe updated the diff for D35258: [Plugins] Add a slim plugin API to work together with the new PM.

[NFC] clang-format

Mar 2 2018, 11:02 AM
philip.pfaffe updated the diff for D35258: [Plugins] Add a slim plugin API to work together with the new PM.

Improve Layering: Move New-PM Pass Plugins into the Passes library

Mar 2 2018, 11:01 AM

Feb 28 2018

philip.pfaffe updated the diff for D43872: [Polly][Acc] Fix for PR33208.

My bad, I uploaded the reverse diff ...

Feb 28 2018, 8:11 AM · Restricted Project
philip.pfaffe added a comment to D43872: [Polly][Acc] Fix for PR33208.

It was in the PR, but I'll mention it in the commit.

Feb 28 2018, 7:58 AM · Restricted Project
philip.pfaffe added a comment to D43872: [Polly][Acc] Fix for PR33208.

This is a fix for existing currently failing testcases, I think that should cover it.

Feb 28 2018, 7:48 AM · Restricted Project
philip.pfaffe updated subscribers of D43872: [Polly][Acc] Fix for PR33208.
Feb 28 2018, 7:42 AM · Restricted Project
philip.pfaffe created D43872: [Polly][Acc] Fix for PR33208.
Feb 28 2018, 7:40 AM · Restricted Project

Jan 11 2018

philip.pfaffe added a comment to D35258: [Plugins] Add a slim plugin API to work together with the new PM.

Humble Ping

Jan 11 2018, 8:20 AM

Dec 6 2017

philip.pfaffe updated the diff for D40803: Port SCEVAffinator to the isl c++ bindings.

Remove isl ownership markers

Dec 6 2017, 11:11 AM

Dec 4 2017

philip.pfaffe created D40803: Port SCEVAffinator to the isl c++ bindings.
Dec 4 2017, 1:27 PM
philip.pfaffe created D40783: [Polly] Drop overapproximation of kernel arguments.
Dec 4 2017, 6:08 AM

Dec 2 2017

philip.pfaffe added inline comments to D39971: Port ScopInfo to the isl cpp bindings.
Dec 2 2017, 6:45 AM

Dec 1 2017

philip.pfaffe added inline comments to D39904: [Polly] [PPCGCodeGeneration] Teach `SubtreeValues` to include possibly required base pointers, not remove all base pointers blindly..
Dec 1 2017, 4:37 AM
philip.pfaffe requested changes to D38353: [Polly] [PPCGCodeGeneration] Fix offset computation for managed arrays..

Actually the tests requires some changes.

Dec 1 2017, 4:33 AM

Nov 30 2017

philip.pfaffe added a comment to D40122: Add isl to LLVM repository..

I think this should be an RFC on llvm-dev. This needs a wider audience and a discussion that goes beyond the technical.

Nov 30 2017, 5:30 AM

Nov 28 2017

philip.pfaffe added a comment to D40207: [WIP] [DebugIR] Revive the old DebugIR pass for tip-of-tree LLVM..

I think you have to recommit this, since phabricator doesn't send out the original patch if you forget to add llvm-commits up front :(

Nov 28 2017, 9:37 AM
philip.pfaffe added inline comments to D40492: Handle Top-Level-Regions in polly::isHoistableLoad.
Nov 28 2017, 5:58 AM · Restricted Project

Nov 27 2017

philip.pfaffe added a comment to D40492: Handle Top-Level-Regions in polly::isHoistableLoad.

@bollu: Can you verify that this fixes your issue?

Nov 27 2017, 7:49 AM · Restricted Project

Nov 20 2017

philip.pfaffe added a comment to D40207: [WIP] [DebugIR] Revive the old DebugIR pass for tip-of-tree LLVM..

If this is going back into LLVM, it should also be a NewPM Pass.

Nov 20 2017, 3:19 AM

Nov 19 2017

philip.pfaffe added inline comments to D39971: Port ScopInfo to the isl cpp bindings.
Nov 19 2017, 2:16 PM

Nov 17 2017

philip.pfaffe added inline comments to D39904: [Polly] [PPCGCodeGeneration] Teach `SubtreeValues` to include possibly required base pointers, not remove all base pointers blindly..
Nov 17 2017, 3:44 AM
philip.pfaffe updated the diff for D39971: Port ScopInfo to the isl cpp bindings.

Address review comments.

Nov 17 2017, 3:32 AM

Nov 16 2017

philip.pfaffe accepted D40126: [Polly] [PCGCodeGen] Change printf to outs() to prevent garbled output. [NFC].

Okay. In that case LGTM.

Nov 16 2017, 8:50 AM
philip.pfaffe added a comment to D40126: [Polly] [PCGCodeGen] Change printf to outs() to prevent garbled output. [NFC].

Shouldn't this actually go to errs() instead?

Nov 16 2017, 7:21 AM