grosser (Tobias Grosser)
User

Projects

User does not belong to any projects.

User Details

User Since
Jul 30 2014, 11:40 AM (147 w, 3 d)

Recent Activity

Today

grosser committed rL304075: Delinearize memory accesses that reference parameters coming from function calls.
Delinearize memory accesses that reference parameters coming from function calls
Sat, May 27, 8:19 AM
grosser committed rL304074: Allow side-effect free function calls in valid affine SCEVs.
Allow side-effect free function calls in valid affine SCEVs
Sat, May 27, 8:19 AM
grosser committed rL304073: [SCEV] Assume parameters coming from function calls contain IVs.
[SCEV] Assume parameters coming from function calls contain IVs
Sat, May 27, 8:18 AM
grosser committed rL304070: Update some tests to changes in isl's internal representation.
Update some tests to changes in isl's internal representation
Sat, May 27, 4:33 AM
grosser committed rL304069: Update isl to isl-0.18-662-g17e172e.
Update isl to isl-0.18-662-g17e172e
Sat, May 27, 4:10 AM
grosser committed rL304068: Update information on isl C++ bindings in Polly release notes.
Update information on isl C++ bindings in Polly release notes
Sat, May 27, 4:01 AM

Yesterday

grosser added a comment to D33163: [Polly] Added the list of Instructions to output in ScopInfo pass.

Thank you Nandini, this looks good. I committed the change!

Fri, May 26, 9:41 PM
grosser committed rL304062: [Polly] Added the list of Instructions to output in ScopInfo pass.
[Polly] Added the list of Instructions to output in ScopInfo pass
Fri, May 26, 9:40 PM
grosser closed D33163: [Polly] Added the list of Instructions to output in ScopInfo pass by committing rL304062: [Polly] Added the list of Instructions to output in ScopInfo pass.
Fri, May 26, 9:40 PM
grosser committed rL304007: [ScopInfo] Translate mapToDimension to isl C++ [NFC].
[ScopInfo] Translate mapToDimension to isl C++ [NFC]
Fri, May 26, 10:22 AM

Thu, May 25

grosser added a comment to D33523: [Polly] [OpenMP] Update Subtree values for nested parallel loop.

Hi Mohan,

Thu, May 25, 10:50 PM · Unknown Object (Project)
grosser added a comment to D33163: [Polly] Added the list of Instructions to output in ScopInfo pass.

Hi Nandini,

Thu, May 25, 9:50 PM

Wed, May 24

grosser committed rL303839: Drop newline in docs builder to see if Polly docs are updated.
Drop newline in docs builder to see if Polly docs are updated
Wed, May 24, 10:38 PM
grosser committed rL303807: [ScopInfo] Tighten compute out introduced in r303404.
[ScopInfo] Tighten compute out introduced in r303404
Wed, May 24, 2:24 PM
grosser resigned from D33281: [Doc] Update how to install graphviz on macOS.

Gonna be dropping out as reviewer as I don't really have MacOSX and there are enough reviewers in the discussion. My two cents: I would likely also avoid to put too specific instruction into the documentation for the reasons others pointed out.

Wed, May 24, 12:12 PM
grosser accepted D33411: [Polly] Add handling of Top Level Regions.

LGTM, Philip feel free to commit.

Wed, May 24, 10:45 AM · Unknown Object (Project)
grosser accepted D33487: [Polly][DeLICM] Partial writes for PHIs..

This LGTM, in terms of implementation. So we can commit it to experiment with it. However, performance wise this seems to not work well. My gemm kernel regresses from 0.8 seconds in the optimal case (without pattern matching opts) to 4.4 seconds with partial writes.

Wed, May 24, 4:23 AM · Unknown Object (Project)

Tue, May 23

grosser committed rL303725: Add a permanent Polly Hangouts URL.
Add a permanent Polly Hangouts URL
Tue, May 23, 11:46 PM
grosser committed rL303724: Add new C++ bindings to release notes.
Add new C++ bindings to release notes
Tue, May 23, 11:46 PM
grosser committed rL303723: Update version number in Polly release notes.
Update version number in Polly release notes
Tue, May 23, 11:46 PM
grosser committed rL303722: Header file to help forcibly link GPURuntime.
Header file to help forcibly link GPURuntime
Tue, May 23, 11:46 PM
grosser closed D33198: Header file to help forcibly link GPURuntime by committing rL303722: Header file to help forcibly link GPURuntime.
Tue, May 23, 11:46 PM · Unknown Object (Project)
grosser accepted D33198: Header file to help forcibly link GPURuntime.

This LGTM. Should I commit it?

Tue, May 23, 6:28 AM · Unknown Object (Project)
grosser committed rL303615: [ScopInfo] Translate foldAccessRelation to isl C++ [NFC].
[ScopInfo] Translate foldAccessRelation to isl C++ [NFC]
Tue, May 23, 12:23 AM
grosser committed rL303612: [ScopInfo] Translate buildMemIntrinsicAccessRelation to isl C++ [NFC].
[ScopInfo] Translate buildMemIntrinsicAccessRelation to isl C++ [NFC]
Tue, May 23, 12:07 AM
grosser committed rL303611: [ScopInfo] Translate assumeNoOutOfBound to isl C++ [NFC].
[ScopInfo] Translate assumeNoOutOfBound to isl C++ [NFC]
Tue, May 23, 12:07 AM
grosser committed rL303610: [ScopInfo] Translate applyAndSetFAD to isl C++.
[ScopInfo] Translate applyAndSetFAD to isl C++
Tue, May 23, 12:07 AM

Mon, May 22

grosser committed rL303608: [ScopInfo] Translate isReadOnly to isl C++.
[ScopInfo] Translate isReadOnly to isl C++
Mon, May 22, 11:41 PM
grosser committed rL303600: [ScopInfo] Simplify domains early.
[ScopInfo] Simplify domains early
Mon, May 22, 9:26 PM
grosser added a comment to D33198: Header file to help forcibly link GPURuntime.

Hi Sanjay,

Mon, May 22, 8:28 PM · Unknown Object (Project)
grosser accepted D33423: [Polly][NewPM] Port CodeGen to the new PM.

LGTM

Mon, May 22, 8:23 PM · Unknown Object (Project)
grosser accepted D33422: [Polly][NewPM] Port IslAst to the new ScopPassManager.

LGTM.

Mon, May 22, 8:22 PM · Unknown Object (Project)
grosser accepted D33421: [Polly][NewPM] Port DependenceInfo to the new ScopPassManager..
Mon, May 22, 8:21 PM · Unknown Object (Project)
grosser added a comment to D33411: [Polly] Add handling of Top Level Regions.

Very cool! I will have a look later. Could you possibly add some test cases?

Mon, May 22, 7:54 AM · Unknown Object (Project)
grosser accepted D33387: [Polly] [Docs] Use ReadTheDocs theme if available..

Nice!

Mon, May 22, 1:04 AM
grosser added a dependent revision for D33386: [Polly] [Docs] Fix Sphinx documentation CMake check.: D33387: [Polly] [Docs] Use ReadTheDocs theme if available..
Mon, May 22, 1:02 AM
grosser removed a dependent revision for D33299: [Polly][CMake] Use the CMake Package instead of llvm-config in out-of-tree builds: D33387: [Polly] [Docs] Use ReadTheDocs theme if available..
Mon, May 22, 1:02 AM
grosser edited dependencies for D33387: [Polly] [Docs] Use ReadTheDocs theme if available., added: 1; removed: 1.
Mon, May 22, 1:02 AM
grosser added a dependent revision for D33299: [Polly][CMake] Use the CMake Package instead of llvm-config in out-of-tree builds: D33387: [Polly] [Docs] Use ReadTheDocs theme if available..
Mon, May 22, 1:02 AM
grosser added a dependency for D33387: [Polly] [Docs] Use ReadTheDocs theme if available.: D33299: [Polly][CMake] Use the CMake Package instead of llvm-config in out-of-tree builds.
Mon, May 22, 1:02 AM
grosser removed a dependent revision for D33387: [Polly] [Docs] Use ReadTheDocs theme if available.: D33299: [Polly][CMake] Use the CMake Package instead of llvm-config in out-of-tree builds.
Mon, May 22, 1:01 AM
grosser removed a dependency for D33299: [Polly][CMake] Use the CMake Package instead of llvm-config in out-of-tree builds: D33387: [Polly] [Docs] Use ReadTheDocs theme if available..
Mon, May 22, 1:01 AM
grosser added a dependent revision for D33387: [Polly] [Docs] Use ReadTheDocs theme if available.: D33299: [Polly][CMake] Use the CMake Package instead of llvm-config in out-of-tree builds.
Mon, May 22, 1:01 AM
grosser added a dependency for D33299: [Polly][CMake] Use the CMake Package instead of llvm-config in out-of-tree builds: D33387: [Polly] [Docs] Use ReadTheDocs theme if available..
Mon, May 22, 1:01 AM
grosser accepted D33386: [Polly] [Docs] Fix Sphinx documentation CMake check..

LGTM.

Mon, May 22, 12:59 AM

Sun, May 21

grosser committed rL303514: [ScopInfo] Translate updateDimensionality to isl C++ [NFC].
[ScopInfo] Translate updateDimensionality to isl C++ [NFC]
Sun, May 21, 1:38 PM
grosser committed rL303512: [isl++] add isl_constraint to C++ bindings [NFC].
[isl++] add isl_constraint to C++ bindings [NFC]
Sun, May 21, 1:23 PM
grosser committed rL303511: [ScopInfo] Translate wrapConstantDimensions to isl C++ [NFC].
[ScopInfo] Translate wrapConstantDimensions to isl C++ [NFC]
Sun, May 21, 1:23 PM
grosser committed rL303510: [ScopInfo] Translate addRangeBoundsToSet to isl C++ [NFC].
[ScopInfo] Translate addRangeBoundsToSet to isl C++ [NFC]
Sun, May 21, 1:23 PM
grosser committed rL303509: Fix include order to stop clang-format complains.
Fix include order to stop clang-format complains
Sun, May 21, 9:34 AM
grosser committed rL303508: [ScheduleOptimizer] Move schedule construction to isl C++ [NFC].
[ScheduleOptimizer] Move schedule construction to isl C++ [NFC]
Sun, May 21, 9:21 AM
grosser committed rL303507: [Simplify] Move to isl C++.
[Simplify] Move to isl C++
Sun, May 21, 9:12 AM
grosser committed rL303506: [isl++] Export (almost) all functions from isl.
[isl++] Export (almost) all functions from isl
Sun, May 21, 9:00 AM
grosser committed rL303505: [isl++] Rebase isl C++ bindings on top of 29aee98ce.
[isl++] Rebase isl C++ bindings on top of 29aee98ce
Sun, May 21, 8:59 AM
grosser committed rL303503: [isl++] Move isl raw_ostream printers into separate header.
[isl++] Move isl raw_ostream printers into separate header
Sun, May 21, 6:16 AM

Fri, May 19

grosser added a comment to D33163: [Polly] Added the list of Instructions to output in ScopInfo pass.

Hi Nandini,

Fri, May 19, 9:58 PM
grosser committed rL303485: Fix buildbots after r303429.
Fix buildbots after r303429
Fri, May 19, 9:35 PM
grosser committed rL303420: [ScopDetection] Allow detection of full functions.
[ScopDetection] Allow detection of full functions
Fri, May 19, 5:26 AM

Thu, May 18

grosser committed rL303405: [ScopInfo] Fix typo in documentation.
[ScopInfo] Fix typo in documentation
Thu, May 18, 9:15 PM
grosser committed rL303404: [ScopInfo] Gracefully handle long compile times.
[ScopInfo] Gracefully handle long compile times
Thu, May 18, 8:58 PM
grosser accepted D33255: [Polly][CodeGen] Support partial write accesses..
Thu, May 18, 2:50 PM · Unknown Object (Project)

Wed, May 17

grosser added a comment to D32893: Fix handling of signWrappedSets in access relations.

I do not see a specific reason why we prefer "|" over "||". Just now saw the difference. Maxililian, what is your opinion here?

Wed, May 17, 10:02 PM · Unknown Object (Project)

Tue, May 16

grosser added a comment to D33163: [Polly] Added the list of Instructions to output in ScopInfo pass.

@nandini12396 : Hey Nandini, any update on this? ;-)

Tue, May 16, 9:51 PM

Mon, May 15

grosser added a comment to D32539: [Polly][NewPM][WIP] Add a ScopPassManager.

I tried to fix the buildbots. Can you fix it if I broke something?

Mon, May 15, 8:14 AM · Unknown Object (Project)
grosser added a comment to D32539: [Polly][NewPM][WIP] Add a ScopPassManager.

It seems you forgot to commit the unit tests. Also, please update the commit message in the future.

Mon, May 15, 7:54 AM · Unknown Object (Project)
grosser requested changes to D33198: Header file to help forcibly link GPURuntime.

Otherwise, this looks good.

Mon, May 15, 7:53 AM · Unknown Object (Project)
grosser committed rL303066: Drop nonexisting ScopPassManager directory.
Drop nonexisting ScopPassManager directory
Mon, May 15, 7:26 AM
grosser committed rL303065: Adjust formatting.
Adjust formatting
Mon, May 15, 7:25 AM
grosser accepted D33192: [Polly][CMake] Exclude isl_config from the polly-check-format target..

LGTM

Mon, May 15, 6:21 AM · Unknown Object (Project)
grosser added a comment to D33005: [Polly]Fix code generation bug with Polly vectorizer.

There is no guarantee that every statement must be vectorized. The only requirement is that we should not crash or assert. Feel free to add vectorization support for regions, but my advice is to delay further support until we have evidence that this could speed up performance. ;-)

Mon, May 15, 1:53 AM
grosser added a comment to D33005: [Polly]Fix code generation bug with Polly vectorizer.

Can we just fall back to generating non-vectorized code in these cases. This code path certainly should not break, but as we do not use the Polly vectorizer by default, it is likely not worth spending time on optimizing execution time performance before we find a test case where this is really useful.

Mon, May 15, 12:30 AM

Sun, May 14

grosser added a comment to D32894: Stripping invalid debug information before verification can prevent code-generation.

Hi Sanjay,

Sun, May 14, 10:42 PM · Unknown Object (Project)
grosser committed rL303007: [Simplify] Remove some leftover dead code.
[Simplify] Remove some leftover dead code
Sun, May 14, 2:34 AM
grosser committed rL303006: [Polly] Fix code generation of llvm.expect intrinsic.
[Polly] Fix code generation of llvm.expect intrinsic
Sun, May 14, 2:23 AM
grosser closed D32992: [Polly] Fix code generation of llvm.expect intrinsic by committing rL303006: [Polly] Fix code generation of llvm.expect intrinsic.
Sun, May 14, 2:23 AM

Sat, May 13

grosser added a comment to D32992: [Polly] Fix code generation of llvm.expect intrinsic.

Great. Looks good. Can you also update the commit message to summarize the discussions we had and to explain why we choose the solution now implemented.

Sat, May 13, 11:25 PM
grosser accepted D33143: [Polly][Simplify] Remove identical write removal. NFC..

Ah, so this is just a cleanup. LGTM them.

Sat, May 13, 5:23 AM · Unknown Object (Project)
grosser added a comment to D33138: [Polly][WIP] Make the pattern matching work with modified memory accesses.

Hi Roman,

Sat, May 13, 3:08 AM
grosser added a comment to D33142: [Polly][Simplify] Remove writes that are overwritten..

Btw, this improves the gemm performance nicely from 4s to 1.3 seconds, which matches the results we obtain without pattern matching based opts.

Sat, May 13, 3:02 AM · Unknown Object (Project)
grosser added a comment to D33143: [Polly][Simplify] Remove identical write removal. NFC..

Hi Michael,

Sat, May 13, 2:58 AM · Unknown Object (Project)
grosser accepted D33142: [Polly][Simplify] Remove writes that are overwritten..

L:GTM

Sat, May 13, 2:53 AM · Unknown Object (Project)
grosser accepted D32539: [Polly][NewPM][WIP] Add a ScopPassManager.
Sat, May 13, 2:43 AM · Unknown Object (Project)
grosser accepted D32538: [Polly][NewPM] Port ScopInfo to the new PassManager.
Sat, May 13, 2:43 AM · Unknown Object (Project)
grosser requested changes to D32992: [Polly] Fix code generation of llvm.expect intrinsic.
Sat, May 13, 2:42 AM
grosser added a comment to D32992: [Polly] Fix code generation of llvm.expect intrinsic.

Hi Annanay,

Sat, May 13, 2:42 AM

Fri, May 12

grosser added a comment to D32538: [Polly][NewPM] Port ScopInfo to the new PassManager.

This patch looks good to me. I understand, we can currently not have execution test coverage. I suggest to commit this exceptionally without test cases to at least get compile time test coverage and to also facilitate further testing with the pass manager plugins.

Fri, May 12, 5:49 AM · Unknown Object (Project)
grosser accepted D31459: [Polly][NewPM] Port ScopDetection to the new PassManager.

This patch looks good to me. I understand, we can currently not have execution test coverage. I suggest to commit this exceptionally without test cases to at least get compile time test coverage and to also facilitate further testing with the pass manager plugins.

Fri, May 12, 5:48 AM · Unknown Object (Project)
grosser added a comment to D32539: [Polly][NewPM][WIP] Add a ScopPassManager.
  1. The ability to actually run (and test) individual passes with 'opt'
  2. The ability to run polly in -O3

    I assume 2) is only possible after your patch has been accepted by Chandler. As mentioned earlier, Polly-specific #ifdefs might be a good idea to early on establish a full flow.

    I am not clear what is missing for 1).
  1. is tied very closely to PassBuilders ability to accept injected plugin passes, yes. On the other hand, while it would be technically possible to forward declare the polly passes in PassBuilder (and the PassRegistry.def), that would pose a severe violation of layering, because we'd need to make lib/Passes depend on polly, both logically as well as in a build-system sense. This dependency doesn't currently exist: LLVM_LINK_POLLY_INTO_TOOLS (which would be the flag we'd #ifdef on), only affects tools. Forward declaring Polly in the PassBuilder would however pull polly dependencies into the PassBuilder.

    The good news is I've talked to Chandler about this some more, and there will be progress on the Plugins patch soon!
Fri, May 12, 5:47 AM · Unknown Object (Project)
grosser added a comment to D32894: Stripping invalid debug information before verification can prevent code-generation.

@sanyam: Sorry, I am a little busy today, but can you please try to generate parallel code from C (with debug info) . I am almost certain it will also crash the same way (at least for the first bug). I should reply to your other question tomorrow.

Fri, May 12, 12:52 AM · Unknown Object (Project)

Thu, May 11

grosser added a comment to D32894: Stripping invalid debug information before verification can prevent code-generation.

Interesting. As another data point, can you try a single-dimensional parallel loop, where you add two vectors and see if the polly openmp backend crashes here as well?

Thu, May 11, 11:40 AM · Unknown Object (Project)
grosser accepted D33089: [Polly] Generate more 'canonical' induction variable if we can prove there is no overflow..

LGTM, assuming you update the commit message accordingly.

Thu, May 11, 10:52 AM

Wed, May 10

grosser added a comment to D33089: [Polly] Generate more 'canonical' induction variable if we can prove there is no overflow..

I just posted to bug reports that cover are not directly related to this patch, but touch a very similar topic: http://llvm.org/PR32997, http://llvm.org/PR32998

Wed, May 10, 11:43 PM
grosser added a comment to D33089: [Polly] Generate more 'canonical' induction variable if we can prove there is no overflow..

I don't think checking with mayOverflowOnIV is needed. The types we use today are anyway just a guess, as we have no idea what type is needed. Consequently checking for possible overflows -- while we already may have some overflows just gives a wrong sense of security.

Wed, May 10, 11:26 PM
grosser added a comment to D32739: [Polly] JSONImporter misses checks whether the data it imports makes sense..

@niosega : Any updates here?

Wed, May 10, 11:50 AM
grosser added a comment to D31859: Initializing NVPTX backend within Polly.

Sorry, I misread the last message. I understand this patch is indeed needed. I am fine for it to be committed as it is. I leave the last word to Michael, as he already started the review.

Wed, May 10, 10:57 AM · Unknown Object (Project)
grosser accepted D33054: [Fix] [Fortran Support] Fix variable name & make testcase activate on release.

buildbots should send an email.

Wed, May 10, 10:35 AM
grosser committed rL302636: [Polly] Canonicalize arrays according to base-ptr equivalence class.
[Polly] Canonicalize arrays according to base-ptr equivalence class
Wed, May 10, 4:13 AM
grosser closed D28518: [Polly] Canonicalize arrays according to base-ptr equivalence class by committing rL302636: [Polly] Canonicalize arrays according to base-ptr equivalence class.
Wed, May 10, 4:13 AM · Unknown Object (Project)

Tue, May 9

grosser added a comment to D32992: [Polly] Fix code generation of llvm.expect intrinsic.

Hi Annay,

Tue, May 9, 10:39 PM
grosser committed rL302621: Add noreturn attribute to avoid warnings about missing initialization.
Add noreturn attribute to avoid warnings about missing initialization
Tue, May 9, 10:34 PM