Page MenuHomePhabricator

jingyue (Jingyue Wu)
User

Projects

User does not belong to any projects.

User Details

User Since
Mar 27 2014, 11:59 PM (277 w, 5 d)

Recent Activity

Nov 30 2016

jingyue added inline comments to D9819: [TTI/CostModel] improve TTI::getGEPCost and use it in CostModel::getInstructionCost.
Nov 30 2016, 3:07 PM

Aug 22 2016

jingyue added a comment to D23749: [NVPTX] Add NVPTXHoistAddrSpaceCast pass..

LGTM after you address David's comment -- A post-dominates B doesn't guarantee execution that reaches B will arrive A.

Aug 22 2016, 12:03 PM

Aug 18 2016

jingyue accepted D23694: [NVPTX] Switch nvptx-use-infer-addrspace to true..

Yay! Thanks for testing this!

Aug 18 2016, 3:45 PM

Jul 15 2016

jingyue committed rL275596: [ReassociateGEP] Update tests to allow missing "inbounds" on certain GEPs..
[ReassociateGEP] Update tests to allow missing "inbounds" on certain GEPs.
Jul 15 2016, 11:54 AM

Jul 11 2016

jingyue committed rL275083: [SLSR] Call getPointerSizeInBits with the correct address space..
[SLSR] Call getPointerSizeInBits with the correct address space.
Jul 11 2016, 11:20 AM

Jul 9 2016

jingyue committed rL274982: [SLSR] Fix crash on handling 128-bit integers..
[SLSR] Fix crash on handling 128-bit integers.
Jul 9 2016, 12:20 PM

Jul 8 2016

jingyue committed rL274940: [TTI] Expose TTI::getGEPCost and use it in SLSR and NaryReassociate..
[TTI] Expose TTI::getGEPCost and use it in SLSR and NaryReassociate.
Jul 8 2016, 2:55 PM

Jun 24 2016

jingyue accepted D21421: [NVPTX] Improve lowering of byval args of device functions..
Jun 24 2016, 3:48 PM

Jun 15 2016

jingyue added inline comments to D21421: [NVPTX] Improve lowering of byval args of device functions..
Jun 15 2016, 8:32 PM

May 25 2016

jingyue accepted D20644: [NVPTX] Added NVVMIntrRange pass .
May 25 2016, 3:31 PM
jingyue added a comment to D20644: [NVPTX] Added NVVMIntrRange pass .

Have you considered letting Clang (instead of a late-stage IR pass) add these ranges? These ranges are useful for some target-independent IR passes, e.g. those using ValueTracking (D4150).

May 25 2016, 2:50 PM

May 4 2016

jingyue committed rL268522: [NVPTX:doc] Update code examples to conform to the latest IR..
[NVPTX:doc] Update code examples to conform to the latest IR.
May 4 2016, 10:41 AM

Apr 29 2016

jingyue accepted D19732: DivergenceAnalysis: Fix crash with no return blocks.
Apr 29 2016, 5:50 PM

Apr 28 2016

jingyue accepted D19615: [NVPTX] Fix sign/zero-extending ldg/ldu instruction selection.

Looks great! Thanks.

Apr 28 2016, 1:17 PM
jingyue accepted D19646: DivergenceAnalysis: Fix crash with unreachable blocks.
Apr 28 2016, 12:41 PM
jingyue added inline comments to D19646: DivergenceAnalysis: Fix crash with unreachable blocks.
Apr 28 2016, 10:45 AM
jingyue requested changes to D19646: DivergenceAnalysis: Fix crash with unreachable blocks.

Thanks! I completely missed this case :(

Apr 28 2016, 10:14 AM
jingyue added inline comments to D19615: [NVPTX] Fix sign/zero-extending ldg/ldu instruction selection.
Apr 28 2016, 9:59 AM
jingyue added a comment to D19615: [NVPTX] Fix sign/zero-extending ldg/ldu instruction selection.

Awesome! Thanks Justin.

Apr 28 2016, 9:17 AM

Apr 26 2016

jingyue accepted D19518: Add optimization bisect opt-in calls for NVPTX passes.

Oh, I see what you mean now.

So NVPTXPeephole should be guarded by CodeGenOpt::None and thus is safe to skip.

The optnone-llc.ll test verifies that nothing is skipped which is run at -O0, so as long as this pass is being run at -O0 we can't add the call to skip it for "optnone" functions and bisection. If the target machine code were updated to skip the peephole pass at -O0 then the skip check could be added.

Apr 26 2016, 4:08 PM
jingyue committed rL267619: [NVPTX] Fix some usages of CodeGenOpt::None..
[NVPTX] Fix some usages of CodeGenOpt::None.
Apr 26 2016, 4:05 PM
jingyue requested changes to D19518: Add optimization bisect opt-in calls for NVPTX passes.

Oh, I see what you mean now.

Apr 26 2016, 11:00 AM
jingyue added a comment to D19518: Add optimization bisect opt-in calls for NVPTX passes.

Hi Andrew,

Apr 26 2016, 10:35 AM
jingyue accepted D19521: SLSR: Use UnknownAddressSpace instead of default.

Maybe change the title to be "... instead of default for pure arithmetic instructions". Otherwise, LGTM!

Apr 26 2016, 10:28 AM

Apr 12 2016

jingyue accepted D19013: [DivergenceAnalysis] Treat PHI with incoming undef as constant.

Thanks!

Apr 12 2016, 10:12 AM

Apr 4 2016

jingyue added a comment to D18053: [NVPTX] Handle ldg created from sign-/zero-extended load.

Should we land this? It will fix PR26185.

Apr 4 2016, 10:36 PM

Mar 31 2016

jingyue accepted D17872: [NVPTX] Add a truncate DAG node to some calls..

Other than that FIXME, pretty straightforward. LGTM

Mar 31 2016, 4:01 PM

Mar 29 2016

jingyue committed rL264839: [docs] Add gpucc publication and tutorial..
[docs] Add gpucc publication and tutorial.
Mar 29 2016, 10:32 PM

Mar 22 2016

jingyue closed D12246: [NVPTX] change threading intrinsics from noduplicate to convergent.

D18168 duplicates this and is submitted.

Mar 22 2016, 3:24 PM
jingyue accepted D18168: [NVVM] Remove noduplicate attribute from synchronizing intrinsics..

Thanks for working on this long overdue feature!

Mar 22 2016, 12:01 PM

Mar 20 2016

jingyue committed rL263916: [NVPTX] Adds a new address space inference pass..
[NVPTX] Adds a new address space inference pass.
Mar 20 2016, 2:04 PM
jingyue closed D17965: [NVPTX] Adds a new address space inference pass..
Mar 20 2016, 2:04 PM
jingyue added a comment to D17965: [NVPTX] Adds a new address space inference pass..

All comments addressed. Submitting...

Mar 20 2016, 1:52 PM
jingyue updated the diff for D17965: [NVPTX] Adds a new address space inference pass..

Some more minor changes.

Mar 20 2016, 1:45 PM

Mar 18 2016

jingyue added inline comments to D17965: [NVPTX] Adds a new address space inference pass..
Mar 18 2016, 5:32 PM
jingyue updated the diff for D17965: [NVPTX] Adds a new address space inference pass..

jlebar@'s comments

Mar 18 2016, 5:31 PM

Mar 16 2016

jingyue added a comment to D17965: [NVPTX] Adds a new address space inference pass..

Thanks a lot for the review! I understand it's a lot of work :) I answered one of your high-level questions. I'm still OOO and will get to other comments later.

Mar 16 2016, 1:19 AM

Mar 15 2016

jingyue updated subscribers of D17872: [NVPTX] Add a truncate DAG node to some calls..
Mar 15 2016, 10:56 AM
jingyue resigned from D17872: [NVPTX] Add a truncate DAG node to some calls..

I am not clear with this part either. jholewinski@, can you comment on this?

Mar 15 2016, 10:56 AM
jingyue abandoned D9136: Reassociate GEP operands for loop invariant code motion.

Sorry, I pressed the wrong button. I meant to say "needs revision". Feel free to reclaim this patch.

Mar 15 2016, 10:54 AM
jingyue abandoned D10327: Reassociate in favor of grouping previously paired operands.

Pressed the wrong button. Meant to say "needs revision". Feel free to reclaim it.

Mar 15 2016, 10:53 AM
jingyue abandoned D9149: [SeparateConstOffsetFromGEP] Bail there is only one GEP offset.

Oops... I pressed the wrong button. I meant to say "need revision". jmolly@, feel free to reclaim it.

Mar 15 2016, 10:52 AM
jingyue abandoned D12414: [NVPTX] add an NVPTX-specific alias analysis.

It doesn't work with the new alias analysis infrastructure.

Mar 15 2016, 10:42 AM
jingyue added a comment to D14000: [CUDA] Allow code generation for functions with target attributes that don't match compilation mode..

Is this patch obsolete? Are you still trying to push it in?

Mar 15 2016, 10:33 AM
jingyue commandeered D10327: Reassociate in favor of grouping previously paired operands.
Mar 15 2016, 10:31 AM
jingyue commandeered D9149: [SeparateConstOffsetFromGEP] Bail there is only one GEP offset.
Mar 15 2016, 10:31 AM
jingyue commandeered D9136: Reassociate GEP operands for loop invariant code motion.
Mar 15 2016, 10:30 AM
jingyue updated subscribers of D9136: Reassociate GEP operands for loop invariant code motion.
Mar 15 2016, 10:30 AM
jingyue resigned from D9168: [NVPTX] Check if callsite is defined when computing argument allignment.
Mar 15 2016, 10:30 AM

Mar 11 2016

jingyue added a reviewer for D18053: [NVPTX] Handle ldg created from sign-/zero-extended load: jlebar.

I am OOO and maybe unable to review this until next week.

Mar 11 2016, 3:55 AM

Mar 8 2016

jingyue retitled D17965: [NVPTX] Adds a new address space inference pass. from to [NVPTX] Adds a new address space inference pass..
Mar 8 2016, 11:07 AM

Mar 3 2016

jingyue added inline comments to D17872: [NVPTX] Add a truncate DAG node to some calls..
Mar 3 2016, 5:25 PM

Feb 23 2016

jingyue committed rL261706: [doc] Obtaining help on LLVM's CUDA support..
[doc] Obtaining help on LLVM's CUDA support.
Feb 23 2016, 3:39 PM

Feb 20 2016

jingyue accepted D17317: [attrs] Handle convergent CallSites..

Happy!

Feb 20 2016, 9:06 PM

Feb 16 2016

jingyue accepted D17314: Update langref to indicate that calls may be convergent..
Feb 16 2016, 10:03 PM
jingyue added inline comments to D17317: [attrs] Handle convergent CallSites..
Feb 16 2016, 10:02 PM
jingyue accepted D17318: [NVPTX] Annotate convergent intrinsics as convergent..
Feb 16 2016, 9:09 PM
jingyue accepted D17319: [NVPTX] Test that MachineSink won't sink across llvm.cuda.syncthreads..
Feb 16 2016, 8:53 PM

Feb 12 2016

jingyue added a comment to D17128: [SimplifyCFG] Don't fold conditional branches that contain calls to convergent functions..

Lgtm

Feb 12 2016, 1:04 PM

Feb 11 2016

jingyue accepted D17127: [LoopRotate] Don't perform loop rotation if the loop header calls a convergent function..
Feb 11 2016, 5:39 PM
jingyue added inline comments to D17128: [SimplifyCFG] Don't fold conditional branches that contain calls to convergent functions..
Feb 11 2016, 5:20 PM
jingyue added a comment to D16870: [CUDA] Tweak attribute-based overload resolution to match nvcc behavior. .

I'll defer to Justin's approval.

Feb 11 2016, 4:31 PM
jingyue accepted D17125: Initialize CodeMetrics' member variables inline with definitions..
Feb 11 2016, 10:21 AM

Feb 9 2016

jingyue accepted D17012: Update document about convergent attribute..

Just a reminder if you haven't done that already, double-check how the web page looks like before you commit.

Feb 9 2016, 1:34 PM
jingyue accepted D17013: Add convergent-removing bits to FunctionAttrs pass..
Feb 9 2016, 1:20 PM

Feb 8 2016

jingyue added inline comments to D17013: Add convergent-removing bits to FunctionAttrs pass..
Feb 8 2016, 7:26 PM
jingyue updated subscribers of D17012: Update document about convergent attribute..
Feb 8 2016, 7:15 PM

Feb 5 2016

jingyue added a comment to D16941: [NVPTX] Mark nvvm synchronizing intrinsics as convergent..

I was referring to this paragraph

Barriers are executed on a per-warp basis as if all the threads in a warp are active. Thus, if any thread in a warp executes a bar instruction, it is as if all the threads in the warp have executed the bar instruction. All threads in the warp are stalled until the barrier completes, and the arrival count for the barrier is incremented by the warp size (not the number of active threads in the warp). In conditionally executed code, a bar instruction should only be used if it is known that all threads evaluate the condition identically (the warp does not diverge). Since barriers are executed on a per-warp basis, the optional thread count must be a multiple of the warp size.

Read more at: http://docs.nvidia.com/cuda/parallel-thread-execution/index.html#ixzz3zMFDtpKf

Feb 5 2016, 8:38 PM
jingyue added a comment to D16941: [NVPTX] Mark nvvm synchronizing intrinsics as convergent..

LGTM, but do you have a test where LLVM generates wrong code if __syncthreads is not marked convergent?

Feb 5 2016, 6:04 PM
jingyue resigned from D14976: Fixed a failure in cost calculation for vector GEP.

Committed in r254408.

Feb 5 2016, 5:49 PM
jingyue added inline comments to D16870: [CUDA] Tweak attribute-based overload resolution to match nvcc behavior. .
Feb 5 2016, 11:00 AM

Feb 4 2016

jingyue accepted D16870: [CUDA] Tweak attribute-based overload resolution to match nvcc behavior. .
Feb 4 2016, 11:17 AM

Feb 3 2016

jingyue added inline comments to D16870: [CUDA] Tweak attribute-based overload resolution to match nvcc behavior. .
Feb 3 2016, 9:40 PM
jingyue committed rL259749: [NVPTX] Disable performance optimizations when OptLevel==None.
[NVPTX] Disable performance optimizations when OptLevel==None
Feb 3 2016, 8:19 PM
jingyue closed D16874: [NVPTX] Disable performance optimizations when OptLevel==None.
Feb 3 2016, 8:19 PM
jingyue added inline comments to D16874: [NVPTX] Disable performance optimizations when OptLevel==None.
Feb 3 2016, 8:18 PM
jingyue updated the diff for D16874: [NVPTX] Disable performance optimizations when OptLevel==None.

More comments.

Feb 3 2016, 8:18 PM
jingyue retitled D16874: [NVPTX] Disable performance optimizations when OptLevel==None from to [NVPTX] Disable performance optimizations when OptLevel==None.
Feb 3 2016, 6:08 PM
jingyue added inline comments to D16870: [CUDA] Tweak attribute-based overload resolution to match nvcc behavior. .
Feb 3 2016, 5:43 PM

Jan 30 2016

jingyue committed rL259307: [doc] improve the doc for CUDA.
[doc] improve the doc for CUDA
Jan 30 2016, 3:52 PM

Jan 22 2016

jingyue committed rL258536: [NVPTX] expand mul_lohi to mul_lo and mul_hi.
[NVPTX] expand mul_lohi to mul_lo and mul_hi
Jan 22 2016, 11:51 AM
jingyue closed D16479: [NVPTX] expand mul_lohi to mul_lo and mul_hi.
Jan 22 2016, 11:51 AM
jingyue retitled D16479: [NVPTX] expand mul_lohi to mul_lo and mul_hi from to [NVPTX] expand mul_lohi to mul_lo and mul_hi.
Jan 22 2016, 11:34 AM

Dec 18 2015

jingyue committed rL256036: [DivergenceAnalysis] fix a bug in computing influence regions.
[DivergenceAnalysis] fix a bug in computing influence regions
Dec 18 2015, 1:47 PM
jingyue committed rL256035: [NaryReassociate] allow candidate to have a different type.
[NaryReassociate] allow candidate to have a different type
Dec 18 2015, 1:40 PM
jingyue closed D15618: [NaryReassociate] allow candidate to have a different type by committing rL256035: [NaryReassociate] allow candidate to have a different type.
Dec 18 2015, 1:39 PM
jingyue added inline comments to D15618: [NaryReassociate] allow candidate to have a different type.
Dec 18 2015, 1:38 PM

Dec 17 2015

jingyue added a reviewer for D15618: [NaryReassociate] allow candidate to have a different type: tra.
Dec 17 2015, 11:25 AM
jingyue retitled D15618: [NaryReassociate] allow candidate to have a different type from to [NaryReassociate] allow candidate to have a different type.
Dec 17 2015, 11:06 AM

Nov 29 2015

jingyue requested changes to D14976: Fixed a failure in cost calculation for vector GEP.

LGTM with some minors

Nov 29 2015, 12:14 PM

Nov 25 2015

jingyue added inline comments to D14976: Fixed a failure in cost calculation for vector GEP.
Nov 25 2015, 9:49 AM

Nov 18 2015

jingyue committed rL253509: [doc] fix a wrong link.
[doc] fix a wrong link
Nov 18 2015, 2:04 PM

Nov 17 2015

jingyue abandoned D14452: [CUDA] mixed-mode support.
Nov 17 2015, 6:02 PM
jingyue committed rL253427: [doc] simplify the doc on compiling CUDA.
[doc] simplify the doc on compiling CUDA
Nov 17 2015, 6:01 PM

Nov 10 2015

jingyue committed rL252660: [doc] Compile CUDA with LLVM.
[doc] Compile CUDA with LLVM
Nov 10 2015, 2:38 PM
jingyue closed D14370: [doc] Compile CUDA with LLVM.
Nov 10 2015, 2:38 PM
jingyue updated the diff for D14370: [doc] Compile CUDA with LLVM.

Replace the link to the raw diff with more instructions.

Nov 10 2015, 2:36 PM
jingyue updated the diff for D14370: [doc] Compile CUDA with LLVM.

Simplify the command lines and header file inclusion

Nov 10 2015, 1:18 PM

Nov 6 2015

jingyue added a comment to D14370: [doc] Compile CUDA with LLVM.

My biggest concern is to avoid giving users the false impression that what is described here is an officially supported long-term interface from clang. Would it be accurate to say that this document is meant for "LLVM developers" (or otherwise people working inside LLVM)?

Other than clarifying that, the content LGTM.

Nov 6 2015, 10:56 PM
jingyue updated the diff for D14370: [doc] Compile CUDA with LLVM.

simplify the doc

Nov 6 2015, 10:53 PM