similar to: [LLVMdev] ValueTracking's GetUnderlyingObject vs. ScheduleDAGInstrs' getUnderlyingObject

Hi Matthew, > I'm investigating a problem with "Machine Instruction Scheduling" that is rooted > in incorrect alias information. > > Things go wrong in the "Merge disjoint stack slots"[1] pass when a store > instruction fails to get updated after its stack slot is merged. As a result, > when "Machine Instruction Scheduling" runs, it incorrectly

I notice that GetUnderlyingObject has a few checks, but alloca isn't one of them. Then it fall backs to SimplifyInstruction which doesn't know about alloca so falls back to just trying to constant fold it. This seems a little silly since I assume alloca can't be constant folded. Should we just detect this early in GetUnderlyingObject and stop? ~Craig -------------- next part

Yep. Makes sense to me. There's nothing to simplify or constant-fold about an alloca. -Hal On 04/12/2017 04:23 PM, Craig Topper wrote: > Ping > > ~Craig > > On Fri, Apr 7, 2017 at 1:25 PM, Craig Topper <craig.topper at gmail.com > <mailto:craig.topper at gmail.com>> wrote: > > I notice that GetUnderlyingObject has a few checks, but alloca >

Hi, Can anyone help me to understand the ScheduleDAGInstrs::buildSchedGraph() method? I find the handling of AliasChain is disturbing since: 1. A new alias chain add deps to all possibly aliasing SUs, and then clears those lists. 2. When AliasChain is present, the addChainDependency() method is called, but the target hook areMemAccessesTriviallyDisjoint() called inside

Hello again, Sorry -- I think I found the problem somewhere else. I was a bit confused and missed the fact that adjustChainDeps() is called a few lines down and does just what I wanted :-) I would like to instead ask another question: Why is I->isCtrl() used in code like // Iterate over chain dependencies only. for (SUnit::const_succ_iterator I = SUb->Succs.begin(), E =

Hello, While tuning the MIScheduler for my target, I discovered a code that unnecessarily restricts the scheduler. I think this is a bug, but I would appreciate a second opinion. In file ScheduleDAGInstrs.cpp, the function MIsNeedChainEdge determines whether two MachineInstrs are ordered by a memory dependence. It first runs through the standard criteria (Do both instructions access memory?

After discussing with Nick Lewycky in the IRC channel, here comes a less aggressive patch. We were worried that a constant pointer could alias with a GlobalValue. Also, if one pointer could be a GlobalValue and the other a GlobalAlias with the GlobalValue as aliasee. However, I was not able to produce a test where this happens wihout the getUnderlyingObject() returning the same value. It

Hans Wennborg wrote: > After discussing with Nick Lewycky in the IRC channel, here comes a less > aggressive patch. > > We were worried that a constant pointer could alias with a GlobalValue. > Also, if one pointer could be a GlobalValue and the other a GlobalAlias > with the GlobalValue as aliasee. > However, I was not able to produce a test where this happens wihout the

This is the email I sent last week. ---------- Forwarded message ---------- From: Akira Hatanaka <ahatanak at gmail.com> Date: Wed, May 8, 2013 at 7:04 PM Subject: [PATCH] Minor fix to StackColoring to avoid needlessly clearing mem operands. To: LLVM Developers Mailing List <llvmdev at cs.uiuc.edu> The following code snippet taken from StackColoring::remapInstructions clears a mem

The following code snippet taken from StackColoring::remapInstructions clears a mem operand if it can't guarantee whether the memoperand's underlying value aliases with the merged allocas: // Update the MachineMemOperand to use the new alloca. 522 for (MachineInstr::mmo_iterator MM = I->memoperands_begin(), .... // Climb up and find the original alloca. 532 V =

Dan Gohman wrote: > On Nov 4, 2009, at 6:43 AM, Hans Wennborg wrote: > >> Here is another change I'd like to suggest to the BasicAliasAnalysis. >> >> LLVM fails to remove the dead store in the following code: >> >> %t = type { i32 } >> >> define void @f(%t* noalias nocapture %stuff ) { >> %p = getelementptr inbounds %t* %stuff, i32 0,

On Nov 4, 2009, at 6:43 AM, Hans Wennborg wrote: > Here is another change I'd like to suggest to the BasicAliasAnalysis. > > LLVM fails to remove the dead store in the following code: > > %t = type { i32 } > > define void @f(%t* noalias nocapture %stuff ) { > %p = getelementptr inbounds %t* %stuff, i32 0, i32 0 > > store i32 1, i32* %p; <-- This

Hi, Thank you for the reply. >It looks to me like we can choose any subset of edges here and be correct. We're basically trying to prune/pinch the DAG edges here. They can easily blow up with AA sched. I would guess that isCtrl() edges are good ones to bypass because they could be a low-latecy edges, whereas true data dependencies from a load are expected to be >higher latency, so they

----- Original Message ----- > From: "Chris Lattner" <clattner at apple.com> > To: "Chandler Carruth" <chandlerc at gmail.com> > Cc: "LLVM Developers Mailing List" <llvmdev at cs.uiuc.edu>, "Hal Finkel" <hfinkel at anl.gov>, "Justin Bogner" > <mail at justinbogner.com>, "Duncan Exon Smith"

I am stuck trying to adapt my out-of-target implementation to build on SVN head (actually the v3.8 branch, rev #257626). This is currently working on the v3.7.1 sources, but the changes to 'llvm::ScheduleDAGInstrs' have me stumped as to how to revise my implementation to track the changes to this class. Our 'SHAVEAsmScheduler' derives from 'ScheduleDAGInstrs' and uses

On Mon, Jul 13, 2015 at 10:21 PM Chris Lattner <clattner at apple.com> wrote: > > > On Jul 13, 2015, at 8:19 PM, Chandler Carruth <chandlerc at gmail.com> > wrote: > > > > Ok folks, > > > > I wrote up the general high-level thoughts I have about stateful AA in a > separate thread. But we need to sort out the completely and horribly broken >

Hi all, One of the reasons the Livermore Loops couldn't be vectorized is that it was using global structures to hold the arrays. Today, I'm investigating why is that so and how to fix it. My investigation brought me to LoopVectorizationLegality::canVectorizeMemory(): if (WriteObjects.count(*it)) { DEBUG(dbgs() << "LV: Found a possible read/write reorder:"

Dan Gohman wrote: > Hello, > > On Nov 4, 2009, at 1:51 AM, Hans Wennborg wrote: >> >> / Hans >> Index: lib/Analysis/BasicAliasAnalysis.cpp >> =================================================================== >> --- lib/Analysis/BasicAliasAnalysis.cpp (revision 86023) >> +++ lib/Analysis/BasicAliasAnalysis.cpp (working copy) >> @@ -633,6 +633,15

Replying here, but several of the questions raised boil down to "couldn't you make the usage of GetUnderlyingObject conservatively correct?". I'll try and address that. I think this *is* the right approach, but I think it is very hard to do without effectively disabling this part of GlobalsModRef. That is, the easy ways are likely to fire very frequently IMO. The core idea is

Ok folks, I wrote up the general high-level thoughts I have about stateful AA in a separate thread. But we need to sort out the completely and horribly broken aspects of GlobalsModRef today, and the practical steps forward. This email is totally about the practical stuff. Now, as to why I emailed this group of people and with this subject, the only pass pipeline that includes GlobalsModRef, is