[llvm-commits] [llvm] r162912 - in /llvm/trunk: lib/Transforms/Scalar/LoopRotation.cpp test/Transforms/LoopRotate/multiple-exits.ll
Duncan Sands
baldrick at free.fr
Sat Sep 1 01:42:38 PDT 2012
Hi Benjamin, probably this caused this expensive checks buildbot failure:
http://lab.llvm.org:8011/builders/dragonegg-x86_64-linux-gcc-4.6-self-host-checks/builds/1535
The failure is:
DominatorTree is not up to date!
Computed:
=============================--------------------------------
Inorder Dominator Tree: DFSNumbers invalid: 6 slow queries.
...
I'm working on a testcase. Unfortunately I haven't yet been able to reproduce
with "opt" on a dump of the bitcode from the failing example.
Ciao, Duncan.
On 30/08/12 17:39, Benjamin Kramer wrote:
> Author: d0k
> Date: Thu Aug 30 10:39:42 2012
> New Revision: 162912
>
> URL: http://llvm.org/viewvc/llvm-project?rev=162912&view=rev
> Log:
> LoopRotate: Also rotate loops with multiple exits.
>
> The old PHI updating code in loop-rotate was replaced with SSAUpdater a while
> ago, it has no problems with comples PHIs. What had to be fixed is detecting
> whether a loop was already rotated and updating dominators when multiple exits
> were present.
>
> This change increases overall code size a bit, mostly due to additional loop
> unrolling opportunities. Passes test-suite and selfhost with -verify-dom-info.
> Fixes PR7447.
>
> Thanks to Andy for the input on the domtree updating code.
>
> Added:
> llvm/trunk/test/Transforms/LoopRotate/multiple-exits.ll
> Modified:
> llvm/trunk/lib/Transforms/Scalar/LoopRotation.cpp
>
> Modified: llvm/trunk/lib/Transforms/Scalar/LoopRotation.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopRotation.cpp?rev=162912&r1=162911&r2=162912&view=diff
> ==============================================================================
> --- llvm/trunk/lib/Transforms/Scalar/LoopRotation.cpp (original)
> +++ llvm/trunk/lib/Transforms/Scalar/LoopRotation.cpp Thu Aug 30 10:39:42 2012
> @@ -24,6 +24,7 @@
> #include "llvm/Transforms/Utils/BasicBlockUtils.h"
> #include "llvm/Transforms/Utils/SSAUpdater.h"
> #include "llvm/Transforms/Utils/ValueMapper.h"
> +#include "llvm/Support/CFG.h"
> #include "llvm/Support/Debug.h"
> #include "llvm/ADT/Statistic.h"
> using namespace llvm;
> @@ -256,6 +257,7 @@
> return false;
>
> BasicBlock *OrigHeader = L->getHeader();
> + BasicBlock *OrigLatch = L->getLoopLatch();
>
> BranchInst *BI = dyn_cast<BranchInst>(OrigHeader->getTerminator());
> if (BI == 0 || BI->isUnconditional())
> @@ -267,13 +269,9 @@
> if (!L->isLoopExiting(OrigHeader))
> return false;
>
> - // Updating PHInodes in loops with multiple exits adds complexity.
> - // Keep it simple, and restrict loop rotation to loops with one exit only.
> - // In future, lift this restriction and support for multiple exits if
> - // required.
> - SmallVector<BasicBlock*, 8> ExitBlocks;
> - L->getExitBlocks(ExitBlocks);
> - if (ExitBlocks.size() > 1)
> + // If the loop latch already contains a branch that leaves the loop then the
> + // loop is already rotated.
> + if (OrigLatch == 0 || L->isLoopExiting(OrigLatch))
> return false;
>
> // Check size of original header and reject loop if it is very big.
> @@ -286,11 +284,10 @@
>
> // Now, this loop is suitable for rotation.
> BasicBlock *OrigPreheader = L->getLoopPreheader();
> - BasicBlock *OrigLatch = L->getLoopLatch();
>
> // If the loop could not be converted to canonical form, it must have an
> // indirectbr in it, just give up.
> - if (OrigPreheader == 0 || OrigLatch == 0)
> + if (OrigPreheader == 0)
> return false;
>
> // Anything ScalarEvolution may know about this loop or the PHI nodes
> @@ -298,6 +295,8 @@
> if (ScalarEvolution *SE = getAnalysisIfAvailable<ScalarEvolution>())
> SE->forgetLoop(L);
>
> + DEBUG(dbgs() << "LoopRotation: rotating "; L->dump());
> +
> // Find new Loop header. NewHeader is a Header's one and only successor
> // that is inside loop. Header's other successor is outside the
> // loop. Otherwise loop is not suitable for rotation.
> @@ -408,10 +407,16 @@
> // Update DominatorTree to reflect the CFG change we just made. Then split
> // edges as necessary to preserve LoopSimplify form.
> if (DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>()) {
> - // Since OrigPreheader now has the conditional branch to Exit block, it is
> - // the dominator of Exit.
> - DT->changeImmediateDominator(Exit, OrigPreheader);
> - DT->changeImmediateDominator(NewHeader, OrigPreheader);
> + // Everything that was dominated by the old loop header is now dominated
> + // by the original loop preheader. Conceptually the header was merged
> + // into the preheader, even though we reuse the actual block as a new
> + // loop latch.
> + DomTreeNode *OrigHeaderNode = DT->getNode(OrigHeader);
> + SmallVector<DomTreeNode *, 8> HeaderChildren(OrigHeaderNode->begin(),
> + OrigHeaderNode->end());
> + DomTreeNode *OrigPreheaderNode = DT->getNode(OrigPreheader);
> + for (unsigned I = 0, E = HeaderChildren.size(); I != E; ++I)
> + DT->changeImmediateDominator(HeaderChildren[I], OrigPreheaderNode);
>
> // Update OrigHeader to be dominated by the new header block.
> DT->changeImmediateDominator(OrigHeader, OrigLatch);
> @@ -440,6 +445,46 @@
> // Update OrigHeader to be dominated by the new header block.
> DT->changeImmediateDominator(NewHeader, OrigPreheader);
> DT->changeImmediateDominator(OrigHeader, OrigLatch);
> +
> + // Brute force incremental dominator tree update. Call
> + // findNearestCommonDominator on all CFG predecessors of each child of the
> + // original header.
> + DomTreeNode *OrigHeaderNode = DT->getNode(OrigHeader);
> + SmallVector<DomTreeNode *, 8> WorkList(OrigHeaderNode->begin(),
> + OrigHeaderNode->end());
> + while (!WorkList.empty()) {
> + DomTreeNode *Node = WorkList.pop_back_val();
> + BasicBlock *BB = Node->getBlock();
> + BasicBlock *NearestDom = 0;
> + for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB); PI != PE;
> + ++PI) {
> + BasicBlock *Pred = *PI;
> +
> + // We have to process predecessors of a node before we touch the
> + // actual node. If one of the predecessors is in our worklist, put it
> + // and the currently processed node on the worklist and go processing
> + // the predecessor.
> + SmallVectorImpl<DomTreeNode *>::iterator I =
> + std::find(WorkList.begin(), WorkList.end(), DT->getNode(Pred));
> + if (I != WorkList.end()) {
> + WorkList.push_back(Node);
> + std::swap(*I, WorkList.back());
> + // The predecessor is now at the end of the worklist.
> + NearestDom = 0;
> + break;
> + }
> +
> + // On the first iteration start with Pred, on the other iterations we
> + // narrow it down to the nearest common dominator.
> + if (!NearestDom)
> + NearestDom = Pred;
> + else
> + NearestDom = DT->findNearestCommonDominator(NearestDom, Pred);
> + }
> +
> + if (NearestDom)
> + DT->changeImmediateDominator(BB, NearestDom);
> + }
> }
> }
>
> @@ -452,6 +497,8 @@
> // emitted code isn't too gross in this common case.
> MergeBlockIntoPredecessor(OrigHeader, this);
>
> + DEBUG(dbgs() << "LoopRotation: into "; L->dump());
> +
> ++NumRotated;
> return true;
> }
>
> Added: llvm/trunk/test/Transforms/LoopRotate/multiple-exits.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopRotate/multiple-exits.ll?rev=162912&view=auto
> ==============================================================================
> --- llvm/trunk/test/Transforms/LoopRotate/multiple-exits.ll (added)
> +++ llvm/trunk/test/Transforms/LoopRotate/multiple-exits.ll Thu Aug 30 10:39:42 2012
> @@ -0,0 +1,200 @@
> +; RUN: opt -S -loop-rotate < %s -verify-loop-info -verify-dom-info | FileCheck %s
> +
> +target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
> +target triple = "x86_64-apple-macosx10.8.0"
> +
> +; PR7447
> +define i32 @test1([100 x i32]* nocapture %a) nounwind readonly {
> +entry:
> + br label %for.cond
> +
> +for.cond: ; preds = %for.cond1, %entry
> + %sum.0 = phi i32 [ 0, %entry ], [ %sum.1, %for.cond1 ]
> + %i.0 = phi i1 [ true, %entry ], [ false, %for.cond1 ]
> + br i1 %i.0, label %for.cond1, label %return
> +
> +for.cond1: ; preds = %for.cond, %land.rhs
> + %sum.1 = phi i32 [ %add, %land.rhs ], [ %sum.0, %for.cond ]
> + %i.1 = phi i32 [ %inc, %land.rhs ], [ 0, %for.cond ]
> + %cmp2 = icmp ult i32 %i.1, 100
> + br i1 %cmp2, label %land.rhs, label %for.cond
> +
> +land.rhs: ; preds = %for.cond1
> + %conv = zext i32 %i.1 to i64
> + %arrayidx = getelementptr inbounds [100 x i32]* %a, i64 0, i64 %conv
> + %0 = load i32* %arrayidx, align 4
> + %add = add i32 %0, %sum.1
> + %cmp4 = icmp ugt i32 %add, 1000
> + %inc = add i32 %i.1, 1
> + br i1 %cmp4, label %return, label %for.cond1
> +
> +return: ; preds = %for.cond, %land.rhs
> + %retval.0 = phi i32 [ 1000, %land.rhs ], [ %sum.0, %for.cond ]
> + ret i32 %retval.0
> +
> +; CHECK: @test1
> +; CHECK: for.cond1.preheader:
> +; CHECK: %sum.04 = phi i32 [ 0, %entry ], [ %sum.1.lcssa, %for.cond.loopexit ]
> +; CHECK: br label %for.cond1
> +
> +; CHECK: for.cond1:
> +; CHECK: %sum.1 = phi i32 [ %add, %land.rhs ], [ %sum.04, %for.cond1.preheader ]
> +; CHECK: %i.1 = phi i32 [ %inc, %land.rhs ], [ 0, %for.cond1.preheader ]
> +; CHECK: %cmp2 = icmp ult i32 %i.1, 100
> +; CHECK: br i1 %cmp2, label %land.rhs, label %for.cond.loopexit
> +}
> +
> +define void @test2(i32 %x) nounwind {
> +entry:
> + br label %for.cond
> +
> +for.cond: ; preds = %if.end, %entry
> + %i.0 = phi i32 [ 0, %entry ], [ %inc, %if.end ]
> + %cmp = icmp eq i32 %i.0, %x
> + br i1 %cmp, label %return.loopexit, label %for.body
> +
> +for.body: ; preds = %for.cond
> + %call = tail call i32 @foo(i32 %i.0) nounwind
> + %tobool = icmp eq i32 %call, 0
> + br i1 %tobool, label %if.end, label %a
> +
> +if.end: ; preds = %for.body
> + %call1 = tail call i32 @foo(i32 42) nounwind
> + %inc = add i32 %i.0, 1
> + br label %for.cond
> +
> +a: ; preds = %for.body
> + %call2 = tail call i32 @bar(i32 1) nounwind
> + br label %return
> +
> +return.loopexit: ; preds = %for.cond
> + br label %return
> +
> +return: ; preds = %return.loopexit, %a
> + ret void
> +
> +; CHECK: @test2
> +; CHECK: if.end:
> +; CHECK: %inc = add i32 %i.02, 1
> +; CHECK: %cmp = icmp eq i32 %inc, %x
> +; CHECK: br i1 %cmp, label %for.cond.return.loopexit_crit_edge, label %for.body
> +}
> +
> +declare i32 @foo(i32)
> +
> +declare i32 @bar(i32)
> +
> + at _ZTIi = external constant i8*
> +
> +; Verify dominators.
> +define void @test3(i32 %x) {
> +entry:
> + %cmp2 = icmp eq i32 0, %x
> + br i1 %cmp2, label %try.cont.loopexit, label %for.body.lr.ph
> +
> +for.body.lr.ph: ; preds = %entry
> + br label %for.body
> +
> +for.body: ; preds = %for.body.lr.ph, %for.inc
> + %i.03 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.inc ]
> + invoke void @_Z3fooi(i32 %i.03)
> + to label %for.inc unwind label %lpad
> +
> +for.inc: ; preds = %for.body
> + %inc = add i32 %i.03, 1
> + %cmp = icmp eq i32 %inc, %x
> + br i1 %cmp, label %for.cond.try.cont.loopexit_crit_edge, label %for.body
> +
> +lpad: ; preds = %for.body
> + %0 = landingpad { i8*, i32 } personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*)
> + catch i8* bitcast (i8** @_ZTIi to i8*)
> + %1 = extractvalue { i8*, i32 } %0, 0
> + %2 = extractvalue { i8*, i32 } %0, 1
> + %3 = tail call i32 @llvm.eh.typeid.for(i8* bitcast (i8** @_ZTIi to i8*)) nounwind
> + %matches = icmp eq i32 %2, %3
> + br i1 %matches, label %catch, label %eh.resume
> +
> +catch: ; preds = %lpad
> + %4 = tail call i8* @__cxa_begin_catch(i8* %1) nounwind
> + br i1 true, label %invoke.cont2.loopexit, label %for.body.i.lr.ph
> +
> +for.body.i.lr.ph: ; preds = %catch
> + br label %for.body.i
> +
> +for.body.i: ; preds = %for.body.i.lr.ph, %for.inc.i
> + %i.0.i1 = phi i32 [ 0, %for.body.i.lr.ph ], [ %inc.i, %for.inc.i ]
> + invoke void @_Z3fooi(i32 %i.0.i1)
> + to label %for.inc.i unwind label %lpad.i
> +
> +for.inc.i: ; preds = %for.body.i
> + %inc.i = add i32 %i.0.i1, 1
> + %cmp.i = icmp eq i32 %inc.i, 0
> + br i1 %cmp.i, label %for.cond.i.invoke.cont2.loopexit_crit_edge, label %for.body.i
> +
> +lpad.i: ; preds = %for.body.i
> + %5 = landingpad { i8*, i32 } personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*)
> + catch i8* bitcast (i8** @_ZTIi to i8*)
> + %6 = extractvalue { i8*, i32 } %5, 0
> + %7 = extractvalue { i8*, i32 } %5, 1
> + %matches.i = icmp eq i32 %7, %3
> + br i1 %matches.i, label %catch.i, label %lpad1.body
> +
> +catch.i: ; preds = %lpad.i
> + %8 = tail call i8* @__cxa_begin_catch(i8* %6) nounwind
> + invoke void @_Z3barj(i32 0)
> + to label %invoke.cont2.i unwind label %lpad1.i
> +
> +invoke.cont2.i: ; preds = %catch.i
> + tail call void @__cxa_end_catch() nounwind
> + br label %invoke.cont2
> +
> +lpad1.i: ; preds = %catch.i
> + %9 = landingpad { i8*, i32 } personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*)
> + cleanup
> + %10 = extractvalue { i8*, i32 } %9, 0
> + %11 = extractvalue { i8*, i32 } %9, 1
> + tail call void @__cxa_end_catch() nounwind
> + br label %lpad1.body
> +
> +for.cond.i.invoke.cont2.loopexit_crit_edge: ; preds = %for.inc.i
> + br label %invoke.cont2.loopexit
> +
> +invoke.cont2.loopexit: ; preds = %for.cond.i.invoke.cont2.loopexit_crit_edge, %catch
> + br label %invoke.cont2
> +
> +invoke.cont2: ; preds = %invoke.cont2.loopexit, %invoke.cont2.i
> + tail call void @__cxa_end_catch() nounwind
> + br label %try.cont
> +
> +for.cond.try.cont.loopexit_crit_edge: ; preds = %for.inc
> + br label %try.cont.loopexit
> +
> +try.cont.loopexit: ; preds = %for.cond.try.cont.loopexit_crit_edge, %entry
> + br label %try.cont
> +
> +try.cont: ; preds = %try.cont.loopexit, %invoke.cont2
> + ret void
> +
> +lpad1.body: ; preds = %lpad1.i, %lpad.i
> + %exn.slot.0.i = phi i8* [ %10, %lpad1.i ], [ %6, %lpad.i ]
> + %ehselector.slot.0.i = phi i32 [ %11, %lpad1.i ], [ %7, %lpad.i ]
> + tail call void @__cxa_end_catch() nounwind
> + br label %eh.resume
> +
> +eh.resume: ; preds = %lpad1.body, %lpad
> + %exn.slot.0 = phi i8* [ %exn.slot.0.i, %lpad1.body ], [ %1, %lpad ]
> + %ehselector.slot.0 = phi i32 [ %ehselector.slot.0.i, %lpad1.body ], [ %2, %lpad ]
> + %lpad.val = insertvalue { i8*, i32 } undef, i8* %exn.slot.0, 0
> + %lpad.val5 = insertvalue { i8*, i32 } %lpad.val, i32 %ehselector.slot.0, 1
> + resume { i8*, i32 } %lpad.val5
> +}
> +
> +declare void @_Z3fooi(i32)
> +
> +declare i32 @__gxx_personality_v0(...)
> +
> +declare i32 @llvm.eh.typeid.for(i8*) nounwind readnone
> +
> +declare i8* @__cxa_begin_catch(i8*)
> +
> +declare void @__cxa_end_catch()
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
More information about the llvm-commits
mailing list