[llvm] r269119 - Revert r26084: Enable loopreroll for sext of loop control only IV
Lawrence Hu via llvm-commits
llvm-commits at lists.llvm.org
Tue May 10 14:11:09 PDT 2016
Author: lawrence
Date: Tue May 10 16:11:09 2016
New Revision: 269119
URL: http://llvm.org/viewvc/llvm-project?rev=269119&view=rev
Log:
Revert r26084: Enable loopreroll for sext of loop control only IV
Removed:
llvm/trunk/test/Transforms/LoopReroll/indvar_with_ext.ll
Modified:
llvm/trunk/lib/Transforms/Scalar/LoopRerollPass.cpp
Modified: llvm/trunk/lib/Transforms/Scalar/LoopRerollPass.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopRerollPass.cpp?rev=269119&r1=269118&r2=269119&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/LoopRerollPass.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/LoopRerollPass.cpp Tue May 10 16:11:09 2016
@@ -512,11 +512,11 @@ static const SCEVConstant *getIncrmentFa
// Check if an IV is only used to control the loop. There are two cases:
// 1. It only has one use which is loop increment, and the increment is only
-// used by comparison and the PHI (could has sext with nsw in between), and the
-// comparison is only used by branch.
+// used by comparison and the PHI, and the comparison is only used by branch.
// 2. It is used by loop increment and the comparison, the loop increment is
// only used by the PHI, and the comparison is used only by the branch.
bool LoopReroll::isLoopControlIV(Loop *L, Instruction *IV) {
+
unsigned IVUses = IV->getNumUses();
if (IVUses != 2 && IVUses != 1)
return false;
@@ -551,17 +551,9 @@ bool LoopReroll::isLoopControlIV(Loop *L
if (PN != IV)
return false;
}
- // Must be a CMP or an ext (of a value with nsw) then CMP
- else {
- Instruction *UUser = dyn_cast<Instruction>(UU);
- // Skip SExt if we are extending an nsw value
- // TODO: Allow ZExt too
- if (BO->hasNoSignedWrap() && UUser && UUser->getNumUses() == 1 &&
- isa<SExtInst>(UUser))
- UUser = dyn_cast<Instruction>(*(UUser->user_begin()));
- if (!isCompareUsedByBranch(UUser))
- return false;
- }
+ // Must be a CMP
+ else if (!isCompareUsedByBranch(dyn_cast<Instruction>(UU)))
+ return false;
}
} else
return false;
@@ -1169,11 +1161,6 @@ bool LoopReroll::DAGRootTracker::validat
Instruction *UUser = dyn_cast<Instruction>(UU);
// UUser could be compare, PHI or branch
Uses[UUser].set(IL_All);
- // Skip SExt
- if (isa<SExtInst>(UUser)) {
- UUser = dyn_cast<Instruction>(*(UUser->user_begin()));
- Uses[UUser].set(IL_All);
- }
// Is UUser a compare instruction?
if (UU->hasOneUse()) {
Instruction *BI = dyn_cast<BranchInst>(*UUser->user_begin());
@@ -1515,6 +1502,8 @@ void LoopReroll::DAGRootTracker::replace
if (NeedNewIV)
ICSCEV = SE->getMulExpr(IterCount,
SE->getConstant(IterCount->getType(), Scale));
+ else
+ ICSCEV = RealIVSCEV->evaluateAtIteration(IterCount, *SE);
// Iteration count SCEV minus or plus 1
const SCEV *MinusPlus1SCEV =
@@ -1526,25 +1515,17 @@ void LoopReroll::DAGRootTracker::replace
const SCEV *ICMinusPlus1SCEV = SE->getMinusSCEV(ICSCEV, MinusPlus1SCEV);
// Iteration count minus 1
- Instruction *InsertPtr = nullptr;
+ Value *ICMinusPlus1 = nullptr;
if (isa<SCEVConstant>(ICMinusPlus1SCEV)) {
- InsertPtr = BI;
+ ICMinusPlus1 =
+ Expander.expandCodeFor(ICMinusPlus1SCEV, NewIV->getType(), BI);
} else {
BasicBlock *Preheader = L->getLoopPreheader();
if (!Preheader)
Preheader = InsertPreheaderForLoop(L, DT, LI, PreserveLCSSA);
- InsertPtr = Preheader->getTerminator();
- }
-
- if (!isa<PointerType>(NewIV->getType()) && NeedNewIV &&
- (SE->getTypeSizeInBits(NewIV->getType()) <
- SE->getTypeSizeInBits(ICMinusPlus1SCEV->getType()))) {
- IRBuilder<> Builder(BI);
- Builder.SetCurrentDebugLocation(BI->getDebugLoc());
- NewIV = Builder.CreateSExt(NewIV, ICMinusPlus1SCEV->getType());
+ ICMinusPlus1 = Expander.expandCodeFor(
+ ICMinusPlus1SCEV, NewIV->getType(), Preheader->getTerminator());
}
- Value *ICMinusPlus1 = Expander.expandCodeFor(
- ICMinusPlus1SCEV, NewIV->getType(), InsertPtr);
Value *Cond =
new ICmpInst(BI, CmpInst::ICMP_EQ, NewIV, ICMinusPlus1, "exitcond");
@@ -1724,7 +1705,6 @@ bool LoopReroll::runOnLoop(Loop *L, LPPa
const SCEV *LIBETC = SE->getBackedgeTakenCount(L);
const SCEV *IterCount = SE->getAddExpr(LIBETC, SE->getOne(LIBETC->getType()));
- DEBUG(dbgs() << "\n Before Reroll:\n" << *(L->getHeader()) << "\n");
DEBUG(dbgs() << "LRR: iteration count = " << *IterCount << "\n");
// First, we need to find the induction variable with respect to which we can
@@ -1751,7 +1731,6 @@ bool LoopReroll::runOnLoop(Loop *L, LPPa
Changed = true;
break;
}
- DEBUG(dbgs() << "\n After Reroll:\n" << *(L->getHeader()) << "\n");
// Trip count of L has changed so SE must be re-evaluated.
if (Changed)
Removed: llvm/trunk/test/Transforms/LoopReroll/indvar_with_ext.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopReroll/indvar_with_ext.ll?rev=269118&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/LoopReroll/indvar_with_ext.ll (original)
+++ llvm/trunk/test/Transforms/LoopReroll/indvar_with_ext.ll (removed)
@@ -1,186 +0,0 @@
-; RUN: opt -S -loop-reroll %s | FileCheck %s
-target triple = "aarch64--linux-gnu"
-
-define void @test(i32 %n, float* %arrayidx200, float* %arrayidx164, float* %arrayidx172) {
-entry:
- %rem.i = srem i32 %n, 4
- %t22 = load float, float* %arrayidx172, align 4
- %cmp.9 = icmp eq i32 %n, 0
- %t7 = sext i32 %n to i64
- br i1 %cmp.9, label %while.end, label %while.body.preheader
-
-while.body.preheader:
- br label %while.body
-
-while.body:
-;CHECK-LABEL: while.body:
-;CHECK-NEXT: %indvars.iv.i423 = phi i64 [ %indvars.iv.next.i424, %while.body ], [ 0, %while.body.preheader ]
-;CHECK-NEXT: [[T1:%[0-9]+]] = trunc i64 %indvars.iv.i423 to i32
-;CHECK-NEXT: %arrayidx62.i = getelementptr inbounds float, float* %arrayidx200, i64 %indvars.iv.i423
-;CHECK-NEXT: %t1 = load float, float* %arrayidx62.i, align 4
-;CHECK-NEXT: %arrayidx64.i = getelementptr inbounds float, float* %arrayidx164, i64 %indvars.iv.i423
-;CHECK-NEXT: %t2 = load float, float* %arrayidx64.i, align 4
-;CHECK-NEXT: %mul65.i = fmul fast float %t2, %t22
-;CHECK-NEXT: %add66.i = fadd fast float %mul65.i, %t1
-;CHECK-NEXT: store float %add66.i, float* %arrayidx62.i, align 4
-;CHECK-NEXT: %indvars.iv.next.i424 = add i64 %indvars.iv.i423, 1
-;CHECK-NEXT: [[T2:%[0-9]+]] = sext i32 [[T1]] to i64
-;CHECK-NEXT: %exitcond = icmp eq i64 [[T2]], %{{[0-9]+}}
-;CHECK-NEXT: br i1 %exitcond, label %while.end.loopexit, label %while.body
-
- %indvars.iv.i423 = phi i64 [ %indvars.iv.next.i424, %while.body ], [ 0, %while.body.preheader ]
- %i.22.i = phi i32 [ %add103.i, %while.body ], [ %rem.i, %while.body.preheader ]
- %arrayidx62.i = getelementptr inbounds float, float* %arrayidx200, i64 %indvars.iv.i423
- %t1 = load float, float* %arrayidx62.i, align 4
- %arrayidx64.i = getelementptr inbounds float, float* %arrayidx164, i64 %indvars.iv.i423
- %t2 = load float, float* %arrayidx64.i, align 4
- %mul65.i = fmul fast float %t2, %t22
- %add66.i = fadd fast float %mul65.i, %t1
- store float %add66.i, float* %arrayidx62.i, align 4
- %t3 = add nsw i64 %indvars.iv.i423, 1
- %arrayidx71.i = getelementptr inbounds float, float* %arrayidx200, i64 %t3
- %t4 = load float, float* %arrayidx71.i, align 4
- %arrayidx74.i = getelementptr inbounds float, float* %arrayidx164, i64 %t3
- %t5 = load float, float* %arrayidx74.i, align 4
- %mul75.i = fmul fast float %t5, %t22
- %add76.i = fadd fast float %mul75.i, %t4
- store float %add76.i, float* %arrayidx71.i, align 4
- %add103.i = add nsw i32 %i.22.i, 2
- %t6 = sext i32 %add103.i to i64
- %cmp58.i = icmp slt i64 %t6, %t7
- %indvars.iv.next.i424 = add i64 %indvars.iv.i423, 2
- br i1 %cmp58.i, label %while.body, label %while.end.loopexit
-
-while.end.loopexit:
- br label %while.end
-
-while.end:
- ret void
-}
-
-; Function Attrs: noinline norecurse nounwind
-define i32 @test2(i64 %n, i32* nocapture %x, i32* nocapture readonly %y) {
-entry:
- %cmp18 = icmp sgt i64 %n, 0
- br i1 %cmp18, label %for.body.preheader, label %for.end
-
-for.body.preheader: ; preds = %entry
- br label %for.body
-
-for.body: ; preds = %for.body.preheader, %for.body
-
-;CHECK: for.body:
-;CHECK-NEXT: %indvar = phi i64 [ %indvar.next, %for.body ], [ 0, %for.body.preheader ]
-;CHECK-NEXT: %arrayidx = getelementptr inbounds i32, i32* %y, i64 %indvar
-;CHECK-NEXT: [[T1:%[0-9]+]] = load i32, i32* %arrayidx, align 4
-;CHECK-NEXT: %arrayidx3 = getelementptr inbounds i32, i32* %x, i64 %indvar
-;CHECK-NEXT: store i32 [[T1]], i32* %arrayidx3, align 4
-;CHECK-NEXT: %indvar.next = add i64 %indvar, 1
-;CHECK-NEXT: %exitcond = icmp eq i64 %indvar, %{{[0-9]+}}
-;CHECK-NEXT: br i1 %exitcond, label %for.end.loopexit, label %for.body
-
- %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
- %arrayidx = getelementptr inbounds i32, i32* %y, i64 %indvars.iv
- %0 = load i32, i32* %arrayidx, align 4
- %arrayidx3 = getelementptr inbounds i32, i32* %x, i64 %indvars.iv
- store i32 %0, i32* %arrayidx3, align 4
- %1 = or i64 %indvars.iv, 1
- %arrayidx5 = getelementptr inbounds i32, i32* %y, i64 %1
- %2 = load i32, i32* %arrayidx5, align 4
- %arrayidx8 = getelementptr inbounds i32, i32* %x, i64 %1
- store i32 %2, i32* %arrayidx8, align 4
- %indvars.iv.next = add nuw nsw i64 %indvars.iv, 2
- %cmp = icmp slt i64 %indvars.iv.next, %n
- br i1 %cmp, label %for.body, label %for.end.loopexit
-
-for.end.loopexit: ; preds = %for.body
- br label %for.end
-
-for.end: ; preds = %for.end.loopexit, %entry
- ret i32 0
-}
-
-; Function Attrs: noinline norecurse nounwind
-define i32 @test3(i32 %n, i32* nocapture %x, i32* nocapture readonly %y) {
-entry:
- %cmp21 = icmp sgt i32 %n, 0
- br i1 %cmp21, label %for.body.preheader, label %for.end
-
-for.body.preheader: ; preds = %entry
- br label %for.body
-
-for.body: ; preds = %for.body.preheader, %for.body
-
-;CHECK: for.body:
-;CHECK: %add12 = add i8 %i.022, 2
-;CHECK-NEXT: %conv = sext i8 %add12 to i32
-;CHECK-NEXT: %cmp = icmp slt i32 %conv, %n
-;CHECK-NEXT: br i1 %cmp, label %for.body, label %for.end.loopexit
-
- %conv23 = phi i32 [ %conv, %for.body ], [ 0, %for.body.preheader ]
- %i.022 = phi i8 [ %add12, %for.body ], [ 0, %for.body.preheader ]
- %idxprom = sext i8 %i.022 to i64
- %arrayidx = getelementptr inbounds i32, i32* %y, i64 %idxprom
- %0 = load i32, i32* %arrayidx, align 4
- %arrayidx3 = getelementptr inbounds i32, i32* %x, i64 %idxprom
- store i32 %0, i32* %arrayidx3, align 4
- %add = or i32 %conv23, 1
- %idxprom5 = sext i32 %add to i64
- %arrayidx6 = getelementptr inbounds i32, i32* %y, i64 %idxprom5
- %1 = load i32, i32* %arrayidx6, align 4
- %arrayidx10 = getelementptr inbounds i32, i32* %x, i64 %idxprom5
- store i32 %1, i32* %arrayidx10, align 4
- %add12 = add i8 %i.022, 2
- %conv = sext i8 %add12 to i32
- %cmp = icmp slt i32 %conv, %n
- br i1 %cmp, label %for.body, label %for.end.loopexit
-
-for.end.loopexit: ; preds = %for.body
- br label %for.end
-
-for.end: ; preds = %for.end.loopexit, %entry
- ret i32 0
-}
-
-; Function Attrs: noinline norecurse nounwind
-define i32 @test4(i64 %n, i32* nocapture %x, i32* nocapture readonly %y) {
-entry:
- %cmp18 = icmp eq i64 %n, 0
- br i1 %cmp18, label %for.end, label %for.body.preheader
-
-for.body.preheader: ; preds = %entry
- br label %for.body
-
-for.body: ; preds = %for.body.preheader, %for.body
-
-;CHECK: for.body:
-;CHECK-NEXT: %indvar = phi i64 [ %indvar.next, %for.body ], [ 0, %for.body.preheader ]
-;CHECK-NEXT: %arrayidx = getelementptr inbounds i32, i32* %y, i64 %indvar
-;CHECK-NEXT: [[T1:%[0-9]+]] = load i32, i32* %arrayidx, align 4
-;CHECK-NEXT: %arrayidx3 = getelementptr inbounds i32, i32* %x, i64 %indvar
-;CHECK-NEXT: store i32 [[T1]], i32* %arrayidx3, align 4
-;CHECK-NEXT: %indvar.next = add i64 %indvar, 1
-;CHECK-NEXT: %exitcond = icmp eq i64 %indvar, %{{[0-9]+}}
-;CHECK-NEXT: br i1 %exitcond, label %for.end.loopexit, label %for.body
-
- %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
- %arrayidx = getelementptr inbounds i32, i32* %y, i64 %indvars.iv
- %0 = load i32, i32* %arrayidx, align 4
- %arrayidx3 = getelementptr inbounds i32, i32* %x, i64 %indvars.iv
- store i32 %0, i32* %arrayidx3, align 4
- %1 = or i64 %indvars.iv, 1
- %arrayidx5 = getelementptr inbounds i32, i32* %y, i64 %1
- %2 = load i32, i32* %arrayidx5, align 4
- %arrayidx8 = getelementptr inbounds i32, i32* %x, i64 %1
- store i32 %2, i32* %arrayidx8, align 4
- %indvars.iv.next = add nuw nsw i64 %indvars.iv, 2
- %cmp = icmp ult i64 %indvars.iv.next, %n
- br i1 %cmp, label %for.body, label %for.end.loopexit
-
-for.end.loopexit: ; preds = %for.body
- br label %for.end
-
-for.end: ; preds = %for.end.loopexit, %entry
- ret i32 0
-}
-
More information about the llvm-commits
mailing list