[llvm] r320789 - [SCEV] Fix the movement of insertion point in expander. PR35406.
Serguei Katkov via llvm-commits
llvm-commits at lists.llvm.org
Thu Dec 14 21:24:42 PST 2017
Author: skatkov
Date: Thu Dec 14 21:24:42 2017
New Revision: 320789
URL: http://llvm.org/viewvc/llvm-project?rev=320789&view=rev
Log:
[SCEV] Fix the movement of insertion point in expander. PR35406.
We cannot move the insertion point to header if SCEV contains div/rem
operations due to they may go over check for zero denominator.
Reviewers: sanjoy, mkazantsev, sebpop
Reviewed By: sebpop
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D41229
Added:
llvm/trunk/test/Transforms/IndVarSimplify/pr35406.ll
Modified:
llvm/trunk/lib/Analysis/ScalarEvolutionExpander.cpp
Modified: llvm/trunk/lib/Analysis/ScalarEvolutionExpander.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ScalarEvolutionExpander.cpp?rev=320789&r1=320788&r2=320789&view=diff
==============================================================================
--- llvm/trunk/lib/Analysis/ScalarEvolutionExpander.cpp (original)
+++ llvm/trunk/lib/Analysis/ScalarEvolutionExpander.cpp Thu Dec 14 21:24:42 2017
@@ -1732,10 +1732,28 @@ Value *SCEVExpander::expand(const SCEV *
InsertPt = &*L->getHeader()->getFirstInsertionPt();
}
} else {
+ // We can move insertion point only if there is no div or rem operations
+ // otherwise we are risky to move it over the check for zero denominator.
+ auto SafeToHoist = [](const SCEV *S) {
+ return !SCEVExprContains(S, [](const SCEV *S) {
+ if (const auto *D = dyn_cast<SCEVUDivExpr>(S)) {
+ if (const auto *SC = dyn_cast<SCEVConstant>(D->getRHS()))
+ // Division by non-zero constants can be hoisted.
+ return SC->getValue()->isZero();
+ // All other divisions should not be moved as they may be
+ // divisions by zero and should be kept within the
+ // conditions of the surrounding loops that guard their
+ // execution (see PR35406).
+ return true;
+ }
+ return false;
+ });
+ };
// If the SCEV is computable at this level, insert it into the header
// after the PHIs (and after any other instructions that we've inserted
// there) so that it is guaranteed to dominate any user inside the loop.
- if (L && SE.hasComputableLoopEvolution(S, L) && !PostIncLoops.count(L))
+ if (L && SE.hasComputableLoopEvolution(S, L) && !PostIncLoops.count(L) &&
+ SafeToHoist(S))
InsertPt = &*L->getHeader()->getFirstInsertionPt();
while (InsertPt->getIterator() != Builder.GetInsertPoint() &&
(isInsertedInstruction(InsertPt) ||
Added: llvm/trunk/test/Transforms/IndVarSimplify/pr35406.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/IndVarSimplify/pr35406.ll?rev=320789&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/IndVarSimplify/pr35406.ll (added)
+++ llvm/trunk/test/Transforms/IndVarSimplify/pr35406.ll Thu Dec 14 21:24:42 2017
@@ -0,0 +1,88 @@
+; RUN: opt -S -indvars %s | FileCheck %s
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128-ni:1"
+target triple = "x86_64-unknown-linux-gnu"
+
+define i32 @testDiv(i8* %p, i64* %p1) {
+; CHECK-LABEL: @testDiv
+entry:
+ br label %loop1
+
+loop1:
+ %local_0_ = phi i32 [ 8, %entry ], [ %9, %loop2.exit ]
+ %local_2_ = phi i32 [ 63864, %entry ], [ %local_2_43, %loop2.exit ]
+ %local_3_ = phi i32 [ 51, %entry ], [ %local_3_44, %loop2.exit ]
+; CHECK-NOT: udiv
+ %0 = udiv i32 14, %local_0_
+ %1 = icmp ugt i32 %local_0_, 14
+ br i1 %1, label %exit, label %general_case24
+
+; CHECK-LABEL: general_case24
+general_case24:
+ %2 = udiv i32 60392, %0
+ br i1 false, label %loop2, label %loop2.exit
+
+loop2:
+ %local_1_56 = phi i32 [ %2, %general_case24 ], [ %3, %loop2 ]
+ %local_2_57 = phi i32 [ 1, %general_case24 ], [ %7, %loop2 ]
+ %3 = add i32 %local_1_56, -1
+ %4 = load atomic i64, i64* %p1 unordered, align 8
+ %5 = sext i32 %3 to i64
+ %6 = sub i64 %4, %5
+ store atomic i64 %6, i64* %p1 unordered, align 8
+ %7 = add nuw nsw i32 %local_2_57, 1
+ %8 = icmp ugt i32 %local_2_57, 7
+ br i1 %8, label %loop2.exit, label %loop2
+
+loop2.exit:
+ %local_2_43 = phi i32 [ %local_2_, %general_case24 ], [ 9, %loop2 ]
+ %local_3_44 = phi i32 [ %local_3_, %general_case24 ], [ %local_1_56, %loop2 ]
+ %9 = add nuw nsw i32 %local_0_, 1
+ %10 = icmp ugt i32 %local_0_, 129
+ br i1 %10, label %exit, label %loop1
+
+exit:
+ ret i32 0
+}
+
+define i32 @testRem(i8* %p, i64* %p1) {
+; CHECK-LABEL: @testRem
+entry:
+ br label %loop1
+
+loop1:
+ %local_0_ = phi i32 [ 8, %entry ], [ %9, %loop2.exit ]
+ %local_2_ = phi i32 [ 63864, %entry ], [ %local_2_43, %loop2.exit ]
+ %local_3_ = phi i32 [ 51, %entry ], [ %local_3_44, %loop2.exit ]
+; CHECK: udiv
+; CHECK-NOT: udiv
+ %0 = udiv i32 14, %local_0_
+ %1 = icmp ugt i32 %local_0_, 14
+ br i1 %1, label %exit, label %general_case24
+
+; CHECK-LABEL: general_case24
+general_case24:
+ %2 = urem i32 60392, %0
+ br i1 false, label %loop2, label %loop2.exit
+
+loop2:
+ %local_1_56 = phi i32 [ %2, %general_case24 ], [ %3, %loop2 ]
+ %local_2_57 = phi i32 [ 1, %general_case24 ], [ %7, %loop2 ]
+ %3 = add i32 %local_1_56, -1
+ %4 = load atomic i64, i64* %p1 unordered, align 8
+ %5 = sext i32 %3 to i64
+ %6 = sub i64 %4, %5
+ store atomic i64 %6, i64* %p1 unordered, align 8
+ %7 = add nuw nsw i32 %local_2_57, 1
+ %8 = icmp ugt i32 %local_2_57, 7
+ br i1 %8, label %loop2.exit, label %loop2
+
+loop2.exit:
+ %local_2_43 = phi i32 [ %local_2_, %general_case24 ], [ 9, %loop2 ]
+ %local_3_44 = phi i32 [ %local_3_, %general_case24 ], [ %local_1_56, %loop2 ]
+ %9 = add nuw nsw i32 %local_0_, 1
+ %10 = icmp ugt i32 %local_0_, 129
+ br i1 %10, label %exit, label %loop1
+
+exit:
+ ret i32 0
+}
More information about the llvm-commits
mailing list