[llvm-branch-commits] [llvm] 8b330f1 - [SCEV] Add missing type check into getRangeForAffineNoSelfWrappingAR
Max Kazantsev via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Mon Dec 14 23:55:31 PST 2020
Author: Max Kazantsev
Date: 2020-12-15T14:50:32+07:00
New Revision: 8b330f1f6919a2ac85eeda753ad8d1090468e23f
URL: https://github.com/llvm/llvm-project/commit/8b330f1f6919a2ac85eeda753ad8d1090468e23f
DIFF: https://github.com/llvm/llvm-project/commit/8b330f1f6919a2ac85eeda753ad8d1090468e23f.diff
LOG: [SCEV] Add missing type check into getRangeForAffineNoSelfWrappingAR
We make type widening without checking if it's needed. Bail if the max
iteration count is wider than AR's type.
Added:
Modified:
llvm/lib/Analysis/ScalarEvolution.cpp
llvm/test/Transforms/IndVarSimplify/2020-12-15-trunc-bug-expensive-range-inference.ll
Removed:
################################################################################
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index 3005a44b44ef..071b569d3f17 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -6006,6 +6006,9 @@ ConstantRange ScalarEvolution::getRangeForAffineNoSelfWrappingAR(
// iteration count estimate, and we might infer nw from some exit for which we
// do not know max exit count (or any other side reasoning).
// TODO: Turn into assert at some point.
+ if (getTypeSizeInBits(MaxBECount->getType()) >
+ getTypeSizeInBits(AddRec->getType()))
+ return ConstantRange::getFull(BitWidth);
MaxBECount = getNoopOrZeroExtend(MaxBECount, AddRec->getType());
const SCEV *RangeWidth = getMinusOne(AddRec->getType());
const SCEV *StepAbs = getUMinExpr(Step, getNegativeSCEV(Step));
diff --git a/llvm/test/Transforms/IndVarSimplify/2020-12-15-trunc-bug-expensive-range-inference.ll b/llvm/test/Transforms/IndVarSimplify/2020-12-15-trunc-bug-expensive-range-inference.ll
index dd105f40544e..a1ad4d07fab2 100644
--- a/llvm/test/Transforms/IndVarSimplify/2020-12-15-trunc-bug-expensive-range-inference.ll
+++ b/llvm/test/Transforms/IndVarSimplify/2020-12-15-trunc-bug-expensive-range-inference.ll
@@ -1,7 +1,5 @@
; RUN: opt < %s -indvars -S -scalar-evolution-use-expensive-range-sharpening | FileCheck %s
; RUN: opt < %s -passes=indvars -S -scalar-evolution-use-expensive-range-sharpening | FileCheck %s
-; REQUIRES: asserts
-; XFAIL: *
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128-ni:1-p2:32:8:8:32-ni:2"
target triple = "x86_64-unknown-linux-gnu"
More information about the llvm-branch-commits
mailing list