From c9cf35055f7b0fd5814836686b4a550dc69bd11c Mon Sep 17 00:00:00 2001 From: Dale Johannesen Date: Wed, 29 Apr 2009 16:38:47 +0000 Subject: [PATCH] Fix recent regression in gcc.dg/pr26719.c (6835035). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@70386 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Analysis/ScalarEvolution.cpp | 7 ++- .../2009-04-28-no-reduce-mul.ll | 43 +++++++++++++++++++ 2 files changed, 48 insertions(+), 2 deletions(-) create mode 100644 test/Transforms/LoopStrengthReduce/2009-04-28-no-reduce-mul.ll diff --git a/lib/Analysis/ScalarEvolution.cpp b/lib/Analysis/ScalarEvolution.cpp index 7f632179257..42d2fdd3629 100644 --- a/lib/Analysis/ScalarEvolution.cpp +++ b/lib/Analysis/ScalarEvolution.cpp @@ -812,11 +812,14 @@ SCEVHandle ScalarEvolution::getSignExtendExpr(const SCEVHandle &Op, SCEVHandle Step = AR->getStepRecurrence(*this); // Check whether the backedge-taken count can be losslessly casted to - // the addrec's type. The count is always unsigned. + // the addrec's type. The count needs to be the same whether sign + // extended or zero extended. SCEVHandle CastedBECount = getTruncateOrZeroExtend(BECount, Start->getType()); if (BECount == - getTruncateOrZeroExtend(CastedBECount, BECount->getType())) { + getTruncateOrZeroExtend(CastedBECount, BECount->getType()) && + BECount == + getTruncateOrSignExtend(CastedBECount, BECount->getType())) { const Type *WideTy = IntegerType::get(getTypeSizeInBits(Start->getType()) * 2); SCEVHandle SMul = diff --git a/test/Transforms/LoopStrengthReduce/2009-04-28-no-reduce-mul.ll b/test/Transforms/LoopStrengthReduce/2009-04-28-no-reduce-mul.ll new file mode 100644 index 00000000000..153a1811e63 --- /dev/null +++ b/test/Transforms/LoopStrengthReduce/2009-04-28-no-reduce-mul.ll @@ -0,0 +1,43 @@ +; RUN: llvm-as < %s | opt -loop-reduce | llvm-dis | grep mul | count 3 +; The multiply in bb2 must not be reduced to an add, as the sext causes the +; %1 argument to become negative after a while. +; ModuleID = '' +target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128" +target triple = "i386-apple-darwin9.6" +@table = common global [32 x [256 x i32]] zeroinitializer, align 32 ; <[32 x [256 x i32]]*> [#uses=2] + +define i32 @main() nounwind { +bb4.thread: + br label %bb2 + +bb2: ; preds = %bb4, %bb2, %bb4.thread + %i.0.reg2mem.0.ph = phi i32 [ 0, %bb4.thread ], [ %i.0.reg2mem.0.ph, %bb2 ], [ %indvar.next9, %bb4 ] ; [#uses=4] + %j.0.reg2mem.0 = phi i32 [ 0, %bb4.thread ], [ %indvar.next, %bb2 ], [ 0, %bb4 ] ; [#uses=3] + %0 = trunc i32 %j.0.reg2mem.0 to i8 ; [#uses=1] + %1 = sext i8 %0 to i32 ; [#uses=1] + %2 = mul i32 %1, %i.0.reg2mem.0.ph ; [#uses=1] + %3 = getelementptr [32 x [256 x i32]]* @table, i32 0, i32 %i.0.reg2mem.0.ph, i32 %j.0.reg2mem.0 ; [#uses=1] + store i32 %2, i32* %3, align 4 + %indvar.next = add i32 %j.0.reg2mem.0, 1 ; [#uses=2] + %exitcond = icmp eq i32 %indvar.next, 256 ; [#uses=1] + br i1 %exitcond, label %bb4, label %bb2 + +bb4: ; preds = %bb2 + %indvar.next9 = add i32 %i.0.reg2mem.0.ph, 1 ; [#uses=2] + %exitcond10 = icmp eq i32 %indvar.next9, 32 ; [#uses=1] + br i1 %exitcond10, label %bb5, label %bb2 + +bb5: ; preds = %bb4 + %4 = load i32* getelementptr ([32 x [256 x i32]]* @table, i32 0, i32 9, i32 132), align 16 ; [#uses=1] + %5 = icmp eq i32 %4, -1116 ; [#uses=1] + br i1 %5, label %bb7, label %bb6 + +bb6: ; preds = %bb5 + tail call void @abort() noreturn nounwind + unreachable + +bb7: ; preds = %bb5 + ret i32 0 +} + +declare void @abort() noreturn nounwind -- 2.34.1