forked from OSchip/llvm-project
When turning (ashr(shl(x, n), n)) into sext(trunc(x)), the width of the
type to truncate to should be the number of bits of the value that are preserved, not the number that are clobbered with sign-extension. This fixes regressions in ldecod. llvm-svn: 69704
This commit is contained in:
parent
a8e8bc24b3
commit
e14efcc9f4
|
@ -2015,7 +2015,7 @@ SCEVHandle ScalarEvolutionsImpl::createSCEV(Value *V) {
|
|||
if (Instruction *L = dyn_cast<Instruction>(U->getOperand(0)))
|
||||
if (L->getOpcode() == Instruction::Shl &&
|
||||
L->getOperand(1) == U->getOperand(1)) {
|
||||
uint64_t Amt = CI->getZExtValue();
|
||||
uint64_t Amt = getTypeSizeInBits(U->getType()) - CI->getZExtValue();
|
||||
return
|
||||
SE.getSignExtendExpr(SE.getTruncateExpr(getSCEV(L->getOperand(0)),
|
||||
IntegerType::get(Amt)),
|
||||
|
|
|
@ -0,0 +1,29 @@
|
|||
; RUN: llvm-as < %s | opt -analyze -scalar-evolution -disable-output \
|
||||
; RUN: | grep {signextend \{0,+,199\}<bb> to i64} | count 2
|
||||
|
||||
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"
|
||||
|
||||
define i64 @foo(i64* nocapture %x, i64 %n) nounwind {
|
||||
entry:
|
||||
%t0 = icmp sgt i64 %n, 0 ; <i1> [#uses=1]
|
||||
br i1 %t0, label %bb, label %return
|
||||
|
||||
bb: ; preds = %bb, %entry
|
||||
%i.01 = phi i64 [ 0, %entry ], [ %indvar.next, %bb ] ; <i32> [#uses=2]
|
||||
%t1 = shl i64 %i.01, 7 ; <i32> [#uses=1]
|
||||
%t2 = ashr i64 %t1, 7 ; <i32> [#uses=1]
|
||||
%s1 = shl i64 %i.01, 5 ; <i32> [#uses=1]
|
||||
%s2 = ashr i64 %s1, 5 ; <i32> [#uses=1]
|
||||
%t3 = getelementptr i64* %x, i64 %i.01 ; <i64*> [#uses=1]
|
||||
store i64 0, i64* %t3, align 1
|
||||
%indvar.next = add i64 %i.01, 199 ; <i32> [#uses=2]
|
||||
%exitcond = icmp eq i64 %indvar.next, %n ; <i1> [#uses=1]
|
||||
br i1 %exitcond, label %return, label %bb
|
||||
|
||||
return: ; preds = %bb, %entry
|
||||
%p = phi i64 [ 0, %entry ], [ %t2, %bb ]
|
||||
%q = phi i64 [ 0, %entry ], [ %s2, %bb ]
|
||||
%v = xor i64 %p, %q
|
||||
ret i64 %v
|
||||
}
|
Loading…
Reference in New Issue