// Construct the extended SCEV: (Ext ix (Trunc iy (Expr) to ix) to iy)
// for each of StartVal and Accum
- auto GetExtendedExpr = [&](const SCEV *Expr) -> const SCEV * {
+ auto getExtendedExpr = [&](const SCEV *Expr,
+ bool CreateSignExtend) -> const SCEV * {
assert(isLoopInvariant(Expr, L) && "Expr is expected to be invariant");
const SCEV *TruncatedExpr = getTruncateExpr(Expr, TruncTy);
const SCEV *ExtendedExpr =
- Signed ? getSignExtendExpr(TruncatedExpr, Expr->getType())
- : getZeroExtendExpr(TruncatedExpr, Expr->getType());
+ CreateSignExtend ? getSignExtendExpr(TruncatedExpr, Expr->getType())
+ : getZeroExtendExpr(TruncatedExpr, Expr->getType());
return ExtendedExpr;
};
// Given:
// ExtendedExpr = (Ext ix (Trunc iy (Expr) to ix) to iy
- // = GetExtendedExpr(Expr)
+ // = getExtendedExpr(Expr)
// Determine whether the predicate P: Expr == ExtendedExpr
// is known to be false at compile time
auto PredIsKnownFalse = [&](const SCEV *Expr,
isKnownPredicate(ICmpInst::ICMP_NE, Expr, ExtendedExpr);
};
- const SCEV *StartExtended = GetExtendedExpr(StartVal);
+ const SCEV *StartExtended = getExtendedExpr(StartVal, Signed);
if (PredIsKnownFalse(StartVal, StartExtended)) {
DEBUG(dbgs() << "P2 is compile-time false\n";);
return None;
}
- const SCEV *AccumExtended = GetExtendedExpr(Accum);
+ // The Step is always Signed (because the overflow checks are either
+ // NSSW or NUSW)
+ const SCEV *AccumExtended = getExtendedExpr(Accum, /*CreateSignExtend=*/true);
if (PredIsKnownFalse(Accum, AccumExtended)) {
DEBUG(dbgs() << "P3 is compile-time false\n";);
return None;
; Same as above, but for checking the SCEV "zext(trunc(%p.09)) + %step".
; Here we expect the following two predicates to be added for runtime checking:
; 1) {0,+,(trunc i32 %step to i8)}<%for.body> Added Flags: <nusw>
-; 2) Equal predicate: %step == (zext i8 (trunc i32 %step to i8) to i32)
+; 2) Equal predicate: %step == (sext i8 (trunc i32 %step to i8) to i32)
;
; int a[N];
; void doit2(int n, int step) {
; CHECK-NOT: %mul = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 {{.*}}, i8 {{.*}})
; CHECK: %[[TEST:[0-9]+]] = or i1 {{.*}}, %mul.overflow
; CHECK: %[[NTEST:[0-9]+]] = or i1 false, %[[TEST]]
-; CHECK: %ident.check = icmp ne i32 {{.*}}, %{{.*}}
+; CHECK: %[[EXT:[0-9]+]] = sext i8 {{.*}} to i32
+; CHECK: %ident.check = icmp ne i32 {{.*}}, %[[EXT]]
; CHECK: %{{.*}} = or i1 %[[NTEST]], %ident.check
; CHECK-NOT: %mul = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 {{.*}}, i8 {{.*}})
; CHECK: vector.body: