1 ; RUN: opt -S -loop-reduce < %s | FileCheck %s
2 ; Scaling factor in addressing mode are costly.
3 ; Make loop-reduce prefer unscaled accesses.
4 ; <rdar://problem/13806271>
5 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-n32:64-S128"
6 target triple = "arm64-apple-ios7.0.0"
8 ; Function Attrs: nounwind ssp
9 define void @mulDouble(double* nocapture %a, double* nocapture %b, double* nocapture %c) {
14 for.body: ; preds = %for.body, %entry
15 ; CHECK: [[IV:%[^ ]+]] = phi i64 [ [[IVNEXT:%[^,]+]], %for.body ], [ 0, %entry ]
16 ; Only one induction variable should have been generated.
18 %indvars.iv = phi i64 [ 1, %entry ], [ %indvars.iv.next, %for.body ]
19 %tmp = add nsw i64 %indvars.iv, -1
20 %arrayidx = getelementptr inbounds double, double* %b, i64 %tmp
21 %tmp1 = load double, double* %arrayidx, align 8
22 ; The induction variable should carry the scaling factor: 1 * 8 = 8.
23 ; CHECK: [[IVNEXT]] = add nuw nsw i64 [[IV]], 8
24 %indvars.iv.next = add i64 %indvars.iv, 1
25 %arrayidx2 = getelementptr inbounds double, double* %c, i64 %indvars.iv.next
26 %tmp2 = load double, double* %arrayidx2, align 8
27 %mul = fmul double %tmp1, %tmp2
28 %arrayidx4 = getelementptr inbounds double, double* %a, i64 %indvars.iv
29 store double %mul, double* %arrayidx4, align 8
30 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
31 ; Comparison should be 19 * 8 = 152.
32 ; CHECK: icmp eq i32 {{%[^,]+}}, 152
33 %exitcond = icmp eq i32 %lftr.wideiv, 20
34 br i1 %exitcond, label %for.end, label %for.body
36 for.end: ; preds = %for.body