1 ; RUN: opt < %s -loop-reduce -mtriple=x86_64-- -S | FileCheck %s -check-prefix=BOTH -check-prefix=INSN
2 ; RUN: opt < %s -loop-reduce -mtriple=x86_64-- -lsr-insns-cost=false -S | FileCheck %s -check-prefix=BOTH -check-prefix=REGS
3 ; RUN: llc < %s -O2 -mtriple=x86_64-- -lsr-insns-cost -asm-verbose=0 | FileCheck %s
5 ; OPT checks that LSR prefers less instructions to less registers.
6 ; For x86 LSR should prefer complicated address to new lsr induction
10 ; INSN: getelementptr i32, i32* %x, i64 %indvars.iv
11 ; INSN: getelementptr i32, i32* %y, i64 %indvars.iv
12 ; INSN: getelementptr i32, i32* %q, i64 %indvars.iv
13 ; REGS: %lsr.iv4 = phi
14 ; REGS: %lsr.iv2 = phi
15 ; REGS: %lsr.iv1 = phi
16 ; REGS: getelementptr i32, i32* %lsr.iv1, i64 1
17 ; REGS: getelementptr i32, i32* %lsr.iv2, i64 1
18 ; REGS: getelementptr i32, i32* %lsr.iv4, i64 1
20 ; LLC checks that LSR prefers less instructions to less registers.
21 ; LSR should prefer complicated address to additonal add instructions.
24 ; CHECK-NEXT: movl (%r{{.+}},
25 ; CHECK-NEXT: addl (%r{{.+}},
26 ; CHECK-NEXT: movl %e{{.+}}, (%r{{.+}},
28 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
30 ; Function Attrs: norecurse nounwind uwtable
31 define void @foo(i32* nocapture readonly %x, i32* nocapture readonly %y, i32* nocapture %q, i32 %n) {
33 %cmp10 = icmp sgt i32 %n, 0
34 br i1 %cmp10, label %for.body.preheader, label %for.cond.cleanup
36 for.body.preheader: ; preds = %entry
37 %wide.trip.count = zext i32 %n to i64
40 for.cond.cleanup.loopexit: ; preds = %for.body
41 br label %for.cond.cleanup
43 for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry
46 for.body: ; preds = %for.body, %for.body.preheader
47 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
48 %arrayidx = getelementptr inbounds i32, i32* %x, i64 %indvars.iv
49 %tmp = load i32, i32* %arrayidx, align 4
50 %arrayidx2 = getelementptr inbounds i32, i32* %y, i64 %indvars.iv
51 %tmp1 = load i32, i32* %arrayidx2, align 4
52 %add = add nsw i32 %tmp1, %tmp
53 %arrayidx4 = getelementptr inbounds i32, i32* %q, i64 %indvars.iv
54 store i32 %add, i32* %arrayidx4, align 4
55 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
56 %exitcond = icmp eq i64 %indvars.iv.next, %wide.trip.count
57 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body