1; RUN: opt < %s -S -loop-unroll -unroll-max-iteration-count-to-analyze=100 -unroll-threshold=12 -unroll-max-percent-threshold-boost=400 | FileCheck %s
2; RUN: opt < %s -S -passes='require<opt-remark-emit>,loop(loop-unroll-full)' -unroll-max-iteration-count-to-analyze=100 -unroll-threshold=12 -unroll-max-percent-threshold-boost=400 | FileCheck %s
3target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
4
5@known_constant = internal unnamed_addr constant [10 x i32] [i32 0, i32 0, i32 0, i32 0, i32 1, i32 0, i32 0, i32 0, i32 0, i32 0], align 16
6
7; If a load becomes a constant after loop unrolling, we sometimes can simplify
8; CFG. This test verifies that we handle such cases.
9; After one operand in an instruction is constant-folded and the
10; instruction is simplified, the other operand might become dead.
11; In this test we have::
12; for i in 1..10:
13;   r += A[i] * B[i]
14; A[i] is 0 almost at every iteration, so there is no need in loading B[i] at
15; all.
16
17
18; CHECK-LABEL: @unroll_dce
19; CHECK-NOT:   br i1 %exitcond, label %for.end, label %for.body
20define i32 @unroll_dce(i32* noalias nocapture readonly %b) {
21entry:
22  br label %for.body
23
24for.body:                                         ; preds = %for.body, %entry
25  %iv.0 = phi i64 [ 0, %entry ], [ %iv.1, %for.body ]
26  %r.0 = phi i32 [ 0, %entry ], [ %r.1, %for.body ]
27  %arrayidx1 = getelementptr inbounds [10 x i32], [10 x i32]* @known_constant, i64 0, i64 %iv.0
28  %x1 = load i32, i32* %arrayidx1, align 4
29  %arrayidx2 = getelementptr inbounds i32, i32* %b, i64 %iv.0
30  %x2 = load i32, i32* %arrayidx2, align 4
31  %mul = mul i32 %x1, %x2
32  %r.1 = add i32 %mul, %r.0
33  %iv.1 = add nuw nsw i64 %iv.0, 1
34  %exitcond = icmp eq i64 %iv.1, 10
35  br i1 %exitcond, label %for.end, label %for.body
36
37for.end:                                          ; preds = %for.body
38  ret i32 %r.1
39}
40