1; RUN: llc -march=hexagon -mcpu=hexagonv5 < %s | FileCheck %s
2;
3; Remove the unnecessary 'add' instruction used for the hardware loop setup.
4
5; CHECK: [[OP0:r[0-9]+]] = add([[OP1:r[0-9]+]], #-[[OP2:[0-9]+]]
6; CHECK-NOT: add([[OP0]], #[[OP2]])
7; CHECK: lsr([[OP1]], #{{[0-9]+}})
8; CHECK: loop0
9
10define void @matrix_mul_matrix(i32 %N, i32* nocapture %C, i16* nocapture readnone %A, i16* nocapture readnone %B) #0 {
11entry:
12  %cmp4 = icmp eq i32 %N, 0
13  br i1 %cmp4, label %for.end, label %for.body.preheader
14
15for.body.preheader:
16  %maxval = add i32 %N, -7
17  %0 = icmp sgt i32 %maxval, 0
18  br i1 %0, label %for.body.preheader9, label %for.body.ur.preheader
19
20for.body.preheader9:
21  br label %for.body
22
23for.body:
24  %arrayidx.phi = phi i32* [ %arrayidx.inc.7, %for.body ], [ %C, %for.body.preheader9 ]
25  %i.05 = phi i32 [ %inc.7, %for.body ], [ 0, %for.body.preheader9 ]
26  store i32 %i.05, i32* %arrayidx.phi, align 4
27  %inc = add i32 %i.05, 1
28  %arrayidx.inc = getelementptr i32, i32* %arrayidx.phi, i32 1
29  store i32 %inc, i32* %arrayidx.inc, align 4
30  %inc.1 = add i32 %i.05, 2
31  %arrayidx.inc.1 = getelementptr i32, i32* %arrayidx.phi, i32 2
32  store i32 %inc.1, i32* %arrayidx.inc.1, align 4
33  %inc.2 = add i32 %i.05, 3
34  %arrayidx.inc.2 = getelementptr i32, i32* %arrayidx.phi, i32 3
35  store i32 %inc.2, i32* %arrayidx.inc.2, align 4
36  %inc.3 = add i32 %i.05, 4
37  %arrayidx.inc.3 = getelementptr i32, i32* %arrayidx.phi, i32 4
38  store i32 %inc.3, i32* %arrayidx.inc.3, align 4
39  %inc.4 = add i32 %i.05, 5
40  %arrayidx.inc.4 = getelementptr i32, i32* %arrayidx.phi, i32 5
41  store i32 %inc.4, i32* %arrayidx.inc.4, align 4
42  %inc.5 = add i32 %i.05, 6
43  %arrayidx.inc.5 = getelementptr i32, i32* %arrayidx.phi, i32 6
44  store i32 %inc.5, i32* %arrayidx.inc.5, align 4
45  %inc.6 = add i32 %i.05, 7
46  %arrayidx.inc.6 = getelementptr i32, i32* %arrayidx.phi, i32 7
47  store i32 %inc.6, i32* %arrayidx.inc.6, align 4
48  %inc.7 = add i32 %i.05, 8
49  %exitcond.7 = icmp slt i32 %inc.7, %maxval
50  %arrayidx.inc.7 = getelementptr i32, i32* %arrayidx.phi, i32 8
51  br i1 %exitcond.7, label %for.body, label %for.end.loopexit.ur-lcssa
52
53for.end.loopexit.ur-lcssa:
54  %1 = icmp eq i32 %inc.7, %N
55  br i1 %1, label %for.end, label %for.body.ur.preheader
56
57for.body.ur.preheader:
58  %arrayidx.phi.ur.ph = phi i32* [ %C, %for.body.preheader ], [ %arrayidx.inc.7, %for.end.loopexit.ur-lcssa ]
59  %i.05.ur.ph = phi i32 [ 0, %for.body.preheader ], [ %inc.7, %for.end.loopexit.ur-lcssa ]
60  br label %for.body.ur
61
62for.body.ur:
63  %arrayidx.phi.ur = phi i32* [ %arrayidx.inc.ur, %for.body.ur ], [ %arrayidx.phi.ur.ph, %for.body.ur.preheader ]
64  %i.05.ur = phi i32 [ %inc.ur, %for.body.ur ], [ %i.05.ur.ph, %for.body.ur.preheader ]
65  store i32 %i.05.ur, i32* %arrayidx.phi.ur, align 4
66  %inc.ur = add i32 %i.05.ur, 1
67  %exitcond.ur = icmp eq i32 %inc.ur, %N
68  %arrayidx.inc.ur = getelementptr i32, i32* %arrayidx.phi.ur, i32 1
69  br i1 %exitcond.ur, label %for.end.loopexit, label %for.body.ur
70
71for.end.loopexit:
72  br label %for.end
73
74for.end:
75  ret void
76}
77