1; RUN: llc < %s -mtriple=powerpc64le-unknown-linux-gnu -verify-machineinstrs -mcpu=pwr8 | FileCheck %s --check-prefixes=CHECK,CHECK-PWR8
2; RUN: llc < %s -mtriple=powerpc64le-unknown-linux-gnu -verify-machineinstrs -mcpu=a2q | FileCheck %s --check-prefixes=CHECK,CHECK-A2Q
3
4; Verify that we do NOT generate the mtctr instruction for loop trip counts < 4
5; The latency of the mtctr is only justified if there are more than 4 comparisons that are removed as a result.
6
7@a = common local_unnamed_addr global i32 0, align 4
8@b = common local_unnamed_addr global i32 0, align 4
9@c = common local_unnamed_addr global i32 0, align 4
10@d = common local_unnamed_addr global i32 0, align 4
11@e = common local_unnamed_addr global i32 0, align 4
12@f = common local_unnamed_addr global i32 0, align 4
13@arr = common local_unnamed_addr global [5 x i32] zeroinitializer, align 4
14
15; Function Attrs: norecurse nounwind readonly
16define signext i32 @testTripCount2(i32 signext %a) {
17
18; CHECK-LABEL: testTripCount2:
19; CHECK-NOT: mtctr
20; CHECK: blr
21
22entry:
23  br label %for.body
24
25for.cond.cleanup:                                 ; preds = %for.body
26  ret i32 %add
27
28for.body:                                         ; preds = %entry, %for.body
29  %indvars.iv = phi i64 [ 1, %entry ], [ %indvars.iv.next, %for.body ]
30  %Sum.05 = phi i32 [ 0, %entry ], [ %add, %for.body ]
31  %arrayidx = getelementptr inbounds [5 x i32], [5 x i32]* @arr, i64 0, i64 %indvars.iv
32  %0 = load i32, i32* %arrayidx, align 4
33  %add = add nsw i32 %0, %Sum.05
34  %indvars.iv.next = add nsw i64 %indvars.iv, -1
35  %tobool = icmp eq i64 %indvars.iv, 0
36  br i1 %tobool, label %for.cond.cleanup, label %for.body
37}
38
39; Function Attrs: norecurse nounwind readonly
40define signext i32 @testTripCount3(i32 signext %a) {
41
42; CHECK-LABEL: testTripCount3:
43; CHECK-NOT: mtctr
44; CHECK: blr
45
46entry:
47  br label %for.body
48
49for.cond.cleanup:                                 ; preds = %for.body
50  ret i32 %add
51
52for.body:                                         ; preds = %entry, %for.body
53  %indvars.iv = phi i64 [ 2, %entry ], [ %indvars.iv.next, %for.body ]
54  %Sum.05 = phi i32 [ 0, %entry ], [ %add, %for.body ]
55  %arrayidx = getelementptr inbounds [5 x i32], [5 x i32]* @arr, i64 0, i64 %indvars.iv
56  %0 = load i32, i32* %arrayidx, align 4
57  %add = add nsw i32 %0, %Sum.05
58  %indvars.iv.next = add nsw i64 %indvars.iv, -1
59  %tobool = icmp eq i64 %indvars.iv, 0
60  br i1 %tobool, label %for.cond.cleanup, label %for.body
61}
62
63; Function Attrs: norecurse nounwind readonly
64
65define signext i32 @testTripCount4(i32 signext %a) {
66
67; CHECK-LABEL: testTripCount4:
68; CHECK: mtctr
69; CHECK: bdnz
70
71entry:
72  br label %for.body
73
74for.cond.cleanup:                                 ; preds = %for.body
75  ret i32 %add
76
77for.body:                                         ; preds = %entry, %for.body
78  %indvars.iv = phi i64 [ 3, %entry ], [ %indvars.iv.next, %for.body ]
79  %Sum.05 = phi i32 [ 0, %entry ], [ %add, %for.body ]
80  %arrayidx = getelementptr inbounds [5 x i32], [5 x i32]* @arr, i64 0, i64 %indvars.iv
81  %0 = load i32, i32* %arrayidx, align 4
82  %add = add nsw i32 %0, %Sum.05
83  %indvars.iv.next = add nsw i64 %indvars.iv, -1
84  %tobool = icmp eq i64 %indvars.iv, 0
85  br i1 %tobool, label %for.cond.cleanup, label %for.body
86}
87
88; Function Attrs: norecurse nounwind
89define signext i32 @testTripCount2NonSmallLoop() {
90; CHECK-LABEL: testTripCount2NonSmallLoop:
91; CHECK: blt
92; CHECK: beq
93; CHECK: blr
94
95entry:
96  %.pre = load i32, i32* @a, align 4
97  br label %for.body
98
99for.body:                                         ; preds = %entry, %if.end
100  %0 = phi i32 [ %.pre, %entry ], [ %1, %if.end ]
101  %dec4 = phi i32 [ 1, %entry ], [ %dec, %if.end ]
102  %b.03 = phi i8 [ 0, %entry ], [ %b.1, %if.end ]
103  %tobool1 = icmp eq i32 %0, 0
104  br i1 %tobool1, label %if.end, label %if.then
105
106if.then:                                          ; preds = %for.body
107  store i32 2, i32* @a, align 4
108  br label %if.end
109
110if.end:                                           ; preds = %for.body, %if.then
111  %1 = phi i32 [ 2, %if.then ], [ 0, %for.body ]
112  %b.1 = phi i8 [ 2, %if.then ], [ %b.03, %for.body ]
113  %dec = add nsw i32 %dec4, -1
114  %tobool = icmp eq i32 %dec4, 0
115  br i1 %tobool, label %for.end, label %for.body
116
117for.end:                                          ; preds = %if.end
118  %conv = zext i8 %b.1 to i32
119  ret i32 %conv
120}
121
122; On core a2q, IssueWidth is 1. On core pwr8, IssueWidth is 8.
123; a2q should use mtctr, but pwr8 should not use mtctr.
124define signext i32 @testTripCount5() {
125; CHECK-LABEL: testTripCount5:
126; CHECK-PWR8-NOT: mtctr
127; CHECK-A2Q: mtctr
128
129entry:
130  %.prea = load i32, i32* @a, align 4
131  %.preb = load i32, i32* @b, align 4
132  %.prec = load i32, i32* @c, align 4
133  %.pred = load i32, i32* @d, align 4
134  %.pree = load i32, i32* @e, align 4
135  %.pref = load i32, i32* @f, align 4
136  br label %for.body
137
138for.body:                                 ; preds = %entry, %for.body
139  %indvars.iv = phi i64 [ 2, %entry ], [ %indvars.iv.next, %for.body ]
140  %0 = phi i32 [ %.prea, %entry ], [ %6, %for.body ]
141  %1 = phi i32 [ %.preb, %entry ], [ %7, %for.body ]
142  %2 = phi i32 [ %.prec, %entry ], [ %8, %for.body ]
143  %3 = phi i32 [ %.pred, %entry ], [ %9, %for.body ]
144  %4 = phi i32 [ %.pree, %entry ], [ %10, %for.body ]
145  %5 = phi i32 [ %.pref, %entry ], [ %11, %for.body ]
146  %6 = add i32 %0, 1
147  %7 = add i32 %1, 1
148  %8 = add i32 %2, 1
149  %9 = add i32 %3, 1
150  %10 = add i32 %4, 1
151  %11 = add i32 %5, 1
152  %indvars.iv.next = add nsw i64 %indvars.iv, -1
153  %tobool = icmp eq i64 %indvars.iv, 0
154  br i1 %tobool, label %for.end, label %for.body
155
156for.end:                                 ; preds = %for.body
157  store i32 %6, i32* @a, align 4
158  store i32 %7, i32* @b, align 4
159  store i32 %8, i32* @c, align 4
160  store i32 %9, i32* @d, align 4
161  store i32 %10, i32* @e, align 4
162  store i32 %11, i32* @f, align 4
163  ret i32 0
164}
165
166