xref: /llvm-project/llvm/test/CodeGen/ARM/lsr-unfolded-offset.ll (revision 44587e2f7e732604cd6340061d40ac21e7e188e5)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -regalloc=greedy -arm-atomic-cfg-tidy=0 < %s | FileCheck %s
3
4; LSR shouldn't introduce more induction variables than needed, increasing
5; register pressure and therefore spilling. There is more room for improvement
6; here.
7
8; CHECK: sub sp, #{{40|36|32|28|24}}
9
10; CHECK: %for.inc
11; CHECK: adds    r6, #1
12; CHECK: adds    r4, #24
13; CHECK: cmp     r1, r6
14; CHECK: bne     LBB0_3
15
16
17target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:32-f32:32:32-f64:32:32-v64:32:64-v128:32:128-a0:0:32-n32"
18target triple = "thumbv7-apple-ios"
19
20%struct.partition_entry = type { i32, i32, i64, i64 }
21
22define i32 @partition_overlap_check(%struct.partition_entry* nocapture %part, i32 %num_entries) nounwind readonly optsize ssp "frame-pointer"="all" {
23entry:
24  %cmp79 = icmp sgt i32 %num_entries, 0
25  br i1 %cmp79, label %outer.loop, label %for.end72
26
27outer.loop:                                 ; preds = %for.inc69, %entry
28  %overlap.081 = phi i32 [ %overlap.4, %for.inc69 ], [ 0, %entry ]
29  %0 = phi i32 [ %inc71, %for.inc69 ], [ 0, %entry ]
30  %offset = getelementptr %struct.partition_entry, %struct.partition_entry* %part, i32 %0, i32 2
31  %len = getelementptr %struct.partition_entry, %struct.partition_entry* %part, i32 %0, i32 3
32  %tmp5 = load i64, i64* %offset, align 4
33  %tmp15 = load i64, i64* %len, align 4
34  %add = add nsw i64 %tmp15, %tmp5
35  br label %inner.loop
36
37inner.loop:                                       ; preds = %for.inc, %outer.loop
38  %overlap.178 = phi i32 [ %overlap.081, %outer.loop ], [ %overlap.4, %for.inc ]
39  %1 = phi i32 [ 0, %outer.loop ], [ %inc, %for.inc ]
40  %cmp23 = icmp eq i32 %0, %1
41  br i1 %cmp23, label %for.inc, label %if.end
42
43if.end:                                           ; preds = %inner.loop
44  %len39 = getelementptr %struct.partition_entry, %struct.partition_entry* %part, i32 %1, i32 3
45  %offset28 = getelementptr %struct.partition_entry, %struct.partition_entry* %part, i32 %1, i32 2
46  %tmp29 = load i64, i64* %offset28, align 4
47  %tmp40 = load i64, i64* %len39, align 4
48  %add41 = add nsw i64 %tmp40, %tmp29
49  %cmp44 = icmp sge i64 %tmp29, %tmp5
50  %cmp47 = icmp slt i64 %tmp29, %add
51  %or.cond = and i1 %cmp44, %cmp47
52  %overlap.2 = select i1 %or.cond, i32 1, i32 %overlap.178
53  %cmp52 = icmp sle i64 %add41, %add
54  %cmp56 = icmp sgt i64 %add41, %tmp5
55  %or.cond74 = and i1 %cmp52, %cmp56
56  %overlap.3 = select i1 %or.cond74, i32 1, i32 %overlap.2
57  %cmp61 = icmp sgt i64 %tmp29, %tmp5
58  %cmp65 = icmp slt i64 %add41, %add
59  %or.cond75 = or i1 %cmp61, %cmp65
60  br i1 %or.cond75, label %for.inc, label %if.then66
61
62if.then66:                                        ; preds = %if.end
63  br label %for.inc
64
65for.inc:                                          ; preds = %if.end, %if.then66, %inner.loop
66  %overlap.4 = phi i32 [ %overlap.178, %inner.loop ], [ 1, %if.then66 ], [ %overlap.3, %if.end ]
67  %inc = add nsw i32 %1, 1
68  %exitcond = icmp eq i32 %inc, %num_entries
69  br i1 %exitcond, label %for.inc69, label %inner.loop
70
71for.inc69:                                        ; preds = %for.inc
72  %inc71 = add nsw i32 %0, 1
73  %exitcond83 = icmp eq i32 %inc71, %num_entries
74  br i1 %exitcond83, label %for.end72, label %outer.loop
75
76for.end72:                                        ; preds = %for.inc69, %entry
77  %overlap.0.lcssa = phi i32 [ 0, %entry ], [ %overlap.4, %for.inc69 ]
78  ret i32 %overlap.0.lcssa
79}
80