xref: /llvm-project/llvm/test/Transforms/LoopUnroll/X86/high-cost-expansion.ll (revision eecb99c5f66c8491766628a2925587e20f3b1dbd)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -mtriple=x86_64-unknown-linux-gnu -passes='loop-unroll' -unroll-runtime -S < %s 2>&1 | FileCheck %s
3
4define void @mask-high(i64 %arg, ptr dereferenceable(4) %arg1) {
5; CHECK-LABEL: @mask-high(
6; CHECK-NEXT:  bb:
7; CHECK-NEXT:    [[I:%.*]] = load i32, ptr [[ARG1:%.*]], align 4
8; CHECK-NEXT:    [[I2:%.*]] = sext i32 [[I]] to i64
9; CHECK-NEXT:    [[I3:%.*]] = and i64 [[ARG:%.*]], -16
10; CHECK-NEXT:    [[I4:%.*]] = or disjoint i64 1, [[I3]]
11; CHECK-NEXT:    [[I5:%.*]] = icmp sgt i64 [[I4]], [[I2]]
12; CHECK-NEXT:    br i1 [[I5]], label [[BB10:%.*]], label [[BB6_PREHEADER:%.*]]
13; CHECK:       bb6.preheader:
14; CHECK-NEXT:    br label [[BB6:%.*]]
15; CHECK:       bb6:
16; CHECK-NEXT:    [[I7:%.*]] = phi i64 [ [[I8:%.*]], [[BB6]] ], [ [[I4]], [[BB6_PREHEADER]] ]
17; CHECK-NEXT:    [[I8]] = add i64 [[I7]], 1
18; CHECK-NEXT:    [[I9:%.*]] = icmp slt i64 [[I7]], [[I2]]
19; CHECK-NEXT:    br i1 [[I9]], label [[BB6]], label [[BB10_LOOPEXIT:%.*]]
20; CHECK:       bb10.loopexit:
21; CHECK-NEXT:    br label [[BB10]]
22; CHECK:       bb10:
23; CHECK-NEXT:    ret void
24;
25bb:
26  %i = load i32, ptr %arg1, align 4
27  %i2 = sext i32 %i to i64
28  %i3 = and i64 %arg, -16
29  %i4 = or disjoint i64 1, %i3
30  %i5 = icmp sgt i64 %i4, %i2
31  br i1 %i5, label %bb10, label %bb6
32
33bb6:                                              ; preds = %bb6, %bb
34  %i7 = phi i64 [ %i4, %bb ], [ %i8, %bb6 ]
35  %i8 = add i64 %i7, 1
36  %i9 = icmp slt i64 %i7, %i2
37  br i1 %i9, label %bb6, label %bb10
38
39bb10:                                             ; preds = %bb6, %bb
40  ret void
41}
42
43
44define void @mask-low(i64 %arg, ptr dereferenceable(4) %arg1) {
45; CHECK-LABEL: @mask-low(
46; CHECK-NEXT:  bb:
47; CHECK-NEXT:    [[I:%.*]] = load i32, ptr [[ARG1:%.*]], align 4
48; CHECK-NEXT:    [[I2:%.*]] = sext i32 [[I]] to i64
49; CHECK-NEXT:    [[I3:%.*]] = and i64 [[ARG:%.*]], 16
50; CHECK-NEXT:    [[I4:%.*]] = add i64 1, [[I3]]
51; CHECK-NEXT:    [[I5:%.*]] = icmp sgt i64 [[I4]], [[I2]]
52; CHECK-NEXT:    br i1 [[I5]], label [[BB10:%.*]], label [[BB6_PREHEADER:%.*]]
53; CHECK:       bb6.preheader:
54; CHECK-NEXT:    br label [[BB6:%.*]]
55; CHECK:       bb6:
56; CHECK-NEXT:    [[I7:%.*]] = phi i64 [ [[I8:%.*]], [[BB6]] ], [ [[I4]], [[BB6_PREHEADER]] ]
57; CHECK-NEXT:    [[I8]] = add i64 [[I7]], 1
58; CHECK-NEXT:    [[I9:%.*]] = icmp slt i64 [[I7]], [[I2]]
59; CHECK-NEXT:    br i1 [[I9]], label [[BB6]], label [[BB10_LOOPEXIT:%.*]]
60; CHECK:       bb10.loopexit:
61; CHECK-NEXT:    br label [[BB10]]
62; CHECK:       bb10:
63; CHECK-NEXT:    ret void
64;
65bb:
66  %i = load i32, ptr %arg1, align 4
67  %i2 = sext i32 %i to i64
68  %i3 = and i64 %arg, 16
69  %i4 = add i64 1, %i3
70  %i5 = icmp sgt i64 %i4, %i2
71  br i1 %i5, label %bb10, label %bb6
72
73bb6:                                              ; preds = %bb6, %bb
74  %i7 = phi i64 [ %i4, %bb ], [ %i8, %bb6 ]
75  %i8 = add i64 %i7, 1
76  %i9 = icmp slt i64 %i7, %i2
77  br i1 %i9, label %bb6, label %bb10
78
79bb10:                                             ; preds = %bb6, %bb
80  ret void
81}
82