xref: /llvm-project/llvm/test/CodeGen/RISCV/unroll-loop-cse.ll (revision 9122c5235ec85ce0c0ad337e862b006e7b349d84)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=riscv32 -verify-machineinstrs < %s | FileCheck %s
3; RUN: llc -mtriple=riscv64 -verify-machineinstrs < %s | FileCheck %s
4
5@x = global [6 x i32] [i32 0, i32 1, i32 2, i32 3, i32 4, i32 5], align 4
6@check = global [6 x i32] [i32 0, i32 1, i32 2, i32 3, i32 4, i32 5], align 4
7
8; This test case checks whether the base address of an array is repeatedly
9; rematerialised within a unrolled loop.
10define signext i32 @unroll_loop_cse() {
11; CHECK-LABEL: unroll_loop_cse:
12; CHECK:       # %bb.0:
13; CHECK-NEXT:    lui a0, %hi(x)
14; CHECK-NEXT:    lui a1, %hi(check)
15; CHECK-NEXT:    lw a2, %lo(x)(a0)
16; CHECK-NEXT:    lw a1, %lo(check)(a1)
17; CHECK-NEXT:    li a0, 1
18; CHECK-NEXT:    bne a2, a1, .LBB0_6
19; CHECK-NEXT:  # %bb.1:
20; CHECK-NEXT:    lui a1, %hi(x)
21; CHECK-NEXT:    addi a1, a1, %lo(x)
22; CHECK-NEXT:    lui a2, %hi(check)
23; CHECK-NEXT:    addi a2, a2, %lo(check)
24; CHECK-NEXT:    lw a3, 4(a1)
25; CHECK-NEXT:    lw a4, 4(a2)
26; CHECK-NEXT:    bne a3, a4, .LBB0_6
27; CHECK-NEXT:  # %bb.2:
28; CHECK-NEXT:    lw a3, 8(a1)
29; CHECK-NEXT:    lw a4, 8(a2)
30; CHECK-NEXT:    bne a3, a4, .LBB0_6
31; CHECK-NEXT:  # %bb.3:
32; CHECK-NEXT:    lw a3, 12(a1)
33; CHECK-NEXT:    lw a4, 12(a2)
34; CHECK-NEXT:    bne a3, a4, .LBB0_6
35; CHECK-NEXT:  # %bb.4:
36; CHECK-NEXT:    lw a3, 16(a1)
37; CHECK-NEXT:    lw a4, 16(a2)
38; CHECK-NEXT:    bne a3, a4, .LBB0_6
39; CHECK-NEXT:  # %bb.5:
40; CHECK-NEXT:    lw a0, 20(a1)
41; CHECK-NEXT:    lw a1, 20(a2)
42; CHECK-NEXT:    xor a0, a0, a1
43; CHECK-NEXT:    snez a0, a0
44; CHECK-NEXT:  .LBB0_6:
45; CHECK-NEXT:    ret
46  %1 = load i32, ptr @x, align 4
47  %2 = load i32, ptr @check, align 4
48  %3 = icmp eq i32 %1, %2
49  br i1 %3, label %4, label %25
50
514:
52  %5 = load i32, ptr getelementptr inbounds ([6 x i32], ptr @x, i64 0, i64 1), align 4
53  %6 = load i32, ptr getelementptr inbounds ([6 x i32], ptr @check, i64 0, i64 1), align 4
54  %7 = icmp eq i32 %5, %6
55  br i1 %7, label %8, label %25
56
578:
58  %9 = load i32, ptr getelementptr inbounds ([6 x i32], ptr @x, i64 0, i64 2), align 4
59  %10 = load i32, ptr getelementptr inbounds ([6 x i32], ptr @check, i64 0, i64 2), align 4
60  %11 = icmp eq i32 %9, %10
61  br i1 %11, label %12, label %25
62
6312:
64  %13 = load i32, ptr getelementptr inbounds ([6 x i32], ptr @x, i64 0, i64 3), align 4
65  %14 = load i32, ptr getelementptr inbounds ([6 x i32], ptr @check, i64 0, i64 3), align 4
66  %15 = icmp eq i32 %13, %14
67  br i1 %15, label %16, label %25
68
6916:
70  %17 = load i32, ptr getelementptr inbounds ([6 x i32], ptr @x, i64 0, i64 4), align 4
71  %18 = load i32, ptr getelementptr inbounds ([6 x i32], ptr @check, i64 0, i64 4), align 4
72  %19 = icmp eq i32 %17, %18
73  br i1 %19, label %20, label %25
74
7520:
76  %21 = load i32, ptr getelementptr inbounds ([6 x i32], ptr @x, i64 0, i64 5), align 4
77  %22 = load i32, ptr getelementptr inbounds ([6 x i32], ptr @check, i64 0, i64 5), align 4
78  %23 = icmp ne i32 %21, %22
79  %24 = zext i1 %23 to i32
80  br label %25
81
8225:
83  %26 = phi i32 [ 1, %0 ], [ 1, %4 ], [ 1, %8 ], [ 1, %12 ], [ 1, %16 ], [ %24, %20 ]
84  ret i32 %26
85}
86