1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=riscv64 | FileCheck %s
3
4
5; Make sure we don't emit a pair of shift for the zext in the preheader. We
6; can tell that bit 31 is 0 in the preheader and rely on %n already being
7; sign extended without adding zeros explicitly.
8define void @test1(ptr nocapture noundef %a, i32 noundef signext %n) {
9; CHECK-LABEL: test1:
10; CHECK:       # %bb.0: # %entry
11; CHECK-NEXT:    blez a1, .LBB0_2
12; CHECK-NEXT:  .LBB0_1: # %for.body
13; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
14; CHECK-NEXT:    lw a2, 0(a0)
15; CHECK-NEXT:    addiw a2, a2, 4
16; CHECK-NEXT:    sw a2, 0(a0)
17; CHECK-NEXT:    addi a1, a1, -1
18; CHECK-NEXT:    addi a0, a0, 4
19; CHECK-NEXT:    bnez a1, .LBB0_1
20; CHECK-NEXT:  .LBB0_2: # %for.cond.cleanup
21; CHECK-NEXT:    ret
22entry:
23  %cmp3 = icmp sgt i32 %n, 0
24  br i1 %cmp3, label %for.body.preheader, label %for.cond.cleanup
25
26for.body.preheader:                               ; preds = %entry
27  %wide.trip.count = zext i32 %n to i64
28  br label %for.body
29
30for.cond.cleanup:                                 ; preds = %for.body, %entry
31  ret void
32
33for.body:                                         ; preds = %for.body.preheader, %for.body
34  %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
35  %arrayidx = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
36  %0 = load i32, ptr %arrayidx, align 4
37  %add = add nsw i32 %0, 4
38  store i32 %add, ptr %arrayidx, align 4
39  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
40  %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count
41  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
42}
43
44; Make sure we convert the 4294967294 in for.body.preheader.new to -2 based on
45; the upper 33 bits being zero by the dominating condition %cmp3.
46define void @test2(ptr nocapture noundef %a, i32 noundef signext %n) {
47; CHECK-LABEL: test2:
48; CHECK:       # %bb.0: # %entry
49; CHECK-NEXT:    blez a1, .LBB1_7
50; CHECK-NEXT:  # %bb.1: # %for.body.preheader
51; CHECK-NEXT:    li a3, 1
52; CHECK-NEXT:    andi a2, a1, 1
53; CHECK-NEXT:    bne a1, a3, .LBB1_3
54; CHECK-NEXT:  # %bb.2:
55; CHECK-NEXT:    li a3, 0
56; CHECK-NEXT:    j .LBB1_5
57; CHECK-NEXT:  .LBB1_3: # %for.body.preheader.new
58; CHECK-NEXT:    li a3, 0
59; CHECK-NEXT:    andi a1, a1, -2
60; CHECK-NEXT:    addi a4, a0, 4
61; CHECK-NEXT:  .LBB1_4: # %for.body
62; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
63; CHECK-NEXT:    lw a5, -4(a4)
64; CHECK-NEXT:    lw a6, 0(a4)
65; CHECK-NEXT:    addiw a5, a5, 4
66; CHECK-NEXT:    sw a5, -4(a4)
67; CHECK-NEXT:    addiw a5, a6, 4
68; CHECK-NEXT:    sw a5, 0(a4)
69; CHECK-NEXT:    addi a3, a3, 2
70; CHECK-NEXT:    addi a4, a4, 8
71; CHECK-NEXT:    bne a1, a3, .LBB1_4
72; CHECK-NEXT:  .LBB1_5: # %for.cond.cleanup.loopexit.unr-lcssa
73; CHECK-NEXT:    beqz a2, .LBB1_7
74; CHECK-NEXT:  # %bb.6: # %for.body.epil
75; CHECK-NEXT:    slli a1, a3, 2
76; CHECK-NEXT:    add a0, a0, a1
77; CHECK-NEXT:    lw a1, 0(a0)
78; CHECK-NEXT:    addiw a1, a1, 4
79; CHECK-NEXT:    sw a1, 0(a0)
80; CHECK-NEXT:  .LBB1_7: # %for.cond.cleanup
81; CHECK-NEXT:    ret
82entry:
83  %cmp3 = icmp sgt i32 %n, 0
84  br i1 %cmp3, label %for.body.preheader, label %for.cond.cleanup
85
86for.body.preheader:                               ; preds = %entry
87  %wide.trip.count = zext i32 %n to i64
88  %xtraiter = and i64 %wide.trip.count, 1
89  %0 = icmp eq i32 %n, 1
90  br i1 %0, label %for.cond.cleanup.loopexit.unr-lcssa, label %for.body.preheader.new
91
92for.body.preheader.new:                           ; preds = %for.body.preheader
93  %unroll_iter = and i64 %wide.trip.count, 4294967294
94  br label %for.body
95
96for.cond.cleanup.loopexit.unr-lcssa:              ; preds = %for.body, %for.body.preheader
97  %indvars.iv.unr = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next.1, %for.body ]
98  %lcmp.mod.not = icmp eq i64 %xtraiter, 0
99  br i1 %lcmp.mod.not, label %for.cond.cleanup, label %for.body.epil
100
101for.body.epil:                                    ; preds = %for.cond.cleanup.loopexit.unr-lcssa
102  %arrayidx.epil = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.unr
103  %1 = load i32, ptr %arrayidx.epil, align 4
104  %add.epil = add nsw i32 %1, 4
105  store i32 %add.epil, ptr %arrayidx.epil, align 4
106  br label %for.cond.cleanup
107
108for.cond.cleanup:                                 ; preds = %for.body.epil, %for.cond.cleanup.loopexit.unr-lcssa, %entry
109  ret void
110
111for.body:                                         ; preds = %for.body, %for.body.preheader.new
112  %indvars.iv = phi i64 [ 0, %for.body.preheader.new ], [ %indvars.iv.next.1, %for.body ]
113  %niter = phi i64 [ 0, %for.body.preheader.new ], [ %niter.next.1, %for.body ]
114  %arrayidx = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
115  %2 = load i32, ptr %arrayidx, align 4
116  %add = add nsw i32 %2, 4
117  store i32 %add, ptr %arrayidx, align 4
118  %indvars.iv.next = or i64 %indvars.iv, 1
119  %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
120  %3 = load i32, ptr %arrayidx.1, align 4
121  %add.1 = add nsw i32 %3, 4
122  store i32 %add.1, ptr %arrayidx.1, align 4
123  %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv, 2
124  %niter.next.1 = add i64 %niter, 2
125  %niter.ncmp.1 = icmp eq i64 %niter.next.1, %unroll_iter
126  br i1 %niter.ncmp.1, label %for.cond.cleanup.loopexit.unr-lcssa, label %for.body
127}
128