1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve -verify-machineinstrs %s -o - | FileCheck %s
3
4define void @arm_min_q31(i32* nocapture readonly %pSrc, i32 %blockSize, i32* nocapture %pResult, i32* nocapture %pIndex) {
5; CHECK-LABEL: arm_min_q31:
6; CHECK:       @ %bb.0: @ %entry
7; CHECK-NEXT:    .save {r4, r5, r6, r7, r8, r9, r10, r11, lr}
8; CHECK-NEXT:    push.w {r4, r5, r6, r7, r8, r9, r10, r11, lr}
9; CHECK-NEXT:    ldr.w r12, [r0]
10; CHECK-NEXT:    subs.w r9, r1, #1
11; CHECK-NEXT:    beq .LBB0_3
12; CHECK-NEXT:  @ %bb.1: @ %while.body.preheader
13; CHECK-NEXT:    subs r7, r1, #2
14; CHECK-NEXT:    and r8, r9, #3
15; CHECK-NEXT:    cmp r7, #3
16; CHECK-NEXT:    bhs .LBB0_4
17; CHECK-NEXT:  @ %bb.2:
18; CHECK-NEXT:    movs r6, #0
19; CHECK-NEXT:    b .LBB0_6
20; CHECK-NEXT:  .LBB0_3:
21; CHECK-NEXT:    movs r6, #0
22; CHECK-NEXT:    b .LBB0_10
23; CHECK-NEXT:  .LBB0_4: @ %while.body.preheader.new
24; CHECK-NEXT:    bic r7, r9, #3
25; CHECK-NEXT:    movs r6, #1
26; CHECK-NEXT:    subs r7, #4
27; CHECK-NEXT:    add.w lr, r6, r7, lsr #2
28; CHECK-NEXT:    movs r6, #0
29; CHECK-NEXT:    movs r7, #4
30; CHECK-NEXT:  .LBB0_5: @ %while.body
31; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
32; CHECK-NEXT:    ldr r10, [r0, #16]!
33; CHECK-NEXT:    sub.w r9, r9, #4
34; CHECK-NEXT:    ldrd r5, r4, [r0, #-12]
35; CHECK-NEXT:    ldr r11, [r0, #-4]
36; CHECK-NEXT:    cmp r12, r5
37; CHECK-NEXT:    it gt
38; CHECK-NEXT:    subgt r6, r7, #3
39; CHECK-NEXT:    csel r5, r5, r12, gt
40; CHECK-NEXT:    cmp r5, r4
41; CHECK-NEXT:    it gt
42; CHECK-NEXT:    subgt r6, r7, #2
43; CHECK-NEXT:    csel r5, r4, r5, gt
44; CHECK-NEXT:    cmp r5, r11
45; CHECK-NEXT:    it gt
46; CHECK-NEXT:    subgt r6, r7, #1
47; CHECK-NEXT:    csel r5, r11, r5, gt
48; CHECK-NEXT:    cmp r5, r10
49; CHECK-NEXT:    csel r6, r7, r6, gt
50; CHECK-NEXT:    add.w r7, r7, #4
51; CHECK-NEXT:    csel r12, r10, r5, gt
52; CHECK-NEXT:    le lr, .LBB0_5
53; CHECK-NEXT:  .LBB0_6: @ %while.end.loopexit.unr-lcssa
54; CHECK-NEXT:    cmp.w r8, #0
55; CHECK-NEXT:    beq .LBB0_10
56; CHECK-NEXT:  @ %bb.7: @ %while.body.epil
57; CHECK-NEXT:    ldr r7, [r0, #4]
58; CHECK-NEXT:    sub.w r1, r1, r9
59; CHECK-NEXT:    cmp r12, r7
60; CHECK-NEXT:    csel r6, r1, r6, gt
61; CHECK-NEXT:    csel r12, r7, r12, gt
62; CHECK-NEXT:    cmp.w r8, #1
63; CHECK-NEXT:    beq .LBB0_10
64; CHECK-NEXT:  @ %bb.8: @ %while.body.epil.1
65; CHECK-NEXT:    ldr r7, [r0, #8]
66; CHECK-NEXT:    cmp r12, r7
67; CHECK-NEXT:    csinc r6, r6, r1, le
68; CHECK-NEXT:    csel r12, r7, r12, gt
69; CHECK-NEXT:    cmp.w r8, #2
70; CHECK-NEXT:    beq .LBB0_10
71; CHECK-NEXT:  @ %bb.9: @ %while.body.epil.2
72; CHECK-NEXT:    ldr r0, [r0, #12]
73; CHECK-NEXT:    cmp r12, r0
74; CHECK-NEXT:    it gt
75; CHECK-NEXT:    addgt r6, r1, #2
76; CHECK-NEXT:    csel r12, r0, r12, gt
77; CHECK-NEXT:  .LBB0_10: @ %while.end
78; CHECK-NEXT:    str.w r12, [r2]
79; CHECK-NEXT:    str r6, [r3]
80; CHECK-NEXT:    pop.w {r4, r5, r6, r7, r8, r9, r10, r11, pc}
81entry:
82  %0 = load i32, i32* %pSrc, align 4
83  %blkCnt.015 = add i32 %blockSize, -1
84  %cmp.not17 = icmp eq i32 %blkCnt.015, 0
85  br i1 %cmp.not17, label %while.end, label %while.body.preheader
86
87while.body.preheader:                             ; preds = %entry
88  %1 = add i32 %blockSize, -2
89  %xtraiter = and i32 %blkCnt.015, 3
90  %2 = icmp ult i32 %1, 3
91  br i1 %2, label %while.end.loopexit.unr-lcssa, label %while.body.preheader.new
92
93while.body.preheader.new:                         ; preds = %while.body.preheader
94  %unroll_iter = and i32 %blkCnt.015, -4
95  br label %while.body
96
97while.body:                                       ; preds = %while.body, %while.body.preheader.new
98  %pSrc.addr.021.pn = phi i32* [ %pSrc, %while.body.preheader.new ], [ %pSrc.addr.021.3, %while.body ]
99  %blkCnt.020 = phi i32 [ %blkCnt.015, %while.body.preheader.new ], [ %blkCnt.0.3, %while.body ]
100  %outIndex.019 = phi i32 [ 0, %while.body.preheader.new ], [ %spec.select14.3, %while.body ]
101  %out.018 = phi i32 [ %0, %while.body.preheader.new ], [ %spec.select.3, %while.body ]
102  %niter = phi i32 [ %unroll_iter, %while.body.preheader.new ], [ %niter.nsub.3, %while.body ]
103  %pSrc.addr.021 = getelementptr inbounds i32, i32* %pSrc.addr.021.pn, i32 1
104  %3 = load i32, i32* %pSrc.addr.021, align 4
105  %cmp2 = icmp sgt i32 %out.018, %3
106  %sub3 = sub i32 %blockSize, %blkCnt.020
107  %spec.select = select i1 %cmp2, i32 %3, i32 %out.018
108  %spec.select14 = select i1 %cmp2, i32 %sub3, i32 %outIndex.019
109  %blkCnt.0 = add i32 %blkCnt.020, -1
110  %pSrc.addr.021.1 = getelementptr inbounds i32, i32* %pSrc.addr.021.pn, i32 2
111  %4 = load i32, i32* %pSrc.addr.021.1, align 4
112  %cmp2.1 = icmp sgt i32 %spec.select, %4
113  %sub3.1 = sub i32 %blockSize, %blkCnt.0
114  %spec.select.1 = select i1 %cmp2.1, i32 %4, i32 %spec.select
115  %spec.select14.1 = select i1 %cmp2.1, i32 %sub3.1, i32 %spec.select14
116  %blkCnt.0.1 = add i32 %blkCnt.020, -2
117  %pSrc.addr.021.2 = getelementptr inbounds i32, i32* %pSrc.addr.021.pn, i32 3
118  %5 = load i32, i32* %pSrc.addr.021.2, align 4
119  %cmp2.2 = icmp sgt i32 %spec.select.1, %5
120  %sub3.2 = sub i32 %blockSize, %blkCnt.0.1
121  %spec.select.2 = select i1 %cmp2.2, i32 %5, i32 %spec.select.1
122  %spec.select14.2 = select i1 %cmp2.2, i32 %sub3.2, i32 %spec.select14.1
123  %blkCnt.0.2 = add i32 %blkCnt.020, -3
124  %pSrc.addr.021.3 = getelementptr inbounds i32, i32* %pSrc.addr.021.pn, i32 4
125  %6 = load i32, i32* %pSrc.addr.021.3, align 4
126  %cmp2.3 = icmp sgt i32 %spec.select.2, %6
127  %sub3.3 = sub i32 %blockSize, %blkCnt.0.2
128  %spec.select.3 = select i1 %cmp2.3, i32 %6, i32 %spec.select.2
129  %spec.select14.3 = select i1 %cmp2.3, i32 %sub3.3, i32 %spec.select14.2
130  %blkCnt.0.3 = add i32 %blkCnt.020, -4
131  %niter.nsub.3 = add i32 %niter, -4
132  %niter.ncmp.3 = icmp eq i32 %niter.nsub.3, 0
133  br i1 %niter.ncmp.3, label %while.end.loopexit.unr-lcssa, label %while.body
134
135while.end.loopexit.unr-lcssa:                     ; preds = %while.body, %while.body.preheader
136  %spec.select.lcssa.ph = phi i32 [ undef, %while.body.preheader ], [ %spec.select.3, %while.body ]
137  %spec.select14.lcssa.ph = phi i32 [ undef, %while.body.preheader ], [ %spec.select14.3, %while.body ]
138  %pSrc.addr.021.pn.unr = phi i32* [ %pSrc, %while.body.preheader ], [ %pSrc.addr.021.3, %while.body ]
139  %blkCnt.020.unr = phi i32 [ %blkCnt.015, %while.body.preheader ], [ %blkCnt.0.3, %while.body ]
140  %outIndex.019.unr = phi i32 [ 0, %while.body.preheader ], [ %spec.select14.3, %while.body ]
141  %out.018.unr = phi i32 [ %0, %while.body.preheader ], [ %spec.select.3, %while.body ]
142  %lcmp.mod.not = icmp eq i32 %xtraiter, 0
143  br i1 %lcmp.mod.not, label %while.end, label %while.body.epil
144
145while.body.epil:                                  ; preds = %while.end.loopexit.unr-lcssa
146  %pSrc.addr.021.epil = getelementptr inbounds i32, i32* %pSrc.addr.021.pn.unr, i32 1
147  %7 = load i32, i32* %pSrc.addr.021.epil, align 4
148  %cmp2.epil = icmp sgt i32 %out.018.unr, %7
149  %sub3.epil = sub i32 %blockSize, %blkCnt.020.unr
150  %spec.select.epil = select i1 %cmp2.epil, i32 %7, i32 %out.018.unr
151  %spec.select14.epil = select i1 %cmp2.epil, i32 %sub3.epil, i32 %outIndex.019.unr
152  %epil.iter.cmp.not = icmp eq i32 %xtraiter, 1
153  br i1 %epil.iter.cmp.not, label %while.end, label %while.body.epil.1
154
155while.end:                                        ; preds = %while.end.loopexit.unr-lcssa, %while.body.epil.2, %while.body.epil.1, %while.body.epil, %entry
156  %out.0.lcssa = phi i32 [ %0, %entry ], [ %spec.select.lcssa.ph, %while.end.loopexit.unr-lcssa ], [ %spec.select.epil, %while.body.epil ], [ %spec.select.epil.1, %while.body.epil.1 ], [ %spec.select.epil.2, %while.body.epil.2 ]
157  %outIndex.0.lcssa = phi i32 [ 0, %entry ], [ %spec.select14.lcssa.ph, %while.end.loopexit.unr-lcssa ], [ %spec.select14.epil, %while.body.epil ], [ %spec.select14.epil.1, %while.body.epil.1 ], [ %spec.select14.epil.2, %while.body.epil.2 ]
158  store i32 %out.0.lcssa, i32* %pResult, align 4
159  store i32 %outIndex.0.lcssa, i32* %pIndex, align 4
160  ret void
161
162while.body.epil.1:                                ; preds = %while.body.epil
163  %blkCnt.0.epil = add i32 %blkCnt.020.unr, -1
164  %pSrc.addr.021.epil.1 = getelementptr inbounds i32, i32* %pSrc.addr.021.pn.unr, i32 2
165  %8 = load i32, i32* %pSrc.addr.021.epil.1, align 4
166  %cmp2.epil.1 = icmp sgt i32 %spec.select.epil, %8
167  %sub3.epil.1 = sub i32 %blockSize, %blkCnt.0.epil
168  %spec.select.epil.1 = select i1 %cmp2.epil.1, i32 %8, i32 %spec.select.epil
169  %spec.select14.epil.1 = select i1 %cmp2.epil.1, i32 %sub3.epil.1, i32 %spec.select14.epil
170  %epil.iter.cmp.1.not = icmp eq i32 %xtraiter, 2
171  br i1 %epil.iter.cmp.1.not, label %while.end, label %while.body.epil.2
172
173while.body.epil.2:                                ; preds = %while.body.epil.1
174  %blkCnt.0.epil.1 = add i32 %blkCnt.020.unr, -2
175  %pSrc.addr.021.epil.2 = getelementptr inbounds i32, i32* %pSrc.addr.021.pn.unr, i32 3
176  %9 = load i32, i32* %pSrc.addr.021.epil.2, align 4
177  %cmp2.epil.2 = icmp sgt i32 %spec.select.epil.1, %9
178  %sub3.epil.2 = sub i32 %blockSize, %blkCnt.0.epil.1
179  %spec.select.epil.2 = select i1 %cmp2.epil.2, i32 %9, i32 %spec.select.epil.1
180  %spec.select14.epil.2 = select i1 %cmp2.epil.2, i32 %sub3.epil.2, i32 %spec.select14.epil.1
181  br label %while.end
182}
183