xref: /llvm-project/llvm/test/CodeGen/ARM/ParallelDSP/pr42729.ll (revision debbf8049d221966fd484c546fa2ddef11daec3c)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -mtriple=thumbv7-unknown-linux-android -arm-parallel-dsp -S %s -o - | FileCheck %s
3
4define void @undef_no_return(i16* %a) {
5; CHECK-LABEL: @undef_no_return(
6; CHECK-NEXT:  entry:
7; CHECK-NEXT:    [[INCDEC_PTR21:%.*]] = getelementptr inbounds i16, i16* [[A:%.*]], i32 3
8; CHECK-NEXT:    [[INCDEC_PTR29:%.*]] = getelementptr inbounds i16, i16* [[A]], i32 4
9; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
10; CHECK:       for.body:
11; CHECK-NEXT:    [[TMP0:%.*]] = load i16, i16* [[INCDEC_PTR21]], align 2
12; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i16* [[INCDEC_PTR21]] to i32*
13; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 2
14; CHECK-NEXT:    [[TMP3:%.*]] = trunc i32 [[TMP2]] to i16
15; CHECK-NEXT:    [[TMP4:%.*]] = sext i16 [[TMP3]] to i32
16; CHECK-NEXT:    [[TMP5:%.*]] = lshr i32 [[TMP2]], 16
17; CHECK-NEXT:    [[TMP6:%.*]] = trunc i32 [[TMP5]] to i16
18; CHECK-NEXT:    [[TMP7:%.*]] = sext i16 [[TMP6]] to i32
19; CHECK-NEXT:    [[CONV25:%.*]] = sext i16 [[TMP0]] to i32
20; CHECK-NEXT:    [[UGLYGEP15:%.*]] = getelementptr i8, i8* undef, i32 undef
21; CHECK-NEXT:    [[UGLYGEP1516:%.*]] = bitcast i8* [[UGLYGEP15]] to i16*
22; CHECK-NEXT:    [[SCEVGEP17:%.*]] = getelementptr i16, i16* [[UGLYGEP1516]], i32 7
23; CHECK-NEXT:    [[TMP8:%.*]] = load i16, i16* [[SCEVGEP17]], align 2
24; CHECK-NEXT:    [[UGLYGEP12:%.*]] = getelementptr i8, i8* undef, i32 undef
25; CHECK-NEXT:    [[UGLYGEP1213:%.*]] = bitcast i8* [[UGLYGEP12]] to i16*
26; CHECK-NEXT:    [[SCEVGEP14:%.*]] = getelementptr i16, i16* [[UGLYGEP1213]], i32 6
27; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i16* [[SCEVGEP14]] to i32*
28; CHECK-NEXT:    [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 2
29; CHECK-NEXT:    [[TMP11:%.*]] = trunc i32 [[TMP10]] to i16
30; CHECK-NEXT:    [[TMP12:%.*]] = call i32 @llvm.arm.smladx(i32 [[TMP10]], i32 [[TMP2]], i32 undef)
31; CHECK-NEXT:    [[TMP13:%.*]] = sext i16 [[TMP11]] to i32
32; CHECK-NEXT:    [[TMP14:%.*]] = lshr i32 [[TMP10]], 16
33; CHECK-NEXT:    [[TMP15:%.*]] = trunc i32 [[TMP14]] to i16
34; CHECK-NEXT:    [[TMP16:%.*]] = sext i16 [[TMP15]] to i32
35; CHECK-NEXT:    [[CONV31:%.*]] = sext i16 [[TMP8]] to i32
36; CHECK-NEXT:    [[TMP17:%.*]] = load i16, i16* [[INCDEC_PTR29]], align 2
37; CHECK-NEXT:    [[CONV33:%.*]] = sext i16 [[TMP17]] to i32
38; CHECK-NEXT:    [[TMP18:%.*]] = load i16, i16* [[SCEVGEP14]], align 2
39; CHECK-NEXT:    [[CONV39:%.*]] = sext i16 [[TMP18]] to i32
40; CHECK-NEXT:    [[MUL_I287_NEG_NEG:%.*]] = mul nsw i32 [[TMP16]], [[TMP4]]
41; CHECK-NEXT:    [[MUL_I283_NEG_NEG:%.*]] = mul nsw i32 [[TMP13]], [[TMP7]]
42; CHECK-NEXT:    [[REASS_ADD408:%.*]] = add i32 undef, [[MUL_I287_NEG_NEG]]
43; CHECK-NEXT:    [[REASS_ADD409:%.*]] = add i32 [[REASS_ADD408]], [[MUL_I283_NEG_NEG]]
44; CHECK-NEXT:    br label [[FOR_BODY]]
45;
46entry:
47  %incdec.ptr21 = getelementptr inbounds i16, i16* %a, i32 3
48  %incdec.ptr29 = getelementptr inbounds i16, i16* %a, i32 4
49  br label %for.body
50
51for.body:
52  %0 = load i16, i16* %incdec.ptr21, align 2
53  %conv25 = sext i16 %0 to i32
54  %uglygep15 = getelementptr i8, i8* undef, i32 undef
55  %uglygep1516 = bitcast i8* %uglygep15 to i16*
56  %scevgep17 = getelementptr i16, i16* %uglygep1516, i32 7
57  %1 = load i16, i16* %scevgep17, align 2
58  %conv31 = sext i16 %1 to i32
59  %2 = load i16, i16* %incdec.ptr29, align 2
60  %conv33 = sext i16 %2 to i32
61  %uglygep12 = getelementptr i8, i8* undef, i32 undef
62  %uglygep1213 = bitcast i8* %uglygep12 to i16*
63  %scevgep14 = getelementptr i16, i16* %uglygep1213, i32 6
64  %3 = load i16, i16* %scevgep14, align 2
65  %conv39 = sext i16 %3 to i32
66  %mul.i287.neg.neg = mul nsw i32 %conv31, %conv25
67  %mul.i283.neg.neg = mul nsw i32 %conv39, %conv33
68  %reass.add408 = add i32 undef, %mul.i287.neg.neg
69  %reass.add409 = add i32 %reass.add408, %mul.i283.neg.neg
70  br label %for.body
71}
72
73define i32 @return(i16* %a, i8* %b, i32 %N) {
74; CHECK-LABEL: @return(
75; CHECK-NEXT:  entry:
76; CHECK-NEXT:    [[INCDEC_PTR21:%.*]] = getelementptr inbounds i16, i16* [[A:%.*]], i32 3
77; CHECK-NEXT:    [[INCDEC_PTR29:%.*]] = getelementptr inbounds i16, i16* [[A]], i32 4
78; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
79; CHECK:       for.body:
80; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[N:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
81; CHECK-NEXT:    [[ACC:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[TMP12:%.*]], [[FOR_BODY]] ]
82; CHECK-NEXT:    [[TMP0:%.*]] = load i16, i16* [[INCDEC_PTR21]], align 2
83; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i16* [[INCDEC_PTR21]] to i32*
84; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 2
85; CHECK-NEXT:    [[TMP3:%.*]] = trunc i32 [[TMP2]] to i16
86; CHECK-NEXT:    [[TMP4:%.*]] = sext i16 [[TMP3]] to i32
87; CHECK-NEXT:    [[TMP5:%.*]] = lshr i32 [[TMP2]], 16
88; CHECK-NEXT:    [[TMP6:%.*]] = trunc i32 [[TMP5]] to i16
89; CHECK-NEXT:    [[TMP7:%.*]] = sext i16 [[TMP6]] to i32
90; CHECK-NEXT:    [[CONV25:%.*]] = sext i16 [[TMP0]] to i32
91; CHECK-NEXT:    [[UGLYGEP15:%.*]] = getelementptr i8, i8* [[B:%.*]], i32 0
92; CHECK-NEXT:    [[UGLYGEP1516:%.*]] = bitcast i8* [[UGLYGEP15]] to i16*
93; CHECK-NEXT:    [[B_IDX:%.*]] = add nuw nsw i32 [[IV]], 1
94; CHECK-NEXT:    [[SCEVGEP17:%.*]] = getelementptr i16, i16* [[UGLYGEP1516]], i32 [[B_IDX]]
95; CHECK-NEXT:    [[SCEVGEP14:%.*]] = getelementptr i16, i16* [[UGLYGEP1516]], i32 [[IV]]
96; CHECK-NEXT:    [[TMP8:%.*]] = load i16, i16* [[SCEVGEP17]], align 2
97; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i16* [[SCEVGEP14]] to i32*
98; CHECK-NEXT:    [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 2
99; CHECK-NEXT:    [[TMP11:%.*]] = trunc i32 [[TMP10]] to i16
100; CHECK-NEXT:    [[TMP12]] = call i32 @llvm.arm.smladx(i32 [[TMP10]], i32 [[TMP2]], i32 [[ACC]])
101; CHECK-NEXT:    [[TMP13:%.*]] = sext i16 [[TMP11]] to i32
102; CHECK-NEXT:    [[TMP14:%.*]] = lshr i32 [[TMP10]], 16
103; CHECK-NEXT:    [[TMP15:%.*]] = trunc i32 [[TMP14]] to i16
104; CHECK-NEXT:    [[TMP16:%.*]] = sext i16 [[TMP15]] to i32
105; CHECK-NEXT:    [[CONV31:%.*]] = sext i16 [[TMP8]] to i32
106; CHECK-NEXT:    [[TMP17:%.*]] = load i16, i16* [[INCDEC_PTR29]], align 2
107; CHECK-NEXT:    [[CONV33:%.*]] = sext i16 [[TMP17]] to i32
108; CHECK-NEXT:    [[TMP18:%.*]] = load i16, i16* [[SCEVGEP14]], align 2
109; CHECK-NEXT:    [[CONV39:%.*]] = sext i16 [[TMP18]] to i32
110; CHECK-NEXT:    [[MUL_I287_NEG_NEG:%.*]] = mul nsw i32 [[TMP16]], [[TMP4]]
111; CHECK-NEXT:    [[MUL_I283_NEG_NEG:%.*]] = mul nsw i32 [[TMP13]], [[TMP7]]
112; CHECK-NEXT:    [[REASS_ADD408:%.*]] = add i32 [[ACC]], [[MUL_I287_NEG_NEG]]
113; CHECK-NEXT:    [[REASS_ADD409:%.*]] = add i32 [[REASS_ADD408]], [[MUL_I283_NEG_NEG]]
114; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], -1
115; CHECK-NEXT:    [[CMP:%.*]] = icmp ne i32 [[IV_NEXT]], 0
116; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[EXIT:%.*]]
117; CHECK:       exit:
118; CHECK-NEXT:    ret i32 [[TMP12]]
119;
120entry:
121  %incdec.ptr21 = getelementptr inbounds i16, i16* %a, i32 3
122  %incdec.ptr29 = getelementptr inbounds i16, i16* %a, i32 4
123  br label %for.body
124
125for.body:
126  %iv = phi i32 [ %N, %entry ], [ %iv.next, %for.body ]
127  %acc = phi i32 [ 0, %entry ], [ %reass.add409, %for.body ]
128  %0 = load i16, i16* %incdec.ptr21, align 2
129  %conv25 = sext i16 %0 to i32
130  %uglygep15 = getelementptr i8, i8* %b, i32 0
131  %uglygep1516 = bitcast i8* %uglygep15 to i16*
132  %b.idx = add nuw nsw i32 %iv, 1
133  %scevgep17 = getelementptr i16, i16* %uglygep1516, i32 %b.idx
134  %scevgep14 = getelementptr i16, i16* %uglygep1516, i32 %iv
135  %1 = load i16, i16* %scevgep17, align 2
136  %conv31 = sext i16 %1 to i32
137  %2 = load i16, i16* %incdec.ptr29, align 2
138  %conv33 = sext i16 %2 to i32
139  %3 = load i16, i16* %scevgep14, align 2
140  %conv39 = sext i16 %3 to i32
141  %mul.i287.neg.neg = mul nsw i32 %conv31, %conv25
142  %mul.i283.neg.neg = mul nsw i32 %conv39, %conv33
143  %reass.add408 = add i32 %acc, %mul.i287.neg.neg
144  %reass.add409 = add i32 %reass.add408, %mul.i283.neg.neg
145  %iv.next = add nuw nsw i32 %iv, -1
146  %cmp = icmp ne i32 %iv.next, 0
147  br i1 %cmp, label %for.body, label %exit
148
149exit:
150  ret i32 %reass.add409
151}
152