xref: /llvm-project/llvm/test/Transforms/InstCombine/ssub-with-overflow.ll (revision 38fffa630ee80163dc65e759392ad29798905679) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes=instcombine -S | FileCheck %s
3
4declare { <2 x i32>, <2 x i1> } @llvm.ssub.with.overflow.v2i32(<2 x i32>, <2 x i32>)
5
6declare { <2 x i8>, <2 x i1> } @llvm.ssub.with.overflow.v2i8(<2 x i8>, <2 x i8>)
7
8declare { i32, i1 } @llvm.ssub.with.overflow.i32(i32, i32)
9
10declare { i8, i1 } @llvm.ssub.with.overflow.i8(i8, i8)
11
12define { i32, i1 } @simple_fold(i32 %x) {
13; CHECK-LABEL: @simple_fold(
14; CHECK-NEXT:    [[B:%.*]] = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 [[X:%.*]], i32 -20)
15; CHECK-NEXT:    ret { i32, i1 } [[B]]
16;
17  %a = sub nsw i32 %x, 7
18  %b = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 %a, i32 13)
19  ret { i32, i1 } %b
20}
21
22define { i32, i1 } @fold_mixed_signs(i32 %x) {
23; CHECK-LABEL: @fold_mixed_signs(
24; CHECK-NEXT:    [[B:%.*]] = add nsw i32 [[X:%.*]], -6
25; CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { i32, i1 } { i32 poison, i1 false }, i32 [[B]], 0
26; CHECK-NEXT:    ret { i32, i1 } [[TMP1]]
27;
28  %a = sub nsw i32 %x, 13
29  %b = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 %a, i32 -7)
30  ret { i32, i1 } %b
31}
32
33define { i8, i1 } @fold_on_constant_sub_no_overflow(i8 %x) {
34; CHECK-LABEL: @fold_on_constant_sub_no_overflow(
35; CHECK-NEXT:    [[B:%.*]] = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[X:%.*]], i8 -128)
36; CHECK-NEXT:    ret { i8, i1 } [[B]]
37;
38  %a = sub nsw i8 %x, 100
39  %b = tail call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 %a, i8 28)
40  ret { i8, i1 } %b
41}
42
43define { i8, i1 } @no_fold_on_constant_sub_overflow(i8 %x) {
44; CHECK-LABEL: @no_fold_on_constant_sub_overflow(
45; CHECK-NEXT:    [[A:%.*]] = add nsw i8 [[X:%.*]], -100
46; CHECK-NEXT:    [[B:%.*]] = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[A]], i8 -29)
47; CHECK-NEXT:    ret { i8, i1 } [[B]]
48;
49  %a = sub nsw i8 %x, 100
50  %b = tail call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 %a, i8 29)
51  ret { i8, i1 } %b
52}
53
54define { <2 x i32>, <2 x i1> } @fold_simple_splat_constant(<2 x i32> %x) {
55; CHECK-LABEL: @fold_simple_splat_constant(
56; CHECK-NEXT:    [[B:%.*]] = call { <2 x i32>, <2 x i1> } @llvm.sadd.with.overflow.v2i32(<2 x i32> [[X:%.*]], <2 x i32> splat (i32 -42))
57; CHECK-NEXT:    ret { <2 x i32>, <2 x i1> } [[B]]
58;
59  %a = sub nsw <2 x i32> %x, <i32 12, i32 12>
60  %b = tail call { <2 x i32>, <2 x i1> } @llvm.ssub.with.overflow.v2i32(<2 x i32> %a, <2 x i32> <i32 30, i32 30>)
61  ret { <2 x i32>, <2 x i1> } %b
62}
63
64define { <2 x i32>, <2 x i1> } @no_fold_splat_undef_constant(<2 x i32> %x) {
65; CHECK-LABEL: @no_fold_splat_undef_constant(
66; CHECK-NEXT:    [[A:%.*]] = add <2 x i32> [[X:%.*]], <i32 -12, i32 undef>
67; CHECK-NEXT:    [[B:%.*]] = call { <2 x i32>, <2 x i1> } @llvm.sadd.with.overflow.v2i32(<2 x i32> [[A]], <2 x i32> splat (i32 -30))
68; CHECK-NEXT:    ret { <2 x i32>, <2 x i1> } [[B]]
69;
70  %a = sub nsw <2 x i32> %x, <i32 12, i32 undef>
71  %b = tail call { <2 x i32>, <2 x i1> } @llvm.ssub.with.overflow.v2i32(<2 x i32> %a, <2 x i32> <i32 30, i32 30>)
72  ret { <2 x i32>, <2 x i1> } %b
73}
74
75define { <2 x i32>, <2 x i1> } @no_fold_splat_not_constant(<2 x i32> %x, <2 x i32> %y) {
76; CHECK-LABEL: @no_fold_splat_not_constant(
77; CHECK-NEXT:    [[A:%.*]] = sub nsw <2 x i32> [[X:%.*]], [[Y:%.*]]
78; CHECK-NEXT:    [[B:%.*]] = call { <2 x i32>, <2 x i1> } @llvm.sadd.with.overflow.v2i32(<2 x i32> [[A]], <2 x i32> splat (i32 -30))
79; CHECK-NEXT:    ret { <2 x i32>, <2 x i1> } [[B]]
80;
81  %a = sub nsw <2 x i32> %x, %y
82  %b = tail call { <2 x i32>, <2 x i1> } @llvm.ssub.with.overflow.v2i32(<2 x i32> %a, <2 x i32> <i32 30, i32 30>)
83  ret { <2 x i32>, <2 x i1> } %b
84}
85
86define { i32, i1 } @fold_nuwnsw(i32 %x) {
87; CHECK-LABEL: @fold_nuwnsw(
88; CHECK-NEXT:    [[B:%.*]] = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 [[X:%.*]], i32 -42)
89; CHECK-NEXT:    ret { i32, i1 } [[B]]
90;
91  %a = sub nuw nsw i32 %x, 12
92  %b = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 %a, i32 30)
93  ret { i32, i1 } %b
94}
95
96define { i32, i1 } @no_fold_nuw(i32 %x) {
97; CHECK-LABEL: @no_fold_nuw(
98; CHECK-NEXT:    [[A:%.*]] = add i32 [[X:%.*]], -12
99; CHECK-NEXT:    [[B:%.*]] = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 [[A]], i32 -30)
100; CHECK-NEXT:    ret { i32, i1 } [[B]]
101;
102  %a = sub nuw i32 %x, 12
103  %b = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 %a, i32 30)
104  ret { i32, i1 } %b
105}
106
107define { i32, i1 } @no_fold_wrapped_sub(i32 %x) {
108; CHECK-LABEL: @no_fold_wrapped_sub(
109; CHECK-NEXT:    [[A:%.*]] = add i32 [[X:%.*]], -12
110; CHECK-NEXT:    [[B:%.*]] = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 30, i32 [[A]])
111; CHECK-NEXT:    ret { i32, i1 } [[B]]
112;
113  %a = sub i32 %x, 12
114  %b = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 30, i32 %a)
115  ret { i32, i1 } %b
116}
117
118define { i32, i1 } @fold_add_simple(i32 %x) {
119; CHECK-LABEL: @fold_add_simple(
120; CHECK-NEXT:    [[B:%.*]] = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 [[X:%.*]], i32 -42)
121; CHECK-NEXT:    ret { i32, i1 } [[B]]
122;
123  %a = add nsw i32 %x, -12
124  %b = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 %a, i32 30)
125  ret { i32, i1 } %b
126}
127
128define { <2 x i32>, <2 x i1> } @keep_ssubo_undef(<2 x i32> %x) {
129; CHECK-LABEL: @keep_ssubo_undef(
130; CHECK-NEXT:    [[A:%.*]] = tail call { <2 x i32>, <2 x i1> } @llvm.ssub.with.overflow.v2i32(<2 x i32> [[X:%.*]], <2 x i32> <i32 30, i32 undef>)
131; CHECK-NEXT:    ret { <2 x i32>, <2 x i1> } [[A]]
132;
133  %a = tail call { <2 x i32>, <2 x i1> } @llvm.ssub.with.overflow.v2i32(<2 x i32> %x, <2 x i32> <i32 30, i32 undef>)
134  ret { <2 x i32>, <2 x i1> } %a
135}
136
137define { <2 x i32>, <2 x i1> } @keep_ssubo_non_splat(<2 x i32> %x) {
138; CHECK-LABEL: @keep_ssubo_non_splat(
139; CHECK-NEXT:    [[A:%.*]] = call { <2 x i32>, <2 x i1> } @llvm.sadd.with.overflow.v2i32(<2 x i32> [[X:%.*]], <2 x i32> <i32 -30, i32 -31>)
140; CHECK-NEXT:    ret { <2 x i32>, <2 x i1> } [[A]]
141;
142  %a = tail call { <2 x i32>, <2 x i1> } @llvm.ssub.with.overflow.v2i32(<2 x i32> %x, <2 x i32> <i32 30, i32 31>)
143  ret { <2 x i32>, <2 x i1> } %a
144}
145
146define { <2 x i8>, <2 x i1> } @keep_ssubo_one_element_is_128(<2 x i8> %x) {
147; CHECK-LABEL: @keep_ssubo_one_element_is_128(
148; CHECK-NEXT:    [[A:%.*]] = tail call { <2 x i8>, <2 x i1> } @llvm.ssub.with.overflow.v2i8(<2 x i8> [[X:%.*]], <2 x i8> <i8 0, i8 -128>)
149; CHECK-NEXT:    ret { <2 x i8>, <2 x i1> } [[A]]
150;
151  %a = tail call { <2 x i8>, <2 x i1> } @llvm.ssub.with.overflow.v2i8(<2 x i8> %x, <2 x i8> <i8 0, i8 -128>)
152  ret { <2 x i8>, <2 x i1> } %a
153}
154
155define { i8, i1 } @keep_ssubo_128(i8 %x) {
156; CHECK-LABEL: @keep_ssubo_128(
157; CHECK-NEXT:    [[A:%.*]] = tail call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 [[X:%.*]], i8 -128)
158; CHECK-NEXT:    ret { i8, i1 } [[A]]
159;
160  %a = tail call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 %x, i8 -128)
161  ret { i8, i1 } %a
162}
163