aboutsummaryrefslogtreecommitdiff
path: root/llvm/test/CodeGen/WebAssembly/wide-arithmetic.ll
blob: 71974b012a2b62e7e5f0d03120fd3c8621c57d1d (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mattr=+wide-arithmetic < %s | FileCheck %s

target triple = "wasm32-unknown-unknown"

define i128 @add_i128(i128 %a, i128 %b) {
; CHECK-LABEL: add_i128:
; CHECK:         .functype add_i128 (i32, i64, i64, i64, i64) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    local.get 3
; CHECK-NEXT:    local.get 4
; CHECK-NEXT:    i64.add128
; CHECK-NEXT:    local.set 3
; CHECK-NEXT:    local.set 4
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 3
; CHECK-NEXT:    i64.store 8
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 4
; CHECK-NEXT:    i64.store 0
; CHECK-NEXT:    # fallthrough-return
  %c = add i128 %a, %b
  ret i128 %c
}

define i128 @sub_i128(i128 %a, i128 %b) {
; CHECK-LABEL: sub_i128:
; CHECK:         .functype sub_i128 (i32, i64, i64, i64, i64) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    local.get 3
; CHECK-NEXT:    local.get 4
; CHECK-NEXT:    i64.sub128
; CHECK-NEXT:    local.set 3
; CHECK-NEXT:    local.set 4
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 3
; CHECK-NEXT:    i64.store 8
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 4
; CHECK-NEXT:    i64.store 0
; CHECK-NEXT:    # fallthrough-return
  %c = sub i128 %a, %b
  ret i128 %c
}

define i128 @mul_i128(i128 %a, i128 %b) {
; CHECK-LABEL: mul_i128:
; CHECK:         .functype mul_i128 (i32, i64, i64, i64, i64) -> ()
; CHECK-NEXT:    .local i64
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 3
; CHECK-NEXT:    i64.mul_wide_u
; CHECK-NEXT:    local.set 5
; CHECK-NEXT:    i64.store 0
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 5
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 4
; CHECK-NEXT:    i64.mul
; CHECK-NEXT:    i64.add
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    local.get 3
; CHECK-NEXT:    i64.mul
; CHECK-NEXT:    i64.add
; CHECK-NEXT:    i64.store 8
; CHECK-NEXT:    # fallthrough-return
  %c = mul i128 %a, %b
  ret i128 %c
}

define i128 @i64_mul_wide_s(i64 %a, i64 %b) {
; CHECK-LABEL: i64_mul_wide_s:
; CHECK:         .functype i64_mul_wide_s (i32, i64, i64) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    i64.mul_wide_s
; CHECK-NEXT:    local.set 1
; CHECK-NEXT:    local.set 2
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    i64.store 8
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    i64.store 0
; CHECK-NEXT:    # fallthrough-return
  %a128 = sext i64 %a to i128
  %b128 = sext i64 %b to i128
  %c = mul i128 %a128, %b128
  ret i128 %c
}

define i128 @i64_mul_wide_u(i64 %a, i64 %b) {
; CHECK-LABEL: i64_mul_wide_u:
; CHECK:         .functype i64_mul_wide_u (i32, i64, i64) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    i64.mul_wide_u
; CHECK-NEXT:    local.set 1
; CHECK-NEXT:    local.set 2
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    i64.store 8
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    i64.store 0
; CHECK-NEXT:    # fallthrough-return
  %a128 = zext i64 %a to i128
  %b128 = zext i64 %b to i128
  %c = mul i128 %a128, %b128
  ret i128 %c
}

define i64 @mul_i128_only_lo(i128 %a, i128 %b) {
; CHECK-LABEL: mul_i128_only_lo:
; CHECK:         .functype mul_i128_only_lo (i64, i64, i64, i64) -> (i64)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    i64.mul
; CHECK-NEXT:    # fallthrough-return
  %c = mul i128 %a, %b
  %d = trunc i128 %c to i64
  ret i64 %d
}

declare { i64, i1 } @llvm.sadd.with.overflow.i64(i64, i64)
declare { i64, i1 } @llvm.uadd.with.overflow.i64(i64, i64)

; This is a codegen test to see the effect of overflowing adds on signed
; integers with wide-arithmetic enabled. At this time it doesn't actually
; generate anything differently than without wide-arithmetic but this has also
; been useful for evaluating the proposal.
define { i64, i1 } @add_wide_s(i64 %a, i64 %b) {
; CHECK-LABEL: add_wide_s:
; CHECK:         .functype add_wide_s (i32, i64, i64) -> ()
; CHECK-NEXT:    .local i64
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    i64.add
; CHECK-NEXT:    local.tee 3
; CHECK-NEXT:    i64.store 0
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    i64.const 0
; CHECK-NEXT:    i64.lt_s
; CHECK-NEXT:    local.get 3
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    i64.lt_s
; CHECK-NEXT:    i32.xor
; CHECK-NEXT:    i32.store8 8
; CHECK-NEXT:    # fallthrough-return
  %pair = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
  ret { i64, i1 } %pair
}

define { i64, i1 } @add_wide_u(i64 %a, i64 %b) {
; CHECK-LABEL: add_wide_u:
; CHECK:         .functype add_wide_u (i32, i64, i64) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    i64.const 0
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    i64.const 0
; CHECK-NEXT:    i64.add128
; CHECK-NEXT:    local.set 1
; CHECK-NEXT:    local.set 2
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    i64.store8 8
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    i64.store 0
; CHECK-NEXT:    # fallthrough-return
  %pair = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
  ret { i64, i1 } %pair
}

; This is a model of a hypothetical `i64.add_wide3_u` instruction using LLVM
; intrinsics. In theory this should optimize better (to the equivalent below)
; but it doesn't currently.
define { i64, i64 } @add_wide3_u_via_intrinsics(i64 %a, i64 %b, i64 %c) {
; CHECK-LABEL: add_wide3_u_via_intrinsics:
; CHECK:         .functype add_wide3_u_via_intrinsics (i32, i64, i64, i64) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    i64.const 0
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    i64.const 0
; CHECK-NEXT:    i64.add128
; CHECK-NEXT:    local.set 2
; CHECK-NEXT:    i64.const 0
; CHECK-NEXT:    local.get 3
; CHECK-NEXT:    i64.const 0
; CHECK-NEXT:    i64.add128
; CHECK-NEXT:    local.set 1
; CHECK-NEXT:    i64.store 0
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    i64.add
; CHECK-NEXT:    i64.store 8
; CHECK-NEXT:    # fallthrough-return
  %pair = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
  %t0 = extractvalue { i64, i1 } %pair, 0
  %carry1 = extractvalue { i64, i1 } %pair, 1

  %pair2 = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %t0, i64 %c)
  %ret1 = extractvalue { i64, i1 } %pair2, 0
  %carry2 = extractvalue { i64, i1 } %pair2, 1

  %carry1_64 = zext i1 %carry1 to i64
  %carry2_64 = zext i1 %carry2 to i64
  %ret2 = add i64 %carry1_64, %carry2_64

  %r0 = insertvalue { i64, i64 } poison, i64 %ret1, 0
  %r1 = insertvalue { i64, i64 } %r0, i64 %ret2, 1
  ret { i64, i64 } %r1
}

; This is a model of a hypothetical `i64.add_wide3_u` instruction using 128-bit
; integer addition. This optimizes better than the above currently.
define { i64, i64 } @add_wide3_u_via_i128(i64 %a, i64 %b, i64 %c) {
; CHECK-LABEL: add_wide3_u_via_i128:
; CHECK:         .functype add_wide3_u_via_i128 (i32, i64, i64, i64) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    i64.const 0
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    i64.const 0
; CHECK-NEXT:    i64.add128
; CHECK-NEXT:    local.get 3
; CHECK-NEXT:    i64.const 0
; CHECK-NEXT:    i64.add128
; CHECK-NEXT:    local.set 1
; CHECK-NEXT:    local.set 2
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    i64.store 8
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 2
; CHECK-NEXT:    i64.store 0
; CHECK-NEXT:    # fallthrough-return
  %a128 = zext i64 %a to i128
  %b128 = zext i64 %b to i128
  %c128 = zext i64 %c to i128
  %t0 = add i128 %a128, %b128
  %t1 = add i128 %t0, %c128
  %result = trunc i128 %t1 to i64
  %t2 = lshr i128 %t1, 64
  %carry = trunc i128 %t2 to i64

  %ret0 = insertvalue { i64, i64 } poison, i64 %result, 0
  %ret1 = insertvalue { i64, i64 } %ret0, i64 %carry, 1
  ret { i64, i64 } %ret1
}