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
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
|
; RUN: llc < %s -mtriple=i686-- -mattr=+mmx,+sse2 | FileCheck -check-prefix=X32 %s
; RUN: llc < %s -mtriple=x86_64-- -mattr=+mmx,+sse2 | FileCheck -check-prefix=X64 %s
;; A basic sanity check to make sure that MMX arithmetic actually compiles.
;; First is a straight translation of the original with bitcasts as needed.
; X32-LABEL: test0
; X64-LABEL: test0
define void @test0(x86_mmx* %A, x86_mmx* %B) {
entry:
%tmp1 = load x86_mmx, x86_mmx* %A
%tmp3 = load x86_mmx, x86_mmx* %B
%tmp1a = bitcast x86_mmx %tmp1 to <8 x i8>
%tmp3a = bitcast x86_mmx %tmp3 to <8 x i8>
%tmp4 = add <8 x i8> %tmp1a, %tmp3a
%tmp4a = bitcast <8 x i8> %tmp4 to x86_mmx
store x86_mmx %tmp4a, x86_mmx* %A
%tmp7 = load x86_mmx, x86_mmx* %B
%tmp12 = tail call x86_mmx @llvm.x86.mmx.padds.b(x86_mmx %tmp4a, x86_mmx %tmp7)
store x86_mmx %tmp12, x86_mmx* %A
%tmp16 = load x86_mmx, x86_mmx* %B
%tmp21 = tail call x86_mmx @llvm.x86.mmx.paddus.b(x86_mmx %tmp12, x86_mmx %tmp16)
store x86_mmx %tmp21, x86_mmx* %A
%tmp27 = load x86_mmx, x86_mmx* %B
%tmp21a = bitcast x86_mmx %tmp21 to <8 x i8>
%tmp27a = bitcast x86_mmx %tmp27 to <8 x i8>
%tmp28 = sub <8 x i8> %tmp21a, %tmp27a
%tmp28a = bitcast <8 x i8> %tmp28 to x86_mmx
store x86_mmx %tmp28a, x86_mmx* %A
%tmp31 = load x86_mmx, x86_mmx* %B
%tmp36 = tail call x86_mmx @llvm.x86.mmx.psubs.b(x86_mmx %tmp28a, x86_mmx %tmp31)
store x86_mmx %tmp36, x86_mmx* %A
%tmp40 = load x86_mmx, x86_mmx* %B
%tmp45 = tail call x86_mmx @llvm.x86.mmx.psubus.b(x86_mmx %tmp36, x86_mmx %tmp40)
store x86_mmx %tmp45, x86_mmx* %A
%tmp51 = load x86_mmx, x86_mmx* %B
%tmp45a = bitcast x86_mmx %tmp45 to <8 x i8>
%tmp51a = bitcast x86_mmx %tmp51 to <8 x i8>
%tmp52 = mul <8 x i8> %tmp45a, %tmp51a
%tmp52a = bitcast <8 x i8> %tmp52 to x86_mmx
store x86_mmx %tmp52a, x86_mmx* %A
%tmp57 = load x86_mmx, x86_mmx* %B
%tmp57a = bitcast x86_mmx %tmp57 to <8 x i8>
%tmp58 = and <8 x i8> %tmp52, %tmp57a
%tmp58a = bitcast <8 x i8> %tmp58 to x86_mmx
store x86_mmx %tmp58a, x86_mmx* %A
%tmp63 = load x86_mmx, x86_mmx* %B
%tmp63a = bitcast x86_mmx %tmp63 to <8 x i8>
%tmp64 = or <8 x i8> %tmp58, %tmp63a
%tmp64a = bitcast <8 x i8> %tmp64 to x86_mmx
store x86_mmx %tmp64a, x86_mmx* %A
%tmp69 = load x86_mmx, x86_mmx* %B
%tmp69a = bitcast x86_mmx %tmp69 to <8 x i8>
%tmp64b = bitcast x86_mmx %tmp64a to <8 x i8>
%tmp70 = xor <8 x i8> %tmp64b, %tmp69a
%tmp70a = bitcast <8 x i8> %tmp70 to x86_mmx
store x86_mmx %tmp70a, x86_mmx* %A
tail call void @llvm.x86.mmx.emms()
ret void
}
; X32-LABEL: test1
; X64-LABEL: test1
define void @test1(x86_mmx* %A, x86_mmx* %B) {
entry:
%tmp1 = load x86_mmx, x86_mmx* %A
%tmp3 = load x86_mmx, x86_mmx* %B
%tmp1a = bitcast x86_mmx %tmp1 to <2 x i32>
%tmp3a = bitcast x86_mmx %tmp3 to <2 x i32>
%tmp4 = add <2 x i32> %tmp1a, %tmp3a
%tmp4a = bitcast <2 x i32> %tmp4 to x86_mmx
store x86_mmx %tmp4a, x86_mmx* %A
%tmp9 = load x86_mmx, x86_mmx* %B
%tmp9a = bitcast x86_mmx %tmp9 to <2 x i32>
%tmp10 = sub <2 x i32> %tmp4, %tmp9a
%tmp10a = bitcast <2 x i32> %tmp4 to x86_mmx
store x86_mmx %tmp10a, x86_mmx* %A
%tmp15 = load x86_mmx, x86_mmx* %B
%tmp10b = bitcast x86_mmx %tmp10a to <2 x i32>
%tmp15a = bitcast x86_mmx %tmp15 to <2 x i32>
%tmp16 = mul <2 x i32> %tmp10b, %tmp15a
%tmp16a = bitcast <2 x i32> %tmp16 to x86_mmx
store x86_mmx %tmp16a, x86_mmx* %A
%tmp21 = load x86_mmx, x86_mmx* %B
%tmp16b = bitcast x86_mmx %tmp16a to <2 x i32>
%tmp21a = bitcast x86_mmx %tmp21 to <2 x i32>
%tmp22 = and <2 x i32> %tmp16b, %tmp21a
%tmp22a = bitcast <2 x i32> %tmp22 to x86_mmx
store x86_mmx %tmp22a, x86_mmx* %A
%tmp27 = load x86_mmx, x86_mmx* %B
%tmp22b = bitcast x86_mmx %tmp22a to <2 x i32>
%tmp27a = bitcast x86_mmx %tmp27 to <2 x i32>
%tmp28 = or <2 x i32> %tmp22b, %tmp27a
%tmp28a = bitcast <2 x i32> %tmp28 to x86_mmx
store x86_mmx %tmp28a, x86_mmx* %A
%tmp33 = load x86_mmx, x86_mmx* %B
%tmp28b = bitcast x86_mmx %tmp28a to <2 x i32>
%tmp33a = bitcast x86_mmx %tmp33 to <2 x i32>
%tmp34 = xor <2 x i32> %tmp28b, %tmp33a
%tmp34a = bitcast <2 x i32> %tmp34 to x86_mmx
store x86_mmx %tmp34a, x86_mmx* %A
tail call void @llvm.x86.mmx.emms( )
ret void
}
; X32-LABEL: test2
; X64-LABEL: test2
define void @test2(x86_mmx* %A, x86_mmx* %B) {
entry:
%tmp1 = load x86_mmx, x86_mmx* %A
%tmp3 = load x86_mmx, x86_mmx* %B
%tmp1a = bitcast x86_mmx %tmp1 to <4 x i16>
%tmp3a = bitcast x86_mmx %tmp3 to <4 x i16>
%tmp4 = add <4 x i16> %tmp1a, %tmp3a
%tmp4a = bitcast <4 x i16> %tmp4 to x86_mmx
store x86_mmx %tmp4a, x86_mmx* %A
%tmp7 = load x86_mmx, x86_mmx* %B
%tmp12 = tail call x86_mmx @llvm.x86.mmx.padds.w(x86_mmx %tmp4a, x86_mmx %tmp7)
store x86_mmx %tmp12, x86_mmx* %A
%tmp16 = load x86_mmx, x86_mmx* %B
%tmp21 = tail call x86_mmx @llvm.x86.mmx.paddus.w(x86_mmx %tmp12, x86_mmx %tmp16)
store x86_mmx %tmp21, x86_mmx* %A
%tmp27 = load x86_mmx, x86_mmx* %B
%tmp21a = bitcast x86_mmx %tmp21 to <4 x i16>
%tmp27a = bitcast x86_mmx %tmp27 to <4 x i16>
%tmp28 = sub <4 x i16> %tmp21a, %tmp27a
%tmp28a = bitcast <4 x i16> %tmp28 to x86_mmx
store x86_mmx %tmp28a, x86_mmx* %A
%tmp31 = load x86_mmx, x86_mmx* %B
%tmp36 = tail call x86_mmx @llvm.x86.mmx.psubs.w(x86_mmx %tmp28a, x86_mmx %tmp31)
store x86_mmx %tmp36, x86_mmx* %A
%tmp40 = load x86_mmx, x86_mmx* %B
%tmp45 = tail call x86_mmx @llvm.x86.mmx.psubus.w(x86_mmx %tmp36, x86_mmx %tmp40)
store x86_mmx %tmp45, x86_mmx* %A
%tmp51 = load x86_mmx, x86_mmx* %B
%tmp45a = bitcast x86_mmx %tmp45 to <4 x i16>
%tmp51a = bitcast x86_mmx %tmp51 to <4 x i16>
%tmp52 = mul <4 x i16> %tmp45a, %tmp51a
%tmp52a = bitcast <4 x i16> %tmp52 to x86_mmx
store x86_mmx %tmp52a, x86_mmx* %A
%tmp55 = load x86_mmx, x86_mmx* %B
%tmp60 = tail call x86_mmx @llvm.x86.mmx.pmulh.w(x86_mmx %tmp52a, x86_mmx %tmp55)
store x86_mmx %tmp60, x86_mmx* %A
%tmp64 = load x86_mmx, x86_mmx* %B
%tmp69 = tail call x86_mmx @llvm.x86.mmx.pmadd.wd(x86_mmx %tmp60, x86_mmx %tmp64)
%tmp70 = bitcast x86_mmx %tmp69 to x86_mmx
store x86_mmx %tmp70, x86_mmx* %A
%tmp75 = load x86_mmx, x86_mmx* %B
%tmp70a = bitcast x86_mmx %tmp70 to <4 x i16>
%tmp75a = bitcast x86_mmx %tmp75 to <4 x i16>
%tmp76 = and <4 x i16> %tmp70a, %tmp75a
%tmp76a = bitcast <4 x i16> %tmp76 to x86_mmx
store x86_mmx %tmp76a, x86_mmx* %A
%tmp81 = load x86_mmx, x86_mmx* %B
%tmp76b = bitcast x86_mmx %tmp76a to <4 x i16>
%tmp81a = bitcast x86_mmx %tmp81 to <4 x i16>
%tmp82 = or <4 x i16> %tmp76b, %tmp81a
%tmp82a = bitcast <4 x i16> %tmp82 to x86_mmx
store x86_mmx %tmp82a, x86_mmx* %A
%tmp87 = load x86_mmx, x86_mmx* %B
%tmp82b = bitcast x86_mmx %tmp82a to <4 x i16>
%tmp87a = bitcast x86_mmx %tmp87 to <4 x i16>
%tmp88 = xor <4 x i16> %tmp82b, %tmp87a
%tmp88a = bitcast <4 x i16> %tmp88 to x86_mmx
store x86_mmx %tmp88a, x86_mmx* %A
tail call void @llvm.x86.mmx.emms( )
ret void
}
; X32-LABEL: test3
define <1 x i64> @test3(<1 x i64>* %a, <1 x i64>* %b, i32 %count) nounwind {
entry:
%tmp2942 = icmp eq i32 %count, 0
br i1 %tmp2942, label %bb31, label %bb26
bb26:
; X32: addl
; X32: adcl
%i.037.0 = phi i32 [ 0, %entry ], [ %tmp25, %bb26 ]
%sum.035.0 = phi <1 x i64> [ zeroinitializer, %entry ], [ %tmp22, %bb26 ]
%tmp13 = getelementptr <1 x i64>, <1 x i64>* %b, i32 %i.037.0
%tmp14 = load <1 x i64>, <1 x i64>* %tmp13
%tmp18 = getelementptr <1 x i64>, <1 x i64>* %a, i32 %i.037.0
%tmp19 = load <1 x i64>, <1 x i64>* %tmp18
%tmp21 = add <1 x i64> %tmp19, %tmp14
%tmp22 = add <1 x i64> %tmp21, %sum.035.0
%tmp25 = add i32 %i.037.0, 1
%tmp29 = icmp ult i32 %tmp25, %count
br i1 %tmp29, label %bb26, label %bb31
bb31:
%sum.035.1 = phi <1 x i64> [ zeroinitializer, %entry ], [ %tmp22, %bb26 ]
ret <1 x i64> %sum.035.1
}
; There are no MMX operations here, so we use XMM or i64.
; X64-LABEL: ti8
define void @ti8(double %a, double %b) nounwind {
entry:
%tmp1 = bitcast double %a to <8 x i8>
%tmp2 = bitcast double %b to <8 x i8>
%tmp3 = add <8 x i8> %tmp1, %tmp2
; X64: paddb
store <8 x i8> %tmp3, <8 x i8>* null
ret void
}
; X64-LABEL: ti16
define void @ti16(double %a, double %b) nounwind {
entry:
%tmp1 = bitcast double %a to <4 x i16>
%tmp2 = bitcast double %b to <4 x i16>
%tmp3 = add <4 x i16> %tmp1, %tmp2
; X64: paddw
store <4 x i16> %tmp3, <4 x i16>* null
ret void
}
; X64-LABEL: ti32
define void @ti32(double %a, double %b) nounwind {
entry:
%tmp1 = bitcast double %a to <2 x i32>
%tmp2 = bitcast double %b to <2 x i32>
%tmp3 = add <2 x i32> %tmp1, %tmp2
; X64: paddd
store <2 x i32> %tmp3, <2 x i32>* null
ret void
}
; X64-LABEL: ti64
define void @ti64(double %a, double %b) nounwind {
entry:
%tmp1 = bitcast double %a to <1 x i64>
%tmp2 = bitcast double %b to <1 x i64>
%tmp3 = add <1 x i64> %tmp1, %tmp2
; X64: addq
store <1 x i64> %tmp3, <1 x i64>* null
ret void
}
; MMX intrinsics calls get us MMX instructions.
; X64-LABEL: ti8a
define void @ti8a(double %a, double %b) nounwind {
entry:
%tmp1 = bitcast double %a to x86_mmx
; X64: movdq2q
%tmp2 = bitcast double %b to x86_mmx
; X64: movdq2q
%tmp3 = tail call x86_mmx @llvm.x86.mmx.padd.b(x86_mmx %tmp1, x86_mmx %tmp2)
store x86_mmx %tmp3, x86_mmx* null
ret void
}
; X64-LABEL: ti16a
define void @ti16a(double %a, double %b) nounwind {
entry:
%tmp1 = bitcast double %a to x86_mmx
; X64: movdq2q
%tmp2 = bitcast double %b to x86_mmx
; X64: movdq2q
%tmp3 = tail call x86_mmx @llvm.x86.mmx.padd.w(x86_mmx %tmp1, x86_mmx %tmp2)
store x86_mmx %tmp3, x86_mmx* null
ret void
}
; X64-LABEL: ti32a
define void @ti32a(double %a, double %b) nounwind {
entry:
%tmp1 = bitcast double %a to x86_mmx
; X64: movdq2q
%tmp2 = bitcast double %b to x86_mmx
; X64: movdq2q
%tmp3 = tail call x86_mmx @llvm.x86.mmx.padd.d(x86_mmx %tmp1, x86_mmx %tmp2)
store x86_mmx %tmp3, x86_mmx* null
ret void
}
; X64-LABEL: ti64a
define void @ti64a(double %a, double %b) nounwind {
entry:
%tmp1 = bitcast double %a to x86_mmx
; X64: movdq2q
%tmp2 = bitcast double %b to x86_mmx
; X64: movdq2q
%tmp3 = tail call x86_mmx @llvm.x86.mmx.padd.q(x86_mmx %tmp1, x86_mmx %tmp2)
store x86_mmx %tmp3, x86_mmx* null
ret void
}
declare x86_mmx @llvm.x86.mmx.padd.b(x86_mmx, x86_mmx)
declare x86_mmx @llvm.x86.mmx.padd.w(x86_mmx, x86_mmx)
declare x86_mmx @llvm.x86.mmx.padd.d(x86_mmx, x86_mmx)
declare x86_mmx @llvm.x86.mmx.padd.q(x86_mmx, x86_mmx)
declare x86_mmx @llvm.x86.mmx.paddus.b(x86_mmx, x86_mmx)
declare x86_mmx @llvm.x86.mmx.psubus.b(x86_mmx, x86_mmx)
declare x86_mmx @llvm.x86.mmx.paddus.w(x86_mmx, x86_mmx)
declare x86_mmx @llvm.x86.mmx.psubus.w(x86_mmx, x86_mmx)
declare x86_mmx @llvm.x86.mmx.pmulh.w(x86_mmx, x86_mmx)
declare x86_mmx @llvm.x86.mmx.pmadd.wd(x86_mmx, x86_mmx)
declare void @llvm.x86.mmx.emms()
declare x86_mmx @llvm.x86.mmx.padds.b(x86_mmx, x86_mmx)
declare x86_mmx @llvm.x86.mmx.padds.w(x86_mmx, x86_mmx)
declare x86_mmx @llvm.x86.mmx.psubs.b(x86_mmx, x86_mmx)
declare x86_mmx @llvm.x86.mmx.psubs.w(x86_mmx, x86_mmx)
|
