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
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
|
; RUN: opt < %s -basicaa -slp-vectorizer -slp-threshold=-100 -instcombine -dce -S -mtriple=i386-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32-S128"
; Make sure we order the operands of commutative operations so that we get
; bigger vectorizable trees.
; CHECK-LABEL: shuffle_operands1
; CHECK: load <2 x double>
; CHECK: fadd <2 x double>
define void @shuffle_operands1(double * noalias %from, double * noalias %to,
double %v1, double %v2) {
%from_1 = getelementptr double *%from, i64 1
%v0_1 = load double * %from
%v0_2 = load double * %from_1
%v1_1 = fadd double %v0_1, %v1
%v1_2 = fadd double %v2, %v0_2
%to_2 = getelementptr double * %to, i64 1
store double %v1_1, double *%to
store double %v1_2, double *%to_2
ret void
}
; CHECK-LABEL: shuffle_preserve_broadcast
; CHECK: %[[BCAST:[a-z0-9]+]] = insertelement <2 x double> undef, double %v0_1
; CHECK: = insertelement <2 x double> %[[BCAST]], double %v0_1
define void @shuffle_preserve_broadcast(double * noalias %from,
double * noalias %to,
double %v1, double %v2) {
entry:
br label %lp
lp:
%p = phi double [ 1.000000e+00, %lp ], [ 0.000000e+00, %entry ]
%from_1 = getelementptr double *%from, i64 1
%v0_1 = load double * %from
%v0_2 = load double * %from_1
%v1_1 = fadd double %v0_1, %p
%v1_2 = fadd double %v0_1, %v0_2
%to_2 = getelementptr double * %to, i64 1
store double %v1_1, double *%to
store double %v1_2, double *%to_2
br i1 undef, label %lp, label %ext
ext:
ret void
}
; CHECK-LABEL: shuffle_preserve_broadcast2
; CHECK: %[[BCAST:[a-z0-9]+]] = insertelement <2 x double> undef, double %v0_1
; CHECK: = insertelement <2 x double> %[[BCAST]], double %v0_1
define void @shuffle_preserve_broadcast2(double * noalias %from,
double * noalias %to,
double %v1, double %v2) {
entry:
br label %lp
lp:
%p = phi double [ 1.000000e+00, %lp ], [ 0.000000e+00, %entry ]
%from_1 = getelementptr double *%from, i64 1
%v0_1 = load double * %from
%v0_2 = load double * %from_1
%v1_1 = fadd double %p, %v0_1
%v1_2 = fadd double %v0_2, %v0_1
%to_2 = getelementptr double * %to, i64 1
store double %v1_1, double *%to
store double %v1_2, double *%to_2
br i1 undef, label %lp, label %ext
ext:
ret void
}
; CHECK-LABEL: shuffle_preserve_broadcast3
; CHECK: %[[BCAST:[a-z0-9]+]] = insertelement <2 x double> undef, double %v0_1
; CHECK: = insertelement <2 x double> %[[BCAST]], double %v0_1
define void @shuffle_preserve_broadcast3(double * noalias %from,
double * noalias %to,
double %v1, double %v2) {
entry:
br label %lp
lp:
%p = phi double [ 1.000000e+00, %lp ], [ 0.000000e+00, %entry ]
%from_1 = getelementptr double *%from, i64 1
%v0_1 = load double * %from
%v0_2 = load double * %from_1
%v1_1 = fadd double %p, %v0_1
%v1_2 = fadd double %v0_1, %v0_2
%to_2 = getelementptr double * %to, i64 1
store double %v1_1, double *%to
store double %v1_2, double *%to_2
br i1 undef, label %lp, label %ext
ext:
ret void
}
; CHECK-LABEL: shuffle_preserve_broadcast4
; CHECK: %[[BCAST:[a-z0-9]+]] = insertelement <2 x double> undef, double %v0_1
; CHECK: = insertelement <2 x double> %[[BCAST]], double %v0_1
define void @shuffle_preserve_broadcast4(double * noalias %from,
double * noalias %to,
double %v1, double %v2) {
entry:
br label %lp
lp:
%p = phi double [ 1.000000e+00, %lp ], [ 0.000000e+00, %entry ]
%from_1 = getelementptr double *%from, i64 1
%v0_1 = load double * %from
%v0_2 = load double * %from_1
%v1_1 = fadd double %v0_2, %v0_1
%v1_2 = fadd double %p, %v0_1
%to_2 = getelementptr double * %to, i64 1
store double %v1_1, double *%to
store double %v1_2, double *%to_2
br i1 undef, label %lp, label %ext
ext:
ret void
}
; CHECK-LABEL: shuffle_preserve_broadcast5
; CHECK: %[[BCAST:[a-z0-9]+]] = insertelement <2 x double> undef, double %v0_1
; CHECK: = insertelement <2 x double> %[[BCAST]], double %v0_1
define void @shuffle_preserve_broadcast5(double * noalias %from,
double * noalias %to,
double %v1, double %v2) {
entry:
br label %lp
lp:
%p = phi double [ 1.000000e+00, %lp ], [ 0.000000e+00, %entry ]
%from_1 = getelementptr double *%from, i64 1
%v0_1 = load double * %from
%v0_2 = load double * %from_1
%v1_1 = fadd double %v0_1, %v0_2
%v1_2 = fadd double %p, %v0_1
%to_2 = getelementptr double * %to, i64 1
store double %v1_1, double *%to
store double %v1_2, double *%to_2
br i1 undef, label %lp, label %ext
ext:
ret void
}
; CHECK-LABEL: shuffle_preserve_broadcast6
; CHECK: %[[BCAST:[a-z0-9]+]] = insertelement <2 x double> undef, double %v0_1
; CHECK: = insertelement <2 x double> %[[BCAST]], double %v0_1
define void @shuffle_preserve_broadcast6(double * noalias %from,
double * noalias %to,
double %v1, double %v2) {
entry:
br label %lp
lp:
%p = phi double [ 1.000000e+00, %lp ], [ 0.000000e+00, %entry ]
%from_1 = getelementptr double *%from, i64 1
%v0_1 = load double * %from
%v0_2 = load double * %from_1
%v1_1 = fadd double %v0_1, %v0_2
%v1_2 = fadd double %v0_1, %p
%to_2 = getelementptr double * %to, i64 1
store double %v1_1, double *%to
store double %v1_2, double *%to_2
br i1 undef, label %lp, label %ext
ext:
ret void
}
; Make sure we don't scramble operands when we reorder them and destroy
; 'good' source order.
; CHECK-LABEL: good_load_order
; CHECK: %[[V1:[0-9]+]] = load <4 x float>*
; CHECK: %[[V2:[0-9]+]] = insertelement <4 x float> undef, float %1, i32 0
; CHECK: %[[V3:[0-9]+]] = shufflevector <4 x float> %[[V2]], <4 x float> %[[V1]], <4 x i32> <i32 0, i32 4, i32 5, i32 6>
; CHECK: = fmul <4 x float> %[[V1]], %[[V3]]
@a = common global [32000 x float] zeroinitializer, align 16
define void @good_load_order() {
entry:
br label %for.cond1.preheader
for.cond1.preheader:
%0 = load float* getelementptr inbounds ([32000 x float]* @a, i64 0, i64 0), align 16
br label %for.body3
for.body3:
%1 = phi float [ %0, %for.cond1.preheader ], [ %10, %for.body3 ]
%indvars.iv = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next, %for.body3 ]
%2 = add nsw i64 %indvars.iv, 1
%arrayidx = getelementptr inbounds [32000 x float]* @a, i64 0, i64 %2
%3 = load float* %arrayidx, align 4
%arrayidx5 = getelementptr inbounds [32000 x float]* @a, i64 0, i64 %indvars.iv
%mul6 = fmul float %3, %1
store float %mul6, float* %arrayidx5, align 4
%4 = add nsw i64 %indvars.iv, 2
%arrayidx11 = getelementptr inbounds [32000 x float]* @a, i64 0, i64 %4
%5 = load float* %arrayidx11, align 4
%mul15 = fmul float %5, %3
store float %mul15, float* %arrayidx, align 4
%6 = add nsw i64 %indvars.iv, 3
%arrayidx21 = getelementptr inbounds [32000 x float]* @a, i64 0, i64 %6
%7 = load float* %arrayidx21, align 4
%mul25 = fmul float %7, %5
store float %mul25, float* %arrayidx11, align 4
%8 = add nsw i64 %indvars.iv, 4
%arrayidx31 = getelementptr inbounds [32000 x float]* @a, i64 0, i64 %8
%9 = load float* %arrayidx31, align 4
%mul35 = fmul float %9, %7
store float %mul35, float* %arrayidx21, align 4
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 5
%arrayidx41 = getelementptr inbounds [32000 x float]* @a, i64 0, i64 %indvars.iv.next
%10 = load float* %arrayidx41, align 4
%mul45 = fmul float %10, %9
store float %mul45, float* %arrayidx31, align 4
%11 = trunc i64 %indvars.iv.next to i32
%cmp2 = icmp slt i32 %11, 31995
br i1 %cmp2, label %for.body3, label %for.end
for.end:
ret void
}
; Check vectorization of following code for double data type-
; c[0] = a[0]+b[0];
; c[1] = b[1]+a[1]; // swapped b[1] and a[1]
; CHECK-LABEL: load_reorder_double
; CHECK: load <2 x double>*
; CHECK: fadd <2 x double>
define void @load_reorder_double(double* nocapture %c, double* noalias nocapture readonly %a, double* noalias nocapture readonly %b){
%1 = load double* %a
%2 = load double* %b
%3 = fadd double %1, %2
store double %3, double* %c
%4 = getelementptr inbounds double* %b, i64 1
%5 = load double* %4
%6 = getelementptr inbounds double* %a, i64 1
%7 = load double* %6
%8 = fadd double %5, %7
%9 = getelementptr inbounds double* %c, i64 1
store double %8, double* %9
ret void
}
; Check vectorization of following code for float data type-
; c[0] = a[0]+b[0];
; c[1] = b[1]+a[1]; // swapped b[1] and a[1]
; c[2] = a[2]+b[2];
; c[3] = a[3]+b[3];
; CHECK-LABEL: load_reorder_float
; CHECK: load <4 x float>*
; CHECK: fadd <4 x float>
define void @load_reorder_float(float* nocapture %c, float* noalias nocapture readonly %a, float* noalias nocapture readonly %b){
%1 = load float* %a
%2 = load float* %b
%3 = fadd float %1, %2
store float %3, float* %c
%4 = getelementptr inbounds float* %b, i64 1
%5 = load float* %4
%6 = getelementptr inbounds float* %a, i64 1
%7 = load float* %6
%8 = fadd float %5, %7
%9 = getelementptr inbounds float* %c, i64 1
store float %8, float* %9
%10 = getelementptr inbounds float* %a, i64 2
%11 = load float* %10
%12 = getelementptr inbounds float* %b, i64 2
%13 = load float* %12
%14 = fadd float %11, %13
%15 = getelementptr inbounds float* %c, i64 2
store float %14, float* %15
%16 = getelementptr inbounds float* %a, i64 3
%17 = load float* %16
%18 = getelementptr inbounds float* %b, i64 3
%19 = load float* %18
%20 = fadd float %17, %19
%21 = getelementptr inbounds float* %c, i64 3
store float %20, float* %21
ret void
}
; Check we properly reorder the below code so that it gets vectorized optimally-
; a[0] = (b[0]+c[0])+d[0];
; a[1] = d[1]+(b[1]+c[1]);
; a[2] = (b[2]+c[2])+d[2];
; a[3] = (b[3]+c[3])+d[3];
; CHECK-LABEL: opcode_reorder
; CHECK: load <4 x float>*
; CHECK: fadd <4 x float>
define void @opcode_reorder(float* noalias nocapture %a, float* noalias nocapture readonly %b,
float* noalias nocapture readonly %c,float* noalias nocapture readonly %d){
%1 = load float* %b
%2 = load float* %c
%3 = fadd float %1, %2
%4 = load float* %d
%5 = fadd float %3, %4
store float %5, float* %a
%6 = getelementptr inbounds float* %d, i64 1
%7 = load float* %6
%8 = getelementptr inbounds float* %b, i64 1
%9 = load float* %8
%10 = getelementptr inbounds float* %c, i64 1
%11 = load float* %10
%12 = fadd float %9, %11
%13 = fadd float %7, %12
%14 = getelementptr inbounds float* %a, i64 1
store float %13, float* %14
%15 = getelementptr inbounds float* %b, i64 2
%16 = load float* %15
%17 = getelementptr inbounds float* %c, i64 2
%18 = load float* %17
%19 = fadd float %16, %18
%20 = getelementptr inbounds float* %d, i64 2
%21 = load float* %20
%22 = fadd float %19, %21
%23 = getelementptr inbounds float* %a, i64 2
store float %22, float* %23
%24 = getelementptr inbounds float* %b, i64 3
%25 = load float* %24
%26 = getelementptr inbounds float* %c, i64 3
%27 = load float* %26
%28 = fadd float %25, %27
%29 = getelementptr inbounds float* %d, i64 3
%30 = load float* %29
%31 = fadd float %28, %30
%32 = getelementptr inbounds float* %a, i64 3
store float %31, float* %32
ret void
}
|
