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
|
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -passes="default<O3>" -S < %s | FileCheck %s --check-prefix=SSE
; RUN: opt -passes="default<O3>" -S -mattr=avx < %s | FileCheck %s --check-prefix=AVX
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64--"
%union.ElementWiseAccess = type { <4 x float> }
$getAt = comdat any
define dso_local noundef <4 x float> @ConvertVectors_ByRef(ptr noundef nonnull align 16 dereferenceable(16) %0) #0 {
; SSE-LABEL: @ConvertVectors_ByRef(
; SSE-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr [[TMP0:%.*]], align 16
; SSE-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 2>
; SSE-NEXT: ret <4 x float> [[TMP3]]
;
; AVX-LABEL: @ConvertVectors_ByRef(
; AVX-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr [[TMP0:%.*]], align 16
; AVX-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 2>
; AVX-NEXT: ret <4 x float> [[TMP3]]
;
%2 = alloca ptr, align 8
%3 = alloca <4 x float>, align 16
store ptr %0, ptr %2, align 8
%4 = load ptr, ptr %2, align 8
%5 = call noundef nonnull align 16 dereferenceable(16) ptr @castToElementWiseAccess_ByRef(ptr noundef nonnull align 16 dereferenceable(16) %4)
%6 = call noundef float @getAt(ptr noundef nonnull align 16 dereferenceable(16) %5, i32 noundef 0)
%7 = insertelement <4 x float> undef, float %6, i32 0
%8 = load ptr, ptr %2, align 8
%9 = call noundef nonnull align 16 dereferenceable(16) ptr @castToElementWiseAccess_ByRef(ptr noundef nonnull align 16 dereferenceable(16) %8)
%10 = call noundef float @getAt(ptr noundef nonnull align 16 dereferenceable(16) %9, i32 noundef 1)
%11 = insertelement <4 x float> %7, float %10, i32 1
%12 = load ptr, ptr %2, align 8
%13 = call noundef nonnull align 16 dereferenceable(16) ptr @castToElementWiseAccess_ByRef(ptr noundef nonnull align 16 dereferenceable(16) %12)
%14 = call noundef float @getAt(ptr noundef nonnull align 16 dereferenceable(16) %13, i32 noundef 2)
%15 = insertelement <4 x float> %11, float %14, i32 2
%16 = load ptr, ptr %2, align 8
%17 = call noundef nonnull align 16 dereferenceable(16) ptr @castToElementWiseAccess_ByRef(ptr noundef nonnull align 16 dereferenceable(16) %16)
%18 = call noundef float @getAt(ptr noundef nonnull align 16 dereferenceable(16) %17, i32 noundef 2)
%19 = insertelement <4 x float> %15, float %18, i32 3
store <4 x float> %19, ptr %3, align 16
%20 = load <4 x float>, ptr %3, align 16
ret <4 x float> %20
}
define noundef <4 x float> @ConvertVectors_ByVal(ptr noundef nonnull align 16 dereferenceable(16) %V) #0 {
; SSE-LABEL: @ConvertVectors_ByVal(
; SSE-NEXT: entry:
; SSE-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr [[V:%.*]], align 16
; SSE-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[V]], i64 8
; SSE-NEXT: [[V_VAL421:%.*]] = load i64, ptr [[TMP1]], align 8
; SSE-NEXT: [[TMP2:%.*]] = trunc i64 [[V_VAL421]] to i32
; SSE-NEXT: [[TMP3:%.*]] = bitcast i32 [[TMP2]] to float
; SSE-NEXT: [[VECINIT11:%.*]] = insertelement <4 x float> [[TMP0]], float [[TMP3]], i64 2
; SSE-NEXT: [[VECINIT16:%.*]] = insertelement <4 x float> [[VECINIT11]], float [[TMP3]], i64 3
; SSE-NEXT: ret <4 x float> [[VECINIT16]]
;
; AVX-LABEL: @ConvertVectors_ByVal(
; AVX-NEXT: entry:
; AVX-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr [[V:%.*]], align 16
; AVX-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[V]], i64 8
; AVX-NEXT: [[V_VAL421:%.*]] = load i64, ptr [[TMP1]], align 8
; AVX-NEXT: [[TMP2:%.*]] = trunc i64 [[V_VAL421]] to i32
; AVX-NEXT: [[TMP3:%.*]] = bitcast i32 [[TMP2]] to float
; AVX-NEXT: [[VECINIT11:%.*]] = insertelement <4 x float> [[TMP0]], float [[TMP3]], i64 2
; AVX-NEXT: [[VECINIT16:%.*]] = insertelement <4 x float> [[VECINIT11]], float [[TMP3]], i64 3
; AVX-NEXT: ret <4 x float> [[VECINIT16]]
;
entry:
%V.addr = alloca ptr, align 8
%.compoundliteral = alloca <4 x float>, align 16
%ref.tmp = alloca %union.ElementWiseAccess, align 16
%ref.tmp2 = alloca %union.ElementWiseAccess, align 16
%ref.tmp7 = alloca %union.ElementWiseAccess, align 16
%ref.tmp12 = alloca %union.ElementWiseAccess, align 16
store ptr %V, ptr %V.addr, align 8
call void @llvm.lifetime.start.p0(i64 16, ptr %ref.tmp) #4
%0 = load ptr, ptr %V.addr, align 8
%call = call { double, double } @castToElementWiseAccess_ByVal(ptr noundef nonnull align 16 dereferenceable(16) %0)
%coerce.dive = getelementptr inbounds %union.ElementWiseAccess, ptr %ref.tmp, i32 0, i32 0
%1 = getelementptr inbounds { double, double }, ptr %coerce.dive, i32 0, i32 0
%2 = extractvalue { double, double } %call, 0
store double %2, ptr %1, align 16
%3 = getelementptr inbounds { double, double }, ptr %coerce.dive, i32 0, i32 1
%4 = extractvalue { double, double } %call, 1
store double %4, ptr %3, align 8
%call1 = call noundef float @ElementWiseAccess5getAt(ptr noundef nonnull align 16 dereferenceable(16) %ref.tmp, i32 noundef 0)
%vecinit = insertelement <4 x float> undef, float %call1, i32 0
call void @llvm.lifetime.start.p0(i64 16, ptr %ref.tmp2) #4
%5 = load ptr, ptr %V.addr, align 8
%call3 = call { double, double } @castToElementWiseAccess_ByVal(ptr noundef nonnull align 16 dereferenceable(16) %5)
%coerce.dive4 = getelementptr inbounds %union.ElementWiseAccess, ptr %ref.tmp2, i32 0, i32 0
%6 = getelementptr inbounds { double, double }, ptr %coerce.dive4, i32 0, i32 0
%7 = extractvalue { double, double } %call3, 0
store double %7, ptr %6, align 16
%8 = getelementptr inbounds { double, double }, ptr %coerce.dive4, i32 0, i32 1
%9 = extractvalue { double, double } %call3, 1
store double %9, ptr %8, align 8
%call5 = call noundef float @ElementWiseAccess5getAt(ptr noundef nonnull align 16 dereferenceable(16) %ref.tmp2, i32 noundef 1)
%vecinit6 = insertelement <4 x float> %vecinit, float %call5, i32 1
call void @llvm.lifetime.start.p0(i64 16, ptr %ref.tmp7) #4
%10 = load ptr, ptr %V.addr, align 8
%call8 = call { double, double } @castToElementWiseAccess_ByVal(ptr noundef nonnull align 16 dereferenceable(16) %10)
%coerce.dive9 = getelementptr inbounds %union.ElementWiseAccess, ptr %ref.tmp7, i32 0, i32 0
%11 = getelementptr inbounds { double, double }, ptr %coerce.dive9, i32 0, i32 0
%12 = extractvalue { double, double } %call8, 0
store double %12, ptr %11, align 16
%13 = getelementptr inbounds { double, double }, ptr %coerce.dive9, i32 0, i32 1
%14 = extractvalue { double, double } %call8, 1
store double %14, ptr %13, align 8
%call10 = call noundef float @ElementWiseAccess5getAt(ptr noundef nonnull align 16 dereferenceable(16) %ref.tmp7, i32 noundef 2)
%vecinit11 = insertelement <4 x float> %vecinit6, float %call10, i32 2
call void @llvm.lifetime.start.p0(i64 16, ptr %ref.tmp12) #4
%15 = load ptr, ptr %V.addr, align 8
%call13 = call { double, double } @castToElementWiseAccess_ByVal(ptr noundef nonnull align 16 dereferenceable(16) %15)
%coerce.dive14 = getelementptr inbounds %union.ElementWiseAccess, ptr %ref.tmp12, i32 0, i32 0
%16 = getelementptr inbounds { double, double }, ptr %coerce.dive14, i32 0, i32 0
%17 = extractvalue { double, double } %call13, 0
store double %17, ptr %16, align 16
%18 = getelementptr inbounds { double, double }, ptr %coerce.dive14, i32 0, i32 1
%19 = extractvalue { double, double } %call13, 1
store double %19, ptr %18, align 8
%call15 = call noundef float @ElementWiseAccess5getAt(ptr noundef nonnull align 16 dereferenceable(16) %ref.tmp12, i32 noundef 2)
%vecinit16 = insertelement <4 x float> %vecinit11, float %call15, i32 3
store <4 x float> %vecinit16, ptr %.compoundliteral, align 16
%20 = load <4 x float>, ptr %.compoundliteral, align 16
call void @llvm.lifetime.end.p0(i64 16, ptr %ref.tmp12) #4
call void @llvm.lifetime.end.p0(i64 16, ptr %ref.tmp7) #4
call void @llvm.lifetime.end.p0(i64 16, ptr %ref.tmp2) #4
call void @llvm.lifetime.end.p0(i64 16, ptr %ref.tmp) #4
ret <4 x float> %20
}
define internal { double, double } @castToElementWiseAccess_ByVal(ptr noundef nonnull align 16 dereferenceable(16) %t) #1 {
entry:
%retval = alloca %union.ElementWiseAccess, align 16
%t.addr = alloca ptr, align 8
store ptr %t, ptr %t.addr, align 8
%0 = load ptr, ptr %t.addr, align 8
call void @llvm.memcpy.p0.p0.i64(ptr align 16 %retval, ptr align 16 %0, i64 16, i1 false)
%coerce.dive = getelementptr inbounds %union.ElementWiseAccess, ptr %retval, i32 0, i32 0
%1 = load { double, double }, ptr %coerce.dive, align 16
ret { double, double } %1
}
declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #2
declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #2
declare void @llvm.memcpy.p0.p0.i64(ptr noalias nocapture writeonly, ptr noalias nocapture readonly, i64, i1 immarg) #3
define internal noundef nonnull align 16 dereferenceable(16) ptr @castToElementWiseAccess_ByRef(ptr noundef nonnull align 16 dereferenceable(16) %0) #1 {
%2 = alloca ptr, align 8
store ptr %0, ptr %2, align 8
%3 = load ptr, ptr %2, align 8
ret ptr %3
}
define linkonce_odr dso_local noundef float @getAt(ptr noundef nonnull align 16 dereferenceable(16) %0, i32 noundef %1) #1 comdat align 2 {
%3 = alloca ptr, align 8
%4 = alloca i32, align 4
store ptr %0, ptr %3, align 8
store i32 %1, ptr %4, align 4
%5 = load ptr, ptr %3, align 8
%6 = load i32, ptr %4, align 4
%7 = sext i32 %6 to i64
%8 = getelementptr inbounds [4 x float], ptr %5, i64 0, i64 %7
%9 = load float, ptr %8, align 4
ret float %9
}
define linkonce_odr noundef float @ElementWiseAccess5getAt(ptr noundef nonnull align 16 dereferenceable(16) %this, i32 noundef %i) #1 align 2 {
entry:
%this.addr = alloca ptr, align 8
%i.addr = alloca i32, align 4
store ptr %this, ptr %this.addr, align 8
store i32 %i, ptr %i.addr, align 4
%this1 = load ptr, ptr %this.addr, align 8
%0 = load i32, ptr %i.addr, align 4
%idxprom = sext i32 %0 to i64
%arrayidx = getelementptr inbounds [4 x float], ptr %this1, i64 0, i64 %idxprom
%1 = load float, ptr %arrayidx, align 4
ret float %1
}
; Vector combine + SLP should form a narrow load and a vector cast
define void @PR51397(ptr noundef %dst, ptr noundef %srcp) {
; SSE-LABEL: @PR51397(
; SSE-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[SRCP:%.*]], align 16
; SSE-NEXT: [[TMP2:%.*]] = sitofp <4 x i32> [[TMP1]] to <4 x float>
; SSE-NEXT: store <4 x float> [[TMP2]], ptr [[DST:%.*]], align 16
; SSE-NEXT: ret void
;
; AVX-LABEL: @PR51397(
; AVX-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[SRCP:%.*]], align 16
; AVX-NEXT: [[TMP2:%.*]] = sitofp <4 x i32> [[TMP1]] to <4 x float>
; AVX-NEXT: store <4 x float> [[TMP2]], ptr [[DST:%.*]], align 16
; AVX-NEXT: ret void
;
%src = load <8 x i32>, ptr %srcp, align 16
%vecext = extractelement <8 x i32> %src, i32 0
%conv = sitofp i32 %vecext to float
%vecinit = insertelement <4 x float> undef, float %conv, i32 0
%vecext1 = extractelement <8 x i32> %src, i32 1
%conv2 = sitofp i32 %vecext1 to float
%vecinit3 = insertelement <4 x float> %vecinit, float %conv2, i32 1
%vecext4 = extractelement <8 x i32> %src, i32 2
%conv5 = sitofp i32 %vecext4 to float
%vecinit6 = insertelement <4 x float> %vecinit3, float %conv5, i32 2
%vecext7 = extractelement <8 x i32> %src, i32 3
%conv8 = sitofp i32 %vecext7 to float
%vecinit9 = insertelement <4 x float> %vecinit6, float %conv8, i32 3
store <4 x float> %vecinit9, ptr %dst, align 16
ret void
}
|
