1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -mtriple=x86_64-unknown -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=SSE
3; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX256
4; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=bdver1 -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX256
5; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX256
6; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skylake-avx512 -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX512
7
8target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
9
10@srcA64 = common global [8 x double] zeroinitializer, align 64
11@srcB64 = common global [8 x double] zeroinitializer, align 64
12@srcC64 = common global [8 x double] zeroinitializer, align 64
13@srcA32 = common global [16 x float] zeroinitializer, align 64
14@srcB32 = common global [16 x float] zeroinitializer, align 64
15@srcC32 = common global [16 x float] zeroinitializer, align 64
16@dst64 = common global [8 x double] zeroinitializer, align 64
17@dst32 = common global [16 x float] zeroinitializer, align 64
18
19declare float @llvm.copysign.f32(float, float)
20declare double @llvm.copysign.f64(double, double)
21
22;
23; CHECK
24;
25
26define void @fcopysign_2f64() #0 {
27; CHECK-LABEL: @fcopysign_2f64(
28; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcA64 to <2 x double>*), align 8
29; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcB64 to <2 x double>*), align 8
30; CHECK-NEXT:    [[TMP3:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP1]], <2 x double> [[TMP2]])
31; CHECK-NEXT:    store <2 x double> [[TMP3]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
32; CHECK-NEXT:    ret void
33;
34  %a0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 0), align 8
35  %a1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 1), align 8
36  %b0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 0), align 8
37  %b1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 1), align 8
38  %fcopysign0 = call double @llvm.copysign.f64(double %a0, double %b0)
39  %fcopysign1 = call double @llvm.copysign.f64(double %a1, double %b1)
40  store double %fcopysign0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
41  store double %fcopysign1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
42  ret void
43}
44
45define void @fcopysign_4f64() #0 {
46; SSE-LABEL: @fcopysign_4f64(
47; SSE-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcA64 to <2 x double>*), align 8
48; SSE-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 2) to <2 x double>*), align 8
49; SSE-NEXT:    [[TMP3:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcB64 to <2 x double>*), align 8
50; SSE-NEXT:    [[TMP4:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 2) to <2 x double>*), align 8
51; SSE-NEXT:    [[TMP5:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP1]], <2 x double> [[TMP3]])
52; SSE-NEXT:    [[TMP6:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP2]], <2 x double> [[TMP4]])
53; SSE-NEXT:    store <2 x double> [[TMP5]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
54; SSE-NEXT:    store <2 x double> [[TMP6]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 8
55; SSE-NEXT:    ret void
56;
57; AVX-LABEL: @fcopysign_4f64(
58; AVX-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @srcA64 to <4 x double>*), align 8
59; AVX-NEXT:    [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @srcB64 to <4 x double>*), align 8
60; AVX-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.copysign.v4f64(<4 x double> [[TMP1]], <4 x double> [[TMP2]])
61; AVX-NEXT:    store <4 x double> [[TMP3]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
62; AVX-NEXT:    ret void
63;
64  %a0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 0), align 8
65  %a1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 1), align 8
66  %a2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 2), align 8
67  %a3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 3), align 8
68  %b0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 0), align 8
69  %b1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 1), align 8
70  %b2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 2), align 8
71  %b3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 3), align 8
72  %fcopysign0 = call double @llvm.copysign.f64(double %a0, double %b0)
73  %fcopysign1 = call double @llvm.copysign.f64(double %a1, double %b1)
74  %fcopysign2 = call double @llvm.copysign.f64(double %a2, double %b2)
75  %fcopysign3 = call double @llvm.copysign.f64(double %a3, double %b3)
76  store double %fcopysign0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
77  store double %fcopysign1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
78  store double %fcopysign2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
79  store double %fcopysign3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
80  ret void
81}
82
83define void @fcopysign_8f64() #0 {
84; SSE-LABEL: @fcopysign_8f64(
85; SSE-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcA64 to <2 x double>*), align 4
86; SSE-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 2) to <2 x double>*), align 4
87; SSE-NEXT:    [[TMP3:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 4) to <2 x double>*), align 4
88; SSE-NEXT:    [[TMP4:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 6) to <2 x double>*), align 4
89; SSE-NEXT:    [[TMP5:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @srcB64 to <2 x double>*), align 4
90; SSE-NEXT:    [[TMP6:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 2) to <2 x double>*), align 4
91; SSE-NEXT:    [[TMP7:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 4) to <2 x double>*), align 4
92; SSE-NEXT:    [[TMP8:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 6) to <2 x double>*), align 4
93; SSE-NEXT:    [[TMP9:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP1]], <2 x double> [[TMP5]])
94; SSE-NEXT:    [[TMP10:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP2]], <2 x double> [[TMP6]])
95; SSE-NEXT:    [[TMP11:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP3]], <2 x double> [[TMP7]])
96; SSE-NEXT:    [[TMP12:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP4]], <2 x double> [[TMP8]])
97; SSE-NEXT:    store <2 x double> [[TMP9]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 4
98; SSE-NEXT:    store <2 x double> [[TMP10]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 4
99; SSE-NEXT:    store <2 x double> [[TMP11]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 4
100; SSE-NEXT:    store <2 x double> [[TMP12]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 4
101; SSE-NEXT:    ret void
102;
103; AVX256-LABEL: @fcopysign_8f64(
104; AVX256-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @srcA64 to <4 x double>*), align 4
105; AVX256-NEXT:    [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 4) to <4 x double>*), align 4
106; AVX256-NEXT:    [[TMP3:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @srcB64 to <4 x double>*), align 4
107; AVX256-NEXT:    [[TMP4:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 4) to <4 x double>*), align 4
108; AVX256-NEXT:    [[TMP5:%.*]] = call <4 x double> @llvm.copysign.v4f64(<4 x double> [[TMP1]], <4 x double> [[TMP3]])
109; AVX256-NEXT:    [[TMP6:%.*]] = call <4 x double> @llvm.copysign.v4f64(<4 x double> [[TMP2]], <4 x double> [[TMP4]])
110; AVX256-NEXT:    store <4 x double> [[TMP5]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 4
111; AVX256-NEXT:    store <4 x double> [[TMP6]], <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <4 x double>*), align 4
112; AVX256-NEXT:    ret void
113;
114; AVX512-LABEL: @fcopysign_8f64(
115; AVX512-NEXT:    [[TMP1:%.*]] = load <8 x double>, <8 x double>* bitcast ([8 x double]* @srcA64 to <8 x double>*), align 4
116; AVX512-NEXT:    [[TMP2:%.*]] = load <8 x double>, <8 x double>* bitcast ([8 x double]* @srcB64 to <8 x double>*), align 4
117; AVX512-NEXT:    [[TMP3:%.*]] = call <8 x double> @llvm.copysign.v8f64(<8 x double> [[TMP1]], <8 x double> [[TMP2]])
118; AVX512-NEXT:    store <8 x double> [[TMP3]], <8 x double>* bitcast ([8 x double]* @dst64 to <8 x double>*), align 4
119; AVX512-NEXT:    ret void
120;
121  %a0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 0), align 4
122  %a1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 1), align 4
123  %a2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 2), align 4
124  %a3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 3), align 4
125  %a4 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 4), align 4
126  %a5 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 5), align 4
127  %a6 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 6), align 4
128  %a7 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcA64, i32 0, i64 7), align 4
129  %b0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 0), align 4
130  %b1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 1), align 4
131  %b2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 2), align 4
132  %b3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 3), align 4
133  %b4 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 4), align 4
134  %b5 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 5), align 4
135  %b6 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 6), align 4
136  %b7 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @srcB64, i32 0, i64 7), align 4
137  %fcopysign0 = call double @llvm.copysign.f64(double %a0, double %b0)
138  %fcopysign1 = call double @llvm.copysign.f64(double %a1, double %b1)
139  %fcopysign2 = call double @llvm.copysign.f64(double %a2, double %b2)
140  %fcopysign3 = call double @llvm.copysign.f64(double %a3, double %b3)
141  %fcopysign4 = call double @llvm.copysign.f64(double %a4, double %b4)
142  %fcopysign5 = call double @llvm.copysign.f64(double %a5, double %b5)
143  %fcopysign6 = call double @llvm.copysign.f64(double %a6, double %b6)
144  %fcopysign7 = call double @llvm.copysign.f64(double %a7, double %b7)
145  store double %fcopysign0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 4
146  store double %fcopysign1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 4
147  store double %fcopysign2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 4
148  store double %fcopysign3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 4
149  store double %fcopysign4, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 4
150  store double %fcopysign5, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 4
151  store double %fcopysign6, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 4
152  store double %fcopysign7, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 4
153  ret void
154}
155
156define void @fcopysign_4f32() #0 {
157; CHECK-LABEL: @fcopysign_4f32(
158; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcA32 to <4 x float>*), align 4
159; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcB32 to <4 x float>*), align 4
160; CHECK-NEXT:    [[TMP3:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP1]], <4 x float> [[TMP2]])
161; CHECK-NEXT:    store <4 x float> [[TMP3]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
162; CHECK-NEXT:    ret void
163;
164  %a0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 0), align 4
165  %a1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 1), align 4
166  %a2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 2), align 4
167  %a3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 3), align 4
168  %b0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 0), align 4
169  %b1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 1), align 4
170  %b2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 2), align 4
171  %b3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 3), align 4
172  %fcopysign0 = call float @llvm.copysign.f32(float %a0, float %b0)
173  %fcopysign1 = call float @llvm.copysign.f32(float %a1, float %b1)
174  %fcopysign2 = call float @llvm.copysign.f32(float %a2, float %b2)
175  %fcopysign3 = call float @llvm.copysign.f32(float %a3, float %b3)
176  store float %fcopysign0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
177  store float %fcopysign1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
178  store float %fcopysign2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
179  store float %fcopysign3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
180  ret void
181}
182
183define void @fcopysign_8f32() #0 {
184; SSE-LABEL: @fcopysign_8f32(
185; SSE-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcA32 to <4 x float>*), align 4
186; SSE-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 4) to <4 x float>*), align 4
187; SSE-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcB32 to <4 x float>*), align 4
188; SSE-NEXT:    [[TMP4:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 4) to <4 x float>*), align 4
189; SSE-NEXT:    [[TMP5:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP1]], <4 x float> [[TMP3]])
190; SSE-NEXT:    [[TMP6:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP2]], <4 x float> [[TMP4]])
191; SSE-NEXT:    store <4 x float> [[TMP5]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
192; SSE-NEXT:    store <4 x float> [[TMP6]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
193; SSE-NEXT:    ret void
194;
195; AVX-LABEL: @fcopysign_8f32(
196; AVX-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @srcA32 to <8 x float>*), align 4
197; AVX-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @srcB32 to <8 x float>*), align 4
198; AVX-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.copysign.v8f32(<8 x float> [[TMP1]], <8 x float> [[TMP2]])
199; AVX-NEXT:    store <8 x float> [[TMP3]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
200; AVX-NEXT:    ret void
201;
202  %a0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 0), align 4
203  %a1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 1), align 4
204  %a2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 2), align 4
205  %a3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 3), align 4
206  %a4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 4), align 4
207  %a5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 5), align 4
208  %a6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 6), align 4
209  %a7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 7), align 4
210  %b0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 0), align 4
211  %b1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 1), align 4
212  %b2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 2), align 4
213  %b3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 3), align 4
214  %b4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 4), align 4
215  %b5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 5), align 4
216  %b6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 6), align 4
217  %b7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 7), align 4
218  %fcopysign0 = call float @llvm.copysign.f32(float %a0, float %b0)
219  %fcopysign1 = call float @llvm.copysign.f32(float %a1, float %b1)
220  %fcopysign2 = call float @llvm.copysign.f32(float %a2, float %b2)
221  %fcopysign3 = call float @llvm.copysign.f32(float %a3, float %b3)
222  %fcopysign4 = call float @llvm.copysign.f32(float %a4, float %b4)
223  %fcopysign5 = call float @llvm.copysign.f32(float %a5, float %b5)
224  %fcopysign6 = call float @llvm.copysign.f32(float %a6, float %b6)
225  %fcopysign7 = call float @llvm.copysign.f32(float %a7, float %b7)
226  store float %fcopysign0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
227  store float %fcopysign1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
228  store float %fcopysign2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
229  store float %fcopysign3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
230  store float %fcopysign4, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
231  store float %fcopysign5, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
232  store float %fcopysign6, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
233  store float %fcopysign7, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
234  ret void
235}
236
237define void @fcopysign_16f32() #0 {
238; SSE-LABEL: @fcopysign_16f32(
239; SSE-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcA32 to <4 x float>*), align 4
240; SSE-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 4) to <4 x float>*), align 4
241; SSE-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 8) to <4 x float>*), align 4
242; SSE-NEXT:    [[TMP4:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 12) to <4 x float>*), align 4
243; SSE-NEXT:    [[TMP5:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @srcB32 to <4 x float>*), align 4
244; SSE-NEXT:    [[TMP6:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 4) to <4 x float>*), align 4
245; SSE-NEXT:    [[TMP7:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 8) to <4 x float>*), align 4
246; SSE-NEXT:    [[TMP8:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 12) to <4 x float>*), align 4
247; SSE-NEXT:    [[TMP9:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP1]], <4 x float> [[TMP5]])
248; SSE-NEXT:    [[TMP10:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP2]], <4 x float> [[TMP6]])
249; SSE-NEXT:    [[TMP11:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP3]], <4 x float> [[TMP7]])
250; SSE-NEXT:    [[TMP12:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP4]], <4 x float> [[TMP8]])
251; SSE-NEXT:    store <4 x float> [[TMP9]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
252; SSE-NEXT:    store <4 x float> [[TMP10]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
253; SSE-NEXT:    store <4 x float> [[TMP11]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <4 x float>*), align 4
254; SSE-NEXT:    store <4 x float> [[TMP12]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12) to <4 x float>*), align 4
255; SSE-NEXT:    ret void
256;
257; AVX256-LABEL: @fcopysign_16f32(
258; AVX256-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @srcA32 to <8 x float>*), align 4
259; AVX256-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 8) to <8 x float>*), align 4
260; AVX256-NEXT:    [[TMP3:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @srcB32 to <8 x float>*), align 4
261; AVX256-NEXT:    [[TMP4:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 8) to <8 x float>*), align 4
262; AVX256-NEXT:    [[TMP5:%.*]] = call <8 x float> @llvm.copysign.v8f32(<8 x float> [[TMP1]], <8 x float> [[TMP3]])
263; AVX256-NEXT:    [[TMP6:%.*]] = call <8 x float> @llvm.copysign.v8f32(<8 x float> [[TMP2]], <8 x float> [[TMP4]])
264; AVX256-NEXT:    store <8 x float> [[TMP5]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
265; AVX256-NEXT:    store <8 x float> [[TMP6]], <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <8 x float>*), align 4
266; AVX256-NEXT:    ret void
267;
268; AVX512-LABEL: @fcopysign_16f32(
269; AVX512-NEXT:    [[TMP1:%.*]] = load <16 x float>, <16 x float>* bitcast ([16 x float]* @srcA32 to <16 x float>*), align 4
270; AVX512-NEXT:    [[TMP2:%.*]] = load <16 x float>, <16 x float>* bitcast ([16 x float]* @srcB32 to <16 x float>*), align 4
271; AVX512-NEXT:    [[TMP3:%.*]] = call <16 x float> @llvm.copysign.v16f32(<16 x float> [[TMP1]], <16 x float> [[TMP2]])
272; AVX512-NEXT:    store <16 x float> [[TMP3]], <16 x float>* bitcast ([16 x float]* @dst32 to <16 x float>*), align 4
273; AVX512-NEXT:    ret void
274;
275  %a0  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64  0), align 4
276  %a1  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64  1), align 4
277  %a2  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64  2), align 4
278  %a3  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64  3), align 4
279  %a4  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64  4), align 4
280  %a5  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64  5), align 4
281  %a6  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64  6), align 4
282  %a7  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64  7), align 4
283  %a8  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64  8), align 4
284  %a9  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64  9), align 4
285  %a10 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 10), align 4
286  %a11 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 11), align 4
287  %a12 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 12), align 4
288  %a13 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 13), align 4
289  %a14 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 14), align 4
290  %a15 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcA32, i32 0, i64 15), align 4
291  %b0  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64  0), align 4
292  %b1  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64  1), align 4
293  %b2  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64  2), align 4
294  %b3  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64  3), align 4
295  %b4  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64  4), align 4
296  %b5  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64  5), align 4
297  %b6  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64  6), align 4
298  %b7  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64  7), align 4
299  %b8  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64  8), align 4
300  %b9  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64  9), align 4
301  %b10 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 10), align 4
302  %b11 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 11), align 4
303  %b12 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 12), align 4
304  %b13 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 13), align 4
305  %b14 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 14), align 4
306  %b15 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @srcB32, i32 0, i64 15), align 4
307  %fcopysign0  = call float @llvm.copysign.f32(float %a0 , float %b0 )
308  %fcopysign1  = call float @llvm.copysign.f32(float %a1 , float %b1 )
309  %fcopysign2  = call float @llvm.copysign.f32(float %a2 , float %b2 )
310  %fcopysign3  = call float @llvm.copysign.f32(float %a3 , float %b3 )
311  %fcopysign4  = call float @llvm.copysign.f32(float %a4 , float %b4 )
312  %fcopysign5  = call float @llvm.copysign.f32(float %a5 , float %b5 )
313  %fcopysign6  = call float @llvm.copysign.f32(float %a6 , float %b6 )
314  %fcopysign7  = call float @llvm.copysign.f32(float %a7 , float %b7 )
315  %fcopysign8  = call float @llvm.copysign.f32(float %a8 , float %b8 )
316  %fcopysign9  = call float @llvm.copysign.f32(float %a9 , float %b9 )
317  %fcopysign10 = call float @llvm.copysign.f32(float %a10, float %b10)
318  %fcopysign11 = call float @llvm.copysign.f32(float %a11, float %b11)
319  %fcopysign12 = call float @llvm.copysign.f32(float %a12, float %b12)
320  %fcopysign13 = call float @llvm.copysign.f32(float %a13, float %b13)
321  %fcopysign14 = call float @llvm.copysign.f32(float %a14, float %b14)
322  %fcopysign15 = call float @llvm.copysign.f32(float %a15, float %b15)
323  store float %fcopysign0 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64  0), align 4
324  store float %fcopysign1 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64  1), align 4
325  store float %fcopysign2 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64  2), align 4
326  store float %fcopysign3 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64  3), align 4
327  store float %fcopysign4 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64  4), align 4
328  store float %fcopysign5 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64  5), align 4
329  store float %fcopysign6 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64  6), align 4
330  store float %fcopysign7 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64  7), align 4
331  store float %fcopysign8 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64  8), align 4
332  store float %fcopysign9 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64  9), align 4
333  store float %fcopysign10, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 10), align 4
334  store float %fcopysign11, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 11), align 4
335  store float %fcopysign12, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12), align 4
336  store float %fcopysign13, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 13), align 4
337  store float %fcopysign14, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 14), align 4
338  store float %fcopysign15, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 15), align 4
339  ret void
340}
341
342attributes #0 = { nounwind }
343