1 // RUN: %clang_cc1 -x c++ %s -O0 -triple=x86_64-apple-darwin -target-feature +avx2 -fmax-type-align=16 -emit-llvm -o - -Werror | FileCheck %s
2 // rdar://16254558
3 
4 typedef float AVX2Float __attribute__((__vector_size__(32)));
5 
6 
7 volatile float TestAlign(void)
8 {
9        volatile AVX2Float *p = new AVX2Float;
10         *p = *p;
11         AVX2Float r = *p;
12         return r[0];
13 }
14 
15 // CHECK: [[R:%.*]] = alloca <8 x float>, align 32
16 // CHECK-NEXT:  [[CALL:%.*]] = call i8* @_Znwm(i64 32)
17 // CHECK-NEXT:  [[ZERO:%.*]] = bitcast i8* [[CALL]] to <8 x float>*
18 // CHECK-NEXT:  store <8 x float>* [[ZERO]], <8 x float>** [[P:%.*]], align 8
19 // CHECK-NEXT:  [[ONE:%.*]] = load <8 x float>*, <8 x float>** [[P]], align 8
20 // CHECK-NEXT:  [[TWO:%.*]] = load volatile <8 x float>, <8 x float>* [[ONE]], align 16
21 // CHECK-NEXT:  [[THREE:%.*]] = load <8 x float>*, <8 x float>** [[P]], align 8
22 // CHECK-NEXT:  store volatile <8 x float> [[TWO]], <8 x float>* [[THREE]], align 16
23 // CHECK-NEXT:  [[FOUR:%.*]] = load <8 x float>*, <8 x float>** [[P]], align 8
24 // CHECK-NEXT:  [[FIVE:%.*]] = load volatile <8 x float>, <8 x float>* [[FOUR]], align 16
25 // CHECK-NEXT:  store <8 x float> [[FIVE]], <8 x float>* [[R]], align 32
26 // CHECK-NEXT:  [[SIX:%.*]] = load <8 x float>, <8 x float>* [[R]], align 32
27 // CHECK-NEXT:  [[VECEXT:%.*]] = extractelement <8 x float> [[SIX]], i32 0
28 // CHECK-NEXT:  ret float [[VECEXT]]
29 
30 typedef float AVX2Float_Explicitly_aligned __attribute__((__vector_size__(32))) __attribute__((aligned (32)));
31 
32 typedef AVX2Float_Explicitly_aligned AVX2Float_indirect;
33 
34 typedef AVX2Float_indirect AVX2Float_use_existing_align;
35 
36 volatile float TestAlign2(void)
37 {
38        volatile AVX2Float_use_existing_align *p = new AVX2Float_use_existing_align;
39         *p = *p;
40         AVX2Float_use_existing_align r = *p;
41         return r[0];
42 }
43 
44 // CHECK: [[R:%.*]] = alloca <8 x float>, align 32
45 // CHECK-NEXT:  [[CALL:%.*]] = call i8* @_Znwm(i64 32)
46 // CHECK-NEXT:  [[ZERO:%.*]] = bitcast i8* [[CALL]] to <8 x float>*
47 // CHECK-NEXT:  store <8 x float>* [[ZERO]], <8 x float>** [[P:%.*]], align 8
48 // CHECK-NEXT:  [[ONE:%.*]] = load <8 x float>*, <8 x float>** [[P]], align 8
49 // CHECK-NEXT:  [[TWO:%.*]] = load volatile <8 x float>, <8 x float>* [[ONE]], align 32
50 // CHECK-NEXT:  [[THREE:%.*]] = load <8 x float>*, <8 x float>** [[P]], align 8
51 // CHECK-NEXT:  store volatile <8 x float> [[TWO]], <8 x float>* [[THREE]], align 32
52 // CHECK-NEXT:  [[FOUR:%.*]] = load <8 x float>*, <8 x float>** [[P]], align 8
53 // CHECK-NEXT:  [[FIVE:%.*]] = load volatile <8 x float>, <8 x float>* [[FOUR]], align 32
54 // CHECK-NEXT:  store <8 x float> [[FIVE]], <8 x float>* [[R]], align 32
55 // CHECK-NEXT:  [[SIX:%.*]] = load <8 x float>, <8 x float>* [[R]], align 32
56 // CHECK-NEXT:  [[VECEXT:%.*]] = extractelement <8 x float> [[SIX]], i32 0
57 // CHECK-NEXT:  ret float [[VECEXT]]
58