1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt < %s -interleaved-access -S | FileCheck %s
4 target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
5 target triple = "aarch64--linux-gnu"
7 define <4 x float> @vld2(<8 x float>* %pSrc) {
10 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast <8 x float>* [[PSRC:%.*]] to <4 x float>*
11 ; CHECK-NEXT: [[LDN:%.*]] = call { <4 x float>, <4 x float> } @llvm.aarch64.neon.ld2.v4f32.p0v4f32(<4 x float>* [[TMP0]])
12 ; CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <4 x float>, <4 x float> } [[LDN]], 1
13 ; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <4 x float>, <4 x float> } [[LDN]], 1
14 ; CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <4 x float>, <4 x float> } [[LDN]], 0
15 ; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <4 x float>, <4 x float> } [[LDN]], 0
16 ; CHECK-NEXT: [[L26:%.*]] = fmul fast <4 x float> [[TMP3]], [[TMP4]]
17 ; CHECK-NEXT: [[L43:%.*]] = fmul fast <4 x float> [[TMP1]], [[TMP2]]
18 ; CHECK-NEXT: [[L6:%.*]] = fadd fast <4 x float> [[L43]], [[L26]]
19 ; CHECK-NEXT: ret <4 x float> [[L6]]
22 %wide.vec = load <8 x float>, <8 x float>* %pSrc, align 4
23 %l2 = fmul fast <8 x float> %wide.vec, %wide.vec
24 %l3 = shufflevector <8 x float> %l2, <8 x float> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
25 %l4 = fmul fast <8 x float> %wide.vec, %wide.vec
26 %l5 = shufflevector <8 x float> %l4, <8 x float> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
27 %l6 = fadd fast <4 x float> %l5, %l3
31 define <4 x float> @vld3(<12 x float>* %pSrc) {
34 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast <12 x float>* [[PSRC:%.*]] to <4 x float>*
35 ; CHECK-NEXT: [[LDN:%.*]] = call { <4 x float>, <4 x float>, <4 x float> } @llvm.aarch64.neon.ld3.v4f32.p0v4f32(<4 x float>* [[TMP0]])
36 ; CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float> } [[LDN]], 2
37 ; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float> } [[LDN]], 2
38 ; CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float> } [[LDN]], 1
39 ; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float> } [[LDN]], 1
40 ; CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float> } [[LDN]], 0
41 ; CHECK-NEXT: [[TMP6:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float> } [[LDN]], 0
42 ; CHECK-NEXT: [[L29:%.*]] = fmul fast <4 x float> [[TMP5]], [[TMP6]]
43 ; CHECK-NEXT: [[L46:%.*]] = fmul fast <4 x float> [[TMP3]], [[TMP4]]
44 ; CHECK-NEXT: [[L6:%.*]] = fadd fast <4 x float> [[L46]], [[L29]]
45 ; CHECK-NEXT: [[L73:%.*]] = fmul fast <4 x float> [[TMP1]], [[TMP2]]
46 ; CHECK-NEXT: [[L9:%.*]] = fadd fast <4 x float> [[L6]], [[L73]]
47 ; CHECK-NEXT: ret <4 x float> [[L9]]
50 %wide.vec = load <12 x float>, <12 x float>* %pSrc, align 4
51 %l2 = fmul fast <12 x float> %wide.vec, %wide.vec
52 %l3 = shufflevector <12 x float> %l2, <12 x float> undef, <4 x i32> <i32 0, i32 3, i32 6, i32 9>
53 %l4 = fmul fast <12 x float> %wide.vec, %wide.vec
54 %l5 = shufflevector <12 x float> %l4, <12 x float> undef, <4 x i32> <i32 1, i32 4, i32 7, i32 10>
55 %l6 = fadd fast <4 x float> %l5, %l3
56 %l7 = fmul fast <12 x float> %wide.vec, %wide.vec
57 %l8 = shufflevector <12 x float> %l7, <12 x float> undef, <4 x i32> <i32 2, i32 5, i32 8, i32 11>
58 %l9 = fadd fast <4 x float> %l6, %l8
62 define <4 x float> @vld4(<16 x float>* %pSrc) {
65 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast <16 x float>* [[PSRC:%.*]] to <4 x float>*
66 ; CHECK-NEXT: [[LDN:%.*]] = call { <4 x float>, <4 x float>, <4 x float>, <4 x float> } @llvm.aarch64.neon.ld4.v4f32.p0v4f32(<4 x float>* [[TMP0]])
67 ; CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float>, <4 x float> } [[LDN]], 3
68 ; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float>, <4 x float> } [[LDN]], 3
69 ; CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float>, <4 x float> } [[LDN]], 2
70 ; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float>, <4 x float> } [[LDN]], 2
71 ; CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float>, <4 x float> } [[LDN]], 1
72 ; CHECK-NEXT: [[TMP6:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float>, <4 x float> } [[LDN]], 1
73 ; CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float>, <4 x float> } [[LDN]], 0
74 ; CHECK-NEXT: [[TMP8:%.*]] = extractvalue { <4 x float>, <4 x float>, <4 x float>, <4 x float> } [[LDN]], 0
75 ; CHECK-NEXT: [[L312:%.*]] = fmul fast <4 x float> [[TMP7]], [[TMP8]]
76 ; CHECK-NEXT: [[L59:%.*]] = fmul fast <4 x float> [[TMP5]], [[TMP6]]
77 ; CHECK-NEXT: [[L7:%.*]] = fadd fast <4 x float> [[L59]], [[L312]]
78 ; CHECK-NEXT: [[L86:%.*]] = fmul fast <4 x float> [[TMP3]], [[TMP4]]
79 ; CHECK-NEXT: [[L103:%.*]] = fmul fast <4 x float> [[TMP1]], [[TMP2]]
80 ; CHECK-NEXT: [[L12:%.*]] = fadd fast <4 x float> [[L103]], [[L86]]
81 ; CHECK-NEXT: ret <4 x float> [[L12]]
84 %wide.vec = load <16 x float>, <16 x float>* %pSrc, align 4
85 %l3 = fmul fast <16 x float> %wide.vec, %wide.vec
86 %l4 = shufflevector <16 x float> %l3, <16 x float> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
87 %l5 = fmul fast <16 x float> %wide.vec, %wide.vec
88 %l6 = shufflevector <16 x float> %l5, <16 x float> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>
89 %l7 = fadd fast <4 x float> %l6, %l4
90 %l8 = fmul fast <16 x float> %wide.vec, %wide.vec
91 %l9 = shufflevector <16 x float> %l8, <16 x float> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>
92 %l10 = fmul fast <16 x float> %wide.vec, %wide.vec
93 %l11 = shufflevector <16 x float> %l10, <16 x float> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>
94 %l12 = fadd fast <4 x float> %l11, %l9
98 define <4 x float> @twosrc(<8 x float>* %pSrc1, <8 x float>* %pSrc2) {
99 ; CHECK-LABEL: @twosrc(
101 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast <8 x float>* [[PSRC1:%.*]] to <4 x float>*
102 ; CHECK-NEXT: [[LDN:%.*]] = call { <4 x float>, <4 x float> } @llvm.aarch64.neon.ld2.v4f32.p0v4f32(<4 x float>* [[TMP0]])
103 ; CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <4 x float>, <4 x float> } [[LDN]], 1
104 ; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <4 x float>, <4 x float> } [[LDN]], 0
105 ; CHECK-NEXT: [[TMP3:%.*]] = bitcast <8 x float>* [[PSRC2:%.*]] to <4 x float>*
106 ; CHECK-NEXT: [[LDN7:%.*]] = call { <4 x float>, <4 x float> } @llvm.aarch64.neon.ld2.v4f32.p0v4f32(<4 x float>* [[TMP3]])
107 ; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <4 x float>, <4 x float> } [[LDN7]], 0
108 ; CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <4 x float>, <4 x float> } [[LDN7]], 1
109 ; CHECK-NEXT: [[L46:%.*]] = fmul fast <4 x float> [[TMP4]], [[TMP2]]
110 ; CHECK-NEXT: [[L63:%.*]] = fmul fast <4 x float> [[TMP5]], [[TMP1]]
111 ; CHECK-NEXT: [[L8:%.*]] = fadd fast <4 x float> [[L63]], [[L46]]
112 ; CHECK-NEXT: ret <4 x float> [[L8]]
115 %wide.vec = load <8 x float>, <8 x float>* %pSrc1, align 4
116 %wide.vec26 = load <8 x float>, <8 x float>* %pSrc2, align 4
117 %l4 = fmul fast <8 x float> %wide.vec26, %wide.vec
118 %l5 = shufflevector <8 x float> %l4, <8 x float> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
119 %l6 = fmul fast <8 x float> %wide.vec26, %wide.vec
120 %l7 = shufflevector <8 x float> %l6, <8 x float> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
121 %l8 = fadd fast <4 x float> %l7, %l5
125 define <4 x float> @twosrc2(<8 x float>* %pSrc1, <8 x float>* %pSrc2) {
126 ; CHECK-LABEL: @twosrc2(
128 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast <8 x float>* [[PSRC1:%.*]] to <4 x float>*
129 ; CHECK-NEXT: [[LDN:%.*]] = call { <4 x float>, <4 x float> } @llvm.aarch64.neon.ld2.v4f32.p0v4f32(<4 x float>* [[TMP0]])
130 ; CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <4 x float>, <4 x float> } [[LDN]], 1
131 ; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <4 x float>, <4 x float> } [[LDN]], 0
132 ; CHECK-NEXT: [[TMP3:%.*]] = bitcast <8 x float>* [[PSRC2:%.*]] to <4 x float>*
133 ; CHECK-NEXT: [[LDN4:%.*]] = call { <4 x float>, <4 x float> } @llvm.aarch64.neon.ld2.v4f32.p0v4f32(<4 x float>* [[TMP3]])
134 ; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <4 x float>, <4 x float> } [[LDN4]], 0
135 ; CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <4 x float>, <4 x float> } [[LDN4]], 1
136 ; CHECK-NEXT: [[L43:%.*]] = fmul fast <4 x float> [[TMP4]], [[TMP2]]
137 ; CHECK-NEXT: [[L6:%.*]] = fmul fast <4 x float> [[TMP5]], [[TMP1]]
138 ; CHECK-NEXT: [[L8:%.*]] = fadd fast <4 x float> [[L6]], [[L43]]
139 ; CHECK-NEXT: ret <4 x float> [[L8]]
142 %wide.vec = load <8 x float>, <8 x float>* %pSrc1, align 4
143 %wide.vec26 = load <8 x float>, <8 x float>* %pSrc2, align 4
144 %l4 = fmul fast <8 x float> %wide.vec26, %wide.vec
145 %l5 = shufflevector <8 x float> %l4, <8 x float> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
146 %s1 = shufflevector <8 x float> %wide.vec26, <8 x float> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
147 %s2 = shufflevector <8 x float> %wide.vec, <8 x float> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
148 %l6 = fmul fast <4 x float> %s1, %s2
149 %l8 = fadd fast <4 x float> %l6, %l5