1 ; RUN: llc < %s -mattr=+neon | FileCheck %s
2 target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:32-f32:32:32-f64:32:32-v64:64:64-v128:128:128-a0:0:32"
3 target triple = "thumbv7-elf"
5 define i32 @vget_lanes8(ptr %A) nounwind {
6 ;CHECK-LABEL: vget_lanes8:
8 %tmp1 = load <8 x i8>, ptr %A
9 %tmp2 = extractelement <8 x i8> %tmp1, i32 1
10 %tmp3 = sext i8 %tmp2 to i32
14 define i32 @vget_lanes16(ptr %A) nounwind {
15 ;CHECK-LABEL: vget_lanes16:
17 %tmp1 = load <4 x i16>, ptr %A
18 %tmp2 = extractelement <4 x i16> %tmp1, i32 1
19 %tmp3 = sext i16 %tmp2 to i32
23 define i32 @vget_laneu8(ptr %A) nounwind {
24 ;CHECK-LABEL: vget_laneu8:
26 %tmp1 = load <8 x i8>, ptr %A
27 %tmp2 = extractelement <8 x i8> %tmp1, i32 1
28 %tmp3 = zext i8 %tmp2 to i32
32 define i32 @vget_laneu16(ptr %A) nounwind {
33 ;CHECK-LABEL: vget_laneu16:
35 %tmp1 = load <4 x i16>, ptr %A
36 %tmp2 = extractelement <4 x i16> %tmp1, i32 1
37 %tmp3 = zext i16 %tmp2 to i32
41 ; Do a vector add to keep the extraction from being done directly from memory.
42 define i32 @vget_lanei32(ptr %A) nounwind {
43 ;CHECK-LABEL: vget_lanei32:
45 %tmp1 = load <2 x i32>, ptr %A
46 %tmp2 = add <2 x i32> %tmp1, %tmp1
47 %tmp3 = extractelement <2 x i32> %tmp2, i32 1
51 define i32 @vgetQ_lanes8(ptr %A) nounwind {
52 ;CHECK-LABEL: vgetQ_lanes8:
54 %tmp1 = load <16 x i8>, ptr %A
55 %tmp2 = extractelement <16 x i8> %tmp1, i32 1
56 %tmp3 = sext i8 %tmp2 to i32
60 define i32 @vgetQ_lanes16(ptr %A) nounwind {
61 ;CHECK-LABEL: vgetQ_lanes16:
63 %tmp1 = load <8 x i16>, ptr %A
64 %tmp2 = extractelement <8 x i16> %tmp1, i32 1
65 %tmp3 = sext i16 %tmp2 to i32
69 define i32 @vgetQ_laneu8(ptr %A) nounwind {
70 ;CHECK-LABEL: vgetQ_laneu8:
72 %tmp1 = load <16 x i8>, ptr %A
73 %tmp2 = extractelement <16 x i8> %tmp1, i32 1
74 %tmp3 = zext i8 %tmp2 to i32
78 define i32 @vgetQ_laneu16(ptr %A) nounwind {
79 ;CHECK-LABEL: vgetQ_laneu16:
81 %tmp1 = load <8 x i16>, ptr %A
82 %tmp2 = extractelement <8 x i16> %tmp1, i32 1
83 %tmp3 = zext i16 %tmp2 to i32
87 ; Do a vector add to keep the extraction from being done directly from memory.
88 define i32 @vgetQ_lanei32(ptr %A) nounwind {
89 ;CHECK-LABEL: vgetQ_lanei32:
91 %tmp1 = load <4 x i32>, ptr %A
92 %tmp2 = add <4 x i32> %tmp1, %tmp1
93 %tmp3 = extractelement <4 x i32> %tmp2, i32 1
97 define arm_aapcs_vfpcc void @test_vget_laneu16() nounwind {
99 ; CHECK: vmov.u16 r0, d{{.*}}[1]
100 %arg0_uint16x4_t = alloca <4 x i16> ; <ptr> [#uses=1]
101 %out_uint16_t = alloca i16 ; <ptr> [#uses=1]
102 %"alloca point" = bitcast i32 0 to i32 ; <i32> [#uses=0]
103 %0 = load <4 x i16>, ptr %arg0_uint16x4_t, align 8 ; <<4 x i16>> [#uses=1]
104 %1 = extractelement <4 x i16> %0, i32 1 ; <i16> [#uses=1]
106 store i16 %2, ptr %out_uint16_t, align 2
109 return: ; preds = %entry
113 define arm_aapcs_vfpcc void @test_vget_laneu8() nounwind {
115 ; CHECK: vmov.u8 r0, d{{.*}}[1]
116 %arg0_uint8x8_t = alloca <8 x i8> ; <ptr> [#uses=1]
117 %out_uint8_t = alloca i8 ; <ptr> [#uses=1]
118 %"alloca point" = bitcast i32 0 to i32 ; <i32> [#uses=0]
119 %0 = load <8 x i8>, ptr %arg0_uint8x8_t, align 8 ; <<8 x i8>> [#uses=1]
120 %1 = extractelement <8 x i8> %0, i32 1 ; <i8> [#uses=1]
122 store i8 %2, ptr %out_uint8_t, align 1
125 return: ; preds = %entry
129 define arm_aapcs_vfpcc void @test_vgetQ_laneu16() nounwind {
131 ; CHECK: vmov.u16 r0, d{{.*}}[1]
132 %arg0_uint16x8_t = alloca <8 x i16> ; <ptr> [#uses=1]
133 %out_uint16_t = alloca i16 ; <ptr> [#uses=1]
134 %"alloca point" = bitcast i32 0 to i32 ; <i32> [#uses=0]
135 %0 = load <8 x i16>, ptr %arg0_uint16x8_t, align 16 ; <<8 x i16>> [#uses=1]
136 %1 = extractelement <8 x i16> %0, i32 1 ; <i16> [#uses=1]
138 store i16 %2, ptr %out_uint16_t, align 2
141 return: ; preds = %entry
145 define arm_aapcs_vfpcc void @test_vgetQ_laneu8() nounwind {
147 ; CHECK: vmov.u8 r0, d{{.*}}[1]
148 %arg0_uint8x16_t = alloca <16 x i8> ; <ptr> [#uses=1]
149 %out_uint8_t = alloca i8 ; <ptr> [#uses=1]
150 %"alloca point" = bitcast i32 0 to i32 ; <i32> [#uses=0]
151 %0 = load <16 x i8>, ptr %arg0_uint8x16_t, align 16 ; <<16 x i8>> [#uses=1]
152 %1 = extractelement <16 x i8> %0, i32 1 ; <i8> [#uses=1]
154 store i8 %2, ptr %out_uint8_t, align 1
157 return: ; preds = %entry
161 define <8 x i8> @vset_lane8(ptr %A, i8 %B) nounwind {
162 ;CHECK-LABEL: vset_lane8:
164 %tmp1 = load <8 x i8>, ptr %A
165 %tmp2 = insertelement <8 x i8> %tmp1, i8 %B, i32 1
169 define <4 x i16> @vset_lane16(ptr %A, i16 %B) nounwind {
170 ;CHECK-LABEL: vset_lane16:
172 %tmp1 = load <4 x i16>, ptr %A
173 %tmp2 = insertelement <4 x i16> %tmp1, i16 %B, i32 1
177 define <2 x i32> @vset_lane32(ptr %A, i32 %B) nounwind {
178 ;CHECK-LABEL: vset_lane32:
180 %tmp1 = load <2 x i32>, ptr %A
181 %tmp2 = insertelement <2 x i32> %tmp1, i32 %B, i32 1
185 define <16 x i8> @vsetQ_lane8(ptr %A, i8 %B) nounwind {
186 ;CHECK-LABEL: vsetQ_lane8:
188 %tmp1 = load <16 x i8>, ptr %A
189 %tmp2 = insertelement <16 x i8> %tmp1, i8 %B, i32 1
193 define <8 x i16> @vsetQ_lane16(ptr %A, i16 %B) nounwind {
194 ;CHECK-LABEL: vsetQ_lane16:
196 %tmp1 = load <8 x i16>, ptr %A
197 %tmp2 = insertelement <8 x i16> %tmp1, i16 %B, i32 1
201 define <4 x i32> @vsetQ_lane32(ptr %A, i32 %B) nounwind {
202 ;CHECK-LABEL: vsetQ_lane32:
203 ;CHECK: vmov.32 d{{.*}}[1], r1
204 %tmp1 = load <4 x i32>, ptr %A
205 %tmp2 = insertelement <4 x i32> %tmp1, i32 %B, i32 1
209 define arm_aapcs_vfpcc <2 x float> @test_vset_lanef32(float %arg0_float32_t, <2 x float> %arg1_float32x2_t) nounwind {
210 ;CHECK-LABEL: test_vset_lanef32:
211 ;CHECK: vmov.f32 s3, s0
212 ;CHECK: vmov.f64 d0, d1
214 %0 = insertelement <2 x float> %arg1_float32x2_t, float %arg0_float32_t, i32 1 ; <<2 x float>> [#uses=1]
218 ; The llvm extractelement instruction does not require that the lane number
219 ; be an immediate constant. Make sure a variable lane number is handled.
221 define i32 @vget_variable_lanes8(ptr %A, i32 %B) nounwind {
222 %tmp1 = load <8 x i8>, ptr %A
223 %tmp2 = extractelement <8 x i8> %tmp1, i32 %B
224 %tmp3 = sext i8 %tmp2 to i32
228 define i32 @vgetQ_variable_lanei32(ptr %A, i32 %B) nounwind {
229 %tmp1 = load <4 x i32>, ptr %A
230 %tmp2 = add <4 x i32> %tmp1, %tmp1
231 %tmp3 = extractelement <4 x i32> %tmp2, i32 %B