1 ; RUN: llc < %s -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -mattr=+neon | FileCheck %s
2 ; arm64 has no equivalent vbsl intrinsic, always using the and/or IR. The final
3 ; two tests are duplicated by ARM64's vselect.ll test.
5 declare <2 x double> @llvm.arm.neon.vbsl.v2f64(<2 x double>, <2 x double>, <2 x double>)
7 declare <8 x i16> @llvm.arm.neon.vbsl.v8i16(<8 x i16>, <8 x i16>, <8 x i16>)
9 declare <16 x i8> @llvm.arm.neon.vbsl.v16i8(<16 x i8>, <16 x i8>, <16 x i8>)
11 declare <4 x float> @llvm.arm.neon.vbsl.v4f32(<4 x float>, <4 x float>, <4 x float>)
13 declare <2 x i64> @llvm.arm.neon.vbsl.v2i64(<2 x i64>, <2 x i64>, <2 x i64>)
15 declare <4 x i32> @llvm.arm.neon.vbsl.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
17 declare <4 x i16> @llvm.arm.neon.vbsl.v4i16(<4 x i16>, <4 x i16>, <4 x i16>)
19 declare <8 x i8> @llvm.arm.neon.vbsl.v8i8(<8 x i8>, <8 x i8>, <8 x i8>)
21 declare <1 x double> @llvm.arm.neon.vbsl.v1f64(<1 x double>, <1 x double>, <1 x double>)
23 declare <2 x float> @llvm.arm.neon.vbsl.v2f32(<2 x float>, <2 x float>, <2 x float>)
25 declare <1 x i64> @llvm.arm.neon.vbsl.v1i64(<1 x i64>, <1 x i64>, <1 x i64>)
27 declare <2 x i32> @llvm.arm.neon.vbsl.v2i32(<2 x i32>, <2 x i32>, <2 x i32>)
29 define <8 x i8> @test_vbsl_s8(<8 x i8> %v1, <8 x i8> %v2, <8 x i8> %v3) {
30 ; CHECK-LABEL: test_vbsl_s8:
31 ; CHECK: bsl {{v[0-9]+}}.8b, {{v[0-9]+}}.8b, {{v[0-9]+}}.8b
33 %vbsl.i = tail call <8 x i8> @llvm.arm.neon.vbsl.v8i8(<8 x i8> %v1, <8 x i8> %v2, <8 x i8> %v3)
37 define <8 x i8> @test_vbsl_s16(<4 x i16> %v1, <4 x i16> %v2, <4 x i16> %v3) {
38 ; CHECK-LABEL: test_vbsl_s16:
39 ; CHECK: bsl {{v[0-9]+}}.8b, {{v[0-9]+}}.8b, {{v[0-9]+}}.8b
41 %vbsl3.i = tail call <4 x i16> @llvm.arm.neon.vbsl.v4i16(<4 x i16> %v1, <4 x i16> %v2, <4 x i16> %v3)
42 %0 = bitcast <4 x i16> %vbsl3.i to <8 x i8>
46 define <2 x i32> @test_vbsl_s32(<2 x i32> %v1, <2 x i32> %v2, <2 x i32> %v3) {
47 ; CHECK-LABEL: test_vbsl_s32:
48 ; CHECK: bsl {{v[0-9]+}}.8b, {{v[0-9]+}}.8b, {{v[0-9]+}}.8b
50 %vbsl3.i = tail call <2 x i32> @llvm.arm.neon.vbsl.v2i32(<2 x i32> %v1, <2 x i32> %v2, <2 x i32> %v3)
51 ret <2 x i32> %vbsl3.i
54 define <1 x i64> @test_vbsl_s64(<1 x i64> %v1, <1 x i64> %v2, <1 x i64> %v3) {
55 ; CHECK-LABEL: test_vbsl_s64:
56 ; CHECK: bsl {{v[0-9]+}}.8b, {{v[0-9]+}}.8b, {{v[0-9]+}}.8b
58 %vbsl3.i = tail call <1 x i64> @llvm.arm.neon.vbsl.v1i64(<1 x i64> %v1, <1 x i64> %v2, <1 x i64> %v3)
59 ret <1 x i64> %vbsl3.i
62 define <8 x i8> @test_vbsl_u8(<8 x i8> %v1, <8 x i8> %v2, <8 x i8> %v3) {
63 ; CHECK-LABEL: test_vbsl_u8:
64 ; CHECK: bsl {{v[0-9]+}}.8b, {{v[0-9]+}}.8b, {{v[0-9]+}}.8b
66 %vbsl.i = tail call <8 x i8> @llvm.arm.neon.vbsl.v8i8(<8 x i8> %v1, <8 x i8> %v2, <8 x i8> %v3)
70 define <4 x i16> @test_vbsl_u16(<4 x i16> %v1, <4 x i16> %v2, <4 x i16> %v3) {
71 ; CHECK-LABEL: test_vbsl_u16:
72 ; CHECK: bsl {{v[0-9]+}}.8b, {{v[0-9]+}}.8b, {{v[0-9]+}}.8b
74 %vbsl3.i = tail call <4 x i16> @llvm.arm.neon.vbsl.v4i16(<4 x i16> %v1, <4 x i16> %v2, <4 x i16> %v3)
75 ret <4 x i16> %vbsl3.i
78 define <2 x i32> @test_vbsl_u32(<2 x i32> %v1, <2 x i32> %v2, <2 x i32> %v3) {
79 ; CHECK-LABEL: test_vbsl_u32:
80 ; CHECK: bsl {{v[0-9]+}}.8b, {{v[0-9]+}}.8b, {{v[0-9]+}}.8b
82 %vbsl3.i = tail call <2 x i32> @llvm.arm.neon.vbsl.v2i32(<2 x i32> %v1, <2 x i32> %v2, <2 x i32> %v3)
83 ret <2 x i32> %vbsl3.i
86 define <1 x i64> @test_vbsl_u64(<1 x i64> %v1, <1 x i64> %v2, <1 x i64> %v3) {
87 ; CHECK-LABEL: test_vbsl_u64:
88 ; CHECK: bsl {{v[0-9]+}}.8b, {{v[0-9]+}}.8b, {{v[0-9]+}}.8b
90 %vbsl3.i = tail call <1 x i64> @llvm.arm.neon.vbsl.v1i64(<1 x i64> %v1, <1 x i64> %v2, <1 x i64> %v3)
91 ret <1 x i64> %vbsl3.i
94 define <2 x float> @test_vbsl_f32(<2 x float> %v1, <2 x float> %v2, <2 x float> %v3) {
95 ; CHECK-LABEL: test_vbsl_f32:
96 ; CHECK: bsl {{v[0-9]+}}.8b, {{v[0-9]+}}.8b, {{v[0-9]+}}.8b
98 %vbsl3.i = tail call <2 x float> @llvm.arm.neon.vbsl.v2f32(<2 x float> %v1, <2 x float> %v2, <2 x float> %v3)
99 ret <2 x float> %vbsl3.i
102 define <1 x double> @test_vbsl_f64(<1 x i64> %v1, <1 x double> %v2, <1 x double> %v3) {
103 ; CHECK-LABEL: test_vbsl_f64:
104 ; CHECK: bsl {{v[0-9]+}}.8b, {{v[0-9]+}}.8b, {{v[0-9]+}}.8b
106 %vbsl.i = bitcast <1 x i64> %v1 to <1 x double>
107 %vbsl3.i = tail call <1 x double> @llvm.arm.neon.vbsl.v1f64(<1 x double> %vbsl.i, <1 x double> %v2, <1 x double> %v3)
108 ret <1 x double> %vbsl3.i
111 define <8 x i8> @test_vbsl_p8(<8 x i8> %v1, <8 x i8> %v2, <8 x i8> %v3) {
112 ; CHECK-LABEL: test_vbsl_p8:
113 ; CHECK: bsl {{v[0-9]+}}.8b, {{v[0-9]+}}.8b, {{v[0-9]+}}.8b
115 %vbsl.i = tail call <8 x i8> @llvm.arm.neon.vbsl.v8i8(<8 x i8> %v1, <8 x i8> %v2, <8 x i8> %v3)
119 define <4 x i16> @test_vbsl_p16(<4 x i16> %v1, <4 x i16> %v2, <4 x i16> %v3) {
120 ; CHECK-LABEL: test_vbsl_p16:
121 ; CHECK: bsl {{v[0-9]+}}.8b, {{v[0-9]+}}.8b, {{v[0-9]+}}.8b
123 %vbsl3.i = tail call <4 x i16> @llvm.arm.neon.vbsl.v4i16(<4 x i16> %v1, <4 x i16> %v2, <4 x i16> %v3)
124 ret <4 x i16> %vbsl3.i
127 define <16 x i8> @test_vbslq_s8(<16 x i8> %v1, <16 x i8> %v2, <16 x i8> %v3) {
128 ; CHECK-LABEL: test_vbslq_s8:
129 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
131 %vbsl.i = tail call <16 x i8> @llvm.arm.neon.vbsl.v16i8(<16 x i8> %v1, <16 x i8> %v2, <16 x i8> %v3)
132 ret <16 x i8> %vbsl.i
135 define <8 x i16> @test_vbslq_s16(<8 x i16> %v1, <8 x i16> %v2, <8 x i16> %v3) {
136 ; CHECK-LABEL: test_vbslq_s16:
137 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
139 %vbsl3.i = tail call <8 x i16> @llvm.arm.neon.vbsl.v8i16(<8 x i16> %v1, <8 x i16> %v2, <8 x i16> %v3)
140 ret <8 x i16> %vbsl3.i
143 define <4 x i32> @test_vbslq_s32(<4 x i32> %v1, <4 x i32> %v2, <4 x i32> %v3) {
144 ; CHECK-LABEL: test_vbslq_s32:
145 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
147 %vbsl3.i = tail call <4 x i32> @llvm.arm.neon.vbsl.v4i32(<4 x i32> %v1, <4 x i32> %v2, <4 x i32> %v3)
148 ret <4 x i32> %vbsl3.i
151 define <2 x i64> @test_vbslq_s64(<2 x i64> %v1, <2 x i64> %v2, <2 x i64> %v3) {
152 ; CHECK-LABEL: test_vbslq_s64:
153 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
155 %vbsl3.i = tail call <2 x i64> @llvm.arm.neon.vbsl.v2i64(<2 x i64> %v1, <2 x i64> %v2, <2 x i64> %v3)
156 ret <2 x i64> %vbsl3.i
159 define <16 x i8> @test_vbslq_u8(<16 x i8> %v1, <16 x i8> %v2, <16 x i8> %v3) {
160 ; CHECK-LABEL: test_vbslq_u8:
161 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
163 %vbsl.i = tail call <16 x i8> @llvm.arm.neon.vbsl.v16i8(<16 x i8> %v1, <16 x i8> %v2, <16 x i8> %v3)
164 ret <16 x i8> %vbsl.i
167 define <8 x i16> @test_vbslq_u16(<8 x i16> %v1, <8 x i16> %v2, <8 x i16> %v3) {
168 ; CHECK-LABEL: test_vbslq_u16:
169 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
171 %vbsl3.i = tail call <8 x i16> @llvm.arm.neon.vbsl.v8i16(<8 x i16> %v1, <8 x i16> %v2, <8 x i16> %v3)
172 ret <8 x i16> %vbsl3.i
175 define <4 x i32> @test_vbslq_u32(<4 x i32> %v1, <4 x i32> %v2, <4 x i32> %v3) {
176 ; CHECK-LABEL: test_vbslq_u32:
177 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
179 %vbsl3.i = tail call <4 x i32> @llvm.arm.neon.vbsl.v4i32(<4 x i32> %v1, <4 x i32> %v2, <4 x i32> %v3)
180 ret <4 x i32> %vbsl3.i
183 define <2 x i64> @test_vbslq_u64(<2 x i64> %v1, <2 x i64> %v2, <2 x i64> %v3) {
184 ; CHECK-LABEL: test_vbslq_u64:
185 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
187 %vbsl3.i = tail call <2 x i64> @llvm.arm.neon.vbsl.v2i64(<2 x i64> %v1, <2 x i64> %v2, <2 x i64> %v3)
188 ret <2 x i64> %vbsl3.i
191 define <4 x float> @test_vbslq_f32(<4 x i32> %v1, <4 x float> %v2, <4 x float> %v3) {
192 ; CHECK-LABEL: test_vbslq_f32:
193 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
195 %vbsl.i = bitcast <4 x i32> %v1 to <4 x float>
196 %vbsl3.i = tail call <4 x float> @llvm.arm.neon.vbsl.v4f32(<4 x float> %vbsl.i, <4 x float> %v2, <4 x float> %v3)
197 ret <4 x float> %vbsl3.i
200 define <16 x i8> @test_vbslq_p8(<16 x i8> %v1, <16 x i8> %v2, <16 x i8> %v3) {
201 ; CHECK-LABEL: test_vbslq_p8:
202 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
204 %vbsl.i = tail call <16 x i8> @llvm.arm.neon.vbsl.v16i8(<16 x i8> %v1, <16 x i8> %v2, <16 x i8> %v3)
205 ret <16 x i8> %vbsl.i
208 define <8 x i16> @test_vbslq_p16(<8 x i16> %v1, <8 x i16> %v2, <8 x i16> %v3) {
209 ; CHECK-LABEL: test_vbslq_p16:
210 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
212 %vbsl3.i = tail call <8 x i16> @llvm.arm.neon.vbsl.v8i16(<8 x i16> %v1, <8 x i16> %v2, <8 x i16> %v3)
213 ret <8 x i16> %vbsl3.i
216 define <2 x double> @test_vbslq_f64(<2 x i64> %v1, <2 x double> %v2, <2 x double> %v3) {
217 ; CHECK-LABEL: test_vbslq_f64:
218 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
220 %vbsl.i = bitcast <2 x i64> %v1 to <2 x double>
221 %vbsl3.i = tail call <2 x double> @llvm.arm.neon.vbsl.v2f64(<2 x double> %vbsl.i, <2 x double> %v2, <2 x double> %v3)
222 ret <2 x double> %vbsl3.i
225 define <2 x double> @test_bsl_v2f64(<2 x i1> %v1, <2 x double> %v2, <2 x double> %v3) {
226 ; CHECK-LABEL: test_bsl_v2f64:
227 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
228 %1 = select <2 x i1> %v1, <2 x double> %v2, <2 x double> %v3
232 define <4 x float> @test_bsl_v4f32(<4 x i1> %v1, <4 x float> %v2, <4 x float> %v3) {
233 ; CHECK-LABEL: test_bsl_v4f32:
234 ; CHECK: bsl {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
235 %1 = select <4 x i1> %v1, <4 x float> %v2, <4 x float> %v3