1 ; RUN: llc < %s -mtriple=arm64-eabi -aarch64-neon-syntax=apple -asm-verbose=false | FileCheck %s
3 define signext i8 @test_vmaxv_s8(<8 x i8> %a1) {
5 ; CHECK: smaxv.8b b[[REGNUM:[0-9]+]], v0
6 ; CHECK-NEXT: smov.b w0, v[[REGNUM]][0]
9 %vmaxv.i = tail call i32 @llvm.aarch64.neon.smaxv.i32.v8i8(<8 x i8> %a1)
10 %0 = trunc i32 %vmaxv.i to i8
14 define signext i16 @test_vmaxv_s16(<4 x i16> %a1) {
15 ; CHECK: test_vmaxv_s16
16 ; CHECK: smaxv.4h h[[REGNUM:[0-9]+]], v0
17 ; CHECK-NEXT: smov.h w0, v[[REGNUM]][0]
20 %vmaxv.i = tail call i32 @llvm.aarch64.neon.smaxv.i32.v4i16(<4 x i16> %a1)
21 %0 = trunc i32 %vmaxv.i to i16
25 define i32 @test_vmaxv_s32(<2 x i32> %a1) {
26 ; CHECK: test_vmaxv_s32
27 ; 2 x i32 is not supported by the ISA, thus, this is a special case
28 ; CHECK: smaxp.2s v[[REGNUM:[0-9]+]], v0, v0
29 ; CHECK-NEXT: fmov w0, s[[REGNUM]]
32 %vmaxv.i = tail call i32 @llvm.aarch64.neon.smaxv.i32.v2i32(<2 x i32> %a1)
36 define signext i8 @test_vmaxvq_s8(<16 x i8> %a1) {
37 ; CHECK: test_vmaxvq_s8
38 ; CHECK: smaxv.16b b[[REGNUM:[0-9]+]], v0
39 ; CHECK-NEXT: smov.b w0, v[[REGNUM]][0]
42 %vmaxv.i = tail call i32 @llvm.aarch64.neon.smaxv.i32.v16i8(<16 x i8> %a1)
43 %0 = trunc i32 %vmaxv.i to i8
47 define signext i16 @test_vmaxvq_s16(<8 x i16> %a1) {
48 ; CHECK: test_vmaxvq_s16
49 ; CHECK: smaxv.8h h[[REGNUM:[0-9]+]], v0
50 ; CHECK-NEXT: smov.h w0, v[[REGNUM]][0]
53 %vmaxv.i = tail call i32 @llvm.aarch64.neon.smaxv.i32.v8i16(<8 x i16> %a1)
54 %0 = trunc i32 %vmaxv.i to i16
58 define i32 @test_vmaxvq_s32(<4 x i32> %a1) {
59 ; CHECK: test_vmaxvq_s32
60 ; CHECK: smaxv.4s [[REGNUM:s[0-9]+]], v0
61 ; CHECK-NEXT: fmov w0, [[REGNUM]]
64 %vmaxv.i = tail call i32 @llvm.aarch64.neon.smaxv.i32.v4i32(<4 x i32> %a1)
68 define <8 x i8> @test_vmaxv_s8_used_by_laneop(<8 x i8> %a1, <8 x i8> %a2) {
69 ; CHECK-LABEL: test_vmaxv_s8_used_by_laneop:
70 ; CHECK: smaxv.8b b[[REGNUM:[0-9]+]], v1
71 ; CHECK-NEXT: mov.b v0[3], v[[REGNUM]][0]
74 %0 = tail call i32 @llvm.aarch64.neon.smaxv.i32.v8i8(<8 x i8> %a2)
75 %1 = trunc i32 %0 to i8
76 %2 = insertelement <8 x i8> %a1, i8 %1, i32 3
80 define <4 x i16> @test_vmaxv_s16_used_by_laneop(<4 x i16> %a1, <4 x i16> %a2) {
81 ; CHECK-LABEL: test_vmaxv_s16_used_by_laneop:
82 ; CHECK: smaxv.4h h[[REGNUM:[0-9]+]], v1
83 ; CHECK-NEXT: mov.h v0[3], v[[REGNUM]][0]
86 %0 = tail call i32 @llvm.aarch64.neon.smaxv.i32.v4i16(<4 x i16> %a2)
87 %1 = trunc i32 %0 to i16
88 %2 = insertelement <4 x i16> %a1, i16 %1, i32 3
92 define <2 x i32> @test_vmaxv_s32_used_by_laneop(<2 x i32> %a1, <2 x i32> %a2) {
93 ; CHECK-LABEL: test_vmaxv_s32_used_by_laneop:
94 ; CHECK: smaxp.2s v[[REGNUM:[0-9]+]], v1, v1
95 ; CHECK-NEXT: mov.s v0[1], v[[REGNUM]][0]
98 %0 = tail call i32 @llvm.aarch64.neon.smaxv.i32.v2i32(<2 x i32> %a2)
99 %1 = insertelement <2 x i32> %a1, i32 %0, i32 1
103 define <16 x i8> @test_vmaxvq_s8_used_by_laneop(<16 x i8> %a1, <16 x i8> %a2) {
104 ; CHECK-LABEL: test_vmaxvq_s8_used_by_laneop:
105 ; CHECK: smaxv.16b b[[REGNUM:[0-9]+]], v1
106 ; CHECK-NEXT: mov.b v0[3], v[[REGNUM]][0]
109 %0 = tail call i32 @llvm.aarch64.neon.smaxv.i32.v16i8(<16 x i8> %a2)
110 %1 = trunc i32 %0 to i8
111 %2 = insertelement <16 x i8> %a1, i8 %1, i32 3
115 define <8 x i16> @test_vmaxvq_s16_used_by_laneop(<8 x i16> %a1, <8 x i16> %a2) {
116 ; CHECK-LABEL: test_vmaxvq_s16_used_by_laneop:
117 ; CHECK: smaxv.8h h[[REGNUM:[0-9]+]], v1
118 ; CHECK-NEXT: mov.h v0[3], v[[REGNUM]][0]
121 %0 = tail call i32 @llvm.aarch64.neon.smaxv.i32.v8i16(<8 x i16> %a2)
122 %1 = trunc i32 %0 to i16
123 %2 = insertelement <8 x i16> %a1, i16 %1, i32 3
127 define <4 x i32> @test_vmaxvq_s32_used_by_laneop(<4 x i32> %a1, <4 x i32> %a2) {
128 ; CHECK-LABEL: test_vmaxvq_s32_used_by_laneop:
129 ; CHECK: smaxv.4s s[[REGNUM:[0-9]+]], v1
130 ; CHECK-NEXT: mov.s v0[3], v[[REGNUM]][0]
133 %0 = tail call i32 @llvm.aarch64.neon.smaxv.i32.v4i32(<4 x i32> %a2)
134 %1 = insertelement <4 x i32> %a1, i32 %0, i32 3
138 declare i32 @llvm.aarch64.neon.smaxv.i32.v4i32(<4 x i32>)
139 declare i32 @llvm.aarch64.neon.smaxv.i32.v8i16(<8 x i16>)
140 declare i32 @llvm.aarch64.neon.smaxv.i32.v16i8(<16 x i8>)
141 declare i32 @llvm.aarch64.neon.smaxv.i32.v2i32(<2 x i32>)
142 declare i32 @llvm.aarch64.neon.smaxv.i32.v4i16(<4 x i16>)
143 declare i32 @llvm.aarch64.neon.smaxv.i32.v8i8(<8 x i8>)