1 ; RUN: llc -march=arm64 -aarch64-neon-syntax=apple < %s | 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 declare i32 @llvm.aarch64.neon.smaxv.i32.v4i32(<4 x i32>)
69 declare i32 @llvm.aarch64.neon.smaxv.i32.v8i16(<8 x i16>)
70 declare i32 @llvm.aarch64.neon.smaxv.i32.v16i8(<16 x i8>)
71 declare i32 @llvm.aarch64.neon.smaxv.i32.v2i32(<2 x i32>)
72 declare i32 @llvm.aarch64.neon.smaxv.i32.v4i16(<4 x i16>)
73 declare i32 @llvm.aarch64.neon.smaxv.i32.v8i8(<8 x i8>)