1; RUN: llc < %s -mtriple=arm64-eabi -aarch64-neon-syntax=apple -asm-verbose=false | FileCheck %s 2; RUN: llc < %s -global-isel=1 -mtriple=arm64-eabi -aarch64-neon-syntax=apple -asm-verbose=false | FileCheck %s 3 4define i32 @vmax_u8x8(<8 x i8> %a) nounwind ssp { 5; CHECK-LABEL: vmax_u8x8: 6; CHECK: umaxv.8b b[[REG:[0-9]+]], v0 7; CHECK: fmov [[REG2:w[0-9]+]], s[[REG]] 8; CHECK-NOT: and 9; CHECK: cbz [[REG2]], 10entry: 11 %vmaxv.i = tail call i32 @llvm.aarch64.neon.umaxv.i32.v8i8(<8 x i8> %a) nounwind 12 %tmp = trunc i32 %vmaxv.i to i8 13 %tobool = icmp eq i8 %tmp, 0 14 br i1 %tobool, label %return, label %if.then 15 16if.then: 17 %call1 = tail call i32 @bar() nounwind 18 br label %return 19 20return: 21 %retval.0 = phi i32 [ %call1, %if.then ], [ 0, %entry ] 22 ret i32 %retval.0 23} 24 25declare i32 @bar(...) 26 27define i32 @vmax_u4x16(<4 x i16> %a) nounwind ssp { 28; CHECK-LABEL: vmax_u4x16: 29; CHECK: umaxv.4h h[[REG:[0-9]+]], v0 30; CHECK: fmov [[REG2:w[0-9]+]], s[[REG]] 31; CHECK-NOT: and 32; CHECK: cbz [[REG2]], 33entry: 34 %vmaxv.i = tail call i32 @llvm.aarch64.neon.umaxv.i32.v4i16(<4 x i16> %a) nounwind 35 %tmp = trunc i32 %vmaxv.i to i16 36 %tobool = icmp eq i16 %tmp, 0 37 br i1 %tobool, label %return, label %if.then 38 39if.then: 40 %call1 = tail call i32 @bar() nounwind 41 br label %return 42 43return: 44 %retval.0 = phi i32 [ %call1, %if.then ], [ 0, %entry ] 45 ret i32 %retval.0 46} 47 48define i32 @vmax_u8x16(<8 x i16> %a) nounwind ssp { 49; CHECK-LABEL: vmax_u8x16: 50; CHECK: umaxv.8h h[[REG:[0-9]+]], v0 51; CHECK: fmov [[REG2:w[0-9]+]], s[[REG]] 52; CHECK-NOT: and 53; CHECK: cbz [[REG2]], 54entry: 55 %vmaxv.i = tail call i32 @llvm.aarch64.neon.umaxv.i32.v8i16(<8 x i16> %a) nounwind 56 %tmp = trunc i32 %vmaxv.i to i16 57 %tobool = icmp eq i16 %tmp, 0 58 br i1 %tobool, label %return, label %if.then 59 60if.then: 61 %call1 = tail call i32 @bar() nounwind 62 br label %return 63 64return: 65 %retval.0 = phi i32 [ %call1, %if.then ], [ 0, %entry ] 66 ret i32 %retval.0 67} 68 69define i32 @vmax_u16x8(<16 x i8> %a) nounwind ssp { 70; CHECK-LABEL: vmax_u16x8: 71; CHECK: umaxv.16b b[[REG:[0-9]+]], v0 72; CHECK: fmov [[REG2:w[0-9]+]], s[[REG]] 73; CHECK-NOT: and 74; CHECK: cbz [[REG2]], 75entry: 76 %vmaxv.i = tail call i32 @llvm.aarch64.neon.umaxv.i32.v16i8(<16 x i8> %a) nounwind 77 %tmp = trunc i32 %vmaxv.i to i8 78 %tobool = icmp eq i8 %tmp, 0 79 br i1 %tobool, label %return, label %if.then 80 81if.then: 82 %call1 = tail call i32 @bar() nounwind 83 br label %return 84 85return: 86 %retval.0 = phi i32 [ %call1, %if.then ], [ 0, %entry ] 87 ret i32 %retval.0 88} 89 90define <8 x i8> @test_vmaxv_u8_used_by_laneop(<8 x i8> %a1, <8 x i8> %a2) { 91; CHECK-LABEL: test_vmaxv_u8_used_by_laneop: 92; CHECK: umaxv.8b b[[REGNUM:[0-9]+]], v1 93; CHECK-NEXT: mov.b v0[3], v[[REGNUM]][0] 94; CHECK-NEXT: ret 95entry: 96 %0 = tail call i32 @llvm.aarch64.neon.umaxv.i32.v8i8(<8 x i8> %a2) 97 %1 = trunc i32 %0 to i8 98 %2 = insertelement <8 x i8> %a1, i8 %1, i32 3 99 ret <8 x i8> %2 100} 101 102define <4 x i16> @test_vmaxv_u16_used_by_laneop(<4 x i16> %a1, <4 x i16> %a2) { 103; CHECK-LABEL: test_vmaxv_u16_used_by_laneop: 104; CHECK: umaxv.4h h[[REGNUM:[0-9]+]], v1 105; CHECK-NEXT: mov.h v0[3], v[[REGNUM]][0] 106; CHECK-NEXT: ret 107entry: 108 %0 = tail call i32 @llvm.aarch64.neon.umaxv.i32.v4i16(<4 x i16> %a2) 109 %1 = trunc i32 %0 to i16 110 %2 = insertelement <4 x i16> %a1, i16 %1, i32 3 111 ret <4 x i16> %2 112} 113 114define <2 x i32> @test_vmaxv_u32_used_by_laneop(<2 x i32> %a1, <2 x i32> %a2) { 115; CHECK-LABEL: test_vmaxv_u32_used_by_laneop: 116; CHECK: umaxp.2s v[[REGNUM:[0-9]+]], v1, v1 117; CHECK-NEXT: mov.s v0[1], v[[REGNUM]][0] 118; CHECK-NEXT: ret 119entry: 120 %0 = tail call i32 @llvm.aarch64.neon.umaxv.i32.v2i32(<2 x i32> %a2) 121 %1 = insertelement <2 x i32> %a1, i32 %0, i32 1 122 ret <2 x i32> %1 123} 124 125define <16 x i8> @test_vmaxvq_u8_used_by_laneop(<16 x i8> %a1, <16 x i8> %a2) { 126; CHECK-LABEL: test_vmaxvq_u8_used_by_laneop: 127; CHECK: umaxv.16b b[[REGNUM:[0-9]+]], v1 128; CHECK-NEXT: mov.b v0[3], v[[REGNUM]][0] 129; CHECK-NEXT: ret 130entry: 131 %0 = tail call i32 @llvm.aarch64.neon.umaxv.i32.v16i8(<16 x i8> %a2) 132 %1 = trunc i32 %0 to i8 133 %2 = insertelement <16 x i8> %a1, i8 %1, i32 3 134 ret <16 x i8> %2 135} 136 137define <8 x i16> @test_vmaxvq_u16_used_by_laneop(<8 x i16> %a1, <8 x i16> %a2) { 138; CHECK-LABEL: test_vmaxvq_u16_used_by_laneop: 139; CHECK: umaxv.8h h[[REGNUM:[0-9]+]], v1 140; CHECK-NEXT: mov.h v0[3], v[[REGNUM]][0] 141; CHECK-NEXT: ret 142entry: 143 %0 = tail call i32 @llvm.aarch64.neon.umaxv.i32.v8i16(<8 x i16> %a2) 144 %1 = trunc i32 %0 to i16 145 %2 = insertelement <8 x i16> %a1, i16 %1, i32 3 146 ret <8 x i16> %2 147} 148 149define <4 x i32> @test_vmaxvq_u32_used_by_laneop(<4 x i32> %a1, <4 x i32> %a2) { 150; CHECK-LABEL: test_vmaxvq_u32_used_by_laneop: 151; CHECK: umaxv.4s s[[REGNUM:[0-9]+]], v1 152; CHECK-NEXT: mov.s v0[3], v[[REGNUM]][0] 153; CHECK-NEXT: ret 154entry: 155 %0 = tail call i32 @llvm.aarch64.neon.umaxv.i32.v4i32(<4 x i32> %a2) 156 %1 = insertelement <4 x i32> %a1, i32 %0, i32 3 157 ret <4 x i32> %1 158} 159 160declare i32 @llvm.aarch64.neon.umaxv.i32.v16i8(<16 x i8>) nounwind readnone 161declare i32 @llvm.aarch64.neon.umaxv.i32.v8i16(<8 x i16>) nounwind readnone 162declare i32 @llvm.aarch64.neon.umaxv.i32.v4i16(<4 x i16>) nounwind readnone 163declare i32 @llvm.aarch64.neon.umaxv.i32.v8i8(<8 x i8>) nounwind readnone 164declare i32 @llvm.aarch64.neon.umaxv.i32.v2i32(<2 x i32>) nounwind readnone 165declare i32 @llvm.aarch64.neon.umaxv.i32.v4i32(<4 x i32>) nounwind readnone 166