1; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s 2; RUN: llc < %s -march=arm -mattr=+neon -regalloc=basic | FileCheck %s 3 4define <8 x i8> @vcgts8(<8 x i8>* %A, <8 x i8>* %B) nounwind { 5;CHECK: vcgts8: 6;CHECK: vcgt.s8 7 %tmp1 = load <8 x i8>* %A 8 %tmp2 = load <8 x i8>* %B 9 %tmp3 = icmp sgt <8 x i8> %tmp1, %tmp2 10 %tmp4 = sext <8 x i1> %tmp3 to <8 x i8> 11 ret <8 x i8> %tmp4 12} 13 14define <4 x i16> @vcgts16(<4 x i16>* %A, <4 x i16>* %B) nounwind { 15;CHECK: vcgts16: 16;CHECK: vcgt.s16 17 %tmp1 = load <4 x i16>* %A 18 %tmp2 = load <4 x i16>* %B 19 %tmp3 = icmp sgt <4 x i16> %tmp1, %tmp2 20 %tmp4 = sext <4 x i1> %tmp3 to <4 x i16> 21 ret <4 x i16> %tmp4 22} 23 24define <2 x i32> @vcgts32(<2 x i32>* %A, <2 x i32>* %B) nounwind { 25;CHECK: vcgts32: 26;CHECK: vcgt.s32 27 %tmp1 = load <2 x i32>* %A 28 %tmp2 = load <2 x i32>* %B 29 %tmp3 = icmp sgt <2 x i32> %tmp1, %tmp2 30 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32> 31 ret <2 x i32> %tmp4 32} 33 34define <8 x i8> @vcgtu8(<8 x i8>* %A, <8 x i8>* %B) nounwind { 35;CHECK: vcgtu8: 36;CHECK: vcgt.u8 37 %tmp1 = load <8 x i8>* %A 38 %tmp2 = load <8 x i8>* %B 39 %tmp3 = icmp ugt <8 x i8> %tmp1, %tmp2 40 %tmp4 = sext <8 x i1> %tmp3 to <8 x i8> 41 ret <8 x i8> %tmp4 42} 43 44define <4 x i16> @vcgtu16(<4 x i16>* %A, <4 x i16>* %B) nounwind { 45;CHECK: vcgtu16: 46;CHECK: vcgt.u16 47 %tmp1 = load <4 x i16>* %A 48 %tmp2 = load <4 x i16>* %B 49 %tmp3 = icmp ugt <4 x i16> %tmp1, %tmp2 50 %tmp4 = sext <4 x i1> %tmp3 to <4 x i16> 51 ret <4 x i16> %tmp4 52} 53 54define <2 x i32> @vcgtu32(<2 x i32>* %A, <2 x i32>* %B) nounwind { 55;CHECK: vcgtu32: 56;CHECK: vcgt.u32 57 %tmp1 = load <2 x i32>* %A 58 %tmp2 = load <2 x i32>* %B 59 %tmp3 = icmp ugt <2 x i32> %tmp1, %tmp2 60 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32> 61 ret <2 x i32> %tmp4 62} 63 64define <2 x i32> @vcgtf32(<2 x float>* %A, <2 x float>* %B) nounwind { 65;CHECK: vcgtf32: 66;CHECK: vcgt.f32 67 %tmp1 = load <2 x float>* %A 68 %tmp2 = load <2 x float>* %B 69 %tmp3 = fcmp ogt <2 x float> %tmp1, %tmp2 70 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32> 71 ret <2 x i32> %tmp4 72} 73 74define <16 x i8> @vcgtQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind { 75;CHECK: vcgtQs8: 76;CHECK: vcgt.s8 77 %tmp1 = load <16 x i8>* %A 78 %tmp2 = load <16 x i8>* %B 79 %tmp3 = icmp sgt <16 x i8> %tmp1, %tmp2 80 %tmp4 = sext <16 x i1> %tmp3 to <16 x i8> 81 ret <16 x i8> %tmp4 82} 83 84define <8 x i16> @vcgtQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind { 85;CHECK: vcgtQs16: 86;CHECK: vcgt.s16 87 %tmp1 = load <8 x i16>* %A 88 %tmp2 = load <8 x i16>* %B 89 %tmp3 = icmp sgt <8 x i16> %tmp1, %tmp2 90 %tmp4 = sext <8 x i1> %tmp3 to <8 x i16> 91 ret <8 x i16> %tmp4 92} 93 94define <4 x i32> @vcgtQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind { 95;CHECK: vcgtQs32: 96;CHECK: vcgt.s32 97 %tmp1 = load <4 x i32>* %A 98 %tmp2 = load <4 x i32>* %B 99 %tmp3 = icmp sgt <4 x i32> %tmp1, %tmp2 100 %tmp4 = sext <4 x i1> %tmp3 to <4 x i32> 101 ret <4 x i32> %tmp4 102} 103 104define <16 x i8> @vcgtQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind { 105;CHECK: vcgtQu8: 106;CHECK: vcgt.u8 107 %tmp1 = load <16 x i8>* %A 108 %tmp2 = load <16 x i8>* %B 109 %tmp3 = icmp ugt <16 x i8> %tmp1, %tmp2 110 %tmp4 = sext <16 x i1> %tmp3 to <16 x i8> 111 ret <16 x i8> %tmp4 112} 113 114define <8 x i16> @vcgtQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind { 115;CHECK: vcgtQu16: 116;CHECK: vcgt.u16 117 %tmp1 = load <8 x i16>* %A 118 %tmp2 = load <8 x i16>* %B 119 %tmp3 = icmp ugt <8 x i16> %tmp1, %tmp2 120 %tmp4 = sext <8 x i1> %tmp3 to <8 x i16> 121 ret <8 x i16> %tmp4 122} 123 124define <4 x i32> @vcgtQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind { 125;CHECK: vcgtQu32: 126;CHECK: vcgt.u32 127 %tmp1 = load <4 x i32>* %A 128 %tmp2 = load <4 x i32>* %B 129 %tmp3 = icmp ugt <4 x i32> %tmp1, %tmp2 130 %tmp4 = sext <4 x i1> %tmp3 to <4 x i32> 131 ret <4 x i32> %tmp4 132} 133 134define <4 x i32> @vcgtQf32(<4 x float>* %A, <4 x float>* %B) nounwind { 135;CHECK: vcgtQf32: 136;CHECK: vcgt.f32 137 %tmp1 = load <4 x float>* %A 138 %tmp2 = load <4 x float>* %B 139 %tmp3 = fcmp ogt <4 x float> %tmp1, %tmp2 140 %tmp4 = sext <4 x i1> %tmp3 to <4 x i32> 141 ret <4 x i32> %tmp4 142} 143 144define <2 x i32> @vacgtf32(<2 x float>* %A, <2 x float>* %B) nounwind { 145;CHECK: vacgtf32: 146;CHECK: vacgt.f32 147 %tmp1 = load <2 x float>* %A 148 %tmp2 = load <2 x float>* %B 149 %tmp3 = call <2 x i32> @llvm.arm.neon.vacgtd(<2 x float> %tmp1, <2 x float> %tmp2) 150 ret <2 x i32> %tmp3 151} 152 153define <4 x i32> @vacgtQf32(<4 x float>* %A, <4 x float>* %B) nounwind { 154;CHECK: vacgtQf32: 155;CHECK: vacgt.f32 156 %tmp1 = load <4 x float>* %A 157 %tmp2 = load <4 x float>* %B 158 %tmp3 = call <4 x i32> @llvm.arm.neon.vacgtq(<4 x float> %tmp1, <4 x float> %tmp2) 159 ret <4 x i32> %tmp3 160} 161 162; rdar://7923010 163define <4 x i32> @vcgt_zext(<4 x float>* %A, <4 x float>* %B) nounwind { 164;CHECK: vcgt_zext: 165;CHECK: vmov.i32 [[Q0:q[0-9]+]], #0x1 166;CHECK: vcgt.f32 [[Q1:q[0-9]+]] 167;CHECK: vand [[Q2:q[0-9]+]], [[Q1]], [[Q0]] 168 %tmp1 = load <4 x float>* %A 169 %tmp2 = load <4 x float>* %B 170 %tmp3 = fcmp ogt <4 x float> %tmp1, %tmp2 171 %tmp4 = zext <4 x i1> %tmp3 to <4 x i32> 172 ret <4 x i32> %tmp4 173} 174 175declare <2 x i32> @llvm.arm.neon.vacgtd(<2 x float>, <2 x float>) nounwind readnone 176declare <4 x i32> @llvm.arm.neon.vacgtq(<4 x float>, <4 x float>) nounwind readnone 177 178define <8 x i8> @vcgti8Z(<8 x i8>* %A) nounwind { 179;CHECK: vcgti8Z: 180;CHECK-NOT: vmov 181;CHECK-NOT: vmvn 182;CHECK: vcgt.s8 183 %tmp1 = load <8 x i8>* %A 184 %tmp3 = icmp sgt <8 x i8> %tmp1, <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0> 185 %tmp4 = sext <8 x i1> %tmp3 to <8 x i8> 186 ret <8 x i8> %tmp4 187} 188 189define <8 x i8> @vclti8Z(<8 x i8>* %A) nounwind { 190;CHECK: vclti8Z: 191;CHECK-NOT: vmov 192;CHECK-NOT: vmvn 193;CHECK: vclt.s8 194 %tmp1 = load <8 x i8>* %A 195 %tmp3 = icmp slt <8 x i8> %tmp1, <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0> 196 %tmp4 = sext <8 x i1> %tmp3 to <8 x i8> 197 ret <8 x i8> %tmp4 198} 199