1 ; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s 2 3 define <8 x i8> @test_vextd(<8 x i8>* %A, <8 x i8>* %B) nounwind { 4 ;CHECK: test_vextd: 5 ;CHECK: vext 6 %tmp1 = load <8 x i8>* %A 7 %tmp2 = load <8 x i8>* %B 8 %tmp3 = shufflevector <8 x i8> %tmp1, <8 x i8> %tmp2, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10> 9 ret <8 x i8> %tmp3 10 } 11 12 define <8 x i8> @test_vextRd(<8 x i8>* %A, <8 x i8>* %B) nounwind { 13 ;CHECK: test_vextRd: 14 ;CHECK: vext 15 %tmp1 = load <8 x i8>* %A 16 %tmp2 = load <8 x i8>* %B 17 %tmp3 = shufflevector <8 x i8> %tmp1, <8 x i8> %tmp2, <8 x i32> <i32 13, i32 14, i32 15, i32 0, i32 1, i32 2, i32 3, i32 4> 18 ret <8 x i8> %tmp3 19 } 20 21 define <16 x i8> @test_vextq(<16 x i8>* %A, <16 x i8>* %B) nounwind { 22 ;CHECK: test_vextq: 23 ;CHECK: vext 24 %tmp1 = load <16 x i8>* %A 25 %tmp2 = load <16 x i8>* %B 26 %tmp3 = shufflevector <16 x i8> %tmp1, <16 x i8> %tmp2, <16 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18> 27 ret <16 x i8> %tmp3 28 } 29 30 define <16 x i8> @test_vextRq(<16 x i8>* %A, <16 x i8>* %B) nounwind { 31 ;CHECK: test_vextRq: 32 ;CHECK: vext 33 %tmp1 = load <16 x i8>* %A 34 %tmp2 = load <16 x i8>* %B 35 %tmp3 = shufflevector <16 x i8> %tmp1, <16 x i8> %tmp2, <16 x i32> <i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6> 36 ret <16 x i8> %tmp3 37 } 38 39 define <4 x i16> @test_vextd16(<4 x i16>* %A, <4 x i16>* %B) nounwind { 40 ;CHECK: test_vextd16: 41 ;CHECK: vext 42 %tmp1 = load <4 x i16>* %A 43 %tmp2 = load <4 x i16>* %B 44 %tmp3 = shufflevector <4 x i16> %tmp1, <4 x i16> %tmp2, <4 x i32> <i32 3, i32 4, i32 5, i32 6> 45 ret <4 x i16> %tmp3 46 } 47 48 define <4 x i32> @test_vextq32(<4 x i32>* %A, <4 x i32>* %B) nounwind { 49 ;CHECK: test_vextq32: 50 ;CHECK: vext 51 %tmp1 = load <4 x i32>* %A 52 %tmp2 = load <4 x i32>* %B 53 %tmp3 = shufflevector <4 x i32> %tmp1, <4 x i32> %tmp2, <4 x i32> <i32 3, i32 4, i32 5, i32 6> 54 ret <4 x i32> %tmp3 55 } 56 57 ; Undef shuffle indices should not prevent matching to VEXT: 58 59 define <8 x i8> @test_vextd_undef(<8 x i8>* %A, <8 x i8>* %B) nounwind { 60 ;CHECK: test_vextd_undef: 61 ;CHECK: vext 62 %tmp1 = load <8 x i8>* %A 63 %tmp2 = load <8 x i8>* %B 64 %tmp3 = shufflevector <8 x i8> %tmp1, <8 x i8> %tmp2, <8 x i32> <i32 3, i32 undef, i32 undef, i32 6, i32 7, i32 8, i32 9, i32 10> 65 ret <8 x i8> %tmp3 66 } 67 68 define <16 x i8> @test_vextRq_undef(<16 x i8>* %A, <16 x i8>* %B) nounwind { 69 ;CHECK: test_vextRq_undef: 70 ;CHECK: vext 71 %tmp1 = load <16 x i8>* %A 72 %tmp2 = load <16 x i8>* %B 73 %tmp3 = shufflevector <16 x i8> %tmp1, <16 x i8> %tmp2, <16 x i32> <i32 23, i32 24, i32 25, i32 26, i32 undef, i32 undef, i32 29, i32 30, i32 31, i32 0, i32 1, i32 2, i32 3, i32 4, i32 undef, i32 6> 74 ret <16 x i8> %tmp3 75 } 76 77 define <16 x i8> @test_vextq_undef_op2(<16 x i8> %a) nounwind { 78 ;CHECK: test_vextq_undef_op2: 79 ;CHECK: vext 80 entry: 81 %tmp1 = shufflevector <16 x i8> %a, <16 x i8> undef, <16 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 0, i32 1> 82 ret <16 x i8> %tmp1 83 } 84 85 define <8 x i8> @test_vextd_undef_op2(<8 x i8> %a) nounwind { 86 ;CHECK: test_vextd_undef_op2: 87 ;CHECK: vext 88 entry: 89 %tmp1 = shufflevector <8 x i8> %a, <8 x i8> undef, <8 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1> 90 ret <8 x i8> %tmp1 91 } 92 93 94 define <16 x i8> @test_vextq_undef_op2_undef(<16 x i8> %a) nounwind { 95 ;CHECK: test_vextq_undef_op2_undef: 96 ;CHECK: vext 97 entry: 98 %tmp1 = shufflevector <16 x i8> %a, <16 x i8> undef, <16 x i32> <i32 2, i32 3, i32 4, i32 undef, i32 undef, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 0, i32 1> 99 ret <16 x i8> %tmp1 100 } 101 102 define <8 x i8> @test_vextd_undef_op2_undef(<8 x i8> %a) nounwind { 103 ;CHECK: test_vextd_undef_op2_undef: 104 ;CHECK: vext 105 entry: 106 %tmp1 = shufflevector <8 x i8> %a, <8 x i8> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 6, i32 7, i32 undef, i32 1> 107 ret <8 x i8> %tmp1 108 } 109 110 ; Tests for ReconstructShuffle function. Indices have to be carefully 111 ; chosen to reach lowering phase as a BUILD_VECTOR. 112 113 ; One vector needs vext, the other can be handled by extract_subvector 114 ; Also checks interleaving of sources is handled correctly. 115 ; Essence: a vext is used on %A and something saner than stack load/store for final result. 116 define <4 x i16> @test_interleaved(<8 x i16>* %A, <8 x i16>* %B) nounwind { 117 ;CHECK: test_interleaved: 118 ;CHECK: vext.16 119 ;CHECK-NOT: vext.16 120 ;CHECK: vzip.16 121 %tmp1 = load <8 x i16>* %A 122 %tmp2 = load <8 x i16>* %B 123 %tmp3 = shufflevector <8 x i16> %tmp1, <8 x i16> %tmp2, <4 x i32> <i32 3, i32 8, i32 5, i32 9> 124 ret <4 x i16> %tmp3 125 } 126 127 ; An undef in the shuffle list should still be optimizable 128 define <4 x i16> @test_undef(<8 x i16>* %A, <8 x i16>* %B) nounwind { 129 ;CHECK: test_undef: 130 ;CHECK: vzip.16 131 %tmp1 = load <8 x i16>* %A 132 %tmp2 = load <8 x i16>* %B 133 %tmp3 = shufflevector <8 x i16> %tmp1, <8 x i16> %tmp2, <4 x i32> <i32 undef, i32 8, i32 5, i32 9> 134 ret <4 x i16> %tmp3 135 } 136 137 ; We should ignore a build_vector with more than two sources. 138 ; Use illegal <32 x i16> type to produce such a shuffle after legalizing types. 139 ; Try to look for fallback to stack expansion. 140 define <4 x i16> @test_multisource(<32 x i16>* %B) nounwind { 141 ;CHECK: test_multisource: 142 ;CHECK: vst1.16 143 %tmp1 = load <32 x i16>* %B 144 %tmp2 = shufflevector <32 x i16> %tmp1, <32 x i16> undef, <4 x i32> <i32 0, i32 8, i32 16, i32 24> 145 ret <4 x i16> %tmp2 146 } 147 148 ; We don't handle shuffles using more than half of a 128-bit vector. 149 ; Again, test for fallback to stack expansion 150 define <4 x i16> @test_largespan(<8 x i16>* %B) nounwind { 151 ;CHECK: test_largespan: 152 ;CHECK: vst1.16 153 %tmp1 = load <8 x i16>* %B 154 %tmp2 = shufflevector <8 x i16> %tmp1, <8 x i16> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> 155 ret <4 x i16> %tmp2 156 } 157 158 ; The actual shuffle code only handles some cases, make sure we check 159 ; this rather than blindly emitting a VECTOR_SHUFFLE (infinite 160 ; lowering loop can result otherwise). 161 define <8 x i16> @test_illegal(<8 x i16>* %A, <8 x i16>* %B) nounwind { 162 ;CHECK: test_illegal: 163 ;CHECK: vst1.16 164 %tmp1 = load <8 x i16>* %A 165 %tmp2 = load <8 x i16>* %B 166 %tmp3 = shufflevector <8 x i16> %tmp1, <8 x i16> %tmp2, <8 x i32> <i32 0, i32 7, i32 5, i32 13, i32 3, i32 2, i32 2, i32 9> 167 ret <8 x i16> %tmp3 168 } 169 170 ; PR11129 171 ; Make sure this doesn't crash 172 define arm_aapcscc void @test_elem_mismatch(<2 x i64>* nocapture %src, <4 x i16>* nocapture %dest) nounwind { 173 ; CHECK: test_elem_mismatch: 174 ; CHECK: vstr 175 %tmp0 = load <2 x i64>* %src, align 16 176 %tmp1 = bitcast <2 x i64> %tmp0 to <4 x i32> 177 %tmp2 = extractelement <4 x i32> %tmp1, i32 0 178 %tmp3 = extractelement <4 x i32> %tmp1, i32 2 179 %tmp4 = trunc i32 %tmp2 to i16 180 %tmp5 = trunc i32 %tmp3 to i16 181 %tmp6 = insertelement <4 x i16> undef, i16 %tmp4, i32 0 182 %tmp7 = insertelement <4 x i16> %tmp6, i16 %tmp5, i32 1 183 store <4 x i16> %tmp7, <4 x i16>* %dest, align 4 184 ret void 185 } 186