1; RUN: llc -mtriple=arm-eabi -mattr=+neon %s -o - -lower-interleaved-accesses=false | FileCheck %s 2 3define <8 x i8> @test_vextd(<8 x i8>* %A, <8 x i8>* %B) nounwind { 4;CHECK-LABEL: test_vextd: 5;CHECK: vext 6 %tmp1 = load <8 x i8>, <8 x i8>* %A 7 %tmp2 = load <8 x i8>, <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 12define <8 x i8> @test_vextRd(<8 x i8>* %A, <8 x i8>* %B) nounwind { 13;CHECK-LABEL: test_vextRd: 14;CHECK: vext 15 %tmp1 = load <8 x i8>, <8 x i8>* %A 16 %tmp2 = load <8 x i8>, <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 21define <16 x i8> @test_vextq(<16 x i8>* %A, <16 x i8>* %B) nounwind { 22;CHECK-LABEL: test_vextq: 23;CHECK: vext 24 %tmp1 = load <16 x i8>, <16 x i8>* %A 25 %tmp2 = load <16 x i8>, <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 30define <16 x i8> @test_vextRq(<16 x i8>* %A, <16 x i8>* %B) nounwind { 31;CHECK-LABEL: test_vextRq: 32;CHECK: vext 33 %tmp1 = load <16 x i8>, <16 x i8>* %A 34 %tmp2 = load <16 x i8>, <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 39define <4 x i16> @test_vextd16(<4 x i16>* %A, <4 x i16>* %B) nounwind { 40;CHECK-LABEL: test_vextd16: 41;CHECK: vext 42 %tmp1 = load <4 x i16>, <4 x i16>* %A 43 %tmp2 = load <4 x i16>, <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 48define <4 x i32> @test_vextq32(<4 x i32>* %A, <4 x i32>* %B) nounwind { 49;CHECK-LABEL: test_vextq32: 50;CHECK: vext 51 %tmp1 = load <4 x i32>, <4 x i32>* %A 52 %tmp2 = load <4 x i32>, <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 59define <8 x i8> @test_vextd_undef(<8 x i8>* %A, <8 x i8>* %B) nounwind { 60;CHECK-LABEL: test_vextd_undef: 61;CHECK: vext 62 %tmp1 = load <8 x i8>, <8 x i8>* %A 63 %tmp2 = load <8 x i8>, <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 68define <16 x i8> @test_vextRq_undef(<16 x i8>* %A, <16 x i8>* %B) nounwind { 69;CHECK-LABEL: test_vextRq_undef: 70;CHECK: vext 71 %tmp1 = load <16 x i8>, <16 x i8>* %A 72 %tmp2 = load <16 x i8>, <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 77define <16 x i8> @test_vextq_undef_op2(<16 x i8> %a) nounwind { 78;CHECK-LABEL: test_vextq_undef_op2: 79;CHECK: vext 80entry: 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 85define <8 x i8> @test_vextd_undef_op2(<8 x i8> %a) nounwind { 86;CHECK-LABEL: test_vextd_undef_op2: 87;CHECK: vext 88entry: 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 94define <16 x i8> @test_vextq_undef_op2_undef(<16 x i8> %a) nounwind { 95;CHECK-LABEL: test_vextq_undef_op2_undef: 96;CHECK: vext 97entry: 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 102define <8 x i8> @test_vextd_undef_op2_undef(<8 x i8> %a) nounwind { 103;CHECK-LABEL: test_vextd_undef_op2_undef: 104;CHECK: vext 105entry: 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. 116define <4 x i16> @test_interleaved(<8 x i16>* %A, <8 x i16>* %B) nounwind { 117;CHECK-LABEL: test_interleaved: 118;CHECK: vext.16 119;CHECK-NOT: vext.16 120;CHECK: vzip.16 121 %tmp1 = load <8 x i16>, <8 x i16>* %A 122 %tmp2 = load <8 x i16>, <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 128define <4 x i16> @test_undef(<8 x i16>* %A, <8 x i16>* %B) nounwind { 129;CHECK-LABEL: test_undef: 130;CHECK: vzip.16 131 %tmp1 = load <8 x i16>, <8 x i16>* %A 132 %tmp2 = load <8 x i16>, <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 by-element inserts. 140define <4 x i16> @test_multisource(<32 x i16>* %B) nounwind { 141;CHECK-LABEL: test_multisource: 142;CHECK: vmov.16 [[REG:d[0-9]+]][0] 143;CHECK: vmov.16 [[REG]][1] 144;CHECK: vmov.16 [[REG]][2] 145;CHECK: vmov.16 [[REG]][3] 146 %tmp1 = load <32 x i16>, <32 x i16>* %B 147 %tmp2 = shufflevector <32 x i16> %tmp1, <32 x i16> undef, <4 x i32> <i32 0, i32 8, i32 16, i32 24> 148 ret <4 x i16> %tmp2 149} 150 151; We don't handle shuffles using more than half of a 128-bit vector. 152; Again, test for fallback to by-element inserts. 153define <4 x i16> @test_largespan(<8 x i16>* %B) nounwind { 154;CHECK-LABEL: test_largespan: 155;CHECK: vmov.16 [[REG:d[0-9]+]][0] 156;CHECK: vmov.16 [[REG]][1] 157;CHECK: vmov.16 [[REG]][2] 158;CHECK: vmov.16 [[REG]][3] 159 %tmp1 = load <8 x i16>, <8 x i16>* %B 160 %tmp2 = shufflevector <8 x i16> %tmp1, <8 x i16> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> 161 ret <4 x i16> %tmp2 162} 163 164; The actual shuffle code only handles some cases, make sure we check 165; this rather than blindly emitting a VECTOR_SHUFFLE (infinite 166; lowering loop can result otherwise). 167define <8 x i16> @test_illegal(<8 x i16>* %A, <8 x i16>* %B) nounwind { 168;CHECK-LABEL: test_illegal: 169;CHECK: vmov.16 [[REG:d[0-9]+]][0] 170;CHECK: vmov.16 [[REG]][1] 171;CHECK: vmov.16 [[REG]][2] 172;CHECK: vmov.16 [[REG]][3] 173;CHECK: vmov.16 [[REG2:d[0-9]+]][0] 174;CHECK: vmov.16 [[REG2]][1] 175;CHECK: vmov.16 [[REG2]][2] 176;CHECK: vmov.16 [[REG2]][3] 177 %tmp1 = load <8 x i16>, <8 x i16>* %A 178 %tmp2 = load <8 x i16>, <8 x i16>* %B 179 %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> 180 ret <8 x i16> %tmp3 181} 182 183; PR11129 184; Make sure this doesn't crash 185define arm_aapcscc void @test_elem_mismatch(<2 x i64>* nocapture %src, <4 x i16>* nocapture %dest) nounwind { 186; CHECK-LABEL: test_elem_mismatch: 187; CHECK: vstr 188 %tmp0 = load <2 x i64>, <2 x i64>* %src, align 16 189 %tmp1 = bitcast <2 x i64> %tmp0 to <4 x i32> 190 %tmp2 = extractelement <4 x i32> %tmp1, i32 0 191 %tmp3 = extractelement <4 x i32> %tmp1, i32 2 192 %tmp4 = trunc i32 %tmp2 to i16 193 %tmp5 = trunc i32 %tmp3 to i16 194 %tmp6 = insertelement <4 x i16> undef, i16 %tmp4, i32 0 195 %tmp7 = insertelement <4 x i16> %tmp6, i16 %tmp5, i32 1 196 store <4 x i16> %tmp7, <4 x i16>* %dest, align 4 197 ret void 198} 199 200define <4 x i32> @test_reverse_and_extract(<2 x i32>* %A) { 201entry: 202 ; CHECK-LABEL: test_reverse_and_extract 203 ; CHECK-NOT: vtrn 204 ; CHECK: vrev 205 ; CHECK: vext 206 %tmp1 = load <2 x i32>, <2 x i32>* %A 207 %0 = shufflevector <2 x i32> %tmp1, <2 x i32> undef, <4 x i32> <i32 undef, i32 undef, i32 1, i32 0> 208 ret <4 x i32> %0 209} 210 211define <4 x i32> @test_dup_and_extract(<2 x i32>* %A) { 212entry: 213 ; CHECK-LABEL: test_dup_and_extract 214 ; CHECK-NOT: vtrn 215 ; CHECK: vdup 216 ; CHECK: vext 217 %tmp1 = load <2 x i32>, <2 x i32>* %A 218 %0 = shufflevector <2 x i32> %tmp1, <2 x i32> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 1> 219 ret <4 x i32> %0 220} 221 222define <4 x i32> @test_zip_and_extract(<2 x i32>* %A) { 223entry: 224 ; CHECK-LABEL: test_zip_and_extract 225 ; CHECK: vzip 226 ; CHECK: vext 227 %tmp1 = load <2 x i32>, <2 x i32>* %A 228 %0 = shufflevector <2 x i32> %tmp1, <2 x i32> undef, <4 x i32> <i32 1, i32 1, i32 0, i32 1> 229 ret <4 x i32> %0 230} 231