Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 1 | //===- IntrinsicsARM.td - Defines ARM intrinsics -----------*- tablegen -*-===// |
| 2 | // |
Andrew Walbran | 16937d0 | 2019-10-22 13:54:20 +0100 | [diff] [blame^] | 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | // |
| 9 | // This file defines all of the ARM-specific intrinsics. |
| 10 | // |
| 11 | //===----------------------------------------------------------------------===// |
| 12 | |
| 13 | |
| 14 | //===----------------------------------------------------------------------===// |
| 15 | // TLS |
| 16 | |
| 17 | let TargetPrefix = "arm" in { // All intrinsics start with "llvm.arm.". |
| 18 | |
| 19 | // A space-consuming intrinsic primarily for testing ARMConstantIslands. The |
| 20 | // first argument is the number of bytes this "instruction" takes up, the second |
| 21 | // and return value are essentially chains, used to force ordering during ISel. |
| 22 | def int_arm_space : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 23 | |
| 24 | // 16-bit multiplications |
| 25 | def int_arm_smulbb : GCCBuiltin<"__builtin_arm_smulbb">, |
| 26 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 27 | def int_arm_smulbt : GCCBuiltin<"__builtin_arm_smulbt">, |
| 28 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 29 | def int_arm_smultb : GCCBuiltin<"__builtin_arm_smultb">, |
| 30 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 31 | def int_arm_smultt : GCCBuiltin<"__builtin_arm_smultt">, |
| 32 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 33 | def int_arm_smulwb : GCCBuiltin<"__builtin_arm_smulwb">, |
| 34 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 35 | def int_arm_smulwt : GCCBuiltin<"__builtin_arm_smulwt">, |
| 36 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 37 | |
| 38 | //===----------------------------------------------------------------------===// |
| 39 | // Saturating Arithmetic |
| 40 | |
| 41 | def int_arm_qadd : GCCBuiltin<"__builtin_arm_qadd">, |
| 42 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], |
| 43 | [Commutative, IntrNoMem]>; |
| 44 | def int_arm_qsub : GCCBuiltin<"__builtin_arm_qsub">, |
| 45 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 46 | def int_arm_ssat : GCCBuiltin<"__builtin_arm_ssat">, |
| 47 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 48 | def int_arm_usat : GCCBuiltin<"__builtin_arm_usat">, |
| 49 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 50 | |
| 51 | // Accumulating multiplications |
| 52 | def int_arm_smlabb : GCCBuiltin<"__builtin_arm_smlabb">, |
| 53 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], |
| 54 | [IntrNoMem]>; |
| 55 | def int_arm_smlabt : GCCBuiltin<"__builtin_arm_smlabt">, |
| 56 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], |
| 57 | [IntrNoMem]>; |
| 58 | def int_arm_smlatb : GCCBuiltin<"__builtin_arm_smlatb">, |
| 59 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], |
| 60 | [IntrNoMem]>; |
| 61 | def int_arm_smlatt : GCCBuiltin<"__builtin_arm_smlatt">, |
| 62 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], |
| 63 | [IntrNoMem]>; |
| 64 | def int_arm_smlawb : GCCBuiltin<"__builtin_arm_smlawb">, |
| 65 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], |
| 66 | [IntrNoMem]>; |
| 67 | def int_arm_smlawt : GCCBuiltin<"__builtin_arm_smlawt">, |
| 68 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], |
| 69 | [IntrNoMem]>; |
| 70 | |
| 71 | // Parallel 16-bit saturation |
| 72 | def int_arm_ssat16 : GCCBuiltin<"__builtin_arm_ssat16">, |
| 73 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 74 | def int_arm_usat16 : GCCBuiltin<"__builtin_arm_usat16">, |
| 75 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 76 | |
| 77 | // Packing and unpacking |
| 78 | def int_arm_sxtab16 : GCCBuiltin<"__builtin_arm_sxtab16">, |
| 79 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 80 | def int_arm_sxtb16 : GCCBuiltin<"__builtin_arm_sxtb16">, |
| 81 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem]>; |
| 82 | def int_arm_uxtab16 : GCCBuiltin<"__builtin_arm_uxtab16">, |
| 83 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 84 | def int_arm_uxtb16 : GCCBuiltin<"__builtin_arm_uxtb16">, |
| 85 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem]>; |
| 86 | |
| 87 | // Parallel selection, reads the GE flags. |
| 88 | def int_arm_sel : GCCBuiltin<"__builtin_arm_sel">, |
| 89 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrReadMem]>; |
| 90 | |
| 91 | // Parallel 8-bit addition and subtraction |
| 92 | def int_arm_qadd8 : GCCBuiltin<"__builtin_arm_qadd8">, |
| 93 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 94 | def int_arm_qsub8 : GCCBuiltin<"__builtin_arm_qsub8">, |
| 95 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 96 | // Writes to the GE bits. |
| 97 | def int_arm_sadd8 : GCCBuiltin<"__builtin_arm_sadd8">, |
| 98 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 99 | def int_arm_shadd8 : GCCBuiltin<"__builtin_arm_shadd8">, |
| 100 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 101 | def int_arm_shsub8 : GCCBuiltin<"__builtin_arm_shsub8">, |
| 102 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 103 | // Writes to the GE bits. |
| 104 | def int_arm_ssub8 : GCCBuiltin<"__builtin_arm_ssub8">, |
| 105 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 106 | // Writes to the GE bits. |
| 107 | def int_arm_uadd8 : GCCBuiltin<"__builtin_arm_uadd8">, |
| 108 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 109 | def int_arm_uhadd8 : GCCBuiltin<"__builtin_arm_uhadd8">, |
| 110 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 111 | def int_arm_uhsub8 : GCCBuiltin<"__builtin_arm_uhsub8">, |
| 112 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 113 | def int_arm_uqadd8 : GCCBuiltin<"__builtin_arm_uqadd8">, |
| 114 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 115 | def int_arm_uqsub8 : GCCBuiltin<"__builtin_arm_uqsub8">, |
| 116 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 117 | // Writes to the GE bits. |
| 118 | def int_arm_usub8 : GCCBuiltin<"__builtin_arm_usub8">, |
| 119 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 120 | |
| 121 | // Sum of 8-bit absolute differences |
| 122 | def int_arm_usad8 : GCCBuiltin<"__builtin_arm_usad8">, |
| 123 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 124 | def int_arm_usada8 : GCCBuiltin<"__builtin_arm_usada8">, |
| 125 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], |
| 126 | [IntrNoMem]>; |
| 127 | |
| 128 | // Parallel 16-bit addition and subtraction |
| 129 | def int_arm_qadd16 : GCCBuiltin<"__builtin_arm_qadd16">, |
| 130 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 131 | def int_arm_qasx : GCCBuiltin<"__builtin_arm_qasx">, |
| 132 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 133 | def int_arm_qsax : GCCBuiltin<"__builtin_arm_qsax">, |
| 134 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 135 | def int_arm_qsub16 : GCCBuiltin<"__builtin_arm_qsub16">, |
| 136 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 137 | // Writes to the GE bits. |
| 138 | def int_arm_sadd16 : GCCBuiltin<"__builtin_arm_sadd16">, |
| 139 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 140 | // Writes to the GE bits. |
| 141 | def int_arm_sasx : GCCBuiltin<"__builtin_arm_sasx">, |
| 142 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 143 | def int_arm_shadd16 : GCCBuiltin<"__builtin_arm_shadd16">, |
| 144 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 145 | def int_arm_shasx : GCCBuiltin<"__builtin_arm_shasx">, |
| 146 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 147 | def int_arm_shsax : GCCBuiltin<"__builtin_arm_shsax">, |
| 148 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 149 | def int_arm_shsub16 : GCCBuiltin<"__builtin_arm_shsub16">, |
| 150 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 151 | // Writes to the GE bits. |
| 152 | def int_arm_ssax : GCCBuiltin<"__builtin_arm_ssax">, |
| 153 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 154 | // Writes to the GE bits. |
| 155 | def int_arm_ssub16 : GCCBuiltin<"__builtin_arm_ssub16">, |
| 156 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 157 | // Writes to the GE bits. |
| 158 | def int_arm_uadd16 : GCCBuiltin<"__builtin_arm_uadd16">, |
| 159 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 160 | // Writes to the GE bits. |
| 161 | def int_arm_uasx : GCCBuiltin<"__builtin_arm_uasx">, |
| 162 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 163 | def int_arm_uhadd16 : GCCBuiltin<"__builtin_arm_uhadd16">, |
| 164 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 165 | def int_arm_uhasx : GCCBuiltin<"__builtin_arm_uhasx">, |
| 166 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 167 | def int_arm_uhsax : GCCBuiltin<"__builtin_arm_uhsax">, |
| 168 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 169 | def int_arm_uhsub16 : GCCBuiltin<"__builtin_arm_uhsub16">, |
| 170 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 171 | def int_arm_uqadd16 : GCCBuiltin<"__builtin_arm_uqadd16">, |
| 172 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 173 | def int_arm_uqasx : GCCBuiltin<"__builtin_arm_uqasx">, |
| 174 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 175 | def int_arm_uqsax : GCCBuiltin<"__builtin_arm_uqsax">, |
| 176 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 177 | def int_arm_uqsub16 : GCCBuiltin<"__builtin_arm_uqsub16">, |
| 178 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 179 | // Writes to the GE bits. |
| 180 | def int_arm_usax : GCCBuiltin<"__builtin_arm_usax">, |
| 181 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 182 | // Writes to the GE bits. |
| 183 | def int_arm_usub16 : GCCBuiltin<"__builtin_arm_usub16">, |
| 184 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], []>; |
| 185 | |
| 186 | // Parallel 16-bit multiplication |
| 187 | def int_arm_smlad : GCCBuiltin<"__builtin_arm_smlad">, |
| 188 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], |
| 189 | [IntrNoMem]>; |
| 190 | def int_arm_smladx : GCCBuiltin<"__builtin_arm_smladx">, |
| 191 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], |
| 192 | [IntrNoMem]>; |
| 193 | def int_arm_smlald : GCCBuiltin<"__builtin_arm_smlald">, |
| 194 | Intrinsic<[llvm_i64_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i64_ty], |
| 195 | [IntrNoMem]>; |
| 196 | def int_arm_smlaldx : GCCBuiltin<"__builtin_arm_smlaldx">, |
| 197 | Intrinsic<[llvm_i64_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i64_ty], |
| 198 | [IntrNoMem]>; |
| 199 | def int_arm_smlsd : GCCBuiltin<"__builtin_arm_smlsd">, |
| 200 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], |
| 201 | [IntrNoMem]>; |
| 202 | def int_arm_smlsdx : GCCBuiltin<"__builtin_arm_smlsdx">, |
| 203 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], |
| 204 | [IntrNoMem]>; |
| 205 | def int_arm_smlsld : GCCBuiltin<"__builtin_arm_smlsld">, |
| 206 | Intrinsic<[llvm_i64_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i64_ty], |
| 207 | [IntrNoMem]>; |
| 208 | def int_arm_smlsldx : GCCBuiltin<"__builtin_arm_smlsldx">, |
| 209 | Intrinsic<[llvm_i64_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i64_ty], |
| 210 | [IntrNoMem]>; |
| 211 | def int_arm_smuad : GCCBuiltin<"__builtin_arm_smuad">, |
| 212 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 213 | def int_arm_smuadx : GCCBuiltin<"__builtin_arm_smuadx">, |
| 214 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 215 | def int_arm_smusd : GCCBuiltin<"__builtin_arm_smusd">, |
| 216 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 217 | def int_arm_smusdx : GCCBuiltin<"__builtin_arm_smusdx">, |
| 218 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>; |
| 219 | |
| 220 | |
| 221 | //===----------------------------------------------------------------------===// |
| 222 | // Load, Store and Clear exclusive |
| 223 | |
| 224 | def int_arm_ldrex : Intrinsic<[llvm_i32_ty], [llvm_anyptr_ty]>; |
| 225 | def int_arm_strex : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_anyptr_ty]>; |
| 226 | |
| 227 | def int_arm_ldaex : Intrinsic<[llvm_i32_ty], [llvm_anyptr_ty]>; |
| 228 | def int_arm_stlex : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_anyptr_ty]>; |
| 229 | |
| 230 | def int_arm_clrex : Intrinsic<[]>; |
| 231 | |
| 232 | def int_arm_strexd : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, |
| 233 | llvm_ptr_ty]>; |
| 234 | def int_arm_ldrexd : Intrinsic<[llvm_i32_ty, llvm_i32_ty], [llvm_ptr_ty]>; |
| 235 | |
| 236 | def int_arm_stlexd : Intrinsic<[llvm_i32_ty], |
| 237 | [llvm_i32_ty, llvm_i32_ty, llvm_ptr_ty]>; |
| 238 | def int_arm_ldaexd : Intrinsic<[llvm_i32_ty, llvm_i32_ty], [llvm_ptr_ty]>; |
| 239 | |
| 240 | //===----------------------------------------------------------------------===// |
| 241 | // Data barrier instructions |
| 242 | def int_arm_dmb : GCCBuiltin<"__builtin_arm_dmb">, MSBuiltin<"__dmb">, |
| 243 | Intrinsic<[], [llvm_i32_ty]>; |
| 244 | def int_arm_dsb : GCCBuiltin<"__builtin_arm_dsb">, MSBuiltin<"__dsb">, |
| 245 | Intrinsic<[], [llvm_i32_ty]>; |
| 246 | def int_arm_isb : GCCBuiltin<"__builtin_arm_isb">, MSBuiltin<"__isb">, |
| 247 | Intrinsic<[], [llvm_i32_ty]>; |
| 248 | |
| 249 | //===----------------------------------------------------------------------===// |
| 250 | // VFP |
| 251 | |
| 252 | def int_arm_get_fpscr : GCCBuiltin<"__builtin_arm_get_fpscr">, |
| 253 | Intrinsic<[llvm_i32_ty], [], []>; |
| 254 | def int_arm_set_fpscr : GCCBuiltin<"__builtin_arm_set_fpscr">, |
| 255 | Intrinsic<[], [llvm_i32_ty], []>; |
| 256 | def int_arm_vcvtr : Intrinsic<[llvm_float_ty], [llvm_anyfloat_ty], |
| 257 | [IntrNoMem]>; |
| 258 | def int_arm_vcvtru : Intrinsic<[llvm_float_ty], [llvm_anyfloat_ty], |
| 259 | [IntrNoMem]>; |
| 260 | |
| 261 | //===----------------------------------------------------------------------===// |
| 262 | // Coprocessor |
| 263 | |
| 264 | def int_arm_ldc : GCCBuiltin<"__builtin_arm_ldc">, |
| 265 | Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_ptr_ty], []>; |
| 266 | def int_arm_ldcl : GCCBuiltin<"__builtin_arm_ldcl">, |
| 267 | Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_ptr_ty], []>; |
| 268 | def int_arm_ldc2 : GCCBuiltin<"__builtin_arm_ldc2">, |
| 269 | Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_ptr_ty], []>; |
| 270 | def int_arm_ldc2l : GCCBuiltin<"__builtin_arm_ldc2l">, |
| 271 | Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_ptr_ty], []>; |
| 272 | |
| 273 | def int_arm_stc : GCCBuiltin<"__builtin_arm_stc">, |
| 274 | Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_ptr_ty], []>; |
| 275 | def int_arm_stcl : GCCBuiltin<"__builtin_arm_stcl">, |
| 276 | Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_ptr_ty], []>; |
Andrew Scull | cdfcccc | 2018-10-05 20:58:37 +0100 | [diff] [blame] | 277 | def int_arm_stc2 : GCCBuiltin<"__builtin_arm_stc2">, |
Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 278 | Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_ptr_ty], []>; |
| 279 | def int_arm_stc2l : GCCBuiltin<"__builtin_arm_stc2l">, |
| 280 | Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_ptr_ty], []>; |
| 281 | |
| 282 | // Move to coprocessor |
| 283 | def int_arm_mcr : GCCBuiltin<"__builtin_arm_mcr">, |
| 284 | Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, |
| 285 | llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], []>; |
| 286 | def int_arm_mcr2 : GCCBuiltin<"__builtin_arm_mcr2">, |
| 287 | Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, |
| 288 | llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], []>; |
| 289 | |
| 290 | // Move from coprocessor |
| 291 | def int_arm_mrc : GCCBuiltin<"__builtin_arm_mrc">, |
| 292 | MSBuiltin<"_MoveFromCoprocessor">, |
| 293 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, |
| 294 | llvm_i32_ty, llvm_i32_ty], []>; |
| 295 | def int_arm_mrc2 : GCCBuiltin<"__builtin_arm_mrc2">, |
| 296 | MSBuiltin<"_MoveFromCoprocessor2">, |
| 297 | Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, |
| 298 | llvm_i32_ty, llvm_i32_ty], []>; |
| 299 | |
| 300 | // Coprocessor data processing |
| 301 | def int_arm_cdp : GCCBuiltin<"__builtin_arm_cdp">, |
| 302 | Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, |
| 303 | llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], []>; |
| 304 | def int_arm_cdp2 : GCCBuiltin<"__builtin_arm_cdp2">, |
| 305 | Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, |
| 306 | llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], []>; |
| 307 | |
| 308 | // Move from two registers to coprocessor |
| 309 | def int_arm_mcrr : Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, |
| 310 | llvm_i32_ty, llvm_i32_ty], []>; |
| 311 | def int_arm_mcrr2 : Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, |
| 312 | llvm_i32_ty, llvm_i32_ty], []>; |
| 313 | |
| 314 | def int_arm_mrrc : Intrinsic<[llvm_i32_ty, llvm_i32_ty], [llvm_i32_ty, |
| 315 | llvm_i32_ty, llvm_i32_ty], []>; |
| 316 | def int_arm_mrrc2 : Intrinsic<[llvm_i32_ty, llvm_i32_ty], [llvm_i32_ty, |
| 317 | llvm_i32_ty, llvm_i32_ty], []>; |
| 318 | |
| 319 | //===----------------------------------------------------------------------===// |
| 320 | // CRC32 |
| 321 | |
| 322 | def int_arm_crc32b : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], |
| 323 | [IntrNoMem]>; |
| 324 | def int_arm_crc32cb : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], |
| 325 | [IntrNoMem]>; |
| 326 | def int_arm_crc32h : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], |
| 327 | [IntrNoMem]>; |
| 328 | def int_arm_crc32ch : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], |
| 329 | [IntrNoMem]>; |
| 330 | def int_arm_crc32w : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], |
| 331 | [IntrNoMem]>; |
| 332 | def int_arm_crc32cw : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], |
| 333 | [IntrNoMem]>; |
| 334 | |
| 335 | //===----------------------------------------------------------------------===// |
| 336 | // HINT |
| 337 | |
| 338 | def int_arm_hint : Intrinsic<[], [llvm_i32_ty]>; |
| 339 | def int_arm_dbg : Intrinsic<[], [llvm_i32_ty]>; |
| 340 | |
| 341 | //===----------------------------------------------------------------------===// |
| 342 | // UND (reserved undefined sequence) |
| 343 | |
| 344 | def int_arm_undefined : Intrinsic<[], [llvm_i32_ty]>; |
| 345 | |
| 346 | //===----------------------------------------------------------------------===// |
| 347 | // Advanced SIMD (NEON) |
| 348 | |
| 349 | // The following classes do not correspond directly to GCC builtins. |
| 350 | class Neon_1Arg_Intrinsic |
| 351 | : Intrinsic<[llvm_anyvector_ty], [LLVMMatchType<0>], [IntrNoMem]>; |
| 352 | class Neon_1Arg_Narrow_Intrinsic |
| 353 | : Intrinsic<[llvm_anyvector_ty], [LLVMExtendedType<0>], [IntrNoMem]>; |
| 354 | class Neon_2Arg_Intrinsic |
| 355 | : Intrinsic<[llvm_anyvector_ty], [LLVMMatchType<0>, LLVMMatchType<0>], |
| 356 | [IntrNoMem]>; |
| 357 | class Neon_2Arg_Narrow_Intrinsic |
| 358 | : Intrinsic<[llvm_anyvector_ty], [LLVMExtendedType<0>, LLVMExtendedType<0>], |
| 359 | [IntrNoMem]>; |
| 360 | class Neon_2Arg_Long_Intrinsic |
| 361 | : Intrinsic<[llvm_anyvector_ty], [LLVMTruncatedType<0>, LLVMTruncatedType<0>], |
| 362 | [IntrNoMem]>; |
| 363 | class Neon_3Arg_Intrinsic |
| 364 | : Intrinsic<[llvm_anyvector_ty], |
| 365 | [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>], |
| 366 | [IntrNoMem]>; |
| 367 | class Neon_3Arg_Long_Intrinsic |
| 368 | : Intrinsic<[llvm_anyvector_ty], |
| 369 | [LLVMMatchType<0>, LLVMTruncatedType<0>, LLVMTruncatedType<0>], |
| 370 | [IntrNoMem]>; |
Andrew Scull | cdfcccc | 2018-10-05 20:58:37 +0100 | [diff] [blame] | 371 | |
| 372 | class Neon_1FloatArg_Intrinsic |
| 373 | : Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>; |
| 374 | |
Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 375 | class Neon_CvtFxToFP_Intrinsic |
| 376 | : Intrinsic<[llvm_anyfloat_ty], [llvm_anyint_ty, llvm_i32_ty], [IntrNoMem]>; |
| 377 | class Neon_CvtFPToFx_Intrinsic |
| 378 | : Intrinsic<[llvm_anyint_ty], [llvm_anyfloat_ty, llvm_i32_ty], [IntrNoMem]>; |
| 379 | class Neon_CvtFPtoInt_1Arg_Intrinsic |
| 380 | : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; |
| 381 | |
| 382 | class Neon_Compare_Intrinsic |
| 383 | : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty, LLVMMatchType<1>], |
| 384 | [IntrNoMem]>; |
| 385 | |
| 386 | // The table operands for VTBL and VTBX consist of 1 to 4 v8i8 vectors. |
| 387 | // Besides the table, VTBL has one other v8i8 argument and VTBX has two. |
| 388 | // Overall, the classes range from 2 to 6 v8i8 arguments. |
| 389 | class Neon_Tbl2Arg_Intrinsic |
| 390 | : Intrinsic<[llvm_v8i8_ty], |
| 391 | [llvm_v8i8_ty, llvm_v8i8_ty], [IntrNoMem]>; |
| 392 | class Neon_Tbl3Arg_Intrinsic |
| 393 | : Intrinsic<[llvm_v8i8_ty], |
| 394 | [llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty], [IntrNoMem]>; |
| 395 | class Neon_Tbl4Arg_Intrinsic |
| 396 | : Intrinsic<[llvm_v8i8_ty], |
| 397 | [llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty], |
| 398 | [IntrNoMem]>; |
| 399 | class Neon_Tbl5Arg_Intrinsic |
| 400 | : Intrinsic<[llvm_v8i8_ty], |
| 401 | [llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty, |
| 402 | llvm_v8i8_ty], [IntrNoMem]>; |
| 403 | class Neon_Tbl6Arg_Intrinsic |
| 404 | : Intrinsic<[llvm_v8i8_ty], |
| 405 | [llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty, |
| 406 | llvm_v8i8_ty, llvm_v8i8_ty], [IntrNoMem]>; |
| 407 | |
| 408 | // Arithmetic ops |
| 409 | |
| 410 | let IntrProperties = [IntrNoMem, Commutative] in { |
| 411 | |
| 412 | // Vector Add. |
| 413 | def int_arm_neon_vhadds : Neon_2Arg_Intrinsic; |
| 414 | def int_arm_neon_vhaddu : Neon_2Arg_Intrinsic; |
| 415 | def int_arm_neon_vrhadds : Neon_2Arg_Intrinsic; |
| 416 | def int_arm_neon_vrhaddu : Neon_2Arg_Intrinsic; |
| 417 | def int_arm_neon_vqadds : Neon_2Arg_Intrinsic; |
| 418 | def int_arm_neon_vqaddu : Neon_2Arg_Intrinsic; |
| 419 | def int_arm_neon_vraddhn : Neon_2Arg_Narrow_Intrinsic; |
| 420 | |
| 421 | // Vector Multiply. |
| 422 | def int_arm_neon_vmulp : Neon_2Arg_Intrinsic; |
| 423 | def int_arm_neon_vqdmulh : Neon_2Arg_Intrinsic; |
| 424 | def int_arm_neon_vqrdmulh : Neon_2Arg_Intrinsic; |
| 425 | def int_arm_neon_vmulls : Neon_2Arg_Long_Intrinsic; |
| 426 | def int_arm_neon_vmullu : Neon_2Arg_Long_Intrinsic; |
| 427 | def int_arm_neon_vmullp : Neon_2Arg_Long_Intrinsic; |
| 428 | def int_arm_neon_vqdmull : Neon_2Arg_Long_Intrinsic; |
| 429 | |
| 430 | // Vector Maximum. |
| 431 | def int_arm_neon_vmaxs : Neon_2Arg_Intrinsic; |
| 432 | def int_arm_neon_vmaxu : Neon_2Arg_Intrinsic; |
| 433 | def int_arm_neon_vmaxnm : Neon_2Arg_Intrinsic; |
| 434 | |
| 435 | // Vector Minimum. |
| 436 | def int_arm_neon_vmins : Neon_2Arg_Intrinsic; |
| 437 | def int_arm_neon_vminu : Neon_2Arg_Intrinsic; |
| 438 | def int_arm_neon_vminnm : Neon_2Arg_Intrinsic; |
| 439 | |
| 440 | // Vector Reciprocal Step. |
| 441 | def int_arm_neon_vrecps : Neon_2Arg_Intrinsic; |
| 442 | |
| 443 | // Vector Reciprocal Square Root Step. |
| 444 | def int_arm_neon_vrsqrts : Neon_2Arg_Intrinsic; |
| 445 | } |
| 446 | |
| 447 | // Vector Subtract. |
| 448 | def int_arm_neon_vhsubs : Neon_2Arg_Intrinsic; |
| 449 | def int_arm_neon_vhsubu : Neon_2Arg_Intrinsic; |
| 450 | def int_arm_neon_vqsubs : Neon_2Arg_Intrinsic; |
| 451 | def int_arm_neon_vqsubu : Neon_2Arg_Intrinsic; |
| 452 | def int_arm_neon_vrsubhn : Neon_2Arg_Narrow_Intrinsic; |
| 453 | |
| 454 | // Vector Absolute Compare. |
| 455 | def int_arm_neon_vacge : Neon_Compare_Intrinsic; |
| 456 | def int_arm_neon_vacgt : Neon_Compare_Intrinsic; |
| 457 | |
| 458 | // Vector Absolute Differences. |
| 459 | def int_arm_neon_vabds : Neon_2Arg_Intrinsic; |
| 460 | def int_arm_neon_vabdu : Neon_2Arg_Intrinsic; |
| 461 | |
| 462 | // Vector Pairwise Add. |
| 463 | def int_arm_neon_vpadd : Neon_2Arg_Intrinsic; |
| 464 | |
| 465 | // Vector Pairwise Add Long. |
| 466 | // Note: This is different than the other "long" NEON intrinsics because |
| 467 | // the result vector has half as many elements as the source vector. |
| 468 | // The source and destination vector types must be specified separately. |
| 469 | def int_arm_neon_vpaddls : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], |
| 470 | [IntrNoMem]>; |
| 471 | def int_arm_neon_vpaddlu : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], |
| 472 | [IntrNoMem]>; |
| 473 | |
| 474 | // Vector Pairwise Add and Accumulate Long. |
| 475 | // Note: This is similar to vpaddl but the destination vector also appears |
| 476 | // as the first argument. |
| 477 | def int_arm_neon_vpadals : Intrinsic<[llvm_anyvector_ty], |
| 478 | [LLVMMatchType<0>, llvm_anyvector_ty], |
| 479 | [IntrNoMem]>; |
| 480 | def int_arm_neon_vpadalu : Intrinsic<[llvm_anyvector_ty], |
| 481 | [LLVMMatchType<0>, llvm_anyvector_ty], |
| 482 | [IntrNoMem]>; |
| 483 | |
| 484 | // Vector Pairwise Maximum and Minimum. |
| 485 | def int_arm_neon_vpmaxs : Neon_2Arg_Intrinsic; |
| 486 | def int_arm_neon_vpmaxu : Neon_2Arg_Intrinsic; |
| 487 | def int_arm_neon_vpmins : Neon_2Arg_Intrinsic; |
| 488 | def int_arm_neon_vpminu : Neon_2Arg_Intrinsic; |
| 489 | |
| 490 | // Vector Shifts: |
| 491 | // |
| 492 | // The various saturating and rounding vector shift operations need to be |
| 493 | // represented by intrinsics in LLVM, and even the basic VSHL variable shift |
| 494 | // operation cannot be safely translated to LLVM's shift operators. VSHL can |
| 495 | // be used for both left and right shifts, or even combinations of the two, |
| 496 | // depending on the signs of the shift amounts. It also has well-defined |
| 497 | // behavior for shift amounts that LLVM leaves undefined. Only basic shifts |
| 498 | // by constants can be represented with LLVM's shift operators. |
| 499 | // |
| 500 | // The shift counts for these intrinsics are always vectors, even for constant |
| 501 | // shifts, where the constant is replicated. For consistency with VSHL (and |
| 502 | // other variable shift instructions), left shifts have positive shift counts |
| 503 | // and right shifts have negative shift counts. This convention is also used |
| 504 | // for constant right shift intrinsics, and to help preserve sanity, the |
| 505 | // intrinsic names use "shift" instead of either "shl" or "shr". Where |
| 506 | // applicable, signed and unsigned versions of the intrinsics are |
| 507 | // distinguished with "s" and "u" suffixes. A few NEON shift instructions, |
| 508 | // such as VQSHLU, take signed operands but produce unsigned results; these |
| 509 | // use a "su" suffix. |
| 510 | |
| 511 | // Vector Shift. |
| 512 | def int_arm_neon_vshifts : Neon_2Arg_Intrinsic; |
| 513 | def int_arm_neon_vshiftu : Neon_2Arg_Intrinsic; |
| 514 | |
| 515 | // Vector Rounding Shift. |
| 516 | def int_arm_neon_vrshifts : Neon_2Arg_Intrinsic; |
| 517 | def int_arm_neon_vrshiftu : Neon_2Arg_Intrinsic; |
| 518 | def int_arm_neon_vrshiftn : Neon_2Arg_Narrow_Intrinsic; |
| 519 | |
| 520 | // Vector Saturating Shift. |
| 521 | def int_arm_neon_vqshifts : Neon_2Arg_Intrinsic; |
| 522 | def int_arm_neon_vqshiftu : Neon_2Arg_Intrinsic; |
| 523 | def int_arm_neon_vqshiftsu : Neon_2Arg_Intrinsic; |
| 524 | def int_arm_neon_vqshiftns : Neon_2Arg_Narrow_Intrinsic; |
| 525 | def int_arm_neon_vqshiftnu : Neon_2Arg_Narrow_Intrinsic; |
| 526 | def int_arm_neon_vqshiftnsu : Neon_2Arg_Narrow_Intrinsic; |
| 527 | |
| 528 | // Vector Saturating Rounding Shift. |
| 529 | def int_arm_neon_vqrshifts : Neon_2Arg_Intrinsic; |
| 530 | def int_arm_neon_vqrshiftu : Neon_2Arg_Intrinsic; |
| 531 | def int_arm_neon_vqrshiftns : Neon_2Arg_Narrow_Intrinsic; |
| 532 | def int_arm_neon_vqrshiftnu : Neon_2Arg_Narrow_Intrinsic; |
| 533 | def int_arm_neon_vqrshiftnsu : Neon_2Arg_Narrow_Intrinsic; |
| 534 | |
| 535 | // Vector Shift and Insert. |
| 536 | def int_arm_neon_vshiftins : Neon_3Arg_Intrinsic; |
| 537 | |
| 538 | // Vector Absolute Value and Saturating Absolute Value. |
| 539 | def int_arm_neon_vabs : Neon_1Arg_Intrinsic; |
| 540 | def int_arm_neon_vqabs : Neon_1Arg_Intrinsic; |
| 541 | |
| 542 | // Vector Saturating Negate. |
| 543 | def int_arm_neon_vqneg : Neon_1Arg_Intrinsic; |
| 544 | |
| 545 | // Vector Count Leading Sign/Zero Bits. |
| 546 | def int_arm_neon_vcls : Neon_1Arg_Intrinsic; |
| 547 | |
| 548 | // Vector Reciprocal Estimate. |
| 549 | def int_arm_neon_vrecpe : Neon_1Arg_Intrinsic; |
| 550 | |
| 551 | // Vector Reciprocal Square Root Estimate. |
| 552 | def int_arm_neon_vrsqrte : Neon_1Arg_Intrinsic; |
| 553 | |
| 554 | // Vector Conversions Between Floating-point and Integer |
| 555 | def int_arm_neon_vcvtau : Neon_CvtFPtoInt_1Arg_Intrinsic; |
| 556 | def int_arm_neon_vcvtas : Neon_CvtFPtoInt_1Arg_Intrinsic; |
| 557 | def int_arm_neon_vcvtnu : Neon_CvtFPtoInt_1Arg_Intrinsic; |
| 558 | def int_arm_neon_vcvtns : Neon_CvtFPtoInt_1Arg_Intrinsic; |
| 559 | def int_arm_neon_vcvtpu : Neon_CvtFPtoInt_1Arg_Intrinsic; |
| 560 | def int_arm_neon_vcvtps : Neon_CvtFPtoInt_1Arg_Intrinsic; |
| 561 | def int_arm_neon_vcvtmu : Neon_CvtFPtoInt_1Arg_Intrinsic; |
| 562 | def int_arm_neon_vcvtms : Neon_CvtFPtoInt_1Arg_Intrinsic; |
| 563 | |
| 564 | // Vector Conversions Between Floating-point and Fixed-point. |
| 565 | def int_arm_neon_vcvtfp2fxs : Neon_CvtFPToFx_Intrinsic; |
| 566 | def int_arm_neon_vcvtfp2fxu : Neon_CvtFPToFx_Intrinsic; |
| 567 | def int_arm_neon_vcvtfxs2fp : Neon_CvtFxToFP_Intrinsic; |
| 568 | def int_arm_neon_vcvtfxu2fp : Neon_CvtFxToFP_Intrinsic; |
| 569 | |
| 570 | // Vector Conversions Between Half-Precision and Single-Precision. |
| 571 | def int_arm_neon_vcvtfp2hf |
| 572 | : Intrinsic<[llvm_v4i16_ty], [llvm_v4f32_ty], [IntrNoMem]>; |
| 573 | def int_arm_neon_vcvthf2fp |
| 574 | : Intrinsic<[llvm_v4f32_ty], [llvm_v4i16_ty], [IntrNoMem]>; |
| 575 | |
| 576 | // Narrowing Saturating Vector Moves. |
| 577 | def int_arm_neon_vqmovns : Neon_1Arg_Narrow_Intrinsic; |
| 578 | def int_arm_neon_vqmovnu : Neon_1Arg_Narrow_Intrinsic; |
| 579 | def int_arm_neon_vqmovnsu : Neon_1Arg_Narrow_Intrinsic; |
| 580 | |
| 581 | // Vector Table Lookup. |
| 582 | // The first 1-4 arguments are the table. |
| 583 | def int_arm_neon_vtbl1 : Neon_Tbl2Arg_Intrinsic; |
| 584 | def int_arm_neon_vtbl2 : Neon_Tbl3Arg_Intrinsic; |
| 585 | def int_arm_neon_vtbl3 : Neon_Tbl4Arg_Intrinsic; |
| 586 | def int_arm_neon_vtbl4 : Neon_Tbl5Arg_Intrinsic; |
| 587 | |
| 588 | // Vector Table Extension. |
| 589 | // Some elements of the destination vector may not be updated, so the original |
| 590 | // value of that vector is passed as the first argument. The next 1-4 |
| 591 | // arguments after that are the table. |
| 592 | def int_arm_neon_vtbx1 : Neon_Tbl3Arg_Intrinsic; |
| 593 | def int_arm_neon_vtbx2 : Neon_Tbl4Arg_Intrinsic; |
| 594 | def int_arm_neon_vtbx3 : Neon_Tbl5Arg_Intrinsic; |
| 595 | def int_arm_neon_vtbx4 : Neon_Tbl6Arg_Intrinsic; |
| 596 | |
Andrew Scull | cdfcccc | 2018-10-05 20:58:37 +0100 | [diff] [blame] | 597 | // Vector and Scalar Rounding. |
| 598 | def int_arm_neon_vrintn : Neon_1FloatArg_Intrinsic; |
Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 599 | def int_arm_neon_vrintx : Neon_1Arg_Intrinsic; |
| 600 | def int_arm_neon_vrinta : Neon_1Arg_Intrinsic; |
| 601 | def int_arm_neon_vrintz : Neon_1Arg_Intrinsic; |
| 602 | def int_arm_neon_vrintm : Neon_1Arg_Intrinsic; |
| 603 | def int_arm_neon_vrintp : Neon_1Arg_Intrinsic; |
| 604 | |
| 605 | // De-interleaving vector loads from N-element structures. |
| 606 | // Source operands are the address and alignment. |
| 607 | def int_arm_neon_vld1 : Intrinsic<[llvm_anyvector_ty], |
| 608 | [llvm_anyptr_ty, llvm_i32_ty], |
| 609 | [IntrReadMem, IntrArgMemOnly]>; |
| 610 | def int_arm_neon_vld2 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>], |
| 611 | [llvm_anyptr_ty, llvm_i32_ty], |
| 612 | [IntrReadMem, IntrArgMemOnly]>; |
| 613 | def int_arm_neon_vld3 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>, |
| 614 | LLVMMatchType<0>], |
| 615 | [llvm_anyptr_ty, llvm_i32_ty], |
| 616 | [IntrReadMem, IntrArgMemOnly]>; |
| 617 | def int_arm_neon_vld4 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>, |
| 618 | LLVMMatchType<0>, LLVMMatchType<0>], |
| 619 | [llvm_anyptr_ty, llvm_i32_ty], |
| 620 | [IntrReadMem, IntrArgMemOnly]>; |
| 621 | |
Andrew Scull | cdfcccc | 2018-10-05 20:58:37 +0100 | [diff] [blame] | 622 | def int_arm_neon_vld1x2 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>], |
| 623 | [LLVMAnyPointerType<LLVMMatchType<0>>], |
| 624 | [IntrReadMem, IntrArgMemOnly]>; |
| 625 | def int_arm_neon_vld1x3 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>, |
| 626 | LLVMMatchType<0>], |
| 627 | [LLVMAnyPointerType<LLVMMatchType<0>>], |
| 628 | [IntrReadMem, IntrArgMemOnly]>; |
| 629 | def int_arm_neon_vld1x4 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>, |
| 630 | LLVMMatchType<0>, LLVMMatchType<0>], |
| 631 | [LLVMAnyPointerType<LLVMMatchType<0>>], |
| 632 | [IntrReadMem, IntrArgMemOnly]>; |
| 633 | |
Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 634 | // Vector load N-element structure to one lane. |
| 635 | // Source operands are: the address, the N input vectors (since only one |
| 636 | // lane is assigned), the lane number, and the alignment. |
| 637 | def int_arm_neon_vld2lane : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>], |
| 638 | [llvm_anyptr_ty, LLVMMatchType<0>, |
| 639 | LLVMMatchType<0>, llvm_i32_ty, |
| 640 | llvm_i32_ty], [IntrReadMem, IntrArgMemOnly]>; |
| 641 | def int_arm_neon_vld3lane : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>, |
| 642 | LLVMMatchType<0>], |
| 643 | [llvm_anyptr_ty, LLVMMatchType<0>, |
| 644 | LLVMMatchType<0>, LLVMMatchType<0>, |
| 645 | llvm_i32_ty, llvm_i32_ty], |
| 646 | [IntrReadMem, IntrArgMemOnly]>; |
| 647 | def int_arm_neon_vld4lane : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>, |
| 648 | LLVMMatchType<0>, LLVMMatchType<0>], |
| 649 | [llvm_anyptr_ty, LLVMMatchType<0>, |
| 650 | LLVMMatchType<0>, LLVMMatchType<0>, |
| 651 | LLVMMatchType<0>, llvm_i32_ty, |
| 652 | llvm_i32_ty], [IntrReadMem, IntrArgMemOnly]>; |
| 653 | |
Andrew Scull | cdfcccc | 2018-10-05 20:58:37 +0100 | [diff] [blame] | 654 | // Vector load N-element structure to all lanes. |
| 655 | // Source operands are the address and alignment. |
| 656 | def int_arm_neon_vld2dup : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>], |
| 657 | [llvm_anyptr_ty, llvm_i32_ty], |
| 658 | [IntrReadMem, IntrArgMemOnly]>; |
| 659 | def int_arm_neon_vld3dup : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>, |
| 660 | LLVMMatchType<0>], |
| 661 | [llvm_anyptr_ty, llvm_i32_ty], |
| 662 | [IntrReadMem, IntrArgMemOnly]>; |
| 663 | def int_arm_neon_vld4dup : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>, |
| 664 | LLVMMatchType<0>, LLVMMatchType<0>], |
| 665 | [llvm_anyptr_ty, llvm_i32_ty], |
| 666 | [IntrReadMem, IntrArgMemOnly]>; |
| 667 | |
Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 668 | // Interleaving vector stores from N-element structures. |
| 669 | // Source operands are: the address, the N vectors, and the alignment. |
| 670 | def int_arm_neon_vst1 : Intrinsic<[], |
| 671 | [llvm_anyptr_ty, llvm_anyvector_ty, |
| 672 | llvm_i32_ty], [IntrArgMemOnly]>; |
| 673 | def int_arm_neon_vst2 : Intrinsic<[], |
| 674 | [llvm_anyptr_ty, llvm_anyvector_ty, |
| 675 | LLVMMatchType<1>, llvm_i32_ty], |
| 676 | [IntrArgMemOnly]>; |
| 677 | def int_arm_neon_vst3 : Intrinsic<[], |
| 678 | [llvm_anyptr_ty, llvm_anyvector_ty, |
| 679 | LLVMMatchType<1>, LLVMMatchType<1>, |
| 680 | llvm_i32_ty], [IntrArgMemOnly]>; |
| 681 | def int_arm_neon_vst4 : Intrinsic<[], |
| 682 | [llvm_anyptr_ty, llvm_anyvector_ty, |
| 683 | LLVMMatchType<1>, LLVMMatchType<1>, |
| 684 | LLVMMatchType<1>, llvm_i32_ty], |
| 685 | [IntrArgMemOnly]>; |
| 686 | |
Andrew Scull | cdfcccc | 2018-10-05 20:58:37 +0100 | [diff] [blame] | 687 | def int_arm_neon_vst1x2 : Intrinsic<[], |
| 688 | [llvm_anyptr_ty, llvm_anyvector_ty, |
| 689 | LLVMMatchType<1>], |
| 690 | [IntrArgMemOnly, NoCapture<0>]>; |
| 691 | def int_arm_neon_vst1x3 : Intrinsic<[], |
| 692 | [llvm_anyptr_ty, llvm_anyvector_ty, |
| 693 | LLVMMatchType<1>, LLVMMatchType<1>], |
| 694 | [IntrArgMemOnly, NoCapture<0>]>; |
| 695 | def int_arm_neon_vst1x4 : Intrinsic<[], |
| 696 | [llvm_anyptr_ty, llvm_anyvector_ty, |
| 697 | LLVMMatchType<1>, LLVMMatchType<1>, |
| 698 | LLVMMatchType<1>], |
| 699 | [IntrArgMemOnly, NoCapture<0>]>; |
| 700 | |
Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 701 | // Vector store N-element structure from one lane. |
| 702 | // Source operands are: the address, the N vectors, the lane number, and |
| 703 | // the alignment. |
| 704 | def int_arm_neon_vst2lane : Intrinsic<[], |
| 705 | [llvm_anyptr_ty, llvm_anyvector_ty, |
| 706 | LLVMMatchType<1>, llvm_i32_ty, |
| 707 | llvm_i32_ty], [IntrArgMemOnly]>; |
| 708 | def int_arm_neon_vst3lane : Intrinsic<[], |
| 709 | [llvm_anyptr_ty, llvm_anyvector_ty, |
| 710 | LLVMMatchType<1>, LLVMMatchType<1>, |
| 711 | llvm_i32_ty, llvm_i32_ty], |
| 712 | [IntrArgMemOnly]>; |
| 713 | def int_arm_neon_vst4lane : Intrinsic<[], |
| 714 | [llvm_anyptr_ty, llvm_anyvector_ty, |
| 715 | LLVMMatchType<1>, LLVMMatchType<1>, |
| 716 | LLVMMatchType<1>, llvm_i32_ty, |
| 717 | llvm_i32_ty], [IntrArgMemOnly]>; |
| 718 | |
| 719 | // Vector bitwise select. |
| 720 | def int_arm_neon_vbsl : Intrinsic<[llvm_anyvector_ty], |
| 721 | [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>], |
| 722 | [IntrNoMem]>; |
| 723 | |
| 724 | |
| 725 | // Crypto instructions |
| 726 | class AES_1Arg_Intrinsic : Intrinsic<[llvm_v16i8_ty], |
| 727 | [llvm_v16i8_ty], [IntrNoMem]>; |
| 728 | class AES_2Arg_Intrinsic : Intrinsic<[llvm_v16i8_ty], |
| 729 | [llvm_v16i8_ty, llvm_v16i8_ty], |
| 730 | [IntrNoMem]>; |
| 731 | |
| 732 | class SHA_1Arg_Intrinsic : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], |
| 733 | [IntrNoMem]>; |
| 734 | class SHA_2Arg_Intrinsic : Intrinsic<[llvm_v4i32_ty], |
| 735 | [llvm_v4i32_ty, llvm_v4i32_ty], |
| 736 | [IntrNoMem]>; |
| 737 | class SHA_3Arg_i32_Intrinsic : Intrinsic<[llvm_v4i32_ty], |
| 738 | [llvm_v4i32_ty, llvm_i32_ty, llvm_v4i32_ty], |
| 739 | [IntrNoMem]>; |
| 740 | class SHA_3Arg_v4i32_Intrinsic : Intrinsic<[llvm_v4i32_ty], |
| 741 | [llvm_v4i32_ty, llvm_v4i32_ty,llvm_v4i32_ty], |
| 742 | [IntrNoMem]>; |
| 743 | |
| 744 | def int_arm_neon_aesd : AES_2Arg_Intrinsic; |
| 745 | def int_arm_neon_aese : AES_2Arg_Intrinsic; |
| 746 | def int_arm_neon_aesimc : AES_1Arg_Intrinsic; |
| 747 | def int_arm_neon_aesmc : AES_1Arg_Intrinsic; |
| 748 | def int_arm_neon_sha1h : SHA_1Arg_Intrinsic; |
| 749 | def int_arm_neon_sha1su1 : SHA_2Arg_Intrinsic; |
| 750 | def int_arm_neon_sha256su0 : SHA_2Arg_Intrinsic; |
| 751 | def int_arm_neon_sha1c : SHA_3Arg_i32_Intrinsic; |
| 752 | def int_arm_neon_sha1m : SHA_3Arg_i32_Intrinsic; |
| 753 | def int_arm_neon_sha1p : SHA_3Arg_i32_Intrinsic; |
| 754 | def int_arm_neon_sha1su0: SHA_3Arg_v4i32_Intrinsic; |
| 755 | def int_arm_neon_sha256h: SHA_3Arg_v4i32_Intrinsic; |
| 756 | def int_arm_neon_sha256h2: SHA_3Arg_v4i32_Intrinsic; |
| 757 | def int_arm_neon_sha256su1: SHA_3Arg_v4i32_Intrinsic; |
| 758 | |
Andrew Scull | cdfcccc | 2018-10-05 20:58:37 +0100 | [diff] [blame] | 759 | // Armv8.2-A dot product instructions |
| 760 | class Neon_Dot_Intrinsic |
| 761 | : Intrinsic<[llvm_anyvector_ty], |
| 762 | [LLVMMatchType<0>, llvm_anyvector_ty, |
| 763 | LLVMMatchType<1>], |
| 764 | [IntrNoMem]>; |
| 765 | def int_arm_neon_udot : Neon_Dot_Intrinsic; |
| 766 | def int_arm_neon_sdot : Neon_Dot_Intrinsic; |
| 767 | |
| 768 | |
Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 769 | } // end TargetPrefix |