1//===-- SystemZInstrInfo.td - General SystemZ instructions ----*- tblgen-*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9 10//===----------------------------------------------------------------------===// 11// Stack allocation 12//===----------------------------------------------------------------------===// 13 14def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i64imm:$amt), 15 [(callseq_start timm:$amt)]>; 16def ADJCALLSTACKUP : Pseudo<(outs), (ins i64imm:$amt1, i64imm:$amt2), 17 [(callseq_end timm:$amt1, timm:$amt2)]>; 18 19let hasSideEffects = 0 in { 20 // Takes as input the value of the stack pointer after a dynamic allocation 21 // has been made. Sets the output to the address of the dynamically- 22 // allocated area itself, skipping the outgoing arguments. 23 // 24 // This expands to an LA or LAY instruction. We restrict the offset 25 // to the range of LA and keep the LAY range in reserve for when 26 // the size of the outgoing arguments is added. 27 def ADJDYNALLOC : Pseudo<(outs GR64:$dst), (ins dynalloc12only:$src), 28 [(set GR64:$dst, dynalloc12only:$src)]>; 29} 30 31//===----------------------------------------------------------------------===// 32// Control flow instructions 33//===----------------------------------------------------------------------===// 34 35// A return instruction (br %r14). 36let isReturn = 1, isTerminator = 1, isBarrier = 1, hasCtrlDep = 1 in 37 def Return : Alias<2, (outs), (ins), [(z_retflag)]>; 38 39// Unconditional branches. R1 is the condition-code mask (all 1s). 40let isBranch = 1, isTerminator = 1, isBarrier = 1, R1 = 15 in { 41 let isIndirectBranch = 1 in 42 def BR : InstRR<0x07, (outs), (ins ADDR64:$R2), 43 "br\t$R2", [(brind ADDR64:$R2)]>; 44 45 // An assembler extended mnemonic for BRC. 46 def J : InstRI<0xA74, (outs), (ins brtarget16:$I2), "j\t$I2", 47 [(br bb:$I2)]>; 48 49 // An assembler extended mnemonic for BRCL. (The extension is "G" 50 // rather than "L" because "JL" is "Jump if Less".) 51 def JG : InstRIL<0xC04, (outs), (ins brtarget32:$I2), "jg\t$I2", []>; 52} 53 54// Conditional branches. It's easier for LLVM to handle these branches 55// in their raw BRC/BRCL form, with the 4-bit condition-code mask being 56// the first operand. It seems friendlier to use mnemonic forms like 57// JE and JLH when writing out the assembly though. 58let isBranch = 1, isTerminator = 1, Uses = [CC] in { 59 let isCodeGenOnly = 1, CCMaskFirst = 1 in { 60 def BRC : InstRI<0xA74, (outs), (ins cond4:$valid, cond4:$R1, 61 brtarget16:$I2), "j$R1\t$I2", 62 [(z_br_ccmask cond4:$valid, cond4:$R1, bb:$I2)]>; 63 def BRCL : InstRIL<0xC04, (outs), (ins cond4:$valid, cond4:$R1, 64 brtarget32:$I2), "jg$R1\t$I2", []>; 65 } 66 def AsmBRC : InstRI<0xA74, (outs), (ins imm32zx4:$R1, brtarget16:$I2), 67 "brc\t$R1, $I2", []>; 68 def AsmBRCL : InstRIL<0xC04, (outs), (ins imm32zx4:$R1, brtarget32:$I2), 69 "brcl\t$R1, $I2", []>; 70 def AsmBCR : InstRR<0x07, (outs), (ins imm32zx4:$R1, GR64:$R2), 71 "bcr\t$R1, $R2", []>; 72} 73 74// Fused compare-and-branch instructions. As for normal branches, 75// we handle these instructions internally in their raw CRJ-like form, 76// but use assembly macros like CRJE when writing them out. 77// 78// These instructions do not use or clobber the condition codes. 79// We nevertheless pretend that they clobber CC, so that we can lower 80// them to separate comparisons and BRCLs if the branch ends up being 81// out of range. 82multiclass CompareBranches<Operand ccmask, string pos1, string pos2> { 83 let isBranch = 1, isTerminator = 1, Defs = [CC] in { 84 def RJ : InstRIEb<0xEC76, (outs), (ins GR32:$R1, GR32:$R2, ccmask:$M3, 85 brtarget16:$RI4), 86 "crj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>; 87 def GRJ : InstRIEb<0xEC64, (outs), (ins GR64:$R1, GR64:$R2, ccmask:$M3, 88 brtarget16:$RI4), 89 "cgrj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>; 90 def IJ : InstRIEc<0xEC7E, (outs), (ins GR32:$R1, imm32sx8:$I2, ccmask:$M3, 91 brtarget16:$RI4), 92 "cij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>; 93 def GIJ : InstRIEc<0xEC7C, (outs), (ins GR64:$R1, imm64sx8:$I2, ccmask:$M3, 94 brtarget16:$RI4), 95 "cgij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>; 96 def LRJ : InstRIEb<0xEC77, (outs), (ins GR32:$R1, GR32:$R2, ccmask:$M3, 97 brtarget16:$RI4), 98 "clrj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>; 99 def LGRJ : InstRIEb<0xEC65, (outs), (ins GR64:$R1, GR64:$R2, ccmask:$M3, 100 brtarget16:$RI4), 101 "clgrj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>; 102 def LIJ : InstRIEc<0xEC7F, (outs), (ins GR32:$R1, imm32zx8:$I2, ccmask:$M3, 103 brtarget16:$RI4), 104 "clij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>; 105 def LGIJ : InstRIEc<0xEC7D, (outs), (ins GR64:$R1, imm64zx8:$I2, ccmask:$M3, 106 brtarget16:$RI4), 107 "clgij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>; 108 } 109} 110let isCodeGenOnly = 1 in 111 defm C : CompareBranches<cond4, "$M3", "">; 112defm AsmC : CompareBranches<imm32zx4, "", "$M3, ">; 113 114// Define AsmParser mnemonics for each general condition-code mask 115// (integer or floating-point) 116multiclass CondExtendedMnemonic<bits<4> ccmask, string name> { 117 let isBranch = 1, isTerminator = 1, R1 = ccmask in { 118 def J : InstRI<0xA74, (outs), (ins brtarget16:$I2), 119 "j"##name##"\t$I2", []>; 120 def JG : InstRIL<0xC04, (outs), (ins brtarget32:$I2), 121 "jg"##name##"\t$I2", []>; 122 def BR : InstRR<0x07, (outs), (ins ADDR64:$R2), "b"##name##"r\t$R2", []>; 123 } 124 def LOCR : FixedCondUnaryRRF<"locr"##name, 0xB9F2, GR32, GR32, ccmask>; 125 def LOCGR : FixedCondUnaryRRF<"locgr"##name, 0xB9E2, GR64, GR64, ccmask>; 126 def LOC : FixedCondUnaryRSY<"loc"##name, 0xEBF2, GR32, ccmask, 4>; 127 def LOCG : FixedCondUnaryRSY<"locg"##name, 0xEBE2, GR64, ccmask, 8>; 128 def STOC : FixedCondStoreRSY<"stoc"##name, 0xEBF3, GR32, ccmask, 4>; 129 def STOCG : FixedCondStoreRSY<"stocg"##name, 0xEBE3, GR64, ccmask, 8>; 130} 131defm AsmO : CondExtendedMnemonic<1, "o">; 132defm AsmH : CondExtendedMnemonic<2, "h">; 133defm AsmNLE : CondExtendedMnemonic<3, "nle">; 134defm AsmL : CondExtendedMnemonic<4, "l">; 135defm AsmNHE : CondExtendedMnemonic<5, "nhe">; 136defm AsmLH : CondExtendedMnemonic<6, "lh">; 137defm AsmNE : CondExtendedMnemonic<7, "ne">; 138defm AsmE : CondExtendedMnemonic<8, "e">; 139defm AsmNLH : CondExtendedMnemonic<9, "nlh">; 140defm AsmHE : CondExtendedMnemonic<10, "he">; 141defm AsmNL : CondExtendedMnemonic<11, "nl">; 142defm AsmLE : CondExtendedMnemonic<12, "le">; 143defm AsmNH : CondExtendedMnemonic<13, "nh">; 144defm AsmNO : CondExtendedMnemonic<14, "no">; 145 146// Define AsmParser mnemonics for each integer condition-code mask. 147// This is like the list above, except that condition 3 is not possible 148// and that the low bit of the mask is therefore always 0. This means 149// that each condition has two names. Conditions "o" and "no" are not used. 150// 151// We don't make one of the two names an alias of the other because 152// we need the custom parsing routines to select the correct register class. 153multiclass IntCondExtendedMnemonicA<bits<4> ccmask, string name> { 154 let isBranch = 1, isTerminator = 1, M3 = ccmask in { 155 def CR : InstRIEb<0xEC76, (outs), (ins GR32:$R1, GR32:$R2, 156 brtarget16:$RI4), 157 "crj"##name##"\t$R1, $R2, $RI4", []>; 158 def CGR : InstRIEb<0xEC64, (outs), (ins GR64:$R1, GR64:$R2, 159 brtarget16:$RI4), 160 "cgrj"##name##"\t$R1, $R2, $RI4", []>; 161 def CI : InstRIEc<0xEC7E, (outs), (ins GR32:$R1, imm32sx8:$I2, 162 brtarget16:$RI4), 163 "cij"##name##"\t$R1, $I2, $RI4", []>; 164 def CGI : InstRIEc<0xEC7C, (outs), (ins GR64:$R1, imm64sx8:$I2, 165 brtarget16:$RI4), 166 "cgij"##name##"\t$R1, $I2, $RI4", []>; 167 def CLR : InstRIEb<0xEC77, (outs), (ins GR32:$R1, GR32:$R2, 168 brtarget16:$RI4), 169 "clrj"##name##"\t$R1, $R2, $RI4", []>; 170 def CLGR : InstRIEb<0xEC65, (outs), (ins GR64:$R1, GR64:$R2, 171 brtarget16:$RI4), 172 "clgrj"##name##"\t$R1, $R2, $RI4", []>; 173 def CLI : InstRIEc<0xEC7F, (outs), (ins GR32:$R1, imm32zx8:$I2, 174 brtarget16:$RI4), 175 "clij"##name##"\t$R1, $I2, $RI4", []>; 176 def CLGI : InstRIEc<0xEC7D, (outs), (ins GR64:$R1, imm64zx8:$I2, 177 brtarget16:$RI4), 178 "clgij"##name##"\t$R1, $I2, $RI4", []>; 179 } 180} 181multiclass IntCondExtendedMnemonic<bits<4> ccmask, string name1, string name2> 182 : IntCondExtendedMnemonicA<ccmask, name1> { 183 let isAsmParserOnly = 1 in 184 defm Alt : IntCondExtendedMnemonicA<ccmask, name2>; 185} 186defm AsmJH : IntCondExtendedMnemonic<2, "h", "nle">; 187defm AsmJL : IntCondExtendedMnemonic<4, "l", "nhe">; 188defm AsmJLH : IntCondExtendedMnemonic<6, "lh", "ne">; 189defm AsmJE : IntCondExtendedMnemonic<8, "e", "nlh">; 190defm AsmJHE : IntCondExtendedMnemonic<10, "he", "nl">; 191defm AsmJLE : IntCondExtendedMnemonic<12, "le", "nh">; 192 193// Decrement a register and branch if it is nonzero. These don't clobber CC, 194// but we might need to split long branches into sequences that do. 195let Defs = [CC] in { 196 def BRCT : BranchUnaryRI<"brct", 0xA76, GR32>; 197 def BRCTG : BranchUnaryRI<"brctg", 0xA77, GR64>; 198} 199 200//===----------------------------------------------------------------------===// 201// Select instructions 202//===----------------------------------------------------------------------===// 203 204def Select32Mux : SelectWrapper<GRX32>, Requires<[FeatureHighWord]>; 205def Select32 : SelectWrapper<GR32>; 206def Select64 : SelectWrapper<GR64>; 207 208// We don't define 32-bit Mux stores because the low-only STOC should 209// always be used if possible. 210defm CondStore8Mux : CondStores<GRX32, nonvolatile_truncstorei8, 211 nonvolatile_anyextloadi8, bdxaddr20only>, 212 Requires<[FeatureHighWord]>; 213defm CondStore16Mux : CondStores<GRX32, nonvolatile_truncstorei16, 214 nonvolatile_anyextloadi16, bdxaddr20only>, 215 Requires<[FeatureHighWord]>; 216defm CondStore8 : CondStores<GR32, nonvolatile_truncstorei8, 217 nonvolatile_anyextloadi8, bdxaddr20only>; 218defm CondStore16 : CondStores<GR32, nonvolatile_truncstorei16, 219 nonvolatile_anyextloadi16, bdxaddr20only>; 220defm CondStore32 : CondStores<GR32, nonvolatile_store, 221 nonvolatile_load, bdxaddr20only>; 222 223defm : CondStores64<CondStore8, CondStore8Inv, nonvolatile_truncstorei8, 224 nonvolatile_anyextloadi8, bdxaddr20only>; 225defm : CondStores64<CondStore16, CondStore16Inv, nonvolatile_truncstorei16, 226 nonvolatile_anyextloadi16, bdxaddr20only>; 227defm : CondStores64<CondStore32, CondStore32Inv, nonvolatile_truncstorei32, 228 nonvolatile_anyextloadi32, bdxaddr20only>; 229defm CondStore64 : CondStores<GR64, nonvolatile_store, 230 nonvolatile_load, bdxaddr20only>; 231 232//===----------------------------------------------------------------------===// 233// Call instructions 234//===----------------------------------------------------------------------===// 235 236let isCall = 1, Defs = [R14D, CC] in { 237 def CallBRASL : Alias<6, (outs), (ins pcrel32:$I2, variable_ops), 238 [(z_call pcrel32:$I2)]>; 239 def CallBASR : Alias<2, (outs), (ins ADDR64:$R2, variable_ops), 240 [(z_call ADDR64:$R2)]>; 241} 242 243// Sibling calls. Indirect sibling calls must be via R1, since R2 upwards 244// are argument registers and since branching to R0 is a no-op. 245let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in { 246 def CallJG : Alias<6, (outs), (ins pcrel32:$I2), 247 [(z_sibcall pcrel32:$I2)]>; 248 let Uses = [R1D] in 249 def CallBR : Alias<2, (outs), (ins), [(z_sibcall R1D)]>; 250} 251 252// TLS calls. These will be lowered into a call to __tls_get_offset, 253// with an extra relocation specifying the TLS symbol. 254let isCall = 1, Defs = [R14D, CC] in { 255 def TLS_GDCALL : Alias<6, (outs), (ins tlssym:$I2, variable_ops), 256 [(z_tls_gdcall tglobaltlsaddr:$I2)]>; 257 def TLS_LDCALL : Alias<6, (outs), (ins tlssym:$I2, variable_ops), 258 [(z_tls_ldcall tglobaltlsaddr:$I2)]>; 259} 260 261// Define the general form of the call instructions for the asm parser. 262// These instructions don't hard-code %r14 as the return address register. 263// Allow an optional TLS marker symbol to generate TLS call relocations. 264let isCall = 1, Defs = [CC] in { 265 def BRAS : InstRI<0xA75, (outs), (ins GR64:$R1, brtarget16tls:$I2), 266 "bras\t$R1, $I2", []>; 267 def BRASL : InstRIL<0xC05, (outs), (ins GR64:$R1, brtarget32tls:$I2), 268 "brasl\t$R1, $I2", []>; 269 def BASR : InstRR<0x0D, (outs), (ins GR64:$R1, ADDR64:$R2), 270 "basr\t$R1, $R2", []>; 271} 272 273//===----------------------------------------------------------------------===// 274// Move instructions 275//===----------------------------------------------------------------------===// 276 277// Register moves. 278let hasSideEffects = 0 in { 279 // Expands to LR, RISBHG or RISBLG, depending on the choice of registers. 280 def LRMux : UnaryRRPseudo<"l", null_frag, GRX32, GRX32>, 281 Requires<[FeatureHighWord]>; 282 def LR : UnaryRR <"l", 0x18, null_frag, GR32, GR32>; 283 def LGR : UnaryRRE<"lg", 0xB904, null_frag, GR64, GR64>; 284} 285let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in { 286 def LTR : UnaryRR <"lt", 0x12, null_frag, GR32, GR32>; 287 def LTGR : UnaryRRE<"ltg", 0xB902, null_frag, GR64, GR64>; 288} 289 290// Move on condition. 291let isCodeGenOnly = 1, Uses = [CC] in { 292 def LOCR : CondUnaryRRF<"loc", 0xB9F2, GR32, GR32>; 293 def LOCGR : CondUnaryRRF<"locg", 0xB9E2, GR64, GR64>; 294} 295let Uses = [CC] in { 296 def AsmLOCR : AsmCondUnaryRRF<"loc", 0xB9F2, GR32, GR32>; 297 def AsmLOCGR : AsmCondUnaryRRF<"locg", 0xB9E2, GR64, GR64>; 298} 299 300// Immediate moves. 301let hasSideEffects = 0, isAsCheapAsAMove = 1, isMoveImm = 1, 302 isReMaterializable = 1 in { 303 // 16-bit sign-extended immediates. LHIMux expands to LHI or IIHF, 304 // deopending on the choice of register. 305 def LHIMux : UnaryRIPseudo<bitconvert, GRX32, imm32sx16>, 306 Requires<[FeatureHighWord]>; 307 def LHI : UnaryRI<"lhi", 0xA78, bitconvert, GR32, imm32sx16>; 308 def LGHI : UnaryRI<"lghi", 0xA79, bitconvert, GR64, imm64sx16>; 309 310 // Other 16-bit immediates. 311 def LLILL : UnaryRI<"llill", 0xA5F, bitconvert, GR64, imm64ll16>; 312 def LLILH : UnaryRI<"llilh", 0xA5E, bitconvert, GR64, imm64lh16>; 313 def LLIHL : UnaryRI<"llihl", 0xA5D, bitconvert, GR64, imm64hl16>; 314 def LLIHH : UnaryRI<"llihh", 0xA5C, bitconvert, GR64, imm64hh16>; 315 316 // 32-bit immediates. 317 def LGFI : UnaryRIL<"lgfi", 0xC01, bitconvert, GR64, imm64sx32>; 318 def LLILF : UnaryRIL<"llilf", 0xC0F, bitconvert, GR64, imm64lf32>; 319 def LLIHF : UnaryRIL<"llihf", 0xC0E, bitconvert, GR64, imm64hf32>; 320} 321 322// Register loads. 323let canFoldAsLoad = 1, SimpleBDXLoad = 1 in { 324 // Expands to L, LY or LFH, depending on the choice of register. 325 def LMux : UnaryRXYPseudo<"l", load, GRX32, 4>, 326 Requires<[FeatureHighWord]>; 327 defm L : UnaryRXPair<"l", 0x58, 0xE358, load, GR32, 4>; 328 def LFH : UnaryRXY<"lfh", 0xE3CA, load, GRH32, 4>, 329 Requires<[FeatureHighWord]>; 330 def LG : UnaryRXY<"lg", 0xE304, load, GR64, 8>; 331 332 // These instructions are split after register allocation, so we don't 333 // want a custom inserter. 334 let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in { 335 def L128 : Pseudo<(outs GR128:$dst), (ins bdxaddr20only128:$src), 336 [(set GR128:$dst, (load bdxaddr20only128:$src))]>; 337 } 338} 339let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in { 340 def LT : UnaryRXY<"lt", 0xE312, load, GR32, 4>; 341 def LTG : UnaryRXY<"ltg", 0xE302, load, GR64, 8>; 342} 343 344let canFoldAsLoad = 1 in { 345 def LRL : UnaryRILPC<"lrl", 0xC4D, aligned_load, GR32>; 346 def LGRL : UnaryRILPC<"lgrl", 0xC48, aligned_load, GR64>; 347} 348 349// Load on condition. 350let isCodeGenOnly = 1, Uses = [CC] in { 351 def LOC : CondUnaryRSY<"loc", 0xEBF2, nonvolatile_load, GR32, 4>; 352 def LOCG : CondUnaryRSY<"locg", 0xEBE2, nonvolatile_load, GR64, 8>; 353} 354let Uses = [CC] in { 355 def AsmLOC : AsmCondUnaryRSY<"loc", 0xEBF2, GR32, 4>; 356 def AsmLOCG : AsmCondUnaryRSY<"locg", 0xEBE2, GR64, 8>; 357} 358 359// Register stores. 360let SimpleBDXStore = 1 in { 361 // Expands to ST, STY or STFH, depending on the choice of register. 362 def STMux : StoreRXYPseudo<store, GRX32, 4>, 363 Requires<[FeatureHighWord]>; 364 defm ST : StoreRXPair<"st", 0x50, 0xE350, store, GR32, 4>; 365 def STFH : StoreRXY<"stfh", 0xE3CB, store, GRH32, 4>, 366 Requires<[FeatureHighWord]>; 367 def STG : StoreRXY<"stg", 0xE324, store, GR64, 8>; 368 369 // These instructions are split after register allocation, so we don't 370 // want a custom inserter. 371 let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in { 372 def ST128 : Pseudo<(outs), (ins GR128:$src, bdxaddr20only128:$dst), 373 [(store GR128:$src, bdxaddr20only128:$dst)]>; 374 } 375} 376def STRL : StoreRILPC<"strl", 0xC4F, aligned_store, GR32>; 377def STGRL : StoreRILPC<"stgrl", 0xC4B, aligned_store, GR64>; 378 379// Store on condition. 380let isCodeGenOnly = 1, Uses = [CC] in { 381 def STOC : CondStoreRSY<"stoc", 0xEBF3, GR32, 4>; 382 def STOCG : CondStoreRSY<"stocg", 0xEBE3, GR64, 8>; 383} 384let Uses = [CC] in { 385 def AsmSTOC : AsmCondStoreRSY<"stoc", 0xEBF3, GR32, 4>; 386 def AsmSTOCG : AsmCondStoreRSY<"stocg", 0xEBE3, GR64, 8>; 387} 388 389// 8-bit immediate stores to 8-bit fields. 390defm MVI : StoreSIPair<"mvi", 0x92, 0xEB52, truncstorei8, imm32zx8trunc>; 391 392// 16-bit immediate stores to 16-, 32- or 64-bit fields. 393def MVHHI : StoreSIL<"mvhhi", 0xE544, truncstorei16, imm32sx16trunc>; 394def MVHI : StoreSIL<"mvhi", 0xE54C, store, imm32sx16>; 395def MVGHI : StoreSIL<"mvghi", 0xE548, store, imm64sx16>; 396 397// Memory-to-memory moves. 398let mayLoad = 1, mayStore = 1 in 399 defm MVC : MemorySS<"mvc", 0xD2, z_mvc, z_mvc_loop>; 400 401// String moves. 402let mayLoad = 1, mayStore = 1, Defs = [CC] in 403 defm MVST : StringRRE<"mvst", 0xB255, z_stpcpy>; 404 405//===----------------------------------------------------------------------===// 406// Sign extensions 407//===----------------------------------------------------------------------===// 408// 409// Note that putting these before zero extensions mean that we will prefer 410// them for anyextload*. There's not really much to choose between the two 411// either way, but signed-extending loads have a short LH and a long LHY, 412// while zero-extending loads have only the long LLH. 413// 414//===----------------------------------------------------------------------===// 415 416// 32-bit extensions from registers. 417let hasSideEffects = 0 in { 418 def LBR : UnaryRRE<"lb", 0xB926, sext8, GR32, GR32>; 419 def LHR : UnaryRRE<"lh", 0xB927, sext16, GR32, GR32>; 420} 421 422// 64-bit extensions from registers. 423let hasSideEffects = 0 in { 424 def LGBR : UnaryRRE<"lgb", 0xB906, sext8, GR64, GR64>; 425 def LGHR : UnaryRRE<"lgh", 0xB907, sext16, GR64, GR64>; 426 def LGFR : UnaryRRE<"lgf", 0xB914, sext32, GR64, GR32>; 427} 428let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in 429 def LTGFR : UnaryRRE<"ltgf", 0xB912, null_frag, GR64, GR32>; 430 431// Match 32-to-64-bit sign extensions in which the source is already 432// in a 64-bit register. 433def : Pat<(sext_inreg GR64:$src, i32), 434 (LGFR (EXTRACT_SUBREG GR64:$src, subreg_l32))>; 435 436// 32-bit extensions from 8-bit memory. LBMux expands to LB or LBH, 437// depending on the choice of register. 438def LBMux : UnaryRXYPseudo<"lb", asextloadi8, GRX32, 1>, 439 Requires<[FeatureHighWord]>; 440def LB : UnaryRXY<"lb", 0xE376, asextloadi8, GR32, 1>; 441def LBH : UnaryRXY<"lbh", 0xE3C0, asextloadi8, GRH32, 1>, 442 Requires<[FeatureHighWord]>; 443 444// 32-bit extensions from 16-bit memory. LHMux expands to LH or LHH, 445// depending on the choice of register. 446def LHMux : UnaryRXYPseudo<"lh", asextloadi16, GRX32, 2>, 447 Requires<[FeatureHighWord]>; 448defm LH : UnaryRXPair<"lh", 0x48, 0xE378, asextloadi16, GR32, 2>; 449def LHH : UnaryRXY<"lhh", 0xE3C4, asextloadi16, GRH32, 2>, 450 Requires<[FeatureHighWord]>; 451def LHRL : UnaryRILPC<"lhrl", 0xC45, aligned_asextloadi16, GR32>; 452 453// 64-bit extensions from memory. 454def LGB : UnaryRXY<"lgb", 0xE377, asextloadi8, GR64, 1>; 455def LGH : UnaryRXY<"lgh", 0xE315, asextloadi16, GR64, 2>; 456def LGF : UnaryRXY<"lgf", 0xE314, asextloadi32, GR64, 4>; 457def LGHRL : UnaryRILPC<"lghrl", 0xC44, aligned_asextloadi16, GR64>; 458def LGFRL : UnaryRILPC<"lgfrl", 0xC4C, aligned_asextloadi32, GR64>; 459let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in 460 def LTGF : UnaryRXY<"ltgf", 0xE332, asextloadi32, GR64, 4>; 461 462//===----------------------------------------------------------------------===// 463// Zero extensions 464//===----------------------------------------------------------------------===// 465 466// 32-bit extensions from registers. 467let hasSideEffects = 0 in { 468 // Expands to LLCR or RISB[LH]G, depending on the choice of registers. 469 def LLCRMux : UnaryRRPseudo<"llc", zext8, GRX32, GRX32>, 470 Requires<[FeatureHighWord]>; 471 def LLCR : UnaryRRE<"llc", 0xB994, zext8, GR32, GR32>; 472 // Expands to LLHR or RISB[LH]G, depending on the choice of registers. 473 def LLHRMux : UnaryRRPseudo<"llh", zext16, GRX32, GRX32>, 474 Requires<[FeatureHighWord]>; 475 def LLHR : UnaryRRE<"llh", 0xB995, zext16, GR32, GR32>; 476} 477 478// 64-bit extensions from registers. 479let hasSideEffects = 0 in { 480 def LLGCR : UnaryRRE<"llgc", 0xB984, zext8, GR64, GR64>; 481 def LLGHR : UnaryRRE<"llgh", 0xB985, zext16, GR64, GR64>; 482 def LLGFR : UnaryRRE<"llgf", 0xB916, zext32, GR64, GR32>; 483} 484 485// Match 32-to-64-bit zero extensions in which the source is already 486// in a 64-bit register. 487def : Pat<(and GR64:$src, 0xffffffff), 488 (LLGFR (EXTRACT_SUBREG GR64:$src, subreg_l32))>; 489 490// 32-bit extensions from 8-bit memory. LLCMux expands to LLC or LLCH, 491// depending on the choice of register. 492def LLCMux : UnaryRXYPseudo<"llc", azextloadi8, GRX32, 1>, 493 Requires<[FeatureHighWord]>; 494def LLC : UnaryRXY<"llc", 0xE394, azextloadi8, GR32, 1>; 495def LLCH : UnaryRXY<"llch", 0xE3C2, azextloadi8, GRH32, 1>, 496 Requires<[FeatureHighWord]>; 497 498// 32-bit extensions from 16-bit memory. LLHMux expands to LLH or LLHH, 499// depending on the choice of register. 500def LLHMux : UnaryRXYPseudo<"llh", azextloadi16, GRX32, 2>, 501 Requires<[FeatureHighWord]>; 502def LLH : UnaryRXY<"llh", 0xE395, azextloadi16, GR32, 2>; 503def LLHH : UnaryRXY<"llhh", 0xE3C6, azextloadi16, GRH32, 2>, 504 Requires<[FeatureHighWord]>; 505def LLHRL : UnaryRILPC<"llhrl", 0xC42, aligned_azextloadi16, GR32>; 506 507// 64-bit extensions from memory. 508def LLGC : UnaryRXY<"llgc", 0xE390, azextloadi8, GR64, 1>; 509def LLGH : UnaryRXY<"llgh", 0xE391, azextloadi16, GR64, 2>; 510def LLGF : UnaryRXY<"llgf", 0xE316, azextloadi32, GR64, 4>; 511def LLGHRL : UnaryRILPC<"llghrl", 0xC46, aligned_azextloadi16, GR64>; 512def LLGFRL : UnaryRILPC<"llgfrl", 0xC4E, aligned_azextloadi32, GR64>; 513 514//===----------------------------------------------------------------------===// 515// Truncations 516//===----------------------------------------------------------------------===// 517 518// Truncations of 64-bit registers to 32-bit registers. 519def : Pat<(i32 (trunc GR64:$src)), 520 (EXTRACT_SUBREG GR64:$src, subreg_l32)>; 521 522// Truncations of 32-bit registers to 8-bit memory. STCMux expands to 523// STC, STCY or STCH, depending on the choice of register. 524def STCMux : StoreRXYPseudo<truncstorei8, GRX32, 1>, 525 Requires<[FeatureHighWord]>; 526defm STC : StoreRXPair<"stc", 0x42, 0xE372, truncstorei8, GR32, 1>; 527def STCH : StoreRXY<"stch", 0xE3C3, truncstorei8, GRH32, 1>, 528 Requires<[FeatureHighWord]>; 529 530// Truncations of 32-bit registers to 16-bit memory. STHMux expands to 531// STH, STHY or STHH, depending on the choice of register. 532def STHMux : StoreRXYPseudo<truncstorei16, GRX32, 1>, 533 Requires<[FeatureHighWord]>; 534defm STH : StoreRXPair<"sth", 0x40, 0xE370, truncstorei16, GR32, 2>; 535def STHH : StoreRXY<"sthh", 0xE3C7, truncstorei16, GRH32, 2>, 536 Requires<[FeatureHighWord]>; 537def STHRL : StoreRILPC<"sthrl", 0xC47, aligned_truncstorei16, GR32>; 538 539// Truncations of 64-bit registers to memory. 540defm : StoreGR64Pair<STC, STCY, truncstorei8>; 541defm : StoreGR64Pair<STH, STHY, truncstorei16>; 542def : StoreGR64PC<STHRL, aligned_truncstorei16>; 543defm : StoreGR64Pair<ST, STY, truncstorei32>; 544def : StoreGR64PC<STRL, aligned_truncstorei32>; 545 546//===----------------------------------------------------------------------===// 547// Multi-register moves 548//===----------------------------------------------------------------------===// 549 550// Multi-register loads. 551def LMG : LoadMultipleRSY<"lmg", 0xEB04, GR64>; 552 553// Multi-register stores. 554def STMG : StoreMultipleRSY<"stmg", 0xEB24, GR64>; 555 556//===----------------------------------------------------------------------===// 557// Byte swaps 558//===----------------------------------------------------------------------===// 559 560// Byte-swapping register moves. 561let hasSideEffects = 0 in { 562 def LRVR : UnaryRRE<"lrv", 0xB91F, bswap, GR32, GR32>; 563 def LRVGR : UnaryRRE<"lrvg", 0xB90F, bswap, GR64, GR64>; 564} 565 566// Byte-swapping loads. Unlike normal loads, these instructions are 567// allowed to access storage more than once. 568def LRV : UnaryRXY<"lrv", 0xE31E, loadu<bswap, nonvolatile_load>, GR32, 4>; 569def LRVG : UnaryRXY<"lrvg", 0xE30F, loadu<bswap, nonvolatile_load>, GR64, 8>; 570 571// Likewise byte-swapping stores. 572def STRV : StoreRXY<"strv", 0xE33E, storeu<bswap, nonvolatile_store>, GR32, 4>; 573def STRVG : StoreRXY<"strvg", 0xE32F, storeu<bswap, nonvolatile_store>, 574 GR64, 8>; 575 576//===----------------------------------------------------------------------===// 577// Load address instructions 578//===----------------------------------------------------------------------===// 579 580// Load BDX-style addresses. 581let hasSideEffects = 0, isAsCheapAsAMove = 1, isReMaterializable = 1, 582 DispKey = "la" in { 583 let DispSize = "12" in 584 def LA : InstRX<0x41, (outs GR64:$R1), (ins laaddr12pair:$XBD2), 585 "la\t$R1, $XBD2", 586 [(set GR64:$R1, laaddr12pair:$XBD2)]>; 587 let DispSize = "20" in 588 def LAY : InstRXY<0xE371, (outs GR64:$R1), (ins laaddr20pair:$XBD2), 589 "lay\t$R1, $XBD2", 590 [(set GR64:$R1, laaddr20pair:$XBD2)]>; 591} 592 593// Load a PC-relative address. There's no version of this instruction 594// with a 16-bit offset, so there's no relaxation. 595let hasSideEffects = 0, isAsCheapAsAMove = 1, isMoveImm = 1, 596 isReMaterializable = 1 in { 597 def LARL : InstRIL<0xC00, (outs GR64:$R1), (ins pcrel32:$I2), 598 "larl\t$R1, $I2", 599 [(set GR64:$R1, pcrel32:$I2)]>; 600} 601 602// Load the Global Offset Table address. This will be lowered into a 603// larl $R1, _GLOBAL_OFFSET_TABLE_ 604// instruction. 605def GOT : Alias<6, (outs GR64:$R1), (ins), 606 [(set GR64:$R1, (global_offset_table))]>; 607 608//===----------------------------------------------------------------------===// 609// Absolute and Negation 610//===----------------------------------------------------------------------===// 611 612let Defs = [CC] in { 613 let CCValues = 0xF, CompareZeroCCMask = 0x8 in { 614 def LPR : UnaryRR <"lp", 0x10, z_iabs, GR32, GR32>; 615 def LPGR : UnaryRRE<"lpg", 0xB900, z_iabs, GR64, GR64>; 616 } 617 let CCValues = 0xE, CompareZeroCCMask = 0xE in 618 def LPGFR : UnaryRRE<"lpgf", 0xB910, null_frag, GR64, GR32>; 619} 620def : Pat<(z_iabs32 GR32:$src), (LPR GR32:$src)>; 621def : Pat<(z_iabs64 GR64:$src), (LPGR GR64:$src)>; 622defm : SXU<z_iabs, LPGFR>; 623defm : SXU<z_iabs64, LPGFR>; 624 625let Defs = [CC] in { 626 let CCValues = 0xF, CompareZeroCCMask = 0x8 in { 627 def LNR : UnaryRR <"ln", 0x11, z_inegabs, GR32, GR32>; 628 def LNGR : UnaryRRE<"lng", 0xB901, z_inegabs, GR64, GR64>; 629 } 630 let CCValues = 0xE, CompareZeroCCMask = 0xE in 631 def LNGFR : UnaryRRE<"lngf", 0xB911, null_frag, GR64, GR32>; 632} 633def : Pat<(z_inegabs32 GR32:$src), (LNR GR32:$src)>; 634def : Pat<(z_inegabs64 GR64:$src), (LNGR GR64:$src)>; 635defm : SXU<z_inegabs, LNGFR>; 636defm : SXU<z_inegabs64, LNGFR>; 637 638let Defs = [CC] in { 639 let CCValues = 0xF, CompareZeroCCMask = 0x8 in { 640 def LCR : UnaryRR <"lc", 0x13, ineg, GR32, GR32>; 641 def LCGR : UnaryRRE<"lcg", 0xB903, ineg, GR64, GR64>; 642 } 643 let CCValues = 0xE, CompareZeroCCMask = 0xE in 644 def LCGFR : UnaryRRE<"lcgf", 0xB913, null_frag, GR64, GR32>; 645} 646defm : SXU<ineg, LCGFR>; 647 648//===----------------------------------------------------------------------===// 649// Insertion 650//===----------------------------------------------------------------------===// 651 652let isCodeGenOnly = 1 in 653 defm IC32 : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR32, azextloadi8, 1>; 654defm IC : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR64, azextloadi8, 1>; 655 656defm : InsertMem<"inserti8", IC32, GR32, azextloadi8, bdxaddr12pair>; 657defm : InsertMem<"inserti8", IC32Y, GR32, azextloadi8, bdxaddr20pair>; 658 659defm : InsertMem<"inserti8", IC, GR64, azextloadi8, bdxaddr12pair>; 660defm : InsertMem<"inserti8", ICY, GR64, azextloadi8, bdxaddr20pair>; 661 662// Insertions of a 16-bit immediate, leaving other bits unaffected. 663// We don't have or_as_insert equivalents of these operations because 664// OI is available instead. 665// 666// IIxMux expands to II[LH]x, depending on the choice of register. 667def IILMux : BinaryRIPseudo<insertll, GRX32, imm32ll16>, 668 Requires<[FeatureHighWord]>; 669def IIHMux : BinaryRIPseudo<insertlh, GRX32, imm32lh16>, 670 Requires<[FeatureHighWord]>; 671def IILL : BinaryRI<"iill", 0xA53, insertll, GR32, imm32ll16>; 672def IILH : BinaryRI<"iilh", 0xA52, insertlh, GR32, imm32lh16>; 673def IIHL : BinaryRI<"iihl", 0xA51, insertll, GRH32, imm32ll16>; 674def IIHH : BinaryRI<"iihh", 0xA50, insertlh, GRH32, imm32lh16>; 675def IILL64 : BinaryAliasRI<insertll, GR64, imm64ll16>; 676def IILH64 : BinaryAliasRI<insertlh, GR64, imm64lh16>; 677def IIHL64 : BinaryAliasRI<inserthl, GR64, imm64hl16>; 678def IIHH64 : BinaryAliasRI<inserthh, GR64, imm64hh16>; 679 680// ...likewise for 32-bit immediates. For GR32s this is a general 681// full-width move. (We use IILF rather than something like LLILF 682// for 32-bit moves because IILF leaves the upper 32 bits of the 683// GR64 unchanged.) 684let isAsCheapAsAMove = 1, isMoveImm = 1, isReMaterializable = 1 in { 685 def IIFMux : UnaryRIPseudo<bitconvert, GRX32, uimm32>, 686 Requires<[FeatureHighWord]>; 687 def IILF : UnaryRIL<"iilf", 0xC09, bitconvert, GR32, uimm32>; 688 def IIHF : UnaryRIL<"iihf", 0xC08, bitconvert, GRH32, uimm32>; 689} 690def IILF64 : BinaryAliasRIL<insertlf, GR64, imm64lf32>; 691def IIHF64 : BinaryAliasRIL<inserthf, GR64, imm64hf32>; 692 693// An alternative model of inserthf, with the first operand being 694// a zero-extended value. 695def : Pat<(or (zext32 GR32:$src), imm64hf32:$imm), 696 (IIHF64 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_l32), 697 imm64hf32:$imm)>; 698 699//===----------------------------------------------------------------------===// 700// Addition 701//===----------------------------------------------------------------------===// 702 703// Plain addition. 704let Defs = [CC], CCValues = 0xF, CompareZeroCCMask = 0x8 in { 705 // Addition of a register. 706 let isCommutable = 1 in { 707 defm AR : BinaryRRAndK<"a", 0x1A, 0xB9F8, add, GR32, GR32>; 708 defm AGR : BinaryRREAndK<"ag", 0xB908, 0xB9E8, add, GR64, GR64>; 709 } 710 def AGFR : BinaryRRE<"agf", 0xB918, null_frag, GR64, GR32>; 711 712 // Addition of signed 16-bit immediates. 713 defm AHIMux : BinaryRIAndKPseudo<"ahimux", add, GRX32, imm32sx16>; 714 defm AHI : BinaryRIAndK<"ahi", 0xA7A, 0xECD8, add, GR32, imm32sx16>; 715 defm AGHI : BinaryRIAndK<"aghi", 0xA7B, 0xECD9, add, GR64, imm64sx16>; 716 717 // Addition of signed 32-bit immediates. 718 def AFIMux : BinaryRIPseudo<add, GRX32, simm32>, 719 Requires<[FeatureHighWord]>; 720 def AFI : BinaryRIL<"afi", 0xC29, add, GR32, simm32>; 721 def AIH : BinaryRIL<"aih", 0xCC8, add, GRH32, simm32>, 722 Requires<[FeatureHighWord]>; 723 def AGFI : BinaryRIL<"agfi", 0xC28, add, GR64, imm64sx32>; 724 725 // Addition of memory. 726 defm AH : BinaryRXPair<"ah", 0x4A, 0xE37A, add, GR32, asextloadi16, 2>; 727 defm A : BinaryRXPair<"a", 0x5A, 0xE35A, add, GR32, load, 4>; 728 def AGF : BinaryRXY<"agf", 0xE318, add, GR64, asextloadi32, 4>; 729 def AG : BinaryRXY<"ag", 0xE308, add, GR64, load, 8>; 730 731 // Addition to memory. 732 def ASI : BinarySIY<"asi", 0xEB6A, add, imm32sx8>; 733 def AGSI : BinarySIY<"agsi", 0xEB7A, add, imm64sx8>; 734} 735defm : SXB<add, GR64, AGFR>; 736 737// Addition producing a carry. 738let Defs = [CC] in { 739 // Addition of a register. 740 let isCommutable = 1 in { 741 defm ALR : BinaryRRAndK<"al", 0x1E, 0xB9FA, addc, GR32, GR32>; 742 defm ALGR : BinaryRREAndK<"alg", 0xB90A, 0xB9EA, addc, GR64, GR64>; 743 } 744 def ALGFR : BinaryRRE<"algf", 0xB91A, null_frag, GR64, GR32>; 745 746 // Addition of signed 16-bit immediates. 747 def ALHSIK : BinaryRIE<"alhsik", 0xECDA, addc, GR32, imm32sx16>, 748 Requires<[FeatureDistinctOps]>; 749 def ALGHSIK : BinaryRIE<"alghsik", 0xECDB, addc, GR64, imm64sx16>, 750 Requires<[FeatureDistinctOps]>; 751 752 // Addition of unsigned 32-bit immediates. 753 def ALFI : BinaryRIL<"alfi", 0xC2B, addc, GR32, uimm32>; 754 def ALGFI : BinaryRIL<"algfi", 0xC2A, addc, GR64, imm64zx32>; 755 756 // Addition of memory. 757 defm AL : BinaryRXPair<"al", 0x5E, 0xE35E, addc, GR32, load, 4>; 758 def ALGF : BinaryRXY<"algf", 0xE31A, addc, GR64, azextloadi32, 4>; 759 def ALG : BinaryRXY<"alg", 0xE30A, addc, GR64, load, 8>; 760} 761defm : ZXB<addc, GR64, ALGFR>; 762 763// Addition producing and using a carry. 764let Defs = [CC], Uses = [CC] in { 765 // Addition of a register. 766 def ALCR : BinaryRRE<"alc", 0xB998, adde, GR32, GR32>; 767 def ALCGR : BinaryRRE<"alcg", 0xB988, adde, GR64, GR64>; 768 769 // Addition of memory. 770 def ALC : BinaryRXY<"alc", 0xE398, adde, GR32, load, 4>; 771 def ALCG : BinaryRXY<"alcg", 0xE388, adde, GR64, load, 8>; 772} 773 774//===----------------------------------------------------------------------===// 775// Subtraction 776//===----------------------------------------------------------------------===// 777 778// Plain subtraction. Although immediate forms exist, we use the 779// add-immediate instruction instead. 780let Defs = [CC], CCValues = 0xF, CompareZeroCCMask = 0x8 in { 781 // Subtraction of a register. 782 defm SR : BinaryRRAndK<"s", 0x1B, 0xB9F9, sub, GR32, GR32>; 783 def SGFR : BinaryRRE<"sgf", 0xB919, null_frag, GR64, GR32>; 784 defm SGR : BinaryRREAndK<"sg", 0xB909, 0xB9E9, sub, GR64, GR64>; 785 786 // Subtraction of memory. 787 defm SH : BinaryRXPair<"sh", 0x4B, 0xE37B, sub, GR32, asextloadi16, 2>; 788 defm S : BinaryRXPair<"s", 0x5B, 0xE35B, sub, GR32, load, 4>; 789 def SGF : BinaryRXY<"sgf", 0xE319, sub, GR64, asextloadi32, 4>; 790 def SG : BinaryRXY<"sg", 0xE309, sub, GR64, load, 8>; 791} 792defm : SXB<sub, GR64, SGFR>; 793 794// Subtraction producing a carry. 795let Defs = [CC] in { 796 // Subtraction of a register. 797 defm SLR : BinaryRRAndK<"sl", 0x1F, 0xB9FB, subc, GR32, GR32>; 798 def SLGFR : BinaryRRE<"slgf", 0xB91B, null_frag, GR64, GR32>; 799 defm SLGR : BinaryRREAndK<"slg", 0xB90B, 0xB9EB, subc, GR64, GR64>; 800 801 // Subtraction of unsigned 32-bit immediates. These don't match 802 // subc because we prefer addc for constants. 803 def SLFI : BinaryRIL<"slfi", 0xC25, null_frag, GR32, uimm32>; 804 def SLGFI : BinaryRIL<"slgfi", 0xC24, null_frag, GR64, imm64zx32>; 805 806 // Subtraction of memory. 807 defm SL : BinaryRXPair<"sl", 0x5F, 0xE35F, subc, GR32, load, 4>; 808 def SLGF : BinaryRXY<"slgf", 0xE31B, subc, GR64, azextloadi32, 4>; 809 def SLG : BinaryRXY<"slg", 0xE30B, subc, GR64, load, 8>; 810} 811defm : ZXB<subc, GR64, SLGFR>; 812 813// Subtraction producing and using a carry. 814let Defs = [CC], Uses = [CC] in { 815 // Subtraction of a register. 816 def SLBR : BinaryRRE<"slb", 0xB999, sube, GR32, GR32>; 817 def SLBGR : BinaryRRE<"slbg", 0xB989, sube, GR64, GR64>; 818 819 // Subtraction of memory. 820 def SLB : BinaryRXY<"slb", 0xE399, sube, GR32, load, 4>; 821 def SLBG : BinaryRXY<"slbg", 0xE389, sube, GR64, load, 8>; 822} 823 824//===----------------------------------------------------------------------===// 825// AND 826//===----------------------------------------------------------------------===// 827 828let Defs = [CC] in { 829 // ANDs of a register. 830 let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in { 831 defm NR : BinaryRRAndK<"n", 0x14, 0xB9F4, and, GR32, GR32>; 832 defm NGR : BinaryRREAndK<"ng", 0xB980, 0xB9E4, and, GR64, GR64>; 833 } 834 835 let isConvertibleToThreeAddress = 1 in { 836 // ANDs of a 16-bit immediate, leaving other bits unaffected. 837 // The CC result only reflects the 16-bit field, not the full register. 838 // 839 // NIxMux expands to NI[LH]x, depending on the choice of register. 840 def NILMux : BinaryRIPseudo<and, GRX32, imm32ll16c>, 841 Requires<[FeatureHighWord]>; 842 def NIHMux : BinaryRIPseudo<and, GRX32, imm32lh16c>, 843 Requires<[FeatureHighWord]>; 844 def NILL : BinaryRI<"nill", 0xA57, and, GR32, imm32ll16c>; 845 def NILH : BinaryRI<"nilh", 0xA56, and, GR32, imm32lh16c>; 846 def NIHL : BinaryRI<"nihl", 0xA55, and, GRH32, imm32ll16c>; 847 def NIHH : BinaryRI<"nihh", 0xA54, and, GRH32, imm32lh16c>; 848 def NILL64 : BinaryAliasRI<and, GR64, imm64ll16c>; 849 def NILH64 : BinaryAliasRI<and, GR64, imm64lh16c>; 850 def NIHL64 : BinaryAliasRI<and, GR64, imm64hl16c>; 851 def NIHH64 : BinaryAliasRI<and, GR64, imm64hh16c>; 852 853 // ANDs of a 32-bit immediate, leaving other bits unaffected. 854 // The CC result only reflects the 32-bit field, which means we can 855 // use it as a zero indicator for i32 operations but not otherwise. 856 let CCValues = 0xC, CompareZeroCCMask = 0x8 in { 857 // Expands to NILF or NIHF, depending on the choice of register. 858 def NIFMux : BinaryRIPseudo<and, GRX32, uimm32>, 859 Requires<[FeatureHighWord]>; 860 def NILF : BinaryRIL<"nilf", 0xC0B, and, GR32, uimm32>; 861 def NIHF : BinaryRIL<"nihf", 0xC0A, and, GRH32, uimm32>; 862 } 863 def NILF64 : BinaryAliasRIL<and, GR64, imm64lf32c>; 864 def NIHF64 : BinaryAliasRIL<and, GR64, imm64hf32c>; 865 } 866 867 // ANDs of memory. 868 let CCValues = 0xC, CompareZeroCCMask = 0x8 in { 869 defm N : BinaryRXPair<"n", 0x54, 0xE354, and, GR32, load, 4>; 870 def NG : BinaryRXY<"ng", 0xE380, and, GR64, load, 8>; 871 } 872 873 // AND to memory 874 defm NI : BinarySIPair<"ni", 0x94, 0xEB54, null_frag, imm32zx8>; 875 876 // Block AND. 877 let mayLoad = 1, mayStore = 1 in 878 defm NC : MemorySS<"nc", 0xD4, z_nc, z_nc_loop>; 879} 880defm : RMWIByte<and, bdaddr12pair, NI>; 881defm : RMWIByte<and, bdaddr20pair, NIY>; 882 883//===----------------------------------------------------------------------===// 884// OR 885//===----------------------------------------------------------------------===// 886 887let Defs = [CC] in { 888 // ORs of a register. 889 let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in { 890 defm OR : BinaryRRAndK<"o", 0x16, 0xB9F6, or, GR32, GR32>; 891 defm OGR : BinaryRREAndK<"og", 0xB981, 0xB9E6, or, GR64, GR64>; 892 } 893 894 // ORs of a 16-bit immediate, leaving other bits unaffected. 895 // The CC result only reflects the 16-bit field, not the full register. 896 // 897 // OIxMux expands to OI[LH]x, depending on the choice of register. 898 def OILMux : BinaryRIPseudo<or, GRX32, imm32ll16>, 899 Requires<[FeatureHighWord]>; 900 def OIHMux : BinaryRIPseudo<or, GRX32, imm32lh16>, 901 Requires<[FeatureHighWord]>; 902 def OILL : BinaryRI<"oill", 0xA5B, or, GR32, imm32ll16>; 903 def OILH : BinaryRI<"oilh", 0xA5A, or, GR32, imm32lh16>; 904 def OIHL : BinaryRI<"oihl", 0xA59, or, GRH32, imm32ll16>; 905 def OIHH : BinaryRI<"oihh", 0xA58, or, GRH32, imm32lh16>; 906 def OILL64 : BinaryAliasRI<or, GR64, imm64ll16>; 907 def OILH64 : BinaryAliasRI<or, GR64, imm64lh16>; 908 def OIHL64 : BinaryAliasRI<or, GR64, imm64hl16>; 909 def OIHH64 : BinaryAliasRI<or, GR64, imm64hh16>; 910 911 // ORs of a 32-bit immediate, leaving other bits unaffected. 912 // The CC result only reflects the 32-bit field, which means we can 913 // use it as a zero indicator for i32 operations but not otherwise. 914 let CCValues = 0xC, CompareZeroCCMask = 0x8 in { 915 // Expands to OILF or OIHF, depending on the choice of register. 916 def OIFMux : BinaryRIPseudo<or, GRX32, uimm32>, 917 Requires<[FeatureHighWord]>; 918 def OILF : BinaryRIL<"oilf", 0xC0D, or, GR32, uimm32>; 919 def OIHF : BinaryRIL<"oihf", 0xC0C, or, GRH32, uimm32>; 920 } 921 def OILF64 : BinaryAliasRIL<or, GR64, imm64lf32>; 922 def OIHF64 : BinaryAliasRIL<or, GR64, imm64hf32>; 923 924 // ORs of memory. 925 let CCValues = 0xC, CompareZeroCCMask = 0x8 in { 926 defm O : BinaryRXPair<"o", 0x56, 0xE356, or, GR32, load, 4>; 927 def OG : BinaryRXY<"og", 0xE381, or, GR64, load, 8>; 928 } 929 930 // OR to memory 931 defm OI : BinarySIPair<"oi", 0x96, 0xEB56, null_frag, imm32zx8>; 932 933 // Block OR. 934 let mayLoad = 1, mayStore = 1 in 935 defm OC : MemorySS<"oc", 0xD6, z_oc, z_oc_loop>; 936} 937defm : RMWIByte<or, bdaddr12pair, OI>; 938defm : RMWIByte<or, bdaddr20pair, OIY>; 939 940//===----------------------------------------------------------------------===// 941// XOR 942//===----------------------------------------------------------------------===// 943 944let Defs = [CC] in { 945 // XORs of a register. 946 let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in { 947 defm XR : BinaryRRAndK<"x", 0x17, 0xB9F7, xor, GR32, GR32>; 948 defm XGR : BinaryRREAndK<"xg", 0xB982, 0xB9E7, xor, GR64, GR64>; 949 } 950 951 // XORs of a 32-bit immediate, leaving other bits unaffected. 952 // The CC result only reflects the 32-bit field, which means we can 953 // use it as a zero indicator for i32 operations but not otherwise. 954 let CCValues = 0xC, CompareZeroCCMask = 0x8 in { 955 // Expands to XILF or XIHF, depending on the choice of register. 956 def XIFMux : BinaryRIPseudo<xor, GRX32, uimm32>, 957 Requires<[FeatureHighWord]>; 958 def XILF : BinaryRIL<"xilf", 0xC07, xor, GR32, uimm32>; 959 def XIHF : BinaryRIL<"xihf", 0xC06, xor, GRH32, uimm32>; 960 } 961 def XILF64 : BinaryAliasRIL<xor, GR64, imm64lf32>; 962 def XIHF64 : BinaryAliasRIL<xor, GR64, imm64hf32>; 963 964 // XORs of memory. 965 let CCValues = 0xC, CompareZeroCCMask = 0x8 in { 966 defm X : BinaryRXPair<"x",0x57, 0xE357, xor, GR32, load, 4>; 967 def XG : BinaryRXY<"xg", 0xE382, xor, GR64, load, 8>; 968 } 969 970 // XOR to memory 971 defm XI : BinarySIPair<"xi", 0x97, 0xEB57, null_frag, imm32zx8>; 972 973 // Block XOR. 974 let mayLoad = 1, mayStore = 1 in 975 defm XC : MemorySS<"xc", 0xD7, z_xc, z_xc_loop>; 976} 977defm : RMWIByte<xor, bdaddr12pair, XI>; 978defm : RMWIByte<xor, bdaddr20pair, XIY>; 979 980//===----------------------------------------------------------------------===// 981// Multiplication 982//===----------------------------------------------------------------------===// 983 984// Multiplication of a register. 985let isCommutable = 1 in { 986 def MSR : BinaryRRE<"ms", 0xB252, mul, GR32, GR32>; 987 def MSGR : BinaryRRE<"msg", 0xB90C, mul, GR64, GR64>; 988} 989def MSGFR : BinaryRRE<"msgf", 0xB91C, null_frag, GR64, GR32>; 990defm : SXB<mul, GR64, MSGFR>; 991 992// Multiplication of a signed 16-bit immediate. 993def MHI : BinaryRI<"mhi", 0xA7C, mul, GR32, imm32sx16>; 994def MGHI : BinaryRI<"mghi", 0xA7D, mul, GR64, imm64sx16>; 995 996// Multiplication of a signed 32-bit immediate. 997def MSFI : BinaryRIL<"msfi", 0xC21, mul, GR32, simm32>; 998def MSGFI : BinaryRIL<"msgfi", 0xC20, mul, GR64, imm64sx32>; 999 1000// Multiplication of memory. 1001defm MH : BinaryRXPair<"mh", 0x4C, 0xE37C, mul, GR32, asextloadi16, 2>; 1002defm MS : BinaryRXPair<"ms", 0x71, 0xE351, mul, GR32, load, 4>; 1003def MSGF : BinaryRXY<"msgf", 0xE31C, mul, GR64, asextloadi32, 4>; 1004def MSG : BinaryRXY<"msg", 0xE30C, mul, GR64, load, 8>; 1005 1006// Multiplication of a register, producing two results. 1007def MLGR : BinaryRRE<"mlg", 0xB986, z_umul_lohi64, GR128, GR64>; 1008 1009// Multiplication of memory, producing two results. 1010def MLG : BinaryRXY<"mlg", 0xE386, z_umul_lohi64, GR128, load, 8>; 1011 1012//===----------------------------------------------------------------------===// 1013// Division and remainder 1014//===----------------------------------------------------------------------===// 1015 1016// Division and remainder, from registers. 1017def DSGFR : BinaryRRE<"dsgf", 0xB91D, z_sdivrem32, GR128, GR32>; 1018def DSGR : BinaryRRE<"dsg", 0xB90D, z_sdivrem64, GR128, GR64>; 1019def DLR : BinaryRRE<"dl", 0xB997, z_udivrem32, GR128, GR32>; 1020def DLGR : BinaryRRE<"dlg", 0xB987, z_udivrem64, GR128, GR64>; 1021 1022// Division and remainder, from memory. 1023def DSGF : BinaryRXY<"dsgf", 0xE31D, z_sdivrem32, GR128, load, 4>; 1024def DSG : BinaryRXY<"dsg", 0xE30D, z_sdivrem64, GR128, load, 8>; 1025def DL : BinaryRXY<"dl", 0xE397, z_udivrem32, GR128, load, 4>; 1026def DLG : BinaryRXY<"dlg", 0xE387, z_udivrem64, GR128, load, 8>; 1027 1028//===----------------------------------------------------------------------===// 1029// Shifts 1030//===----------------------------------------------------------------------===// 1031 1032// Shift left. 1033let hasSideEffects = 0 in { 1034 defm SLL : BinaryRSAndK<"sll", 0x89, 0xEBDF, shl, GR32>; 1035 def SLLG : BinaryRSY<"sllg", 0xEB0D, shl, GR64>; 1036} 1037 1038// Logical shift right. 1039let hasSideEffects = 0 in { 1040 defm SRL : BinaryRSAndK<"srl", 0x88, 0xEBDE, srl, GR32>; 1041 def SRLG : BinaryRSY<"srlg", 0xEB0C, srl, GR64>; 1042} 1043 1044// Arithmetic shift right. 1045let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in { 1046 defm SRA : BinaryRSAndK<"sra", 0x8A, 0xEBDC, sra, GR32>; 1047 def SRAG : BinaryRSY<"srag", 0xEB0A, sra, GR64>; 1048} 1049 1050// Rotate left. 1051let hasSideEffects = 0 in { 1052 def RLL : BinaryRSY<"rll", 0xEB1D, rotl, GR32>; 1053 def RLLG : BinaryRSY<"rllg", 0xEB1C, rotl, GR64>; 1054} 1055 1056// Rotate second operand left and inserted selected bits into first operand. 1057// These can act like 32-bit operands provided that the constant start and 1058// end bits (operands 2 and 3) are in the range [32, 64). 1059let Defs = [CC] in { 1060 let isCodeGenOnly = 1 in 1061 def RISBG32 : RotateSelectRIEf<"risbg", 0xEC55, GR32, GR32>; 1062 let CCValues = 0xE, CompareZeroCCMask = 0xE in 1063 def RISBG : RotateSelectRIEf<"risbg", 0xEC55, GR64, GR64>; 1064} 1065 1066// On zEC12 we have a variant of RISBG that does not set CC. 1067let Predicates = [FeatureMiscellaneousExtensions] in 1068 def RISBGN : RotateSelectRIEf<"risbgn", 0xEC59, GR64, GR64>; 1069 1070// Forms of RISBG that only affect one word of the destination register. 1071// They do not set CC. 1072let Predicates = [FeatureHighWord] in { 1073 def RISBMux : RotateSelectRIEfPseudo<GRX32, GRX32>; 1074 def RISBLL : RotateSelectAliasRIEf<GR32, GR32>; 1075 def RISBLH : RotateSelectAliasRIEf<GR32, GRH32>; 1076 def RISBHL : RotateSelectAliasRIEf<GRH32, GR32>; 1077 def RISBHH : RotateSelectAliasRIEf<GRH32, GRH32>; 1078 def RISBLG : RotateSelectRIEf<"risblg", 0xEC51, GR32, GR64>; 1079 def RISBHG : RotateSelectRIEf<"risbhg", 0xEC5D, GRH32, GR64>; 1080} 1081 1082// Rotate second operand left and perform a logical operation with selected 1083// bits of the first operand. The CC result only describes the selected bits, 1084// so isn't useful for a full comparison against zero. 1085let Defs = [CC] in { 1086 def RNSBG : RotateSelectRIEf<"rnsbg", 0xEC54, GR64, GR64>; 1087 def ROSBG : RotateSelectRIEf<"rosbg", 0xEC56, GR64, GR64>; 1088 def RXSBG : RotateSelectRIEf<"rxsbg", 0xEC57, GR64, GR64>; 1089} 1090 1091//===----------------------------------------------------------------------===// 1092// Comparison 1093//===----------------------------------------------------------------------===// 1094 1095// Signed comparisons. We put these before the unsigned comparisons because 1096// some of the signed forms have COMPARE AND BRANCH equivalents whereas none 1097// of the unsigned forms do. 1098let Defs = [CC], CCValues = 0xE in { 1099 // Comparison with a register. 1100 def CR : CompareRR <"c", 0x19, z_scmp, GR32, GR32>; 1101 def CGFR : CompareRRE<"cgf", 0xB930, null_frag, GR64, GR32>; 1102 def CGR : CompareRRE<"cg", 0xB920, z_scmp, GR64, GR64>; 1103 1104 // Comparison with a signed 16-bit immediate. 1105 def CHI : CompareRI<"chi", 0xA7E, z_scmp, GR32, imm32sx16>; 1106 def CGHI : CompareRI<"cghi", 0xA7F, z_scmp, GR64, imm64sx16>; 1107 1108 // Comparison with a signed 32-bit immediate. CFIMux expands to CFI or CIH, 1109 // depending on the choice of register. 1110 def CFIMux : CompareRIPseudo<z_scmp, GRX32, simm32>, 1111 Requires<[FeatureHighWord]>; 1112 def CFI : CompareRIL<"cfi", 0xC2D, z_scmp, GR32, simm32>; 1113 def CIH : CompareRIL<"cih", 0xCCD, z_scmp, GRH32, simm32>, 1114 Requires<[FeatureHighWord]>; 1115 def CGFI : CompareRIL<"cgfi", 0xC2C, z_scmp, GR64, imm64sx32>; 1116 1117 // Comparison with memory. 1118 defm CH : CompareRXPair<"ch", 0x49, 0xE379, z_scmp, GR32, asextloadi16, 2>; 1119 def CMux : CompareRXYPseudo<z_scmp, GRX32, load, 4>, 1120 Requires<[FeatureHighWord]>; 1121 defm C : CompareRXPair<"c", 0x59, 0xE359, z_scmp, GR32, load, 4>; 1122 def CHF : CompareRXY<"chf", 0xE3CD, z_scmp, GRH32, load, 4>, 1123 Requires<[FeatureHighWord]>; 1124 def CGH : CompareRXY<"cgh", 0xE334, z_scmp, GR64, asextloadi16, 2>; 1125 def CGF : CompareRXY<"cgf", 0xE330, z_scmp, GR64, asextloadi32, 4>; 1126 def CG : CompareRXY<"cg", 0xE320, z_scmp, GR64, load, 8>; 1127 def CHRL : CompareRILPC<"chrl", 0xC65, z_scmp, GR32, aligned_asextloadi16>; 1128 def CRL : CompareRILPC<"crl", 0xC6D, z_scmp, GR32, aligned_load>; 1129 def CGHRL : CompareRILPC<"cghrl", 0xC64, z_scmp, GR64, aligned_asextloadi16>; 1130 def CGFRL : CompareRILPC<"cgfrl", 0xC6C, z_scmp, GR64, aligned_asextloadi32>; 1131 def CGRL : CompareRILPC<"cgrl", 0xC68, z_scmp, GR64, aligned_load>; 1132 1133 // Comparison between memory and a signed 16-bit immediate. 1134 def CHHSI : CompareSIL<"chhsi", 0xE554, z_scmp, asextloadi16, imm32sx16>; 1135 def CHSI : CompareSIL<"chsi", 0xE55C, z_scmp, load, imm32sx16>; 1136 def CGHSI : CompareSIL<"cghsi", 0xE558, z_scmp, load, imm64sx16>; 1137} 1138defm : SXB<z_scmp, GR64, CGFR>; 1139 1140// Unsigned comparisons. 1141let Defs = [CC], CCValues = 0xE, IsLogical = 1 in { 1142 // Comparison with a register. 1143 def CLR : CompareRR <"cl", 0x15, z_ucmp, GR32, GR32>; 1144 def CLGFR : CompareRRE<"clgf", 0xB931, null_frag, GR64, GR32>; 1145 def CLGR : CompareRRE<"clg", 0xB921, z_ucmp, GR64, GR64>; 1146 1147 // Comparison with an unsigned 32-bit immediate. CLFIMux expands to CLFI 1148 // or CLIH, depending on the choice of register. 1149 def CLFIMux : CompareRIPseudo<z_ucmp, GRX32, uimm32>, 1150 Requires<[FeatureHighWord]>; 1151 def CLFI : CompareRIL<"clfi", 0xC2F, z_ucmp, GR32, uimm32>; 1152 def CLIH : CompareRIL<"clih", 0xCCF, z_ucmp, GRH32, uimm32>, 1153 Requires<[FeatureHighWord]>; 1154 def CLGFI : CompareRIL<"clgfi", 0xC2E, z_ucmp, GR64, imm64zx32>; 1155 1156 // Comparison with memory. 1157 def CLMux : CompareRXYPseudo<z_ucmp, GRX32, load, 4>, 1158 Requires<[FeatureHighWord]>; 1159 defm CL : CompareRXPair<"cl", 0x55, 0xE355, z_ucmp, GR32, load, 4>; 1160 def CLHF : CompareRXY<"clhf", 0xE3CF, z_ucmp, GRH32, load, 4>, 1161 Requires<[FeatureHighWord]>; 1162 def CLGF : CompareRXY<"clgf", 0xE331, z_ucmp, GR64, azextloadi32, 4>; 1163 def CLG : CompareRXY<"clg", 0xE321, z_ucmp, GR64, load, 8>; 1164 def CLHRL : CompareRILPC<"clhrl", 0xC67, z_ucmp, GR32, 1165 aligned_azextloadi16>; 1166 def CLRL : CompareRILPC<"clrl", 0xC6F, z_ucmp, GR32, 1167 aligned_load>; 1168 def CLGHRL : CompareRILPC<"clghrl", 0xC66, z_ucmp, GR64, 1169 aligned_azextloadi16>; 1170 def CLGFRL : CompareRILPC<"clgfrl", 0xC6E, z_ucmp, GR64, 1171 aligned_azextloadi32>; 1172 def CLGRL : CompareRILPC<"clgrl", 0xC6A, z_ucmp, GR64, 1173 aligned_load>; 1174 1175 // Comparison between memory and an unsigned 8-bit immediate. 1176 defm CLI : CompareSIPair<"cli", 0x95, 0xEB55, z_ucmp, azextloadi8, imm32zx8>; 1177 1178 // Comparison between memory and an unsigned 16-bit immediate. 1179 def CLHHSI : CompareSIL<"clhhsi", 0xE555, z_ucmp, azextloadi16, imm32zx16>; 1180 def CLFHSI : CompareSIL<"clfhsi", 0xE55D, z_ucmp, load, imm32zx16>; 1181 def CLGHSI : CompareSIL<"clghsi", 0xE559, z_ucmp, load, imm64zx16>; 1182} 1183defm : ZXB<z_ucmp, GR64, CLGFR>; 1184 1185// Memory-to-memory comparison. 1186let mayLoad = 1, Defs = [CC] in 1187 defm CLC : MemorySS<"clc", 0xD5, z_clc, z_clc_loop>; 1188 1189// String comparison. 1190let mayLoad = 1, Defs = [CC] in 1191 defm CLST : StringRRE<"clst", 0xB25D, z_strcmp>; 1192 1193// Test under mask. 1194let Defs = [CC] in { 1195 // TMxMux expands to TM[LH]x, depending on the choice of register. 1196 def TMLMux : CompareRIPseudo<z_tm_reg, GRX32, imm32ll16>, 1197 Requires<[FeatureHighWord]>; 1198 def TMHMux : CompareRIPseudo<z_tm_reg, GRX32, imm32lh16>, 1199 Requires<[FeatureHighWord]>; 1200 def TMLL : CompareRI<"tmll", 0xA71, z_tm_reg, GR32, imm32ll16>; 1201 def TMLH : CompareRI<"tmlh", 0xA70, z_tm_reg, GR32, imm32lh16>; 1202 def TMHL : CompareRI<"tmhl", 0xA73, z_tm_reg, GRH32, imm32ll16>; 1203 def TMHH : CompareRI<"tmhh", 0xA72, z_tm_reg, GRH32, imm32lh16>; 1204 1205 def TMLL64 : CompareAliasRI<z_tm_reg, GR64, imm64ll16>; 1206 def TMLH64 : CompareAliasRI<z_tm_reg, GR64, imm64lh16>; 1207 def TMHL64 : CompareAliasRI<z_tm_reg, GR64, imm64hl16>; 1208 def TMHH64 : CompareAliasRI<z_tm_reg, GR64, imm64hh16>; 1209 1210 defm TM : CompareSIPair<"tm", 0x91, 0xEB51, z_tm_mem, anyextloadi8, imm32zx8>; 1211} 1212 1213//===----------------------------------------------------------------------===// 1214// Prefetch 1215//===----------------------------------------------------------------------===// 1216 1217def PFD : PrefetchRXY<"pfd", 0xE336, z_prefetch>; 1218def PFDRL : PrefetchRILPC<"pfdrl", 0xC62, z_prefetch>; 1219 1220//===----------------------------------------------------------------------===// 1221// Atomic operations 1222//===----------------------------------------------------------------------===// 1223 1224// A serialization instruction that acts as a barrier for all memory 1225// accesses, which expands to "bcr 14, 0". 1226let hasSideEffects = 1 in 1227def Serialize : Alias<2, (outs), (ins), [(z_serialize)]>; 1228 1229let Predicates = [FeatureInterlockedAccess1], Defs = [CC] in { 1230 def LAA : LoadAndOpRSY<"laa", 0xEBF8, atomic_load_add_32, GR32>; 1231 def LAAG : LoadAndOpRSY<"laag", 0xEBE8, atomic_load_add_64, GR64>; 1232 def LAAL : LoadAndOpRSY<"laal", 0xEBFA, null_frag, GR32>; 1233 def LAALG : LoadAndOpRSY<"laalg", 0xEBEA, null_frag, GR64>; 1234 def LAN : LoadAndOpRSY<"lan", 0xEBF4, atomic_load_and_32, GR32>; 1235 def LANG : LoadAndOpRSY<"lang", 0xEBE4, atomic_load_and_64, GR64>; 1236 def LAO : LoadAndOpRSY<"lao", 0xEBF6, atomic_load_or_32, GR32>; 1237 def LAOG : LoadAndOpRSY<"laog", 0xEBE6, atomic_load_or_64, GR64>; 1238 def LAX : LoadAndOpRSY<"lax", 0xEBF7, atomic_load_xor_32, GR32>; 1239 def LAXG : LoadAndOpRSY<"laxg", 0xEBE7, atomic_load_xor_64, GR64>; 1240} 1241 1242def ATOMIC_SWAPW : AtomicLoadWBinaryReg<z_atomic_swapw>; 1243def ATOMIC_SWAP_32 : AtomicLoadBinaryReg32<atomic_swap_32>; 1244def ATOMIC_SWAP_64 : AtomicLoadBinaryReg64<atomic_swap_64>; 1245 1246def ATOMIC_LOADW_AR : AtomicLoadWBinaryReg<z_atomic_loadw_add>; 1247def ATOMIC_LOADW_AFI : AtomicLoadWBinaryImm<z_atomic_loadw_add, simm32>; 1248let Predicates = [FeatureNoInterlockedAccess1] in { 1249 def ATOMIC_LOAD_AR : AtomicLoadBinaryReg32<atomic_load_add_32>; 1250 def ATOMIC_LOAD_AHI : AtomicLoadBinaryImm32<atomic_load_add_32, imm32sx16>; 1251 def ATOMIC_LOAD_AFI : AtomicLoadBinaryImm32<atomic_load_add_32, simm32>; 1252 def ATOMIC_LOAD_AGR : AtomicLoadBinaryReg64<atomic_load_add_64>; 1253 def ATOMIC_LOAD_AGHI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx16>; 1254 def ATOMIC_LOAD_AGFI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx32>; 1255} 1256 1257def ATOMIC_LOADW_SR : AtomicLoadWBinaryReg<z_atomic_loadw_sub>; 1258def ATOMIC_LOAD_SR : AtomicLoadBinaryReg32<atomic_load_sub_32>; 1259def ATOMIC_LOAD_SGR : AtomicLoadBinaryReg64<atomic_load_sub_64>; 1260 1261def ATOMIC_LOADW_NR : AtomicLoadWBinaryReg<z_atomic_loadw_and>; 1262def ATOMIC_LOADW_NILH : AtomicLoadWBinaryImm<z_atomic_loadw_and, imm32lh16c>; 1263let Predicates = [FeatureNoInterlockedAccess1] in { 1264 def ATOMIC_LOAD_NR : AtomicLoadBinaryReg32<atomic_load_and_32>; 1265 def ATOMIC_LOAD_NILL : AtomicLoadBinaryImm32<atomic_load_and_32, 1266 imm32ll16c>; 1267 def ATOMIC_LOAD_NILH : AtomicLoadBinaryImm32<atomic_load_and_32, 1268 imm32lh16c>; 1269 def ATOMIC_LOAD_NILF : AtomicLoadBinaryImm32<atomic_load_and_32, uimm32>; 1270 def ATOMIC_LOAD_NGR : AtomicLoadBinaryReg64<atomic_load_and_64>; 1271 def ATOMIC_LOAD_NILL64 : AtomicLoadBinaryImm64<atomic_load_and_64, 1272 imm64ll16c>; 1273 def ATOMIC_LOAD_NILH64 : AtomicLoadBinaryImm64<atomic_load_and_64, 1274 imm64lh16c>; 1275 def ATOMIC_LOAD_NIHL64 : AtomicLoadBinaryImm64<atomic_load_and_64, 1276 imm64hl16c>; 1277 def ATOMIC_LOAD_NIHH64 : AtomicLoadBinaryImm64<atomic_load_and_64, 1278 imm64hh16c>; 1279 def ATOMIC_LOAD_NILF64 : AtomicLoadBinaryImm64<atomic_load_and_64, 1280 imm64lf32c>; 1281 def ATOMIC_LOAD_NIHF64 : AtomicLoadBinaryImm64<atomic_load_and_64, 1282 imm64hf32c>; 1283} 1284 1285def ATOMIC_LOADW_OR : AtomicLoadWBinaryReg<z_atomic_loadw_or>; 1286def ATOMIC_LOADW_OILH : AtomicLoadWBinaryImm<z_atomic_loadw_or, imm32lh16>; 1287let Predicates = [FeatureNoInterlockedAccess1] in { 1288 def ATOMIC_LOAD_OR : AtomicLoadBinaryReg32<atomic_load_or_32>; 1289 def ATOMIC_LOAD_OILL : AtomicLoadBinaryImm32<atomic_load_or_32, imm32ll16>; 1290 def ATOMIC_LOAD_OILH : AtomicLoadBinaryImm32<atomic_load_or_32, imm32lh16>; 1291 def ATOMIC_LOAD_OILF : AtomicLoadBinaryImm32<atomic_load_or_32, uimm32>; 1292 def ATOMIC_LOAD_OGR : AtomicLoadBinaryReg64<atomic_load_or_64>; 1293 def ATOMIC_LOAD_OILL64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64ll16>; 1294 def ATOMIC_LOAD_OILH64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lh16>; 1295 def ATOMIC_LOAD_OIHL64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hl16>; 1296 def ATOMIC_LOAD_OIHH64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hh16>; 1297 def ATOMIC_LOAD_OILF64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lf32>; 1298 def ATOMIC_LOAD_OIHF64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hf32>; 1299} 1300 1301def ATOMIC_LOADW_XR : AtomicLoadWBinaryReg<z_atomic_loadw_xor>; 1302def ATOMIC_LOADW_XILF : AtomicLoadWBinaryImm<z_atomic_loadw_xor, uimm32>; 1303let Predicates = [FeatureNoInterlockedAccess1] in { 1304 def ATOMIC_LOAD_XR : AtomicLoadBinaryReg32<atomic_load_xor_32>; 1305 def ATOMIC_LOAD_XILF : AtomicLoadBinaryImm32<atomic_load_xor_32, uimm32>; 1306 def ATOMIC_LOAD_XGR : AtomicLoadBinaryReg64<atomic_load_xor_64>; 1307 def ATOMIC_LOAD_XILF64 : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64lf32>; 1308 def ATOMIC_LOAD_XIHF64 : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64hf32>; 1309} 1310 1311def ATOMIC_LOADW_NRi : AtomicLoadWBinaryReg<z_atomic_loadw_nand>; 1312def ATOMIC_LOADW_NILHi : AtomicLoadWBinaryImm<z_atomic_loadw_nand, 1313 imm32lh16c>; 1314def ATOMIC_LOAD_NRi : AtomicLoadBinaryReg32<atomic_load_nand_32>; 1315def ATOMIC_LOAD_NILLi : AtomicLoadBinaryImm32<atomic_load_nand_32, 1316 imm32ll16c>; 1317def ATOMIC_LOAD_NILHi : AtomicLoadBinaryImm32<atomic_load_nand_32, 1318 imm32lh16c>; 1319def ATOMIC_LOAD_NILFi : AtomicLoadBinaryImm32<atomic_load_nand_32, uimm32>; 1320def ATOMIC_LOAD_NGRi : AtomicLoadBinaryReg64<atomic_load_nand_64>; 1321def ATOMIC_LOAD_NILL64i : AtomicLoadBinaryImm64<atomic_load_nand_64, 1322 imm64ll16c>; 1323def ATOMIC_LOAD_NILH64i : AtomicLoadBinaryImm64<atomic_load_nand_64, 1324 imm64lh16c>; 1325def ATOMIC_LOAD_NIHL64i : AtomicLoadBinaryImm64<atomic_load_nand_64, 1326 imm64hl16c>; 1327def ATOMIC_LOAD_NIHH64i : AtomicLoadBinaryImm64<atomic_load_nand_64, 1328 imm64hh16c>; 1329def ATOMIC_LOAD_NILF64i : AtomicLoadBinaryImm64<atomic_load_nand_64, 1330 imm64lf32c>; 1331def ATOMIC_LOAD_NIHF64i : AtomicLoadBinaryImm64<atomic_load_nand_64, 1332 imm64hf32c>; 1333 1334def ATOMIC_LOADW_MIN : AtomicLoadWBinaryReg<z_atomic_loadw_min>; 1335def ATOMIC_LOAD_MIN_32 : AtomicLoadBinaryReg32<atomic_load_min_32>; 1336def ATOMIC_LOAD_MIN_64 : AtomicLoadBinaryReg64<atomic_load_min_64>; 1337 1338def ATOMIC_LOADW_MAX : AtomicLoadWBinaryReg<z_atomic_loadw_max>; 1339def ATOMIC_LOAD_MAX_32 : AtomicLoadBinaryReg32<atomic_load_max_32>; 1340def ATOMIC_LOAD_MAX_64 : AtomicLoadBinaryReg64<atomic_load_max_64>; 1341 1342def ATOMIC_LOADW_UMIN : AtomicLoadWBinaryReg<z_atomic_loadw_umin>; 1343def ATOMIC_LOAD_UMIN_32 : AtomicLoadBinaryReg32<atomic_load_umin_32>; 1344def ATOMIC_LOAD_UMIN_64 : AtomicLoadBinaryReg64<atomic_load_umin_64>; 1345 1346def ATOMIC_LOADW_UMAX : AtomicLoadWBinaryReg<z_atomic_loadw_umax>; 1347def ATOMIC_LOAD_UMAX_32 : AtomicLoadBinaryReg32<atomic_load_umax_32>; 1348def ATOMIC_LOAD_UMAX_64 : AtomicLoadBinaryReg64<atomic_load_umax_64>; 1349 1350def ATOMIC_CMP_SWAPW 1351 : Pseudo<(outs GR32:$dst), (ins bdaddr20only:$addr, GR32:$cmp, GR32:$swap, 1352 ADDR32:$bitshift, ADDR32:$negbitshift, 1353 uimm32:$bitsize), 1354 [(set GR32:$dst, 1355 (z_atomic_cmp_swapw bdaddr20only:$addr, GR32:$cmp, GR32:$swap, 1356 ADDR32:$bitshift, ADDR32:$negbitshift, 1357 uimm32:$bitsize))]> { 1358 let Defs = [CC]; 1359 let mayLoad = 1; 1360 let mayStore = 1; 1361 let usesCustomInserter = 1; 1362} 1363 1364let Defs = [CC] in { 1365 defm CS : CmpSwapRSPair<"cs", 0xBA, 0xEB14, atomic_cmp_swap_32, GR32>; 1366 def CSG : CmpSwapRSY<"csg", 0xEB30, atomic_cmp_swap_64, GR64>; 1367} 1368 1369//===----------------------------------------------------------------------===// 1370// Transactional execution 1371//===----------------------------------------------------------------------===// 1372 1373let Predicates = [FeatureTransactionalExecution] in { 1374 // Transaction Begin 1375 let hasSideEffects = 1, mayStore = 1, 1376 usesCustomInserter = 1, Defs = [CC] in { 1377 def TBEGIN : InstSIL<0xE560, 1378 (outs), (ins bdaddr12only:$BD1, imm32zx16:$I2), 1379 "tbegin\t$BD1, $I2", 1380 [(z_tbegin bdaddr12only:$BD1, imm32zx16:$I2)]>; 1381 def TBEGIN_nofloat : Pseudo<(outs), (ins bdaddr12only:$BD1, imm32zx16:$I2), 1382 [(z_tbegin_nofloat bdaddr12only:$BD1, 1383 imm32zx16:$I2)]>; 1384 def TBEGINC : InstSIL<0xE561, 1385 (outs), (ins bdaddr12only:$BD1, imm32zx16:$I2), 1386 "tbeginc\t$BD1, $I2", 1387 [(int_s390_tbeginc bdaddr12only:$BD1, 1388 imm32zx16:$I2)]>; 1389 } 1390 1391 // Transaction End 1392 let hasSideEffects = 1, Defs = [CC], BD2 = 0 in 1393 def TEND : InstS<0xB2F8, (outs), (ins), "tend", [(z_tend)]>; 1394 1395 // Transaction Abort 1396 let hasSideEffects = 1, isTerminator = 1, isBarrier = 1 in 1397 def TABORT : InstS<0xB2FC, (outs), (ins bdaddr12only:$BD2), 1398 "tabort\t$BD2", 1399 [(int_s390_tabort bdaddr12only:$BD2)]>; 1400 1401 // Nontransactional Store 1402 let hasSideEffects = 1 in 1403 def NTSTG : StoreRXY<"ntstg", 0xE325, int_s390_ntstg, GR64, 8>; 1404 1405 // Extract Transaction Nesting Depth 1406 let hasSideEffects = 1 in 1407 def ETND : InherentRRE<"etnd", 0xB2EC, GR32, (int_s390_etnd)>; 1408} 1409 1410//===----------------------------------------------------------------------===// 1411// Processor assist 1412//===----------------------------------------------------------------------===// 1413 1414let Predicates = [FeatureProcessorAssist] in { 1415 let hasSideEffects = 1, R4 = 0 in 1416 def PPA : InstRRF<0xB2E8, (outs), (ins GR64:$R1, GR64:$R2, imm32zx4:$R3), 1417 "ppa\t$R1, $R2, $R3", []>; 1418 def : Pat<(int_s390_ppa_txassist GR32:$src), 1419 (PPA (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_l32), 1420 0, 1)>; 1421} 1422 1423//===----------------------------------------------------------------------===// 1424// Miscellaneous Instructions. 1425//===----------------------------------------------------------------------===// 1426 1427// Extract CC into bits 29 and 28 of a register. 1428let Uses = [CC] in 1429 def IPM : InherentRRE<"ipm", 0xB222, GR32, (z_ipm)>; 1430 1431// Read a 32-bit access register into a GR32. As with all GR32 operations, 1432// the upper 32 bits of the enclosing GR64 remain unchanged, which is useful 1433// when a 64-bit address is stored in a pair of access registers. 1434def EAR : InstRRE<0xB24F, (outs GR32:$R1), (ins access_reg:$R2), 1435 "ear\t$R1, $R2", 1436 [(set GR32:$R1, (z_extract_access access_reg:$R2))]>; 1437 1438// Find leftmost one, AKA count leading zeros. The instruction actually 1439// returns a pair of GR64s, the first giving the number of leading zeros 1440// and the second giving a copy of the source with the leftmost one bit 1441// cleared. We only use the first result here. 1442let Defs = [CC] in { 1443 def FLOGR : UnaryRRE<"flog", 0xB983, null_frag, GR128, GR64>; 1444} 1445def : Pat<(ctlz GR64:$src), 1446 (EXTRACT_SUBREG (FLOGR GR64:$src), subreg_h64)>; 1447 1448// Population count. Counts bits set per byte. 1449let Predicates = [FeaturePopulationCount], Defs = [CC] in { 1450 def POPCNT : InstRRE<0xB9E1, (outs GR64:$R1), (ins GR64:$R2), 1451 "popcnt\t$R1, $R2", 1452 [(set GR64:$R1, (z_popcnt GR64:$R2))]>; 1453} 1454 1455// Use subregs to populate the "don't care" bits in a 32-bit to 64-bit anyext. 1456def : Pat<(i64 (anyext GR32:$src)), 1457 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_l32)>; 1458 1459// Extend GR32s and GR64s to GR128s. 1460let usesCustomInserter = 1 in { 1461 def AEXT128_64 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>; 1462 def ZEXT128_32 : Pseudo<(outs GR128:$dst), (ins GR32:$src), []>; 1463 def ZEXT128_64 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>; 1464} 1465 1466// Search a block of memory for a character. 1467let mayLoad = 1, Defs = [CC] in 1468 defm SRST : StringRRE<"srst", 0xb25e, z_search_string>; 1469 1470// Other instructions for inline assembly 1471let hasSideEffects = 1, Defs = [CC], mayStore = 1 in 1472 def STCK : InstS<0xB205, (outs), (ins bdaddr12only:$BD2), 1473 "stck\t$BD2", 1474 []>; 1475let hasSideEffects = 1, Defs = [CC], mayStore = 1 in 1476 def STCKF : InstS<0xB27C, (outs), (ins bdaddr12only:$BD2), 1477 "stckf\t$BD2", 1478 []>; 1479let hasSideEffects = 1, Defs = [CC], mayStore = 1 in 1480 def STCKE : InstS<0xB278, (outs), (ins bdaddr12only:$BD2), 1481 "stcke\t$BD2", 1482 []>; 1483let hasSideEffects = 1, Defs = [CC], mayStore = 1 in 1484 def STFLE : InstS<0xB2B0, (outs), (ins bdaddr12only:$BD2), 1485 "stfle\t$BD2", 1486 []>; 1487 1488 1489 1490//===----------------------------------------------------------------------===// 1491// Peepholes. 1492//===----------------------------------------------------------------------===// 1493 1494// Use AL* for GR64 additions of unsigned 32-bit values. 1495defm : ZXB<add, GR64, ALGFR>; 1496def : Pat<(add GR64:$src1, imm64zx32:$src2), 1497 (ALGFI GR64:$src1, imm64zx32:$src2)>; 1498def : Pat<(add GR64:$src1, (azextloadi32 bdxaddr20only:$addr)), 1499 (ALGF GR64:$src1, bdxaddr20only:$addr)>; 1500 1501// Use SL* for GR64 subtractions of unsigned 32-bit values. 1502defm : ZXB<sub, GR64, SLGFR>; 1503def : Pat<(add GR64:$src1, imm64zx32n:$src2), 1504 (SLGFI GR64:$src1, imm64zx32n:$src2)>; 1505def : Pat<(sub GR64:$src1, (azextloadi32 bdxaddr20only:$addr)), 1506 (SLGF GR64:$src1, bdxaddr20only:$addr)>; 1507 1508// Optimize sign-extended 1/0 selects to -1/0 selects. This is important 1509// for vector legalization. 1510def : Pat<(sra (shl (i32 (z_select_ccmask 1, 0, imm32zx4:$valid, imm32zx4:$cc)), 1511 (i32 31)), 1512 (i32 31)), 1513 (Select32 (LHI -1), (LHI 0), imm32zx4:$valid, imm32zx4:$cc)>; 1514def : Pat<(sra (shl (i64 (anyext (i32 (z_select_ccmask 1, 0, imm32zx4:$valid, 1515 imm32zx4:$cc)))), 1516 (i32 63)), 1517 (i32 63)), 1518 (Select64 (LGHI -1), (LGHI 0), imm32zx4:$valid, imm32zx4:$cc)>; 1519 1520// Peepholes for turning scalar operations into block operations. 1521defm : BlockLoadStore<anyextloadi8, i32, MVCSequence, NCSequence, OCSequence, 1522 XCSequence, 1>; 1523defm : BlockLoadStore<anyextloadi16, i32, MVCSequence, NCSequence, OCSequence, 1524 XCSequence, 2>; 1525defm : BlockLoadStore<load, i32, MVCSequence, NCSequence, OCSequence, 1526 XCSequence, 4>; 1527defm : BlockLoadStore<anyextloadi8, i64, MVCSequence, NCSequence, 1528 OCSequence, XCSequence, 1>; 1529defm : BlockLoadStore<anyextloadi16, i64, MVCSequence, NCSequence, OCSequence, 1530 XCSequence, 2>; 1531defm : BlockLoadStore<anyextloadi32, i64, MVCSequence, NCSequence, OCSequence, 1532 XCSequence, 4>; 1533defm : BlockLoadStore<load, i64, MVCSequence, NCSequence, OCSequence, 1534 XCSequence, 8>; 1535