1;; Represents the possible widths of an element when used in an operation. 2(type VecElementWidth (enum 3 (E8) 4 (E16) 5 (E32) 6 (E64) 7)) 8 9;; Vector Register Group Multiplier (LMUL) 10;; 11;; The LMUL setting specifies how we should group registers together. LMUL can 12;; also be a fractional value, reducing the number of bits used in a single 13;; vector register. Fractional LMUL is used to increase the number of effective 14;; usable vector register groups when operating on mixed-width values. 15(type VecLmul (enum 16 (LmulF8) 17 (LmulF4) 18 (LmulF2) 19 (Lmul1) 20 (Lmul2) 21 (Lmul4) 22 (Lmul8) 23)) 24 25;; Tail Mode 26;; 27;; The tail mode specifies how the tail elements of a vector register are handled. 28(type VecTailMode (enum 29 ;; Tail Agnostic means that the tail elements are left in an undefined state. 30 (Agnostic) 31 ;; Tail Undisturbed means that the tail elements are left in their original values. 32 (Undisturbed) 33)) 34 35;; Mask Mode 36;; 37;; The mask mode specifies how the masked elements of a vector register are handled. 38(type VecMaskMode (enum 39 ;; Mask Agnostic means that the masked out elements are left in an undefined state. 40 (Agnostic) 41 ;; Mask Undisturbed means that the masked out elements are left in their original values. 42 (Undisturbed) 43)) 44 45;; Application Vector Length (AVL) 46;; 47;; This setting specifies the number of elements that are going to be processed 48;; in a single instruction. Note: We may end up processing fewer elements than 49;; the AVL setting, if they don't fit in a single register. 50(type VecAvl (enum 51 ;; Static AVL emits a `vsetivli` that uses a constant value 52 (Static (size UImm5)) 53 ;; TODO: Add a dynamic, register based AVL mode when we are able to properly test it 54)) 55 56(type VType (primitive VType)) 57(type VState (primitive VState)) 58 59 60;; Vector Opcode Category 61;; 62;; These categories are used to determine the type of operands that are allowed in the 63;; instruction. 64(type VecOpCategory (enum 65 (OPIVV) 66 (OPFVV) 67 (OPMVV) 68 (OPIVI) 69 (OPIVX) 70 (OPFVF) 71 (OPMVX) 72 (OPCFG) 73)) 74 75;; Vector Opcode Masking 76;; 77;; When masked, the instruction will only operate on the elements that are dictated by 78;; the mask register. Currently this is always fixed to v0. 79(type VecOpMasking (enum 80 (Enabled (reg Reg)) 81 (Disabled) 82)) 83 84(decl pure masked (VReg) VecOpMasking) 85(rule (masked reg) (VecOpMasking.Enabled reg)) 86 87(decl pure unmasked () VecOpMasking) 88(rule (unmasked) (VecOpMasking.Disabled)) 89 90;; Register to Register ALU Ops 91(type VecAluOpRRR (enum 92 ;; Vector-Vector Opcodes 93 (VaddVV) 94 (VsaddVV) 95 (VsadduVV) 96 (VwaddVV) 97 (VwaddWV) 98 (VwadduVV) 99 (VwadduWV) 100 (VsubVV) 101 (VwsubVV) 102 (VwsubWV) 103 (VwsubuVV) 104 (VwsubuWV) 105 (VssubVV) 106 (VssubuVV) 107 (VmulVV) 108 (VmulhVV) 109 (VmulhuVV) 110 (VsmulVV) 111 (VsllVV) 112 (VsrlVV) 113 (VsraVV) 114 (VandVV) 115 (VorVV) 116 (VxorVV) 117 (VmaxVV) 118 (VmaxuVV) 119 (VminVV) 120 (VminuVV) 121 (VfaddVV) 122 (VfsubVV) 123 (VfmulVV) 124 (VfdivVV) 125 (VfminVV) 126 (VfmaxVV) 127 (VfsgnjVV) 128 (VfsgnjnVV) 129 (VfsgnjxVV) 130 (VmergeVVM) 131 (VredmaxuVS) 132 (VredminuVS) 133 (VrgatherVV) 134 (VcompressVM) 135 (VmseqVV) 136 (VmsneVV) 137 (VmsltuVV) 138 (VmsltVV) 139 (VmsleuVV) 140 (VmsleVV) 141 (VmfeqVV) 142 (VmfneVV) 143 (VmfltVV) 144 (VmfleVV) 145 (VmandMM) 146 (VmorMM) 147 (VmnandMM) 148 (VmnorMM) 149 150 151 ;; Vector-Scalar Opcodes 152 (VaddVX) 153 (VsaddVX) 154 (VsadduVX) 155 (VwaddVX) 156 (VwaddWX) 157 (VwadduVX) 158 (VwadduWX) 159 (VsubVX) 160 (VrsubVX) 161 (VwsubVX) 162 (VwsubWX) 163 (VwsubuVX) 164 (VwsubuWX) 165 (VssubVX) 166 (VssubuVX) 167 (VmulVX) 168 (VmulhVX) 169 (VmulhuVX) 170 (VsmulVX) 171 (VsllVX) 172 (VsrlVX) 173 (VsraVX) 174 (VandVX) 175 (VorVX) 176 (VxorVX) 177 (VmaxVX) 178 (VmaxuVX) 179 (VminVX) 180 (VminuVX) 181 (VslidedownVX) 182 (VfaddVF) 183 (VfsubVF) 184 (VfrsubVF) 185 (VfmulVF) 186 (VfdivVF) 187 (VfsgnjVF) 188 (VfrdivVF) 189 (VmergeVXM) 190 (VfmergeVFM) 191 (VrgatherVX) 192 (VmseqVX) 193 (VmsneVX) 194 (VmsltuVX) 195 (VmsltVX) 196 (VmsleuVX) 197 (VmsleVX) 198 (VmsgtuVX) 199 (VmsgtVX) 200 (VmfeqVF) 201 (VmfneVF) 202 (VmfltVF) 203 (VmfleVF) 204 (VmfgtVF) 205 (VmfgeVF) 206)) 207 208 209 210;; Register-Imm ALU Ops that modify the destination register 211(type VecAluOpRRRImm5 (enum 212 (VslideupVI) 213)) 214 215;; Register-Register ALU Ops that modify the destination register 216(type VecAluOpRRRR (enum 217 ;; Vector-Vector Opcodes 218 (VmaccVV) 219 (VnmsacVV) 220 (VfmaccVV) 221 (VfnmaccVV) 222 (VfmsacVV) 223 (VfnmsacVV) 224 225 ;; Vector-Scalar Opcodes 226 (VmaccVX) 227 (VnmsacVX) 228 (VfmaccVF) 229 (VfnmaccVF) 230 (VfmsacVF) 231 (VfnmsacVF) 232 (Vslide1upVX) 233)) 234 235;; Register-Imm ALU Ops 236(type VecAluOpRRImm5 (enum 237 ;; Regular VI Opcodes 238 (VaddVI) 239 (VsaddVI) 240 (VsadduVI) 241 (VrsubVI) 242 (VsllVI) 243 (VsrlVI) 244 (VsraVI) 245 (VandVI) 246 (VorVI) 247 (VxorVI) 248 (VssrlVI) 249 (VslidedownVI) 250 (VmergeVIM) 251 (VrgatherVI) 252 ;; This opcode represents multiple instructions `vmv1r`/`vmv2r`/`vmv4r`/etc... 253 ;; The immediate field specifies how many registers should be copied. 254 (VmvrV) 255 (VnclipWI) 256 (VnclipuWI) 257 (VmseqVI) 258 (VmsneVI) 259 (VmsleuVI) 260 (VmsleVI) 261 (VmsgtuVI) 262 (VmsgtVI) 263)) 264 265;; Imm only ALU Ops 266(type VecAluOpRImm5 (enum 267 (VmvVI) 268)) 269 270;; These are all of the special cases that have weird encodings. They are all 271;; single source, single destination instructions, and usually use one of 272;; the two source registers as auxiliary encoding space. 273(type VecAluOpRR (enum 274 (VmvSX) 275 (VmvXS) 276 (VfmvSF) 277 (VfmvFS) 278 ;; vmv.v* is special in that vs2 must be v0 (and is ignored) otherwise the instruction is illegal. 279 (VmvVV) 280 (VmvVX) 281 (VfmvVF) 282 (VfsqrtV) 283 (VsextVF2) 284 (VsextVF4) 285 (VsextVF8) 286 (VzextVF2) 287 (VzextVF4) 288 (VzextVF8) 289 (VfcvtxufV) 290 (VfcvtxfV) 291 (VfcvtrtzxufV) 292 (VfcvtrtzxfV) 293 (VfcvtfxuV) 294 (VfcvtfxV) 295 (VfwcvtffV) 296 (VfncvtffW) 297)) 298 299;; Returns the canonical destination type for a VecAluOpRRImm5. 300(decl pure vec_alu_rr_dst_type (VecAluOpRR) Type) 301(extern constructor vec_alu_rr_dst_type vec_alu_rr_dst_type) 302 303 304;; Vector Addressing Mode 305(type VecAMode (enum 306 ;; Vector unit-stride operations access elements stored contiguously in memory 307 ;; starting from the base effective address. 308 (UnitStride 309 (base AMode)) 310 ;; TODO: Constant Stride 311 ;; TODO: Indexed Operations 312)) 313 314 315;; Builds a static VState matching a SIMD type. 316;; The VState is guaranteed to be static with AVL set to the number of lanes. 317;; Element size is set to the size of the type. 318;; LMUL is set to 1. 319;; Tail mode is set to agnostic. 320;; Mask mode is set to agnostic. 321(decl pure vstate_from_type (Type) VState) 322(extern constructor vstate_from_type vstate_from_type) 323(convert Type VState vstate_from_type) 324 325;; Alters the LMUL of a VState to mf2 326(decl pure vstate_mf2 (VState) VState) 327(extern constructor vstate_mf2 vstate_mf2) 328 329;; Extracts an element width from a SIMD type. 330(decl pure element_width_from_type (Type) VecElementWidth) 331(rule (element_width_from_type ty) 332 (if-let $I8 (lane_type ty)) 333 (VecElementWidth.E8)) 334(rule (element_width_from_type ty) 335 (if-let $I16 (lane_type ty)) 336 (VecElementWidth.E16)) 337(rule (element_width_from_type ty) 338 (if-let $I32 (lane_type ty)) 339 (VecElementWidth.E32)) 340(rule (element_width_from_type ty) 341 (if-let $F32 (lane_type ty)) 342 (VecElementWidth.E32)) 343(rule (element_width_from_type ty) 344 (if-let $I64 (lane_type ty)) 345 (VecElementWidth.E64)) 346(rule (element_width_from_type ty) 347 (if-let $F64 (lane_type ty)) 348 (VecElementWidth.E64)) 349 350(decl pure min_vec_reg_size () u64) 351(extern constructor min_vec_reg_size min_vec_reg_size) 352 353;; An extractor that matches any type that is known to fit in a single vector 354;; register. 355(decl ty_vec_fits_in_register (Type) Type) 356(extern extractor ty_vec_fits_in_register ty_vec_fits_in_register) 357 358;;;; Instruction Helpers ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; 359 360;; As noted in the RISC-V Vector Extension Specification, rs2 is the first 361;; source register and rs1 is the second source register. This is the opposite 362;; of the usual RISC-V register order. 363;; See Section 10.1 of the RISC-V Vector Extension Specification. 364 365 366;; Helper for emitting `MInst.VecAluRRRR` instructions. 367;; These instructions modify the destination register. 368(decl vec_alu_rrrr (VecAluOpRRRR VReg VReg Reg VecOpMasking VState) VReg) 369(rule (vec_alu_rrrr op vd_src vs2 vs1 mask vstate) 370 (let ((vd WritableVReg (temp_writable_vreg)) 371 (_ Unit (emit (MInst.VecAluRRRR op vd vd_src vs2 vs1 mask vstate)))) 372 vd)) 373 374;; Helper for emitting `MInst.VecAluRRRImm5` instructions. 375;; These instructions modify the destination register. 376(decl vec_alu_rrr_imm5 (VecAluOpRRRImm5 VReg VReg Imm5 VecOpMasking VState) VReg) 377(rule (vec_alu_rrr_imm5 op vd_src vs2 imm mask vstate) 378 (let ((vd WritableVReg (temp_writable_vreg)) 379 (_ Unit (emit (MInst.VecAluRRRImm5 op vd vd_src vs2 imm mask vstate)))) 380 vd)) 381 382;; Helper for emitting `MInst.VecAluRRRImm5` instructions where the immediate 383;; is zero extended instead of sign extended. 384(decl vec_alu_rrr_uimm5 (VecAluOpRRRImm5 VReg VReg UImm5 VecOpMasking VState) VReg) 385(rule (vec_alu_rrr_uimm5 op vd_src vs2 imm mask vstate) 386 (vec_alu_rrr_imm5 op vd_src vs2 (uimm5_bitcast_to_imm5 imm) mask vstate)) 387 388;; Helper for emitting `MInst.VecAluRRR` instructions. 389(decl vec_alu_rrr (VecAluOpRRR Reg Reg VecOpMasking VState) Reg) 390(rule (vec_alu_rrr op vs2 vs1 mask vstate) 391 (let ((vd WritableVReg (temp_writable_vreg)) 392 (_ Unit (emit (MInst.VecAluRRR op vd vs2 vs1 mask vstate)))) 393 vd)) 394 395;; Helper for emitting `MInst.VecAluRRImm5` instructions. 396(decl vec_alu_rr_imm5 (VecAluOpRRImm5 Reg Imm5 VecOpMasking VState) Reg) 397(rule (vec_alu_rr_imm5 op vs2 imm mask vstate) 398 (let ((vd WritableVReg (temp_writable_vreg)) 399 (_ Unit (emit (MInst.VecAluRRImm5 op vd vs2 imm mask vstate)))) 400 vd)) 401 402;; Helper for emitting `MInst.VecAluRRImm5` instructions where the immediate 403;; is zero extended instead of sign extended. 404(decl vec_alu_rr_uimm5 (VecAluOpRRImm5 Reg UImm5 VecOpMasking VState) Reg) 405(rule (vec_alu_rr_uimm5 op vs2 imm mask vstate) 406 (vec_alu_rr_imm5 op vs2 (uimm5_bitcast_to_imm5 imm) mask vstate)) 407 408;; Helper for emitting `MInst.VecAluRRImm5` instructions that use the Imm5 as 409;; auxiliary encoding space. 410(decl vec_alu_rr (VecAluOpRR Reg VecOpMasking VState) Reg) 411(rule (vec_alu_rr op vs mask vstate) 412 (let ((vd WritableReg (temp_writable_reg (vec_alu_rr_dst_type op))) 413 (_ Unit (emit (MInst.VecAluRR op vd vs mask vstate)))) 414 vd)) 415 416;; Helper for emitting `MInst.VecAluRImm5` instructions. 417(decl vec_alu_r_imm5 (VecAluOpRImm5 Imm5 VecOpMasking VState) Reg) 418(rule (vec_alu_r_imm5 op imm mask vstate) 419 (let ((vd WritableVReg (temp_writable_vreg)) 420 (_ Unit (emit (MInst.VecAluRImm5 op vd imm mask vstate)))) 421 vd)) 422 423;; Helper for emitting `MInst.VecLoad` instructions. 424(decl vec_load (VecElementWidth VecAMode MemFlags VecOpMasking VState) Reg) 425(rule (vec_load eew from flags mask vstate) 426 (let ((vd WritableVReg (temp_writable_vreg)) 427 (_ Unit (emit (MInst.VecLoad eew vd from flags mask vstate)))) 428 vd)) 429 430;; Helper for emitting `MInst.VecStore` instructions. 431(decl vec_store (VecElementWidth VecAMode VReg MemFlags VecOpMasking VState) InstOutput) 432(rule (vec_store eew to from flags mask vstate) 433 (side_effect 434 (SideEffectNoResult.Inst (MInst.VecStore eew to from flags mask vstate)))) 435 436;; Helper for emitting the `vadd.vv` instruction. 437(decl rv_vadd_vv (VReg VReg VecOpMasking VState) VReg) 438(rule (rv_vadd_vv vs2 vs1 mask vstate) 439 (vec_alu_rrr (VecAluOpRRR.VaddVV) vs2 vs1 mask vstate)) 440 441;; Helper for emitting the `vadd.vx` instruction. 442(decl rv_vadd_vx (VReg XReg VecOpMasking VState) VReg) 443(rule (rv_vadd_vx vs2 vs1 mask vstate) 444 (vec_alu_rrr (VecAluOpRRR.VaddVX) vs2 vs1 mask vstate)) 445 446;; Helper for emitting the `vadd.vi` instruction. 447(decl rv_vadd_vi (VReg Imm5 VecOpMasking VState) VReg) 448(rule (rv_vadd_vi vs2 imm mask vstate) 449 (vec_alu_rr_imm5 (VecAluOpRRImm5.VaddVI) vs2 imm mask vstate)) 450 451;; Helper for emitting the `vsadd.vv` instruction. 452(decl rv_vsadd_vv (VReg VReg VecOpMasking VState) VReg) 453(rule (rv_vsadd_vv vs2 vs1 mask vstate) 454 (vec_alu_rrr (VecAluOpRRR.VsaddVV) vs2 vs1 mask vstate)) 455 456;; Helper for emitting the `vsadd.vx` instruction. 457(decl rv_vsadd_vx (VReg XReg VecOpMasking VState) VReg) 458(rule (rv_vsadd_vx vs2 vs1 mask vstate) 459 (vec_alu_rrr (VecAluOpRRR.VsaddVX) vs2 vs1 mask vstate)) 460 461;; Helper for emitting the `vsadd.vi` instruction. 462(decl rv_vsadd_vi (VReg Imm5 VecOpMasking VState) VReg) 463(rule (rv_vsadd_vi vs2 imm mask vstate) 464 (vec_alu_rr_imm5 (VecAluOpRRImm5.VsaddVI) vs2 imm mask vstate)) 465 466;; Helper for emitting the `vsaddu.vv` instruction. 467(decl rv_vsaddu_vv (VReg VReg VecOpMasking VState) VReg) 468(rule (rv_vsaddu_vv vs2 vs1 mask vstate) 469 (vec_alu_rrr (VecAluOpRRR.VsadduVV) vs2 vs1 mask vstate)) 470 471;; Helper for emitting the `vsaddu.vx` instruction. 472(decl rv_vsaddu_vx (VReg XReg VecOpMasking VState) VReg) 473(rule (rv_vsaddu_vx vs2 vs1 mask vstate) 474 (vec_alu_rrr (VecAluOpRRR.VsadduVX) vs2 vs1 mask vstate)) 475 476;; Helper for emitting the `vsaddu.vi` instruction. 477(decl rv_vsaddu_vi (VReg Imm5 VecOpMasking VState) VReg) 478(rule (rv_vsaddu_vi vs2 imm mask vstate) 479 (vec_alu_rr_imm5 (VecAluOpRRImm5.VsadduVI) vs2 imm mask vstate)) 480 481;; Helper for emitting the `vwadd.vv` instruction. 482;; 483;; Widening integer add, 2*SEW = SEW + SEW 484(decl rv_vwadd_vv (VReg VReg VecOpMasking VState) VReg) 485(rule (rv_vwadd_vv vs2 vs1 mask vstate) 486 (vec_alu_rrr (VecAluOpRRR.VwaddVV) vs2 vs1 mask vstate)) 487 488;; Helper for emitting the `vwadd.vx` instruction. 489;; 490;; Widening integer add, 2*SEW = SEW + SEW 491(decl rv_vwadd_vx (VReg XReg VecOpMasking VState) VReg) 492(rule (rv_vwadd_vx vs2 vs1 mask vstate) 493 (vec_alu_rrr (VecAluOpRRR.VwaddVX) vs2 vs1 mask vstate)) 494 495;; Helper for emitting the `vwadd.wv` instruction. 496;; 497;; Widening integer add, 2*SEW = 2*SEW + SEW 498(decl rv_vwadd_wv (VReg VReg VecOpMasking VState) VReg) 499(rule (rv_vwadd_wv vs2 vs1 mask vstate) 500 (vec_alu_rrr (VecAluOpRRR.VwaddWV) vs2 vs1 mask vstate)) 501 502;; Helper for emitting the `vwadd.wx` instruction. 503;; 504;; Widening integer add, 2*SEW = 2*SEW + SEW 505(decl rv_vwadd_wx (VReg XReg VecOpMasking VState) VReg) 506(rule (rv_vwadd_wx vs2 vs1 mask vstate) 507 (vec_alu_rrr (VecAluOpRRR.VwaddWX) vs2 vs1 mask vstate)) 508 509;; Helper for emitting the `vwaddu.vv` instruction. 510;; 511;; Widening unsigned integer add, 2*SEW = SEW + SEW 512(decl rv_vwaddu_vv (VReg VReg VecOpMasking VState) VReg) 513(rule (rv_vwaddu_vv vs2 vs1 mask vstate) 514 (vec_alu_rrr (VecAluOpRRR.VwadduVV) vs2 vs1 mask vstate)) 515 516;; Helper for emitting the `vwaddu.vv` instruction. 517;; 518;; Widening unsigned integer add, 2*SEW = SEW + SEW 519(decl rv_vwaddu_vx (VReg XReg VecOpMasking VState) VReg) 520(rule (rv_vwaddu_vx vs2 vs1 mask vstate) 521 (vec_alu_rrr (VecAluOpRRR.VwadduVX) vs2 vs1 mask vstate)) 522 523;; Helper for emitting the `vwaddu.wv` instruction. 524;; 525;; Widening integer add, 2*SEW = 2*SEW + SEW 526(decl rv_vwaddu_wv (VReg VReg VecOpMasking VState) VReg) 527(rule (rv_vwaddu_wv vs2 vs1 mask vstate) 528 (vec_alu_rrr (VecAluOpRRR.VwadduWV) vs2 vs1 mask vstate)) 529 530;; Helper for emitting the `vwaddu.wx` instruction. 531;; 532;; Widening integer add, 2*SEW = 2*SEW + SEW 533(decl rv_vwaddu_wx (VReg XReg VecOpMasking VState) VReg) 534(rule (rv_vwaddu_wx vs2 vs1 mask vstate) 535 (vec_alu_rrr (VecAluOpRRR.VwadduWX) vs2 vs1 mask vstate)) 536 537;; Helper for emitting the `vsub.vv` instruction. 538(decl rv_vsub_vv (VReg VReg VecOpMasking VState) VReg) 539(rule (rv_vsub_vv vs2 vs1 mask vstate) 540 (vec_alu_rrr (VecAluOpRRR.VsubVV) vs2 vs1 mask vstate)) 541 542;; Helper for emitting the `vsub.vx` instruction. 543(decl rv_vsub_vx (VReg XReg VecOpMasking VState) VReg) 544(rule (rv_vsub_vx vs2 vs1 mask vstate) 545 (vec_alu_rrr (VecAluOpRRR.VsubVX) vs2 vs1 mask vstate)) 546 547;; Helper for emitting the `vrsub.vx` instruction. 548(decl rv_vrsub_vx (VReg XReg VecOpMasking VState) VReg) 549(rule (rv_vrsub_vx vs2 vs1 mask vstate) 550 (vec_alu_rrr (VecAluOpRRR.VrsubVX) vs2 vs1 mask vstate)) 551 552;; Helper for emitting the `vwsub.vv` instruction. 553;; 554;; Widening integer sub, 2*SEW = SEW + SEW 555(decl rv_vwsub_vv (VReg VReg VecOpMasking VState) VReg) 556(rule (rv_vwsub_vv vs2 vs1 mask vstate) 557 (vec_alu_rrr (VecAluOpRRR.VwsubVV) vs2 vs1 mask vstate)) 558 559;; Helper for emitting the `vwsub.vx` instruction. 560;; 561;; Widening integer sub, 2*SEW = SEW + SEW 562(decl rv_vwsub_vx (VReg XReg VecOpMasking VState) VReg) 563(rule (rv_vwsub_vx vs2 vs1 mask vstate) 564 (vec_alu_rrr (VecAluOpRRR.VwsubVX) vs2 vs1 mask vstate)) 565 566;; Helper for emitting the `vwsub.wv` instruction. 567;; 568;; Widening integer sub, 2*SEW = 2*SEW + SEW 569(decl rv_vwsub_wv (VReg VReg VecOpMasking VState) VReg) 570(rule (rv_vwsub_wv vs2 vs1 mask vstate) 571 (vec_alu_rrr (VecAluOpRRR.VwsubWV) vs2 vs1 mask vstate)) 572 573;; Helper for emitting the `vwsub.wx` instruction. 574;; 575;; Widening integer sub, 2*SEW = 2*SEW + SEW 576(decl rv_vwsub_wx (VReg XReg VecOpMasking VState) VReg) 577(rule (rv_vwsub_wx vs2 vs1 mask vstate) 578 (vec_alu_rrr (VecAluOpRRR.VwsubWX) vs2 vs1 mask vstate)) 579 580;; Helper for emitting the `vwsubu.vv` instruction. 581;; 582;; Widening unsigned integer sub, 2*SEW = SEW + SEW 583(decl rv_vwsubu_vv (VReg VReg VecOpMasking VState) VReg) 584(rule (rv_vwsubu_vv vs2 vs1 mask vstate) 585 (vec_alu_rrr (VecAluOpRRR.VwsubuVV) vs2 vs1 mask vstate)) 586 587;; Helper for emitting the `vwsubu.vv` instruction. 588;; 589;; Widening unsigned integer sub, 2*SEW = SEW + SEW 590(decl rv_vwsubu_vx (VReg XReg VecOpMasking VState) VReg) 591(rule (rv_vwsubu_vx vs2 vs1 mask vstate) 592 (vec_alu_rrr (VecAluOpRRR.VwsubuVX) vs2 vs1 mask vstate)) 593 594;; Helper for emitting the `vwsubu.wv` instruction. 595;; 596;; Widening integer sub, 2*SEW = 2*SEW + SEW 597(decl rv_vwsubu_wv (VReg VReg VecOpMasking VState) VReg) 598(rule (rv_vwsubu_wv vs2 vs1 mask vstate) 599 (vec_alu_rrr (VecAluOpRRR.VwsubuWV) vs2 vs1 mask vstate)) 600 601;; Helper for emitting the `vwsubu.wx` instruction. 602;; 603;; Widening integer sub, 2*SEW = 2*SEW + SEW 604(decl rv_vwsubu_wx (VReg XReg VecOpMasking VState) VReg) 605(rule (rv_vwsubu_wx vs2 vs1 mask vstate) 606 (vec_alu_rrr (VecAluOpRRR.VwsubuWX) vs2 vs1 mask vstate)) 607 608;; Helper for emitting the `vssub.vv` instruction. 609(decl rv_vssub_vv (VReg VReg VecOpMasking VState) VReg) 610(rule (rv_vssub_vv vs2 vs1 mask vstate) 611 (vec_alu_rrr (VecAluOpRRR.VssubVV) vs2 vs1 mask vstate)) 612 613;; Helper for emitting the `vssub.vx` instruction. 614(decl rv_vssub_vx (VReg XReg VecOpMasking VState) VReg) 615(rule (rv_vssub_vx vs2 vs1 mask vstate) 616 (vec_alu_rrr (VecAluOpRRR.VssubVX) vs2 vs1 mask vstate)) 617 618;; Helper for emitting the `vssubu.vv` instruction. 619(decl rv_vssubu_vv (VReg VReg VecOpMasking VState) VReg) 620(rule (rv_vssubu_vv vs2 vs1 mask vstate) 621 (vec_alu_rrr (VecAluOpRRR.VssubuVV) vs2 vs1 mask vstate)) 622 623;; Helper for emitting the `vssubu.vx` instruction. 624(decl rv_vssubu_vx (VReg XReg VecOpMasking VState) VReg) 625(rule (rv_vssubu_vx vs2 vs1 mask vstate) 626 (vec_alu_rrr (VecAluOpRRR.VssubuVX) vs2 vs1 mask vstate)) 627 628;; Helper for emitting the `vneg.v` pseudo-instruction. 629(decl rv_vneg_v (VReg VecOpMasking VState) VReg) 630(rule (rv_vneg_v vs2 mask vstate) 631 (vec_alu_rrr (VecAluOpRRR.VrsubVX) vs2 (zero_reg) mask vstate)) 632 633;; Helper for emitting the `vrsub.vi` instruction. 634(decl rv_vrsub_vi (VReg Imm5 VecOpMasking VState) VReg) 635(rule (rv_vrsub_vi vs2 imm mask vstate) 636 (vec_alu_rr_imm5 (VecAluOpRRImm5.VrsubVI) vs2 imm mask vstate)) 637 638;; Helper for emitting the `vmul.vv` instruction. 639(decl rv_vmul_vv (VReg VReg VecOpMasking VState) VReg) 640(rule (rv_vmul_vv vs2 vs1 mask vstate) 641 (vec_alu_rrr (VecAluOpRRR.VmulVV) vs2 vs1 mask vstate)) 642 643;; Helper for emitting the `vmul.vx` instruction. 644(decl rv_vmul_vx (VReg XReg VecOpMasking VState) VReg) 645(rule (rv_vmul_vx vs2 vs1 mask vstate) 646 (vec_alu_rrr (VecAluOpRRR.VmulVX) vs2 vs1 mask vstate)) 647 648;; Helper for emitting the `vmulh.vv` instruction. 649(decl rv_vmulh_vv (VReg VReg VecOpMasking VState) VReg) 650(rule (rv_vmulh_vv vs2 vs1 mask vstate) 651 (vec_alu_rrr (VecAluOpRRR.VmulhVV) vs2 vs1 mask vstate)) 652 653;; Helper for emitting the `vmulh.vx` instruction. 654(decl rv_vmulh_vx (VReg XReg VecOpMasking VState) VReg) 655(rule (rv_vmulh_vx vs2 vs1 mask vstate) 656 (vec_alu_rrr (VecAluOpRRR.VmulhVX) vs2 vs1 mask vstate)) 657 658;; Helper for emitting the `vmulhu.vv` instruction. 659(decl rv_vmulhu_vv (VReg VReg VecOpMasking VState) VReg) 660(rule (rv_vmulhu_vv vs2 vs1 mask vstate) 661 (vec_alu_rrr (VecAluOpRRR.VmulhuVV) vs2 vs1 mask vstate)) 662 663;; Helper for emitting the `vmulhu.vx` instruction. 664(decl rv_vmulhu_vx (VReg XReg VecOpMasking VState) VReg) 665(rule (rv_vmulhu_vx vs2 vs1 mask vstate) 666 (vec_alu_rrr (VecAluOpRRR.VmulhuVX) vs2 vs1 mask vstate)) 667 668;; Helper for emitting the `vsmul.vv` instruction. 669;; 670;; Signed saturating and rounding fractional multiply 671;; # vd[i] = clip(roundoff_signed(vs2[i]*vs1[i], SEW-1)) 672(decl rv_vsmul_vv (VReg VReg VecOpMasking VState) VReg) 673(rule (rv_vsmul_vv vs2 vs1 mask vstate) 674 (vec_alu_rrr (VecAluOpRRR.VsmulVV) vs2 vs1 mask vstate)) 675 676;; Helper for emitting the `vsmul.vx` instruction. 677;; 678;; Signed saturating and rounding fractional multiply 679;; # vd[i] = clip(roundoff_signed(vs2[i]*x[rs1], SEW-1)) 680(decl rv_vsmul_vx (VReg XReg VecOpMasking VState) VReg) 681(rule (rv_vsmul_vx vs2 vs1 mask vstate) 682 (vec_alu_rrr (VecAluOpRRR.VsmulVX) vs2 vs1 mask vstate)) 683 684;; Helper for emitting the `vmacc.vv` instruction. 685;; 686;; Integer multiply-add, overwrite addend 687;; # vd[i] = +(vs1[i] * vs2[i]) + vd[i] 688(decl rv_vmacc_vv (VReg VReg VReg VecOpMasking VState) VReg) 689(rule (rv_vmacc_vv vd vs2 vs1 mask vstate) 690 (vec_alu_rrrr (VecAluOpRRRR.VmaccVV) vd vs2 vs1 mask vstate)) 691 692;; Helper for emitting the `vmacc.vx` instruction. 693;; 694;; Integer multiply-add, overwrite addend 695;; # vd[i] = +(x[rs1] * vs2[i]) + vd[i] 696(decl rv_vmacc_vx (VReg VReg XReg VecOpMasking VState) VReg) 697(rule (rv_vmacc_vx vd vs2 vs1 mask vstate) 698 (vec_alu_rrrr (VecAluOpRRRR.VmaccVX) vd vs2 vs1 mask vstate)) 699 700;; Helper for emitting the `vnmsac.vv` instruction. 701;; 702;; Integer multiply-sub, overwrite minuend 703;; # vd[i] = -(vs1[i] * vs2[i]) + vd[i] 704(decl rv_vnmsac_vv (VReg VReg VReg VecOpMasking VState) VReg) 705(rule (rv_vnmsac_vv vd vs2 vs1 mask vstate) 706 (vec_alu_rrrr (VecAluOpRRRR.VnmsacVV) vd vs2 vs1 mask vstate)) 707 708;; Helper for emitting the `vnmsac.vx` instruction. 709;; 710;; Integer multiply-sub, overwrite minuend 711;; # vd[i] = -(x[rs1] * vs2[i]) + vd[i] 712(decl rv_vnmsac_vx (VReg VReg XReg VecOpMasking VState) VReg) 713(rule (rv_vnmsac_vx vd vs2 vs1 mask vstate) 714 (vec_alu_rrrr (VecAluOpRRRR.VnmsacVX) vd vs2 vs1 mask vstate)) 715 716;; Helper for emitting the `sll.vv` instruction. 717(decl rv_vsll_vv (VReg VReg VecOpMasking VState) VReg) 718(rule (rv_vsll_vv vs2 vs1 mask vstate) 719 (vec_alu_rrr (VecAluOpRRR.VsllVV) vs2 vs1 mask vstate)) 720 721;; Helper for emitting the `sll.vx` instruction. 722(decl rv_vsll_vx (VReg XReg VecOpMasking VState) VReg) 723(rule (rv_vsll_vx vs2 vs1 mask vstate) 724 (vec_alu_rrr (VecAluOpRRR.VsllVX) vs2 vs1 mask vstate)) 725 726;; Helper for emitting the `vsll.vi` instruction. 727(decl rv_vsll_vi (VReg UImm5 VecOpMasking VState) VReg) 728(rule (rv_vsll_vi vs2 imm mask vstate) 729 (vec_alu_rr_uimm5 (VecAluOpRRImm5.VsllVI) vs2 imm mask vstate)) 730 731;; Helper for emitting the `srl.vv` instruction. 732(decl rv_vsrl_vv (VReg VReg VecOpMasking VState) VReg) 733(rule (rv_vsrl_vv vs2 vs1 mask vstate) 734 (vec_alu_rrr (VecAluOpRRR.VsrlVV) vs2 vs1 mask vstate)) 735 736;; Helper for emitting the `srl.vx` instruction. 737(decl rv_vsrl_vx (VReg XReg VecOpMasking VState) VReg) 738(rule (rv_vsrl_vx vs2 vs1 mask vstate) 739 (vec_alu_rrr (VecAluOpRRR.VsrlVX) vs2 vs1 mask vstate)) 740 741;; Helper for emitting the `vsrl.vi` instruction. 742(decl rv_vsrl_vi (VReg UImm5 VecOpMasking VState) VReg) 743(rule (rv_vsrl_vi vs2 imm mask vstate) 744 (vec_alu_rr_uimm5 (VecAluOpRRImm5.VsrlVI) vs2 imm mask vstate)) 745 746;; Helper for emitting the `sra.vv` instruction. 747(decl rv_vsra_vv (VReg VReg VecOpMasking VState) VReg) 748(rule (rv_vsra_vv vs2 vs1 mask vstate) 749 (vec_alu_rrr (VecAluOpRRR.VsraVV) vs2 vs1 mask vstate)) 750 751;; Helper for emitting the `sra.vx` instruction. 752(decl rv_vsra_vx (VReg XReg VecOpMasking VState) VReg) 753(rule (rv_vsra_vx vs2 vs1 mask vstate) 754 (vec_alu_rrr (VecAluOpRRR.VsraVX) vs2 vs1 mask vstate)) 755 756;; Helper for emitting the `vsra.vi` instruction. 757(decl rv_vsra_vi (VReg UImm5 VecOpMasking VState) VReg) 758(rule (rv_vsra_vi vs2 imm mask vstate) 759 (vec_alu_rr_uimm5 (VecAluOpRRImm5.VsraVI) vs2 imm mask vstate)) 760 761;; Helper for emitting the `vand.vv` instruction. 762(decl rv_vand_vv (VReg VReg VecOpMasking VState) VReg) 763(rule (rv_vand_vv vs2 vs1 mask vstate) 764 (vec_alu_rrr (VecAluOpRRR.VandVV) vs2 vs1 mask vstate)) 765 766;; Helper for emitting the `vand.vx` instruction. 767(decl rv_vand_vx (VReg XReg VecOpMasking VState) VReg) 768(rule (rv_vand_vx vs2 vs1 mask vstate) 769 (vec_alu_rrr (VecAluOpRRR.VandVX) vs2 vs1 mask vstate)) 770 771;; Helper for emitting the `vand.vi` instruction. 772(decl rv_vand_vi (VReg Imm5 VecOpMasking VState) VReg) 773(rule (rv_vand_vi vs2 imm mask vstate) 774 (vec_alu_rr_imm5 (VecAluOpRRImm5.VandVI) vs2 imm mask vstate)) 775 776;; Helper for emitting the `vor.vv` instruction. 777(decl rv_vor_vv (VReg VReg VecOpMasking VState) VReg) 778(rule (rv_vor_vv vs2 vs1 mask vstate) 779 (vec_alu_rrr (VecAluOpRRR.VorVV) vs2 vs1 mask vstate)) 780 781;; Helper for emitting the `vor.vx` instruction. 782(decl rv_vor_vx (VReg XReg VecOpMasking VState) VReg) 783(rule (rv_vor_vx vs2 vs1 mask vstate) 784 (vec_alu_rrr (VecAluOpRRR.VorVX) vs2 vs1 mask vstate)) 785 786;; Helper for emitting the `vor.vi` instruction. 787(decl rv_vor_vi (VReg Imm5 VecOpMasking VState) VReg) 788(rule (rv_vor_vi vs2 imm mask vstate) 789 (vec_alu_rr_imm5 (VecAluOpRRImm5.VorVI) vs2 imm mask vstate)) 790 791;; Helper for emitting the `vxor.vv` instruction. 792(decl rv_vxor_vv (VReg VReg VecOpMasking VState) VReg) 793(rule (rv_vxor_vv vs2 vs1 mask vstate) 794 (vec_alu_rrr (VecAluOpRRR.VxorVV) vs2 vs1 mask vstate)) 795 796;; Helper for emitting the `vxor.vx` instruction. 797(decl rv_vxor_vx (VReg XReg VecOpMasking VState) VReg) 798(rule (rv_vxor_vx vs2 vs1 mask vstate) 799 (vec_alu_rrr (VecAluOpRRR.VxorVX) vs2 vs1 mask vstate)) 800 801;; Helper for emitting the `vxor.vi` instruction. 802(decl rv_vxor_vi (VReg Imm5 VecOpMasking VState) VReg) 803(rule (rv_vxor_vi vs2 imm mask vstate) 804 (vec_alu_rr_imm5 (VecAluOpRRImm5.VxorVI) vs2 imm mask vstate)) 805 806;; Helper for emitting the `vssrl.vi` instruction. 807;; 808;; vd[i] = (unsigned(vs2[i]) >> imm) + r 809;; 810;; `r` here is the rounding mode currently selected. 811(decl rv_vssrl_vi (VReg UImm5 VecOpMasking VState) VReg) 812(rule (rv_vssrl_vi vs2 imm mask vstate) 813 (vec_alu_rr_uimm5 (VecAluOpRRImm5.VssrlVI) vs2 imm mask vstate)) 814 815;; Helper for emitting the `vnot.v` instruction. 816;; This is just a mnemonic for `vxor.vi vd, vs, -1` 817(decl rv_vnot_v (VReg VecOpMasking VState) VReg) 818(rule (rv_vnot_v vs2 mask vstate) 819 (if-let neg1 (i8_to_imm5 -1)) 820 (rv_vxor_vi vs2 neg1 mask vstate)) 821 822;; Helper for emitting the `vmax.vv` instruction. 823(decl rv_vmax_vv (VReg VReg VecOpMasking VState) VReg) 824(rule (rv_vmax_vv vs2 vs1 mask vstate) 825 (vec_alu_rrr (VecAluOpRRR.VmaxVV) vs2 vs1 mask vstate)) 826 827;; Helper for emitting the `vmax.vx` instruction. 828(decl rv_vmax_vx (VReg XReg VecOpMasking VState) VReg) 829(rule (rv_vmax_vx vs2 vs1 mask vstate) 830 (vec_alu_rrr (VecAluOpRRR.VmaxVX) vs2 vs1 mask vstate)) 831 832;; Helper for emitting the `vmin.vv` instruction. 833(decl rv_vmin_vv (VReg VReg VecOpMasking VState) VReg) 834(rule (rv_vmin_vv vs2 vs1 mask vstate) 835 (vec_alu_rrr (VecAluOpRRR.VminVV) vs2 vs1 mask vstate)) 836 837;; Helper for emitting the `vmin.vx` instruction. 838(decl rv_vmin_vx (VReg XReg VecOpMasking VState) VReg) 839(rule (rv_vmin_vx vs2 vs1 mask vstate) 840 (vec_alu_rrr (VecAluOpRRR.VminVX) vs2 vs1 mask vstate)) 841 842;; Helper for emitting the `vmaxu.vv` instruction. 843(decl rv_vmaxu_vv (VReg VReg VecOpMasking VState) VReg) 844(rule (rv_vmaxu_vv vs2 vs1 mask vstate) 845 (vec_alu_rrr (VecAluOpRRR.VmaxuVV) vs2 vs1 mask vstate)) 846 847;; Helper for emitting the `vmaxu.vx` instruction. 848(decl rv_vmaxu_vx (VReg XReg VecOpMasking VState) VReg) 849(rule (rv_vmaxu_vx vs2 vs1 mask vstate) 850 (vec_alu_rrr (VecAluOpRRR.VmaxuVX) vs2 vs1 mask vstate)) 851 852;; Helper for emitting the `vminu.vv` instruction. 853(decl rv_vminu_vv (VReg VReg VecOpMasking VState) VReg) 854(rule (rv_vminu_vv vs2 vs1 mask vstate) 855 (vec_alu_rrr (VecAluOpRRR.VminuVV) vs2 vs1 mask vstate)) 856 857;; Helper for emitting the `vminu.vx` instruction. 858(decl rv_vminu_vx (VReg XReg VecOpMasking VState) VReg) 859(rule (rv_vminu_vx vs2 vs1 mask vstate) 860 (vec_alu_rrr (VecAluOpRRR.VminuVX) vs2 vs1 mask vstate)) 861 862;; Helper for emitting the `vfadd.vv` instruction. 863(decl rv_vfadd_vv (VReg VReg VecOpMasking VState) VReg) 864(rule (rv_vfadd_vv vs2 vs1 mask vstate) 865 (vec_alu_rrr (VecAluOpRRR.VfaddVV) vs2 vs1 mask vstate)) 866 867;; Helper for emitting the `vfadd.vf` instruction. 868(decl rv_vfadd_vf (VReg FReg VecOpMasking VState) VReg) 869(rule (rv_vfadd_vf vs2 vs1 mask vstate) 870 (vec_alu_rrr (VecAluOpRRR.VfaddVF) vs2 vs1 mask vstate)) 871 872;; Helper for emitting the `vfsub.vv` instruction. 873(decl rv_vfsub_vv (VReg VReg VecOpMasking VState) VReg) 874(rule (rv_vfsub_vv vs2 vs1 mask vstate) 875 (vec_alu_rrr (VecAluOpRRR.VfsubVV) vs2 vs1 mask vstate)) 876 877;; Helper for emitting the `vfsub.vf` instruction. 878(decl rv_vfsub_vf (VReg FReg VecOpMasking VState) VReg) 879(rule (rv_vfsub_vf vs2 vs1 mask vstate) 880 (vec_alu_rrr (VecAluOpRRR.VfsubVF) vs2 vs1 mask vstate)) 881 882;; Helper for emitting the `vfrsub.vf` instruction. 883(decl rv_vfrsub_vf (VReg FReg VecOpMasking VState) VReg) 884(rule (rv_vfrsub_vf vs2 vs1 mask vstate) 885 (vec_alu_rrr (VecAluOpRRR.VfrsubVF) vs2 vs1 mask vstate)) 886 887;; Helper for emitting the `vfmul.vv` instruction. 888(decl rv_vfmul_vv (VReg VReg VecOpMasking VState) VReg) 889(rule (rv_vfmul_vv vs2 vs1 mask vstate) 890 (vec_alu_rrr (VecAluOpRRR.VfmulVV) vs2 vs1 mask vstate)) 891 892;; Helper for emitting the `vfmul.vf` instruction. 893(decl rv_vfmul_vf (VReg FReg VecOpMasking VState) VReg) 894(rule (rv_vfmul_vf vs2 vs1 mask vstate) 895 (vec_alu_rrr (VecAluOpRRR.VfmulVF) vs2 vs1 mask vstate)) 896 897;; Helper for emitting the `vfmacc.vv` instruction. 898;; 899;; FP multiply-accumulate, overwrites addend 900;; # vd[i] = +(vs1[i] * vs2[i]) + vd[i] 901(decl rv_vfmacc_vv (VReg VReg VReg VecOpMasking VState) VReg) 902(rule (rv_vfmacc_vv vd vs2 vs1 mask vstate) 903 (vec_alu_rrrr (VecAluOpRRRR.VfmaccVV) vd vs2 vs1 mask vstate)) 904 905;; Helper for emitting the `vfmacc.vf` instruction. 906;; 907;; FP multiply-accumulate, overwrites addend 908;; # vd[i] = +(f[rs1] * vs2[i]) + vd[i] 909(decl rv_vfmacc_vf (VReg VReg FReg VecOpMasking VState) VReg) 910(rule (rv_vfmacc_vf vd vs2 vs1 mask vstate) 911 (vec_alu_rrrr (VecAluOpRRRR.VfmaccVF) vd vs2 vs1 mask vstate)) 912 913;; Helper for emitting the `vfnmacc.vv` instruction. 914;; 915;; FP negate-(multiply-accumulate), overwrites subtrahend 916;; # vd[i] = -(vs1[i] * vs2[i]) - vd[i] 917(decl rv_vfnmacc_vv (VReg VReg VReg VecOpMasking VState) VReg) 918(rule (rv_vfnmacc_vv vd vs2 vs1 mask vstate) 919 (vec_alu_rrrr (VecAluOpRRRR.VfnmaccVV) vd vs2 vs1 mask vstate)) 920 921;; Helper for emitting the `vfnmacc.vf` instruction. 922;; 923;; FP negate-(multiply-accumulate), overwrites subtrahend 924;; # vd[i] = -(f[rs1] * vs2[i]) - vd[i] 925(decl rv_vfnmacc_vf (VReg VReg FReg VecOpMasking VState) VReg) 926(rule (rv_vfnmacc_vf vd vs2 vs1 mask vstate) 927 (vec_alu_rrrr (VecAluOpRRRR.VfnmaccVF) vd vs2 vs1 mask vstate)) 928 929;; Helper for emitting the `vfmsac.vv` instruction. 930;; 931;; FP multiply-subtract-accumulator, overwrites subtrahend 932;; # vd[i] = +(vs1[i] * vs2[i]) - vd[i] 933(decl rv_vfmsac_vv (VReg VReg VReg VecOpMasking VState) VReg) 934(rule (rv_vfmsac_vv vd vs2 vs1 mask vstate) 935 (vec_alu_rrrr (VecAluOpRRRR.VfmsacVV) vd vs2 vs1 mask vstate)) 936 937;; Helper for emitting the `vfmsac.vf` instruction. 938;; 939;; FP multiply-subtract-accumulator, overwrites subtrahend 940;; # vd[i] = +(f[rs1] * vs2[i]) - vd[i] 941(decl rv_vfmsac_vf (VReg VReg FReg VecOpMasking VState) VReg) 942(rule (rv_vfmsac_vf vd vs2 vs1 mask vstate) 943 (vec_alu_rrrr (VecAluOpRRRR.VfmsacVF) vd vs2 vs1 mask vstate)) 944 945;; Helper for emitting the `vfnmsac.vv` instruction. 946;; 947;; FP negate-(multiply-subtract-accumulator), overwrites minuend 948;; # vd[i] = -(vs1[i] * vs2[i]) + vd[i] 949(decl rv_vfnmsac_vv (VReg VReg VReg VecOpMasking VState) VReg) 950(rule (rv_vfnmsac_vv vd vs2 vs1 mask vstate) 951 (vec_alu_rrrr (VecAluOpRRRR.VfnmsacVV) vd vs2 vs1 mask vstate)) 952 953;; Helper for emitting the `vfnmsac.vf` instruction. 954;; 955;; FP negate-(multiply-subtract-accumulator), overwrites minuend 956;; # vd[i] = -(f[rs1] * vs2[i]) + vd[i] 957(decl rv_vfnmsac_vf (VReg VReg FReg VecOpMasking VState) VReg) 958(rule (rv_vfnmsac_vf vd vs2 vs1 mask vstate) 959 (vec_alu_rrrr (VecAluOpRRRR.VfnmsacVF) vd vs2 vs1 mask vstate)) 960 961;; Helper for emitting the `vfdiv.vv` instruction. 962(decl rv_vfdiv_vv (VReg VReg VecOpMasking VState) VReg) 963(rule (rv_vfdiv_vv vs2 vs1 mask vstate) 964 (vec_alu_rrr (VecAluOpRRR.VfdivVV) vs2 vs1 mask vstate)) 965 966;; Helper for emitting the `vfdiv.vf` instruction. 967(decl rv_vfdiv_vf (VReg FReg VecOpMasking VState) VReg) 968(rule (rv_vfdiv_vf vs2 vs1 mask vstate) 969 (vec_alu_rrr (VecAluOpRRR.VfdivVF) vs2 vs1 mask vstate)) 970 971;; Helper for emitting the `vfrdiv.vf` instruction. 972(decl rv_vfrdiv_vf (VReg FReg VecOpMasking VState) VReg) 973(rule (rv_vfrdiv_vf vs2 vs1 mask vstate) 974 (vec_alu_rrr (VecAluOpRRR.VfrdivVF) vs2 vs1 mask vstate)) 975 976;; Helper for emitting the `vfmin.vv` instruction. 977(decl rv_vfmin_vv (VReg VReg VecOpMasking VState) VReg) 978(rule (rv_vfmin_vv vs2 vs1 mask vstate) 979 (vec_alu_rrr (VecAluOpRRR.VfminVV) vs2 vs1 mask vstate)) 980 981;; Helper for emitting the `vfmax.vv` instruction. 982(decl rv_vfmax_vv (VReg VReg VecOpMasking VState) VReg) 983(rule (rv_vfmax_vv vs2 vs1 mask vstate) 984 (vec_alu_rrr (VecAluOpRRR.VfmaxVV) vs2 vs1 mask vstate)) 985 986;; Helper for emitting the `vfsgnj.vv` ("Floating Point Sign Injection") instruction. 987;; The output of this instruction is `vs2` with the sign bit from `vs1` 988(decl rv_vfsgnj_vv (VReg VReg VecOpMasking VState) VReg) 989(rule (rv_vfsgnj_vv vs2 vs1 mask vstate) 990 (vec_alu_rrr (VecAluOpRRR.VfsgnjVV) vs2 vs1 mask vstate)) 991 992;; Helper for emitting the `vfsgnj.vf` ("Floating Point Sign Injection") instruction. 993(decl rv_vfsgnj_vf (VReg FReg VecOpMasking VState) VReg) 994(rule (rv_vfsgnj_vf vs2 vs1 mask vstate) 995 (vec_alu_rrr (VecAluOpRRR.VfsgnjVF) vs2 vs1 mask vstate)) 996 997;; Helper for emitting the `vfsgnjn.vv` ("Floating Point Sign Injection Negated") instruction. 998;; The output of this instruction is `vs2` with the negated sign bit from `vs1` 999(decl rv_vfsgnjn_vv (VReg VReg VecOpMasking VState) VReg) 1000(rule (rv_vfsgnjn_vv vs2 vs1 mask vstate) 1001 (vec_alu_rrr (VecAluOpRRR.VfsgnjnVV) vs2 vs1 mask vstate)) 1002 1003;; Helper for emitting the `vfneg.v` instruction. 1004;; This instruction is a mnemonic for `vfsgnjn.vv vd, vs, vs` 1005(decl rv_vfneg_v (VReg VecOpMasking VState) VReg) 1006(rule (rv_vfneg_v vs mask vstate) (rv_vfsgnjn_vv vs vs mask vstate)) 1007 1008;; Helper for emitting the `vfsgnjx.vv` ("Floating Point Sign Injection Exclusive") instruction. 1009;; The output of this instruction is `vs2` with the XOR of the sign bits from `vs2` and `vs1`. 1010;; When `vs2 == vs1` this implements `fabs` 1011(decl rv_vfsgnjx_vv (VReg VReg VecOpMasking VState) VReg) 1012(rule (rv_vfsgnjx_vv vs2 vs1 mask vstate) 1013 (vec_alu_rrr (VecAluOpRRR.VfsgnjxVV) vs2 vs1 mask vstate)) 1014 1015;; Helper for emitting the `vfabs.v` instruction. 1016;; This instruction is a mnemonic for `vfsgnjx.vv vd, vs, vs` 1017(decl rv_vfabs_v (VReg VecOpMasking VState) VReg) 1018(rule (rv_vfabs_v vs mask vstate) (rv_vfsgnjx_vv vs vs mask vstate)) 1019 1020;; Helper for emitting the `vfsqrt.v` instruction. 1021;; This instruction splats the F register into all elements of the destination vector. 1022(decl rv_vfsqrt_v (VReg VecOpMasking VState) VReg) 1023(rule (rv_vfsqrt_v vs mask vstate) 1024 (vec_alu_rr (VecAluOpRR.VfsqrtV) vs mask vstate)) 1025 1026;; Helper for emitting the `vfcvt.xu.f.v` instruction. 1027;; This instruction converts a float to an unsigned integer. 1028(decl rv_vfcvt_xu_f_v (VReg VecOpMasking VState) VReg) 1029(rule (rv_vfcvt_xu_f_v vs mask vstate) 1030 (vec_alu_rr (VecAluOpRR.VfcvtxufV) vs mask vstate)) 1031 1032;; Helper for emitting the `vfcvt.x.f.v` instruction. 1033;; This instruction converts a float to a signed integer. 1034(decl rv_vfcvt_x_f_v (VReg VecOpMasking VState) VReg) 1035(rule (rv_vfcvt_x_f_v vs mask vstate) 1036 (vec_alu_rr (VecAluOpRR.VfcvtxfV) vs mask vstate)) 1037 1038;; Helper for emitting the `vfcvt.rtz.xu.f.v` instruction. 1039;; This instruction converts a float to an unsigned integer 1040;; using the Round to Zero (RTZ) rounding mode and ignoring 1041;; the currently set FRM rounding mode. 1042(decl rv_vfcvt_rtz_xu_f_v (VReg VecOpMasking VState) VReg) 1043(rule (rv_vfcvt_rtz_xu_f_v vs mask vstate) 1044 (vec_alu_rr (VecAluOpRR.VfcvtrtzxufV) vs mask vstate)) 1045 1046;; Helper for emitting the `vfcvt.rtz.x.f.v` instruction. 1047;; This instruction converts a float to a signed integer. 1048;; using the Round to Zero (RTZ) rounding mode and ignoring 1049;; the currently set FRM rounding mode. 1050(decl rv_vfcvt_rtz_x_f_v (VReg VecOpMasking VState) VReg) 1051(rule (rv_vfcvt_rtz_x_f_v vs mask vstate) 1052 (vec_alu_rr (VecAluOpRR.VfcvtrtzxfV) vs mask vstate)) 1053 1054;; Helper for emitting the `vfcvt.f.xu.v` instruction. 1055;; This instruction converts a unsigned integer to a float. 1056(decl rv_vfcvt_f_xu_v (VReg VecOpMasking VState) VReg) 1057(rule (rv_vfcvt_f_xu_v vs mask vstate) 1058 (vec_alu_rr (VecAluOpRR.VfcvtfxuV) vs mask vstate)) 1059 1060;; Helper for emitting the `vfcvt.x.f.v` instruction. 1061;; This instruction converts a signed integer to a float. 1062(decl rv_vfcvt_f_x_v (VReg VecOpMasking VState) VReg) 1063(rule (rv_vfcvt_f_x_v vs mask vstate) 1064 (vec_alu_rr (VecAluOpRR.VfcvtfxV) vs mask vstate)) 1065 1066 ;; Helper for emitting the `vfwcvt.f.f.v` instruction. 1067;; Convert single-width float to double-width float. 1068(decl rv_vfwcvt_f_f_v (VReg VecOpMasking VState) VReg) 1069(rule (rv_vfwcvt_f_f_v vs mask vstate) 1070 (vec_alu_rr (VecAluOpRR.VfwcvtffV) vs mask vstate)) 1071 1072;; Helper for emitting the `vfncvt.f.f.w` instruction. 1073;; Convert double-width float to single-width float. 1074(decl rv_vfncvt_f_f_w (VReg VecOpMasking VState) VReg) 1075(rule (rv_vfncvt_f_f_w vs mask vstate) 1076 (vec_alu_rr (VecAluOpRR.VfncvtffW) vs mask vstate)) 1077 1078;; Helper for emitting the `vslidedown.vx` instruction. 1079;; `vslidedown` moves all elements in the vector down by n elements. 1080;; The top most elements are up to the tail policy. 1081(decl rv_vslidedown_vx (VReg XReg VecOpMasking VState) VReg) 1082(rule (rv_vslidedown_vx vs2 vs1 mask vstate) 1083 (vec_alu_rrr (VecAluOpRRR.VslidedownVX) vs2 vs1 mask vstate)) 1084 1085;; Helper for emitting the `vslidedown.vi` instruction. 1086;; Unlike other `vi` instructions the immediate is zero extended. 1087(decl rv_vslidedown_vi (VReg UImm5 VecOpMasking VState) VReg) 1088(rule (rv_vslidedown_vi vs2 imm mask vstate) 1089 (vec_alu_rr_uimm5 (VecAluOpRRImm5.VslidedownVI) vs2 imm mask vstate)) 1090 1091;; Helper for emitting the `vslideup.vi` instruction. 1092;; Unlike other `vi` instructions the immediate is zero extended. 1093;; This is implemented as a 2 source operand instruction, since it only 1094;; partially modifies the destination register. 1095(decl rv_vslideup_vvi (VReg VReg UImm5 VecOpMasking VState) VReg) 1096(rule (rv_vslideup_vvi vd vs2 imm mask vstate) 1097 (vec_alu_rrr_uimm5 (VecAluOpRRRImm5.VslideupVI) vd vs2 imm mask vstate)) 1098 1099;; Helper for emitting the `vslide1up.vx` instruction. 1100;; 1101;; # vd[0]=x[rs1], vd[i+1] = vs2[i] 1102(decl rv_vslide1up_vx (VReg VReg XReg VecOpMasking VState) VReg) 1103(rule (rv_vslide1up_vx vd vs2 rs1 mask vstate) 1104 (vec_alu_rrrr (VecAluOpRRRR.Vslide1upVX) vd vs2 rs1 mask vstate)) 1105 1106;; Helper for emitting the `vmv.x.s` instruction. 1107;; This instruction copies the first element of the source vector to the destination X register. 1108;; Masked versions of this instruction are not supported. 1109(decl rv_vmv_xs (VReg VState) XReg) 1110(rule (rv_vmv_xs vs vstate) 1111 (vec_alu_rr (VecAluOpRR.VmvXS) vs (unmasked) vstate)) 1112 1113;; Helper for emitting the `vfmv.f.s` instruction. 1114;; This instruction copies the first element of the source vector to the destination F register. 1115;; Masked versions of this instruction are not supported. 1116(decl rv_vfmv_fs (VReg VState) FReg) 1117(rule (rv_vfmv_fs vs vstate) 1118 (vec_alu_rr (VecAluOpRR.VfmvFS) vs (unmasked) vstate)) 1119 1120;; Helper for emitting the `vmv.s.x` instruction. 1121;; This instruction copies the source X register into first element of the source vector. 1122;; Masked versions of this instruction are not supported. 1123(decl rv_vmv_sx (XReg VState) VReg) 1124(rule (rv_vmv_sx vs vstate) 1125 (vec_alu_rr (VecAluOpRR.VmvSX) vs (unmasked) vstate)) 1126 1127;; Helper for emitting the `vfmv.s.f` instruction. 1128;; This instruction copies the source F register into first element of the source vector. 1129;; Masked versions of this instruction are not supported. 1130(decl rv_vfmv_sf (FReg VState) VReg) 1131(rule (rv_vfmv_sf vs vstate) 1132 (vec_alu_rr (VecAluOpRR.VfmvSF) vs (unmasked) vstate)) 1133 1134;; Helper for emitting the `vmv.v.x` instruction. 1135;; This instruction splats the X register into all elements of the destination vector. 1136;; Masked versions of this instruction are called `vmerge` 1137(decl rv_vmv_vx (XReg VState) VReg) 1138(rule (rv_vmv_vx vs vstate) 1139 (vec_alu_rr (VecAluOpRR.VmvVX) vs (unmasked) vstate)) 1140 1141;; Helper for emitting the `vfmv.v.f` instruction. 1142;; This instruction splats the F register into all elements of the destination vector. 1143;; Masked versions of this instruction are called `vmerge` 1144(decl rv_vfmv_vf (FReg VState) VReg) 1145(rule (rv_vfmv_vf vs vstate) 1146 (vec_alu_rr (VecAluOpRR.VfmvVF) vs (unmasked) vstate)) 1147 1148;; Helper for emitting the `vmv.v.i` instruction. 1149;; This instruction splat's the immediate value into all elements of the destination vector. 1150;; Masked versions of this instruction are called `vmerge` 1151(decl rv_vmv_vi (Imm5 VState) VReg) 1152(rule (rv_vmv_vi imm vstate) 1153 (vec_alu_r_imm5 (VecAluOpRImm5.VmvVI) imm (unmasked) vstate)) 1154 1155;; Helper for emitting the `vmerge.vvm` instruction. 1156;; This instruction merges the elements of the two source vectors into the destination vector 1157;; based on a mask. Elements are taken from the first source vector if the mask bit is clear, 1158;; and from the second source vector if the mask bit is set. This instruction is always masked. 1159;; 1160;; vd[i] = v0.mask[i] ? vs1[i] : vs2[i] 1161(decl rv_vmerge_vvm (VReg VReg VReg VState) VReg) 1162(rule (rv_vmerge_vvm vs2 vs1 mask vstate) 1163 (vec_alu_rrr (VecAluOpRRR.VmergeVVM) vs2 vs1 (masked mask) vstate)) 1164 1165;; Helper for emitting the `vmerge.vxm` instruction. 1166;; Elements are taken from the first source vector if the mask bit is clear, and from the X 1167;; register if the mask bit is set. This instruction is always masked. 1168;; 1169;; vd[i] = v0.mask[i] ? x[rs1] : vs2[i] 1170(decl rv_vmerge_vxm (VReg XReg VReg VState) VReg) 1171(rule (rv_vmerge_vxm vs2 vs1 mask vstate) 1172 (vec_alu_rrr (VecAluOpRRR.VmergeVXM) vs2 vs1 (masked mask) vstate)) 1173 1174;; Helper for emitting the `vfmerge.vfm` instruction. 1175;; Elements are taken from the first source vector if the mask bit is clear, and from the F 1176;; register if the mask bit is set. This instruction is always masked. 1177;; 1178;; vd[i] = v0.mask[i] ? f[rs1] : vs2[i] 1179(decl rv_vfmerge_vfm (VReg FReg VReg VState) VReg) 1180(rule (rv_vfmerge_vfm vs2 vs1 mask vstate) 1181 (vec_alu_rrr (VecAluOpRRR.VfmergeVFM) vs2 vs1 (masked mask) vstate)) 1182 1183;; Helper for emitting the `vmerge.vim` instruction. 1184;; Elements are taken from the first source vector if the mask bit is clear, and from the 1185;; immediate value if the mask bit is set. This instruction is always masked. 1186;; 1187;; vd[i] = v0.mask[i] ? imm : vs2[i] 1188(decl rv_vmerge_vim (VReg Imm5 VReg VState) VReg) 1189(rule (rv_vmerge_vim vs2 imm mask vstate) 1190 (vec_alu_rr_imm5 (VecAluOpRRImm5.VmergeVIM) vs2 imm (masked mask) vstate)) 1191 1192 1193;; Helper for emitting the `vredminu.vs` instruction. 1194;; 1195;; vd[0] = minu( vs1[0] , vs2[*] ) 1196(decl rv_vredminu_vs (VReg VReg VecOpMasking VState) VReg) 1197(rule (rv_vredminu_vs vs2 vs1 mask vstate) 1198 (vec_alu_rrr (VecAluOpRRR.VredminuVS) vs2 vs1 mask vstate)) 1199 1200;; Helper for emitting the `vredmaxu.vs` instruction. 1201;; 1202;; vd[0] = maxu( vs1[0] , vs2[*] ) 1203(decl rv_vredmaxu_vs (VReg VReg VecOpMasking VState) VReg) 1204(rule (rv_vredmaxu_vs vs2 vs1 mask vstate) 1205 (vec_alu_rrr (VecAluOpRRR.VredmaxuVS) vs2 vs1 mask vstate)) 1206 1207;; Helper for emitting the `vrgather.vv` instruction. 1208;; 1209;; vd[i] = (vs1[i] >= VLMAX) ? 0 : vs2[vs1[i]]; 1210(decl rv_vrgather_vv (VReg VReg VecOpMasking VState) VReg) 1211(rule (rv_vrgather_vv vs2 vs1 mask vstate) 1212 (vec_alu_rrr (VecAluOpRRR.VrgatherVV) vs2 vs1 mask vstate)) 1213 1214;; Helper for emitting the `vrgather.vx` instruction. 1215;; 1216;; vd[i] = (x[rs1] >= VLMAX) ? 0 : vs2[x[rs1]] 1217(decl rv_vrgather_vx (VReg XReg VecOpMasking VState) VReg) 1218(rule (rv_vrgather_vx vs2 vs1 mask vstate) 1219 (vec_alu_rrr (VecAluOpRRR.VrgatherVX) vs2 vs1 mask vstate)) 1220 1221;; Helper for emitting the `vrgather.vi` instruction. 1222(decl rv_vrgather_vi (VReg UImm5 VecOpMasking VState) VReg) 1223(rule (rv_vrgather_vi vs2 imm mask vstate) 1224 (vec_alu_rr_uimm5 (VecAluOpRRImm5.VrgatherVI) vs2 imm mask vstate)) 1225 1226;; Helper for emitting the `vcompress.vm` instruction. 1227;; 1228;; The vector compress instruction allows elements selected by a vector mask 1229;; register from a source vector register group to be packed into contiguous 1230;; elements at the start of the destination vector register group. 1231;; 1232;; The mask register is specified through vs1 1233(decl rv_vcompress_vm (VReg VReg VState) VReg) 1234(rule (rv_vcompress_vm vs2 vs1 vstate) 1235 (vec_alu_rrr (VecAluOpRRR.VcompressVM) vs2 vs1 (unmasked) vstate)) 1236 1237;; Helper for emitting the `vmseq.vv` (Vector Mask Set If Equal) instruction. 1238(decl rv_vmseq_vv (VReg VReg VecOpMasking VState) VReg) 1239(rule (rv_vmseq_vv vs2 vs1 mask vstate) 1240 (vec_alu_rrr (VecAluOpRRR.VmseqVV) vs2 vs1 mask vstate)) 1241 1242;; Helper for emitting the `vmseq.vx` (Vector Mask Set If Equal) instruction. 1243(decl rv_vmseq_vx (VReg XReg VecOpMasking VState) VReg) 1244(rule (rv_vmseq_vx vs2 vs1 mask vstate) 1245 (vec_alu_rrr (VecAluOpRRR.VmseqVX) vs2 vs1 mask vstate)) 1246 1247;; Helper for emitting the `vmseq.vi` (Vector Mask Set If Equal) instruction. 1248(decl rv_vmseq_vi (VReg Imm5 VecOpMasking VState) VReg) 1249(rule (rv_vmseq_vi vs2 imm mask vstate) 1250 (vec_alu_rr_imm5 (VecAluOpRRImm5.VmseqVI) vs2 imm mask vstate)) 1251 1252;; Helper for emitting the `vmsne.vv` (Vector Mask Set If Not Equal) instruction. 1253(decl rv_vmsne_vv (VReg VReg VecOpMasking VState) VReg) 1254(rule (rv_vmsne_vv vs2 vs1 mask vstate) 1255 (vec_alu_rrr (VecAluOpRRR.VmsneVV) vs2 vs1 mask vstate)) 1256 1257;; Helper for emitting the `vmsne.vx` (Vector Mask Set If Not Equal) instruction. 1258(decl rv_vmsne_vx (VReg XReg VecOpMasking VState) VReg) 1259(rule (rv_vmsne_vx vs2 vs1 mask vstate) 1260 (vec_alu_rrr (VecAluOpRRR.VmsneVX) vs2 vs1 mask vstate)) 1261 1262;; Helper for emitting the `vmsne.vi` (Vector Mask Set If Not Equal) instruction. 1263(decl rv_vmsne_vi (VReg Imm5 VecOpMasking VState) VReg) 1264(rule (rv_vmsne_vi vs2 imm mask vstate) 1265 (vec_alu_rr_imm5 (VecAluOpRRImm5.VmsneVI) vs2 imm mask vstate)) 1266 1267;; Helper for emitting the `vmsltu.vv` (Vector Mask Set If Less Than, Unsigned) instruction. 1268(decl rv_vmsltu_vv (VReg VReg VecOpMasking VState) VReg) 1269(rule (rv_vmsltu_vv vs2 vs1 mask vstate) 1270 (vec_alu_rrr (VecAluOpRRR.VmsltuVV) vs2 vs1 mask vstate)) 1271 1272;; Helper for emitting the `vmsltu.vx` (Vector Mask Set If Less Than, Unsigned) instruction. 1273(decl rv_vmsltu_vx (VReg XReg VecOpMasking VState) VReg) 1274(rule (rv_vmsltu_vx vs2 vs1 mask vstate) 1275 (vec_alu_rrr (VecAluOpRRR.VmsltuVX) vs2 vs1 mask vstate)) 1276 1277;; Helper for emitting the `vmslt.vv` (Vector Mask Set If Less Than) instruction. 1278(decl rv_vmslt_vv (VReg VReg VecOpMasking VState) VReg) 1279(rule (rv_vmslt_vv vs2 vs1 mask vstate) 1280 (vec_alu_rrr (VecAluOpRRR.VmsltVV) vs2 vs1 mask vstate)) 1281 1282;; Helper for emitting the `vmslt.vx` (Vector Mask Set If Less Than) instruction. 1283(decl rv_vmslt_vx (VReg XReg VecOpMasking VState) VReg) 1284(rule (rv_vmslt_vx vs2 vs1 mask vstate) 1285 (vec_alu_rrr (VecAluOpRRR.VmsltVX) vs2 vs1 mask vstate)) 1286 1287;; Helper for emitting the `vmsleu.vv` (Vector Mask Set If Less Than or Equal, Unsigned) instruction. 1288(decl rv_vmsleu_vv (VReg VReg VecOpMasking VState) VReg) 1289(rule (rv_vmsleu_vv vs2 vs1 mask vstate) 1290 (vec_alu_rrr (VecAluOpRRR.VmsleuVV) vs2 vs1 mask vstate)) 1291 1292;; Helper for emitting the `vmsleu.vx` (Vector Mask Set If Less Than or Equal, Unsigned) instruction. 1293(decl rv_vmsleu_vx (VReg XReg VecOpMasking VState) VReg) 1294(rule (rv_vmsleu_vx vs2 vs1 mask vstate) 1295 (vec_alu_rrr (VecAluOpRRR.VmsleuVX) vs2 vs1 mask vstate)) 1296 1297;; Helper for emitting the `vmsleu.vi` (Vector Mask Set If Less Than or Equal, Unsigned) instruction. 1298(decl rv_vmsleu_vi (VReg Imm5 VecOpMasking VState) VReg) 1299(rule (rv_vmsleu_vi vs2 imm mask vstate) 1300 (vec_alu_rr_imm5 (VecAluOpRRImm5.VmsleuVI) vs2 imm mask vstate)) 1301 1302;; Helper for emitting the `vmsle.vv` (Vector Mask Set If Less Than or Equal) instruction. 1303(decl rv_vmsle_vv (VReg VReg VecOpMasking VState) VReg) 1304(rule (rv_vmsle_vv vs2 vs1 mask vstate) 1305 (vec_alu_rrr (VecAluOpRRR.VmsleVV) vs2 vs1 mask vstate)) 1306 1307;; Helper for emitting the `vmsle.vx` (Vector Mask Set If Less Than or Equal) instruction. 1308(decl rv_vmsle_vx (VReg XReg VecOpMasking VState) VReg) 1309(rule (rv_vmsle_vx vs2 vs1 mask vstate) 1310 (vec_alu_rrr (VecAluOpRRR.VmsleVX) vs2 vs1 mask vstate)) 1311 1312;; Helper for emitting the `vmsle.vi` (Vector Mask Set If Less Than or Equal) instruction. 1313(decl rv_vmsle_vi (VReg Imm5 VecOpMasking VState) VReg) 1314(rule (rv_vmsle_vi vs2 imm mask vstate) 1315 (vec_alu_rr_imm5 (VecAluOpRRImm5.VmsleVI) vs2 imm mask vstate)) 1316 1317;; Helper for emitting the `vmsgt.vv` (Vector Mask Set If Greater Than, Unsigned) instruction. 1318;; This is an alias for `vmsltu.vv` with the operands inverted. 1319(decl rv_vmsgtu_vv (VReg VReg VecOpMasking VState) VReg) 1320(rule (rv_vmsgtu_vv vs2 vs1 mask vstate) (rv_vmsltu_vv vs1 vs2 mask vstate)) 1321 1322;; Helper for emitting the `vmsgtu.vx` (Vector Mask Set If Greater Than, Unsigned) instruction. 1323(decl rv_vmsgtu_vx (VReg XReg VecOpMasking VState) VReg) 1324(rule (rv_vmsgtu_vx vs2 vs1 mask vstate) 1325 (vec_alu_rrr (VecAluOpRRR.VmsgtuVX) vs2 vs1 mask vstate)) 1326 1327;; Helper for emitting the `vmsgtu.vi` (Vector Mask Set If Greater Than, Unsigned) instruction. 1328(decl rv_vmsgtu_vi (VReg Imm5 VecOpMasking VState) VReg) 1329(rule (rv_vmsgtu_vi vs2 imm mask vstate) 1330 (vec_alu_rr_imm5 (VecAluOpRRImm5.VmsgtuVI) vs2 imm mask vstate)) 1331 1332;; Helper for emitting the `vmsgt.vv` (Vector Mask Set If Greater Than) instruction. 1333;; This is an alias for `vmslt.vv` with the operands inverted. 1334(decl rv_vmsgt_vv (VReg VReg VecOpMasking VState) VReg) 1335(rule (rv_vmsgt_vv vs2 vs1 mask vstate) (rv_vmslt_vv vs1 vs2 mask vstate)) 1336 1337;; Helper for emitting the `vmsgt.vx` (Vector Mask Set If Greater Than) instruction. 1338(decl rv_vmsgt_vx (VReg XReg VecOpMasking VState) VReg) 1339(rule (rv_vmsgt_vx vs2 vs1 mask vstate) 1340 (vec_alu_rrr (VecAluOpRRR.VmsgtVX) vs2 vs1 mask vstate)) 1341 1342;; Helper for emitting the `vmsgt.vi` (Vector Mask Set If Greater Than) instruction. 1343(decl rv_vmsgt_vi (VReg Imm5 VecOpMasking VState) VReg) 1344(rule (rv_vmsgt_vi vs2 imm mask vstate) 1345 (vec_alu_rr_imm5 (VecAluOpRRImm5.VmsgtVI) vs2 imm mask vstate)) 1346 1347;; Helper for emitting the `vmsgeu.vv` (Vector Mask Set If Greater Than or Equal, Unsigned) instruction. 1348;; This is an alias for `vmsleu.vv` with the operands inverted. 1349(decl rv_vmsgeu_vv (VReg VReg VecOpMasking VState) VReg) 1350(rule (rv_vmsgeu_vv vs2 vs1 mask vstate) (rv_vmsleu_vv vs1 vs2 mask vstate)) 1351 1352;; Helper for emitting the `vmsge.vv` (Vector Mask Set If Greater Than or Equal) instruction. 1353;; This is an alias for `vmsle.vv` with the operands inverted. 1354(decl rv_vmsge_vv (VReg VReg VecOpMasking VState) VReg) 1355(rule (rv_vmsge_vv vs2 vs1 mask vstate) (rv_vmsle_vv vs1 vs2 mask vstate)) 1356 1357;; Helper for emitting the `vmfeq.vv` (Vector Mask Set If Float Equal) instruction. 1358(decl rv_vmfeq_vv (VReg VReg VecOpMasking VState) VReg) 1359(rule (rv_vmfeq_vv vs2 vs1 mask vstate) 1360 (vec_alu_rrr (VecAluOpRRR.VmfeqVV) vs2 vs1 mask vstate)) 1361 1362;; Helper for emitting the `vmfeq.vf` (Vector Mask Set If Float Equal) instruction. 1363(decl rv_vmfeq_vf (VReg FReg VecOpMasking VState) VReg) 1364(rule (rv_vmfeq_vf vs2 vs1 mask vstate) 1365 (vec_alu_rrr (VecAluOpRRR.VmfeqVF) vs2 vs1 mask vstate)) 1366 1367;; Helper for emitting the `vmfne.vv` (Vector Mask Set If Float Not Equal) instruction. 1368(decl rv_vmfne_vv (VReg VReg VecOpMasking VState) VReg) 1369(rule (rv_vmfne_vv vs2 vs1 mask vstate) 1370 (vec_alu_rrr (VecAluOpRRR.VmfneVV) vs2 vs1 mask vstate)) 1371 1372;; Helper for emitting the `vmfne.vf` (Vector Mask Set If Float Not Equal) instruction. 1373(decl rv_vmfne_vf (VReg FReg VecOpMasking VState) VReg) 1374(rule (rv_vmfne_vf vs2 vs1 mask vstate) 1375 (vec_alu_rrr (VecAluOpRRR.VmfneVF) vs2 vs1 mask vstate)) 1376 1377;; Helper for emitting the `vmflt.vv` (Vector Mask Set If Float Less Than) instruction. 1378(decl rv_vmflt_vv (VReg VReg VecOpMasking VState) VReg) 1379(rule (rv_vmflt_vv vs2 vs1 mask vstate) 1380 (vec_alu_rrr (VecAluOpRRR.VmfltVV) vs2 vs1 mask vstate)) 1381 1382;; Helper for emitting the `vmflt.vf` (Vector Mask Set If Float Less Than) instruction. 1383(decl rv_vmflt_vf (VReg FReg VecOpMasking VState) VReg) 1384(rule (rv_vmflt_vf vs2 vs1 mask vstate) 1385 (vec_alu_rrr (VecAluOpRRR.VmfltVF) vs2 vs1 mask vstate)) 1386 1387;; Helper for emitting the `vmfle.vv` (Vector Mask Set If Float Less Than Or Equal) instruction. 1388(decl rv_vmfle_vv (VReg VReg VecOpMasking VState) VReg) 1389(rule (rv_vmfle_vv vs2 vs1 mask vstate) 1390 (vec_alu_rrr (VecAluOpRRR.VmfleVV) vs2 vs1 mask vstate)) 1391 1392;; Helper for emitting the `vmfle.vf` (Vector Mask Set If Float Less Than Or Equal) instruction. 1393(decl rv_vmfle_vf (VReg FReg VecOpMasking VState) VReg) 1394(rule (rv_vmfle_vf vs2 vs1 mask vstate) 1395 (vec_alu_rrr (VecAluOpRRR.VmfleVF) vs2 vs1 mask vstate)) 1396 1397;; Helper for emitting the `vmfgt.vv` (Vector Mask Set If Float Greater Than) instruction. 1398;; This is an alias for `vmflt.vv` with the operands inverted. 1399(decl rv_vmfgt_vv (VReg VReg VecOpMasking VState) VReg) 1400(rule (rv_vmfgt_vv vs2 vs1 mask vstate) (rv_vmflt_vv vs1 vs2 mask vstate)) 1401 1402;; Helper for emitting the `vmfgt.vf` (Vector Mask Set If Float Greater Than) instruction. 1403(decl rv_vmfgt_vf (VReg FReg VecOpMasking VState) VReg) 1404(rule (rv_vmfgt_vf vs2 vs1 mask vstate) 1405 (vec_alu_rrr (VecAluOpRRR.VmfgtVF) vs2 vs1 mask vstate)) 1406 1407;; Helper for emitting the `vmfge.vv` (Vector Mask Set If Float Greater Than Or Equal) instruction. 1408;; This is an alias for `vmfle.vv` with the operands inverted. 1409(decl rv_vmfge_vv (VReg VReg VecOpMasking VState) VReg) 1410(rule (rv_vmfge_vv vs2 vs1 mask vstate) (rv_vmfle_vv vs1 vs2 mask vstate)) 1411 1412;; Helper for emitting the `vmfge.vf` (Vector Mask Set If Float Greater Than Or Equal) instruction. 1413(decl rv_vmfge_vf (VReg FReg VecOpMasking VState) VReg) 1414(rule (rv_vmfge_vf vs2 vs1 mask vstate) 1415 (vec_alu_rrr (VecAluOpRRR.VmfgeVF) vs2 vs1 mask vstate)) 1416 1417;; Helper for emitting the `vzext.vf2` instruction. 1418;; Zero-extend SEW/2 source to SEW destination 1419(decl rv_vzext_vf2 (VReg VecOpMasking VState) VReg) 1420(rule (rv_vzext_vf2 vs mask vstate) 1421 (vec_alu_rr (VecAluOpRR.VzextVF2) vs mask vstate)) 1422 1423;; Helper for emitting the `vzext.vf4` instruction. 1424;; Zero-extend SEW/4 source to SEW destination 1425(decl rv_vzext_vf4 (VReg VecOpMasking VState) VReg) 1426(rule (rv_vzext_vf4 vs mask vstate) 1427 (vec_alu_rr (VecAluOpRR.VzextVF4) vs mask vstate)) 1428 1429;; Helper for emitting the `vzext.vf8` instruction. 1430;; Zero-extend SEW/8 source to SEW destination 1431(decl rv_vzext_vf8 (VReg VecOpMasking VState) VReg) 1432(rule (rv_vzext_vf8 vs mask vstate) 1433 (vec_alu_rr (VecAluOpRR.VzextVF8) vs mask vstate)) 1434 1435;; Helper for emitting the `vsext.vf2` instruction. 1436;; Sign-extend SEW/2 source to SEW destination 1437(decl rv_vsext_vf2 (VReg VecOpMasking VState) VReg) 1438(rule (rv_vsext_vf2 vs mask vstate) 1439 (vec_alu_rr (VecAluOpRR.VsextVF2) vs mask vstate)) 1440 1441;; Helper for emitting the `vsext.vf4` instruction. 1442;; Sign-extend SEW/4 source to SEW destination 1443(decl rv_vsext_vf4 (VReg VecOpMasking VState) VReg) 1444(rule (rv_vsext_vf4 vs mask vstate) 1445 (vec_alu_rr (VecAluOpRR.VsextVF4) vs mask vstate)) 1446 1447;; Helper for emitting the `vsext.vf8` instruction. 1448;; Sign-extend SEW/8 source to SEW destination 1449(decl rv_vsext_vf8 (VReg VecOpMasking VState) VReg) 1450(rule (rv_vsext_vf8 vs mask vstate) 1451 (vec_alu_rr (VecAluOpRR.VsextVF8) vs mask vstate)) 1452 1453;; Helper for emitting the `vnclip.wi` instruction. 1454;; 1455;; vd[i] = clip(roundoff_signed(vs2[i], uimm)) 1456(decl rv_vnclip_wi (VReg UImm5 VecOpMasking VState) VReg) 1457(rule (rv_vnclip_wi vs2 imm mask vstate) 1458 (vec_alu_rr_uimm5 (VecAluOpRRImm5.VnclipWI) vs2 imm mask vstate)) 1459 1460;; Helper for emitting the `vnclipu.wi` instruction. 1461;; 1462;; vd[i] = clip(roundoff_unsigned(vs2[i], uimm)) 1463(decl rv_vnclipu_wi (VReg UImm5 VecOpMasking VState) VReg) 1464(rule (rv_vnclipu_wi vs2 imm mask vstate) 1465 (vec_alu_rr_uimm5 (VecAluOpRRImm5.VnclipuWI) vs2 imm mask vstate)) 1466 1467;; Helper for emitting the `vmand.mm` (Mask Bitwise AND) instruction. 1468;; 1469;; vd.mask[i] = vs2.mask[i] && vs1.mask[i] 1470(decl rv_vmand_mm (VReg VReg VState) VReg) 1471(rule (rv_vmand_mm vs2 vs1 vstate) 1472 (vec_alu_rrr (VecAluOpRRR.VmandMM) vs2 vs1 (unmasked) vstate)) 1473 1474;; Helper for emitting the `vmor.mm` (Mask Bitwise OR) instruction. 1475;; 1476;; vd.mask[i] = vs2.mask[i] || vs1.mask[i] 1477(decl rv_vmor_mm (VReg VReg VState) VReg) 1478(rule (rv_vmor_mm vs2 vs1 vstate) 1479 (vec_alu_rrr (VecAluOpRRR.VmorMM) vs2 vs1 (unmasked) vstate)) 1480 1481;; Helper for emitting the `vmnand.mm` (Mask Bitwise NAND) instruction. 1482;; 1483;; vd.mask[i] = !(vs2.mask[i] && vs1.mask[i]) 1484(decl rv_vmnand_mm (VReg VReg VState) VReg) 1485(rule (rv_vmnand_mm vs2 vs1 vstate) 1486 (vec_alu_rrr (VecAluOpRRR.VmnandMM) vs2 vs1 (unmasked) vstate)) 1487 1488;; Helper for emitting the `vmnot.m` (Mask Bitwise NOT) instruction. 1489;; This is an alias for `vmnand.mm vd, vs, vs` 1490;; 1491;; vd.mask[i] = !vs.mask[i] 1492(decl rv_vmnot_m (VReg VState) VReg) 1493(rule (rv_vmnot_m vs vstate) (rv_vmnand_mm vs vs vstate)) 1494 1495;; Helper for emitting the `vmnor.mm` (Mask Bitwise NOR) instruction. 1496;; 1497;; vd.mask[i] = !(vs2.mask[i] || vs1.mask[i]) 1498(decl rv_vmnor_mm (VReg VReg VState) VReg) 1499(rule (rv_vmnor_mm vs2 vs1 vstate) 1500 (vec_alu_rrr (VecAluOpRRR.VmnorMM) vs2 vs1 (unmasked) vstate)) 1501 1502;;;; Multi-Instruction Helpers ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; 1503 1504(decl gen_extractlane (Type VReg u8) Reg) 1505 1506;; When extracting lane 0 for floats, we can use `vfmv.f.s` directly. 1507(rule 3 (gen_extractlane (ty_vec_fits_in_register ty) src 0) 1508 (if (ty_vector_float ty)) 1509 (rv_vfmv_fs src ty)) 1510 1511;; When extracting lane 0 for integers, we can use `vmv.x.s` directly. 1512(rule 2 (gen_extractlane (ty_vec_fits_in_register ty) src 0) 1513 (if (ty_vector_not_float ty)) 1514 (rv_vmv_xs src ty)) 1515 1516;; In the general case, we must first use a `vslidedown` to place the correct lane 1517;; in index 0, and then use the appropriate `vmv` instruction. 1518;; If the index fits into a 5-bit immediate, we can emit a `vslidedown.vi`. 1519(rule 1 (gen_extractlane (ty_vec_fits_in_register ty) src (uimm5_from_u8 idx)) 1520 (gen_extractlane ty (rv_vslidedown_vi src idx (unmasked) ty) 0)) 1521 1522;; Otherwise lower it into an X register. 1523(rule 0 (gen_extractlane (ty_vec_fits_in_register ty) src idx) 1524 (gen_extractlane ty (rv_vslidedown_vx src (imm $I64 idx) (unmasked) ty) 0)) 1525 1526 1527;; Build a vector mask from a u64 1528;; TODO(#6571): We should merge this with the `vconst` rules, and take advantage of 1529;; the other existing `vconst` rules. 1530(decl gen_vec_mask (u64) VReg) 1531 1532;; When the immediate fits in a 5-bit immediate, we can use `vmv.v.i` directly. 1533(rule 1 (gen_vec_mask (imm5_from_u64 imm)) 1534 (rv_vmv_vi imm (vstate_from_type $I64X2))) 1535 1536;; Materialize the mask into an X register, and move it into the bottom of 1537;; the vector register. 1538(rule 0 (gen_vec_mask mask) 1539 (rv_vmv_sx (imm $I64 mask) (vstate_from_type $I64X2))) 1540 1541 1542;; Loads a `VCodeConstant` value into a vector register. For some special `VCodeConstant`s 1543;; we can use a dedicated instruction, otherwise we load the value from the pool. 1544;; 1545;; Type is the preferred type to use when loading the constant. 1546(decl gen_constant (Type VCodeConstant) VReg) 1547 1548;; The fallback case is to load the constant from the pool. 1549(rule (gen_constant ty n) 1550 (vec_load 1551 (element_width_from_type ty) 1552 (VecAMode.UnitStride (gen_const_amode n)) 1553 (mem_flags_trusted) 1554 (unmasked) 1555 ty)) 1556 1557 1558;; Emits a vslidedown instruction that moves half the lanes down. 1559(decl gen_slidedown_half (Type VReg) VReg) 1560 1561;; If the lane count can fit in a 5-bit immediate, we can use `vslidedown.vi`. 1562(rule 1 (gen_slidedown_half (ty_vec_fits_in_register ty) src) 1563 (if-let (uimm5_from_u64 amt) (u64_checked_div (ty_lane_count ty) 2)) 1564 (rv_vslidedown_vi src amt (unmasked) ty)) 1565 1566;; Otherwise lower it into an X register. 1567(rule 0 (gen_slidedown_half (ty_vec_fits_in_register ty) src) 1568 (if-let amt (u64_checked_div (ty_lane_count ty) 2)) 1569 (rv_vslidedown_vx src (imm $I64 amt) (unmasked) ty)) 1570 1571 1572;; Expands a mask into SEW wide lanes. Enabled lanes are set to all ones, disabled 1573;; lanes are set to all zeros. 1574(decl gen_expand_mask (Type VReg) VReg) 1575(rule (gen_expand_mask ty mask) 1576 (if-let zero (i8_to_imm5 0)) 1577 (if-let neg1 (i8_to_imm5 -1)) 1578 (rv_vmerge_vim (rv_vmv_vi zero ty) neg1 mask ty)) 1579 1580 1581;; Builds a vector mask corresponding to the IntCC operation. 1582;; TODO: We are still missing some rules here for immediates. See #6623 1583(decl gen_icmp_mask (Type IntCC Value Value) VReg) 1584 1585;; IntCC.Equal 1586 1587(rule 0 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.Equal) x y) 1588 (rv_vmseq_vv x y (unmasked) ty)) 1589 1590(rule 1 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.Equal) x (splat y)) 1591 (rv_vmseq_vx x y (unmasked) ty)) 1592 1593(rule 2 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.Equal) (splat x) y) 1594 (rv_vmseq_vx y x (unmasked) ty)) 1595 1596(rule 3 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.Equal) x y) 1597 (if-let y_imm (replicated_imm5 y)) 1598 (rv_vmseq_vi x y_imm (unmasked) ty)) 1599 1600(rule 4 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.Equal) x y) 1601 (if-let x_imm (replicated_imm5 x)) 1602 (rv_vmseq_vi y x_imm (unmasked) ty)) 1603 1604;; IntCC.NotEqual 1605 1606(rule 0 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.NotEqual) x y) 1607 (rv_vmsne_vv x y (unmasked) ty)) 1608 1609(rule 1 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.NotEqual) x (splat y)) 1610 (rv_vmsne_vx x y (unmasked) ty)) 1611 1612(rule 2 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.NotEqual) (splat x) y) 1613 (rv_vmsne_vx y x (unmasked) ty)) 1614 1615(rule 3 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.NotEqual) x y) 1616 (if-let y_imm (replicated_imm5 y)) 1617 (rv_vmsne_vi x y_imm (unmasked) ty)) 1618 1619(rule 4 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.NotEqual) x y) 1620 (if-let x_imm (replicated_imm5 x)) 1621 (rv_vmsne_vi y x_imm (unmasked) ty)) 1622 1623;; IntCC.UnsignedLessThan 1624 1625(rule 0 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedLessThan) x y) 1626 (rv_vmsltu_vv x y (unmasked) ty)) 1627 1628(rule 1 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedLessThan) x (splat y)) 1629 (rv_vmsltu_vx x y (unmasked) ty)) 1630 1631(rule 2 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedLessThan) (splat x) y) 1632 (rv_vmsgtu_vx y x (unmasked) ty)) 1633 1634(rule 4 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedLessThan) x y) 1635 (if-let x_imm (replicated_imm5 x)) 1636 (rv_vmsgtu_vi y x_imm (unmasked) ty)) 1637 1638;; IntCC.SignedLessThan 1639 1640(rule 0 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedLessThan) x y) 1641 (rv_vmslt_vv x y (unmasked) ty)) 1642 1643(rule 1 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedLessThan) x (splat y)) 1644 (rv_vmslt_vx x y (unmasked) ty)) 1645 1646(rule 2 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedLessThan) (splat x) y) 1647 (rv_vmsgt_vx y x (unmasked) ty)) 1648 1649(rule 4 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedLessThan) x y) 1650 (if-let x_imm (replicated_imm5 x)) 1651 (rv_vmsgt_vi y x_imm (unmasked) ty)) 1652 1653;; IntCC.UnsignedLessThanOrEqual 1654 1655(rule 0 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedLessThanOrEqual) x y) 1656 (rv_vmsleu_vv x y (unmasked) ty)) 1657 1658(rule 1 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedLessThanOrEqual) x (splat y)) 1659 (rv_vmsleu_vx x y (unmasked) ty)) 1660 1661(rule 3 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedLessThanOrEqual) x y) 1662 (if-let y_imm (replicated_imm5 y)) 1663 (rv_vmsleu_vi x y_imm (unmasked) ty)) 1664 1665;; IntCC.SignedLessThanOrEqual 1666 1667(rule 0 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedLessThanOrEqual) x y) 1668 (rv_vmsle_vv x y (unmasked) ty)) 1669 1670(rule 1 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedLessThanOrEqual) x (splat y)) 1671 (rv_vmsle_vx x y (unmasked) ty)) 1672 1673(rule 3 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedLessThanOrEqual) x y) 1674 (if-let y_imm (replicated_imm5 y)) 1675 (rv_vmsle_vi x y_imm (unmasked) ty)) 1676 1677;; IntCC.UnsignedGreaterThan 1678 1679(rule 0 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedGreaterThan) x y) 1680 (rv_vmsgtu_vv x y (unmasked) ty)) 1681 1682(rule 1 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedGreaterThan) x (splat y)) 1683 (rv_vmsgtu_vx x y (unmasked) ty)) 1684 1685(rule 2 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedGreaterThan) (splat x) y) 1686 (rv_vmsltu_vx y x (unmasked) ty)) 1687 1688(rule 3 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedGreaterThan) x y) 1689 (if-let y_imm (replicated_imm5 y)) 1690 (rv_vmsgtu_vi x y_imm (unmasked) ty)) 1691 1692;; IntCC.SignedGreaterThan 1693 1694(rule 0 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedGreaterThan) x y) 1695 (rv_vmsgt_vv x y (unmasked) ty)) 1696 1697(rule 1 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedGreaterThan) x (splat y)) 1698 (rv_vmsgt_vx x y (unmasked) ty)) 1699 1700(rule 2 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedGreaterThan) (splat x) y) 1701 (rv_vmslt_vx y x (unmasked) ty)) 1702 1703(rule 3 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedGreaterThan) x y) 1704 (if-let y_imm (replicated_imm5 y)) 1705 (rv_vmsgt_vi x y_imm (unmasked) ty)) 1706 1707;; IntCC.UnsignedGreaterThanOrEqual 1708 1709(rule 0 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedGreaterThanOrEqual) x y) 1710 (rv_vmsgeu_vv x y (unmasked) ty)) 1711 1712(rule 2 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedGreaterThanOrEqual) (splat x) y) 1713 (rv_vmsleu_vx y x (unmasked) ty)) 1714 1715(rule 4 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.UnsignedGreaterThanOrEqual) x y) 1716 (if-let x_imm (replicated_imm5 x)) 1717 (rv_vmsleu_vi y x_imm (unmasked) ty)) 1718 1719;; IntCC.SignedGreaterThanOrEqual 1720 1721(rule 0 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedGreaterThanOrEqual) x y) 1722 (rv_vmsge_vv x y (unmasked) ty)) 1723 1724(rule 2 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedGreaterThanOrEqual) (splat x) y) 1725 (rv_vmsle_vx y x (unmasked) ty)) 1726 1727(rule 4 (gen_icmp_mask (ty_vec_fits_in_register ty) (IntCC.SignedGreaterThanOrEqual) x y) 1728 (if-let x_imm (replicated_imm5 x)) 1729 (rv_vmsle_vi y x_imm (unmasked) ty)) 1730 1731 1732 1733;; Builds a vector mask corresponding to the FloatCC operation. 1734(decl gen_fcmp_mask (Type FloatCC Value Value) VReg) 1735 1736;; FloatCC.Equal 1737 1738(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.Equal) x y) 1739 (rv_vmfeq_vv x y (unmasked) ty)) 1740 1741(rule 1 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.Equal) x (splat y)) 1742 (rv_vmfeq_vf x y (unmasked) ty)) 1743 1744(rule 2 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.Equal) (splat x) y) 1745 (rv_vmfeq_vf y x (unmasked) ty)) 1746 1747;; FloatCC.NotEqual 1748;; Note: This is UnorderedNotEqual. It is the only unordered comparison that is not named as such. 1749 1750(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.NotEqual) x y) 1751 (rv_vmfne_vv x y (unmasked) ty)) 1752 1753(rule 1 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.NotEqual) x (splat y)) 1754 (rv_vmfne_vf x y (unmasked) ty)) 1755 1756(rule 2 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.NotEqual) (splat x) y) 1757 (rv_vmfne_vf y x (unmasked) ty)) 1758 1759;; FloatCC.LessThan 1760 1761(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.LessThan) x y) 1762 (rv_vmflt_vv x y (unmasked) ty)) 1763 1764(rule 1 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.LessThan) x (splat y)) 1765 (rv_vmflt_vf x y (unmasked) ty)) 1766 1767(rule 2 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.LessThan) (splat x) y) 1768 (rv_vmfgt_vf y x (unmasked) ty)) 1769 1770;; FloatCC.LessThanOrEqual 1771 1772(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.LessThanOrEqual) x y) 1773 (rv_vmfle_vv x y (unmasked) ty)) 1774 1775(rule 1 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.LessThanOrEqual) x (splat y)) 1776 (rv_vmfle_vf x y (unmasked) ty)) 1777 1778(rule 2 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.LessThanOrEqual) (splat x) y) 1779 (rv_vmfge_vf y x (unmasked) ty)) 1780 1781;; FloatCC.GreaterThan 1782 1783(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.GreaterThan) x y) 1784 (rv_vmfgt_vv x y (unmasked) ty)) 1785 1786(rule 1 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.GreaterThan) x (splat y)) 1787 (rv_vmfgt_vf x y (unmasked) ty)) 1788 1789(rule 2 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.GreaterThan) (splat x) y) 1790 (rv_vmflt_vf y x (unmasked) ty)) 1791 1792;; FloatCC.GreaterThanOrEqual 1793 1794(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.GreaterThanOrEqual) x y) 1795 (rv_vmfge_vv x y (unmasked) ty)) 1796 1797(rule 1 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.GreaterThanOrEqual) x (splat y)) 1798 (rv_vmfge_vf x y (unmasked) ty)) 1799 1800(rule 2 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.GreaterThanOrEqual) (splat x) y) 1801 (rv_vmfle_vf y x (unmasked) ty)) 1802 1803;; FloatCC.Ordered 1804 1805(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.Ordered) x y) 1806 (rv_vmand_mm 1807 (gen_fcmp_mask ty (FloatCC.Equal) x x) 1808 (gen_fcmp_mask ty (FloatCC.Equal) y y) 1809 ty)) 1810 1811;; FloatCC.Unordered 1812 1813(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.Unordered) x y) 1814 (rv_vmor_mm 1815 (gen_fcmp_mask ty (FloatCC.NotEqual) x x) 1816 (gen_fcmp_mask ty (FloatCC.NotEqual) y y) 1817 ty)) 1818 1819;; FloatCC.OrderedNotEqual 1820 1821(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.OrderedNotEqual) x y) 1822 (rv_vmor_mm 1823 (gen_fcmp_mask ty (FloatCC.LessThan) x y) 1824 (gen_fcmp_mask ty (FloatCC.LessThan) y x) 1825 ty)) 1826 1827;; FloatCC.UnorderedOrEqual 1828 1829(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.UnorderedOrEqual) x y) 1830 (rv_vmnor_mm 1831 (gen_fcmp_mask ty (FloatCC.LessThan) x y) 1832 (gen_fcmp_mask ty (FloatCC.LessThan) y x) 1833 ty)) 1834 1835;; FloatCC.UnorderedOrGreaterThan 1836 1837(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.UnorderedOrGreaterThan) x y) 1838 (rv_vmnot_m (gen_fcmp_mask ty (FloatCC.LessThanOrEqual) x y) ty)) 1839 1840;; FloatCC.UnorderedOrGreaterThanOrEqual 1841 1842(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.UnorderedOrGreaterThanOrEqual) x y) 1843 (rv_vmnot_m (gen_fcmp_mask ty (FloatCC.LessThan) x y) ty)) 1844 1845;; FloatCC.UnorderedOrLessThan 1846 1847(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.UnorderedOrLessThan) x y) 1848 (rv_vmnot_m (gen_fcmp_mask ty (FloatCC.GreaterThanOrEqual) x y) ty)) 1849 1850;; FloatCC.UnorderedOrLessThanOrEqual 1851 1852(rule 0 (gen_fcmp_mask (ty_vec_fits_in_register ty) (FloatCC.UnorderedOrLessThanOrEqual) x y) 1853 (rv_vmnot_m (gen_fcmp_mask ty (FloatCC.GreaterThan) x y) ty)) 1854 1855 1856;; Emits a `vfcvt.x.f.v` instruction with the given rounding mode. 1857(decl gen_vfcvt_x_f (VReg FRM VState) VReg) 1858 1859;; We have a special instruction for RTZ 1860(rule 1 (gen_vfcvt_x_f x (FRM.RTZ) vstate) 1861 (rv_vfcvt_rtz_x_f_v x (unmasked) vstate)) 1862 1863;; In the general case we need to first switch into the appropriate rounding mode. 1864(rule 0 (gen_vfcvt_x_f x frm vstate) 1865 (let ( 1866 ;; Set the rounding mode and save the current mode 1867 (saved_frm XReg (rv_fsrmi frm)) 1868 (res VReg (rv_vfcvt_x_f_v x (unmasked) vstate)) 1869 ;; Restore the previous rounding mode 1870 (_ Unit (rv_fsrm saved_frm))) 1871 res)) 1872 1873 1874;; Returns the maximum value integer value that can be represented by a float 1875(decl float_int_max (Type) u64) 1876(rule (float_int_max $F32) 0x4B000000) 1877(rule (float_int_max $F64) 0x4330000000000000) 1878 1879;; Builds the instruction sequence to round a vector register to FRM 1880(decl gen_vec_round (VReg FRM Type) VReg) 1881 1882;; For floating-point round operations, if the input is NaN, +/-infinity, or +/-0, the 1883;; same input is returned as the rounded result; this differs from behavior of 1884;; RISCV fcvt instructions (which round out-of-range values to the nearest 1885;; max or min value), therefore special handling is needed for these values. 1886(rule (gen_vec_round x frm (ty_vec_fits_in_register ty)) 1887 (let ((scalar_ty Type (lane_type ty)) 1888 ;; if x is NaN/+-Infinity/+-Zero or if the exponent is larger than # of bits 1889 ;; in mantissa, the result is the same as src, build a mask for those cases. 1890 ;; (There is an additional fixup for NaN's at the end) 1891 (abs VReg (rv_vfabs_v x (unmasked) ty)) 1892 (max FReg (imm scalar_ty (float_int_max scalar_ty))) 1893 (exact VReg (rv_vmflt_vf abs max (unmasked) ty)) 1894 1895 ;; The rounding is performed by converting from float to integer, with the 1896 ;; desired rounding mode. And then converting back with the default rounding 1897 ;; mode. 1898 (int VReg (gen_vfcvt_x_f x frm ty)) 1899 (cvt VReg (rv_vfcvt_f_x_v int (unmasked) ty)) 1900 ;; Copy the sign bit from the original value. 1901 (signed VReg (rv_vfsgnj_vv cvt x (unmasked) ty)) 1902 1903 ;; We want to return a arithmetic nan if the input is a canonical nan. 1904 ;; Convert them by adding 0.0 to the input. 1905 (float_zero FReg (gen_bitcast (zero_reg) (float_int_of_same_size scalar_ty) scalar_ty)) 1906 (corrected_nan VReg (rv_vfadd_vf x float_zero (unmasked) ty))) 1907 ;; Merge the original value if it does not need rounding, or the rounded value 1908 (rv_vmerge_vvm corrected_nan signed exact ty))) 1909