1 //==- HexagonInstrInfo.td - Target Description for Hexagon -*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the Hexagon instructions in TableGen format.
12 //===----------------------------------------------------------------------===//
14 include "HexagonInstrFormats.td"
15 include "HexagonOperands.td"
17 // Pattern fragment that combines the value type and the register class
18 // into a single parameter.
19 // The pat frags in the definitions below need to have a named register,
20 // otherwise i32 will be assumed regardless of the register class. The
21 // name of the register does not matter.
22 def I1 : PatLeaf<(i1 PredRegs:$R)>;
23 def I32 : PatLeaf<(i32 IntRegs:$R)>;
24 def I64 : PatLeaf<(i64 DoubleRegs:$R)>;
25 def F32 : PatLeaf<(f32 IntRegs:$R)>;
26 def F64 : PatLeaf<(f64 DoubleRegs:$R)>;
28 // Pattern fragments to extract the low and high subregisters from a
30 def LoReg: OutPatFrag<(ops node:$Rs),
31 (EXTRACT_SUBREG (i64 $Rs), subreg_loreg)>;
33 //===----------------------------------------------------------------------===//
35 //===----------------------------------------------------------------------===//
37 //===----------------------------------------------------------------------===//
38 let hasSideEffects = 0, isCompare = 1, InputType = "imm", isExtendable = 1,
40 class T_CMP <string mnemonic, bits<2> MajOp, bit isNot, Operand ImmOp>
41 : ALU32Inst <(outs PredRegs:$dst),
42 (ins IntRegs:$src1, ImmOp:$src2),
43 "$dst = "#!if(isNot, "!","")#mnemonic#"($src1, #$src2)",
44 [], "",ALU32_2op_tc_2early_SLOT0123 >, ImmRegRel {
48 let CextOpcode = mnemonic;
49 let opExtentBits = !if(!eq(mnemonic, "cmp.gtu"), 9, 10);
50 let isExtentSigned = !if(!eq(mnemonic, "cmp.gtu"), 0, 1);
54 let Inst{27-24} = 0b0101;
55 let Inst{23-22} = MajOp;
56 let Inst{21} = !if(!eq(mnemonic, "cmp.gtu"), 0, src2{9});
57 let Inst{20-16} = src1;
58 let Inst{13-5} = src2{8-0};
64 def C2_cmpeqi : T_CMP <"cmp.eq", 0b00, 0, s10Ext>;
65 def C2_cmpgti : T_CMP <"cmp.gt", 0b01, 0, s10Ext>;
66 def C2_cmpgtui : T_CMP <"cmp.gtu", 0b10, 0, u9Ext>;
68 class T_CMP_pat <InstHexagon MI, PatFrag OpNode, PatLeaf ImmPred>
69 : Pat<(i1 (OpNode (i32 IntRegs:$src1), ImmPred:$src2)),
70 (MI IntRegs:$src1, ImmPred:$src2)>;
72 def : T_CMP_pat <C2_cmpeqi, seteq, s10ImmPred>;
73 def : T_CMP_pat <C2_cmpgti, setgt, s10ImmPred>;
74 def : T_CMP_pat <C2_cmpgtui, setugt, u9ImmPred>;
76 //===----------------------------------------------------------------------===//
78 //===----------------------------------------------------------------------===//
79 def SDTHexagonI64I32I32 : SDTypeProfile<1, 2,
80 [SDTCisVT<0, i64>, SDTCisVT<1, i32>, SDTCisSameAs<1, 2>]>;
82 def HexagonCOMBINE : SDNode<"HexagonISD::COMBINE", SDTHexagonI64I32I32>;
84 let hasSideEffects = 0, hasNewValue = 1, InputType = "reg" in
85 class T_ALU32_3op<string mnemonic, bits<3> MajOp, bits<3> MinOp, bit OpsRev,
87 : ALU32_rr<(outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
88 "$Rd = "#mnemonic#"($Rs, $Rt)",
89 [], "", ALU32_3op_tc_1_SLOT0123>, ImmRegRel, PredRel {
90 let isCommutable = IsComm;
91 let BaseOpcode = mnemonic#_rr;
92 let CextOpcode = mnemonic;
100 let Inst{26-24} = MajOp;
101 let Inst{23-21} = MinOp;
102 let Inst{20-16} = !if(OpsRev,Rt,Rs);
103 let Inst{12-8} = !if(OpsRev,Rs,Rt);
107 let hasSideEffects = 0, hasNewValue = 1 in
108 class T_ALU32_3op_pred<string mnemonic, bits<3> MajOp, bits<3> MinOp,
109 bit OpsRev, bit PredNot, bit PredNew>
110 : ALU32_rr<(outs IntRegs:$Rd), (ins PredRegs:$Pu, IntRegs:$Rs, IntRegs:$Rt),
111 "if ("#!if(PredNot,"!","")#"$Pu"#!if(PredNew,".new","")#") "#
112 "$Rd = "#mnemonic#"($Rs, $Rt)",
113 [], "", ALU32_3op_tc_1_SLOT0123>, ImmRegRel, PredNewRel {
114 let isPredicated = 1;
115 let isPredicatedFalse = PredNot;
116 let isPredicatedNew = PredNew;
117 let BaseOpcode = mnemonic#_rr;
118 let CextOpcode = mnemonic;
127 let Inst{26-24} = MajOp;
128 let Inst{23-21} = MinOp;
129 let Inst{20-16} = !if(OpsRev,Rt,Rs);
130 let Inst{13} = PredNew;
131 let Inst{12-8} = !if(OpsRev,Rs,Rt);
132 let Inst{7} = PredNot;
137 class T_ALU32_combineh<string Op1, string Op2, bits<3> MajOp, bits<3> MinOp,
139 : T_ALU32_3op<"", MajOp, MinOp, OpsRev, 0> {
140 let AsmString = "$Rd = combine($Rs"#Op1#", $Rt"#Op2#")";
143 let isCodeGenOnly = 0 in {
144 def A2_combine_hh : T_ALU32_combineh<".h", ".h", 0b011, 0b100, 1>;
145 def A2_combine_hl : T_ALU32_combineh<".h", ".l", 0b011, 0b101, 1>;
146 def A2_combine_lh : T_ALU32_combineh<".l", ".h", 0b011, 0b110, 1>;
147 def A2_combine_ll : T_ALU32_combineh<".l", ".l", 0b011, 0b111, 1>;
150 class T_ALU32_3op_sfx<string mnemonic, string suffix, bits<3> MajOp,
151 bits<3> MinOp, bit OpsRev, bit IsComm>
152 : T_ALU32_3op<"", MajOp, MinOp, OpsRev, IsComm> {
153 let AsmString = "$Rd = "#mnemonic#"($Rs, $Rt)"#suffix;
156 let Defs = [USR_OVF], Itinerary = ALU32_3op_tc_2_SLOT0123,
157 isCodeGenOnly = 0 in {
158 def A2_addsat : T_ALU32_3op_sfx<"add", ":sat", 0b110, 0b010, 0, 1>;
159 def A2_subsat : T_ALU32_3op_sfx<"sub", ":sat", 0b110, 0b110, 1, 0>;
162 multiclass T_ALU32_3op_p<string mnemonic, bits<3> MajOp, bits<3> MinOp,
164 def t : T_ALU32_3op_pred<mnemonic, MajOp, MinOp, OpsRev, 0, 0>;
165 def f : T_ALU32_3op_pred<mnemonic, MajOp, MinOp, OpsRev, 1, 0>;
166 def tnew : T_ALU32_3op_pred<mnemonic, MajOp, MinOp, OpsRev, 0, 1>;
167 def fnew : T_ALU32_3op_pred<mnemonic, MajOp, MinOp, OpsRev, 1, 1>;
170 multiclass T_ALU32_3op_A2<string mnemonic, bits<3> MajOp, bits<3> MinOp,
171 bit OpsRev, bit IsComm> {
172 let isPredicable = 1 in
173 def A2_#NAME : T_ALU32_3op <mnemonic, MajOp, MinOp, OpsRev, IsComm>;
174 defm A2_p#NAME : T_ALU32_3op_p<mnemonic, MajOp, MinOp, OpsRev>;
177 let isCodeGenOnly = 0 in {
178 defm add : T_ALU32_3op_A2<"add", 0b011, 0b000, 0, 1>;
179 defm and : T_ALU32_3op_A2<"and", 0b001, 0b000, 0, 1>;
180 defm or : T_ALU32_3op_A2<"or", 0b001, 0b001, 0, 1>;
181 defm sub : T_ALU32_3op_A2<"sub", 0b011, 0b001, 1, 0>;
182 defm xor : T_ALU32_3op_A2<"xor", 0b001, 0b011, 0, 1>;
185 // Pats for instruction selection.
186 class BinOp32_pat<SDNode Op, InstHexagon MI, ValueType ResT>
187 : Pat<(ResT (Op (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))),
188 (ResT (MI IntRegs:$Rs, IntRegs:$Rt))>;
190 def: BinOp32_pat<add, A2_add, i32>;
191 def: BinOp32_pat<and, A2_and, i32>;
192 def: BinOp32_pat<or, A2_or, i32>;
193 def: BinOp32_pat<sub, A2_sub, i32>;
194 def: BinOp32_pat<xor, A2_xor, i32>;
196 // A few special cases producing register pairs:
197 let OutOperandList = (outs DoubleRegs:$Rd), hasNewValue = 0,
198 isCodeGenOnly = 0 in {
199 def S2_packhl : T_ALU32_3op <"packhl", 0b101, 0b100, 0, 0>;
201 let isPredicable = 1 in
202 def A2_combinew : T_ALU32_3op <"combine", 0b101, 0b000, 0, 0>;
204 // Conditional combinew uses "newt/f" instead of "t/fnew".
205 def C2_ccombinewt : T_ALU32_3op_pred<"combine", 0b101, 0b000, 0, 0, 0>;
206 def C2_ccombinewf : T_ALU32_3op_pred<"combine", 0b101, 0b000, 0, 1, 0>;
207 def C2_ccombinewnewt : T_ALU32_3op_pred<"combine", 0b101, 0b000, 0, 0, 1>;
208 def C2_ccombinewnewf : T_ALU32_3op_pred<"combine", 0b101, 0b000, 0, 1, 1>;
211 let hasSideEffects = 0, hasNewValue = 1, isCompare = 1, InputType = "reg" in
212 class T_ALU32_3op_cmp<string mnemonic, bits<2> MinOp, bit IsNeg, bit IsComm>
213 : ALU32_rr<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt),
214 "$Pd = "#mnemonic#"($Rs, $Rt)",
215 [], "", ALU32_3op_tc_1_SLOT0123>, ImmRegRel {
216 let CextOpcode = mnemonic;
217 let isCommutable = IsComm;
223 let Inst{27-24} = 0b0010;
224 let Inst{22-21} = MinOp;
225 let Inst{20-16} = Rs;
228 let Inst{3-2} = 0b00;
232 let Itinerary = ALU32_3op_tc_2early_SLOT0123, isCodeGenOnly = 0 in {
233 def C2_cmpeq : T_ALU32_3op_cmp< "cmp.eq", 0b00, 0, 1>;
234 def C2_cmpgt : T_ALU32_3op_cmp< "cmp.gt", 0b10, 0, 0>;
235 def C2_cmpgtu : T_ALU32_3op_cmp< "cmp.gtu", 0b11, 0, 0>;
238 // Patfrag to convert the usual comparison patfrags (e.g. setlt) to ones
239 // that reverse the order of the operands.
240 class RevCmp<PatFrag F> : PatFrag<(ops node:$rhs, node:$lhs), F.Fragment>;
242 // Pats for compares. They use PatFrags as operands, not SDNodes,
243 // since seteq/setgt/etc. are defined as ParFrags.
244 class T_cmp32_rr_pat<InstHexagon MI, PatFrag Op, ValueType VT>
245 : Pat<(VT (Op (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))),
246 (VT (MI IntRegs:$Rs, IntRegs:$Rt))>;
248 def: T_cmp32_rr_pat<C2_cmpeq, seteq, i1>;
249 def: T_cmp32_rr_pat<C2_cmpgt, setgt, i1>;
250 def: T_cmp32_rr_pat<C2_cmpgtu, setugt, i1>;
252 def: T_cmp32_rr_pat<C2_cmpgt, RevCmp<setlt>, i1>;
253 def: T_cmp32_rr_pat<C2_cmpgtu, RevCmp<setult>, i1>;
255 let CextOpcode = "MUX", InputType = "reg", hasNewValue = 1,
257 def C2_mux: ALU32_rr<(outs IntRegs:$Rd),
258 (ins PredRegs:$Pu, IntRegs:$Rs, IntRegs:$Rt),
259 "$Rd = mux($Pu, $Rs, $Rt)", [], "", ALU32_3op_tc_1_SLOT0123>, ImmRegRel {
265 let CextOpcode = "mux";
266 let InputType = "reg";
267 let hasSideEffects = 0;
270 let Inst{27-24} = 0b0100;
271 let Inst{20-16} = Rs;
277 def: Pat<(i32 (select (i1 PredRegs:$Pu), (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))),
278 (C2_mux PredRegs:$Pu, IntRegs:$Rs, IntRegs:$Rt)>;
280 // Combines the two immediates into a double register.
281 // Increase complexity to make it greater than any complexity of a combine
282 // that involves a register.
284 let isReMaterializable = 1, isMoveImm = 1, isAsCheapAsAMove = 1,
285 isExtentSigned = 1, isExtendable = 1, opExtentBits = 8, opExtendable = 1,
286 AddedComplexity = 75, isCodeGenOnly = 0 in
287 def A2_combineii: ALU32Inst <(outs DoubleRegs:$Rdd), (ins s8Ext:$s8, s8Imm:$S8),
288 "$Rdd = combine(#$s8, #$S8)",
289 [(set (i64 DoubleRegs:$Rdd),
290 (i64 (HexagonCOMBINE(i32 s8ExtPred:$s8), (i32 s8ImmPred:$S8))))]> {
296 let Inst{27-23} = 0b11000;
297 let Inst{22-16} = S8{7-1};
298 let Inst{13} = S8{0};
303 //===----------------------------------------------------------------------===//
304 // Template class for predicated ADD of a reg and an Immediate value.
305 //===----------------------------------------------------------------------===//
306 let hasNewValue = 1 in
307 class T_Addri_Pred <bit PredNot, bit PredNew>
308 : ALU32_ri <(outs IntRegs:$Rd),
309 (ins PredRegs:$Pu, IntRegs:$Rs, s8Ext:$s8),
310 !if(PredNot, "if (!$Pu", "if ($Pu")#!if(PredNew,".new) $Rd = ",
311 ") $Rd = ")#"add($Rs, #$s8)"> {
317 let isPredicatedNew = PredNew;
320 let Inst{27-24} = 0b0100;
321 let Inst{23} = PredNot;
322 let Inst{22-21} = Pu;
323 let Inst{20-16} = Rs;
324 let Inst{13} = PredNew;
329 //===----------------------------------------------------------------------===//
330 // A2_addi: Add a signed immediate to a register.
331 //===----------------------------------------------------------------------===//
332 let hasNewValue = 1 in
333 class T_Addri <Operand immOp, list<dag> pattern = [] >
334 : ALU32_ri <(outs IntRegs:$Rd),
335 (ins IntRegs:$Rs, immOp:$s16),
336 "$Rd = add($Rs, #$s16)", pattern,
337 //[(set (i32 IntRegs:$Rd), (add (i32 IntRegs:$Rs), (s16ExtPred:$s16)))],
338 "", ALU32_ADDI_tc_1_SLOT0123> {
345 let Inst{27-21} = s16{15-9};
346 let Inst{20-16} = Rs;
347 let Inst{13-5} = s16{8-0};
351 //===----------------------------------------------------------------------===//
352 // Multiclass for ADD of a register and an immediate value.
353 //===----------------------------------------------------------------------===//
354 multiclass Addri_Pred<string mnemonic, bit PredNot> {
355 let isPredicatedFalse = PredNot in {
356 def _c#NAME : T_Addri_Pred<PredNot, 0>;
358 def _cdn#NAME : T_Addri_Pred<PredNot, 1>;
362 let isExtendable = 1, InputType = "imm" in
363 multiclass Addri_base<string mnemonic, SDNode OpNode> {
364 let CextOpcode = mnemonic, BaseOpcode = mnemonic#_ri in {
365 let opExtendable = 2, isExtentSigned = 1, opExtentBits = 16,
367 def NAME : T_Addri< s16Ext, // Rd=add(Rs,#s16)
368 [(set (i32 IntRegs:$Rd),
369 (add IntRegs:$Rs, s16ExtPred:$s16))]>;
371 let opExtendable = 3, isExtentSigned = 1, opExtentBits = 8,
372 hasSideEffects = 0, isPredicated = 1 in {
373 defm Pt : Addri_Pred<mnemonic, 0>;
374 defm NotPt : Addri_Pred<mnemonic, 1>;
379 let isCodeGenOnly = 0 in
380 defm ADD_ri : Addri_base<"add", add>, ImmRegRel, PredNewRel;
382 //===----------------------------------------------------------------------===//
383 // Template class used for the following ALU32 instructions.
386 //===----------------------------------------------------------------------===//
387 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 10,
388 InputType = "imm", hasNewValue = 1 in
389 class T_ALU32ri_logical <string mnemonic, SDNode OpNode, bits<2> MinOp>
390 : ALU32_ri <(outs IntRegs:$Rd),
391 (ins IntRegs:$Rs, s10Ext:$s10),
392 "$Rd = "#mnemonic#"($Rs, #$s10)" ,
393 [(set (i32 IntRegs:$Rd), (OpNode (i32 IntRegs:$Rs), s10ExtPred:$s10))]> {
397 let CextOpcode = mnemonic;
401 let Inst{27-24} = 0b0110;
402 let Inst{23-22} = MinOp;
403 let Inst{21} = s10{9};
404 let Inst{20-16} = Rs;
405 let Inst{13-5} = s10{8-0};
409 let isCodeGenOnly = 0 in {
410 def OR_ri : T_ALU32ri_logical<"or", or, 0b10>, ImmRegRel;
411 def AND_ri : T_ALU32ri_logical<"and", and, 0b00>, ImmRegRel;
414 // Subtract register from immediate
415 // Rd32=sub(#s10,Rs32)
416 let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 10,
417 CextOpcode = "sub", InputType = "imm", hasNewValue = 1, isCodeGenOnly = 0 in
418 def SUB_ri: ALU32_ri <(outs IntRegs:$Rd), (ins s10Ext:$s10, IntRegs:$Rs),
419 "$Rd = sub(#$s10, $Rs)" ,
420 [(set IntRegs:$Rd, (sub s10ExtPred:$s10, IntRegs:$Rs))] > ,
428 let Inst{27-22} = 0b011001;
429 let Inst{21} = s10{9};
430 let Inst{20-16} = Rs;
431 let Inst{13-5} = s10{8-0};
436 let hasSideEffects = 0, isCodeGenOnly = 0 in
437 def A2_nop: ALU32Inst <(outs), (ins), "nop" > {
439 let Inst{27-24} = 0b1111;
441 // Rd = not(Rs) gets mapped to Rd=sub(#-1, Rs).
442 def : Pat<(not (i32 IntRegs:$src1)),
443 (SUB_ri -1, (i32 IntRegs:$src1))>;
445 let hasSideEffects = 0, hasNewValue = 1 in
446 class T_tfr16<bit isHi>
447 : ALU32Inst <(outs IntRegs:$Rx), (ins IntRegs:$src1, u16Imm:$u16),
448 "$Rx"#!if(isHi, ".h", ".l")#" = #$u16",
449 [], "$src1 = $Rx" > {
454 let Inst{27-26} = 0b00;
455 let Inst{25-24} = !if(isHi, 0b10, 0b01);
456 let Inst{23-22} = u16{15-14};
458 let Inst{20-16} = Rx;
459 let Inst{13-0} = u16{13-0};
462 let isCodeGenOnly = 0 in {
463 def A2_tfril: T_tfr16<0>;
464 def A2_tfrih: T_tfr16<1>;
467 // Conditional transfer is an alias to conditional "Rd = add(Rs, #0)".
468 let isPredicated = 1, hasNewValue = 1, opNewValue = 0 in
469 class T_tfr_pred<bit isPredNot, bit isPredNew>
470 : ALU32Inst<(outs IntRegs:$dst),
471 (ins PredRegs:$src1, IntRegs:$src2),
472 "if ("#!if(isPredNot, "!", "")#
473 "$src1"#!if(isPredNew, ".new", "")#
479 let isPredicatedFalse = isPredNot;
480 let isPredicatedNew = isPredNew;
483 let Inst{27-24} = 0b0100;
484 let Inst{23} = isPredNot;
485 let Inst{13} = isPredNew;
488 let Inst{22-21} = src1;
489 let Inst{20-16} = src2;
492 let isPredicable = 1 in
493 class T_tfr : ALU32Inst<(outs IntRegs:$dst), (ins IntRegs:$src),
500 let Inst{27-21} = 0b0000011;
501 let Inst{20-16} = src;
506 let InputType = "reg", hasNewValue = 1, hasSideEffects = 0 in
507 multiclass tfr_base<string CextOp> {
508 let CextOpcode = CextOp, BaseOpcode = CextOp in {
512 def t : T_tfr_pred<0, 0>;
513 def f : T_tfr_pred<1, 0>;
515 def tnew : T_tfr_pred<0, 1>;
516 def fnew : T_tfr_pred<1, 1>;
520 // Assembler mapped to C2_ccombinew[t|f|newt|newf].
521 // Please don't add bits to this instruction as it'll be converted into
522 // 'combine' before object code emission.
523 let isPredicated = 1 in
524 class T_tfrp_pred<bit PredNot, bit PredNew>
525 : ALU32_rr <(outs DoubleRegs:$dst),
526 (ins PredRegs:$src1, DoubleRegs:$src2),
527 "if ("#!if(PredNot, "!", "")#"$src1"
528 #!if(PredNew, ".new", "")#") $dst = $src2" > {
529 let isPredicatedFalse = PredNot;
530 let isPredicatedNew = PredNew;
533 // Assembler mapped to A2_combinew.
534 // Please don't add bits to this instruction as it'll be converted into
535 // 'combine' before object code emission.
536 class T_tfrp : ALU32Inst <(outs DoubleRegs:$dst),
537 (ins DoubleRegs:$src),
540 let hasSideEffects = 0 in
541 multiclass TFR64_base<string BaseName> {
542 let BaseOpcode = BaseName in {
543 let isPredicable = 1 in
546 def t : T_tfrp_pred <0, 0>;
547 def f : T_tfrp_pred <1, 0>;
549 def tnew : T_tfrp_pred <0, 1>;
550 def fnew : T_tfrp_pred <1, 1>;
554 let InputType = "imm", isExtendable = 1, isExtentSigned = 1, opExtentBits = 12,
555 isMoveImm = 1, opExtendable = 2, BaseOpcode = "TFRI", CextOpcode = "TFR",
556 hasSideEffects = 0, isPredicated = 1, hasNewValue = 1 in
557 class T_TFRI_Pred<bit PredNot, bit PredNew>
558 : ALU32_ri<(outs IntRegs:$Rd), (ins PredRegs:$Pu, s12Ext:$s12),
559 "if ("#!if(PredNot,"!","")#"$Pu"#!if(PredNew,".new","")#") $Rd = #$s12",
560 [], "", ALU32_2op_tc_1_SLOT0123>, ImmRegRel, PredNewRel {
561 let isPredicatedFalse = PredNot;
562 let isPredicatedNew = PredNew;
569 let Inst{27-24} = 0b1110;
570 let Inst{23} = PredNot;
571 let Inst{22-21} = Pu;
573 let Inst{19-16,12-5} = s12;
574 let Inst{13} = PredNew;
578 let isCodeGenOnly = 0 in {
579 def C2_cmoveit : T_TFRI_Pred<0, 0>;
580 def C2_cmoveif : T_TFRI_Pred<1, 0>;
581 def C2_cmovenewit : T_TFRI_Pred<0, 1>;
582 def C2_cmovenewif : T_TFRI_Pred<1, 1>;
585 let InputType = "imm", isExtendable = 1, isExtentSigned = 1,
586 CextOpcode = "TFR", BaseOpcode = "TFRI", hasNewValue = 1, opNewValue = 0,
587 isAsCheapAsAMove = 1 , opExtendable = 1, opExtentBits = 16, isMoveImm = 1,
588 isPredicated = 0, isPredicable = 1, isReMaterializable = 1,
590 def A2_tfrsi : ALU32Inst<(outs IntRegs:$Rd), (ins s16Ext:$s16), "$Rd = #$s16",
591 [(set (i32 IntRegs:$Rd), s16ExtPred:$s16)], "", ALU32_2op_tc_1_SLOT0123>,
597 let Inst{27-24} = 0b1000;
598 let Inst{23-22,20-16,13-5} = s16;
602 let isCodeGenOnly = 0 in
603 defm A2_tfr : tfr_base<"TFR">, ImmRegRel, PredNewRel;
604 defm A2_tfrp : TFR64_base<"TFR64">, PredNewRel;
607 let isReMaterializable = 1, isMoveImm = 1, isAsCheapAsAMove = 1 in
608 def A2_tfrpi : ALU64_rr<(outs DoubleRegs:$dst), (ins s8Imm64:$src1),
610 [(set (i64 DoubleRegs:$dst), s8Imm64Pred:$src1)]>;
612 // TODO: see if this instruction can be deleted..
613 let isExtendable = 1, opExtendable = 1, opExtentBits = 6 in
614 def TFRI64_V4 : ALU64_rr<(outs DoubleRegs:$dst), (ins u6Ext:$src1),
617 //===----------------------------------------------------------------------===//
619 //===----------------------------------------------------------------------===//
622 //===----------------------------------------------------------------------===//
624 //===----------------------------------------------------------------------===//
625 // Scalar mux register immediate.
626 let hasSideEffects = 0, isExtentSigned = 1, CextOpcode = "MUX",
627 InputType = "imm", hasNewValue = 1, isExtendable = 1, opExtentBits = 8 in
628 class T_MUX1 <bit MajOp, dag ins, string AsmStr>
629 : ALU32Inst <(outs IntRegs:$Rd), ins, AsmStr>, ImmRegRel {
636 let Inst{27-24} = 0b0011;
637 let Inst{23} = MajOp;
638 let Inst{22-21} = Pu;
639 let Inst{20-16} = Rs;
645 let opExtendable = 2, isCodeGenOnly = 0 in
646 def C2_muxri : T_MUX1<0b1, (ins PredRegs:$Pu, s8Ext:$s8, IntRegs:$Rs),
647 "$Rd = mux($Pu, #$s8, $Rs)">;
649 let opExtendable = 3, isCodeGenOnly = 0 in
650 def C2_muxir : T_MUX1<0b0, (ins PredRegs:$Pu, IntRegs:$Rs, s8Ext:$s8),
651 "$Rd = mux($Pu, $Rs, #$s8)">;
653 def : Pat<(i32 (select I1:$Pu, s8ExtPred:$s8, I32:$Rs)),
654 (C2_muxri I1:$Pu, s8ExtPred:$s8, I32:$Rs)>;
656 def : Pat<(i32 (select I1:$Pu, I32:$Rs, s8ExtPred:$s8)),
657 (C2_muxir I1:$Pu, I32:$Rs, s8ExtPred:$s8)>;
659 // C2_muxii: Scalar mux immediates.
660 let isExtentSigned = 1, hasNewValue = 1, isExtendable = 1,
661 opExtentBits = 8, opExtendable = 2, isCodeGenOnly = 0 in
662 def C2_muxii: ALU32Inst <(outs IntRegs:$Rd),
663 (ins PredRegs:$Pu, s8Ext:$s8, s8Imm:$S8),
664 "$Rd = mux($Pu, #$s8, #$S8)" ,
665 [(set (i32 IntRegs:$Rd),
666 (i32 (select I1:$Pu, s8ExtPred:$s8, s8ImmPred:$S8)))] > {
674 let Inst{27-25} = 0b101;
675 let Inst{24-23} = Pu;
676 let Inst{22-16} = S8{7-1};
677 let Inst{13} = S8{0};
682 //===----------------------------------------------------------------------===//
683 // template class for non-predicated alu32_2op instructions
684 // - aslh, asrh, sxtb, sxth, zxth
685 //===----------------------------------------------------------------------===//
686 let hasNewValue = 1, opNewValue = 0 in
687 class T_ALU32_2op <string mnemonic, bits<3> minOp> :
688 ALU32Inst < (outs IntRegs:$Rd), (ins IntRegs:$Rs),
689 "$Rd = "#mnemonic#"($Rs)", [] > {
695 let Inst{27-24} = 0b0000;
696 let Inst{23-21} = minOp;
699 let Inst{20-16} = Rs;
702 //===----------------------------------------------------------------------===//
703 // template class for predicated alu32_2op instructions
704 // - aslh, asrh, sxtb, sxth, zxtb, zxth
705 //===----------------------------------------------------------------------===//
706 let hasSideEffects = 0, validSubTargets = HasV4SubT,
707 hasNewValue = 1, opNewValue = 0 in
708 class T_ALU32_2op_Pred <string mnemonic, bits<3> minOp, bit isPredNot,
710 ALU32Inst <(outs IntRegs:$Rd), (ins PredRegs:$Pu, IntRegs:$Rs),
711 !if(isPredNot, "if (!$Pu", "if ($Pu")
712 #!if(isPredNew, ".new) ",") ")#"$Rd = "#mnemonic#"($Rs)"> {
719 let Inst{27-24} = 0b0000;
720 let Inst{23-21} = minOp;
722 let Inst{11} = isPredNot;
723 let Inst{10} = isPredNew;
726 let Inst{20-16} = Rs;
729 multiclass ALU32_2op_Pred<string mnemonic, bits<3> minOp, bit PredNot> {
730 let isPredicatedFalse = PredNot in {
731 def NAME : T_ALU32_2op_Pred<mnemonic, minOp, PredNot, 0>;
734 let isPredicatedNew = 1 in
735 def NAME#new : T_ALU32_2op_Pred<mnemonic, minOp, PredNot, 1>;
739 multiclass ALU32_2op_base<string mnemonic, bits<3> minOp> {
740 let BaseOpcode = mnemonic in {
741 let isPredicable = 1, hasSideEffects = 0 in
742 def A2_#NAME : T_ALU32_2op<mnemonic, minOp>;
744 let validSubTargets = HasV4SubT, isPredicated = 1, hasSideEffects = 0 in {
745 defm A4_p#NAME#t : ALU32_2op_Pred<mnemonic, minOp, 0>;
746 defm A4_p#NAME#f : ALU32_2op_Pred<mnemonic, minOp, 1>;
751 let isCodeGenOnly = 0 in {
752 defm aslh : ALU32_2op_base<"aslh", 0b000>, PredNewRel;
753 defm asrh : ALU32_2op_base<"asrh", 0b001>, PredNewRel;
754 defm sxtb : ALU32_2op_base<"sxtb", 0b101>, PredNewRel;
755 defm sxth : ALU32_2op_base<"sxth", 0b111>, PredNewRel;
756 defm zxth : ALU32_2op_base<"zxth", 0b110>, PredNewRel;
759 // Rd=zxtb(Rs): assembler mapped to Rd=and(Rs,#255).
760 // Compiler would want to generate 'zxtb' instead of 'and' becuase 'zxtb' has
761 // predicated forms while 'and' doesn't. Since integrated assembler can't
762 // handle 'mapped' instructions, we need to encode 'zxtb' same as 'and' where
763 // immediate operand is set to '255'.
765 let hasNewValue = 1, opNewValue = 0 in
766 class T_ZXTB: ALU32Inst < (outs IntRegs:$Rd), (ins IntRegs:$Rs),
767 "$Rd = zxtb($Rs)", [] > { // Rd = and(Rs,255)
774 let Inst{27-22} = 0b011000;
776 let Inst{20-16} = Rs;
777 let Inst{21} = s10{9};
778 let Inst{13-5} = s10{8-0};
781 //Rd=zxtb(Rs): assembler mapped to "Rd=and(Rs,#255)
782 multiclass ZXTB_base <string mnemonic, bits<3> minOp> {
783 let BaseOpcode = mnemonic in {
784 let isPredicable = 1, hasSideEffects = 0 in
785 def A2_#NAME : T_ZXTB;
787 let validSubTargets = HasV4SubT, isPredicated = 1, hasSideEffects = 0 in {
788 defm A4_p#NAME#t : ALU32_2op_Pred<mnemonic, minOp, 0>;
789 defm A4_p#NAME#f : ALU32_2op_Pred<mnemonic, minOp, 1>;
794 let isCodeGenOnly=0 in
795 defm zxtb : ZXTB_base<"zxtb",0b100>, PredNewRel;
797 def: Pat<(shl I32:$src1, (i32 16)), (A2_aslh I32:$src1)>;
798 def: Pat<(sra I32:$src1, (i32 16)), (A2_asrh I32:$src1)>;
799 def: Pat<(sext_inreg I32:$src1, i8), (A2_sxtb I32:$src1)>;
800 def: Pat<(sext_inreg I32:$src1, i16), (A2_sxth I32:$src1)>;
803 def VMUX_prr64 : ALU64_rr<(outs DoubleRegs:$dst), (ins PredRegs:$src1,
806 "$dst = vmux($src1, $src2, $src3)",
810 //===----------------------------------------------------------------------===//
812 //===----------------------------------------------------------------------===//
815 //===----------------------------------------------------------------------===//
817 //===----------------------------------------------------------------------===//
819 // SDNode for converting immediate C to C-1.
820 def DEC_CONST_SIGNED : SDNodeXForm<imm, [{
821 // Return the byte immediate const-1 as an SDNode.
822 int32_t imm = N->getSExtValue();
823 return XformSToSM1Imm(imm);
826 // SDNode for converting immediate C to C-1.
827 def DEC_CONST_UNSIGNED : SDNodeXForm<imm, [{
828 // Return the byte immediate const-1 as an SDNode.
829 uint32_t imm = N->getZExtValue();
830 return XformUToUM1Imm(imm);
833 def CTLZ64_rr : SInst<(outs IntRegs:$dst), (ins DoubleRegs:$src1),
835 [(set (i32 IntRegs:$dst), (i32 (trunc (ctlz (i64 DoubleRegs:$src1)))))]>;
837 def CTTZ64_rr : SInst<(outs IntRegs:$dst), (ins DoubleRegs:$src1),
839 [(set (i32 IntRegs:$dst), (i32 (trunc (cttz (i64 DoubleRegs:$src1)))))]>;
841 //===----------------------------------------------------------------------===//
843 //===----------------------------------------------------------------------===//
846 //===----------------------------------------------------------------------===//
848 //===----------------------------------------------------------------------===//// Add.
849 //===----------------------------------------------------------------------===//
851 // Add/Subtract halfword
852 // Rd=add(Rt.L,Rs.[HL])[:sat]
853 // Rd=sub(Rt.L,Rs.[HL])[:sat]
854 // Rd=add(Rt.[LH],Rs.[HL])[:sat][:<16]
855 // Rd=sub(Rt.[LH],Rs.[HL])[:sat][:<16]
856 //===----------------------------------------------------------------------===//
858 let hasNewValue = 1, opNewValue = 0 in
859 class T_XTYPE_ADD_SUB <bits<2> LHbits, bit isSat, bit hasShift, bit isSub>
860 : ALU64Inst <(outs IntRegs:$Rd), (ins IntRegs:$Rt, IntRegs:$Rs),
861 "$Rd = "#!if(isSub,"sub","add")#"($Rt."
862 #!if(hasShift, !if(LHbits{1},"h","l"),"l") #", $Rs."
863 #!if(hasShift, !if(LHbits{0},"h)","l)"), !if(LHbits{1},"h)","l)"))
864 #!if(isSat,":sat","")
865 #!if(hasShift,":<<16",""), [], "", ALU64_tc_1_SLOT23> {
871 let Inst{27-23} = 0b01010;
872 let Inst{22} = hasShift;
873 let Inst{21} = isSub;
875 let Inst{6-5} = LHbits;
878 let Inst{20-16} = Rs;
881 //Rd=sub(Rt.L,Rs.[LH])
882 let isCodeGenOnly = 0 in {
883 def A2_subh_l16_ll : T_XTYPE_ADD_SUB <0b00, 0, 0, 1>;
884 def A2_subh_l16_hl : T_XTYPE_ADD_SUB <0b10, 0, 0, 1>;
887 let isCodeGenOnly = 0 in {
888 //Rd=add(Rt.L,Rs.[LH])
889 def A2_addh_l16_ll : T_XTYPE_ADD_SUB <0b00, 0, 0, 0>;
890 def A2_addh_l16_hl : T_XTYPE_ADD_SUB <0b10, 0, 0, 0>;
893 let Itinerary = ALU64_tc_2_SLOT23, Defs = [USR_OVF], isCodeGenOnly = 0 in {
894 //Rd=sub(Rt.L,Rs.[LH]):sat
895 def A2_subh_l16_sat_ll : T_XTYPE_ADD_SUB <0b00, 1, 0, 1>;
896 def A2_subh_l16_sat_hl : T_XTYPE_ADD_SUB <0b10, 1, 0, 1>;
898 //Rd=add(Rt.L,Rs.[LH]):sat
899 def A2_addh_l16_sat_ll : T_XTYPE_ADD_SUB <0b00, 1, 0, 0>;
900 def A2_addh_l16_sat_hl : T_XTYPE_ADD_SUB <0b10, 1, 0, 0>;
903 //Rd=sub(Rt.[LH],Rs.[LH]):<<16
904 let isCodeGenOnly = 0 in {
905 def A2_subh_h16_ll : T_XTYPE_ADD_SUB <0b00, 0, 1, 1>;
906 def A2_subh_h16_lh : T_XTYPE_ADD_SUB <0b01, 0, 1, 1>;
907 def A2_subh_h16_hl : T_XTYPE_ADD_SUB <0b10, 0, 1, 1>;
908 def A2_subh_h16_hh : T_XTYPE_ADD_SUB <0b11, 0, 1, 1>;
911 //Rd=add(Rt.[LH],Rs.[LH]):<<16
912 let isCodeGenOnly = 0 in {
913 def A2_addh_h16_ll : T_XTYPE_ADD_SUB <0b00, 0, 1, 0>;
914 def A2_addh_h16_lh : T_XTYPE_ADD_SUB <0b01, 0, 1, 0>;
915 def A2_addh_h16_hl : T_XTYPE_ADD_SUB <0b10, 0, 1, 0>;
916 def A2_addh_h16_hh : T_XTYPE_ADD_SUB <0b11, 0, 1, 0>;
919 let Itinerary = ALU64_tc_2_SLOT23, Defs = [USR_OVF], isCodeGenOnly = 0 in {
920 //Rd=sub(Rt.[LH],Rs.[LH]):sat:<<16
921 def A2_subh_h16_sat_ll : T_XTYPE_ADD_SUB <0b00, 1, 1, 1>;
922 def A2_subh_h16_sat_lh : T_XTYPE_ADD_SUB <0b01, 1, 1, 1>;
923 def A2_subh_h16_sat_hl : T_XTYPE_ADD_SUB <0b10, 1, 1, 1>;
924 def A2_subh_h16_sat_hh : T_XTYPE_ADD_SUB <0b11, 1, 1, 1>;
926 //Rd=add(Rt.[LH],Rs.[LH]):sat:<<16
927 def A2_addh_h16_sat_ll : T_XTYPE_ADD_SUB <0b00, 1, 1, 0>;
928 def A2_addh_h16_sat_lh : T_XTYPE_ADD_SUB <0b01, 1, 1, 0>;
929 def A2_addh_h16_sat_hl : T_XTYPE_ADD_SUB <0b10, 1, 1, 0>;
930 def A2_addh_h16_sat_hh : T_XTYPE_ADD_SUB <0b11, 1, 1, 0>;
934 def: Pat<(sext_inreg (add I32:$src1, I32:$src2), i16),
935 (A2_addh_l16_ll I32:$src1, I32:$src2)>;
937 def: Pat<(sra (add (shl I32:$src1, (i32 16)), I32:$src2), (i32 16)),
938 (A2_addh_l16_hl I32:$src1, I32:$src2)>;
940 def: Pat<(shl (add I32:$src1, I32:$src2), (i32 16)),
941 (A2_addh_h16_ll I32:$src1, I32:$src2)>;
943 // Subtract halfword.
944 def: Pat<(sext_inreg (sub I32:$src1, I32:$src2), i16),
945 (A2_subh_l16_ll I32:$src1, I32:$src2)>;
947 def: Pat<(shl (sub I32:$src1, I32:$src2), (i32 16)),
948 (A2_subh_h16_ll I32:$src1, I32:$src2)>;
950 let hasSideEffects = 0, hasNewValue = 1, isCodeGenOnly = 0 in
951 def S2_parityp: ALU64Inst<(outs IntRegs:$Rd),
952 (ins DoubleRegs:$Rs, DoubleRegs:$Rt),
953 "$Rd = parity($Rs, $Rt)", [], "", ALU64_tc_2_SLOT23> {
959 let Inst{27-24} = 0b0000;
960 let Inst{20-16} = Rs;
965 let hasNewValue = 1, opNewValue = 0, hasSideEffects = 0 in
966 class T_XTYPE_MIN_MAX < bit isMax, bit isUnsigned >
967 : ALU64Inst < (outs IntRegs:$Rd), (ins IntRegs:$Rt, IntRegs:$Rs),
968 "$Rd = "#!if(isMax,"max","min")#!if(isUnsigned,"u","")
969 #"($Rt, $Rs)", [], "", ALU64_tc_2_SLOT23> {
976 let Inst{27-23} = 0b01011;
977 let Inst{22-21} = !if(isMax, 0b10, 0b01);
978 let Inst{7} = isUnsigned;
980 let Inst{12-8} = !if(isMax, Rs, Rt);
981 let Inst{20-16} = !if(isMax, Rt, Rs);
984 let isCodeGenOnly = 0 in {
985 def A2_min : T_XTYPE_MIN_MAX < 0, 0 >;
986 def A2_minu : T_XTYPE_MIN_MAX < 0, 1 >;
987 def A2_max : T_XTYPE_MIN_MAX < 1, 0 >;
988 def A2_maxu : T_XTYPE_MIN_MAX < 1, 1 >;
991 // Here, depending on the operand being selected, we'll either generate a
992 // min or max instruction.
994 // (a>b)?a:b --> max(a,b) => Here check performed is '>' and the value selected
995 // is the larger of two. So, the corresponding HexagonInst is passed in 'Inst'.
996 // (a>b)?b:a --> min(a,b) => Here check performed is '>' but the smaller value
997 // is selected and the corresponding HexagonInst is passed in 'SwapInst'.
999 multiclass T_MinMax_pats <PatFrag Op, RegisterClass RC, ValueType VT,
1000 InstHexagon Inst, InstHexagon SwapInst> {
1001 def: Pat<(select (i1 (Op (VT RC:$src1), (VT RC:$src2))),
1002 (VT RC:$src1), (VT RC:$src2)),
1003 (Inst RC:$src1, RC:$src2)>;
1004 def: Pat<(select (i1 (Op (VT RC:$src1), (VT RC:$src2))),
1005 (VT RC:$src2), (VT RC:$src1)),
1006 (SwapInst RC:$src1, RC:$src2)>;
1010 multiclass MinMax_pats <PatFrag Op, InstHexagon Inst, InstHexagon SwapInst> {
1011 defm: T_MinMax_pats<Op, IntRegs, i32, Inst, SwapInst>;
1013 def: Pat<(sext_inreg (i32 (select (i1 (Op (i32 PositiveHalfWord:$src1),
1014 (i32 PositiveHalfWord:$src2))),
1015 (i32 PositiveHalfWord:$src1),
1016 (i32 PositiveHalfWord:$src2))), i16),
1017 (Inst IntRegs:$src1, IntRegs:$src2)>;
1019 def: Pat<(sext_inreg (i32 (select (i1 (Op (i32 PositiveHalfWord:$src1),
1020 (i32 PositiveHalfWord:$src2))),
1021 (i32 PositiveHalfWord:$src2),
1022 (i32 PositiveHalfWord:$src1))), i16),
1023 (SwapInst IntRegs:$src1, IntRegs:$src2)>;
1026 let AddedComplexity = 200 in {
1027 defm: MinMax_pats<setge, A2_max, A2_min>;
1028 defm: MinMax_pats<setgt, A2_max, A2_min>;
1029 defm: MinMax_pats<setle, A2_min, A2_max>;
1030 defm: MinMax_pats<setlt, A2_min, A2_max>;
1031 defm: MinMax_pats<setuge, A2_maxu, A2_minu>;
1032 defm: MinMax_pats<setugt, A2_maxu, A2_minu>;
1033 defm: MinMax_pats<setule, A2_minu, A2_maxu>;
1034 defm: MinMax_pats<setult, A2_minu, A2_maxu>;
1037 class T_cmp64_rr<string mnemonic, bits<3> MinOp, bit IsComm>
1038 : ALU64_rr<(outs PredRegs:$Pd), (ins DoubleRegs:$Rs, DoubleRegs:$Rt),
1039 "$Pd = "#mnemonic#"($Rs, $Rt)", [], "", ALU64_tc_2early_SLOT23> {
1041 let isCommutable = IsComm;
1042 let hasSideEffects = 0;
1048 let IClass = 0b1101;
1049 let Inst{27-21} = 0b0010100;
1050 let Inst{20-16} = Rs;
1051 let Inst{12-8} = Rt;
1052 let Inst{7-5} = MinOp;
1056 let isCodeGenOnly = 0 in {
1057 def C2_cmpeqp : T_cmp64_rr<"cmp.eq", 0b000, 1>;
1058 def C2_cmpgtp : T_cmp64_rr<"cmp.gt", 0b010, 0>;
1059 def C2_cmpgtup : T_cmp64_rr<"cmp.gtu", 0b100, 0>;
1062 class T_cmp64_rr_pat<InstHexagon MI, PatFrag CmpOp>
1063 : Pat<(i1 (CmpOp (i64 DoubleRegs:$Rs), (i64 DoubleRegs:$Rt))),
1064 (i1 (MI DoubleRegs:$Rs, DoubleRegs:$Rt))>;
1066 def: T_cmp64_rr_pat<C2_cmpeqp, seteq>;
1067 def: T_cmp64_rr_pat<C2_cmpgtp, setgt>;
1068 def: T_cmp64_rr_pat<C2_cmpgtup, setugt>;
1069 def: T_cmp64_rr_pat<C2_cmpgtp, RevCmp<setlt>>;
1070 def: T_cmp64_rr_pat<C2_cmpgtup, RevCmp<setult>>;
1072 let isCodeGenOnly = 0 in
1073 def C2_vmux : ALU64_rr<(outs DoubleRegs:$Rd),
1074 (ins PredRegs:$Pu, DoubleRegs:$Rs, DoubleRegs:$Rt),
1075 "$Rd = vmux($Pu, $Rs, $Rt)", [], "", ALU64_tc_1_SLOT23> {
1076 let hasSideEffects = 0;
1083 let IClass = 0b1101;
1084 let Inst{27-24} = 0b0001;
1085 let Inst{20-16} = Rs;
1086 let Inst{12-8} = Rt;
1091 class T_ALU64_rr<string mnemonic, string suffix, bits<4> RegType,
1092 bits<3> MajOp, bits<3> MinOp, bit OpsRev, bit IsComm,
1094 : ALU64_rr<(outs DoubleRegs:$Rd), (ins DoubleRegs:$Rs, DoubleRegs:$Rt),
1095 "$Rd = " #mnemonic# "($Rs, " #Op2Pfx# "$Rt)" #suffix, [],
1096 "", ALU64_tc_1_SLOT23> {
1097 let hasSideEffects = 0;
1098 let isCommutable = IsComm;
1104 let IClass = 0b1101;
1105 let Inst{27-24} = RegType;
1106 let Inst{23-21} = MajOp;
1107 let Inst{20-16} = !if (OpsRev,Rt,Rs);
1108 let Inst{12-8} = !if (OpsRev,Rs,Rt);
1109 let Inst{7-5} = MinOp;
1113 class T_ALU64_arith<string mnemonic, bits<3> MajOp, bits<3> MinOp, bit IsSat,
1114 bit OpsRev, bit IsComm>
1115 : T_ALU64_rr<mnemonic, !if(IsSat,":sat",""), 0b0011, MajOp, MinOp, OpsRev,
1118 let isCodeGenOnly = 0 in {
1119 def A2_addp : T_ALU64_arith<"add", 0b000, 0b111, 0, 0, 1>;
1120 def A2_subp : T_ALU64_arith<"sub", 0b001, 0b111, 0, 1, 0>;
1123 def: Pat<(i64 (add I64:$Rs, I64:$Rt)), (A2_addp I64:$Rs, I64:$Rt)>;
1124 def: Pat<(i64 (sub I64:$Rs, I64:$Rt)), (A2_subp I64:$Rs, I64:$Rt)>;
1126 class T_ALU64_logical<string mnemonic, bits<3> MinOp, bit OpsRev, bit IsComm,
1128 : T_ALU64_rr<mnemonic, "", 0b0011, 0b111, MinOp, OpsRev, IsComm,
1131 let isCodeGenOnly = 0 in {
1132 def A2_andp : T_ALU64_logical<"and", 0b000, 0, 1, 0>;
1133 def A2_orp : T_ALU64_logical<"or", 0b010, 0, 1, 0>;
1134 def A2_xorp : T_ALU64_logical<"xor", 0b100, 0, 1, 0>;
1137 def: Pat<(i64 (and I64:$Rs, I64:$Rt)), (A2_andp I64:$Rs, I64:$Rt)>;
1138 def: Pat<(i64 (or I64:$Rs, I64:$Rt)), (A2_orp I64:$Rs, I64:$Rt)>;
1139 def: Pat<(i64 (xor I64:$Rs, I64:$Rt)), (A2_xorp I64:$Rs, I64:$Rt)>;
1141 //===----------------------------------------------------------------------===//
1143 //===----------------------------------------------------------------------===//
1145 //===----------------------------------------------------------------------===//
1147 //===----------------------------------------------------------------------===//
1149 //===----------------------------------------------------------------------===//
1151 //===----------------------------------------------------------------------===//
1153 //===----------------------------------------------------------------------===//
1155 //===----------------------------------------------------------------------===//
1157 //===----------------------------------------------------------------------===//
1159 //===----------------------------------------------------------------------===//
1161 //===----------------------------------------------------------------------===//
1163 //===----------------------------------------------------------------------===//
1164 // Logical reductions on predicates.
1166 // Looping instructions.
1168 // Pipelined looping instructions.
1170 // Logical operations on predicates.
1171 let hasSideEffects = 0 in
1172 class T_LOGICAL_1OP<string MnOp, bits<2> OpBits>
1173 : CRInst<(outs PredRegs:$Pd), (ins PredRegs:$Ps),
1174 "$Pd = " # MnOp # "($Ps)", [], "", CR_tc_2early_SLOT23> {
1178 let IClass = 0b0110;
1179 let Inst{27-23} = 0b10111;
1180 let Inst{22-21} = OpBits;
1182 let Inst{17-16} = Ps;
1187 let isCodeGenOnly = 0 in {
1188 def C2_any8 : T_LOGICAL_1OP<"any8", 0b00>;
1189 def C2_all8 : T_LOGICAL_1OP<"all8", 0b01>;
1190 def C2_not : T_LOGICAL_1OP<"not", 0b10>;
1193 def: Pat<(i1 (not (i1 PredRegs:$Ps))),
1194 (C2_not PredRegs:$Ps)>;
1196 let hasSideEffects = 0 in
1197 class T_LOGICAL_2OP<string MnOp, bits<3> OpBits, bit IsNeg, bit Rev>
1198 : CRInst<(outs PredRegs:$Pd), (ins PredRegs:$Ps, PredRegs:$Pt),
1199 "$Pd = " # MnOp # "($Ps, " # !if (IsNeg,"!","") # "$Pt)",
1200 [], "", CR_tc_2early_SLOT23> {
1205 let IClass = 0b0110;
1206 let Inst{27-24} = 0b1011;
1207 let Inst{23-21} = OpBits;
1209 let Inst{17-16} = !if(Rev,Pt,Ps); // Rs and Rt are reversed for some
1210 let Inst{13} = 0b0; // instructions.
1211 let Inst{9-8} = !if(Rev,Ps,Pt);
1215 let isCodeGenOnly = 0 in {
1216 def C2_and : T_LOGICAL_2OP<"and", 0b000, 0, 1>;
1217 def C2_or : T_LOGICAL_2OP<"or", 0b001, 0, 1>;
1218 def C2_xor : T_LOGICAL_2OP<"xor", 0b010, 0, 0>;
1219 def C2_andn : T_LOGICAL_2OP<"and", 0b011, 1, 1>;
1220 def C2_orn : T_LOGICAL_2OP<"or", 0b111, 1, 1>;
1223 def: Pat<(i1 (and I1:$Ps, I1:$Pt)), (C2_and I1:$Ps, I1:$Pt)>;
1224 def: Pat<(i1 (or I1:$Ps, I1:$Pt)), (C2_or I1:$Ps, I1:$Pt)>;
1225 def: Pat<(i1 (xor I1:$Ps, I1:$Pt)), (C2_xor I1:$Ps, I1:$Pt)>;
1226 def: Pat<(i1 (and I1:$Ps, (not I1:$Pt))), (C2_andn I1:$Ps, I1:$Pt)>;
1227 def: Pat<(i1 (or I1:$Ps, (not I1:$Pt))), (C2_orn I1:$Ps, I1:$Pt)>;
1229 let hasSideEffects = 0, hasNewValue = 1, isCodeGenOnly = 0 in
1230 def C2_vitpack : SInst<(outs IntRegs:$Rd), (ins PredRegs:$Ps, PredRegs:$Pt),
1231 "$Rd = vitpack($Ps, $Pt)", [], "", S_2op_tc_1_SLOT23> {
1236 let IClass = 0b1000;
1237 let Inst{27-24} = 0b1001;
1238 let Inst{22-21} = 0b00;
1239 let Inst{17-16} = Ps;
1244 let hasSideEffects = 0, isCodeGenOnly = 0 in
1245 def C2_mask : SInst<(outs DoubleRegs:$Rd), (ins PredRegs:$Pt),
1246 "$Rd = mask($Pt)", [], "", S_2op_tc_1_SLOT23> {
1250 let IClass = 0b1000;
1251 let Inst{27-24} = 0b0110;
1256 def VALIGN_rrp : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1,
1259 "$dst = valignb($src1, $src2, $src3)",
1262 def VSPLICE_rrp : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1,
1265 "$dst = vspliceb($src1, $src2, $src3)",
1268 // User control register transfer.
1269 //===----------------------------------------------------------------------===//
1271 //===----------------------------------------------------------------------===//
1273 //===----------------------------------------------------------------------===//
1275 //===----------------------------------------------------------------------===//
1277 def retflag : SDNode<"HexagonISD::RET_FLAG", SDTNone,
1278 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
1279 def eh_return: SDNode<"HexagonISD::EH_RETURN", SDTNone, [SDNPHasChain]>;
1281 def SDHexagonBR_JT: SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
1282 def HexagonBR_JT: SDNode<"HexagonISD::BR_JT", SDHexagonBR_JT, [SDNPHasChain]>;
1284 class CondStr<string CReg, bit True, bit New> {
1285 string S = "if (" # !if(True,"","!") # CReg # !if(New,".new","") # ") ";
1287 class JumpOpcStr<string Mnemonic, bit New, bit Taken> {
1288 string S = Mnemonic # !if(New, !if(Taken,":t",":nt"), "");
1291 let isBranch = 1, isBarrier = 1, Defs = [PC], hasSideEffects = 0,
1293 isExtendable = 1, opExtendable = 0, isExtentSigned = 1,
1294 opExtentBits = 24, opExtentAlign = 2, InputType = "imm" in
1295 class T_JMP<string ExtStr>
1296 : JInst<(outs), (ins brtarget:$dst),
1297 "jump " # ExtStr # "$dst",
1298 [], "", J_tc_2early_SLOT23> {
1300 let IClass = 0b0101;
1302 let Inst{27-25} = 0b100;
1303 let Inst{24-16} = dst{23-15};
1304 let Inst{13-1} = dst{14-2};
1307 let isBranch = 1, Defs = [PC], hasSideEffects = 0, isPredicated = 1,
1308 isExtendable = 1, opExtendable = 1, isExtentSigned = 1,
1309 opExtentBits = 17, opExtentAlign = 2, InputType = "imm" in
1310 class T_JMP_c<bit PredNot, bit isPredNew, bit isTak, string ExtStr>
1311 : JInst<(outs), (ins PredRegs:$src, brtarget:$dst),
1312 CondStr<"$src", !if(PredNot,0,1), isPredNew>.S #
1313 JumpOpcStr<"jump", isPredNew, isTak>.S # " " #
1315 [], "", J_tc_2early_SLOT23>, ImmRegRel {
1316 let isTaken = isTak;
1317 let isPredicatedFalse = PredNot;
1318 let isPredicatedNew = isPredNew;
1322 let IClass = 0b0101;
1324 let Inst{27-24} = 0b1100;
1325 let Inst{21} = PredNot;
1326 let Inst{12} = !if(isPredNew, isTak, zero);
1327 let Inst{11} = isPredNew;
1328 let Inst{9-8} = src;
1329 let Inst{23-22} = dst{16-15};
1330 let Inst{20-16} = dst{14-10};
1331 let Inst{13} = dst{9};
1332 let Inst{7-1} = dst{8-2};
1335 multiclass JMP_Pred<bit PredNot, string ExtStr> {
1336 def NAME : T_JMP_c<PredNot, 0, 0, ExtStr>;
1338 def NAME#newpt : T_JMP_c<PredNot, 1, 1, ExtStr>; // taken
1339 def NAME#new : T_JMP_c<PredNot, 1, 0, ExtStr>; // not taken
1342 multiclass JMP_base<string BaseOp, string ExtStr> {
1343 let BaseOpcode = BaseOp in {
1344 def NAME : T_JMP<ExtStr>;
1345 defm t : JMP_Pred<0, ExtStr>;
1346 defm f : JMP_Pred<1, ExtStr>;
1350 // Jumps to address stored in a register, JUMPR_MISC
1351 // if ([[!]P[.new]]) jumpr[:t/nt] Rs
1352 let isBranch = 1, isIndirectBranch = 1, isBarrier = 1, Defs = [PC],
1353 isPredicable = 1, hasSideEffects = 0, InputType = "reg" in
1355 : JRInst<(outs), (ins IntRegs:$dst),
1356 "jumpr $dst", [], "", J_tc_2early_SLOT2> {
1359 let IClass = 0b0101;
1360 let Inst{27-21} = 0b0010100;
1361 let Inst{20-16} = dst;
1364 let isBranch = 1, isIndirectBranch = 1, Defs = [PC], isPredicated = 1,
1365 hasSideEffects = 0, InputType = "reg" in
1366 class T_JMPr_c <bit PredNot, bit isPredNew, bit isTak>
1367 : JRInst <(outs), (ins PredRegs:$src, IntRegs:$dst),
1368 CondStr<"$src", !if(PredNot,0,1), isPredNew>.S #
1369 JumpOpcStr<"jumpr", isPredNew, isTak>.S # " $dst", [],
1370 "", J_tc_2early_SLOT2> {
1372 let isTaken = isTak;
1373 let isPredicatedFalse = PredNot;
1374 let isPredicatedNew = isPredNew;
1378 let IClass = 0b0101;
1380 let Inst{27-22} = 0b001101;
1381 let Inst{21} = PredNot;
1382 let Inst{20-16} = dst;
1383 let Inst{12} = !if(isPredNew, isTak, zero);
1384 let Inst{11} = isPredNew;
1385 let Inst{9-8} = src;
1388 multiclass JMPR_Pred<bit PredNot> {
1389 def NAME: T_JMPr_c<PredNot, 0, 0>;
1391 def NAME#newpt : T_JMPr_c<PredNot, 1, 1>; // taken
1392 def NAME#new : T_JMPr_c<PredNot, 1, 0>; // not taken
1395 multiclass JMPR_base<string BaseOp> {
1396 let BaseOpcode = BaseOp in {
1398 defm t : JMPR_Pred<0>;
1399 defm f : JMPR_Pred<1>;
1403 let isCall = 1, hasSideEffects = 1 in
1404 class JUMPR_MISC_CALLR<bit isPred, bit isPredNot,
1405 dag InputDag = (ins IntRegs:$Rs)>
1406 : JRInst<(outs), InputDag,
1407 !if(isPred, !if(isPredNot, "if (!$Pu) callr $Rs",
1408 "if ($Pu) callr $Rs"),
1410 [], "", J_tc_2early_SLOT2> {
1413 let isPredicated = isPred;
1414 let isPredicatedFalse = isPredNot;
1416 let IClass = 0b0101;
1417 let Inst{27-25} = 0b000;
1418 let Inst{24-23} = !if (isPred, 0b10, 0b01);
1420 let Inst{21} = isPredNot;
1421 let Inst{9-8} = !if (isPred, Pu, 0b00);
1422 let Inst{20-16} = Rs;
1426 let Defs = VolatileV3.Regs, isCodeGenOnly = 0 in {
1427 def J2_callrt : JUMPR_MISC_CALLR<1, 0, (ins PredRegs:$Pu, IntRegs:$Rs)>;
1428 def J2_callrf : JUMPR_MISC_CALLR<1, 1, (ins PredRegs:$Pu, IntRegs:$Rs)>;
1431 let isTerminator = 1, hasSideEffects = 0, isCodeGenOnly = 0 in {
1432 defm J2_jump : JMP_base<"JMP", "">, PredNewRel;
1434 // Deal with explicit assembly
1435 // - never extened a jump #, always extend a jump ##
1436 let isAsmParserOnly = 1 in {
1437 defm J2_jump_ext : JMP_base<"JMP", "##">;
1438 defm J2_jump_noext : JMP_base<"JMP", "#">;
1441 defm J2_jumpr : JMPR_base<"JMPr">, PredNewRel;
1443 let isReturn = 1, isCodeGenOnly = 1 in
1444 defm JMPret : JMPR_base<"JMPret">, PredNewRel;
1447 def: Pat<(br bb:$dst),
1448 (J2_jump brtarget:$dst)>;
1450 (JMPret (i32 R31))>;
1451 def: Pat<(brcond (i1 PredRegs:$src1), bb:$offset),
1452 (J2_jumpt PredRegs:$src1, bb:$offset)>;
1454 // A return through builtin_eh_return.
1455 let isReturn = 1, isTerminator = 1, isBarrier = 1, hasSideEffects = 0,
1456 isCodeGenOnly = 1, Defs = [PC], Uses = [R28], isPredicable = 0 in
1457 def EH_RETURN_JMPR : T_JMPr;
1459 def: Pat<(eh_return),
1460 (EH_RETURN_JMPR (i32 R31))>;
1461 def: Pat<(HexagonBR_JT (i32 IntRegs:$dst)),
1462 (J2_jumpr IntRegs:$dst)>;
1463 def: Pat<(brind (i32 IntRegs:$dst)),
1464 (J2_jumpr IntRegs:$dst)>;
1466 //===----------------------------------------------------------------------===//
1468 //===----------------------------------------------------------------------===//
1470 //===----------------------------------------------------------------------===//
1472 //===----------------------------------------------------------------------===//
1473 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, AddedComplexity = 20 in
1474 class T_load_io <string mnemonic, RegisterClass RC, bits<4> MajOp,
1476 : LDInst<(outs RC:$dst), (ins IntRegs:$src1, ImmOp:$offset),
1477 "$dst = "#mnemonic#"($src1 + #$offset)", []>, AddrModeRel {
1482 bits<11> offsetBits;
1484 string ImmOpStr = !cast<string>(ImmOp);
1485 let offsetBits = !if (!eq(ImmOpStr, "s11_3Ext"), offset{13-3},
1486 !if (!eq(ImmOpStr, "s11_2Ext"), offset{12-2},
1487 !if (!eq(ImmOpStr, "s11_1Ext"), offset{11-1},
1488 /* s11_0Ext */ offset{10-0})));
1489 let opExtentBits = !if (!eq(ImmOpStr, "s11_3Ext"), 14,
1490 !if (!eq(ImmOpStr, "s11_2Ext"), 13,
1491 !if (!eq(ImmOpStr, "s11_1Ext"), 12,
1492 /* s11_0Ext */ 11)));
1493 let hasNewValue = !if (!eq(ImmOpStr, "s11_3Ext"), 0, 1);
1495 let IClass = 0b1001;
1498 let Inst{26-25} = offsetBits{10-9};
1499 let Inst{24-21} = MajOp;
1500 let Inst{20-16} = src1;
1501 let Inst{13-5} = offsetBits{8-0};
1502 let Inst{4-0} = dst;
1505 let opExtendable = 3, isExtentSigned = 0, isPredicated = 1 in
1506 class T_pload_io <string mnemonic, RegisterClass RC, bits<4>MajOp,
1507 Operand ImmOp, bit isNot, bit isPredNew>
1508 : LDInst<(outs RC:$dst),
1509 (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset),
1510 "if ("#!if(isNot, "!$src1", "$src1")
1511 #!if(isPredNew, ".new", "")
1512 #") $dst = "#mnemonic#"($src2 + #$offset)",
1513 [],"", V2LDST_tc_ld_SLOT01> , AddrModeRel {
1519 string ImmOpStr = !cast<string>(ImmOp);
1521 let offsetBits = !if (!eq(ImmOpStr, "u6_3Ext"), offset{8-3},
1522 !if (!eq(ImmOpStr, "u6_2Ext"), offset{7-2},
1523 !if (!eq(ImmOpStr, "u6_1Ext"), offset{6-1},
1524 /* u6_0Ext */ offset{5-0})));
1525 let opExtentBits = !if (!eq(ImmOpStr, "u6_3Ext"), 9,
1526 !if (!eq(ImmOpStr, "u6_2Ext"), 8,
1527 !if (!eq(ImmOpStr, "u6_1Ext"), 7,
1529 let hasNewValue = !if (!eq(ImmOpStr, "u6_3Ext"), 0, 1);
1530 let isPredicatedNew = isPredNew;
1531 let isPredicatedFalse = isNot;
1533 let IClass = 0b0100;
1537 let Inst{26} = isNot;
1538 let Inst{25} = isPredNew;
1539 let Inst{24-21} = MajOp;
1540 let Inst{20-16} = src2;
1542 let Inst{12-11} = src1;
1543 let Inst{10-5} = offsetBits;
1544 let Inst{4-0} = dst;
1547 let isExtendable = 1, hasSideEffects = 0, addrMode = BaseImmOffset in
1548 multiclass LD_Idxd<string mnemonic, string CextOp, RegisterClass RC,
1549 Operand ImmOp, Operand predImmOp, bits<4>MajOp> {
1550 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
1551 let isPredicable = 1 in
1552 def L2_#NAME#_io : T_load_io <mnemonic, RC, MajOp, ImmOp>;
1555 def L2_p#NAME#t_io : T_pload_io <mnemonic, RC, MajOp, predImmOp, 0, 0>;
1556 def L2_p#NAME#f_io : T_pload_io <mnemonic, RC, MajOp, predImmOp, 1, 0>;
1559 def L2_p#NAME#tnew_io : T_pload_io <mnemonic, RC, MajOp, predImmOp, 0, 1>;
1560 def L2_p#NAME#fnew_io : T_pload_io <mnemonic, RC, MajOp, predImmOp, 1, 1>;
1564 let accessSize = ByteAccess, isCodeGenOnly = 0 in {
1565 defm loadrb: LD_Idxd <"memb", "LDrib", IntRegs, s11_0Ext, u6_0Ext, 0b1000>;
1566 defm loadrub: LD_Idxd <"memub", "LDriub", IntRegs, s11_0Ext, u6_0Ext, 0b1001>;
1569 let accessSize = HalfWordAccess, opExtentAlign = 1, isCodeGenOnly = 0 in {
1570 defm loadrh: LD_Idxd <"memh", "LDrih", IntRegs, s11_1Ext, u6_1Ext, 0b1010>;
1571 defm loadruh: LD_Idxd <"memuh", "LDriuh", IntRegs, s11_1Ext, u6_1Ext, 0b1011>;
1574 let accessSize = WordAccess, opExtentAlign = 2, isCodeGenOnly = 0 in
1575 defm loadri: LD_Idxd <"memw", "LDriw", IntRegs, s11_2Ext, u6_2Ext, 0b1100>;
1577 let accessSize = DoubleWordAccess, opExtentAlign = 3, isCodeGenOnly = 0 in
1578 defm loadrd: LD_Idxd <"memd", "LDrid", DoubleRegs, s11_3Ext, u6_3Ext, 0b1110>;
1580 // Patterns to select load-indexed (i.e. load from base+offset).
1581 multiclass Loadx_pat<PatFrag Load, ValueType VT, PatLeaf ImmPred,
1583 def: Pat<(VT (Load AddrFI:$fi)), (VT (MI AddrFI:$fi, 0))>;
1584 def: Pat<(VT (Load (add (i32 IntRegs:$Rs), ImmPred:$Off))),
1585 (VT (MI IntRegs:$Rs, imm:$Off))>;
1586 def: Pat<(VT (Load (i32 IntRegs:$Rs))), (VT (MI IntRegs:$Rs, 0))>;
1589 def : Pat < (i32 (sextloadi8 ADDRriS11_0:$addr)),
1590 (L2_loadrb_io AddrFI:$addr, 0) >;
1592 def : Pat < (i32 (zextloadi8 ADDRriS11_0:$addr)),
1593 (L2_loadrub_io AddrFI:$addr, 0) >;
1595 def : Pat < (i32 (sextloadi16 ADDRriS11_1:$addr)),
1596 (L2_loadrh_io AddrFI:$addr, 0) >;
1598 def : Pat < (i32 (zextloadi16 ADDRriS11_1:$addr)),
1599 (L2_loadruh_io AddrFI:$addr, 0) >;
1601 def : Pat < (i32 (load ADDRriS11_2:$addr)),
1602 (L2_loadri_io AddrFI:$addr, 0) >;
1604 def : Pat < (i64 (load ADDRriS11_3:$addr)),
1605 (L2_loadrd_io AddrFI:$addr, 0) >;
1607 let AddedComplexity = 20 in {
1608 def : Pat < (i32 (sextloadi8 (add IntRegs:$src1, s11_0ExtPred:$offset))),
1609 (L2_loadrb_io IntRegs:$src1, s11_0ExtPred:$offset) >;
1611 def : Pat < (i32 (zextloadi8 (add IntRegs:$src1, s11_0ExtPred:$offset))),
1612 (L2_loadrub_io IntRegs:$src1, s11_0ExtPred:$offset) >;
1614 def : Pat < (i32 (sextloadi16 (add IntRegs:$src1, s11_1ExtPred:$offset))),
1615 (L2_loadrh_io IntRegs:$src1, s11_1ExtPred:$offset) >;
1617 def : Pat < (i32 (zextloadi16 (add IntRegs:$src1, s11_1ExtPred:$offset))),
1618 (L2_loadruh_io IntRegs:$src1, s11_1ExtPred:$offset) >;
1620 def : Pat < (i32 (load (add IntRegs:$src1, s11_2ExtPred:$offset))),
1621 (L2_loadri_io IntRegs:$src1, s11_2ExtPred:$offset) >;
1623 def : Pat < (i64 (load (add IntRegs:$src1, s11_3ExtPred:$offset))),
1624 (L2_loadrd_io IntRegs:$src1, s11_3ExtPred:$offset) >;
1627 //===----------------------------------------------------------------------===//
1628 // Post increment load
1629 //===----------------------------------------------------------------------===//
1630 //===----------------------------------------------------------------------===//
1631 // Template class for non-predicated post increment loads with immediate offset.
1632 //===----------------------------------------------------------------------===//
1633 let hasSideEffects = 0, addrMode = PostInc in
1634 class T_load_pi <string mnemonic, RegisterClass RC, Operand ImmOp,
1636 : LDInstPI <(outs RC:$dst, IntRegs:$dst2),
1637 (ins IntRegs:$src1, ImmOp:$offset),
1638 "$dst = "#mnemonic#"($src1++#$offset)" ,
1647 string ImmOpStr = !cast<string>(ImmOp);
1648 let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
1649 !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
1650 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
1651 /* s4_0Imm */ offset{3-0})));
1652 let hasNewValue = !if (!eq(ImmOpStr, "s4_3Imm"), 0, 1);
1654 let IClass = 0b1001;
1656 let Inst{27-25} = 0b101;
1657 let Inst{24-21} = MajOp;
1658 let Inst{20-16} = src1;
1659 let Inst{13-12} = 0b00;
1660 let Inst{8-5} = offsetBits;
1661 let Inst{4-0} = dst;
1664 //===----------------------------------------------------------------------===//
1665 // Template class for predicated post increment loads with immediate offset.
1666 //===----------------------------------------------------------------------===//
1667 let isPredicated = 1, hasSideEffects = 0, addrMode = PostInc in
1668 class T_pload_pi <string mnemonic, RegisterClass RC, Operand ImmOp,
1669 bits<4> MajOp, bit isPredNot, bit isPredNew >
1670 : LDInst <(outs RC:$dst, IntRegs:$dst2),
1671 (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset),
1672 !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
1673 ") ")#"$dst = "#mnemonic#"($src2++#$offset)",
1683 let isPredicatedNew = isPredNew;
1684 let isPredicatedFalse = isPredNot;
1686 string ImmOpStr = !cast<string>(ImmOp);
1687 let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
1688 !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
1689 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
1690 /* s4_0Imm */ offset{3-0})));
1691 let hasNewValue = !if (!eq(ImmOpStr, "s4_3Imm"), 0, 1);
1693 let IClass = 0b1001;
1695 let Inst{27-25} = 0b101;
1696 let Inst{24-21} = MajOp;
1697 let Inst{20-16} = src2;
1699 let Inst{12} = isPredNew;
1700 let Inst{11} = isPredNot;
1701 let Inst{10-9} = src1;
1702 let Inst{8-5} = offsetBits;
1703 let Inst{4-0} = dst;
1706 //===----------------------------------------------------------------------===//
1707 // Multiclass for post increment loads with immediate offset.
1708 //===----------------------------------------------------------------------===//
1710 multiclass LD_PostInc <string mnemonic, string BaseOp, RegisterClass RC,
1711 Operand ImmOp, bits<4> MajOp> {
1712 let BaseOpcode = "POST_"#BaseOp in {
1713 let isPredicable = 1 in
1714 def L2_#NAME#_pi : T_load_pi < mnemonic, RC, ImmOp, MajOp>;
1717 def L2_p#NAME#t_pi : T_pload_pi < mnemonic, RC, ImmOp, MajOp, 0, 0>;
1718 def L2_p#NAME#f_pi : T_pload_pi < mnemonic, RC, ImmOp, MajOp, 1, 0>;
1721 def L2_p#NAME#tnew_pi : T_pload_pi < mnemonic, RC, ImmOp, MajOp, 0, 1>;
1722 def L2_p#NAME#fnew_pi : T_pload_pi < mnemonic, RC, ImmOp, MajOp, 1, 1>;
1726 // post increment byte loads with immediate offset
1727 let accessSize = ByteAccess, isCodeGenOnly = 0 in {
1728 defm loadrb : LD_PostInc <"memb", "LDrib", IntRegs, s4_0Imm, 0b1000>;
1729 defm loadrub : LD_PostInc <"memub", "LDriub", IntRegs, s4_0Imm, 0b1001>;
1732 // post increment halfword loads with immediate offset
1733 let accessSize = HalfWordAccess, opExtentAlign = 1, isCodeGenOnly = 0 in {
1734 defm loadrh : LD_PostInc <"memh", "LDrih", IntRegs, s4_1Imm, 0b1010>;
1735 defm loadruh : LD_PostInc <"memuh", "LDriuh", IntRegs, s4_1Imm, 0b1011>;
1738 // post increment word loads with immediate offset
1739 let accessSize = WordAccess, opExtentAlign = 2, isCodeGenOnly = 0 in
1740 defm loadri : LD_PostInc <"memw", "LDriw", IntRegs, s4_2Imm, 0b1100>;
1742 // post increment doubleword loads with immediate offset
1743 let accessSize = DoubleWordAccess, opExtentAlign = 3, isCodeGenOnly = 0 in
1744 defm loadrd : LD_PostInc <"memd", "LDrid", DoubleRegs, s4_3Imm, 0b1110>;
1746 def : Pat< (i32 (extloadi1 ADDRriS11_0:$addr)),
1747 (i32 (L2_loadrb_io AddrFI:$addr, 0)) >;
1749 // Load byte any-extend.
1750 def : Pat < (i32 (extloadi8 ADDRriS11_0:$addr)),
1751 (i32 (L2_loadrb_io AddrFI:$addr, 0)) >;
1753 // Indexed load byte any-extend.
1754 let AddedComplexity = 20 in
1755 def : Pat < (i32 (extloadi8 (add IntRegs:$src1, s11_0ImmPred:$offset))),
1756 (i32 (L2_loadrb_io IntRegs:$src1, s11_0ImmPred:$offset)) >;
1758 def : Pat < (i32 (extloadi16 ADDRriS11_1:$addr)),
1759 (i32 (L2_loadrh_io AddrFI:$addr, 0))>;
1761 let AddedComplexity = 20 in
1762 def : Pat < (i32 (extloadi16 (add IntRegs:$src1, s11_1ImmPred:$offset))),
1763 (i32 (L2_loadrh_io IntRegs:$src1, s11_1ImmPred:$offset)) >;
1765 let AddedComplexity = 10 in
1766 def : Pat < (i32 (zextloadi1 ADDRriS11_0:$addr)),
1767 (i32 (L2_loadrub_io AddrFI:$addr, 0))>;
1769 let AddedComplexity = 20 in
1770 def : Pat < (i32 (zextloadi1 (add IntRegs:$src1, s11_0ImmPred:$offset))),
1771 (i32 (L2_loadrub_io IntRegs:$src1, s11_0ImmPred:$offset))>;
1773 //===----------------------------------------------------------------------===//
1774 // Template class for post increment loads with register offset.
1775 //===----------------------------------------------------------------------===//
1776 let hasSideEffects = 0, addrMode = PostInc in
1777 class T_load_pr <string mnemonic, RegisterClass RC, bits<4> MajOp,
1778 MemAccessSize AccessSz>
1779 : LDInstPI <(outs RC:$dst, IntRegs:$_dst_),
1780 (ins IntRegs:$src1, ModRegs:$src2),
1781 "$dst = "#mnemonic#"($src1++$src2)" ,
1782 [], "$src1 = $_dst_" > {
1787 let accessSize = AccessSz;
1788 let IClass = 0b1001;
1790 let Inst{27-25} = 0b110;
1791 let Inst{24-21} = MajOp;
1792 let Inst{20-16} = src1;
1793 let Inst{13} = src2;
1796 let Inst{4-0} = dst;
1799 let hasNewValue = 1, isCodeGenOnly = 0 in {
1800 def L2_loadrb_pr : T_load_pr <"memb", IntRegs, 0b1000, ByteAccess>;
1801 def L2_loadrub_pr : T_load_pr <"memub", IntRegs, 0b1001, ByteAccess>;
1802 def L2_loadrh_pr : T_load_pr <"memh", IntRegs, 0b1010, HalfWordAccess>;
1803 def L2_loadruh_pr : T_load_pr <"memuh", IntRegs, 0b1011, HalfWordAccess>;
1804 def L2_loadri_pr : T_load_pr <"memw", IntRegs, 0b1100, WordAccess>;
1807 let isCodeGenOnly = 0 in
1808 def L2_loadrd_pr : T_load_pr <"memd", DoubleRegs, 0b1110, DoubleWordAccess>;
1811 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 13,
1812 isPseudo = 1, Defs = [R10,R11,D5], hasSideEffects = 0 in
1813 def LDriw_pred : LDInst2<(outs PredRegs:$dst),
1815 "Error; should not emit",
1818 let Defs = [R29, R30, R31], Uses = [R30], hasSideEffects = 0, isCodeGenOnly = 0 in
1819 def L2_deallocframe : LDInst<(outs), (ins),
1822 let IClass = 0b1001;
1824 let Inst{27-16} = 0b000000011110;
1826 let Inst{4-0} = 0b11110;
1829 // Load / Post increment circular addressing mode.
1830 let Uses = [CS], hasSideEffects = 0, hasNewValue = 1, opNewValue = 0 in
1831 class T_load_pcr<string mnemonic, RegisterClass RC, bits<4> MajOp>
1832 : LDInst <(outs RC:$dst, IntRegs:$_dst_),
1833 (ins IntRegs:$Rz, ModRegs:$Mu),
1834 "$dst = "#mnemonic#"($Rz ++ I:circ($Mu))", [],
1840 let IClass = 0b1001;
1842 let Inst{27-25} = 0b100;
1843 let Inst{24-21} = MajOp;
1844 let Inst{20-16} = Rz;
1849 let Inst{4-0} = dst;
1852 let accessSize = ByteAccess, isCodeGenOnly = 0 in {
1853 def L2_loadrb_pcr : T_load_pcr <"memb", IntRegs, 0b1000>;
1854 def L2_loadrub_pcr : T_load_pcr <"memub", IntRegs, 0b1001>;
1857 let accessSize = HalfWordAccess, isCodeGenOnly = 0 in {
1858 def L2_loadrh_pcr : T_load_pcr <"memh", IntRegs, 0b1010>;
1859 def L2_loadruh_pcr : T_load_pcr <"memuh", IntRegs, 0b1011>;
1862 let accessSize = WordAccess, isCodeGenOnly = 0 in {
1863 def L2_loadri_pcr : T_load_pcr <"memw", IntRegs, 0b1100>;
1866 let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
1867 def L2_loadrd_pcr : T_load_pcr <"memd", DoubleRegs, 0b1110>;
1869 //===----------------------------------------------------------------------===//
1870 // Circular loads with immediate offset.
1871 //===----------------------------------------------------------------------===//
1872 let Uses = [CS], mayLoad = 1, hasSideEffects = 0, hasNewValue = 1 in
1873 class T_load_pci <string mnemonic, RegisterClass RC,
1874 Operand ImmOp, bits<4> MajOp>
1875 : LDInstPI<(outs RC:$dst, IntRegs:$_dst_),
1876 (ins IntRegs:$Rz, ImmOp:$offset, ModRegs:$Mu),
1877 "$dst = "#mnemonic#"($Rz ++ #$offset:circ($Mu))", [],
1885 string ImmOpStr = !cast<string>(ImmOp);
1886 let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
1887 !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
1888 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
1889 /* s4_0Imm */ offset{3-0})));
1890 let IClass = 0b1001;
1891 let Inst{27-25} = 0b100;
1892 let Inst{24-21} = MajOp;
1893 let Inst{20-16} = Rz;
1897 let Inst{8-5} = offsetBits;
1898 let Inst{4-0} = dst;
1901 // Byte variants of circ load
1902 let accessSize = ByteAccess, isCodeGenOnly = 0 in {
1903 def L2_loadrb_pci : T_load_pci <"memb", IntRegs, s4_0Imm, 0b1000>;
1904 def L2_loadrub_pci : T_load_pci <"memub", IntRegs, s4_0Imm, 0b1001>;
1907 // Half word variants of circ load
1908 let accessSize = HalfWordAccess, isCodeGenOnly = 0 in {
1909 def L2_loadrh_pci : T_load_pci <"memh", IntRegs, s4_1Imm, 0b1010>;
1910 def L2_loadruh_pci : T_load_pci <"memuh", IntRegs, s4_1Imm, 0b1011>;
1913 // Word variants of circ load
1914 let accessSize = WordAccess, isCodeGenOnly = 0 in
1915 def L2_loadri_pci : T_load_pci <"memw", IntRegs, s4_2Imm, 0b1100>;
1917 let accessSize = DoubleWordAccess, hasNewValue = 0, isCodeGenOnly = 0 in
1918 def L2_loadrd_pci : T_load_pci <"memd", DoubleRegs, s4_3Imm, 0b1110>;
1920 // L[24]_load[wd]_locked: Load word/double with lock.
1922 class T_load_locked <string mnemonic, RegisterClass RC>
1923 : LD0Inst <(outs RC:$dst),
1925 "$dst = "#mnemonic#"($src)"> {
1928 let IClass = 0b1001;
1929 let Inst{27-21} = 0b0010000;
1930 let Inst{20-16} = src;
1931 let Inst{13-12} = !if (!eq(mnemonic, "memd_locked"), 0b01, 0b00);
1932 let Inst{4-0} = dst;
1934 let hasNewValue = 1, accessSize = WordAccess, opNewValue = 0, isCodeGenOnly = 0 in
1935 def L2_loadw_locked : T_load_locked <"memw_locked", IntRegs>;
1936 let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
1937 def L4_loadd_locked : T_load_locked <"memd_locked", DoubleRegs>;
1938 //===----------------------------------------------------------------------===//
1939 // Bit-reversed loads with auto-increment register
1940 //===----------------------------------------------------------------------===//
1941 let hasSideEffects = 0 in
1942 class T_load_pbr<string mnemonic, RegisterClass RC,
1943 MemAccessSize addrSize, bits<4> majOp>
1945 <(outs RC:$dst, IntRegs:$_dst_),
1946 (ins IntRegs:$Rz, ModRegs:$Mu),
1947 "$dst = "#mnemonic#"($Rz ++ $Mu:brev)" ,
1948 [] , "$Rz = $_dst_" > {
1950 let accessSize = addrSize;
1956 let IClass = 0b1001;
1958 let Inst{27-25} = 0b111;
1959 let Inst{24-21} = majOp;
1960 let Inst{20-16} = Rz;
1964 let Inst{4-0} = dst;
1967 let hasNewValue =1, opNewValue = 0, isCodeGenOnly = 0 in {
1968 def L2_loadrb_pbr : T_load_pbr <"memb", IntRegs, ByteAccess, 0b1000>;
1969 def L2_loadrub_pbr : T_load_pbr <"memub", IntRegs, ByteAccess, 0b1001>;
1970 def L2_loadrh_pbr : T_load_pbr <"memh", IntRegs, HalfWordAccess, 0b1010>;
1971 def L2_loadruh_pbr : T_load_pbr <"memuh", IntRegs, HalfWordAccess, 0b1011>;
1972 def L2_loadri_pbr : T_load_pbr <"memw", IntRegs, WordAccess, 0b1100>;
1975 let isCodeGenOnly = 0 in
1976 def L2_loadrd_pbr : T_load_pbr <"memd", DoubleRegs, DoubleWordAccess, 0b1110>;
1978 //===----------------------------------------------------------------------===//
1980 //===----------------------------------------------------------------------===//
1982 //===----------------------------------------------------------------------===//
1984 //===----------------------------------------------------------------------===//
1985 //===----------------------------------------------------------------------===//
1987 //===----------------------------------------------------------------------===//
1989 //===----------------------------------------------------------------------===//
1991 //===----------------------------------------------------------------------===//
1992 //===----------------------------------------------------------------------===//
1994 //===----------------------------------------------------------------------===//
1996 //===----------------------------------------------------------------------===//
1998 //===----------------------------------------------------------------------===//
2000 //===----------------------------------------------------------------------===//
2002 // MPYS / Multipy signed/unsigned halfwords
2003 //Rd=mpy[u](Rs.[H|L],Rt.[H|L])[:<<1][:rnd][:sat]
2004 //===----------------------------------------------------------------------===//
2006 let hasNewValue = 1, opNewValue = 0 in
2007 class T_M2_mpy < bits<2> LHbits, bit isSat, bit isRnd,
2008 bit hasShift, bit isUnsigned>
2009 : MInst < (outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
2010 "$Rd = "#!if(isUnsigned,"mpyu","mpy")#"($Rs."#!if(LHbits{1},"h","l")
2011 #", $Rt."#!if(LHbits{0},"h)","l)")
2012 #!if(hasShift,":<<1","")
2013 #!if(isRnd,":rnd","")
2014 #!if(isSat,":sat",""),
2015 [], "", M_tc_3x_SLOT23 > {
2020 let IClass = 0b1110;
2022 let Inst{27-24} = 0b1100;
2023 let Inst{23} = hasShift;
2024 let Inst{22} = isUnsigned;
2025 let Inst{21} = isRnd;
2026 let Inst{7} = isSat;
2027 let Inst{6-5} = LHbits;
2029 let Inst{20-16} = Rs;
2030 let Inst{12-8} = Rt;
2033 //Rd=mpy(Rs.[H|L],Rt.[H|L])[:<<1]
2034 let isCodeGenOnly = 0 in {
2035 def M2_mpy_ll_s1: T_M2_mpy<0b00, 0, 0, 1, 0>;
2036 def M2_mpy_ll_s0: T_M2_mpy<0b00, 0, 0, 0, 0>;
2037 def M2_mpy_lh_s1: T_M2_mpy<0b01, 0, 0, 1, 0>;
2038 def M2_mpy_lh_s0: T_M2_mpy<0b01, 0, 0, 0, 0>;
2039 def M2_mpy_hl_s1: T_M2_mpy<0b10, 0, 0, 1, 0>;
2040 def M2_mpy_hl_s0: T_M2_mpy<0b10, 0, 0, 0, 0>;
2041 def M2_mpy_hh_s1: T_M2_mpy<0b11, 0, 0, 1, 0>;
2042 def M2_mpy_hh_s0: T_M2_mpy<0b11, 0, 0, 0, 0>;
2045 //Rd=mpyu(Rs.[H|L],Rt.[H|L])[:<<1]
2046 let isCodeGenOnly = 0 in {
2047 def M2_mpyu_ll_s1: T_M2_mpy<0b00, 0, 0, 1, 1>;
2048 def M2_mpyu_ll_s0: T_M2_mpy<0b00, 0, 0, 0, 1>;
2049 def M2_mpyu_lh_s1: T_M2_mpy<0b01, 0, 0, 1, 1>;
2050 def M2_mpyu_lh_s0: T_M2_mpy<0b01, 0, 0, 0, 1>;
2051 def M2_mpyu_hl_s1: T_M2_mpy<0b10, 0, 0, 1, 1>;
2052 def M2_mpyu_hl_s0: T_M2_mpy<0b10, 0, 0, 0, 1>;
2053 def M2_mpyu_hh_s1: T_M2_mpy<0b11, 0, 0, 1, 1>;
2054 def M2_mpyu_hh_s0: T_M2_mpy<0b11, 0, 0, 0, 1>;
2057 //Rd=mpy(Rs.[H|L],Rt.[H|L])[:<<1]:rnd
2058 let isCodeGenOnly = 0 in {
2059 def M2_mpy_rnd_ll_s1: T_M2_mpy <0b00, 0, 1, 1, 0>;
2060 def M2_mpy_rnd_ll_s0: T_M2_mpy <0b00, 0, 1, 0, 0>;
2061 def M2_mpy_rnd_lh_s1: T_M2_mpy <0b01, 0, 1, 1, 0>;
2062 def M2_mpy_rnd_lh_s0: T_M2_mpy <0b01, 0, 1, 0, 0>;
2063 def M2_mpy_rnd_hl_s1: T_M2_mpy <0b10, 0, 1, 1, 0>;
2064 def M2_mpy_rnd_hl_s0: T_M2_mpy <0b10, 0, 1, 0, 0>;
2065 def M2_mpy_rnd_hh_s1: T_M2_mpy <0b11, 0, 1, 1, 0>;
2066 def M2_mpy_rnd_hh_s0: T_M2_mpy <0b11, 0, 1, 0, 0>;
2069 //Rd=mpy(Rs.[H|L],Rt.[H|L])[:<<1][:sat]
2070 //Rd=mpy(Rs.[H|L],Rt.[H|L])[:<<1][:rnd][:sat]
2071 let Defs = [USR_OVF], isCodeGenOnly = 0 in {
2072 def M2_mpy_sat_ll_s1: T_M2_mpy <0b00, 1, 0, 1, 0>;
2073 def M2_mpy_sat_ll_s0: T_M2_mpy <0b00, 1, 0, 0, 0>;
2074 def M2_mpy_sat_lh_s1: T_M2_mpy <0b01, 1, 0, 1, 0>;
2075 def M2_mpy_sat_lh_s0: T_M2_mpy <0b01, 1, 0, 0, 0>;
2076 def M2_mpy_sat_hl_s1: T_M2_mpy <0b10, 1, 0, 1, 0>;
2077 def M2_mpy_sat_hl_s0: T_M2_mpy <0b10, 1, 0, 0, 0>;
2078 def M2_mpy_sat_hh_s1: T_M2_mpy <0b11, 1, 0, 1, 0>;
2079 def M2_mpy_sat_hh_s0: T_M2_mpy <0b11, 1, 0, 0, 0>;
2081 def M2_mpy_sat_rnd_ll_s1: T_M2_mpy <0b00, 1, 1, 1, 0>;
2082 def M2_mpy_sat_rnd_ll_s0: T_M2_mpy <0b00, 1, 1, 0, 0>;
2083 def M2_mpy_sat_rnd_lh_s1: T_M2_mpy <0b01, 1, 1, 1, 0>;
2084 def M2_mpy_sat_rnd_lh_s0: T_M2_mpy <0b01, 1, 1, 0, 0>;
2085 def M2_mpy_sat_rnd_hl_s1: T_M2_mpy <0b10, 1, 1, 1, 0>;
2086 def M2_mpy_sat_rnd_hl_s0: T_M2_mpy <0b10, 1, 1, 0, 0>;
2087 def M2_mpy_sat_rnd_hh_s1: T_M2_mpy <0b11, 1, 1, 1, 0>;
2088 def M2_mpy_sat_rnd_hh_s0: T_M2_mpy <0b11, 1, 1, 0, 0>;
2091 //===----------------------------------------------------------------------===//
2093 // MPYS / Multipy signed/unsigned halfwords and add/subtract the
2094 // result from the accumulator.
2095 //Rx [-+]= mpy[u](Rs.[H|L],Rt.[H|L])[:<<1][:sat]
2096 //===----------------------------------------------------------------------===//
2098 let hasNewValue = 1, opNewValue = 0 in
2099 class T_M2_mpy_acc < bits<2> LHbits, bit isSat, bit isNac,
2100 bit hasShift, bit isUnsigned >
2101 : MInst_acc<(outs IntRegs:$Rx), (ins IntRegs:$dst2, IntRegs:$Rs, IntRegs:$Rt),
2102 "$Rx "#!if(isNac,"-= ","+= ")#!if(isUnsigned,"mpyu","mpy")
2103 #"($Rs."#!if(LHbits{1},"h","l")
2104 #", $Rt."#!if(LHbits{0},"h)","l)")
2105 #!if(hasShift,":<<1","")
2106 #!if(isSat,":sat",""),
2107 [], "$dst2 = $Rx", M_tc_3x_SLOT23 > {
2112 let IClass = 0b1110;
2113 let Inst{27-24} = 0b1110;
2114 let Inst{23} = hasShift;
2115 let Inst{22} = isUnsigned;
2116 let Inst{21} = isNac;
2117 let Inst{7} = isSat;
2118 let Inst{6-5} = LHbits;
2120 let Inst{20-16} = Rs;
2121 let Inst{12-8} = Rt;
2124 //Rx += mpy(Rs.[H|L],Rt.[H|L])[:<<1]
2125 let isCodeGenOnly = 0 in {
2126 def M2_mpy_acc_ll_s1: T_M2_mpy_acc <0b00, 0, 0, 1, 0>;
2127 def M2_mpy_acc_ll_s0: T_M2_mpy_acc <0b00, 0, 0, 0, 0>;
2128 def M2_mpy_acc_lh_s1: T_M2_mpy_acc <0b01, 0, 0, 1, 0>;
2129 def M2_mpy_acc_lh_s0: T_M2_mpy_acc <0b01, 0, 0, 0, 0>;
2130 def M2_mpy_acc_hl_s1: T_M2_mpy_acc <0b10, 0, 0, 1, 0>;
2131 def M2_mpy_acc_hl_s0: T_M2_mpy_acc <0b10, 0, 0, 0, 0>;
2132 def M2_mpy_acc_hh_s1: T_M2_mpy_acc <0b11, 0, 0, 1, 0>;
2133 def M2_mpy_acc_hh_s0: T_M2_mpy_acc <0b11, 0, 0, 0, 0>;
2136 //Rx += mpyu(Rs.[H|L],Rt.[H|L])[:<<1]
2137 let isCodeGenOnly = 0 in {
2138 def M2_mpyu_acc_ll_s1: T_M2_mpy_acc <0b00, 0, 0, 1, 1>;
2139 def M2_mpyu_acc_ll_s0: T_M2_mpy_acc <0b00, 0, 0, 0, 1>;
2140 def M2_mpyu_acc_lh_s1: T_M2_mpy_acc <0b01, 0, 0, 1, 1>;
2141 def M2_mpyu_acc_lh_s0: T_M2_mpy_acc <0b01, 0, 0, 0, 1>;
2142 def M2_mpyu_acc_hl_s1: T_M2_mpy_acc <0b10, 0, 0, 1, 1>;
2143 def M2_mpyu_acc_hl_s0: T_M2_mpy_acc <0b10, 0, 0, 0, 1>;
2144 def M2_mpyu_acc_hh_s1: T_M2_mpy_acc <0b11, 0, 0, 1, 1>;
2145 def M2_mpyu_acc_hh_s0: T_M2_mpy_acc <0b11, 0, 0, 0, 1>;
2148 //Rx -= mpy(Rs.[H|L],Rt.[H|L])[:<<1]
2149 let isCodeGenOnly = 0 in {
2150 def M2_mpy_nac_ll_s1: T_M2_mpy_acc <0b00, 0, 1, 1, 0>;
2151 def M2_mpy_nac_ll_s0: T_M2_mpy_acc <0b00, 0, 1, 0, 0>;
2152 def M2_mpy_nac_lh_s1: T_M2_mpy_acc <0b01, 0, 1, 1, 0>;
2153 def M2_mpy_nac_lh_s0: T_M2_mpy_acc <0b01, 0, 1, 0, 0>;
2154 def M2_mpy_nac_hl_s1: T_M2_mpy_acc <0b10, 0, 1, 1, 0>;
2155 def M2_mpy_nac_hl_s0: T_M2_mpy_acc <0b10, 0, 1, 0, 0>;
2156 def M2_mpy_nac_hh_s1: T_M2_mpy_acc <0b11, 0, 1, 1, 0>;
2157 def M2_mpy_nac_hh_s0: T_M2_mpy_acc <0b11, 0, 1, 0, 0>;
2160 //Rx -= mpyu(Rs.[H|L],Rt.[H|L])[:<<1]
2161 let isCodeGenOnly = 0 in {
2162 def M2_mpyu_nac_ll_s1: T_M2_mpy_acc <0b00, 0, 1, 1, 1>;
2163 def M2_mpyu_nac_ll_s0: T_M2_mpy_acc <0b00, 0, 1, 0, 1>;
2164 def M2_mpyu_nac_lh_s1: T_M2_mpy_acc <0b01, 0, 1, 1, 1>;
2165 def M2_mpyu_nac_lh_s0: T_M2_mpy_acc <0b01, 0, 1, 0, 1>;
2166 def M2_mpyu_nac_hl_s1: T_M2_mpy_acc <0b10, 0, 1, 1, 1>;
2167 def M2_mpyu_nac_hl_s0: T_M2_mpy_acc <0b10, 0, 1, 0, 1>;
2168 def M2_mpyu_nac_hh_s1: T_M2_mpy_acc <0b11, 0, 1, 1, 1>;
2169 def M2_mpyu_nac_hh_s0: T_M2_mpy_acc <0b11, 0, 1, 0, 1>;
2172 //Rx += mpy(Rs.[H|L],Rt.[H|L])[:<<1]:sat
2173 let isCodeGenOnly = 0 in {
2174 def M2_mpy_acc_sat_ll_s1: T_M2_mpy_acc <0b00, 1, 0, 1, 0>;
2175 def M2_mpy_acc_sat_ll_s0: T_M2_mpy_acc <0b00, 1, 0, 0, 0>;
2176 def M2_mpy_acc_sat_lh_s1: T_M2_mpy_acc <0b01, 1, 0, 1, 0>;
2177 def M2_mpy_acc_sat_lh_s0: T_M2_mpy_acc <0b01, 1, 0, 0, 0>;
2178 def M2_mpy_acc_sat_hl_s1: T_M2_mpy_acc <0b10, 1, 0, 1, 0>;
2179 def M2_mpy_acc_sat_hl_s0: T_M2_mpy_acc <0b10, 1, 0, 0, 0>;
2180 def M2_mpy_acc_sat_hh_s1: T_M2_mpy_acc <0b11, 1, 0, 1, 0>;
2181 def M2_mpy_acc_sat_hh_s0: T_M2_mpy_acc <0b11, 1, 0, 0, 0>;
2184 //Rx -= mpy(Rs.[H|L],Rt.[H|L])[:<<1]:sat
2185 let isCodeGenOnly = 0 in {
2186 def M2_mpy_nac_sat_ll_s1: T_M2_mpy_acc <0b00, 1, 1, 1, 0>;
2187 def M2_mpy_nac_sat_ll_s0: T_M2_mpy_acc <0b00, 1, 1, 0, 0>;
2188 def M2_mpy_nac_sat_lh_s1: T_M2_mpy_acc <0b01, 1, 1, 1, 0>;
2189 def M2_mpy_nac_sat_lh_s0: T_M2_mpy_acc <0b01, 1, 1, 0, 0>;
2190 def M2_mpy_nac_sat_hl_s1: T_M2_mpy_acc <0b10, 1, 1, 1, 0>;
2191 def M2_mpy_nac_sat_hl_s0: T_M2_mpy_acc <0b10, 1, 1, 0, 0>;
2192 def M2_mpy_nac_sat_hh_s1: T_M2_mpy_acc <0b11, 1, 1, 1, 0>;
2193 def M2_mpy_nac_sat_hh_s0: T_M2_mpy_acc <0b11, 1, 1, 0, 0>;
2196 //===----------------------------------------------------------------------===//
2198 // MPYS / Multipy signed/unsigned halfwords and add/subtract the
2199 // result from the 64-bit destination register.
2200 //Rxx [-+]= mpy[u](Rs.[H|L],Rt.[H|L])[:<<1][:sat]
2201 //===----------------------------------------------------------------------===//
2203 class T_M2_mpyd_acc < bits<2> LHbits, bit isNac, bit hasShift, bit isUnsigned>
2204 : MInst_acc<(outs DoubleRegs:$Rxx),
2205 (ins DoubleRegs:$dst2, IntRegs:$Rs, IntRegs:$Rt),
2206 "$Rxx "#!if(isNac,"-= ","+= ")#!if(isUnsigned,"mpyu","mpy")
2207 #"($Rs."#!if(LHbits{1},"h","l")
2208 #", $Rt."#!if(LHbits{0},"h)","l)")
2209 #!if(hasShift,":<<1",""),
2210 [], "$dst2 = $Rxx", M_tc_3x_SLOT23 > {
2215 let IClass = 0b1110;
2217 let Inst{27-24} = 0b0110;
2218 let Inst{23} = hasShift;
2219 let Inst{22} = isUnsigned;
2220 let Inst{21} = isNac;
2222 let Inst{6-5} = LHbits;
2223 let Inst{4-0} = Rxx;
2224 let Inst{20-16} = Rs;
2225 let Inst{12-8} = Rt;
2228 let isCodeGenOnly = 0 in {
2229 def M2_mpyd_acc_hh_s0: T_M2_mpyd_acc <0b11, 0, 0, 0>;
2230 def M2_mpyd_acc_hl_s0: T_M2_mpyd_acc <0b10, 0, 0, 0>;
2231 def M2_mpyd_acc_lh_s0: T_M2_mpyd_acc <0b01, 0, 0, 0>;
2232 def M2_mpyd_acc_ll_s0: T_M2_mpyd_acc <0b00, 0, 0, 0>;
2234 def M2_mpyd_acc_hh_s1: T_M2_mpyd_acc <0b11, 0, 1, 0>;
2235 def M2_mpyd_acc_hl_s1: T_M2_mpyd_acc <0b10, 0, 1, 0>;
2236 def M2_mpyd_acc_lh_s1: T_M2_mpyd_acc <0b01, 0, 1, 0>;
2237 def M2_mpyd_acc_ll_s1: T_M2_mpyd_acc <0b00, 0, 1, 0>;
2239 def M2_mpyd_nac_hh_s0: T_M2_mpyd_acc <0b11, 1, 0, 0>;
2240 def M2_mpyd_nac_hl_s0: T_M2_mpyd_acc <0b10, 1, 0, 0>;
2241 def M2_mpyd_nac_lh_s0: T_M2_mpyd_acc <0b01, 1, 0, 0>;
2242 def M2_mpyd_nac_ll_s0: T_M2_mpyd_acc <0b00, 1, 0, 0>;
2244 def M2_mpyd_nac_hh_s1: T_M2_mpyd_acc <0b11, 1, 1, 0>;
2245 def M2_mpyd_nac_hl_s1: T_M2_mpyd_acc <0b10, 1, 1, 0>;
2246 def M2_mpyd_nac_lh_s1: T_M2_mpyd_acc <0b01, 1, 1, 0>;
2247 def M2_mpyd_nac_ll_s1: T_M2_mpyd_acc <0b00, 1, 1, 0>;
2249 def M2_mpyud_acc_hh_s0: T_M2_mpyd_acc <0b11, 0, 0, 1>;
2250 def M2_mpyud_acc_hl_s0: T_M2_mpyd_acc <0b10, 0, 0, 1>;
2251 def M2_mpyud_acc_lh_s0: T_M2_mpyd_acc <0b01, 0, 0, 1>;
2252 def M2_mpyud_acc_ll_s0: T_M2_mpyd_acc <0b00, 0, 0, 1>;
2254 def M2_mpyud_acc_hh_s1: T_M2_mpyd_acc <0b11, 0, 1, 1>;
2255 def M2_mpyud_acc_hl_s1: T_M2_mpyd_acc <0b10, 0, 1, 1>;
2256 def M2_mpyud_acc_lh_s1: T_M2_mpyd_acc <0b01, 0, 1, 1>;
2257 def M2_mpyud_acc_ll_s1: T_M2_mpyd_acc <0b00, 0, 1, 1>;
2259 def M2_mpyud_nac_hh_s0: T_M2_mpyd_acc <0b11, 1, 0, 1>;
2260 def M2_mpyud_nac_hl_s0: T_M2_mpyd_acc <0b10, 1, 0, 1>;
2261 def M2_mpyud_nac_lh_s0: T_M2_mpyd_acc <0b01, 1, 0, 1>;
2262 def M2_mpyud_nac_ll_s0: T_M2_mpyd_acc <0b00, 1, 0, 1>;
2264 def M2_mpyud_nac_hh_s1: T_M2_mpyd_acc <0b11, 1, 1, 1>;
2265 def M2_mpyud_nac_hl_s1: T_M2_mpyd_acc <0b10, 1, 1, 1>;
2266 def M2_mpyud_nac_lh_s1: T_M2_mpyd_acc <0b01, 1, 1, 1>;
2267 def M2_mpyud_nac_ll_s1: T_M2_mpyd_acc <0b00, 1, 1, 1>;
2270 let hasNewValue = 1, opNewValue = 0 in
2271 class T_MType_mpy <string mnemonic, bits<4> RegTyBits, RegisterClass RC,
2272 bits<3> MajOp, bits<3> MinOp, bit isSat = 0, bit isRnd = 0,
2273 string op2Suffix = "", bit isRaw = 0, bit isHi = 0 >
2274 : MInst <(outs IntRegs:$dst), (ins RC:$src1, RC:$src2),
2276 #"($src1, $src2"#op2Suffix#")"
2277 #!if(MajOp{2}, ":<<1", "")
2278 #!if(isRnd, ":rnd", "")
2279 #!if(isSat, ":sat", "")
2280 #!if(isRaw, !if(isHi, ":raw:hi", ":raw:lo"), ""), [] > {
2285 let IClass = 0b1110;
2287 let Inst{27-24} = RegTyBits;
2288 let Inst{23-21} = MajOp;
2289 let Inst{20-16} = src1;
2291 let Inst{12-8} = src2;
2292 let Inst{7-5} = MinOp;
2293 let Inst{4-0} = dst;
2296 class T_MType_dd <string mnemonic, bits<3> MajOp, bits<3> MinOp,
2297 bit isSat = 0, bit isRnd = 0 >
2298 : T_MType_mpy <mnemonic, 0b1001, DoubleRegs, MajOp, MinOp, isSat, isRnd>;
2300 class T_MType_rr1 <string mnemonic, bits<3> MajOp, bits<3> MinOp,
2301 bit isSat = 0, bit isRnd = 0 >
2302 : T_MType_mpy<mnemonic, 0b1101, IntRegs, MajOp, MinOp, isSat, isRnd>;
2304 class T_MType_rr2 <string mnemonic, bits<3> MajOp, bits<3> MinOp,
2305 bit isSat = 0, bit isRnd = 0, string op2str = "" >
2306 : T_MType_mpy<mnemonic, 0b1101, IntRegs, MajOp, MinOp, isSat, isRnd, op2str>;
2308 let CextOpcode = "mpyi", InputType = "reg", isCodeGenOnly = 0 in
2309 def M2_mpyi : T_MType_rr1 <"mpyi", 0b000, 0b000>, ImmRegRel;
2311 let isCodeGenOnly = 0 in {
2312 def M2_mpy_up : T_MType_rr1 <"mpy", 0b000, 0b001>;
2313 def M2_mpyu_up : T_MType_rr1 <"mpyu", 0b010, 0b001>;
2316 let isCodeGenOnly = 0 in
2317 def M2_dpmpyss_rnd_s0 : T_MType_rr1 <"mpy", 0b001, 0b001, 0, 1>;
2319 let isCodeGenOnly = 0 in {
2320 def M2_hmmpyh_rs1 : T_MType_rr2 <"mpy", 0b101, 0b100, 1, 1, ".h">;
2321 def M2_hmmpyl_rs1 : T_MType_rr2 <"mpy", 0b111, 0b100, 1, 1, ".l">;
2325 let isCodeGenOnly = 0 in {
2326 def M2_mpysu_up : T_MType_rr1 <"mpysu", 0b011, 0b001, 0>;
2327 def M2_mpy_up_s1_sat : T_MType_rr1 <"mpy", 0b111, 0b000, 1>;
2329 def M2_hmmpyh_s1 : T_MType_rr2 <"mpy", 0b101, 0b000, 1, 0, ".h">;
2330 def M2_hmmpyl_s1 : T_MType_rr2 <"mpy", 0b101, 0b001, 1, 0, ".l">;
2333 def: Pat<(i32 (mul I32:$src1, I32:$src2)), (M2_mpyi I32:$src1, I32:$src2)>;
2334 def: Pat<(i32 (mulhs I32:$src1, I32:$src2)), (M2_mpy_up I32:$src1, I32:$src2)>;
2335 def: Pat<(i32 (mulhu I32:$src1, I32:$src2)), (M2_mpyu_up I32:$src1, I32:$src2)>;
2337 let hasNewValue = 1, opNewValue = 0 in
2338 class T_MType_mpy_ri <bit isNeg, Operand ImmOp, list<dag> pattern>
2339 : MInst < (outs IntRegs:$Rd), (ins IntRegs:$Rs, ImmOp:$u8),
2340 "$Rd ="#!if(isNeg, "- ", "+ ")#"mpyi($Rs, #$u8)" ,
2341 pattern, "", M_tc_3x_SLOT23> {
2346 let IClass = 0b1110;
2348 let Inst{27-24} = 0b0000;
2349 let Inst{23} = isNeg;
2352 let Inst{20-16} = Rs;
2353 let Inst{12-5} = u8;
2356 let isExtendable = 1, opExtentBits = 8, opExtendable = 2, isCodeGenOnly = 0 in
2357 def M2_mpysip : T_MType_mpy_ri <0, u8Ext,
2358 [(set (i32 IntRegs:$Rd), (mul IntRegs:$Rs, u8ExtPred:$u8))]>;
2360 let isCodeGenOnly = 0 in
2361 def M2_mpysin : T_MType_mpy_ri <1, u8Imm,
2362 [(set (i32 IntRegs:$Rd), (ineg (mul IntRegs:$Rs,
2365 // Assember mapped to M2_mpyi
2366 let isAsmParserOnly = 1 in
2367 def M2_mpyui : MInst<(outs IntRegs:$dst),
2368 (ins IntRegs:$src1, IntRegs:$src2),
2369 "$dst = mpyui($src1, $src2)">;
2372 // s9 is NOT the same as m9 - but it works.. so far.
2373 // Assembler maps to either Rd=+mpyi(Rs,#u8) or Rd=-mpyi(Rs,#u8)
2374 // depending on the value of m9. See Arch Spec.
2375 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 9,
2376 CextOpcode = "mpyi", InputType = "imm", hasNewValue = 1 in
2377 def M2_mpysmi : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, s9Ext:$src2),
2378 "$dst = mpyi($src1, #$src2)",
2379 [(set (i32 IntRegs:$dst), (mul (i32 IntRegs:$src1),
2380 s9ExtPred:$src2))]>, ImmRegRel;
2382 let hasNewValue = 1, isExtendable = 1, opExtentBits = 8, opExtendable = 3,
2383 InputType = "imm" in
2384 class T_MType_acc_ri <string mnemonic, bits<3> MajOp, Operand ImmOp,
2385 list<dag> pattern = []>
2386 : MInst < (outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, ImmOp:$src3),
2387 "$dst "#mnemonic#"($src2, #$src3)",
2388 pattern, "$src1 = $dst", M_tc_2_SLOT23> {
2393 let IClass = 0b1110;
2395 let Inst{27-26} = 0b00;
2396 let Inst{25-23} = MajOp;
2397 let Inst{20-16} = src2;
2399 let Inst{12-5} = src3;
2400 let Inst{4-0} = dst;
2403 let InputType = "reg", hasNewValue = 1 in
2404 class T_MType_acc_rr <string mnemonic, bits<3> MajOp, bits<3> MinOp,
2405 bit isSwap = 0, list<dag> pattern = [], bit hasNot = 0,
2406 bit isSat = 0, bit isShift = 0>
2407 : MInst < (outs IntRegs:$dst),
2408 (ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3),
2409 "$dst "#mnemonic#"($src2, "#!if(hasNot, "~$src3)","$src3)")
2410 #!if(isShift, ":<<1", "")
2411 #!if(isSat, ":sat", ""),
2412 pattern, "$src1 = $dst", M_tc_2_SLOT23 > {
2417 let IClass = 0b1110;
2419 let Inst{27-24} = 0b1111;
2420 let Inst{23-21} = MajOp;
2421 let Inst{20-16} = !if(isSwap, src3, src2);
2423 let Inst{12-8} = !if(isSwap, src2, src3);
2424 let Inst{7-5} = MinOp;
2425 let Inst{4-0} = dst;
2428 let CextOpcode = "MPYI_acc", Itinerary = M_tc_3x_SLOT23, isCodeGenOnly = 0 in {
2429 def M2_macsip : T_MType_acc_ri <"+= mpyi", 0b010, u8Ext,
2430 [(set (i32 IntRegs:$dst),
2431 (add (mul IntRegs:$src2, u8ExtPred:$src3),
2432 IntRegs:$src1))]>, ImmRegRel;
2434 def M2_maci : T_MType_acc_rr <"+= mpyi", 0b000, 0b000, 0,
2435 [(set (i32 IntRegs:$dst),
2436 (add (mul IntRegs:$src2, IntRegs:$src3),
2437 IntRegs:$src1))]>, ImmRegRel;
2440 let CextOpcode = "ADD_acc", isCodeGenOnly = 0 in {
2441 let isExtentSigned = 1 in
2442 def M2_accii : T_MType_acc_ri <"+= add", 0b100, s8Ext,
2443 [(set (i32 IntRegs:$dst),
2444 (add (add (i32 IntRegs:$src2), s8_16ExtPred:$src3),
2445 (i32 IntRegs:$src1)))]>, ImmRegRel;
2447 def M2_acci : T_MType_acc_rr <"+= add", 0b000, 0b001, 0,
2448 [(set (i32 IntRegs:$dst),
2449 (add (add (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
2450 (i32 IntRegs:$src1)))]>, ImmRegRel;
2453 let CextOpcode = "SUB_acc", isCodeGenOnly = 0 in {
2454 let isExtentSigned = 1 in
2455 def M2_naccii : T_MType_acc_ri <"-= add", 0b101, s8Ext>, ImmRegRel;
2457 def M2_nacci : T_MType_acc_rr <"-= add", 0b100, 0b001, 0>, ImmRegRel;
2460 let Itinerary = M_tc_3x_SLOT23, isCodeGenOnly = 0 in
2461 def M2_macsin : T_MType_acc_ri <"-= mpyi", 0b011, u8Ext>;
2463 let isCodeGenOnly = 0 in {
2464 def M2_xor_xacc : T_MType_acc_rr < "^= xor", 0b100, 0b011, 0>;
2465 def M2_subacc : T_MType_acc_rr <"+= sub", 0b000, 0b011, 1>;
2468 class T_MType_acc_pat1 <InstHexagon MI, SDNode firstOp, SDNode secOp,
2470 : Pat <(secOp IntRegs:$src1, (firstOp IntRegs:$src2, ImmPred:$src3)),
2471 (MI IntRegs:$src1, IntRegs:$src2, ImmPred:$src3)>;
2473 class T_MType_acc_pat2 <InstHexagon MI, SDNode firstOp, SDNode secOp>
2474 : Pat <(i32 (secOp IntRegs:$src1, (firstOp IntRegs:$src2, IntRegs:$src3))),
2475 (MI IntRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
2477 def : T_MType_acc_pat2 <M2_xor_xacc, xor, xor>;
2478 def : T_MType_acc_pat1 <M2_macsin, mul, sub, u8ExtPred>;
2480 def : T_MType_acc_pat1 <M2_naccii, add, sub, s8_16ExtPred>;
2481 def : T_MType_acc_pat2 <M2_nacci, add, sub>;
2482 //===----------------------------------------------------------------------===//
2483 // Template Class -- Multiply signed/unsigned halfwords with and without
2484 // saturation and rounding
2485 //===----------------------------------------------------------------------===//
2486 class T_M2_mpyd < bits<2> LHbits, bit isRnd, bit hasShift, bit isUnsigned >
2487 : MInst < (outs DoubleRegs:$Rdd), (ins IntRegs:$Rs, IntRegs:$Rt),
2488 "$Rdd = "#!if(isUnsigned,"mpyu","mpy")#"($Rs."#!if(LHbits{1},"h","l")
2489 #", $Rt."#!if(LHbits{0},"h)","l)")
2490 #!if(hasShift,":<<1","")
2491 #!if(isRnd,":rnd",""),
2497 let IClass = 0b1110;
2499 let Inst{27-24} = 0b0100;
2500 let Inst{23} = hasShift;
2501 let Inst{22} = isUnsigned;
2502 let Inst{21} = isRnd;
2503 let Inst{6-5} = LHbits;
2504 let Inst{4-0} = Rdd;
2505 let Inst{20-16} = Rs;
2506 let Inst{12-8} = Rt;
2509 let isCodeGenOnly = 0 in {
2510 def M2_mpyd_hh_s0: T_M2_mpyd<0b11, 0, 0, 0>;
2511 def M2_mpyd_hl_s0: T_M2_mpyd<0b10, 0, 0, 0>;
2512 def M2_mpyd_lh_s0: T_M2_mpyd<0b01, 0, 0, 0>;
2513 def M2_mpyd_ll_s0: T_M2_mpyd<0b00, 0, 0, 0>;
2515 def M2_mpyd_hh_s1: T_M2_mpyd<0b11, 0, 1, 0>;
2516 def M2_mpyd_hl_s1: T_M2_mpyd<0b10, 0, 1, 0>;
2517 def M2_mpyd_lh_s1: T_M2_mpyd<0b01, 0, 1, 0>;
2518 def M2_mpyd_ll_s1: T_M2_mpyd<0b00, 0, 1, 0>;
2520 def M2_mpyd_rnd_hh_s0: T_M2_mpyd<0b11, 1, 0, 0>;
2521 def M2_mpyd_rnd_hl_s0: T_M2_mpyd<0b10, 1, 0, 0>;
2522 def M2_mpyd_rnd_lh_s0: T_M2_mpyd<0b01, 1, 0, 0>;
2523 def M2_mpyd_rnd_ll_s0: T_M2_mpyd<0b00, 1, 0, 0>;
2525 def M2_mpyd_rnd_hh_s1: T_M2_mpyd<0b11, 1, 1, 0>;
2526 def M2_mpyd_rnd_hl_s1: T_M2_mpyd<0b10, 1, 1, 0>;
2527 def M2_mpyd_rnd_lh_s1: T_M2_mpyd<0b01, 1, 1, 0>;
2528 def M2_mpyd_rnd_ll_s1: T_M2_mpyd<0b00, 1, 1, 0>;
2530 //Rdd=mpyu(Rs.[HL],Rt.[HL])[:<<1]
2531 def M2_mpyud_hh_s0: T_M2_mpyd<0b11, 0, 0, 1>;
2532 def M2_mpyud_hl_s0: T_M2_mpyd<0b10, 0, 0, 1>;
2533 def M2_mpyud_lh_s0: T_M2_mpyd<0b01, 0, 0, 1>;
2534 def M2_mpyud_ll_s0: T_M2_mpyd<0b00, 0, 0, 1>;
2536 def M2_mpyud_hh_s1: T_M2_mpyd<0b11, 0, 1, 1>;
2537 def M2_mpyud_hl_s1: T_M2_mpyd<0b10, 0, 1, 1>;
2538 def M2_mpyud_lh_s1: T_M2_mpyd<0b01, 0, 1, 1>;
2539 def M2_mpyud_ll_s1: T_M2_mpyd<0b00, 0, 1, 1>;
2541 //===----------------------------------------------------------------------===//
2542 // Template Class for xtype mpy:
2545 // multiply 32X32 and use full result
2546 //===----------------------------------------------------------------------===//
2547 let hasSideEffects = 0 in
2548 class T_XTYPE_mpy64 <string mnemonic, bits<3> MajOp, bits<3> MinOp,
2549 bit isSat, bit hasShift, bit isConj>
2550 : MInst <(outs DoubleRegs:$Rdd),
2551 (ins IntRegs:$Rs, IntRegs:$Rt),
2552 "$Rdd = "#mnemonic#"($Rs, $Rt"#!if(isConj,"*)",")")
2553 #!if(hasShift,":<<1","")
2554 #!if(isSat,":sat",""),
2560 let IClass = 0b1110;
2562 let Inst{27-24} = 0b0101;
2563 let Inst{23-21} = MajOp;
2564 let Inst{20-16} = Rs;
2565 let Inst{12-8} = Rt;
2566 let Inst{7-5} = MinOp;
2567 let Inst{4-0} = Rdd;
2570 //===----------------------------------------------------------------------===//
2571 // Template Class for xtype mpy with accumulation into 64-bit:
2574 // multiply 32X32 and use full result
2575 //===----------------------------------------------------------------------===//
2576 class T_XTYPE_mpy64_acc <string op1, string op2, bits<3> MajOp, bits<3> MinOp,
2577 bit isSat, bit hasShift, bit isConj>
2578 : MInst <(outs DoubleRegs:$Rxx),
2579 (ins DoubleRegs:$dst2, IntRegs:$Rs, IntRegs:$Rt),
2580 "$Rxx "#op2#"= "#op1#"($Rs, $Rt"#!if(isConj,"*)",")")
2581 #!if(hasShift,":<<1","")
2582 #!if(isSat,":sat",""),
2584 [] , "$dst2 = $Rxx" > {
2589 let IClass = 0b1110;
2591 let Inst{27-24} = 0b0111;
2592 let Inst{23-21} = MajOp;
2593 let Inst{20-16} = Rs;
2594 let Inst{12-8} = Rt;
2595 let Inst{7-5} = MinOp;
2596 let Inst{4-0} = Rxx;
2599 // MPY - Multiply and use full result
2600 // Rdd = mpy[u](Rs,Rt)
2601 let isCodeGenOnly = 0 in {
2602 def M2_dpmpyss_s0 : T_XTYPE_mpy64 < "mpy", 0b000, 0b000, 0, 0, 0>;
2603 def M2_dpmpyuu_s0 : T_XTYPE_mpy64 < "mpyu", 0b010, 0b000, 0, 0, 0>;
2605 // Rxx[+-]= mpy[u](Rs,Rt)
2606 def M2_dpmpyss_acc_s0 : T_XTYPE_mpy64_acc < "mpy", "+", 0b000, 0b000, 0, 0, 0>;
2607 def M2_dpmpyss_nac_s0 : T_XTYPE_mpy64_acc < "mpy", "-", 0b001, 0b000, 0, 0, 0>;
2608 def M2_dpmpyuu_acc_s0 : T_XTYPE_mpy64_acc < "mpyu", "+", 0b010, 0b000, 0, 0, 0>;
2609 def M2_dpmpyuu_nac_s0 : T_XTYPE_mpy64_acc < "mpyu", "-", 0b011, 0b000, 0, 0, 0>;
2612 def: Pat<(i64 (mul (i64 (anyext (i32 IntRegs:$src1))),
2613 (i64 (anyext (i32 IntRegs:$src2))))),
2614 (M2_dpmpyuu_s0 IntRegs:$src1, IntRegs:$src2)>;
2616 def: Pat<(i64 (mul (i64 (sext (i32 IntRegs:$src1))),
2617 (i64 (sext (i32 IntRegs:$src2))))),
2618 (M2_dpmpyss_s0 IntRegs:$src1, IntRegs:$src2)>;
2620 def: Pat<(i64 (mul (is_sext_i32:$src1),
2621 (is_sext_i32:$src2))),
2622 (M2_dpmpyss_s0 (LoReg DoubleRegs:$src1), (LoReg DoubleRegs:$src2))>;
2624 // Multiply and accumulate, use full result.
2625 // Rxx[+-]=mpy(Rs,Rt)
2627 def: Pat<(i64 (add (i64 DoubleRegs:$src1),
2628 (mul (i64 (sext (i32 IntRegs:$src2))),
2629 (i64 (sext (i32 IntRegs:$src3)))))),
2630 (M2_dpmpyss_acc_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
2632 def: Pat<(i64 (sub (i64 DoubleRegs:$src1),
2633 (mul (i64 (sext (i32 IntRegs:$src2))),
2634 (i64 (sext (i32 IntRegs:$src3)))))),
2635 (M2_dpmpyss_nac_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
2637 def: Pat<(i64 (add (i64 DoubleRegs:$src1),
2638 (mul (i64 (anyext (i32 IntRegs:$src2))),
2639 (i64 (anyext (i32 IntRegs:$src3)))))),
2640 (M2_dpmpyuu_acc_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
2642 def: Pat<(i64 (add (i64 DoubleRegs:$src1),
2643 (mul (i64 (zext (i32 IntRegs:$src2))),
2644 (i64 (zext (i32 IntRegs:$src3)))))),
2645 (M2_dpmpyuu_acc_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
2647 def: Pat<(i64 (sub (i64 DoubleRegs:$src1),
2648 (mul (i64 (anyext (i32 IntRegs:$src2))),
2649 (i64 (anyext (i32 IntRegs:$src3)))))),
2650 (M2_dpmpyuu_nac_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
2652 def: Pat<(i64 (sub (i64 DoubleRegs:$src1),
2653 (mul (i64 (zext (i32 IntRegs:$src2))),
2654 (i64 (zext (i32 IntRegs:$src3)))))),
2655 (M2_dpmpyuu_nac_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
2657 //===----------------------------------------------------------------------===//
2659 //===----------------------------------------------------------------------===//
2661 //===----------------------------------------------------------------------===//
2663 //===----------------------------------------------------------------------===//
2664 //===----------------------------------------------------------------------===//
2666 //===----------------------------------------------------------------------===//
2668 //===----------------------------------------------------------------------===//
2670 //===----------------------------------------------------------------------===//
2671 //===----------------------------------------------------------------------===//
2673 //===----------------------------------------------------------------------===//
2675 //===----------------------------------------------------------------------===//
2677 //===----------------------------------------------------------------------===//
2678 //===----------------------------------------------------------------------===//
2680 //===----------------------------------------------------------------------===//
2682 //===----------------------------------------------------------------------===//
2684 //===----------------------------------------------------------------------===//
2686 // Store doubleword.
2687 //===----------------------------------------------------------------------===//
2688 // Template class for non-predicated post increment stores with immediate offset
2689 //===----------------------------------------------------------------------===//
2690 let isPredicable = 1, hasSideEffects = 0, addrMode = PostInc in
2691 class T_store_pi <string mnemonic, RegisterClass RC, Operand ImmOp,
2692 bits<4> MajOp, bit isHalf >
2693 : STInst <(outs IntRegs:$_dst_),
2694 (ins IntRegs:$src1, ImmOp:$offset, RC:$src2),
2695 mnemonic#"($src1++#$offset) = $src2"#!if(isHalf, ".h", ""),
2696 [], "$src1 = $_dst_" >,
2703 string ImmOpStr = !cast<string>(ImmOp);
2704 let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
2705 !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
2706 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
2707 /* s4_0Imm */ offset{3-0})));
2708 let isNVStorable = !if (!eq(ImmOpStr, "s4_3Imm"), 0, 1);
2710 let IClass = 0b1010;
2712 let Inst{27-25} = 0b101;
2713 let Inst{24-21} = MajOp;
2714 let Inst{20-16} = src1;
2716 let Inst{12-8} = src2;
2718 let Inst{6-3} = offsetBits;
2722 //===----------------------------------------------------------------------===//
2723 // Template class for predicated post increment stores with immediate offset
2724 //===----------------------------------------------------------------------===//
2725 let isPredicated = 1, hasSideEffects = 0, addrMode = PostInc in
2726 class T_pstore_pi <string mnemonic, RegisterClass RC, Operand ImmOp,
2727 bits<4> MajOp, bit isHalf, bit isPredNot, bit isPredNew >
2728 : STInst <(outs IntRegs:$_dst_),
2729 (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset, RC:$src3),
2730 !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
2731 ") ")#mnemonic#"($src2++#$offset) = $src3"#!if(isHalf, ".h", ""),
2732 [], "$src2 = $_dst_" >,
2740 string ImmOpStr = !cast<string>(ImmOp);
2741 let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
2742 !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
2743 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
2744 /* s4_0Imm */ offset{3-0})));
2746 let isNVStorable = !if (!eq(ImmOpStr, "s4_3Imm"), 0, 1);
2747 let isPredicatedNew = isPredNew;
2748 let isPredicatedFalse = isPredNot;
2750 let IClass = 0b1010;
2752 let Inst{27-25} = 0b101;
2753 let Inst{24-21} = MajOp;
2754 let Inst{20-16} = src2;
2756 let Inst{12-8} = src3;
2757 let Inst{7} = isPredNew;
2758 let Inst{6-3} = offsetBits;
2759 let Inst{2} = isPredNot;
2760 let Inst{1-0} = src1;
2763 multiclass ST_PostInc<string mnemonic, string BaseOp, RegisterClass RC,
2764 Operand ImmOp, bits<4> MajOp, bit isHalf = 0 > {
2766 let BaseOpcode = "POST_"#BaseOp in {
2767 def S2_#NAME#_pi : T_store_pi <mnemonic, RC, ImmOp, MajOp, isHalf>;
2770 def S2_p#NAME#t_pi : T_pstore_pi <mnemonic, RC, ImmOp, MajOp, isHalf, 0, 0>;
2771 def S2_p#NAME#f_pi : T_pstore_pi <mnemonic, RC, ImmOp, MajOp, isHalf, 1, 0>;
2774 def S2_p#NAME#tnew_pi : T_pstore_pi <mnemonic, RC, ImmOp, MajOp,
2776 def S2_p#NAME#fnew_pi : T_pstore_pi <mnemonic, RC, ImmOp, MajOp,
2781 let accessSize = ByteAccess, isCodeGenOnly = 0 in
2782 defm storerb: ST_PostInc <"memb", "STrib", IntRegs, s4_0Imm, 0b1000>;
2784 let accessSize = HalfWordAccess, isCodeGenOnly = 0 in
2785 defm storerh: ST_PostInc <"memh", "STrih", IntRegs, s4_1Imm, 0b1010>;
2787 let accessSize = WordAccess, isCodeGenOnly = 0 in
2788 defm storeri: ST_PostInc <"memw", "STriw", IntRegs, s4_2Imm, 0b1100>;
2790 let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
2791 defm storerd: ST_PostInc <"memd", "STrid", DoubleRegs, s4_3Imm, 0b1110>;
2793 let accessSize = HalfWordAccess, isNVStorable = 0, isCodeGenOnly = 0 in
2794 defm storerf: ST_PostInc <"memh", "STrih_H", IntRegs, s4_1Imm, 0b1011, 1>;
2796 // Patterns for generating stores, where the address takes different forms:
2799 // - simple (base address without offset).
2800 // These would usually be used together (via Storex_pat defined below), but
2801 // in some cases one may want to apply different properties (such as
2802 // AddedComplexity) to the individual patterns.
2803 class Storex_fi_pat<PatFrag Store, PatFrag Value, InstHexagon MI>
2804 : Pat<(Store Value:$Rs, AddrFI:$fi), (MI AddrFI:$fi, 0, Value:$Rs)>;
2805 class Storex_add_pat<PatFrag Store, PatFrag Value, PatFrag ImmPred,
2807 : Pat<(Store Value:$Rt, (add (i32 IntRegs:$Rs), ImmPred:$Off)),
2808 (MI IntRegs:$Rs, imm:$Off, Value:$Rt)>;
2810 multiclass Storex_pat<PatFrag Store, PatFrag Value, PatLeaf ImmPred,
2812 def: Storex_fi_pat <Store, Value, MI>;
2813 def: Storex_add_pat <Store, Value, ImmPred, MI>;
2816 def : Pat<(post_truncsti8 (i32 IntRegs:$src1), IntRegs:$src2,
2817 s4_3ImmPred:$offset),
2818 (S2_storerb_pi IntRegs:$src2, s4_0ImmPred:$offset, IntRegs:$src1)>;
2820 def : Pat<(post_truncsti16 (i32 IntRegs:$src1), IntRegs:$src2,
2821 s4_3ImmPred:$offset),
2822 (S2_storerh_pi IntRegs:$src2, s4_1ImmPred:$offset, IntRegs:$src1)>;
2824 def : Pat<(post_store (i32 IntRegs:$src1), IntRegs:$src2, s4_2ImmPred:$offset),
2825 (S2_storeri_pi IntRegs:$src2, s4_1ImmPred:$offset, IntRegs:$src1)>;
2827 def : Pat<(post_store (i64 DoubleRegs:$src1), IntRegs:$src2,
2828 s4_3ImmPred:$offset),
2829 (S2_storerd_pi IntRegs:$src2, s4_3ImmPred:$offset, DoubleRegs:$src1)>;
2831 //===----------------------------------------------------------------------===//
2832 // Template class for post increment stores with register offset.
2833 //===----------------------------------------------------------------------===//
2834 let isNVStorable = 1 in
2835 class T_store_pr <string mnemonic, RegisterClass RC, bits<3> MajOp,
2836 MemAccessSize AccessSz, bit isHalf = 0>
2837 : STInst <(outs IntRegs:$_dst_),
2838 (ins IntRegs:$src1, ModRegs:$src2, RC:$src3),
2839 mnemonic#"($src1++$src2) = $src3"#!if(isHalf, ".h", ""),
2840 [], "$src1 = $_dst_" > {
2844 let accessSize = AccessSz;
2846 let IClass = 0b1010;
2848 let Inst{27-24} = 0b1101;
2849 let Inst{23-21} = MajOp;
2850 let Inst{20-16} = src1;
2851 let Inst{13} = src2;
2852 let Inst{12-8} = src3;
2856 let isCodeGenOnly = 0 in {
2857 def S2_storerb_pr : T_store_pr<"memb", IntRegs, 0b000, ByteAccess>;
2858 def S2_storerh_pr : T_store_pr<"memh", IntRegs, 0b010, HalfWordAccess>;
2859 def S2_storeri_pr : T_store_pr<"memw", IntRegs, 0b100, WordAccess>;
2860 def S2_storerd_pr : T_store_pr<"memd", DoubleRegs, 0b110, DoubleWordAccess>;
2862 def S2_storerf_pr : T_store_pr<"memh", IntRegs, 0b011, HalfWordAccess, 1>;
2864 let opExtendable = 1, isExtentSigned = 1, isPredicable = 1 in
2865 class T_store_io <string mnemonic, RegisterClass RC, Operand ImmOp,
2866 bits<3>MajOp, bit isH = 0>
2868 (ins IntRegs:$src1, ImmOp:$src2, RC:$src3),
2869 mnemonic#"($src1+#$src2) = $src3"#!if(isH,".h","")>,
2870 AddrModeRel, ImmRegRel {
2872 bits<14> src2; // Actual address offset
2874 bits<11> offsetBits; // Represents offset encoding
2876 string ImmOpStr = !cast<string>(ImmOp);
2878 let opExtentBits = !if (!eq(ImmOpStr, "s11_3Ext"), 14,
2879 !if (!eq(ImmOpStr, "s11_2Ext"), 13,
2880 !if (!eq(ImmOpStr, "s11_1Ext"), 12,
2881 /* s11_0Ext */ 11)));
2882 let offsetBits = !if (!eq(ImmOpStr, "s11_3Ext"), src2{13-3},
2883 !if (!eq(ImmOpStr, "s11_2Ext"), src2{12-2},
2884 !if (!eq(ImmOpStr, "s11_1Ext"), src2{11-1},
2885 /* s11_0Ext */ src2{10-0})));
2886 let IClass = 0b1010;
2889 let Inst{26-25} = offsetBits{10-9};
2891 let Inst{23-21} = MajOp;
2892 let Inst{20-16} = src1;
2893 let Inst{13} = offsetBits{8};
2894 let Inst{12-8} = src3;
2895 let Inst{7-0} = offsetBits{7-0};
2898 let opExtendable = 2, isPredicated = 1 in
2899 class T_pstore_io <string mnemonic, RegisterClass RC, Operand ImmOp,
2900 bits<3>MajOp, bit PredNot, bit isPredNew, bit isH = 0>
2902 (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$src3, RC:$src4),
2903 !if(PredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
2904 ") ")#mnemonic#"($src2+#$src3) = $src4"#!if(isH,".h",""),
2905 [],"",V2LDST_tc_st_SLOT01 >,
2906 AddrModeRel, ImmRegRel {
2909 bits<9> src3; // Actual address offset
2911 bits<6> offsetBits; // Represents offset encoding
2913 let isPredicatedNew = isPredNew;
2914 let isPredicatedFalse = PredNot;
2916 string ImmOpStr = !cast<string>(ImmOp);
2917 let opExtentBits = !if (!eq(ImmOpStr, "u6_3Ext"), 9,
2918 !if (!eq(ImmOpStr, "u6_2Ext"), 8,
2919 !if (!eq(ImmOpStr, "u6_1Ext"), 7,
2921 let offsetBits = !if (!eq(ImmOpStr, "u6_3Ext"), src3{8-3},
2922 !if (!eq(ImmOpStr, "u6_2Ext"), src3{7-2},
2923 !if (!eq(ImmOpStr, "u6_1Ext"), src3{6-1},
2924 /* u6_0Ext */ src3{5-0})));
2925 let IClass = 0b0100;
2928 let Inst{26} = PredNot;
2929 let Inst{25} = isPredNew;
2931 let Inst{23-21} = MajOp;
2932 let Inst{20-16} = src2;
2933 let Inst{13} = offsetBits{5};
2934 let Inst{12-8} = src4;
2935 let Inst{7-3} = offsetBits{4-0};
2936 let Inst{1-0} = src1;
2939 let isExtendable = 1, isNVStorable = 1, hasSideEffects = 0 in
2940 multiclass ST_Idxd<string mnemonic, string CextOp, RegisterClass RC,
2941 Operand ImmOp, Operand predImmOp, bits<3> MajOp, bit isH = 0> {
2942 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
2943 def S2_#NAME#_io : T_store_io <mnemonic, RC, ImmOp, MajOp, isH>;
2946 def S2_p#NAME#t_io : T_pstore_io<mnemonic, RC, predImmOp, MajOp, 0, 0, isH>;
2947 def S2_p#NAME#f_io : T_pstore_io<mnemonic, RC, predImmOp, MajOp, 1, 0, isH>;
2950 def S4_p#NAME#tnew_io : T_pstore_io <mnemonic, RC, predImmOp,
2952 def S4_p#NAME#fnew_io : T_pstore_io <mnemonic, RC, predImmOp,
2957 let addrMode = BaseImmOffset, InputType = "imm", isCodeGenOnly = 0 in {
2958 let accessSize = ByteAccess in
2959 defm storerb: ST_Idxd < "memb", "STrib", IntRegs, s11_0Ext, u6_0Ext, 0b000>;
2961 let accessSize = HalfWordAccess, opExtentAlign = 1 in
2962 defm storerh: ST_Idxd < "memh", "STrih", IntRegs, s11_1Ext, u6_1Ext, 0b010>;
2964 let accessSize = WordAccess, opExtentAlign = 2 in
2965 defm storeri: ST_Idxd < "memw", "STriw", IntRegs, s11_2Ext, u6_2Ext, 0b100>;
2967 let accessSize = DoubleWordAccess, isNVStorable = 0, opExtentAlign = 3 in
2968 defm storerd: ST_Idxd < "memd", "STrid", DoubleRegs, s11_3Ext,
2971 let accessSize = HalfWordAccess, opExtentAlign = 1 in
2972 defm storerf: ST_Idxd < "memh", "STrif", IntRegs, s11_1Ext,
2976 def : Pat<(truncstorei8 (i32 IntRegs:$src1), ADDRriS11_0:$addr),
2977 (S2_storerb_io AddrFI:$addr, 0, (i32 IntRegs:$src1))>;
2979 def : Pat<(truncstorei16 (i32 IntRegs:$src1), ADDRriS11_1:$addr),
2980 (S2_storerh_io AddrFI:$addr, 0, (i32 IntRegs:$src1))>;
2982 def : Pat<(store (i32 IntRegs:$src1), ADDRriS11_2:$addr),
2983 (S2_storeri_io AddrFI:$addr, 0, (i32 IntRegs:$src1))>;
2985 def : Pat<(store (i64 DoubleRegs:$src1), ADDRriS11_3:$addr),
2986 (S2_storerd_io AddrFI:$addr, 0, (i64 DoubleRegs:$src1))>;
2989 let AddedComplexity = 10 in {
2990 def : Pat<(truncstorei8 (i32 IntRegs:$src1), (add IntRegs:$src2,
2991 s11_0ExtPred:$offset)),
2992 (S2_storerb_io IntRegs:$src2, s11_0ImmPred:$offset,
2993 (i32 IntRegs:$src1))>;
2995 def : Pat<(truncstorei16 (i32 IntRegs:$src1), (add IntRegs:$src2,
2996 s11_1ExtPred:$offset)),
2997 (S2_storerh_io IntRegs:$src2, s11_1ImmPred:$offset,
2998 (i32 IntRegs:$src1))>;
3000 def : Pat<(store (i32 IntRegs:$src1), (add IntRegs:$src2,
3001 s11_2ExtPred:$offset)),
3002 (S2_storeri_io IntRegs:$src2, s11_2ImmPred:$offset,
3003 (i32 IntRegs:$src1))>;
3005 def : Pat<(store (i64 DoubleRegs:$src1), (add IntRegs:$src2,
3006 s11_3ExtPred:$offset)),
3007 (S2_storerd_io IntRegs:$src2, s11_3ImmPred:$offset,
3008 (i64 DoubleRegs:$src1))>;
3011 // memh(Rx++#s4:1)=Rt.H
3014 let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 13,
3015 isCodeGenOnly = 1, isPseudo = 1, hasSideEffects = 0 in
3016 def STriw_pred : STInst<(outs),
3017 (ins IntRegs:$addr, s11_2Ext:$off, PredRegs:$src1),
3018 ".error \"should not emit\"", []>;
3020 // S2_allocframe: Allocate stack frame.
3021 let Defs = [R29, R30], Uses = [R29, R31, R30],
3022 hasSideEffects = 0, accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
3023 def S2_allocframe: ST0Inst <
3024 (outs), (ins u11_3Imm:$u11_3),
3025 "allocframe(#$u11_3)" > {
3028 let IClass = 0b1010;
3029 let Inst{27-16} = 0b000010011101;
3030 let Inst{13-11} = 0b000;
3031 let Inst{10-0} = u11_3{13-3};
3034 // S2_storer[bhwdf]_pci: Store byte/half/word/double.
3035 // S2_storer[bhwdf]_pci -> S2_storerbnew_pci
3036 let Uses = [CS], isNVStorable = 1 in
3037 class T_store_pci <string mnemonic, RegisterClass RC,
3038 Operand Imm, bits<4>MajOp,
3039 MemAccessSize AlignSize, string RegSrc = "Rt">
3040 : STInst <(outs IntRegs:$_dst_),
3041 (ins IntRegs:$Rz, Imm:$offset, ModRegs:$Mu, RC:$Rt),
3042 #mnemonic#"($Rz ++ #$offset:circ($Mu)) = $"#RegSrc#"",
3049 let accessSize = AlignSize;
3051 let IClass = 0b1010;
3052 let Inst{27-25} = 0b100;
3053 let Inst{24-21} = MajOp;
3054 let Inst{20-16} = Rz;
3056 let Inst{12-8} = Rt;
3059 !if (!eq(!cast<string>(AlignSize), "DoubleWordAccess"), offset{6-3},
3060 !if (!eq(!cast<string>(AlignSize), "WordAccess"), offset{5-2},
3061 !if (!eq(!cast<string>(AlignSize), "HalfWordAccess"), offset{4-1},
3062 /* ByteAccess */ offset{3-0})));
3066 let isCodeGenOnly = 0 in {
3067 def S2_storerb_pci : T_store_pci<"memb", IntRegs, s4_0Imm, 0b1000,
3069 def S2_storerh_pci : T_store_pci<"memh", IntRegs, s4_1Imm, 0b1010,
3071 def S2_storerf_pci : T_store_pci<"memh", IntRegs, s4_1Imm, 0b1011,
3072 HalfWordAccess, "Rt.h">;
3073 def S2_storeri_pci : T_store_pci<"memw", IntRegs, s4_2Imm, 0b1100,
3075 def S2_storerd_pci : T_store_pci<"memd", DoubleRegs, s4_3Imm, 0b1110,
3079 //===----------------------------------------------------------------------===//
3080 // Circular stores with auto-increment register
3081 //===----------------------------------------------------------------------===//
3082 let Uses = [CS], isNVStorable = 1, isCodeGenOnly = 0 in
3083 class T_store_pcr <string mnemonic, RegisterClass RC, bits<4>MajOp,
3084 MemAccessSize AlignSize, string RegSrc = "Rt">
3085 : STInst <(outs IntRegs:$_dst_),
3086 (ins IntRegs:$Rz, ModRegs:$Mu, RC:$Rt),
3087 #mnemonic#"($Rz ++ I:circ($Mu)) = $"#RegSrc#"",
3094 let accessSize = AlignSize;
3096 let IClass = 0b1010;
3097 let Inst{27-25} = 0b100;
3098 let Inst{24-21} = MajOp;
3099 let Inst{20-16} = Rz;
3101 let Inst{12-8} = Rt;
3106 let isCodeGenOnly = 0 in {
3107 def S2_storerb_pcr : T_store_pcr<"memb", IntRegs, 0b1000, ByteAccess>;
3108 def S2_storerh_pcr : T_store_pcr<"memh", IntRegs, 0b1010, HalfWordAccess>;
3109 def S2_storeri_pcr : T_store_pcr<"memw", IntRegs, 0b1100, WordAccess>;
3110 def S2_storerd_pcr : T_store_pcr<"memd", DoubleRegs, 0b1110, DoubleWordAccess>;
3111 def S2_storerf_pcr : T_store_pcr<"memh", IntRegs, 0b1011,
3112 HalfWordAccess, "Rt.h">;
3115 //===----------------------------------------------------------------------===//
3116 // Bit-reversed stores with auto-increment register
3117 //===----------------------------------------------------------------------===//
3118 let hasSideEffects = 0 in
3119 class T_store_pbr<string mnemonic, RegisterClass RC,
3120 MemAccessSize addrSize, bits<3> majOp,
3123 <(outs IntRegs:$_dst_),
3124 (ins IntRegs:$Rz, ModRegs:$Mu, RC:$src),
3125 #mnemonic#"($Rz ++ $Mu:brev) = $src"#!if (!eq(isHalf, 1), ".h", ""),
3126 [], "$Rz = $_dst_" > {
3128 let accessSize = addrSize;
3134 let IClass = 0b1010;
3136 let Inst{27-24} = 0b1111;
3137 let Inst{23-21} = majOp;
3139 let Inst{20-16} = Rz;
3141 let Inst{12-8} = src;
3144 let isNVStorable = 1, isCodeGenOnly = 0 in {
3145 let BaseOpcode = "S2_storerb_pbr" in
3146 def S2_storerb_pbr : T_store_pbr<"memb", IntRegs, ByteAccess,
3147 0b000>, NewValueRel;
3148 let BaseOpcode = "S2_storerh_pbr" in
3149 def S2_storerh_pbr : T_store_pbr<"memh", IntRegs, HalfWordAccess,
3150 0b010>, NewValueRel;
3151 let BaseOpcode = "S2_storeri_pbr" in
3152 def S2_storeri_pbr : T_store_pbr<"memw", IntRegs, WordAccess,
3153 0b100>, NewValueRel;
3155 let isCodeGenOnly = 0 in {
3156 def S2_storerf_pbr : T_store_pbr<"memh", IntRegs, HalfWordAccess, 0b011, 1>;
3157 def S2_storerd_pbr : T_store_pbr<"memd", DoubleRegs, DoubleWordAccess, 0b110>;
3160 //===----------------------------------------------------------------------===//
3162 //===----------------------------------------------------------------------===//
3164 //===----------------------------------------------------------------------===//
3166 //===----------------------------------------------------------------------===//
3168 def NOT_rr64 : ALU64_rr<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1),
3169 "$dst = not($src1)",
3170 [(set (i64 DoubleRegs:$dst), (not (i64 DoubleRegs:$src1)))]>;
3173 //===----------------------------------------------------------------------===//
3175 //===----------------------------------------------------------------------===//
3177 let hasSideEffects = 0 in
3178 class T_S2op_1 <string mnemonic, bits<4> RegTyBits, RegisterClass RCOut,
3179 RegisterClass RCIn, bits<2> MajOp, bits<3> MinOp, bit isSat>
3180 : SInst <(outs RCOut:$dst), (ins RCIn:$src),
3181 "$dst = "#mnemonic#"($src)"#!if(isSat, ":sat", ""),
3182 [], "", S_2op_tc_1_SLOT23 > {
3186 let IClass = 0b1000;
3188 let Inst{27-24} = RegTyBits;
3189 let Inst{23-22} = MajOp;
3191 let Inst{20-16} = src;
3192 let Inst{7-5} = MinOp;
3193 let Inst{4-0} = dst;
3196 class T_S2op_1_di <string mnemonic, bits<2> MajOp, bits<3> MinOp>
3197 : T_S2op_1 <mnemonic, 0b0100, DoubleRegs, IntRegs, MajOp, MinOp, 0>;
3199 let hasNewValue = 1 in
3200 class T_S2op_1_id <string mnemonic, bits<2> MajOp, bits<3> MinOp, bit isSat = 0>
3201 : T_S2op_1 <mnemonic, 0b1000, IntRegs, DoubleRegs, MajOp, MinOp, isSat>;
3203 let hasNewValue = 1 in
3204 class T_S2op_1_ii <string mnemonic, bits<2> MajOp, bits<3> MinOp, bit isSat = 0>
3205 : T_S2op_1 <mnemonic, 0b1100, IntRegs, IntRegs, MajOp, MinOp, isSat>;
3207 // Sign extend word to doubleword
3208 let isCodeGenOnly = 0 in
3209 def A2_sxtw : T_S2op_1_di <"sxtw", 0b01, 0b000>;
3211 def: Pat <(i64 (sext I32:$src)), (A2_sxtw I32:$src)>;
3213 // Swizzle the bytes of a word
3214 let isCodeGenOnly = 0 in
3215 def A2_swiz : T_S2op_1_ii <"swiz", 0b10, 0b111>;
3218 let Defs = [USR_OVF], isCodeGenOnly = 0 in {
3219 def A2_sat : T_S2op_1_id <"sat", 0b11, 0b000>;
3220 def A2_satb : T_S2op_1_ii <"satb", 0b11, 0b111>;
3221 def A2_satub : T_S2op_1_ii <"satub", 0b11, 0b110>;
3222 def A2_sath : T_S2op_1_ii <"sath", 0b11, 0b100>;
3223 def A2_satuh : T_S2op_1_ii <"satuh", 0b11, 0b101>;
3226 let Itinerary = S_2op_tc_2_SLOT23, isCodeGenOnly = 0 in {
3227 // Absolute value word
3228 def A2_abs : T_S2op_1_ii <"abs", 0b10, 0b100>;
3230 let Defs = [USR_OVF] in
3231 def A2_abssat : T_S2op_1_ii <"abs", 0b10, 0b101, 1>;
3233 // Negate with saturation
3234 let Defs = [USR_OVF] in
3235 def A2_negsat : T_S2op_1_ii <"neg", 0b10, 0b110, 1>;
3238 def: Pat<(i32 (select (i1 (setlt (i32 IntRegs:$src), 0)),
3239 (i32 (sub 0, (i32 IntRegs:$src))),
3240 (i32 IntRegs:$src))),
3241 (A2_abs IntRegs:$src)>;
3243 let AddedComplexity = 50 in
3244 def: Pat<(i32 (xor (add (sra (i32 IntRegs:$src), (i32 31)),
3245 (i32 IntRegs:$src)),
3246 (sra (i32 IntRegs:$src), (i32 31)))),
3247 (A2_abs IntRegs:$src)>;
3249 class T_S2op_2 <string mnemonic, bits<4> RegTyBits, RegisterClass RCOut,
3250 RegisterClass RCIn, bits<3> MajOp, bits<3> MinOp,
3251 bit isSat, bit isRnd, list<dag> pattern = []>
3252 : SInst <(outs RCOut:$dst),
3253 (ins RCIn:$src, u5Imm:$u5),
3254 "$dst = "#mnemonic#"($src, #$u5)"#!if(isSat, ":sat", "")
3255 #!if(isRnd, ":rnd", ""),
3256 pattern, "", S_2op_tc_2_SLOT23> {
3261 let IClass = 0b1000;
3263 let Inst{27-24} = RegTyBits;
3264 let Inst{23-21} = MajOp;
3265 let Inst{20-16} = src;
3267 let Inst{12-8} = u5;
3268 let Inst{7-5} = MinOp;
3269 let Inst{4-0} = dst;
3272 class T_S2op_2_di <string mnemonic, bits<3> MajOp, bits<3> MinOp>
3273 : T_S2op_2 <mnemonic, 0b1000, DoubleRegs, IntRegs, MajOp, MinOp, 0, 0>;
3275 let hasNewValue = 1 in
3276 class T_S2op_2_id <string mnemonic, bits<3> MajOp, bits<3> MinOp>
3277 : T_S2op_2 <mnemonic, 0b1000, IntRegs, DoubleRegs, MajOp, MinOp, 0, 0>;
3279 let hasNewValue = 1 in
3280 class T_S2op_2_ii <string mnemonic, bits<3> MajOp, bits<3> MinOp,
3281 bit isSat = 0, bit isRnd = 0, list<dag> pattern = []>
3282 : T_S2op_2 <mnemonic, 0b1100, IntRegs, IntRegs, MajOp, MinOp,
3283 isSat, isRnd, pattern>;
3285 class T_S2op_shift <string mnemonic, bits<3> MajOp, bits<3> MinOp, SDNode OpNd>
3286 : T_S2op_2_ii <mnemonic, MajOp, MinOp, 0, 0,
3287 [(set (i32 IntRegs:$dst), (OpNd (i32 IntRegs:$src),
3288 (u5ImmPred:$u5)))]>;
3290 // Arithmetic/logical shift right/left by immediate
3291 let Itinerary = S_2op_tc_1_SLOT23, isCodeGenOnly = 0 in {
3292 def S2_asr_i_r : T_S2op_shift <"asr", 0b000, 0b000, sra>;
3293 def S2_lsr_i_r : T_S2op_shift <"lsr", 0b000, 0b001, srl>;
3294 def S2_asl_i_r : T_S2op_shift <"asl", 0b000, 0b010, shl>;
3297 // Shift left by immediate with saturation
3298 let Defs = [USR_OVF], isCodeGenOnly = 0 in
3299 def S2_asl_i_r_sat : T_S2op_2_ii <"asl", 0b010, 0b010, 1>;
3301 // Shift right with round
3302 let isCodeGenOnly = 0 in
3303 def S2_asr_i_r_rnd : T_S2op_2_ii <"asr", 0b010, 0b000, 0, 1>;
3305 def: Pat<(i32 (sra (i32 (add (i32 (sra I32:$src1, u5ImmPred:$src2)),
3308 (S2_asr_i_r_rnd IntRegs:$src1, u5ImmPred:$src2)>;
3310 class T_S2op_3<string opc, bits<2>MajOp, bits<3>minOp, bits<1> sat = 0>
3311 : SInst<(outs DoubleRegs:$Rdd), (ins DoubleRegs:$Rss),
3312 "$Rdd = "#opc#"($Rss)"#!if(!eq(sat, 1),":sat","")> {
3315 let IClass = 0b1000;
3316 let Inst{27-24} = 0;
3317 let Inst{23-22} = MajOp;
3318 let Inst{20-16} = Rss;
3319 let Inst{7-5} = minOp;
3320 let Inst{4-0} = Rdd;
3323 let isCodeGenOnly = 0 in {
3324 def A2_absp : T_S2op_3 <"abs", 0b10, 0b110>;
3325 def A2_negp : T_S2op_3 <"neg", 0b10, 0b101>;
3326 def A2_notp : T_S2op_3 <"not", 0b10, 0b100>;
3329 // Innterleave/deinterleave
3330 let isCodeGenOnly = 0 in {
3331 def S2_interleave : T_S2op_3 <"interleave", 0b11, 0b101>;
3332 def S2_deinterleave : T_S2op_3 <"deinterleave", 0b11, 0b100>;
3335 //===----------------------------------------------------------------------===//
3337 //===----------------------------------------------------------------------===//
3340 let hasSideEffects = 0, hasNewValue = 1 in
3341 class T_COUNT_LEADING<string MnOp, bits<3> MajOp, bits<3> MinOp, bit Is32,
3343 : SInst<Out, Inp, "$Rd = "#MnOp#"($Rs)", [], "", S_2op_tc_1_SLOT23> {
3346 let IClass = 0b1000;
3348 let Inst{26} = Is32;
3349 let Inst{25-24} = 0b00;
3350 let Inst{23-21} = MajOp;
3351 let Inst{20-16} = Rs;
3352 let Inst{7-5} = MinOp;
3356 class T_COUNT_LEADING_32<string MnOp, bits<3> MajOp, bits<3> MinOp>
3357 : T_COUNT_LEADING<MnOp, MajOp, MinOp, 0b1,
3358 (outs IntRegs:$Rd), (ins IntRegs:$Rs)>;
3360 class T_COUNT_LEADING_64<string MnOp, bits<3> MajOp, bits<3> MinOp>
3361 : T_COUNT_LEADING<MnOp, MajOp, MinOp, 0b0,
3362 (outs IntRegs:$Rd), (ins DoubleRegs:$Rs)>;
3364 let isCodeGenOnly = 0 in {
3365 def S2_cl0 : T_COUNT_LEADING_32<"cl0", 0b000, 0b101>;
3366 def S2_cl1 : T_COUNT_LEADING_32<"cl1", 0b000, 0b110>;
3367 def S2_ct0 : T_COUNT_LEADING_32<"ct0", 0b010, 0b100>;
3368 def S2_ct1 : T_COUNT_LEADING_32<"ct1", 0b010, 0b101>;
3369 def S2_cl0p : T_COUNT_LEADING_64<"cl0", 0b010, 0b010>;
3370 def S2_cl1p : T_COUNT_LEADING_64<"cl1", 0b010, 0b100>;
3371 def S2_clb : T_COUNT_LEADING_32<"clb", 0b000, 0b100>;
3372 def S2_clbp : T_COUNT_LEADING_64<"clb", 0b010, 0b000>;
3373 def S2_clbnorm : T_COUNT_LEADING_32<"normamt", 0b000, 0b111>;
3376 def: Pat<(i32 (ctlz I32:$Rs)), (S2_cl0 I32:$Rs)>;
3377 def: Pat<(i32 (ctlz (not I32:$Rs))), (S2_cl1 I32:$Rs)>;
3378 def: Pat<(i32 (cttz I32:$Rs)), (S2_ct0 I32:$Rs)>;
3379 def: Pat<(i32 (cttz (not I32:$Rs))), (S2_ct1 I32:$Rs)>;
3380 def: Pat<(i32 (trunc (ctlz I64:$Rss))), (S2_cl0p I64:$Rss)>;
3381 def: Pat<(i32 (trunc (ctlz (not I64:$Rss)))), (S2_cl1p I64:$Rss)>;
3383 // Bit set/clear/toggle
3385 let hasSideEffects = 0, hasNewValue = 1 in
3386 class T_SCT_BIT_IMM<string MnOp, bits<3> MinOp>
3387 : SInst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, u5Imm:$u5),
3388 "$Rd = "#MnOp#"($Rs, #$u5)", [], "", S_2op_tc_1_SLOT23> {
3392 let IClass = 0b1000;
3393 let Inst{27-21} = 0b1100110;
3394 let Inst{20-16} = Rs;
3396 let Inst{12-8} = u5;
3397 let Inst{7-5} = MinOp;
3401 let hasSideEffects = 0, hasNewValue = 1 in
3402 class T_SCT_BIT_REG<string MnOp, bits<2> MinOp>
3403 : SInst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
3404 "$Rd = "#MnOp#"($Rs, $Rt)", [], "", S_3op_tc_1_SLOT23> {
3408 let IClass = 0b1100;
3409 let Inst{27-22} = 0b011010;
3410 let Inst{20-16} = Rs;
3411 let Inst{12-8} = Rt;
3412 let Inst{7-6} = MinOp;
3416 let isCodeGenOnly = 0 in {
3417 def S2_clrbit_i : T_SCT_BIT_IMM<"clrbit", 0b001>;
3418 def S2_setbit_i : T_SCT_BIT_IMM<"setbit", 0b000>;
3419 def S2_togglebit_i : T_SCT_BIT_IMM<"togglebit", 0b010>;
3420 def S2_clrbit_r : T_SCT_BIT_REG<"clrbit", 0b01>;
3421 def S2_setbit_r : T_SCT_BIT_REG<"setbit", 0b00>;
3422 def S2_togglebit_r : T_SCT_BIT_REG<"togglebit", 0b10>;
3425 def: Pat<(i32 (and (i32 IntRegs:$Rs), (not (shl 1, u5ImmPred:$u5)))),
3426 (S2_clrbit_i IntRegs:$Rs, u5ImmPred:$u5)>;
3427 def: Pat<(i32 (or (i32 IntRegs:$Rs), (shl 1, u5ImmPred:$u5))),
3428 (S2_setbit_i IntRegs:$Rs, u5ImmPred:$u5)>;
3429 def: Pat<(i32 (xor (i32 IntRegs:$Rs), (shl 1, u5ImmPred:$u5))),
3430 (S2_togglebit_i IntRegs:$Rs, u5ImmPred:$u5)>;
3431 def: Pat<(i32 (and (i32 IntRegs:$Rs), (not (shl 1, (i32 IntRegs:$Rt))))),
3432 (S2_clrbit_r IntRegs:$Rs, IntRegs:$Rt)>;
3433 def: Pat<(i32 (or (i32 IntRegs:$Rs), (shl 1, (i32 IntRegs:$Rt)))),
3434 (S2_setbit_r IntRegs:$Rs, IntRegs:$Rt)>;
3435 def: Pat<(i32 (xor (i32 IntRegs:$Rs), (shl 1, (i32 IntRegs:$Rt)))),
3436 (S2_togglebit_r IntRegs:$Rs, IntRegs:$Rt)>;
3440 let hasSideEffects = 0 in
3441 class T_TEST_BIT_IMM<string MnOp, bits<3> MajOp>
3442 : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, u5Imm:$u5),
3443 "$Pd = "#MnOp#"($Rs, #$u5)",
3444 [], "", S_2op_tc_2early_SLOT23> {
3448 let IClass = 0b1000;
3449 let Inst{27-24} = 0b0101;
3450 let Inst{23-21} = MajOp;
3451 let Inst{20-16} = Rs;
3453 let Inst{12-8} = u5;
3457 let hasSideEffects = 0 in
3458 class T_TEST_BIT_REG<string MnOp, bit IsNeg>
3459 : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt),
3460 "$Pd = "#MnOp#"($Rs, $Rt)",
3461 [], "", S_3op_tc_2early_SLOT23> {
3465 let IClass = 0b1100;
3466 let Inst{27-22} = 0b011100;
3467 let Inst{21} = IsNeg;
3468 let Inst{20-16} = Rs;
3469 let Inst{12-8} = Rt;
3473 let isCodeGenOnly = 0 in {
3474 def S2_tstbit_i : T_TEST_BIT_IMM<"tstbit", 0b000>;
3475 def S2_tstbit_r : T_TEST_BIT_REG<"tstbit", 0>;
3478 let AddedComplexity = 20 in { // Complexity greater than cmp reg-imm.
3479 def: Pat<(i1 (setne (and (shl 1, u5ImmPred:$u5), (i32 IntRegs:$Rs)), 0)),
3480 (S2_tstbit_i IntRegs:$Rs, u5ImmPred:$u5)>;
3481 def: Pat<(i1 (setne (and (shl 1, (i32 IntRegs:$Rt)), (i32 IntRegs:$Rs)), 0)),
3482 (S2_tstbit_r IntRegs:$Rs, IntRegs:$Rt)>;
3483 def: Pat<(i1 (trunc (i32 IntRegs:$Rs))),
3484 (S2_tstbit_i IntRegs:$Rs, 0)>;
3485 def: Pat<(i1 (trunc (i64 DoubleRegs:$Rs))),
3486 (S2_tstbit_i (LoReg DoubleRegs:$Rs), 0)>;
3488 let hasSideEffects = 0 in
3489 class T_TEST_BITS_IMM<string MnOp, bits<2> MajOp, bit IsNeg>
3490 : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, u6Imm:$u6),
3491 "$Pd = "#MnOp#"($Rs, #$u6)",
3492 [], "", S_2op_tc_2early_SLOT23> {
3496 let IClass = 0b1000;
3497 let Inst{27-24} = 0b0101;
3498 let Inst{23-22} = MajOp;
3499 let Inst{21} = IsNeg;
3500 let Inst{20-16} = Rs;
3501 let Inst{13-8} = u6;
3505 let hasSideEffects = 0 in
3506 class T_TEST_BITS_REG<string MnOp, bits<2> MajOp, bit IsNeg>
3507 : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt),
3508 "$Pd = "#MnOp#"($Rs, $Rt)",
3509 [], "", S_3op_tc_2early_SLOT23> {
3513 let IClass = 0b1100;
3514 let Inst{27-24} = 0b0111;
3515 let Inst{23-22} = MajOp;
3516 let Inst{21} = IsNeg;
3517 let Inst{20-16} = Rs;
3518 let Inst{12-8} = Rt;
3522 let isCodeGenOnly = 0 in {
3523 def C2_bitsclri : T_TEST_BITS_IMM<"bitsclr", 0b10, 0>;
3524 def C2_bitsclr : T_TEST_BITS_REG<"bitsclr", 0b10, 0>;
3525 def C2_bitsset : T_TEST_BITS_REG<"bitsset", 0b01, 0>;
3528 let AddedComplexity = 20 in { // Complexity greater than compare reg-imm.
3529 def: Pat<(i1 (seteq (and (i32 IntRegs:$Rs), u6ImmPred:$u6), 0)),
3530 (C2_bitsclri IntRegs:$Rs, u6ImmPred:$u6)>;
3531 def: Pat<(i1 (seteq (and (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), 0)),
3532 (C2_bitsclr IntRegs:$Rs, IntRegs:$Rt)>;
3535 let AddedComplexity = 10 in // Complexity greater than compare reg-reg.
3536 def: Pat<(i1 (seteq (and (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), IntRegs:$Rt)),
3537 (C2_bitsset IntRegs:$Rs, IntRegs:$Rt)>;
3539 //===----------------------------------------------------------------------===//
3541 //===----------------------------------------------------------------------===//
3543 //===----------------------------------------------------------------------===//
3545 //===----------------------------------------------------------------------===//
3546 //===----------------------------------------------------------------------===//
3548 //===----------------------------------------------------------------------===//
3550 //===----------------------------------------------------------------------===//
3552 //===----------------------------------------------------------------------===//
3554 //===----------------------------------------------------------------------===//
3556 //===----------------------------------------------------------------------===//
3558 //===----------------------------------------------------------------------===//
3560 //===----------------------------------------------------------------------===//
3562 // Predicate transfer.
3563 let hasSideEffects = 0, hasNewValue = 1, isCodeGenOnly = 0 in
3564 def C2_tfrpr : SInst<(outs IntRegs:$Rd), (ins PredRegs:$Ps),
3565 "$Rd = $Ps", [], "", S_2op_tc_1_SLOT23> {
3569 let IClass = 0b1000;
3570 let Inst{27-24} = 0b1001;
3572 let Inst{17-16} = Ps;
3576 // Transfer general register to predicate.
3577 let hasSideEffects = 0, isCodeGenOnly = 0 in
3578 def C2_tfrrp: SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs),
3579 "$Pd = $Rs", [], "", S_2op_tc_2early_SLOT23> {
3583 let IClass = 0b1000;
3584 let Inst{27-21} = 0b0101010;
3585 let Inst{20-16} = Rs;
3590 //===----------------------------------------------------------------------===//
3592 //===----------------------------------------------------------------------===//
3594 //===----------------------------------------------------------------------===//
3596 //===----------------------------------------------------------------------===//
3597 class S_2OpInstImm<string Mnemonic, bits<3>MajOp, bits<3>MinOp,
3598 Operand Imm, list<dag> pattern = [], bit isRnd = 0>
3599 : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, Imm:$src2),
3600 "$dst = "#Mnemonic#"($src1, #$src2)"#!if(isRnd, ":rnd", ""),
3604 let IClass = 0b1000;
3605 let Inst{27-24} = 0;
3606 let Inst{23-21} = MajOp;
3607 let Inst{20-16} = src1;
3608 let Inst{7-5} = MinOp;
3609 let Inst{4-0} = dst;
3612 class S_2OpInstImmI6<string Mnemonic, SDNode OpNode, bits<3>MinOp>
3613 : S_2OpInstImm<Mnemonic, 0b000, MinOp, u6Imm,
3614 [(set (i64 DoubleRegs:$dst), (OpNode (i64 DoubleRegs:$src1),
3615 u6ImmPred:$src2))]> {
3617 let Inst{13-8} = src2;
3620 // Shift by immediate.
3621 let isCodeGenOnly = 0 in {
3622 def S2_asr_i_p : S_2OpInstImmI6<"asr", sra, 0b000>;
3623 def S2_asl_i_p : S_2OpInstImmI6<"asl", shl, 0b010>;
3624 def S2_lsr_i_p : S_2OpInstImmI6<"lsr", srl, 0b001>;
3627 // Shift left by small amount and add.
3628 let AddedComplexity = 100, hasNewValue = 1, hasSideEffects = 0,
3629 isCodeGenOnly = 0 in
3630 def S2_addasl_rrri: SInst <(outs IntRegs:$Rd),
3631 (ins IntRegs:$Rt, IntRegs:$Rs, u3Imm:$u3),
3632 "$Rd = addasl($Rt, $Rs, #$u3)" ,
3633 [(set (i32 IntRegs:$Rd), (add (i32 IntRegs:$Rt),
3634 (shl (i32 IntRegs:$Rs), u3ImmPred:$u3)))],
3635 "", S_3op_tc_2_SLOT23> {
3641 let IClass = 0b1100;
3643 let Inst{27-21} = 0b0100000;
3644 let Inst{20-16} = Rs;
3646 let Inst{12-8} = Rt;
3651 //===----------------------------------------------------------------------===//
3653 //===----------------------------------------------------------------------===//
3655 //===----------------------------------------------------------------------===//
3657 //===----------------------------------------------------------------------===//
3658 //===----------------------------------------------------------------------===//
3660 //===----------------------------------------------------------------------===//
3662 //===----------------------------------------------------------------------===//
3664 //===----------------------------------------------------------------------===//
3665 //===----------------------------------------------------------------------===//
3667 //===----------------------------------------------------------------------===//
3669 //===----------------------------------------------------------------------===//
3671 //===----------------------------------------------------------------------===//
3673 //===----------------------------------------------------------------------===//
3675 //===----------------------------------------------------------------------===//
3676 def HexagonBARRIER: SDNode<"HexagonISD::BARRIER", SDTNone, [SDNPHasChain]>;
3678 let hasSideEffects = 1, isSoloAX = 1, isCodeGenOnly = 0 in
3679 def BARRIER : SYSInst<(outs), (ins),
3681 [(HexagonBARRIER)],"",ST_tc_st_SLOT0> {
3682 let Inst{31-28} = 0b1010;
3683 let Inst{27-21} = 0b1000000;
3686 //===----------------------------------------------------------------------===//
3688 //===----------------------------------------------------------------------===//
3689 //===----------------------------------------------------------------------===//
3691 //===----------------------------------------------------------------------===//
3693 let isExtendable = 1, isExtentSigned = 1, opExtentBits = 9, opExtentAlign = 2,
3694 opExtendable = 0, hasSideEffects = 0 in
3695 class LOOP_iBase<string mnemonic, Operand brOp, bit mustExtend = 0>
3696 : CRInst<(outs), (ins brOp:$offset, u10Imm:$src2),
3697 #mnemonic#"($offset, #$src2)",
3698 [], "" , CR_tc_3x_SLOT3> {
3702 let IClass = 0b0110;
3704 let Inst{27-22} = 0b100100;
3705 let Inst{21} = !if (!eq(mnemonic, "loop0"), 0b0, 0b1);
3706 let Inst{20-16} = src2{9-5};
3707 let Inst{12-8} = offset{8-4};
3708 let Inst{7-5} = src2{4-2};
3709 let Inst{4-3} = offset{3-2};
3710 let Inst{1-0} = src2{1-0};
3713 let isExtendable = 1, isExtentSigned = 1, opExtentBits = 9, opExtentAlign = 2,
3714 opExtendable = 0, hasSideEffects = 0 in
3715 class LOOP_rBase<string mnemonic, Operand brOp, bit mustExtend = 0>
3716 : CRInst<(outs), (ins brOp:$offset, IntRegs:$src2),
3717 #mnemonic#"($offset, $src2)",
3718 [], "" ,CR_tc_3x_SLOT3> {
3722 let IClass = 0b0110;
3724 let Inst{27-22} = 0b000000;
3725 let Inst{21} = !if (!eq(mnemonic, "loop0"), 0b0, 0b1);
3726 let Inst{20-16} = src2;
3727 let Inst{12-8} = offset{8-4};
3728 let Inst{4-3} = offset{3-2};
3731 multiclass LOOP_ri<string mnemonic> {
3732 def i : LOOP_iBase<mnemonic, brtarget>;
3733 def r : LOOP_rBase<mnemonic, brtarget>;
3737 let Defs = [SA0, LC0, USR], isCodeGenOnly = 0 in
3738 defm J2_loop0 : LOOP_ri<"loop0">;
3740 // Interestingly only loop0's appear to set usr.lpcfg
3741 let Defs = [SA1, LC1], isCodeGenOnly = 0 in
3742 defm J2_loop1 : LOOP_ri<"loop1">;
3744 let isBranch = 1, isTerminator = 1, hasSideEffects = 0,
3745 Defs = [PC, LC0], Uses = [SA0, LC0] in {
3746 def ENDLOOP0 : Endloop<(outs), (ins brtarget:$offset),
3751 let isBranch = 1, isTerminator = 1, hasSideEffects = 0,
3752 Defs = [PC, LC1], Uses = [SA1, LC1] in {
3753 def ENDLOOP1 : Endloop<(outs), (ins brtarget:$offset),
3758 // Pipelined loop instructions, sp[123]loop0
3759 let Defs = [LC0, SA0, P3, USR], hasSideEffects = 0,
3760 isExtentSigned = 1, isExtendable = 1, opExtentBits = 9, opExtentAlign = 2,
3761 opExtendable = 0, isPredicateLate = 1 in
3762 class SPLOOP_iBase<string SP, bits<2> op>
3763 : CRInst <(outs), (ins brtarget:$r7_2, u10Imm:$U10),
3764 "p3 = sp"#SP#"loop0($r7_2, #$U10)" > {
3768 let IClass = 0b0110;
3770 let Inst{22-21} = op;
3771 let Inst{27-23} = 0b10011;
3772 let Inst{20-16} = U10{9-5};
3773 let Inst{12-8} = r7_2{8-4};
3774 let Inst{7-5} = U10{4-2};
3775 let Inst{4-3} = r7_2{3-2};
3776 let Inst{1-0} = U10{1-0};
3779 let Defs = [LC0, SA0, P3, USR], hasSideEffects = 0,
3780 isExtentSigned = 1, isExtendable = 1, opExtentBits = 9, opExtentAlign = 2,
3781 opExtendable = 0, isPredicateLate = 1 in
3782 class SPLOOP_rBase<string SP, bits<2> op>
3783 : CRInst <(outs), (ins brtarget:$r7_2, IntRegs:$Rs),
3784 "p3 = sp"#SP#"loop0($r7_2, $Rs)" > {
3788 let IClass = 0b0110;
3790 let Inst{22-21} = op;
3791 let Inst{27-23} = 0b00001;
3792 let Inst{20-16} = Rs;
3793 let Inst{12-8} = r7_2{8-4};
3794 let Inst{4-3} = r7_2{3-2};
3797 multiclass SPLOOP_ri<string mnemonic, bits<2> op> {
3798 def i : SPLOOP_iBase<mnemonic, op>;
3799 def r : SPLOOP_rBase<mnemonic, op>;
3802 let isCodeGenOnly = 0 in {
3803 defm J2_ploop1s : SPLOOP_ri<"1", 0b01>;
3804 defm J2_ploop2s : SPLOOP_ri<"2", 0b10>;
3805 defm J2_ploop3s : SPLOOP_ri<"3", 0b11>;
3808 // Transfer to/from Control/GPR Guest/GPR
3809 let hasSideEffects = 0 in
3810 class TFR_CR_RS_base<RegisterClass CTRC, RegisterClass RC, bit isDouble>
3811 : CRInst <(outs CTRC:$dst), (ins RC:$src),
3812 "$dst = $src", [], "", CR_tc_3x_SLOT3> {
3816 let IClass = 0b0110;
3818 let Inst{27-25} = 0b001;
3819 let Inst{24} = isDouble;
3820 let Inst{23-21} = 0b001;
3821 let Inst{20-16} = src;
3822 let Inst{4-0} = dst;
3824 let isCodeGenOnly = 0 in
3825 def A2_tfrrcr : TFR_CR_RS_base<CtrRegs, IntRegs, 0b0>;
3826 def : InstAlias<"m0 = $Rs", (A2_tfrrcr C6, IntRegs:$Rs)>;
3827 def : InstAlias<"m1 = $Rs", (A2_tfrrcr C7, IntRegs:$Rs)>;
3829 let hasSideEffects = 0 in
3830 class TFR_RD_CR_base<RegisterClass RC, RegisterClass CTRC, bit isSingle>
3831 : CRInst <(outs RC:$dst), (ins CTRC:$src),
3832 "$dst = $src", [], "", CR_tc_3x_SLOT3> {
3836 let IClass = 0b0110;
3838 let Inst{27-26} = 0b10;
3839 let Inst{25} = isSingle;
3840 let Inst{24-21} = 0b0000;
3841 let Inst{20-16} = src;
3842 let Inst{4-0} = dst;
3845 let hasNewValue = 1, opNewValue = 0, isCodeGenOnly = 0 in
3846 def A2_tfrcrr : TFR_RD_CR_base<IntRegs, CtrRegs, 1>;
3847 def : InstAlias<"$Rd = m0", (A2_tfrcrr IntRegs:$Rd, C6)>;
3848 def : InstAlias<"$Rd = m1", (A2_tfrcrr IntRegs:$Rd, C7)>;
3850 // Y4_trace: Send value to etm trace.
3851 let isSoloAX = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
3852 def Y4_trace: CRInst <(outs), (ins IntRegs:$Rs),
3856 let IClass = 0b0110;
3857 let Inst{27-21} = 0b0010010;
3858 let Inst{20-16} = Rs;
3861 let AddedComplexity = 100, isPredicated = 1 in
3862 def TFR_condset_ri : ALU32_rr<(outs IntRegs:$dst),
3863 (ins PredRegs:$src1, IntRegs:$src2, s12Imm:$src3),
3864 "Error; should not emit",
3865 [(set (i32 IntRegs:$dst),
3866 (i32 (select (i1 PredRegs:$src1), (i32 IntRegs:$src2),
3867 s12ImmPred:$src3)))]>;
3869 let AddedComplexity = 100, isPredicated = 1 in
3870 def TFR_condset_ir : ALU32_rr<(outs IntRegs:$dst),
3871 (ins PredRegs:$src1, s12Imm:$src2, IntRegs:$src3),
3872 "Error; should not emit",
3873 [(set (i32 IntRegs:$dst),
3874 (i32 (select (i1 PredRegs:$src1), s12ImmPred:$src2,
3875 (i32 IntRegs:$src3))))]>;
3877 let AddedComplexity = 100, isPredicated = 1 in
3878 def TFR_condset_ii : ALU32_rr<(outs IntRegs:$dst),
3879 (ins PredRegs:$src1, s12Imm:$src2, s12Imm:$src3),
3880 "Error; should not emit",
3881 [(set (i32 IntRegs:$dst),
3882 (i32 (select (i1 PredRegs:$src1), s12ImmPred:$src2,
3883 s12ImmPred:$src3)))]>;
3885 // Generate frameindex addresses.
3886 let isReMaterializable = 1 in
3887 def TFR_FI : ALU32_ri<(outs IntRegs:$dst), (ins FrameIndex:$src1),
3888 "$dst = add($src1)",
3889 [(set (i32 IntRegs:$dst), ADDRri:$src1)]>;
3891 // Support for generating global address.
3892 // Taken from X86InstrInfo.td.
3893 def SDTHexagonCONST32 : SDTypeProfile<1, 1, [
3897 def HexagonCONST32 : SDNode<"HexagonISD::CONST32", SDTHexagonCONST32>;
3898 def HexagonCONST32_GP : SDNode<"HexagonISD::CONST32_GP", SDTHexagonCONST32>;
3900 // HI/LO Instructions
3901 let isReMaterializable = 1, isMoveImm = 1, hasSideEffects = 0 in
3902 def LO : ALU32_ri<(outs IntRegs:$dst), (ins globaladdress:$global),
3903 "$dst.l = #LO($global)",
3906 let isReMaterializable = 1, isMoveImm = 1, hasSideEffects = 0 in
3907 def HI : ALU32_ri<(outs IntRegs:$dst), (ins globaladdress:$global),
3908 "$dst.h = #HI($global)",
3911 let isReMaterializable = 1, isMoveImm = 1, hasSideEffects = 0 in
3912 def LOi : ALU32_ri<(outs IntRegs:$dst), (ins i32imm:$imm_value),
3913 "$dst.l = #LO($imm_value)",
3917 let isReMaterializable = 1, isMoveImm = 1, hasSideEffects = 0 in
3918 def HIi : ALU32_ri<(outs IntRegs:$dst), (ins i32imm:$imm_value),
3919 "$dst.h = #HI($imm_value)",
3922 let isReMaterializable = 1, isMoveImm = 1, hasSideEffects = 0 in
3923 def LO_jt : ALU32_ri<(outs IntRegs:$dst), (ins jumptablebase:$jt),
3924 "$dst.l = #LO($jt)",
3927 let isReMaterializable = 1, isMoveImm = 1, hasSideEffects = 0 in
3928 def HI_jt : ALU32_ri<(outs IntRegs:$dst), (ins jumptablebase:$jt),
3929 "$dst.h = #HI($jt)",
3933 let isReMaterializable = 1, isMoveImm = 1, hasSideEffects = 0 in
3934 def LO_label : ALU32_ri<(outs IntRegs:$dst), (ins bblabel:$label),
3935 "$dst.l = #LO($label)",
3938 let isReMaterializable = 1, isMoveImm = 1 , hasSideEffects = 0 in
3939 def HI_label : ALU32_ri<(outs IntRegs:$dst), (ins bblabel:$label),
3940 "$dst.h = #HI($label)",
3943 // This pattern is incorrect. When we add small data, we should change
3944 // this pattern to use memw(#foo).
3945 // This is for sdata.
3946 let isMoveImm = 1 in
3947 def CONST32 : LDInst<(outs IntRegs:$dst), (ins globaladdress:$global),
3948 "$dst = CONST32(#$global)",
3949 [(set (i32 IntRegs:$dst),
3950 (load (HexagonCONST32 tglobaltlsaddr:$global)))]>;
3952 // This is for non-sdata.
3953 let isReMaterializable = 1, isMoveImm = 1 in
3954 def CONST32_set : LDInst2<(outs IntRegs:$dst), (ins globaladdress:$global),
3955 "$dst = CONST32(#$global)",
3956 [(set (i32 IntRegs:$dst),
3957 (HexagonCONST32 tglobaladdr:$global))]>;
3959 let isReMaterializable = 1, isMoveImm = 1 in
3960 def CONST32_set_jt : LDInst2<(outs IntRegs:$dst), (ins jumptablebase:$jt),
3961 "$dst = CONST32(#$jt)",
3962 [(set (i32 IntRegs:$dst),
3963 (HexagonCONST32 tjumptable:$jt))]>;
3965 let isReMaterializable = 1, isMoveImm = 1 in
3966 def CONST32GP_set : LDInst2<(outs IntRegs:$dst), (ins globaladdress:$global),
3967 "$dst = CONST32(#$global)",
3968 [(set (i32 IntRegs:$dst),
3969 (HexagonCONST32_GP tglobaladdr:$global))]>;
3971 let isReMaterializable = 1, isMoveImm = 1 in
3972 def CONST32_Int_Real : LDInst2<(outs IntRegs:$dst), (ins i32imm:$global),
3973 "$dst = CONST32(#$global)",
3974 [(set (i32 IntRegs:$dst), imm:$global) ]>;
3976 // Map BlockAddress lowering to CONST32_Int_Real
3977 def : Pat<(HexagonCONST32_GP tblockaddress:$addr),
3978 (CONST32_Int_Real tblockaddress:$addr)>;
3980 let isReMaterializable = 1, isMoveImm = 1 in
3981 def CONST32_Label : LDInst2<(outs IntRegs:$dst), (ins bblabel:$label),
3982 "$dst = CONST32($label)",
3983 [(set (i32 IntRegs:$dst), (HexagonCONST32 bbl:$label))]>;
3985 let isReMaterializable = 1, isMoveImm = 1 in
3986 def CONST64_Int_Real : LDInst2<(outs DoubleRegs:$dst), (ins i64imm:$global),
3987 "$dst = CONST64(#$global)",
3988 [(set (i64 DoubleRegs:$dst), imm:$global) ]>;
3990 def TFR_PdFalse : SInst<(outs PredRegs:$dst), (ins),
3991 "$dst = xor($dst, $dst)",
3992 [(set (i1 PredRegs:$dst), 0)]>;
3994 def MPY_trsext : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
3995 "$dst = mpy($src1, $src2)",
3996 [(set (i32 IntRegs:$dst),
3997 (trunc (i64 (srl (i64 (mul (i64 (sext (i32 IntRegs:$src1))),
3998 (i64 (sext (i32 IntRegs:$src2))))),
4001 // Pseudo instructions.
4002 def SDT_SPCallSeqStart : SDCallSeqStart<[ SDTCisVT<0, i32> ]>;
4004 def SDT_SPCallSeqEnd : SDCallSeqEnd<[ SDTCisVT<0, i32>,
4005 SDTCisVT<1, i32> ]>;
4007 def callseq_end : SDNode<"ISD::CALLSEQ_END", SDT_SPCallSeqEnd,
4008 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
4010 def callseq_start : SDNode<"ISD::CALLSEQ_START", SDT_SPCallSeqStart,
4011 [SDNPHasChain, SDNPOutGlue]>;
4013 def SDT_SPCall : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
4015 def call : SDNode<"HexagonISD::CALL", SDT_SPCall,
4016 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue, SDNPVariadic]>;
4018 // For tailcalls a HexagonTCRet SDNode has 3 SDNode Properties - a chain,
4019 // Optional Flag and Variable Arguments.
4020 // Its 1 Operand has pointer type.
4021 def HexagonTCRet : SDNode<"HexagonISD::TC_RETURN", SDT_SPCall,
4022 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
4024 let Defs = [R29, R30], Uses = [R31, R30, R29] in {
4025 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
4026 "Should never be emitted",
4027 [(callseq_start timm:$amt)]>;
4030 let Defs = [R29, R30, R31], Uses = [R29] in {
4031 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
4032 "Should never be emitted",
4033 [(callseq_end timm:$amt1, timm:$amt2)]>;
4036 let isCall = 1, hasSideEffects = 0,
4037 Defs = [D0, D1, D2, D3, D4, D5, D6, D7, D8, D9, D10,
4038 R22, R23, R28, R31, P0, P1, P2, P3, LC0, LC1, SA0, SA1] in {
4039 def CALL : JInst<(outs), (ins calltarget:$dst),
4043 // Call subroutine indirectly.
4044 let Defs = VolatileV3.Regs, isCodeGenOnly = 0 in
4045 def J2_callr : JUMPR_MISC_CALLR<0, 1>;
4047 // Indirect tail-call.
4048 let isCodeGenOnly = 1, isCall = 1, isReturn = 1 in
4049 def TCRETURNR : T_JMPr;
4051 // Direct tail-calls.
4052 let isCall = 1, isReturn = 1, isBarrier = 1, isPredicable = 0,
4053 isTerminator = 1, isCodeGenOnly = 1 in {
4054 def TCRETURNtg : JInst<(outs), (ins calltarget:$dst), "jump $dst",
4055 [], "", J_tc_2early_SLOT23>;
4056 def TCRETURNtext : JInst<(outs), (ins calltarget:$dst), "jump $dst",
4057 [], "", J_tc_2early_SLOT23>;
4060 // Map call instruction.
4061 def : Pat<(call (i32 IntRegs:$dst)),
4062 (J2_callr (i32 IntRegs:$dst))>, Requires<[HasV2TOnly]>;
4063 def : Pat<(call tglobaladdr:$dst),
4064 (CALL tglobaladdr:$dst)>, Requires<[HasV2TOnly]>;
4065 def : Pat<(call texternalsym:$dst),
4066 (CALL texternalsym:$dst)>, Requires<[HasV2TOnly]>;
4068 def : Pat<(HexagonTCRet tglobaladdr:$dst),
4069 (TCRETURNtg tglobaladdr:$dst)>;
4070 def : Pat<(HexagonTCRet texternalsym:$dst),
4071 (TCRETURNtext texternalsym:$dst)>;
4072 def : Pat<(HexagonTCRet (i32 IntRegs:$dst)),
4073 (TCRETURNR (i32 IntRegs:$dst))>;
4075 // Atomic load and store support
4076 // 8 bit atomic load
4077 def : Pat<(atomic_load_8 ADDRriS11_0:$src1),
4078 (i32 (L2_loadrub_io AddrFI:$src1, 0))>;
4080 def : Pat<(atomic_load_8 (add (i32 IntRegs:$src1), s11_0ImmPred:$offset)),
4081 (i32 (L2_loadrub_io (i32 IntRegs:$src1), s11_0ImmPred:$offset))>;
4083 // 16 bit atomic load
4084 def : Pat<(atomic_load_16 ADDRriS11_1:$src1),
4085 (i32 (L2_loadruh_io AddrFI:$src1, 0))>;
4087 def : Pat<(atomic_load_16 (add (i32 IntRegs:$src1), s11_1ImmPred:$offset)),
4088 (i32 (L2_loadruh_io (i32 IntRegs:$src1), s11_1ImmPred:$offset))>;
4090 def : Pat<(atomic_load_32 ADDRriS11_2:$src1),
4091 (i32 (L2_loadri_io AddrFI:$src1, 0))>;
4093 def : Pat<(atomic_load_32 (add (i32 IntRegs:$src1), s11_2ImmPred:$offset)),
4094 (i32 (L2_loadri_io (i32 IntRegs:$src1), s11_2ImmPred:$offset))>;
4096 // 64 bit atomic load
4097 def : Pat<(atomic_load_64 ADDRriS11_3:$src1),
4098 (i64 (L2_loadrd_io AddrFI:$src1, 0))>;
4100 def : Pat<(atomic_load_64 (add (i32 IntRegs:$src1), s11_3ImmPred:$offset)),
4101 (i64 (L2_loadrd_io (i32 IntRegs:$src1), s11_3ImmPred:$offset))>;
4104 def : Pat<(atomic_store_8 ADDRriS11_0:$src2, (i32 IntRegs:$src1)),
4105 (S2_storerb_io AddrFI:$src2, 0, (i32 IntRegs:$src1))>;
4107 def : Pat<(atomic_store_8 (add (i32 IntRegs:$src2), s11_0ImmPred:$offset),
4108 (i32 IntRegs:$src1)),
4109 (S2_storerb_io (i32 IntRegs:$src2), s11_0ImmPred:$offset,
4110 (i32 IntRegs:$src1))>;
4113 def : Pat<(atomic_store_16 ADDRriS11_1:$src2, (i32 IntRegs:$src1)),
4114 (S2_storerh_io AddrFI:$src2, 0, (i32 IntRegs:$src1))>;
4116 def : Pat<(atomic_store_16 (i32 IntRegs:$src1),
4117 (add (i32 IntRegs:$src2), s11_1ImmPred:$offset)),
4118 (S2_storerh_io (i32 IntRegs:$src2), s11_1ImmPred:$offset,
4119 (i32 IntRegs:$src1))>;
4121 def : Pat<(atomic_store_32 ADDRriS11_2:$src2, (i32 IntRegs:$src1)),
4122 (S2_storeri_io AddrFI:$src2, 0, (i32 IntRegs:$src1))>;
4124 def : Pat<(atomic_store_32 (add (i32 IntRegs:$src2), s11_2ImmPred:$offset),
4125 (i32 IntRegs:$src1)),
4126 (S2_storeri_io (i32 IntRegs:$src2), s11_2ImmPred:$offset,
4127 (i32 IntRegs:$src1))>;
4132 def : Pat<(atomic_store_64 ADDRriS11_3:$src2, (i64 DoubleRegs:$src1)),
4133 (S2_storerd_io AddrFI:$src2, 0, (i64 DoubleRegs:$src1))>;
4135 def : Pat<(atomic_store_64 (add (i32 IntRegs:$src2), s11_3ImmPred:$offset),
4136 (i64 DoubleRegs:$src1)),
4137 (S2_storerd_io (i32 IntRegs:$src2), s11_3ImmPred:$offset,
4138 (i64 DoubleRegs:$src1))>;
4140 // Map from r0 = and(r1, 65535) to r0 = zxth(r1)
4141 def : Pat <(and (i32 IntRegs:$src1), 65535),
4142 (A2_zxth (i32 IntRegs:$src1))>;
4144 // Map from r0 = and(r1, 255) to r0 = zxtb(r1).
4145 def : Pat <(and (i32 IntRegs:$src1), 255),
4146 (A2_zxtb (i32 IntRegs:$src1))>;
4148 // Map Add(p1, true) to p1 = not(p1).
4149 // Add(p1, false) should never be produced,
4150 // if it does, it got to be mapped to NOOP.
4151 def : Pat <(add (i1 PredRegs:$src1), -1),
4152 (C2_not (i1 PredRegs:$src1))>;
4154 // Map from p0 = pnot(p0); r0 = mux(p0, #i, #j) => r0 = mux(p0, #j, #i).
4155 def : Pat <(select (not (i1 PredRegs:$src1)), s8ImmPred:$src2, s8ImmPred:$src3),
4156 (i32 (TFR_condset_ii (i1 PredRegs:$src1), s8ImmPred:$src3,
4159 // Map from p0 = pnot(p0); r0 = select(p0, #i, r1)
4160 // => r0 = TFR_condset_ri(p0, r1, #i)
4161 def : Pat <(select (not (i1 PredRegs:$src1)), s12ImmPred:$src2,
4162 (i32 IntRegs:$src3)),
4163 (i32 (TFR_condset_ri (i1 PredRegs:$src1), (i32 IntRegs:$src3),
4164 s12ImmPred:$src2))>;
4166 // Map from p0 = pnot(p0); r0 = mux(p0, r1, #i)
4167 // => r0 = TFR_condset_ir(p0, #i, r1)
4168 def : Pat <(select (not (i1 PredRegs:$src1)), IntRegs:$src2, s12ImmPred:$src3),
4169 (i32 (TFR_condset_ir (i1 PredRegs:$src1), s12ImmPred:$src3,
4170 (i32 IntRegs:$src2)))>;
4172 // Map from p0 = pnot(p0); if (p0) jump => if (!p0) jump.
4173 def : Pat <(brcond (not (i1 PredRegs:$src1)), bb:$offset),
4174 (J2_jumpf (i1 PredRegs:$src1), bb:$offset)>;
4176 // Map from p2 = pnot(p2); p1 = and(p0, p2) => p1 = and(p0, !p2).
4177 def : Pat <(and (i1 PredRegs:$src1), (not (i1 PredRegs:$src2))),
4178 (i1 (C2_andn (i1 PredRegs:$src1), (i1 PredRegs:$src2)))>;
4181 let AddedComplexity = 100 in
4182 def : Pat <(i64 (zextloadi1 (HexagonCONST32 tglobaladdr:$global))),
4183 (i64 (A2_combinew (A2_tfrsi 0),
4184 (L2_loadrub_io (CONST32_set tglobaladdr:$global), 0)))>,
4187 // Map from i1 loads to 32 bits. This assumes that the i1* is byte aligned.
4188 let AddedComplexity = 10 in
4189 def : Pat <(i32 (zextloadi1 ADDRriS11_0:$addr)),
4190 (i32 (A2_and (i32 (L2_loadrb_io AddrFI:$addr, 0)), (A2_tfrsi 0x1)))>;
4192 // Map from Rdd = sign_extend_inreg(Rss, i32) -> Rdd = A2_sxtw(Rss.lo).
4193 def : Pat <(i64 (sext_inreg (i64 DoubleRegs:$src1), i32)),
4194 (i64 (A2_sxtw (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1), subreg_loreg))))>;
4196 // Map from Rdd = sign_extend_inreg(Rss, i16) -> Rdd = A2_sxtw(SXTH(Rss.lo)).
4197 def : Pat <(i64 (sext_inreg (i64 DoubleRegs:$src1), i16)),
4198 (i64 (A2_sxtw (i32 (A2_sxth (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1),
4199 subreg_loreg))))))>;
4201 // Map from Rdd = sign_extend_inreg(Rss, i8) -> Rdd = A2_sxtw(SXTB(Rss.lo)).
4202 def : Pat <(i64 (sext_inreg (i64 DoubleRegs:$src1), i8)),
4203 (i64 (A2_sxtw (i32 (A2_sxtb (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1),
4204 subreg_loreg))))))>;
4206 // We want to prevent emitting pnot's as much as possible.
4207 // Map brcond with an unsupported setcc to a J2_jumpf.
4208 def : Pat <(brcond (i1 (setne (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4210 (J2_jumpf (C2_cmpeq (i32 IntRegs:$src1), (i32 IntRegs:$src2)),
4213 def : Pat <(brcond (i1 (setne (i32 IntRegs:$src1), s10ImmPred:$src2)),
4215 (J2_jumpf (C2_cmpeqi (i32 IntRegs:$src1), s10ImmPred:$src2), bb:$offset)>;
4217 def : Pat <(brcond (i1 (setne (i1 PredRegs:$src1), (i1 -1))), bb:$offset),
4218 (J2_jumpf (i1 PredRegs:$src1), bb:$offset)>;
4220 def : Pat <(brcond (i1 (setne (i1 PredRegs:$src1), (i1 0))), bb:$offset),
4221 (J2_jumpt (i1 PredRegs:$src1), bb:$offset)>;
4223 // cmp.lt(Rs, Imm) -> !cmp.ge(Rs, Imm) -> !cmp.gt(Rs, Imm-1)
4224 def : Pat <(brcond (i1 (setlt (i32 IntRegs:$src1), s8ImmPred:$src2)),
4226 (J2_jumpf (C2_cmpgti (i32 IntRegs:$src1),
4227 (DEC_CONST_SIGNED s8ImmPred:$src2)), bb:$offset)>;
4229 // cmp.lt(r0, r1) -> cmp.gt(r1, r0)
4230 def : Pat <(brcond (i1 (setlt (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4232 (J2_jumpt (C2_cmpgt (i32 IntRegs:$src2), (i32 IntRegs:$src1)), bb:$offset)>;
4234 def : Pat <(brcond (i1 (setuge (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4236 (J2_jumpf (C2_cmpgtup (i64 DoubleRegs:$src2), (i64 DoubleRegs:$src1)),
4239 def : Pat <(brcond (i1 (setule (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4241 (J2_jumpf (C2_cmpgtu (i32 IntRegs:$src1), (i32 IntRegs:$src2)),
4244 def : Pat <(brcond (i1 (setule (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4246 (J2_jumpf (C2_cmpgtup (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2)),
4249 // Map from a 64-bit select to an emulated 64-bit mux.
4250 // Hexagon does not support 64-bit MUXes; so emulate with combines.
4251 def : Pat <(select (i1 PredRegs:$src1), (i64 DoubleRegs:$src2),
4252 (i64 DoubleRegs:$src3)),
4253 (i64 (A2_combinew (i32 (C2_mux (i1 PredRegs:$src1),
4254 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src2),
4256 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src3),
4258 (i32 (C2_mux (i1 PredRegs:$src1),
4259 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src2),
4261 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src3),
4262 subreg_loreg))))))>;
4264 // Map from a 1-bit select to logical ops.
4265 // From LegalizeDAG.cpp: (B1 ? B2 : B3) <=> (B1 & B2)|(!B1&B3).
4266 def : Pat <(select (i1 PredRegs:$src1), (i1 PredRegs:$src2),
4267 (i1 PredRegs:$src3)),
4268 (C2_or (C2_and (i1 PredRegs:$src1), (i1 PredRegs:$src2)),
4269 (C2_and (C2_not (i1 PredRegs:$src1)), (i1 PredRegs:$src3)))>;
4271 // Map Pd = load(addr) -> Rs = load(addr); Pd = Rs.
4272 def : Pat<(i1 (load ADDRriS11_2:$addr)),
4273 (i1 (C2_tfrrp (i32 (L2_loadrb_io AddrFI:$addr, 0))))>;
4275 // Map for truncating from 64 immediates to 32 bit immediates.
4276 def : Pat<(i32 (trunc (i64 DoubleRegs:$src))),
4277 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src), subreg_loreg))>;
4279 // Map for truncating from i64 immediates to i1 bit immediates.
4280 def : Pat<(i1 (trunc (i64 DoubleRegs:$src))),
4281 (i1 (C2_tfrrp (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
4284 // Map memb(Rs) = Rdd -> memb(Rs) = Rt.
4285 def : Pat<(truncstorei8 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
4286 (S2_storerb_io AddrFI:$addr, 0, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
4289 // Map memh(Rs) = Rdd -> memh(Rs) = Rt.
4290 def : Pat<(truncstorei16 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
4291 (S2_storerh_io AddrFI:$addr, 0, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
4293 // Map memw(Rs) = Rdd -> memw(Rs) = Rt
4294 def : Pat<(truncstorei32 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
4295 (S2_storeri_io AddrFI:$addr, 0, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
4298 // Map memw(Rs) = Rdd -> memw(Rs) = Rt.
4299 def : Pat<(truncstorei32 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
4300 (S2_storeri_io AddrFI:$addr, 0, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
4303 // Map from i1 = constant<-1>; memw(addr) = i1 -> r0 = 1; memw(addr) = r0.
4304 def : Pat<(store (i1 -1), ADDRriS11_2:$addr),
4305 (S2_storerb_io AddrFI:$addr, 0, (A2_tfrsi 1))>;
4308 // Map from i1 = constant<-1>; store i1 -> r0 = 1; store r0.
4309 def : Pat<(store (i1 -1), ADDRriS11_2:$addr),
4310 (S2_storerb_io AddrFI:$addr, 0, (A2_tfrsi 1))>;
4312 // Map from memb(Rs) = Pd -> Rt = mux(Pd, #0, #1); store Rt.
4313 def : Pat<(store (i1 PredRegs:$src1), ADDRriS11_2:$addr),
4314 (S2_storerb_io AddrFI:$addr, 0, (i32 (C2_muxii (i1 PredRegs:$src1), 1, 0)) )>;
4316 // Map Rdd = anyext(Rs) -> Rdd = A2_sxtw(Rs).
4317 // Hexagon_TODO: We can probably use combine but that will cost 2 instructions.
4318 // Better way to do this?
4319 def : Pat<(i64 (anyext (i32 IntRegs:$src1))),
4320 (i64 (A2_sxtw (i32 IntRegs:$src1)))>;
4322 // Map cmple -> cmpgt.
4323 // rs <= rt -> !(rs > rt).
4324 def : Pat<(i1 (setle (i32 IntRegs:$src1), s10ExtPred:$src2)),
4325 (i1 (C2_not (C2_cmpgti (i32 IntRegs:$src1), s10ExtPred:$src2)))>;
4327 // rs <= rt -> !(rs > rt).
4328 def : Pat<(i1 (setle (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4329 (i1 (C2_not (C2_cmpgt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))>;
4331 // Rss <= Rtt -> !(Rss > Rtt).
4332 def : Pat<(i1 (setle (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4333 (i1 (C2_not (C2_cmpgtp (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))))>;
4335 // Map cmpne -> cmpeq.
4336 // Hexagon_TODO: We should improve on this.
4337 // rs != rt -> !(rs == rt).
4338 def : Pat <(i1 (setne (i32 IntRegs:$src1), s10ExtPred:$src2)),
4339 (i1 (C2_not(i1 (C2_cmpeqi (i32 IntRegs:$src1), s10ExtPred:$src2))))>;
4341 // Map cmpne(Rs) -> !cmpeqe(Rs).
4342 // rs != rt -> !(rs == rt).
4343 def : Pat <(i1 (setne (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4344 (i1 (C2_not (i1 (C2_cmpeq (i32 IntRegs:$src1), (i32 IntRegs:$src2)))))>;
4346 // Convert setne back to xor for hexagon since we compute w/ pred registers.
4347 def : Pat <(i1 (setne (i1 PredRegs:$src1), (i1 PredRegs:$src2))),
4348 (i1 (C2_xor (i1 PredRegs:$src1), (i1 PredRegs:$src2)))>;
4350 // Map cmpne(Rss) -> !cmpew(Rss).
4351 // rs != rt -> !(rs == rt).
4352 def : Pat <(i1 (setne (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4353 (i1 (C2_not (i1 (C2_cmpeqp (i64 DoubleRegs:$src1),
4354 (i64 DoubleRegs:$src2)))))>;
4356 // Map cmpge(Rs, Rt) -> !(cmpgt(Rs, Rt).
4357 // rs >= rt -> !(rt > rs).
4358 def : Pat <(i1 (setge (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4359 (i1 (C2_not (i1 (C2_cmpgt (i32 IntRegs:$src2), (i32 IntRegs:$src1)))))>;
4361 // cmpge(Rs, Imm) -> cmpgt(Rs, Imm-1)
4362 def : Pat <(i1 (setge (i32 IntRegs:$src1), s8ExtPred:$src2)),
4363 (i1 (C2_cmpgti (i32 IntRegs:$src1), (DEC_CONST_SIGNED s8ExtPred:$src2)))>;
4365 // Map cmpge(Rss, Rtt) -> !cmpgt(Rtt, Rss).
4366 // rss >= rtt -> !(rtt > rss).
4367 def : Pat <(i1 (setge (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4368 (i1 (C2_not (i1 (C2_cmpgtp (i64 DoubleRegs:$src2),
4369 (i64 DoubleRegs:$src1)))))>;
4371 // Map cmplt(Rs, Imm) -> !cmpge(Rs, Imm).
4372 // !cmpge(Rs, Imm) -> !cmpgt(Rs, Imm-1).
4373 // rs < rt -> !(rs >= rt).
4374 def : Pat <(i1 (setlt (i32 IntRegs:$src1), s8ExtPred:$src2)),
4375 (i1 (C2_not (C2_cmpgti (i32 IntRegs:$src1), (DEC_CONST_SIGNED s8ExtPred:$src2))))>;
4377 // Map cmplt(Rs, Rt) -> cmpgt(Rt, Rs).
4378 // rs < rt -> rt > rs.
4379 // We can let assembler map it, or we can do in the compiler itself.
4380 def : Pat <(i1 (setlt (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4381 (i1 (C2_cmpgt (i32 IntRegs:$src2), (i32 IntRegs:$src1)))>;
4383 // Map cmplt(Rss, Rtt) -> cmpgt(Rtt, Rss).
4384 // rss < rtt -> (rtt > rss).
4385 def : Pat <(i1 (setlt (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4386 (i1 (C2_cmpgtp (i64 DoubleRegs:$src2), (i64 DoubleRegs:$src1)))>;
4388 // Map from cmpltu(Rs, Rd) -> cmpgtu(Rd, Rs)
4389 // rs < rt -> rt > rs.
4390 // We can let assembler map it, or we can do in the compiler itself.
4391 def : Pat <(i1 (setult (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4392 (i1 (C2_cmpgtu (i32 IntRegs:$src2), (i32 IntRegs:$src1)))>;
4394 // Map from cmpltu(Rss, Rdd) -> cmpgtu(Rdd, Rss).
4395 // rs < rt -> rt > rs.
4396 def : Pat <(i1 (setult (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4397 (i1 (C2_cmpgtup (i64 DoubleRegs:$src2), (i64 DoubleRegs:$src1)))>;
4399 // Generate cmpgeu(Rs, #0) -> cmpeq(Rs, Rs)
4400 def : Pat <(i1 (setuge (i32 IntRegs:$src1), 0)),
4401 (i1 (C2_cmpeq (i32 IntRegs:$src1), (i32 IntRegs:$src1)))>;
4403 // Generate cmpgeu(Rs, #u8) -> cmpgtu(Rs, #u8 -1)
4404 def : Pat <(i1 (setuge (i32 IntRegs:$src1), u8ExtPred:$src2)),
4405 (i1 (C2_cmpgtui (i32 IntRegs:$src1), (DEC_CONST_UNSIGNED u8ExtPred:$src2)))>;
4407 // Generate cmpgtu(Rs, #u9)
4408 def : Pat <(i1 (setugt (i32 IntRegs:$src1), u9ExtPred:$src2)),
4409 (i1 (C2_cmpgtui (i32 IntRegs:$src1), u9ExtPred:$src2))>;
4411 // Map from Rs >= Rt -> !(Rt > Rs).
4412 // rs >= rt -> !(rt > rs).
4413 def : Pat <(i1 (setuge (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4414 (i1 (C2_not (C2_cmpgtu (i32 IntRegs:$src2), (i32 IntRegs:$src1))))>;
4416 // Map from Rs >= Rt -> !(Rt > Rs).
4417 // rs >= rt -> !(rt > rs).
4418 def : Pat <(i1 (setuge (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4419 (i1 (C2_not (C2_cmpgtup (i64 DoubleRegs:$src2), (i64 DoubleRegs:$src1))))>;
4421 // Map from cmpleu(Rs, Rt) -> !cmpgtu(Rs, Rt).
4422 // Map from (Rs <= Rt) -> !(Rs > Rt).
4423 def : Pat <(i1 (setule (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4424 (i1 (C2_not (C2_cmpgtu (i32 IntRegs:$src1), (i32 IntRegs:$src2))))>;
4426 // Map from cmpleu(Rss, Rtt) -> !cmpgtu(Rss, Rtt-1).
4427 // Map from (Rs <= Rt) -> !(Rs > Rt).
4428 def : Pat <(i1 (setule (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4429 (i1 (C2_not (C2_cmpgtup (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))))>;
4433 def : Pat <(i32 (sext (i1 PredRegs:$src1))),
4434 (i32 (C2_muxii (i1 PredRegs:$src1), -1, 0))>;
4437 def : Pat <(i64 (sext (i1 PredRegs:$src1))),
4438 (i64 (A2_combinew (A2_tfrsi -1), (C2_muxii (i1 PredRegs:$src1), -1, 0)))>;
4440 // Convert sign-extended load back to load and sign extend.
4442 def: Pat <(i64 (sextloadi8 ADDRriS11_0:$src1)),
4443 (i64 (A2_sxtw (L2_loadrb_io AddrFI:$src1, 0)))>;
4445 // Convert any-extended load back to load and sign extend.
4447 def: Pat <(i64 (extloadi8 ADDRriS11_0:$src1)),
4448 (i64 (A2_sxtw (L2_loadrb_io AddrFI:$src1, 0)))>;
4450 // Convert sign-extended load back to load and sign extend.
4452 def: Pat <(i64 (sextloadi16 ADDRriS11_1:$src1)),
4453 (i64 (A2_sxtw (L2_loadrh_io AddrFI:$src1, 0)))>;
4455 // Convert sign-extended load back to load and sign extend.
4457 def: Pat <(i64 (sextloadi32 ADDRriS11_2:$src1)),
4458 (i64 (A2_sxtw (L2_loadri_io AddrFI:$src1, 0)))>;
4463 def : Pat <(i32 (zext (i1 PredRegs:$src1))),
4464 (i32 (C2_muxii (i1 PredRegs:$src1), 1, 0))>;
4467 def : Pat <(i64 (zext (i1 PredRegs:$src1))),
4468 (i64 (A2_combinew (A2_tfrsi 0), (C2_muxii (i1 PredRegs:$src1), 1, 0)))>,
4472 def : Pat <(i64 (zext (i32 IntRegs:$src1))),
4473 (i64 (A2_combinew (A2_tfrsi 0), (i32 IntRegs:$src1)))>,
4477 def: Pat <(i64 (zextloadi8 ADDRriS11_0:$src1)),
4478 (i64 (A2_combinew (A2_tfrsi 0), (L2_loadrub_io AddrFI:$src1, 0)))>,
4481 let AddedComplexity = 20 in
4482 def: Pat <(i64 (zextloadi8 (add (i32 IntRegs:$src1),
4483 s11_0ExtPred:$offset))),
4484 (i64 (A2_combinew (A2_tfrsi 0), (L2_loadrub_io IntRegs:$src1,
4485 s11_0ExtPred:$offset)))>,
4489 def: Pat <(i64 (zextloadi1 ADDRriS11_0:$src1)),
4490 (i64 (A2_combinew (A2_tfrsi 0), (L2_loadrub_io AddrFI:$src1, 0)))>,
4493 let AddedComplexity = 20 in
4494 def: Pat <(i64 (zextloadi1 (add (i32 IntRegs:$src1),
4495 s11_0ExtPred:$offset))),
4496 (i64 (A2_combinew (A2_tfrsi 0), (L2_loadrub_io IntRegs:$src1,
4497 s11_0ExtPred:$offset)))>,
4501 def: Pat <(i64 (zextloadi16 ADDRriS11_1:$src1)),
4502 (i64 (A2_combinew (A2_tfrsi 0), (L2_loadruh_io AddrFI:$src1, 0)))>,
4505 let AddedComplexity = 20 in
4506 def: Pat <(i64 (zextloadi16 (add (i32 IntRegs:$src1),
4507 s11_1ExtPred:$offset))),
4508 (i64 (A2_combinew (A2_tfrsi 0), (L2_loadruh_io IntRegs:$src1,
4509 s11_1ExtPred:$offset)))>,
4513 def: Pat <(i64 (zextloadi32 ADDRriS11_2:$src1)),
4514 (i64 (A2_combinew (A2_tfrsi 0), (L2_loadri_io AddrFI:$src1, 0)))>,
4517 let AddedComplexity = 100 in
4518 def: Pat <(i64 (zextloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
4519 (i64 (A2_combinew (A2_tfrsi 0), (L2_loadri_io IntRegs:$src1,
4520 s11_2ExtPred:$offset)))>,
4523 let AddedComplexity = 10 in
4524 def: Pat <(i32 (zextloadi1 ADDRriS11_0:$src1)),
4525 (i32 (L2_loadri_io AddrFI:$src1, 0))>;
4527 // Map from Rs = Pd to Pd = mux(Pd, #1, #0)
4528 def : Pat <(i32 (zext (i1 PredRegs:$src1))),
4529 (i32 (C2_muxii (i1 PredRegs:$src1), 1, 0))>;
4531 // Map from Rs = Pd to Pd = mux(Pd, #1, #0)
4532 def : Pat <(i32 (anyext (i1 PredRegs:$src1))),
4533 (i32 (C2_muxii (i1 PredRegs:$src1), 1, 0))>;
4535 // Map from Rss = Pd to Rdd = A2_sxtw (mux(Pd, #1, #0))
4536 def : Pat <(i64 (anyext (i1 PredRegs:$src1))),
4537 (i64 (A2_sxtw (i32 (C2_muxii (i1 PredRegs:$src1), 1, 0))))>;
4540 let AddedComplexity = 100 in
4541 def: Pat<(i64 (or (i64 (shl (i64 DoubleRegs:$srcHigh),
4543 (i64 (zextloadi32 (i32 (add IntRegs:$src2,
4544 s11_2ExtPred:$offset2)))))),
4545 (i64 (A2_combinew (EXTRACT_SUBREG (i64 DoubleRegs:$srcHigh), subreg_loreg),
4546 (L2_loadri_io IntRegs:$src2,
4547 s11_2ExtPred:$offset2)))>;
4549 def: Pat<(i64 (or (i64 (shl (i64 DoubleRegs:$srcHigh),
4551 (i64 (zextloadi32 ADDRriS11_2:$srcLow)))),
4552 (i64 (A2_combinew (EXTRACT_SUBREG (i64 DoubleRegs:$srcHigh), subreg_loreg),
4553 (L2_loadri_io AddrFI:$srcLow, 0)))>;
4555 def: Pat<(i64 (or (i64 (shl (i64 DoubleRegs:$srcHigh),
4557 (i64 (zext (i32 IntRegs:$srcLow))))),
4558 (i64 (A2_combinew (EXTRACT_SUBREG (i64 DoubleRegs:$srcHigh), subreg_loreg),
4561 let AddedComplexity = 100 in
4562 def: Pat<(i64 (or (i64 (shl (i64 DoubleRegs:$srcHigh),
4564 (i64 (zextloadi32 (i32 (add IntRegs:$src2,
4565 s11_2ExtPred:$offset2)))))),
4566 (i64 (A2_combinew (EXTRACT_SUBREG (i64 DoubleRegs:$srcHigh), subreg_loreg),
4567 (L2_loadri_io IntRegs:$src2,
4568 s11_2ExtPred:$offset2)))>;
4570 def: Pat<(i64 (or (i64 (shl (i64 DoubleRegs:$srcHigh),
4572 (i64 (zextloadi32 ADDRriS11_2:$srcLow)))),
4573 (i64 (A2_combinew (EXTRACT_SUBREG (i64 DoubleRegs:$srcHigh), subreg_loreg),
4574 (L2_loadri_io AddrFI:$srcLow, 0)))>;
4576 def: Pat<(i64 (or (i64 (shl (i64 DoubleRegs:$srcHigh),
4578 (i64 (zext (i32 IntRegs:$srcLow))))),
4579 (i64 (A2_combinew (EXTRACT_SUBREG (i64 DoubleRegs:$srcHigh), subreg_loreg),
4582 // Any extended 64-bit load.
4583 // anyext i32 -> i64
4584 def: Pat <(i64 (extloadi32 ADDRriS11_2:$src1)),
4585 (i64 (A2_combinew (A2_tfrsi 0), (L2_loadri_io AddrFI:$src1, 0)))>,
4588 // When there is an offset we should prefer the pattern below over the pattern above.
4589 // The complexity of the above is 13 (gleaned from HexagonGenDAGIsel.inc)
4590 // So this complexity below is comfortably higher to allow for choosing the below.
4591 // If this is not done then we generate addresses such as
4592 // ********************************************
4593 // r1 = add (r0, #4)
4594 // r1 = memw(r1 + #0)
4596 // r1 = memw(r0 + #4)
4597 // ********************************************
4598 let AddedComplexity = 100 in
4599 def: Pat <(i64 (extloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
4600 (i64 (A2_combinew (A2_tfrsi 0), (L2_loadri_io IntRegs:$src1,
4601 s11_2ExtPred:$offset)))>,
4604 // anyext i16 -> i64.
4605 def: Pat <(i64 (extloadi16 ADDRriS11_2:$src1)),
4606 (i64 (A2_combinew (A2_tfrsi 0), (L2_loadrh_io AddrFI:$src1, 0)))>,
4609 let AddedComplexity = 20 in
4610 def: Pat <(i64 (extloadi16 (add (i32 IntRegs:$src1),
4611 s11_1ExtPred:$offset))),
4612 (i64 (A2_combinew (A2_tfrsi 0), (L2_loadrh_io IntRegs:$src1,
4613 s11_1ExtPred:$offset)))>,
4616 // Map from Rdd = zxtw(Rs) -> Rdd = combine(0, Rs).
4617 def : Pat<(i64 (zext (i32 IntRegs:$src1))),
4618 (i64 (A2_combinew (A2_tfrsi 0), (i32 IntRegs:$src1)))>,
4621 // Multiply 64-bit unsigned and use upper result.
4622 def : Pat <(mulhu (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2)),
4637 (A2_combinew (A2_tfrsi 0),
4644 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1),
4646 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src2),
4647 subreg_loreg)))), 32)),
4649 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1), subreg_hireg)),
4650 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src2), subreg_loreg)))),
4651 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1), subreg_loreg)),
4652 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src2), subreg_hireg)))),
4653 32)), subreg_loreg)))),
4654 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1), subreg_hireg)),
4655 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src2), subreg_hireg))))>;
4657 // Multiply 64-bit signed and use upper result.
4658 def : Pat <(mulhs (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2)),
4662 (A2_combinew (A2_tfrsi 0),
4672 (A2_combinew (A2_tfrsi 0),
4679 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1),
4681 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src2),
4682 subreg_loreg)))), 32)),
4684 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1), subreg_hireg)),
4685 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src2), subreg_loreg)))),
4686 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1), subreg_loreg)),
4687 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src2), subreg_hireg)))),
4688 32)), subreg_loreg)))),
4689 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src1), subreg_hireg)),
4690 (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src2), subreg_hireg))))>;
4692 // Hexagon specific ISD nodes.
4693 //def SDTHexagonADJDYNALLOC : SDTypeProfile<1, 2, [SDTCisSameAs<0, 1>]>;
4694 def SDTHexagonADJDYNALLOC : SDTypeProfile<1, 2,
4695 [SDTCisVT<0, i32>, SDTCisVT<1, i32>]>;
4696 def Hexagon_ADJDYNALLOC : SDNode<"HexagonISD::ADJDYNALLOC",
4697 SDTHexagonADJDYNALLOC>;
4698 // Needed to tag these instructions for stack layout.
4699 let usesCustomInserter = 1 in
4700 def ADJDYNALLOC : ALU32_ri<(outs IntRegs:$dst), (ins IntRegs:$src1,
4702 "$dst = add($src1, #$src2)",
4703 [(set (i32 IntRegs:$dst),
4704 (Hexagon_ADJDYNALLOC (i32 IntRegs:$src1),
4705 s16ImmPred:$src2))]>;
4707 def SDTHexagonARGEXTEND : SDTypeProfile<1, 1, [SDTCisVT<0, i32>]>;
4708 def Hexagon_ARGEXTEND : SDNode<"HexagonISD::ARGEXTEND", SDTHexagonARGEXTEND>;
4709 def ARGEXTEND : ALU32_rr <(outs IntRegs:$dst), (ins IntRegs:$src1),
4711 [(set (i32 IntRegs:$dst),
4712 (Hexagon_ARGEXTEND (i32 IntRegs:$src1)))]>;
4714 let AddedComplexity = 100 in
4715 def : Pat<(i32 (sext_inreg (Hexagon_ARGEXTEND (i32 IntRegs:$src1)), i16)),
4716 (COPY (i32 IntRegs:$src1))>;
4718 def HexagonWrapperJT: SDNode<"HexagonISD::WrapperJT", SDTIntUnaryOp>;
4720 def : Pat<(HexagonWrapperJT tjumptable:$dst),
4721 (i32 (CONST32_set_jt tjumptable:$dst))>;
4725 //===----------------------------------------------------------------------===//
4727 // Shift by immediate/register and accumulate/logical
4728 //===----------------------------------------------------------------------===//
4730 // Rx[+-&|]=asr(Rs,#u5)
4731 // Rx[+-&|^]=lsr(Rs,#u5)
4732 // Rx[+-&|^]=asl(Rs,#u5)
4734 let hasNewValue = 1, opNewValue = 0 in
4735 class T_shift_imm_acc_r <string opc1, string opc2, SDNode OpNode1,
4736 SDNode OpNode2, bits<3> majOp, bits<2> minOp>
4737 : SInst_acc<(outs IntRegs:$Rx),
4738 (ins IntRegs:$src1, IntRegs:$Rs, u5Imm:$u5),
4739 "$Rx "#opc2#opc1#"($Rs, #$u5)",
4740 [(set (i32 IntRegs:$Rx),
4741 (OpNode2 (i32 IntRegs:$src1),
4742 (OpNode1 (i32 IntRegs:$Rs), u5ImmPred:$u5)))],
4743 "$src1 = $Rx", S_2op_tc_2_SLOT23> {
4748 let IClass = 0b1000;
4750 let Inst{27-24} = 0b1110;
4751 let Inst{23-22} = majOp{2-1};
4753 let Inst{7} = majOp{0};
4754 let Inst{6-5} = minOp;
4756 let Inst{20-16} = Rs;
4757 let Inst{12-8} = u5;
4760 // Rx[+-&|]=asr(Rs,Rt)
4761 // Rx[+-&|^]=lsr(Rs,Rt)
4762 // Rx[+-&|^]=asl(Rs,Rt)
4764 let hasNewValue = 1, opNewValue = 0 in
4765 class T_shift_reg_acc_r <string opc1, string opc2, SDNode OpNode1,
4766 SDNode OpNode2, bits<2> majOp, bits<2> minOp>
4767 : SInst_acc<(outs IntRegs:$Rx),
4768 (ins IntRegs:$src1, IntRegs:$Rs, IntRegs:$Rt),
4769 "$Rx "#opc2#opc1#"($Rs, $Rt)",
4770 [(set (i32 IntRegs:$Rx),
4771 (OpNode2 (i32 IntRegs:$src1),
4772 (OpNode1 (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))],
4773 "$src1 = $Rx", S_3op_tc_2_SLOT23 > {
4778 let IClass = 0b1100;
4780 let Inst{27-24} = 0b1100;
4781 let Inst{23-22} = majOp;
4782 let Inst{7-6} = minOp;
4784 let Inst{20-16} = Rs;
4785 let Inst{12-8} = Rt;
4788 // Rxx[+-&|]=asr(Rss,#u6)
4789 // Rxx[+-&|^]=lsr(Rss,#u6)
4790 // Rxx[+-&|^]=asl(Rss,#u6)
4792 class T_shift_imm_acc_p <string opc1, string opc2, SDNode OpNode1,
4793 SDNode OpNode2, bits<3> majOp, bits<2> minOp>
4794 : SInst_acc<(outs DoubleRegs:$Rxx),
4795 (ins DoubleRegs:$src1, DoubleRegs:$Rss, u6Imm:$u6),
4796 "$Rxx "#opc2#opc1#"($Rss, #$u6)",
4797 [(set (i64 DoubleRegs:$Rxx),
4798 (OpNode2 (i64 DoubleRegs:$src1),
4799 (OpNode1 (i64 DoubleRegs:$Rss), u6ImmPred:$u6)))],
4800 "$src1 = $Rxx", S_2op_tc_2_SLOT23> {
4805 let IClass = 0b1000;
4807 let Inst{27-24} = 0b0010;
4808 let Inst{23-22} = majOp{2-1};
4809 let Inst{7} = majOp{0};
4810 let Inst{6-5} = minOp;
4811 let Inst{4-0} = Rxx;
4812 let Inst{20-16} = Rss;
4813 let Inst{13-8} = u6;
4817 // Rxx[+-&|]=asr(Rss,Rt)
4818 // Rxx[+-&|^]=lsr(Rss,Rt)
4819 // Rxx[+-&|^]=asl(Rss,Rt)
4820 // Rxx[+-&|^]=lsl(Rss,Rt)
4822 class T_shift_reg_acc_p <string opc1, string opc2, SDNode OpNode1,
4823 SDNode OpNode2, bits<3> majOp, bits<2> minOp>
4824 : SInst_acc<(outs DoubleRegs:$Rxx),
4825 (ins DoubleRegs:$src1, DoubleRegs:$Rss, IntRegs:$Rt),
4826 "$Rxx "#opc2#opc1#"($Rss, $Rt)",
4827 [(set (i64 DoubleRegs:$Rxx),
4828 (OpNode2 (i64 DoubleRegs:$src1),
4829 (OpNode1 (i64 DoubleRegs:$Rss), (i32 IntRegs:$Rt))))],
4830 "$src1 = $Rxx", S_3op_tc_2_SLOT23> {
4835 let IClass = 0b1100;
4837 let Inst{27-24} = 0b1011;
4838 let Inst{23-21} = majOp;
4839 let Inst{20-16} = Rss;
4840 let Inst{12-8} = Rt;
4841 let Inst{7-6} = minOp;
4842 let Inst{4-0} = Rxx;
4845 //===----------------------------------------------------------------------===//
4846 // Multi-class for the shift instructions with logical/arithmetic operators.
4847 //===----------------------------------------------------------------------===//
4849 multiclass xtype_imm_base<string OpcStr1, string OpcStr2, SDNode OpNode1,
4850 SDNode OpNode2, bits<3> majOp, bits<2> minOp > {
4851 def _i_r#NAME : T_shift_imm_acc_r< OpcStr1, OpcStr2, OpNode1,
4852 OpNode2, majOp, minOp >;
4853 def _i_p#NAME : T_shift_imm_acc_p< OpcStr1, OpcStr2, OpNode1,
4854 OpNode2, majOp, minOp >;
4857 multiclass xtype_imm_acc<string opc1, SDNode OpNode, bits<2>minOp> {
4858 let AddedComplexity = 100 in
4859 defm _acc : xtype_imm_base< opc1, "+= ", OpNode, add, 0b001, minOp>;
4861 defm _nac : xtype_imm_base< opc1, "-= ", OpNode, sub, 0b000, minOp>;
4862 defm _and : xtype_imm_base< opc1, "&= ", OpNode, and, 0b010, minOp>;
4863 defm _or : xtype_imm_base< opc1, "|= ", OpNode, or, 0b011, minOp>;
4866 multiclass xtype_xor_imm_acc<string opc1, SDNode OpNode, bits<2>minOp> {
4867 let AddedComplexity = 100 in
4868 defm _xacc : xtype_imm_base< opc1, "^= ", OpNode, xor, 0b100, minOp>;
4871 let isCodeGenOnly = 0 in {
4872 defm S2_asr : xtype_imm_acc<"asr", sra, 0b00>;
4874 defm S2_lsr : xtype_imm_acc<"lsr", srl, 0b01>,
4875 xtype_xor_imm_acc<"lsr", srl, 0b01>;
4877 defm S2_asl : xtype_imm_acc<"asl", shl, 0b10>,
4878 xtype_xor_imm_acc<"asl", shl, 0b10>;
4881 multiclass xtype_reg_acc_r<string opc1, SDNode OpNode, bits<2>minOp> {
4882 let AddedComplexity = 100 in
4883 def _acc : T_shift_reg_acc_r <opc1, "+= ", OpNode, add, 0b11, minOp>;
4885 def _nac : T_shift_reg_acc_r <opc1, "-= ", OpNode, sub, 0b10, minOp>;
4886 def _and : T_shift_reg_acc_r <opc1, "&= ", OpNode, and, 0b01, minOp>;
4887 def _or : T_shift_reg_acc_r <opc1, "|= ", OpNode, or, 0b00, minOp>;
4890 multiclass xtype_reg_acc_p<string opc1, SDNode OpNode, bits<2>minOp> {
4891 let AddedComplexity = 100 in
4892 def _acc : T_shift_reg_acc_p <opc1, "+= ", OpNode, add, 0b110, minOp>;
4894 def _nac : T_shift_reg_acc_p <opc1, "-= ", OpNode, sub, 0b100, minOp>;
4895 def _and : T_shift_reg_acc_p <opc1, "&= ", OpNode, and, 0b010, minOp>;
4896 def _or : T_shift_reg_acc_p <opc1, "|= ", OpNode, or, 0b000, minOp>;
4897 def _xor : T_shift_reg_acc_p <opc1, "^= ", OpNode, xor, 0b011, minOp>;
4900 multiclass xtype_reg_acc<string OpcStr, SDNode OpNode, bits<2> minOp > {
4901 defm _r_r : xtype_reg_acc_r <OpcStr, OpNode, minOp>;
4902 defm _r_p : xtype_reg_acc_p <OpcStr, OpNode, minOp>;
4905 let isCodeGenOnly = 0 in {
4906 defm S2_asl : xtype_reg_acc<"asl", shl, 0b10>;
4907 defm S2_asr : xtype_reg_acc<"asr", sra, 0b00>;
4908 defm S2_lsr : xtype_reg_acc<"lsr", srl, 0b01>;
4909 defm S2_lsl : xtype_reg_acc<"lsl", shl, 0b11>;
4912 //===----------------------------------------------------------------------===//
4913 let hasSideEffects = 0 in
4914 class T_S3op_1 <string mnemonic, RegisterClass RC, bits<2> MajOp, bits<3> MinOp,
4915 bit SwapOps, bit isSat = 0, bit isRnd = 0, bit hasShift = 0>
4916 : SInst <(outs RC:$dst),
4917 (ins DoubleRegs:$src1, DoubleRegs:$src2),
4918 "$dst = "#mnemonic#"($src1, $src2)"#!if(isRnd, ":rnd", "")
4919 #!if(hasShift,":>>1","")
4920 #!if(isSat, ":sat", ""),
4921 [], "", S_3op_tc_2_SLOT23 > {
4926 let IClass = 0b1100;
4928 let Inst{27-24} = 0b0001;
4929 let Inst{23-22} = MajOp;
4930 let Inst{20-16} = !if (SwapOps, src2, src1);
4931 let Inst{12-8} = !if (SwapOps, src1, src2);
4932 let Inst{7-5} = MinOp;
4933 let Inst{4-0} = dst;
4936 class T_S3op_64 <string mnemonic, bits<2> MajOp, bits<3> MinOp, bit SwapOps,
4937 bit isSat = 0, bit isRnd = 0, bit hasShift = 0 >
4938 : T_S3op_1 <mnemonic, DoubleRegs, MajOp, MinOp, SwapOps,
4939 isSat, isRnd, hasShift>;
4941 let isCodeGenOnly = 0 in
4942 def S2_lfsp : T_S3op_64 < "lfs", 0b10, 0b110, 0>;
4944 //===----------------------------------------------------------------------===//
4945 // Template class used by vector shift, vector rotate, vector neg,
4946 // 32-bit shift, 64-bit shifts, etc.
4947 //===----------------------------------------------------------------------===//
4949 let hasSideEffects = 0 in
4950 class T_S3op_3 <string mnemonic, RegisterClass RC, bits<2> MajOp,
4951 bits<2> MinOp, bit isSat = 0, list<dag> pattern = [] >
4952 : SInst <(outs RC:$dst),
4953 (ins RC:$src1, IntRegs:$src2),
4954 "$dst = "#mnemonic#"($src1, $src2)"#!if(isSat, ":sat", ""),
4955 pattern, "", S_3op_tc_1_SLOT23> {
4960 let IClass = 0b1100;
4962 let Inst{27-24} = !if(!eq(!cast<string>(RC), "IntRegs"), 0b0110, 0b0011);
4963 let Inst{23-22} = MajOp;
4964 let Inst{20-16} = src1;
4965 let Inst{12-8} = src2;
4966 let Inst{7-6} = MinOp;
4967 let Inst{4-0} = dst;
4970 let hasNewValue = 1 in
4971 class T_S3op_shift32 <string mnemonic, SDNode OpNode, bits<2> MinOp>
4972 : T_S3op_3 <mnemonic, IntRegs, 0b01, MinOp, 0,
4973 [(set (i32 IntRegs:$dst), (OpNode (i32 IntRegs:$src1),
4974 (i32 IntRegs:$src2)))]>;
4976 let hasNewValue = 1, Itinerary = S_3op_tc_2_SLOT23 in
4977 class T_S3op_shift32_Sat <string mnemonic, bits<2> MinOp>
4978 : T_S3op_3 <mnemonic, IntRegs, 0b00, MinOp, 1, []>;
4981 class T_S3op_shift64 <string mnemonic, SDNode OpNode, bits<2> MinOp>
4982 : T_S3op_3 <mnemonic, DoubleRegs, 0b10, MinOp, 0,
4983 [(set (i64 DoubleRegs:$dst), (OpNode (i64 DoubleRegs:$src1),
4984 (i32 IntRegs:$src2)))]>;
4987 class T_S3op_shiftVect <string mnemonic, bits<2> MajOp, bits<2> MinOp>
4988 : T_S3op_3 <mnemonic, DoubleRegs, MajOp, MinOp, 0, []>;
4991 // Shift by register
4992 // Rdd=[asr|lsr|asl|lsl](Rss,Rt)
4994 let isCodeGenOnly = 0 in {
4995 def S2_asr_r_p : T_S3op_shift64 < "asr", sra, 0b00>;
4996 def S2_lsr_r_p : T_S3op_shift64 < "lsr", srl, 0b01>;
4997 def S2_asl_r_p : T_S3op_shift64 < "asl", shl, 0b10>;
4998 def S2_lsl_r_p : T_S3op_shift64 < "lsl", shl, 0b11>;
5001 // Rd=[asr|lsr|asl|lsl](Rs,Rt)
5003 let isCodeGenOnly = 0 in {
5004 def S2_asr_r_r : T_S3op_shift32<"asr", sra, 0b00>;
5005 def S2_lsr_r_r : T_S3op_shift32<"lsr", srl, 0b01>;
5006 def S2_asl_r_r : T_S3op_shift32<"asl", shl, 0b10>;
5007 def S2_lsl_r_r : T_S3op_shift32<"lsl", shl, 0b11>;
5010 // Shift by register with saturation
5011 // Rd=asr(Rs,Rt):sat
5012 // Rd=asl(Rs,Rt):sat
5014 let Defs = [USR_OVF], isCodeGenOnly = 0 in {
5015 def S2_asr_r_r_sat : T_S3op_shift32_Sat<"asr", 0b00>;
5016 def S2_asl_r_r_sat : T_S3op_shift32_Sat<"asl", 0b10>;
5019 //===----------------------------------------------------------------------===//
5020 // Template class for 'insert bitfield' instructions
5021 //===----------------------------------------------------------------------===//
5022 let hasSideEffects = 0 in
5023 class T_S3op_insert <string mnemonic, RegisterClass RC>
5024 : SInst <(outs RC:$dst),
5025 (ins RC:$src1, RC:$src2, DoubleRegs:$src3),
5026 "$dst = "#mnemonic#"($src2, $src3)" ,
5027 [], "$src1 = $dst", S_3op_tc_1_SLOT23 > {
5032 let IClass = 0b1100;
5034 let Inst{27-26} = 0b10;
5035 let Inst{25-24} = !if(!eq(!cast<string>(RC), "IntRegs"), 0b00, 0b10);
5037 let Inst{20-16} = src2;
5038 let Inst{12-8} = src3;
5039 let Inst{4-0} = dst;
5042 let hasSideEffects = 0 in
5043 class T_S2op_insert <bits<4> RegTyBits, RegisterClass RC, Operand ImmOp>
5044 : SInst <(outs RC:$dst), (ins RC:$dst2, RC:$src1, ImmOp:$src2, ImmOp:$src3),
5045 "$dst = insert($src1, #$src2, #$src3)",
5046 [], "$dst2 = $dst", S_2op_tc_2_SLOT23> {
5053 string ImmOpStr = !cast<string>(ImmOp);
5055 let bit23 = !if (!eq(ImmOpStr, "u6Imm"), src3{5}, 0);
5056 let bit13 = !if (!eq(ImmOpStr, "u6Imm"), src2{5}, 0);
5058 let IClass = 0b1000;
5060 let Inst{27-24} = RegTyBits;
5061 let Inst{23} = bit23;
5062 let Inst{22-21} = src3{4-3};
5063 let Inst{20-16} = src1;
5064 let Inst{13} = bit13;
5065 let Inst{12-8} = src2{4-0};
5066 let Inst{7-5} = src3{2-0};
5067 let Inst{4-0} = dst;
5070 // Rx=insert(Rs,Rtt)
5071 // Rx=insert(Rs,#u5,#U5)
5072 let hasNewValue = 1, isCodeGenOnly = 0 in {
5073 def S2_insert_rp : T_S3op_insert <"insert", IntRegs>;
5074 def S2_insert : T_S2op_insert <0b1111, IntRegs, u5Imm>;
5077 // Rxx=insert(Rss,Rtt)
5078 // Rxx=insert(Rss,#u6,#U6)
5079 let isCodeGenOnly = 0 in {
5080 def S2_insertp_rp : T_S3op_insert<"insert", DoubleRegs>;
5081 def S2_insertp : T_S2op_insert <0b0011, DoubleRegs, u6Imm>;
5084 //===----------------------------------------------------------------------===//
5085 // Template class for 'extract bitfield' instructions
5086 //===----------------------------------------------------------------------===//
5087 let hasNewValue = 1, hasSideEffects = 0 in
5088 class T_S3op_extract <string mnemonic, bits<2> MinOp>
5089 : SInst <(outs IntRegs:$Rd), (ins IntRegs:$Rs, DoubleRegs:$Rtt),
5090 "$Rd = "#mnemonic#"($Rs, $Rtt)",
5091 [], "", S_3op_tc_2_SLOT23 > {
5096 let IClass = 0b1100;
5098 let Inst{27-22} = 0b100100;
5099 let Inst{20-16} = Rs;
5100 let Inst{12-8} = Rtt;
5101 let Inst{7-6} = MinOp;
5105 let hasSideEffects = 0 in
5106 class T_S2op_extract <string mnemonic, bits<4> RegTyBits,
5107 RegisterClass RC, Operand ImmOp>
5108 : SInst <(outs RC:$dst), (ins RC:$src1, ImmOp:$src2, ImmOp:$src3),
5109 "$dst = "#mnemonic#"($src1, #$src2, #$src3)",
5110 [], "", S_2op_tc_2_SLOT23> {
5117 string ImmOpStr = !cast<string>(ImmOp);
5119 let bit23 = !if (!eq(ImmOpStr, "u6Imm"), src3{5},
5120 !if (!eq(mnemonic, "extractu"), 0, 1));
5122 let bit13 = !if (!eq(ImmOpStr, "u6Imm"), src2{5}, 0);
5124 let IClass = 0b1000;
5126 let Inst{27-24} = RegTyBits;
5127 let Inst{23} = bit23;
5128 let Inst{22-21} = src3{4-3};
5129 let Inst{20-16} = src1;
5130 let Inst{13} = bit13;
5131 let Inst{12-8} = src2{4-0};
5132 let Inst{7-5} = src3{2-0};
5133 let Inst{4-0} = dst;
5138 // Rdd=extractu(Rss,Rtt)
5139 // Rdd=extractu(Rss,#u6,#U6)
5140 let isCodeGenOnly = 0 in {
5141 def S2_extractup_rp : T_S3op_64 < "extractu", 0b00, 0b000, 0>;
5142 def S2_extractup : T_S2op_extract <"extractu", 0b0001, DoubleRegs, u6Imm>;
5145 // Rd=extractu(Rs,Rtt)
5146 // Rd=extractu(Rs,#u5,#U5)
5147 let hasNewValue = 1, isCodeGenOnly = 0 in {
5148 def S2_extractu_rp : T_S3op_extract<"extractu", 0b00>;
5149 def S2_extractu : T_S2op_extract <"extractu", 0b1101, IntRegs, u5Imm>;
5152 //===----------------------------------------------------------------------===//
5153 // :raw for of tableindx[bdhw] insns
5154 //===----------------------------------------------------------------------===//
5156 let hasSideEffects = 0, hasNewValue = 1, opNewValue = 0 in
5157 class tableidxRaw<string OpStr, bits<2>MinOp>
5158 : SInst <(outs IntRegs:$Rx),
5159 (ins IntRegs:$_dst_, IntRegs:$Rs, u4Imm:$u4, s6Imm:$S6),
5160 "$Rx = "#OpStr#"($Rs, #$u4, #$S6):raw",
5161 [], "$Rx = $_dst_" > {
5167 let IClass = 0b1000;
5169 let Inst{27-24} = 0b0111;
5170 let Inst{23-22} = MinOp;
5171 let Inst{21} = u4{3};
5172 let Inst{20-16} = Rs;
5173 let Inst{13-8} = S6;
5174 let Inst{7-5} = u4{2-0};
5178 let isCodeGenOnly = 0 in {
5179 def S2_tableidxb : tableidxRaw<"tableidxb", 0b00>;
5180 def S2_tableidxh : tableidxRaw<"tableidxh", 0b01>;
5181 def S2_tableidxw : tableidxRaw<"tableidxw", 0b10>;
5182 def S2_tableidxd : tableidxRaw<"tableidxd", 0b11>;
5185 // Change the sign of the immediate for Rd=-mpyi(Rs,#u8)
5186 def : Pat <(mul (i32 IntRegs:$src1), (ineg n8ImmPred:$src2)),
5187 (i32 (M2_mpysin (i32 IntRegs:$src1), u8ImmPred:$src2))>;
5189 //===----------------------------------------------------------------------===//
5190 // V3 Instructions +
5191 //===----------------------------------------------------------------------===//
5193 include "HexagonInstrInfoV3.td"
5195 //===----------------------------------------------------------------------===//
5196 // V3 Instructions -
5197 //===----------------------------------------------------------------------===//
5199 //===----------------------------------------------------------------------===//
5200 // V4 Instructions +
5201 //===----------------------------------------------------------------------===//
5203 include "HexagonInstrInfoV4.td"
5205 //===----------------------------------------------------------------------===//
5206 // V4 Instructions -
5207 //===----------------------------------------------------------------------===//
5209 //===----------------------------------------------------------------------===//
5210 // V5 Instructions +
5211 //===----------------------------------------------------------------------===//
5213 include "HexagonInstrInfoV5.td"
5215 //===----------------------------------------------------------------------===//
5216 // V5 Instructions -
5217 //===----------------------------------------------------------------------===//