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 $Rs), subreg_loreg)>;
32 def HiReg: OutPatFrag<(ops node:$Rs),
33 (EXTRACT_SUBREG (i64 $Rs), subreg_hireg)>;
35 // SDNode for converting immediate C to C-1.
36 def DEC_CONST_SIGNED : SDNodeXForm<imm, [{
37 // Return the byte immediate const-1 as an SDNode.
38 int32_t imm = N->getSExtValue();
39 return XformSToSM1Imm(imm);
42 // SDNode for converting immediate C to C-2.
43 def DEC2_CONST_SIGNED : SDNodeXForm<imm, [{
44 // Return the byte immediate const-2 as an SDNode.
45 int32_t imm = N->getSExtValue();
46 return XformSToSM2Imm(imm);
49 // SDNode for converting immediate C to C-3.
50 def DEC3_CONST_SIGNED : SDNodeXForm<imm, [{
51 // Return the byte immediate const-3 as an SDNode.
52 int32_t imm = N->getSExtValue();
53 return XformSToSM3Imm(imm);
56 // SDNode for converting immediate C to C-1.
57 def DEC_CONST_UNSIGNED : SDNodeXForm<imm, [{
58 // Return the byte immediate const-1 as an SDNode.
59 uint32_t imm = N->getZExtValue();
60 return XformUToUM1Imm(imm);
63 //===----------------------------------------------------------------------===//
65 //===----------------------------------------------------------------------===//
66 let hasSideEffects = 0, isCompare = 1, InputType = "imm", isExtendable = 1,
68 class T_CMP <string mnemonic, bits<2> MajOp, bit isNot, Operand ImmOp>
69 : ALU32Inst <(outs PredRegs:$dst),
70 (ins IntRegs:$src1, ImmOp:$src2),
71 "$dst = "#!if(isNot, "!","")#mnemonic#"($src1, #$src2)",
72 [], "",ALU32_2op_tc_2early_SLOT0123 >, ImmRegRel {
76 let CextOpcode = mnemonic;
77 let opExtentBits = !if(!eq(mnemonic, "cmp.gtu"), 9, 10);
78 let isExtentSigned = !if(!eq(mnemonic, "cmp.gtu"), 0, 1);
82 let Inst{27-24} = 0b0101;
83 let Inst{23-22} = MajOp;
84 let Inst{21} = !if(!eq(mnemonic, "cmp.gtu"), 0, src2{9});
85 let Inst{20-16} = src1;
86 let Inst{13-5} = src2{8-0};
92 def C2_cmpeqi : T_CMP <"cmp.eq", 0b00, 0, s10Ext>;
93 def C2_cmpgti : T_CMP <"cmp.gt", 0b01, 0, s10Ext>;
94 def C2_cmpgtui : T_CMP <"cmp.gtu", 0b10, 0, u9Ext>;
96 class T_CMP_pat <InstHexagon MI, PatFrag OpNode, PatLeaf ImmPred>
97 : Pat<(i1 (OpNode (i32 IntRegs:$src1), ImmPred:$src2)),
98 (MI IntRegs:$src1, ImmPred:$src2)>;
100 def : T_CMP_pat <C2_cmpeqi, seteq, s10ImmPred>;
101 def : T_CMP_pat <C2_cmpgti, setgt, s10ImmPred>;
102 def : T_CMP_pat <C2_cmpgtui, setugt, u9ImmPred>;
104 //===----------------------------------------------------------------------===//
106 //===----------------------------------------------------------------------===//
107 def SDTHexagonI64I32I32 : SDTypeProfile<1, 2,
108 [SDTCisVT<0, i64>, SDTCisVT<1, i32>, SDTCisSameAs<1, 2>]>;
110 def HexagonCOMBINE : SDNode<"HexagonISD::COMBINE", SDTHexagonI64I32I32>;
112 let hasSideEffects = 0, hasNewValue = 1, InputType = "reg" in
113 class T_ALU32_3op<string mnemonic, bits<3> MajOp, bits<3> MinOp, bit OpsRev,
115 : ALU32_rr<(outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
116 "$Rd = "#mnemonic#"($Rs, $Rt)",
117 [], "", ALU32_3op_tc_1_SLOT0123>, ImmRegRel, PredRel {
118 let isCommutable = IsComm;
119 let BaseOpcode = mnemonic#_rr;
120 let CextOpcode = mnemonic;
128 let Inst{26-24} = MajOp;
129 let Inst{23-21} = MinOp;
130 let Inst{20-16} = !if(OpsRev,Rt,Rs);
131 let Inst{12-8} = !if(OpsRev,Rs,Rt);
135 let hasSideEffects = 0, hasNewValue = 1 in
136 class T_ALU32_3op_pred<string mnemonic, bits<3> MajOp, bits<3> MinOp,
137 bit OpsRev, bit PredNot, bit PredNew>
138 : ALU32_rr<(outs IntRegs:$Rd), (ins PredRegs:$Pu, IntRegs:$Rs, IntRegs:$Rt),
139 "if ("#!if(PredNot,"!","")#"$Pu"#!if(PredNew,".new","")#") "#
140 "$Rd = "#mnemonic#"($Rs, $Rt)",
141 [], "", ALU32_3op_tc_1_SLOT0123>, ImmRegRel, PredNewRel {
142 let isPredicated = 1;
143 let isPredicatedFalse = PredNot;
144 let isPredicatedNew = PredNew;
145 let BaseOpcode = mnemonic#_rr;
146 let CextOpcode = mnemonic;
155 let Inst{26-24} = MajOp;
156 let Inst{23-21} = MinOp;
157 let Inst{20-16} = !if(OpsRev,Rt,Rs);
158 let Inst{13} = PredNew;
159 let Inst{12-8} = !if(OpsRev,Rs,Rt);
160 let Inst{7} = PredNot;
165 class T_ALU32_combineh<string Op1, string Op2, bits<3> MajOp, bits<3> MinOp,
167 : T_ALU32_3op<"", MajOp, MinOp, OpsRev, 0> {
168 let AsmString = "$Rd = combine($Rs"#Op1#", $Rt"#Op2#")";
171 def A2_combine_hh : T_ALU32_combineh<".h", ".h", 0b011, 0b100, 1>;
172 def A2_combine_hl : T_ALU32_combineh<".h", ".l", 0b011, 0b101, 1>;
173 def A2_combine_lh : T_ALU32_combineh<".l", ".h", 0b011, 0b110, 1>;
174 def A2_combine_ll : T_ALU32_combineh<".l", ".l", 0b011, 0b111, 1>;
176 class T_ALU32_3op_sfx<string mnemonic, string suffix, bits<3> MajOp,
177 bits<3> MinOp, bit OpsRev, bit IsComm>
178 : T_ALU32_3op<"", MajOp, MinOp, OpsRev, IsComm> {
179 let AsmString = "$Rd = "#mnemonic#"($Rs, $Rt)"#suffix;
182 def A2_svaddh : T_ALU32_3op<"vaddh", 0b110, 0b000, 0, 1>;
183 def A2_svsubh : T_ALU32_3op<"vsubh", 0b110, 0b100, 1, 0>;
185 let Defs = [USR_OVF], Itinerary = ALU32_3op_tc_2_SLOT0123 in {
186 def A2_svaddhs : T_ALU32_3op_sfx<"vaddh", ":sat", 0b110, 0b001, 0, 1>;
187 def A2_addsat : T_ALU32_3op_sfx<"add", ":sat", 0b110, 0b010, 0, 1>;
188 def A2_svadduhs : T_ALU32_3op_sfx<"vadduh", ":sat", 0b110, 0b011, 0, 1>;
189 def A2_svsubhs : T_ALU32_3op_sfx<"vsubh", ":sat", 0b110, 0b101, 1, 0>;
190 def A2_subsat : T_ALU32_3op_sfx<"sub", ":sat", 0b110, 0b110, 1, 0>;
191 def A2_svsubuhs : T_ALU32_3op_sfx<"vsubuh", ":sat", 0b110, 0b111, 1, 0>;
194 let Itinerary = ALU32_3op_tc_2_SLOT0123 in
195 def A2_svavghs : T_ALU32_3op_sfx<"vavgh", ":rnd", 0b111, 0b001, 0, 1>;
197 def A2_svavgh : T_ALU32_3op<"vavgh", 0b111, 0b000, 0, 1>;
198 def A2_svnavgh : T_ALU32_3op<"vnavgh", 0b111, 0b011, 1, 0>;
200 multiclass T_ALU32_3op_p<string mnemonic, bits<3> MajOp, bits<3> MinOp,
202 def t : T_ALU32_3op_pred<mnemonic, MajOp, MinOp, OpsRev, 0, 0>;
203 def f : T_ALU32_3op_pred<mnemonic, MajOp, MinOp, OpsRev, 1, 0>;
204 def tnew : T_ALU32_3op_pred<mnemonic, MajOp, MinOp, OpsRev, 0, 1>;
205 def fnew : T_ALU32_3op_pred<mnemonic, MajOp, MinOp, OpsRev, 1, 1>;
208 multiclass T_ALU32_3op_A2<string mnemonic, bits<3> MajOp, bits<3> MinOp,
209 bit OpsRev, bit IsComm> {
210 let isPredicable = 1 in
211 def A2_#NAME : T_ALU32_3op <mnemonic, MajOp, MinOp, OpsRev, IsComm>;
212 defm A2_p#NAME : T_ALU32_3op_p<mnemonic, MajOp, MinOp, OpsRev>;
215 defm add : T_ALU32_3op_A2<"add", 0b011, 0b000, 0, 1>;
216 defm and : T_ALU32_3op_A2<"and", 0b001, 0b000, 0, 1>;
217 defm or : T_ALU32_3op_A2<"or", 0b001, 0b001, 0, 1>;
218 defm sub : T_ALU32_3op_A2<"sub", 0b011, 0b001, 1, 0>;
219 defm xor : T_ALU32_3op_A2<"xor", 0b001, 0b011, 0, 1>;
221 // Pats for instruction selection.
222 class BinOp32_pat<SDNode Op, InstHexagon MI, ValueType ResT>
223 : Pat<(ResT (Op (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))),
224 (ResT (MI IntRegs:$Rs, IntRegs:$Rt))>;
226 def: BinOp32_pat<add, A2_add, i32>;
227 def: BinOp32_pat<and, A2_and, i32>;
228 def: BinOp32_pat<or, A2_or, i32>;
229 def: BinOp32_pat<sub, A2_sub, i32>;
230 def: BinOp32_pat<xor, A2_xor, i32>;
232 // A few special cases producing register pairs:
233 let OutOperandList = (outs DoubleRegs:$Rd), hasNewValue = 0 in {
234 def S2_packhl : T_ALU32_3op <"packhl", 0b101, 0b100, 0, 0>;
236 let isPredicable = 1 in
237 def A2_combinew : T_ALU32_3op <"combine", 0b101, 0b000, 0, 0>;
239 // Conditional combinew uses "newt/f" instead of "t/fnew".
240 def C2_ccombinewt : T_ALU32_3op_pred<"combine", 0b101, 0b000, 0, 0, 0>;
241 def C2_ccombinewf : T_ALU32_3op_pred<"combine", 0b101, 0b000, 0, 1, 0>;
242 def C2_ccombinewnewt : T_ALU32_3op_pred<"combine", 0b101, 0b000, 0, 0, 1>;
243 def C2_ccombinewnewf : T_ALU32_3op_pred<"combine", 0b101, 0b000, 0, 1, 1>;
246 let hasSideEffects = 0, hasNewValue = 1, isCompare = 1, InputType = "reg" in
247 class T_ALU32_3op_cmp<string mnemonic, bits<2> MinOp, bit IsNeg, bit IsComm>
248 : ALU32_rr<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt),
249 "$Pd = "#mnemonic#"($Rs, $Rt)",
250 [], "", ALU32_3op_tc_1_SLOT0123>, ImmRegRel {
251 let CextOpcode = mnemonic;
252 let isCommutable = IsComm;
258 let Inst{27-24} = 0b0010;
259 let Inst{22-21} = MinOp;
260 let Inst{20-16} = Rs;
263 let Inst{3-2} = 0b00;
267 let Itinerary = ALU32_3op_tc_2early_SLOT0123 in {
268 def C2_cmpeq : T_ALU32_3op_cmp< "cmp.eq", 0b00, 0, 1>;
269 def C2_cmpgt : T_ALU32_3op_cmp< "cmp.gt", 0b10, 0, 0>;
270 def C2_cmpgtu : T_ALU32_3op_cmp< "cmp.gtu", 0b11, 0, 0>;
273 // Patfrag to convert the usual comparison patfrags (e.g. setlt) to ones
274 // that reverse the order of the operands.
275 class RevCmp<PatFrag F> : PatFrag<(ops node:$rhs, node:$lhs), F.Fragment>;
277 // Pats for compares. They use PatFrags as operands, not SDNodes,
278 // since seteq/setgt/etc. are defined as ParFrags.
279 class T_cmp32_rr_pat<InstHexagon MI, PatFrag Op, ValueType VT>
280 : Pat<(VT (Op (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))),
281 (VT (MI IntRegs:$Rs, IntRegs:$Rt))>;
283 def: T_cmp32_rr_pat<C2_cmpeq, seteq, i1>;
284 def: T_cmp32_rr_pat<C2_cmpgt, setgt, i1>;
285 def: T_cmp32_rr_pat<C2_cmpgtu, setugt, i1>;
287 def: T_cmp32_rr_pat<C2_cmpgt, RevCmp<setlt>, i1>;
288 def: T_cmp32_rr_pat<C2_cmpgtu, RevCmp<setult>, i1>;
290 let CextOpcode = "MUX", InputType = "reg", hasNewValue = 1 in
291 def C2_mux: ALU32_rr<(outs IntRegs:$Rd),
292 (ins PredRegs:$Pu, IntRegs:$Rs, IntRegs:$Rt),
293 "$Rd = mux($Pu, $Rs, $Rt)", [], "", ALU32_3op_tc_1_SLOT0123>, ImmRegRel {
299 let CextOpcode = "mux";
300 let InputType = "reg";
301 let hasSideEffects = 0;
304 let Inst{27-24} = 0b0100;
305 let Inst{20-16} = Rs;
311 def: Pat<(i32 (select (i1 PredRegs:$Pu), (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))),
312 (C2_mux PredRegs:$Pu, IntRegs:$Rs, IntRegs:$Rt)>;
314 // Combines the two immediates into a double register.
315 // Increase complexity to make it greater than any complexity of a combine
316 // that involves a register.
318 let isReMaterializable = 1, isMoveImm = 1, isAsCheapAsAMove = 1,
319 isExtentSigned = 1, isExtendable = 1, opExtentBits = 8, opExtendable = 1,
320 AddedComplexity = 75 in
321 def A2_combineii: ALU32Inst <(outs DoubleRegs:$Rdd), (ins s8Ext:$s8, s8Imm:$S8),
322 "$Rdd = combine(#$s8, #$S8)",
323 [(set (i64 DoubleRegs:$Rdd),
324 (i64 (HexagonCOMBINE(i32 s8ExtPred:$s8), (i32 s8ImmPred:$S8))))]> {
330 let Inst{27-23} = 0b11000;
331 let Inst{22-16} = S8{7-1};
332 let Inst{13} = S8{0};
337 //===----------------------------------------------------------------------===//
338 // Template class for predicated ADD of a reg and an Immediate value.
339 //===----------------------------------------------------------------------===//
340 let hasNewValue = 1, hasSideEffects = 0 in
341 class T_Addri_Pred <bit PredNot, bit PredNew>
342 : ALU32_ri <(outs IntRegs:$Rd),
343 (ins PredRegs:$Pu, IntRegs:$Rs, s8Ext:$s8),
344 !if(PredNot, "if (!$Pu", "if ($Pu")#!if(PredNew,".new) $Rd = ",
345 ") $Rd = ")#"add($Rs, #$s8)"> {
351 let isPredicatedNew = PredNew;
354 let Inst{27-24} = 0b0100;
355 let Inst{23} = PredNot;
356 let Inst{22-21} = Pu;
357 let Inst{20-16} = Rs;
358 let Inst{13} = PredNew;
363 //===----------------------------------------------------------------------===//
364 // A2_addi: Add a signed immediate to a register.
365 //===----------------------------------------------------------------------===//
366 let hasNewValue = 1, hasSideEffects = 0 in
367 class T_Addri <Operand immOp>
368 : ALU32_ri <(outs IntRegs:$Rd),
369 (ins IntRegs:$Rs, immOp:$s16),
370 "$Rd = add($Rs, #$s16)", [], "", ALU32_ADDI_tc_1_SLOT0123> {
377 let Inst{27-21} = s16{15-9};
378 let Inst{20-16} = Rs;
379 let Inst{13-5} = s16{8-0};
383 //===----------------------------------------------------------------------===//
384 // Multiclass for ADD of a register and an immediate value.
385 //===----------------------------------------------------------------------===//
386 multiclass Addri_Pred<string mnemonic, bit PredNot> {
387 let isPredicatedFalse = PredNot in {
388 def NAME : T_Addri_Pred<PredNot, 0>;
390 def NAME#new : T_Addri_Pred<PredNot, 1>;
394 let isExtendable = 1, isExtentSigned = 1, InputType = "imm" in
395 multiclass Addri_base<string mnemonic, SDNode OpNode> {
396 let CextOpcode = mnemonic, BaseOpcode = mnemonic#_ri in {
397 let opExtendable = 2, opExtentBits = 16, isPredicable = 1 in
398 def A2_#NAME : T_Addri<s16Ext>;
400 let opExtendable = 3, opExtentBits = 8, isPredicated = 1 in {
401 defm A2_p#NAME#t : Addri_Pred<mnemonic, 0>;
402 defm A2_p#NAME#f : Addri_Pred<mnemonic, 1>;
407 defm addi : Addri_base<"add", add>, ImmRegRel, PredNewRel;
409 def: Pat<(i32 (add I32:$Rs, s16ExtPred:$s16)),
410 (i32 (A2_addi I32:$Rs, imm:$s16))>;
412 //===----------------------------------------------------------------------===//
413 // Template class used for the following ALU32 instructions.
416 //===----------------------------------------------------------------------===//
417 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 10,
418 InputType = "imm", hasNewValue = 1 in
419 class T_ALU32ri_logical <string mnemonic, SDNode OpNode, bits<2> MinOp>
420 : ALU32_ri <(outs IntRegs:$Rd),
421 (ins IntRegs:$Rs, s10Ext:$s10),
422 "$Rd = "#mnemonic#"($Rs, #$s10)" ,
423 [(set (i32 IntRegs:$Rd), (OpNode (i32 IntRegs:$Rs), s10ExtPred:$s10))]> {
427 let CextOpcode = mnemonic;
431 let Inst{27-24} = 0b0110;
432 let Inst{23-22} = MinOp;
433 let Inst{21} = s10{9};
434 let Inst{20-16} = Rs;
435 let Inst{13-5} = s10{8-0};
439 def A2_orir : T_ALU32ri_logical<"or", or, 0b10>, ImmRegRel;
440 def A2_andir : T_ALU32ri_logical<"and", and, 0b00>, ImmRegRel;
442 // Subtract register from immediate
443 // Rd32=sub(#s10,Rs32)
444 let isExtendable = 1, CextOpcode = "sub", opExtendable = 1, isExtentSigned = 1,
445 opExtentBits = 10, InputType = "imm", hasNewValue = 1, hasSideEffects = 0 in
446 def A2_subri: ALU32_ri <(outs IntRegs:$Rd), (ins s10Ext:$s10, IntRegs:$Rs),
447 "$Rd = sub(#$s10, $Rs)", []>, ImmRegRel {
454 let Inst{27-22} = 0b011001;
455 let Inst{21} = s10{9};
456 let Inst{20-16} = Rs;
457 let Inst{13-5} = s10{8-0};
462 let hasSideEffects = 0 in
463 def A2_nop: ALU32Inst <(outs), (ins), "nop" > {
465 let Inst{27-24} = 0b1111;
468 def: Pat<(sub s10ExtPred:$s10, IntRegs:$Rs),
469 (A2_subri imm:$s10, IntRegs:$Rs)>;
471 // Rd = not(Rs) gets mapped to Rd=sub(#-1, Rs).
472 def: Pat<(not (i32 IntRegs:$src1)),
473 (A2_subri -1, IntRegs:$src1)>;
475 let hasSideEffects = 0, hasNewValue = 1 in
476 class T_tfr16<bit isHi>
477 : ALU32Inst <(outs IntRegs:$Rx), (ins IntRegs:$src1, u16Imm:$u16),
478 "$Rx"#!if(isHi, ".h", ".l")#" = #$u16",
479 [], "$src1 = $Rx" > {
484 let Inst{27-26} = 0b00;
485 let Inst{25-24} = !if(isHi, 0b10, 0b01);
486 let Inst{23-22} = u16{15-14};
488 let Inst{20-16} = Rx;
489 let Inst{13-0} = u16{13-0};
492 def A2_tfril: T_tfr16<0>;
493 def A2_tfrih: T_tfr16<1>;
495 // Conditional transfer is an alias to conditional "Rd = add(Rs, #0)".
496 let isPredicated = 1, hasNewValue = 1, opNewValue = 0 in
497 class T_tfr_pred<bit isPredNot, bit isPredNew>
498 : ALU32Inst<(outs IntRegs:$dst),
499 (ins PredRegs:$src1, IntRegs:$src2),
500 "if ("#!if(isPredNot, "!", "")#
501 "$src1"#!if(isPredNew, ".new", "")#
507 let isPredicatedFalse = isPredNot;
508 let isPredicatedNew = isPredNew;
511 let Inst{27-24} = 0b0100;
512 let Inst{23} = isPredNot;
513 let Inst{13} = isPredNew;
516 let Inst{22-21} = src1;
517 let Inst{20-16} = src2;
520 let isPredicable = 1 in
521 class T_tfr : ALU32Inst<(outs IntRegs:$dst), (ins IntRegs:$src),
528 let Inst{27-21} = 0b0000011;
529 let Inst{20-16} = src;
534 let InputType = "reg", hasNewValue = 1, hasSideEffects = 0 in
535 multiclass tfr_base<string CextOp> {
536 let CextOpcode = CextOp, BaseOpcode = CextOp in {
540 def t : T_tfr_pred<0, 0>;
541 def f : T_tfr_pred<1, 0>;
543 def tnew : T_tfr_pred<0, 1>;
544 def fnew : T_tfr_pred<1, 1>;
548 // Assembler mapped to C2_ccombinew[t|f|newt|newf].
549 // Please don't add bits to this instruction as it'll be converted into
550 // 'combine' before object code emission.
551 let isPredicated = 1 in
552 class T_tfrp_pred<bit PredNot, bit PredNew>
553 : ALU32_rr <(outs DoubleRegs:$dst),
554 (ins PredRegs:$src1, DoubleRegs:$src2),
555 "if ("#!if(PredNot, "!", "")#"$src1"
556 #!if(PredNew, ".new", "")#") $dst = $src2" > {
557 let isPredicatedFalse = PredNot;
558 let isPredicatedNew = PredNew;
561 // Assembler mapped to A2_combinew.
562 // Please don't add bits to this instruction as it'll be converted into
563 // 'combine' before object code emission.
564 class T_tfrp : ALU32Inst <(outs DoubleRegs:$dst),
565 (ins DoubleRegs:$src),
568 let hasSideEffects = 0 in
569 multiclass TFR64_base<string BaseName> {
570 let BaseOpcode = BaseName in {
571 let isPredicable = 1 in
574 def t : T_tfrp_pred <0, 0>;
575 def f : T_tfrp_pred <1, 0>;
577 def tnew : T_tfrp_pred <0, 1>;
578 def fnew : T_tfrp_pred <1, 1>;
582 let InputType = "imm", isExtendable = 1, isExtentSigned = 1, opExtentBits = 12,
583 isMoveImm = 1, opExtendable = 2, BaseOpcode = "TFRI", CextOpcode = "TFR",
584 hasSideEffects = 0, isPredicated = 1, hasNewValue = 1 in
585 class T_TFRI_Pred<bit PredNot, bit PredNew>
586 : ALU32_ri<(outs IntRegs:$Rd), (ins PredRegs:$Pu, s12Ext:$s12),
587 "if ("#!if(PredNot,"!","")#"$Pu"#!if(PredNew,".new","")#") $Rd = #$s12",
588 [], "", ALU32_2op_tc_1_SLOT0123>, ImmRegRel, PredNewRel {
589 let isPredicatedFalse = PredNot;
590 let isPredicatedNew = PredNew;
597 let Inst{27-24} = 0b1110;
598 let Inst{23} = PredNot;
599 let Inst{22-21} = Pu;
601 let Inst{19-16,12-5} = s12;
602 let Inst{13} = PredNew;
606 def C2_cmoveit : T_TFRI_Pred<0, 0>;
607 def C2_cmoveif : T_TFRI_Pred<1, 0>;
608 def C2_cmovenewit : T_TFRI_Pred<0, 1>;
609 def C2_cmovenewif : T_TFRI_Pred<1, 1>;
611 let InputType = "imm", isExtendable = 1, isExtentSigned = 1,
612 CextOpcode = "TFR", BaseOpcode = "TFRI", hasNewValue = 1, opNewValue = 0,
613 isAsCheapAsAMove = 1 , opExtendable = 1, opExtentBits = 16, isMoveImm = 1,
614 isPredicated = 0, isPredicable = 1, isReMaterializable = 1 in
615 def A2_tfrsi : ALU32Inst<(outs IntRegs:$Rd), (ins s16Ext:$s16), "$Rd = #$s16",
616 [(set (i32 IntRegs:$Rd), s16ExtPred:$s16)], "", ALU32_2op_tc_1_SLOT0123>,
622 let Inst{27-24} = 0b1000;
623 let Inst{23-22,20-16,13-5} = s16;
627 defm A2_tfr : tfr_base<"TFR">, ImmRegRel, PredNewRel;
628 let isAsmParserOnly = 1 in
629 defm A2_tfrp : TFR64_base<"TFR64">, PredNewRel;
632 let isReMaterializable = 1, isMoveImm = 1, isAsCheapAsAMove = 1,
633 isAsmParserOnly = 1 in
634 def A2_tfrpi : ALU64_rr<(outs DoubleRegs:$dst), (ins s8Imm64:$src1),
636 [(set (i64 DoubleRegs:$dst), s8Imm64Pred:$src1)]>;
638 // TODO: see if this instruction can be deleted..
639 let isExtendable = 1, opExtendable = 1, opExtentBits = 6,
640 isAsmParserOnly = 1 in
641 def TFRI64_V4 : ALU64_rr<(outs DoubleRegs:$dst), (ins u6Ext:$src1),
644 //===----------------------------------------------------------------------===//
646 //===----------------------------------------------------------------------===//
649 //===----------------------------------------------------------------------===//
651 //===----------------------------------------------------------------------===//
652 // Scalar mux register immediate.
653 let hasSideEffects = 0, isExtentSigned = 1, CextOpcode = "MUX",
654 InputType = "imm", hasNewValue = 1, isExtendable = 1, opExtentBits = 8 in
655 class T_MUX1 <bit MajOp, dag ins, string AsmStr>
656 : ALU32Inst <(outs IntRegs:$Rd), ins, AsmStr>, ImmRegRel {
663 let Inst{27-24} = 0b0011;
664 let Inst{23} = MajOp;
665 let Inst{22-21} = Pu;
666 let Inst{20-16} = Rs;
672 let opExtendable = 2 in
673 def C2_muxri : T_MUX1<0b1, (ins PredRegs:$Pu, s8Ext:$s8, IntRegs:$Rs),
674 "$Rd = mux($Pu, #$s8, $Rs)">;
676 let opExtendable = 3 in
677 def C2_muxir : T_MUX1<0b0, (ins PredRegs:$Pu, IntRegs:$Rs, s8Ext:$s8),
678 "$Rd = mux($Pu, $Rs, #$s8)">;
680 def : Pat<(i32 (select I1:$Pu, s8ExtPred:$s8, I32:$Rs)),
681 (C2_muxri I1:$Pu, s8ExtPred:$s8, I32:$Rs)>;
683 def : Pat<(i32 (select I1:$Pu, I32:$Rs, s8ExtPred:$s8)),
684 (C2_muxir I1:$Pu, I32:$Rs, s8ExtPred:$s8)>;
686 // C2_muxii: Scalar mux immediates.
687 let isExtentSigned = 1, hasNewValue = 1, isExtendable = 1,
688 opExtentBits = 8, opExtendable = 2 in
689 def C2_muxii: ALU32Inst <(outs IntRegs:$Rd),
690 (ins PredRegs:$Pu, s8Ext:$s8, s8Imm:$S8),
691 "$Rd = mux($Pu, #$s8, #$S8)" ,
692 [(set (i32 IntRegs:$Rd),
693 (i32 (select I1:$Pu, s8ExtPred:$s8, s8ImmPred:$S8)))] > {
701 let Inst{27-25} = 0b101;
702 let Inst{24-23} = Pu;
703 let Inst{22-16} = S8{7-1};
704 let Inst{13} = S8{0};
709 //===----------------------------------------------------------------------===//
710 // template class for non-predicated alu32_2op instructions
711 // - aslh, asrh, sxtb, sxth, zxth
712 //===----------------------------------------------------------------------===//
713 let hasNewValue = 1, opNewValue = 0 in
714 class T_ALU32_2op <string mnemonic, bits<3> minOp> :
715 ALU32Inst <(outs IntRegs:$Rd), (ins IntRegs:$Rs),
716 "$Rd = "#mnemonic#"($Rs)", [] > {
722 let Inst{27-24} = 0b0000;
723 let Inst{23-21} = minOp;
726 let Inst{20-16} = Rs;
729 //===----------------------------------------------------------------------===//
730 // template class for predicated alu32_2op instructions
731 // - aslh, asrh, sxtb, sxth, zxtb, zxth
732 //===----------------------------------------------------------------------===//
733 let hasSideEffects = 0, hasNewValue = 1, opNewValue = 0 in
734 class T_ALU32_2op_Pred <string mnemonic, bits<3> minOp, bit isPredNot,
736 ALU32Inst <(outs IntRegs:$Rd), (ins PredRegs:$Pu, IntRegs:$Rs),
737 !if(isPredNot, "if (!$Pu", "if ($Pu")
738 #!if(isPredNew, ".new) ",") ")#"$Rd = "#mnemonic#"($Rs)"> {
745 let Inst{27-24} = 0b0000;
746 let Inst{23-21} = minOp;
748 let Inst{11} = isPredNot;
749 let Inst{10} = isPredNew;
752 let Inst{20-16} = Rs;
755 multiclass ALU32_2op_Pred<string mnemonic, bits<3> minOp, bit PredNot> {
756 let isPredicatedFalse = PredNot in {
757 def NAME : T_ALU32_2op_Pred<mnemonic, minOp, PredNot, 0>;
760 let isPredicatedNew = 1 in
761 def NAME#new : T_ALU32_2op_Pred<mnemonic, minOp, PredNot, 1>;
765 multiclass ALU32_2op_base<string mnemonic, bits<3> minOp> {
766 let BaseOpcode = mnemonic in {
767 let isPredicable = 1, hasSideEffects = 0 in
768 def A2_#NAME : T_ALU32_2op<mnemonic, minOp>;
770 let isPredicated = 1, hasSideEffects = 0 in {
771 defm A4_p#NAME#t : ALU32_2op_Pred<mnemonic, minOp, 0>;
772 defm A4_p#NAME#f : ALU32_2op_Pred<mnemonic, minOp, 1>;
777 defm aslh : ALU32_2op_base<"aslh", 0b000>, PredNewRel;
778 defm asrh : ALU32_2op_base<"asrh", 0b001>, PredNewRel;
779 defm sxtb : ALU32_2op_base<"sxtb", 0b101>, PredNewRel;
780 defm sxth : ALU32_2op_base<"sxth", 0b111>, PredNewRel;
781 defm zxth : ALU32_2op_base<"zxth", 0b110>, PredNewRel;
783 // Rd=zxtb(Rs): assembler mapped to Rd=and(Rs,#255).
784 // Compiler would want to generate 'zxtb' instead of 'and' becuase 'zxtb' has
785 // predicated forms while 'and' doesn't. Since integrated assembler can't
786 // handle 'mapped' instructions, we need to encode 'zxtb' same as 'and' where
787 // immediate operand is set to '255'.
789 let hasNewValue = 1, opNewValue = 0 in
790 class T_ZXTB: ALU32Inst < (outs IntRegs:$Rd), (ins IntRegs:$Rs),
791 "$Rd = zxtb($Rs)", [] > { // Rd = and(Rs,255)
798 let Inst{27-22} = 0b011000;
800 let Inst{20-16} = Rs;
801 let Inst{21} = s10{9};
802 let Inst{13-5} = s10{8-0};
805 //Rd=zxtb(Rs): assembler mapped to "Rd=and(Rs,#255)
806 multiclass ZXTB_base <string mnemonic, bits<3> minOp> {
807 let BaseOpcode = mnemonic in {
808 let isPredicable = 1, hasSideEffects = 0 in
809 def A2_#NAME : T_ZXTB;
811 let isPredicated = 1, hasSideEffects = 0 in {
812 defm A4_p#NAME#t : ALU32_2op_Pred<mnemonic, minOp, 0>;
813 defm A4_p#NAME#f : ALU32_2op_Pred<mnemonic, minOp, 1>;
818 defm zxtb : ZXTB_base<"zxtb",0b100>, PredNewRel;
820 def: Pat<(shl I32:$src1, (i32 16)), (A2_aslh I32:$src1)>;
821 def: Pat<(sra I32:$src1, (i32 16)), (A2_asrh I32:$src1)>;
822 def: Pat<(sext_inreg I32:$src1, i8), (A2_sxtb I32:$src1)>;
823 def: Pat<(sext_inreg I32:$src1, i16), (A2_sxth I32:$src1)>;
825 //===----------------------------------------------------------------------===//
826 // Template class for vector add and avg
827 //===----------------------------------------------------------------------===//
829 class T_VectALU_64 <string opc, bits<3> majOp, bits<3> minOp,
830 bit isSat, bit isRnd, bit isCrnd, bit SwapOps >
831 : ALU64_rr < (outs DoubleRegs:$Rdd),
832 (ins DoubleRegs:$Rss, DoubleRegs:$Rtt),
833 "$Rdd = "#opc#"($Rss, $Rtt)"#!if(isRnd, ":rnd", "")
834 #!if(isCrnd,":crnd","")
835 #!if(isSat, ":sat", ""),
836 [], "", ALU64_tc_2_SLOT23 > {
843 let Inst{27-24} = 0b0011;
844 let Inst{23-21} = majOp;
845 let Inst{20-16} = !if (SwapOps, Rtt, Rss);
846 let Inst{12-8} = !if (SwapOps, Rss, Rtt);
847 let Inst{7-5} = minOp;
851 // ALU64 - Vector add
852 // Rdd=vadd[u][bhw](Rss,Rtt)
853 let Itinerary = ALU64_tc_1_SLOT23 in {
854 def A2_vaddub : T_VectALU_64 < "vaddub", 0b000, 0b000, 0, 0, 0, 0>;
855 def A2_vaddh : T_VectALU_64 < "vaddh", 0b000, 0b010, 0, 0, 0, 0>;
856 def A2_vaddw : T_VectALU_64 < "vaddw", 0b000, 0b101, 0, 0, 0, 0>;
859 // Rdd=vadd[u][bhw](Rss,Rtt):sat
860 let Defs = [USR_OVF] in {
861 def A2_vaddubs : T_VectALU_64 < "vaddub", 0b000, 0b001, 1, 0, 0, 0>;
862 def A2_vaddhs : T_VectALU_64 < "vaddh", 0b000, 0b011, 1, 0, 0, 0>;
863 def A2_vadduhs : T_VectALU_64 < "vadduh", 0b000, 0b100, 1, 0, 0, 0>;
864 def A2_vaddws : T_VectALU_64 < "vaddw", 0b000, 0b110, 1, 0, 0, 0>;
867 // ALU64 - Vector average
868 // Rdd=vavg[u][bhw](Rss,Rtt)
869 let Itinerary = ALU64_tc_1_SLOT23 in {
870 def A2_vavgub : T_VectALU_64 < "vavgub", 0b010, 0b000, 0, 0, 0, 0>;
871 def A2_vavgh : T_VectALU_64 < "vavgh", 0b010, 0b010, 0, 0, 0, 0>;
872 def A2_vavguh : T_VectALU_64 < "vavguh", 0b010, 0b101, 0, 0, 0, 0>;
873 def A2_vavgw : T_VectALU_64 < "vavgw", 0b011, 0b000, 0, 0, 0, 0>;
874 def A2_vavguw : T_VectALU_64 < "vavguw", 0b011, 0b011, 0, 0, 0, 0>;
877 // Rdd=vavg[u][bhw](Rss,Rtt)[:rnd|:crnd]
878 def A2_vavgubr : T_VectALU_64 < "vavgub", 0b010, 0b001, 0, 1, 0, 0>;
879 def A2_vavghr : T_VectALU_64 < "vavgh", 0b010, 0b011, 0, 1, 0, 0>;
880 def A2_vavghcr : T_VectALU_64 < "vavgh", 0b010, 0b100, 0, 0, 1, 0>;
881 def A2_vavguhr : T_VectALU_64 < "vavguh", 0b010, 0b110, 0, 1, 0, 0>;
883 def A2_vavgwr : T_VectALU_64 < "vavgw", 0b011, 0b001, 0, 1, 0, 0>;
884 def A2_vavgwcr : T_VectALU_64 < "vavgw", 0b011, 0b010, 0, 0, 1, 0>;
885 def A2_vavguwr : T_VectALU_64 < "vavguw", 0b011, 0b100, 0, 1, 0, 0>;
887 // Rdd=vnavg[bh](Rss,Rtt)
888 let Itinerary = ALU64_tc_1_SLOT23 in {
889 def A2_vnavgh : T_VectALU_64 < "vnavgh", 0b100, 0b000, 0, 0, 0, 1>;
890 def A2_vnavgw : T_VectALU_64 < "vnavgw", 0b100, 0b011, 0, 0, 0, 1>;
893 // Rdd=vnavg[bh](Rss,Rtt)[:rnd|:crnd]:sat
894 let Defs = [USR_OVF] in {
895 def A2_vnavghr : T_VectALU_64 < "vnavgh", 0b100, 0b001, 1, 1, 0, 1>;
896 def A2_vnavghcr : T_VectALU_64 < "vnavgh", 0b100, 0b010, 1, 0, 1, 1>;
897 def A2_vnavgwr : T_VectALU_64 < "vnavgw", 0b100, 0b100, 1, 1, 0, 1>;
898 def A2_vnavgwcr : T_VectALU_64 < "vnavgw", 0b100, 0b110, 1, 0, 1, 1>;
901 // Rdd=vsub[u][bh](Rss,Rtt)
902 let Itinerary = ALU64_tc_1_SLOT23 in {
903 def A2_vsubub : T_VectALU_64 < "vsubub", 0b001, 0b000, 0, 0, 0, 1>;
904 def A2_vsubh : T_VectALU_64 < "vsubh", 0b001, 0b010, 0, 0, 0, 1>;
905 def A2_vsubw : T_VectALU_64 < "vsubw", 0b001, 0b101, 0, 0, 0, 1>;
908 // Rdd=vsub[u][bh](Rss,Rtt):sat
909 let Defs = [USR_OVF] in {
910 def A2_vsububs : T_VectALU_64 < "vsubub", 0b001, 0b001, 1, 0, 0, 1>;
911 def A2_vsubhs : T_VectALU_64 < "vsubh", 0b001, 0b011, 1, 0, 0, 1>;
912 def A2_vsubuhs : T_VectALU_64 < "vsubuh", 0b001, 0b100, 1, 0, 0, 1>;
913 def A2_vsubws : T_VectALU_64 < "vsubw", 0b001, 0b110, 1, 0, 0, 1>;
916 // Rdd=vmax[u][bhw](Rss,Rtt)
917 def A2_vmaxb : T_VectALU_64 < "vmaxb", 0b110, 0b110, 0, 0, 0, 1>;
918 def A2_vmaxub : T_VectALU_64 < "vmaxub", 0b110, 0b000, 0, 0, 0, 1>;
919 def A2_vmaxh : T_VectALU_64 < "vmaxh", 0b110, 0b001, 0, 0, 0, 1>;
920 def A2_vmaxuh : T_VectALU_64 < "vmaxuh", 0b110, 0b010, 0, 0, 0, 1>;
921 def A2_vmaxw : T_VectALU_64 < "vmaxw", 0b110, 0b011, 0, 0, 0, 1>;
922 def A2_vmaxuw : T_VectALU_64 < "vmaxuw", 0b101, 0b101, 0, 0, 0, 1>;
924 // Rdd=vmin[u][bhw](Rss,Rtt)
925 def A2_vminb : T_VectALU_64 < "vminb", 0b110, 0b111, 0, 0, 0, 1>;
926 def A2_vminub : T_VectALU_64 < "vminub", 0b101, 0b000, 0, 0, 0, 1>;
927 def A2_vminh : T_VectALU_64 < "vminh", 0b101, 0b001, 0, 0, 0, 1>;
928 def A2_vminuh : T_VectALU_64 < "vminuh", 0b101, 0b010, 0, 0, 0, 1>;
929 def A2_vminw : T_VectALU_64 < "vminw", 0b101, 0b011, 0, 0, 0, 1>;
930 def A2_vminuw : T_VectALU_64 < "vminuw", 0b101, 0b100, 0, 0, 0, 1>;
932 //===----------------------------------------------------------------------===//
933 // Template class for vector compare
934 //===----------------------------------------------------------------------===//
935 let hasSideEffects = 0 in
936 class T_vcmp <string Str, bits<4> minOp>
937 : ALU64_rr <(outs PredRegs:$Pd),
938 (ins DoubleRegs:$Rss, DoubleRegs:$Rtt),
939 "$Pd = "#Str#"($Rss, $Rtt)", [],
940 "", ALU64_tc_2early_SLOT23> {
947 let Inst{27-23} = 0b00100;
948 let Inst{13} = minOp{3};
949 let Inst{7-5} = minOp{2-0};
951 let Inst{20-16} = Rss;
952 let Inst{12-8} = Rtt;
955 class T_vcmp_pat<InstHexagon MI, PatFrag Op, ValueType T>
956 : Pat<(i1 (Op (T DoubleRegs:$Rss), (T DoubleRegs:$Rtt))),
957 (i1 (MI DoubleRegs:$Rss, DoubleRegs:$Rtt))>;
959 // Vector compare bytes
960 def A2_vcmpbeq : T_vcmp <"vcmpb.eq", 0b0110>;
961 def A2_vcmpbgtu : T_vcmp <"vcmpb.gtu", 0b0111>;
963 // Vector compare halfwords
964 def A2_vcmpheq : T_vcmp <"vcmph.eq", 0b0011>;
965 def A2_vcmphgt : T_vcmp <"vcmph.gt", 0b0100>;
966 def A2_vcmphgtu : T_vcmp <"vcmph.gtu", 0b0101>;
968 // Vector compare words
969 def A2_vcmpweq : T_vcmp <"vcmpw.eq", 0b0000>;
970 def A2_vcmpwgt : T_vcmp <"vcmpw.gt", 0b0001>;
971 def A2_vcmpwgtu : T_vcmp <"vcmpw.gtu", 0b0010>;
973 def: T_vcmp_pat<A2_vcmpbeq, seteq, v8i8>;
974 def: T_vcmp_pat<A2_vcmpbgtu, setugt, v8i8>;
975 def: T_vcmp_pat<A2_vcmpheq, seteq, v4i16>;
976 def: T_vcmp_pat<A2_vcmphgt, setgt, v4i16>;
977 def: T_vcmp_pat<A2_vcmphgtu, setugt, v4i16>;
978 def: T_vcmp_pat<A2_vcmpweq, seteq, v2i32>;
979 def: T_vcmp_pat<A2_vcmpwgt, setgt, v2i32>;
980 def: T_vcmp_pat<A2_vcmpwgtu, setugt, v2i32>;
982 //===----------------------------------------------------------------------===//
984 //===----------------------------------------------------------------------===//
987 //===----------------------------------------------------------------------===//
989 //===----------------------------------------------------------------------===//
991 //===----------------------------------------------------------------------===//
993 //===----------------------------------------------------------------------===//
996 //===----------------------------------------------------------------------===//
998 //===----------------------------------------------------------------------===//
1000 //===----------------------------------------------------------------------===//
1002 // Add/Subtract halfword
1003 // Rd=add(Rt.L,Rs.[HL])[:sat]
1004 // Rd=sub(Rt.L,Rs.[HL])[:sat]
1005 // Rd=add(Rt.[LH],Rs.[HL])[:sat][:<16]
1006 // Rd=sub(Rt.[LH],Rs.[HL])[:sat][:<16]
1007 //===----------------------------------------------------------------------===//
1009 let hasNewValue = 1, opNewValue = 0 in
1010 class T_XTYPE_ADD_SUB <bits<2> LHbits, bit isSat, bit hasShift, bit isSub>
1011 : ALU64Inst <(outs IntRegs:$Rd), (ins IntRegs:$Rt, IntRegs:$Rs),
1012 "$Rd = "#!if(isSub,"sub","add")#"($Rt."
1013 #!if(hasShift, !if(LHbits{1},"h","l"),"l") #", $Rs."
1014 #!if(hasShift, !if(LHbits{0},"h)","l)"), !if(LHbits{1},"h)","l)"))
1015 #!if(isSat,":sat","")
1016 #!if(hasShift,":<<16",""), [], "", ALU64_tc_1_SLOT23> {
1020 let IClass = 0b1101;
1022 let Inst{27-23} = 0b01010;
1023 let Inst{22} = hasShift;
1024 let Inst{21} = isSub;
1025 let Inst{7} = isSat;
1026 let Inst{6-5} = LHbits;
1028 let Inst{12-8} = Rt;
1029 let Inst{20-16} = Rs;
1032 //Rd=sub(Rt.L,Rs.[LH])
1033 def A2_subh_l16_ll : T_XTYPE_ADD_SUB <0b00, 0, 0, 1>;
1034 def A2_subh_l16_hl : T_XTYPE_ADD_SUB <0b10, 0, 0, 1>;
1036 //Rd=add(Rt.L,Rs.[LH])
1037 def A2_addh_l16_ll : T_XTYPE_ADD_SUB <0b00, 0, 0, 0>;
1038 def A2_addh_l16_hl : T_XTYPE_ADD_SUB <0b10, 0, 0, 0>;
1040 let Itinerary = ALU64_tc_2_SLOT23, Defs = [USR_OVF] in {
1041 //Rd=sub(Rt.L,Rs.[LH]):sat
1042 def A2_subh_l16_sat_ll : T_XTYPE_ADD_SUB <0b00, 1, 0, 1>;
1043 def A2_subh_l16_sat_hl : T_XTYPE_ADD_SUB <0b10, 1, 0, 1>;
1045 //Rd=add(Rt.L,Rs.[LH]):sat
1046 def A2_addh_l16_sat_ll : T_XTYPE_ADD_SUB <0b00, 1, 0, 0>;
1047 def A2_addh_l16_sat_hl : T_XTYPE_ADD_SUB <0b10, 1, 0, 0>;
1050 //Rd=sub(Rt.[LH],Rs.[LH]):<<16
1051 def A2_subh_h16_ll : T_XTYPE_ADD_SUB <0b00, 0, 1, 1>;
1052 def A2_subh_h16_lh : T_XTYPE_ADD_SUB <0b01, 0, 1, 1>;
1053 def A2_subh_h16_hl : T_XTYPE_ADD_SUB <0b10, 0, 1, 1>;
1054 def A2_subh_h16_hh : T_XTYPE_ADD_SUB <0b11, 0, 1, 1>;
1056 //Rd=add(Rt.[LH],Rs.[LH]):<<16
1057 def A2_addh_h16_ll : T_XTYPE_ADD_SUB <0b00, 0, 1, 0>;
1058 def A2_addh_h16_lh : T_XTYPE_ADD_SUB <0b01, 0, 1, 0>;
1059 def A2_addh_h16_hl : T_XTYPE_ADD_SUB <0b10, 0, 1, 0>;
1060 def A2_addh_h16_hh : T_XTYPE_ADD_SUB <0b11, 0, 1, 0>;
1062 let Itinerary = ALU64_tc_2_SLOT23, Defs = [USR_OVF] in {
1063 //Rd=sub(Rt.[LH],Rs.[LH]):sat:<<16
1064 def A2_subh_h16_sat_ll : T_XTYPE_ADD_SUB <0b00, 1, 1, 1>;
1065 def A2_subh_h16_sat_lh : T_XTYPE_ADD_SUB <0b01, 1, 1, 1>;
1066 def A2_subh_h16_sat_hl : T_XTYPE_ADD_SUB <0b10, 1, 1, 1>;
1067 def A2_subh_h16_sat_hh : T_XTYPE_ADD_SUB <0b11, 1, 1, 1>;
1069 //Rd=add(Rt.[LH],Rs.[LH]):sat:<<16
1070 def A2_addh_h16_sat_ll : T_XTYPE_ADD_SUB <0b00, 1, 1, 0>;
1071 def A2_addh_h16_sat_lh : T_XTYPE_ADD_SUB <0b01, 1, 1, 0>;
1072 def A2_addh_h16_sat_hl : T_XTYPE_ADD_SUB <0b10, 1, 1, 0>;
1073 def A2_addh_h16_sat_hh : T_XTYPE_ADD_SUB <0b11, 1, 1, 0>;
1077 def: Pat<(sext_inreg (add I32:$src1, I32:$src2), i16),
1078 (A2_addh_l16_ll I32:$src1, I32:$src2)>;
1080 def: Pat<(sra (add (shl I32:$src1, (i32 16)), I32:$src2), (i32 16)),
1081 (A2_addh_l16_hl I32:$src1, I32:$src2)>;
1083 def: Pat<(shl (add I32:$src1, I32:$src2), (i32 16)),
1084 (A2_addh_h16_ll I32:$src1, I32:$src2)>;
1086 // Subtract halfword.
1087 def: Pat<(sext_inreg (sub I32:$src1, I32:$src2), i16),
1088 (A2_subh_l16_ll I32:$src1, I32:$src2)>;
1090 def: Pat<(shl (sub I32:$src1, I32:$src2), (i32 16)),
1091 (A2_subh_h16_ll I32:$src1, I32:$src2)>;
1093 let hasSideEffects = 0, hasNewValue = 1 in
1094 def S2_parityp: ALU64Inst<(outs IntRegs:$Rd),
1095 (ins DoubleRegs:$Rs, DoubleRegs:$Rt),
1096 "$Rd = parity($Rs, $Rt)", [], "", ALU64_tc_2_SLOT23> {
1101 let IClass = 0b1101;
1102 let Inst{27-24} = 0b0000;
1103 let Inst{20-16} = Rs;
1104 let Inst{12-8} = Rt;
1108 let hasNewValue = 1, opNewValue = 0, hasSideEffects = 0 in
1109 class T_XTYPE_MIN_MAX < bit isMax, bit isUnsigned >
1110 : ALU64Inst < (outs IntRegs:$Rd), (ins IntRegs:$Rt, IntRegs:$Rs),
1111 "$Rd = "#!if(isMax,"max","min")#!if(isUnsigned,"u","")
1112 #"($Rt, $Rs)", [], "", ALU64_tc_2_SLOT23> {
1117 let IClass = 0b1101;
1119 let Inst{27-23} = 0b01011;
1120 let Inst{22-21} = !if(isMax, 0b10, 0b01);
1121 let Inst{7} = isUnsigned;
1123 let Inst{12-8} = !if(isMax, Rs, Rt);
1124 let Inst{20-16} = !if(isMax, Rt, Rs);
1127 def A2_min : T_XTYPE_MIN_MAX < 0, 0 >;
1128 def A2_minu : T_XTYPE_MIN_MAX < 0, 1 >;
1129 def A2_max : T_XTYPE_MIN_MAX < 1, 0 >;
1130 def A2_maxu : T_XTYPE_MIN_MAX < 1, 1 >;
1132 // Here, depending on the operand being selected, we'll either generate a
1133 // min or max instruction.
1135 // (a>b)?a:b --> max(a,b) => Here check performed is '>' and the value selected
1136 // is the larger of two. So, the corresponding HexagonInst is passed in 'Inst'.
1137 // (a>b)?b:a --> min(a,b) => Here check performed is '>' but the smaller value
1138 // is selected and the corresponding HexagonInst is passed in 'SwapInst'.
1140 multiclass T_MinMax_pats <PatFrag Op, RegisterClass RC, ValueType VT,
1141 InstHexagon Inst, InstHexagon SwapInst> {
1142 def: Pat<(select (i1 (Op (VT RC:$src1), (VT RC:$src2))),
1143 (VT RC:$src1), (VT RC:$src2)),
1144 (Inst RC:$src1, RC:$src2)>;
1145 def: Pat<(select (i1 (Op (VT RC:$src1), (VT RC:$src2))),
1146 (VT RC:$src2), (VT RC:$src1)),
1147 (SwapInst RC:$src1, RC:$src2)>;
1151 multiclass MinMax_pats <PatFrag Op, InstHexagon Inst, InstHexagon SwapInst> {
1152 defm: T_MinMax_pats<Op, IntRegs, i32, Inst, SwapInst>;
1154 def: Pat<(sext_inreg (i32 (select (i1 (Op (i32 PositiveHalfWord:$src1),
1155 (i32 PositiveHalfWord:$src2))),
1156 (i32 PositiveHalfWord:$src1),
1157 (i32 PositiveHalfWord:$src2))), i16),
1158 (Inst IntRegs:$src1, IntRegs:$src2)>;
1160 def: Pat<(sext_inreg (i32 (select (i1 (Op (i32 PositiveHalfWord:$src1),
1161 (i32 PositiveHalfWord:$src2))),
1162 (i32 PositiveHalfWord:$src2),
1163 (i32 PositiveHalfWord:$src1))), i16),
1164 (SwapInst IntRegs:$src1, IntRegs:$src2)>;
1167 let AddedComplexity = 200 in {
1168 defm: MinMax_pats<setge, A2_max, A2_min>;
1169 defm: MinMax_pats<setgt, A2_max, A2_min>;
1170 defm: MinMax_pats<setle, A2_min, A2_max>;
1171 defm: MinMax_pats<setlt, A2_min, A2_max>;
1172 defm: MinMax_pats<setuge, A2_maxu, A2_minu>;
1173 defm: MinMax_pats<setugt, A2_maxu, A2_minu>;
1174 defm: MinMax_pats<setule, A2_minu, A2_maxu>;
1175 defm: MinMax_pats<setult, A2_minu, A2_maxu>;
1178 class T_cmp64_rr<string mnemonic, bits<3> MinOp, bit IsComm>
1179 : ALU64_rr<(outs PredRegs:$Pd), (ins DoubleRegs:$Rs, DoubleRegs:$Rt),
1180 "$Pd = "#mnemonic#"($Rs, $Rt)", [], "", ALU64_tc_2early_SLOT23> {
1182 let isCommutable = IsComm;
1183 let hasSideEffects = 0;
1189 let IClass = 0b1101;
1190 let Inst{27-21} = 0b0010100;
1191 let Inst{20-16} = Rs;
1192 let Inst{12-8} = Rt;
1193 let Inst{7-5} = MinOp;
1197 def C2_cmpeqp : T_cmp64_rr<"cmp.eq", 0b000, 1>;
1198 def C2_cmpgtp : T_cmp64_rr<"cmp.gt", 0b010, 0>;
1199 def C2_cmpgtup : T_cmp64_rr<"cmp.gtu", 0b100, 0>;
1201 class T_cmp64_rr_pat<InstHexagon MI, PatFrag CmpOp>
1202 : Pat<(i1 (CmpOp (i64 DoubleRegs:$Rs), (i64 DoubleRegs:$Rt))),
1203 (i1 (MI DoubleRegs:$Rs, DoubleRegs:$Rt))>;
1205 def: T_cmp64_rr_pat<C2_cmpeqp, seteq>;
1206 def: T_cmp64_rr_pat<C2_cmpgtp, setgt>;
1207 def: T_cmp64_rr_pat<C2_cmpgtup, setugt>;
1208 def: T_cmp64_rr_pat<C2_cmpgtp, RevCmp<setlt>>;
1209 def: T_cmp64_rr_pat<C2_cmpgtup, RevCmp<setult>>;
1211 def C2_vmux : ALU64_rr<(outs DoubleRegs:$Rd),
1212 (ins PredRegs:$Pu, DoubleRegs:$Rs, DoubleRegs:$Rt),
1213 "$Rd = vmux($Pu, $Rs, $Rt)", [], "", ALU64_tc_1_SLOT23> {
1214 let hasSideEffects = 0;
1221 let IClass = 0b1101;
1222 let Inst{27-24} = 0b0001;
1223 let Inst{20-16} = Rs;
1224 let Inst{12-8} = Rt;
1229 class T_ALU64_rr<string mnemonic, string suffix, bits<4> RegType,
1230 bits<3> MajOp, bits<3> MinOp, bit OpsRev, bit IsComm,
1232 : ALU64_rr<(outs DoubleRegs:$Rd), (ins DoubleRegs:$Rs, DoubleRegs:$Rt),
1233 "$Rd = " #mnemonic# "($Rs, " #Op2Pfx# "$Rt)" #suffix, [],
1234 "", ALU64_tc_1_SLOT23> {
1235 let hasSideEffects = 0;
1236 let isCommutable = IsComm;
1242 let IClass = 0b1101;
1243 let Inst{27-24} = RegType;
1244 let Inst{23-21} = MajOp;
1245 let Inst{20-16} = !if (OpsRev,Rt,Rs);
1246 let Inst{12-8} = !if (OpsRev,Rs,Rt);
1247 let Inst{7-5} = MinOp;
1251 class T_ALU64_arith<string mnemonic, bits<3> MajOp, bits<3> MinOp, bit IsSat,
1252 bit OpsRev, bit IsComm>
1253 : T_ALU64_rr<mnemonic, !if(IsSat,":sat",""), 0b0011, MajOp, MinOp, OpsRev,
1256 def A2_addp : T_ALU64_arith<"add", 0b000, 0b111, 0, 0, 1>;
1257 def A2_subp : T_ALU64_arith<"sub", 0b001, 0b111, 0, 1, 0>;
1259 def: Pat<(i64 (add I64:$Rs, I64:$Rt)), (A2_addp I64:$Rs, I64:$Rt)>;
1260 def: Pat<(i64 (sub I64:$Rs, I64:$Rt)), (A2_subp I64:$Rs, I64:$Rt)>;
1262 class T_ALU64_logical<string mnemonic, bits<3> MinOp, bit OpsRev, bit IsComm,
1264 : T_ALU64_rr<mnemonic, "", 0b0011, 0b111, MinOp, OpsRev, IsComm,
1267 def A2_andp : T_ALU64_logical<"and", 0b000, 0, 1, 0>;
1268 def A2_orp : T_ALU64_logical<"or", 0b010, 0, 1, 0>;
1269 def A2_xorp : T_ALU64_logical<"xor", 0b100, 0, 1, 0>;
1271 def: Pat<(i64 (and I64:$Rs, I64:$Rt)), (A2_andp I64:$Rs, I64:$Rt)>;
1272 def: Pat<(i64 (or I64:$Rs, I64:$Rt)), (A2_orp I64:$Rs, I64:$Rt)>;
1273 def: Pat<(i64 (xor I64:$Rs, I64:$Rt)), (A2_xorp I64:$Rs, I64:$Rt)>;
1275 //===----------------------------------------------------------------------===//
1277 //===----------------------------------------------------------------------===//
1279 //===----------------------------------------------------------------------===//
1281 //===----------------------------------------------------------------------===//
1283 //===----------------------------------------------------------------------===//
1285 //===----------------------------------------------------------------------===//
1287 //===----------------------------------------------------------------------===//
1289 //===----------------------------------------------------------------------===//
1291 //===----------------------------------------------------------------------===//
1293 //===----------------------------------------------------------------------===//
1295 //===----------------------------------------------------------------------===//
1297 //===----------------------------------------------------------------------===//
1298 // Logical reductions on predicates.
1300 // Looping instructions.
1302 // Pipelined looping instructions.
1304 // Logical operations on predicates.
1305 let hasSideEffects = 0 in
1306 class T_LOGICAL_1OP<string MnOp, bits<2> OpBits>
1307 : CRInst<(outs PredRegs:$Pd), (ins PredRegs:$Ps),
1308 "$Pd = " # MnOp # "($Ps)", [], "", CR_tc_2early_SLOT23> {
1312 let IClass = 0b0110;
1313 let Inst{27-23} = 0b10111;
1314 let Inst{22-21} = OpBits;
1316 let Inst{17-16} = Ps;
1321 def C2_any8 : T_LOGICAL_1OP<"any8", 0b00>;
1322 def C2_all8 : T_LOGICAL_1OP<"all8", 0b01>;
1323 def C2_not : T_LOGICAL_1OP<"not", 0b10>;
1325 def: Pat<(i1 (not (i1 PredRegs:$Ps))),
1326 (C2_not PredRegs:$Ps)>;
1328 let hasSideEffects = 0 in
1329 class T_LOGICAL_2OP<string MnOp, bits<3> OpBits, bit IsNeg, bit Rev>
1330 : CRInst<(outs PredRegs:$Pd), (ins PredRegs:$Ps, PredRegs:$Pt),
1331 "$Pd = " # MnOp # "($Ps, " # !if (IsNeg,"!","") # "$Pt)",
1332 [], "", CR_tc_2early_SLOT23> {
1337 let IClass = 0b0110;
1338 let Inst{27-24} = 0b1011;
1339 let Inst{23-21} = OpBits;
1341 let Inst{17-16} = !if(Rev,Pt,Ps); // Rs and Rt are reversed for some
1342 let Inst{13} = 0b0; // instructions.
1343 let Inst{9-8} = !if(Rev,Ps,Pt);
1347 def C2_and : T_LOGICAL_2OP<"and", 0b000, 0, 1>;
1348 def C2_or : T_LOGICAL_2OP<"or", 0b001, 0, 1>;
1349 def C2_xor : T_LOGICAL_2OP<"xor", 0b010, 0, 0>;
1350 def C2_andn : T_LOGICAL_2OP<"and", 0b011, 1, 1>;
1351 def C2_orn : T_LOGICAL_2OP<"or", 0b111, 1, 1>;
1353 def: Pat<(i1 (and I1:$Ps, I1:$Pt)), (C2_and I1:$Ps, I1:$Pt)>;
1354 def: Pat<(i1 (or I1:$Ps, I1:$Pt)), (C2_or I1:$Ps, I1:$Pt)>;
1355 def: Pat<(i1 (xor I1:$Ps, I1:$Pt)), (C2_xor I1:$Ps, I1:$Pt)>;
1356 def: Pat<(i1 (and I1:$Ps, (not I1:$Pt))), (C2_andn I1:$Ps, I1:$Pt)>;
1357 def: Pat<(i1 (or I1:$Ps, (not I1:$Pt))), (C2_orn I1:$Ps, I1:$Pt)>;
1359 let hasSideEffects = 0, hasNewValue = 1 in
1360 def C2_vitpack : SInst<(outs IntRegs:$Rd), (ins PredRegs:$Ps, PredRegs:$Pt),
1361 "$Rd = vitpack($Ps, $Pt)", [], "", S_2op_tc_1_SLOT23> {
1366 let IClass = 0b1000;
1367 let Inst{27-24} = 0b1001;
1368 let Inst{22-21} = 0b00;
1369 let Inst{17-16} = Ps;
1374 let hasSideEffects = 0 in
1375 def C2_mask : SInst<(outs DoubleRegs:$Rd), (ins PredRegs:$Pt),
1376 "$Rd = mask($Pt)", [], "", S_2op_tc_1_SLOT23> {
1380 let IClass = 0b1000;
1381 let Inst{27-24} = 0b0110;
1386 // User control register transfer.
1387 //===----------------------------------------------------------------------===//
1389 //===----------------------------------------------------------------------===//
1391 //===----------------------------------------------------------------------===//
1393 //===----------------------------------------------------------------------===//
1395 def retflag : SDNode<"HexagonISD::RET_FLAG", SDTNone,
1396 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
1397 def eh_return: SDNode<"HexagonISD::EH_RETURN", SDTNone, [SDNPHasChain]>;
1399 def SDHexagonBR_JT: SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
1400 def HexagonBR_JT: SDNode<"HexagonISD::BR_JT", SDHexagonBR_JT, [SDNPHasChain]>;
1402 class CondStr<string CReg, bit True, bit New> {
1403 string S = "if (" # !if(True,"","!") # CReg # !if(New,".new","") # ") ";
1405 class JumpOpcStr<string Mnemonic, bit New, bit Taken> {
1406 string S = Mnemonic # !if(Taken, ":t", !if(New, ":nt", ""));
1409 let isBranch = 1, isBarrier = 1, Defs = [PC], hasSideEffects = 0,
1411 isExtendable = 1, opExtendable = 0, isExtentSigned = 1,
1412 opExtentBits = 24, opExtentAlign = 2, InputType = "imm" in
1413 class T_JMP<string ExtStr>
1414 : JInst<(outs), (ins brtarget:$dst),
1415 "jump " # ExtStr # "$dst",
1416 [], "", J_tc_2early_SLOT23> {
1418 let IClass = 0b0101;
1420 let Inst{27-25} = 0b100;
1421 let Inst{24-16} = dst{23-15};
1422 let Inst{13-1} = dst{14-2};
1425 let isBranch = 1, Defs = [PC], hasSideEffects = 0, isPredicated = 1,
1426 isExtendable = 1, opExtendable = 1, isExtentSigned = 1,
1427 opExtentBits = 17, opExtentAlign = 2, InputType = "imm" in
1428 class T_JMP_c<bit PredNot, bit isPredNew, bit isTak, string ExtStr>
1429 : JInst<(outs), (ins PredRegs:$src, brtarget:$dst),
1430 CondStr<"$src", !if(PredNot,0,1), isPredNew>.S #
1431 JumpOpcStr<"jump", isPredNew, isTak>.S # " " #
1433 [], "", J_tc_2early_SLOT23>, ImmRegRel {
1434 let isTaken = isTak;
1435 let isPredicatedFalse = PredNot;
1436 let isPredicatedNew = isPredNew;
1440 let IClass = 0b0101;
1442 let Inst{27-24} = 0b1100;
1443 let Inst{21} = PredNot;
1444 let Inst{12} = isTak;
1445 let Inst{11} = isPredNew;
1446 let Inst{9-8} = src;
1447 let Inst{23-22} = dst{16-15};
1448 let Inst{20-16} = dst{14-10};
1449 let Inst{13} = dst{9};
1450 let Inst{7-1} = dst{8-2};
1453 multiclass JMP_Pred<bit PredNot, string ExtStr> {
1454 def NAME : T_JMP_c<PredNot, 0, 0, ExtStr>; // not taken
1456 def NAME#newpt : T_JMP_c<PredNot, 1, 1, ExtStr>; // taken
1457 def NAME#new : T_JMP_c<PredNot, 1, 0, ExtStr>; // not taken
1460 multiclass JMP_base<string BaseOp, string ExtStr> {
1461 let BaseOpcode = BaseOp in {
1462 def NAME : T_JMP<ExtStr>;
1463 defm t : JMP_Pred<0, ExtStr>;
1464 defm f : JMP_Pred<1, ExtStr>;
1468 // Jumps to address stored in a register, JUMPR_MISC
1469 // if ([[!]P[.new]]) jumpr[:t/nt] Rs
1470 let isBranch = 1, isIndirectBranch = 1, isBarrier = 1, Defs = [PC],
1471 isPredicable = 1, hasSideEffects = 0, InputType = "reg" in
1473 : JRInst<(outs), (ins IntRegs:$dst),
1474 "jumpr $dst", [], "", J_tc_2early_SLOT2> {
1477 let IClass = 0b0101;
1478 let Inst{27-21} = 0b0010100;
1479 let Inst{20-16} = dst;
1482 let isBranch = 1, isIndirectBranch = 1, Defs = [PC], isPredicated = 1,
1483 hasSideEffects = 0, InputType = "reg" in
1484 class T_JMPr_c <bit PredNot, bit isPredNew, bit isTak>
1485 : JRInst <(outs), (ins PredRegs:$src, IntRegs:$dst),
1486 CondStr<"$src", !if(PredNot,0,1), isPredNew>.S #
1487 JumpOpcStr<"jumpr", isPredNew, isTak>.S # " $dst", [],
1488 "", J_tc_2early_SLOT2> {
1490 let isTaken = isTak;
1491 let isPredicatedFalse = PredNot;
1492 let isPredicatedNew = isPredNew;
1496 let IClass = 0b0101;
1498 let Inst{27-22} = 0b001101;
1499 let Inst{21} = PredNot;
1500 let Inst{20-16} = dst;
1501 let Inst{12} = isTak;
1502 let Inst{11} = isPredNew;
1503 let Inst{9-8} = src;
1506 multiclass JMPR_Pred<bit PredNot> {
1507 def NAME : T_JMPr_c<PredNot, 0, 0>; // not taken
1509 def NAME#newpt : T_JMPr_c<PredNot, 1, 1>; // taken
1510 def NAME#new : T_JMPr_c<PredNot, 1, 0>; // not taken
1513 multiclass JMPR_base<string BaseOp> {
1514 let BaseOpcode = BaseOp in {
1516 defm t : JMPR_Pred<0>;
1517 defm f : JMPR_Pred<1>;
1521 let isCall = 1, hasSideEffects = 1 in
1522 class JUMPR_MISC_CALLR<bit isPred, bit isPredNot,
1523 dag InputDag = (ins IntRegs:$Rs)>
1524 : JRInst<(outs), InputDag,
1525 !if(isPred, !if(isPredNot, "if (!$Pu) callr $Rs",
1526 "if ($Pu) callr $Rs"),
1528 [], "", J_tc_2early_SLOT2> {
1531 let isPredicated = isPred;
1532 let isPredicatedFalse = isPredNot;
1534 let IClass = 0b0101;
1535 let Inst{27-25} = 0b000;
1536 let Inst{24-23} = !if (isPred, 0b10, 0b01);
1538 let Inst{21} = isPredNot;
1539 let Inst{9-8} = !if (isPred, Pu, 0b00);
1540 let Inst{20-16} = Rs;
1544 let Defs = VolatileV3.Regs in {
1545 def J2_callrt : JUMPR_MISC_CALLR<1, 0, (ins PredRegs:$Pu, IntRegs:$Rs)>;
1546 def J2_callrf : JUMPR_MISC_CALLR<1, 1, (ins PredRegs:$Pu, IntRegs:$Rs)>;
1549 let isTerminator = 1, hasSideEffects = 0 in {
1550 defm J2_jump : JMP_base<"JMP", "">, PredNewRel;
1552 // Deal with explicit assembly
1553 // - never extened a jump #, always extend a jump ##
1554 let isAsmParserOnly = 1 in {
1555 defm J2_jump_ext : JMP_base<"JMP", "##">;
1556 defm J2_jump_noext : JMP_base<"JMP", "#">;
1559 defm J2_jumpr : JMPR_base<"JMPr">, PredNewRel;
1561 let isReturn = 1, isCodeGenOnly = 1 in
1562 defm JMPret : JMPR_base<"JMPret">, PredNewRel;
1565 def: Pat<(br bb:$dst),
1566 (J2_jump brtarget:$dst)>;
1568 (JMPret (i32 R31))>;
1569 def: Pat<(brcond (i1 PredRegs:$src1), bb:$offset),
1570 (J2_jumpt PredRegs:$src1, bb:$offset)>;
1572 // A return through builtin_eh_return.
1573 let isReturn = 1, isTerminator = 1, isBarrier = 1, hasSideEffects = 0,
1574 isCodeGenOnly = 1, Defs = [PC], Uses = [R28], isPredicable = 0 in
1575 def EH_RETURN_JMPR : T_JMPr;
1577 def: Pat<(eh_return),
1578 (EH_RETURN_JMPR (i32 R31))>;
1579 def: Pat<(HexagonBR_JT (i32 IntRegs:$dst)),
1580 (J2_jumpr IntRegs:$dst)>;
1581 def: Pat<(brind (i32 IntRegs:$dst)),
1582 (J2_jumpr IntRegs:$dst)>;
1584 //===----------------------------------------------------------------------===//
1586 //===----------------------------------------------------------------------===//
1588 //===----------------------------------------------------------------------===//
1590 //===----------------------------------------------------------------------===//
1592 // Load - Base with Immediate offset addressing mode
1593 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, AddedComplexity = 20 in
1594 class T_load_io <string mnemonic, RegisterClass RC, bits<4> MajOp,
1596 : LDInst<(outs RC:$dst), (ins IntRegs:$src1, ImmOp:$offset),
1597 "$dst = "#mnemonic#"($src1 + #$offset)", []>, AddrModeRel {
1602 bits<11> offsetBits;
1604 string ImmOpStr = !cast<string>(ImmOp);
1605 let offsetBits = !if (!eq(ImmOpStr, "s11_3Ext"), offset{13-3},
1606 !if (!eq(ImmOpStr, "s11_2Ext"), offset{12-2},
1607 !if (!eq(ImmOpStr, "s11_1Ext"), offset{11-1},
1608 /* s11_0Ext */ offset{10-0})));
1609 let opExtentBits = !if (!eq(ImmOpStr, "s11_3Ext"), 14,
1610 !if (!eq(ImmOpStr, "s11_2Ext"), 13,
1611 !if (!eq(ImmOpStr, "s11_1Ext"), 12,
1612 /* s11_0Ext */ 11)));
1613 let hasNewValue = !if (!eq(!cast<string>(RC), "DoubleRegs"), 0, 1);
1615 let IClass = 0b1001;
1618 let Inst{26-25} = offsetBits{10-9};
1619 let Inst{24-21} = MajOp;
1620 let Inst{20-16} = src1;
1621 let Inst{13-5} = offsetBits{8-0};
1622 let Inst{4-0} = dst;
1625 let opExtendable = 3, isExtentSigned = 0, isPredicated = 1 in
1626 class T_pload_io <string mnemonic, RegisterClass RC, bits<4>MajOp,
1627 Operand ImmOp, bit isNot, bit isPredNew>
1628 : LDInst<(outs RC:$dst),
1629 (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset),
1630 "if ("#!if(isNot, "!$src1", "$src1")
1631 #!if(isPredNew, ".new", "")
1632 #") $dst = "#mnemonic#"($src2 + #$offset)",
1633 [],"", V2LDST_tc_ld_SLOT01> , AddrModeRel {
1639 string ImmOpStr = !cast<string>(ImmOp);
1641 let offsetBits = !if (!eq(ImmOpStr, "u6_3Ext"), offset{8-3},
1642 !if (!eq(ImmOpStr, "u6_2Ext"), offset{7-2},
1643 !if (!eq(ImmOpStr, "u6_1Ext"), offset{6-1},
1644 /* u6_0Ext */ offset{5-0})));
1645 let opExtentBits = !if (!eq(ImmOpStr, "u6_3Ext"), 9,
1646 !if (!eq(ImmOpStr, "u6_2Ext"), 8,
1647 !if (!eq(ImmOpStr, "u6_1Ext"), 7,
1649 let hasNewValue = !if (!eq(ImmOpStr, "u6_3Ext"), 0, 1);
1650 let isPredicatedNew = isPredNew;
1651 let isPredicatedFalse = isNot;
1653 let IClass = 0b0100;
1657 let Inst{26} = isNot;
1658 let Inst{25} = isPredNew;
1659 let Inst{24-21} = MajOp;
1660 let Inst{20-16} = src2;
1662 let Inst{12-11} = src1;
1663 let Inst{10-5} = offsetBits;
1664 let Inst{4-0} = dst;
1667 let isExtendable = 1, hasSideEffects = 0, addrMode = BaseImmOffset in
1668 multiclass LD_Idxd<string mnemonic, string CextOp, RegisterClass RC,
1669 Operand ImmOp, Operand predImmOp, bits<4>MajOp> {
1670 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
1671 let isPredicable = 1 in
1672 def L2_#NAME#_io : T_load_io <mnemonic, RC, MajOp, ImmOp>;
1675 def L2_p#NAME#t_io : T_pload_io <mnemonic, RC, MajOp, predImmOp, 0, 0>;
1676 def L2_p#NAME#f_io : T_pload_io <mnemonic, RC, MajOp, predImmOp, 1, 0>;
1679 def L2_p#NAME#tnew_io : T_pload_io <mnemonic, RC, MajOp, predImmOp, 0, 1>;
1680 def L2_p#NAME#fnew_io : T_pload_io <mnemonic, RC, MajOp, predImmOp, 1, 1>;
1684 let accessSize = ByteAccess in {
1685 defm loadrb: LD_Idxd <"memb", "LDrib", IntRegs, s11_0Ext, u6_0Ext, 0b1000>;
1686 defm loadrub: LD_Idxd <"memub", "LDriub", IntRegs, s11_0Ext, u6_0Ext, 0b1001>;
1689 let accessSize = HalfWordAccess, opExtentAlign = 1 in {
1690 defm loadrh: LD_Idxd <"memh", "LDrih", IntRegs, s11_1Ext, u6_1Ext, 0b1010>;
1691 defm loadruh: LD_Idxd <"memuh", "LDriuh", IntRegs, s11_1Ext, u6_1Ext, 0b1011>;
1694 let accessSize = WordAccess, opExtentAlign = 2 in
1695 defm loadri: LD_Idxd <"memw", "LDriw", IntRegs, s11_2Ext, u6_2Ext, 0b1100>;
1697 let accessSize = DoubleWordAccess, opExtentAlign = 3 in
1698 defm loadrd: LD_Idxd <"memd", "LDrid", DoubleRegs, s11_3Ext, u6_3Ext, 0b1110>;
1700 let accessSize = HalfWordAccess, opExtentAlign = 1 in {
1701 def L2_loadbsw2_io: T_load_io<"membh", IntRegs, 0b0001, s11_1Ext>;
1702 def L2_loadbzw2_io: T_load_io<"memubh", IntRegs, 0b0011, s11_1Ext>;
1705 let accessSize = WordAccess, opExtentAlign = 2 in {
1706 def L2_loadbzw4_io: T_load_io<"memubh", DoubleRegs, 0b0101, s11_2Ext>;
1707 def L2_loadbsw4_io: T_load_io<"membh", DoubleRegs, 0b0111, s11_2Ext>;
1710 // Patterns to select load-indexed (i.e. load from base+offset).
1711 multiclass Loadx_pat<PatFrag Load, ValueType VT, PatLeaf ImmPred,
1713 def: Pat<(VT (Load AddrFI:$fi)), (VT (MI AddrFI:$fi, 0))>;
1714 def: Pat<(VT (Load (add (i32 IntRegs:$Rs), ImmPred:$Off))),
1715 (VT (MI IntRegs:$Rs, imm:$Off))>;
1716 def: Pat<(VT (Load (i32 IntRegs:$Rs))), (VT (MI IntRegs:$Rs, 0))>;
1719 let AddedComplexity = 20 in {
1720 defm: Loadx_pat<load, i32, s11_2ExtPred, L2_loadri_io>;
1721 defm: Loadx_pat<load, i64, s11_3ExtPred, L2_loadrd_io>;
1722 defm: Loadx_pat<atomic_load_8 , i32, s11_0ExtPred, L2_loadrub_io>;
1723 defm: Loadx_pat<atomic_load_16, i32, s11_1ExtPred, L2_loadruh_io>;
1724 defm: Loadx_pat<atomic_load_32, i32, s11_2ExtPred, L2_loadri_io>;
1725 defm: Loadx_pat<atomic_load_64, i64, s11_3ExtPred, L2_loadrd_io>;
1727 defm: Loadx_pat<extloadi1, i32, s11_0ExtPred, L2_loadrub_io>;
1728 defm: Loadx_pat<extloadi8, i32, s11_0ExtPred, L2_loadrub_io>;
1729 defm: Loadx_pat<extloadi16, i32, s11_1ExtPred, L2_loadruh_io>;
1730 defm: Loadx_pat<sextloadi8, i32, s11_0ExtPred, L2_loadrb_io>;
1731 defm: Loadx_pat<sextloadi16, i32, s11_1ExtPred, L2_loadrh_io>;
1732 defm: Loadx_pat<zextloadi1, i32, s11_0ExtPred, L2_loadrub_io>;
1733 defm: Loadx_pat<zextloadi8, i32, s11_0ExtPred, L2_loadrub_io>;
1734 defm: Loadx_pat<zextloadi16, i32, s11_1ExtPred, L2_loadruh_io>;
1738 // Sign-extending loads of i1 need to replicate the lowest bit throughout
1739 // the 32-bit value. Since the loaded value can only be 0 or 1, 0-v should
1741 let AddedComplexity = 20 in
1742 def: Pat<(i32 (sextloadi1 (i32 IntRegs:$Rs))),
1743 (A2_subri 0, (L2_loadrub_io IntRegs:$Rs, 0))>;
1745 //===----------------------------------------------------------------------===//
1746 // Post increment load
1747 //===----------------------------------------------------------------------===//
1748 //===----------------------------------------------------------------------===//
1749 // Template class for non-predicated post increment loads with immediate offset.
1750 //===----------------------------------------------------------------------===//
1751 let hasSideEffects = 0, addrMode = PostInc in
1752 class T_load_pi <string mnemonic, RegisterClass RC, Operand ImmOp,
1754 : LDInstPI <(outs RC:$dst, IntRegs:$dst2),
1755 (ins IntRegs:$src1, ImmOp:$offset),
1756 "$dst = "#mnemonic#"($src1++#$offset)" ,
1765 string ImmOpStr = !cast<string>(ImmOp);
1766 let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
1767 !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
1768 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
1769 /* s4_0Imm */ offset{3-0})));
1770 let hasNewValue = !if (!eq(ImmOpStr, "s4_3Imm"), 0, 1);
1772 let IClass = 0b1001;
1774 let Inst{27-25} = 0b101;
1775 let Inst{24-21} = MajOp;
1776 let Inst{20-16} = src1;
1777 let Inst{13-12} = 0b00;
1778 let Inst{8-5} = offsetBits;
1779 let Inst{4-0} = dst;
1782 //===----------------------------------------------------------------------===//
1783 // Template class for predicated post increment loads with immediate offset.
1784 //===----------------------------------------------------------------------===//
1785 let isPredicated = 1, hasSideEffects = 0, addrMode = PostInc in
1786 class T_pload_pi <string mnemonic, RegisterClass RC, Operand ImmOp,
1787 bits<4> MajOp, bit isPredNot, bit isPredNew >
1788 : LDInst <(outs RC:$dst, IntRegs:$dst2),
1789 (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset),
1790 !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
1791 ") ")#"$dst = "#mnemonic#"($src2++#$offset)",
1801 let isPredicatedNew = isPredNew;
1802 let isPredicatedFalse = isPredNot;
1804 string ImmOpStr = !cast<string>(ImmOp);
1805 let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
1806 !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
1807 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
1808 /* s4_0Imm */ offset{3-0})));
1809 let hasNewValue = !if (!eq(ImmOpStr, "s4_3Imm"), 0, 1);
1811 let IClass = 0b1001;
1813 let Inst{27-25} = 0b101;
1814 let Inst{24-21} = MajOp;
1815 let Inst{20-16} = src2;
1817 let Inst{12} = isPredNew;
1818 let Inst{11} = isPredNot;
1819 let Inst{10-9} = src1;
1820 let Inst{8-5} = offsetBits;
1821 let Inst{4-0} = dst;
1824 //===----------------------------------------------------------------------===//
1825 // Multiclass for post increment loads with immediate offset.
1826 //===----------------------------------------------------------------------===//
1828 multiclass LD_PostInc <string mnemonic, string BaseOp, RegisterClass RC,
1829 Operand ImmOp, bits<4> MajOp> {
1830 let BaseOpcode = "POST_"#BaseOp in {
1831 let isPredicable = 1 in
1832 def L2_#NAME#_pi : T_load_pi < mnemonic, RC, ImmOp, MajOp>;
1835 def L2_p#NAME#t_pi : T_pload_pi < mnemonic, RC, ImmOp, MajOp, 0, 0>;
1836 def L2_p#NAME#f_pi : T_pload_pi < mnemonic, RC, ImmOp, MajOp, 1, 0>;
1839 def L2_p#NAME#tnew_pi : T_pload_pi < mnemonic, RC, ImmOp, MajOp, 0, 1>;
1840 def L2_p#NAME#fnew_pi : T_pload_pi < mnemonic, RC, ImmOp, MajOp, 1, 1>;
1844 // post increment byte loads with immediate offset
1845 let accessSize = ByteAccess in {
1846 defm loadrb : LD_PostInc <"memb", "LDrib", IntRegs, s4_0Imm, 0b1000>;
1847 defm loadrub : LD_PostInc <"memub", "LDriub", IntRegs, s4_0Imm, 0b1001>;
1850 // post increment halfword loads with immediate offset
1851 let accessSize = HalfWordAccess, opExtentAlign = 1 in {
1852 defm loadrh : LD_PostInc <"memh", "LDrih", IntRegs, s4_1Imm, 0b1010>;
1853 defm loadruh : LD_PostInc <"memuh", "LDriuh", IntRegs, s4_1Imm, 0b1011>;
1856 // post increment word loads with immediate offset
1857 let accessSize = WordAccess, opExtentAlign = 2 in
1858 defm loadri : LD_PostInc <"memw", "LDriw", IntRegs, s4_2Imm, 0b1100>;
1860 // post increment doubleword loads with immediate offset
1861 let accessSize = DoubleWordAccess, opExtentAlign = 3 in
1862 defm loadrd : LD_PostInc <"memd", "LDrid", DoubleRegs, s4_3Imm, 0b1110>;
1864 // Rd=memb[u]h(Rx++#s4:1)
1865 // Rdd=memb[u]h(Rx++#s4:2)
1866 let accessSize = HalfWordAccess, opExtentAlign = 1 in {
1867 def L2_loadbsw2_pi : T_load_pi <"membh", IntRegs, s4_1Imm, 0b0001>;
1868 def L2_loadbzw2_pi : T_load_pi <"memubh", IntRegs, s4_1Imm, 0b0011>;
1870 let accessSize = WordAccess, opExtentAlign = 2, hasNewValue = 0 in {
1871 def L2_loadbsw4_pi : T_load_pi <"membh", DoubleRegs, s4_2Imm, 0b0111>;
1872 def L2_loadbzw4_pi : T_load_pi <"memubh", DoubleRegs, s4_2Imm, 0b0101>;
1875 //===----------------------------------------------------------------------===//
1876 // Template class for post increment loads with register offset.
1877 //===----------------------------------------------------------------------===//
1878 let hasSideEffects = 0, addrMode = PostInc in
1879 class T_load_pr <string mnemonic, RegisterClass RC, bits<4> MajOp,
1880 MemAccessSize AccessSz>
1881 : LDInstPI <(outs RC:$dst, IntRegs:$_dst_),
1882 (ins IntRegs:$src1, ModRegs:$src2),
1883 "$dst = "#mnemonic#"($src1++$src2)" ,
1884 [], "$src1 = $_dst_" > {
1889 let accessSize = AccessSz;
1890 let IClass = 0b1001;
1892 let Inst{27-25} = 0b110;
1893 let Inst{24-21} = MajOp;
1894 let Inst{20-16} = src1;
1895 let Inst{13} = src2;
1898 let Inst{4-0} = dst;
1901 let hasNewValue = 1 in {
1902 def L2_loadrb_pr : T_load_pr <"memb", IntRegs, 0b1000, ByteAccess>;
1903 def L2_loadrub_pr : T_load_pr <"memub", IntRegs, 0b1001, ByteAccess>;
1904 def L2_loadrh_pr : T_load_pr <"memh", IntRegs, 0b1010, HalfWordAccess>;
1905 def L2_loadruh_pr : T_load_pr <"memuh", IntRegs, 0b1011, HalfWordAccess>;
1906 def L2_loadri_pr : T_load_pr <"memw", IntRegs, 0b1100, WordAccess>;
1908 def L2_loadbzw2_pr : T_load_pr <"memubh", IntRegs, 0b0011, HalfWordAccess>;
1911 def L2_loadrd_pr : T_load_pr <"memd", DoubleRegs, 0b1110, DoubleWordAccess>;
1912 def L2_loadbzw4_pr : T_load_pr <"memubh", DoubleRegs, 0b0101, WordAccess>;
1915 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 13,
1916 isCodeGenOnly = 1, isPseudo = 1, hasSideEffects = 0 in
1917 def LDriw_pred : LDInst<(outs PredRegs:$dst),
1918 (ins IntRegs:$addr, s11_2Ext:$off),
1919 ".error \"should not emit\"", []>;
1921 let Defs = [R29, R30, R31], Uses = [R30], hasSideEffects = 0 in
1922 def L2_deallocframe : LDInst<(outs), (ins),
1925 let IClass = 0b1001;
1927 let Inst{27-16} = 0b000000011110;
1929 let Inst{4-0} = 0b11110;
1932 // Load / Post increment circular addressing mode.
1933 let Uses = [CS], hasSideEffects = 0 in
1934 class T_load_pcr<string mnemonic, RegisterClass RC, bits<4> MajOp>
1935 : LDInst <(outs RC:$dst, IntRegs:$_dst_),
1936 (ins IntRegs:$Rz, ModRegs:$Mu),
1937 "$dst = "#mnemonic#"($Rz ++ I:circ($Mu))", [],
1943 let hasNewValue = !if (!eq(!cast<string>(RC), "DoubleRegs"), 0, 1);
1944 let IClass = 0b1001;
1946 let Inst{27-25} = 0b100;
1947 let Inst{24-21} = MajOp;
1948 let Inst{20-16} = Rz;
1953 let Inst{4-0} = dst;
1956 let accessSize = ByteAccess in {
1957 def L2_loadrb_pcr : T_load_pcr <"memb", IntRegs, 0b1000>;
1958 def L2_loadrub_pcr : T_load_pcr <"memub", IntRegs, 0b1001>;
1961 let accessSize = HalfWordAccess in {
1962 def L2_loadrh_pcr : T_load_pcr <"memh", IntRegs, 0b1010>;
1963 def L2_loadruh_pcr : T_load_pcr <"memuh", IntRegs, 0b1011>;
1964 def L2_loadbsw2_pcr : T_load_pcr <"membh", IntRegs, 0b0001>;
1965 def L2_loadbzw2_pcr : T_load_pcr <"memubh", IntRegs, 0b0011>;
1968 let accessSize = WordAccess in {
1969 def L2_loadri_pcr : T_load_pcr <"memw", IntRegs, 0b1100>;
1970 let hasNewValue = 0 in {
1971 def L2_loadbzw4_pcr : T_load_pcr <"memubh", DoubleRegs, 0b0101>;
1972 def L2_loadbsw4_pcr : T_load_pcr <"membh", DoubleRegs, 0b0111>;
1976 let accessSize = DoubleWordAccess in
1977 def L2_loadrd_pcr : T_load_pcr <"memd", DoubleRegs, 0b1110>;
1979 //===----------------------------------------------------------------------===//
1980 // Circular loads with immediate offset.
1981 //===----------------------------------------------------------------------===//
1982 let Uses = [CS], mayLoad = 1, hasSideEffects = 0 in
1983 class T_load_pci <string mnemonic, RegisterClass RC,
1984 Operand ImmOp, bits<4> MajOp>
1985 : LDInstPI<(outs RC:$dst, IntRegs:$_dst_),
1986 (ins IntRegs:$Rz, ImmOp:$offset, ModRegs:$Mu),
1987 "$dst = "#mnemonic#"($Rz ++ #$offset:circ($Mu))", [],
1995 string ImmOpStr = !cast<string>(ImmOp);
1996 let hasNewValue = !if (!eq(!cast<string>(RC), "DoubleRegs"), 0, 1);
1997 let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
1998 !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
1999 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
2000 /* s4_0Imm */ offset{3-0})));
2001 let IClass = 0b1001;
2002 let Inst{27-25} = 0b100;
2003 let Inst{24-21} = MajOp;
2004 let Inst{20-16} = Rz;
2008 let Inst{8-5} = offsetBits;
2009 let Inst{4-0} = dst;
2012 // Byte variants of circ load
2013 let accessSize = ByteAccess in {
2014 def L2_loadrb_pci : T_load_pci <"memb", IntRegs, s4_0Imm, 0b1000>;
2015 def L2_loadrub_pci : T_load_pci <"memub", IntRegs, s4_0Imm, 0b1001>;
2018 // Half word variants of circ load
2019 let accessSize = HalfWordAccess in {
2020 def L2_loadrh_pci : T_load_pci <"memh", IntRegs, s4_1Imm, 0b1010>;
2021 def L2_loadruh_pci : T_load_pci <"memuh", IntRegs, s4_1Imm, 0b1011>;
2022 def L2_loadbzw2_pci : T_load_pci <"memubh", IntRegs, s4_1Imm, 0b0011>;
2023 def L2_loadbsw2_pci : T_load_pci <"membh", IntRegs, s4_1Imm, 0b0001>;
2026 // Word variants of circ load
2027 let accessSize = WordAccess in
2028 def L2_loadri_pci : T_load_pci <"memw", IntRegs, s4_2Imm, 0b1100>;
2030 let accessSize = WordAccess, hasNewValue = 0 in {
2031 def L2_loadbzw4_pci : T_load_pci <"memubh", DoubleRegs, s4_2Imm, 0b0101>;
2032 def L2_loadbsw4_pci : T_load_pci <"membh", DoubleRegs, s4_2Imm, 0b0111>;
2035 let accessSize = DoubleWordAccess, hasNewValue = 0 in
2036 def L2_loadrd_pci : T_load_pci <"memd", DoubleRegs, s4_3Imm, 0b1110>;
2038 // L[24]_load[wd]_locked: Load word/double with lock.
2040 class T_load_locked <string mnemonic, RegisterClass RC>
2041 : LD0Inst <(outs RC:$dst),
2043 "$dst = "#mnemonic#"($src)"> {
2046 let IClass = 0b1001;
2047 let Inst{27-21} = 0b0010000;
2048 let Inst{20-16} = src;
2049 let Inst{13-12} = !if (!eq(mnemonic, "memd_locked"), 0b01, 0b00);
2051 let Inst{4-0} = dst;
2053 let hasNewValue = 1, accessSize = WordAccess, opNewValue = 0 in
2054 def L2_loadw_locked : T_load_locked <"memw_locked", IntRegs>;
2055 let accessSize = DoubleWordAccess in
2056 def L4_loadd_locked : T_load_locked <"memd_locked", DoubleRegs>;
2058 // S[24]_store[wd]_locked: Store word/double conditionally.
2059 let isSoloAX = 1, isPredicateLate = 1 in
2060 class T_store_locked <string mnemonic, RegisterClass RC>
2061 : ST0Inst <(outs PredRegs:$Pd), (ins IntRegs:$Rs, RC:$Rt),
2062 mnemonic#"($Rs, $Pd) = $Rt"> {
2067 let IClass = 0b1010;
2068 let Inst{27-23} = 0b00001;
2069 let Inst{22} = !if (!eq(mnemonic, "memw_locked"), 0b0, 0b1);
2071 let Inst{20-16} = Rs;
2072 let Inst{12-8} = Rt;
2076 let accessSize = WordAccess in
2077 def S2_storew_locked : T_store_locked <"memw_locked", IntRegs>;
2079 let accessSize = DoubleWordAccess in
2080 def S4_stored_locked : T_store_locked <"memd_locked", DoubleRegs>;
2082 //===----------------------------------------------------------------------===//
2083 // Bit-reversed loads with auto-increment register
2084 //===----------------------------------------------------------------------===//
2085 let hasSideEffects = 0 in
2086 class T_load_pbr<string mnemonic, RegisterClass RC,
2087 MemAccessSize addrSize, bits<4> majOp>
2089 <(outs RC:$dst, IntRegs:$_dst_),
2090 (ins IntRegs:$Rz, ModRegs:$Mu),
2091 "$dst = "#mnemonic#"($Rz ++ $Mu:brev)" ,
2092 [] , "$Rz = $_dst_" > {
2094 let accessSize = addrSize;
2100 let IClass = 0b1001;
2102 let Inst{27-25} = 0b111;
2103 let Inst{24-21} = majOp;
2104 let Inst{20-16} = Rz;
2108 let Inst{4-0} = dst;
2111 let hasNewValue =1, opNewValue = 0 in {
2112 def L2_loadrb_pbr : T_load_pbr <"memb", IntRegs, ByteAccess, 0b1000>;
2113 def L2_loadrub_pbr : T_load_pbr <"memub", IntRegs, ByteAccess, 0b1001>;
2114 def L2_loadrh_pbr : T_load_pbr <"memh", IntRegs, HalfWordAccess, 0b1010>;
2115 def L2_loadruh_pbr : T_load_pbr <"memuh", IntRegs, HalfWordAccess, 0b1011>;
2116 def L2_loadbsw2_pbr : T_load_pbr <"membh", IntRegs, HalfWordAccess, 0b0001>;
2117 def L2_loadbzw2_pbr : T_load_pbr <"memubh", IntRegs, HalfWordAccess, 0b0011>;
2118 def L2_loadri_pbr : T_load_pbr <"memw", IntRegs, WordAccess, 0b1100>;
2121 def L2_loadbzw4_pbr : T_load_pbr <"memubh", DoubleRegs, WordAccess, 0b0101>;
2122 def L2_loadbsw4_pbr : T_load_pbr <"membh", DoubleRegs, WordAccess, 0b0111>;
2123 def L2_loadrd_pbr : T_load_pbr <"memd", DoubleRegs, DoubleWordAccess, 0b1110>;
2125 //===----------------------------------------------------------------------===//
2127 //===----------------------------------------------------------------------===//
2129 //===----------------------------------------------------------------------===//
2131 //===----------------------------------------------------------------------===//
2132 //===----------------------------------------------------------------------===//
2134 //===----------------------------------------------------------------------===//
2136 //===----------------------------------------------------------------------===//
2138 //===----------------------------------------------------------------------===//
2139 //===----------------------------------------------------------------------===//
2141 //===----------------------------------------------------------------------===//
2143 //===----------------------------------------------------------------------===//
2145 //===----------------------------------------------------------------------===//
2147 //===----------------------------------------------------------------------===//
2149 // MPYS / Multipy signed/unsigned halfwords
2150 //Rd=mpy[u](Rs.[H|L],Rt.[H|L])[:<<1][:rnd][:sat]
2151 //===----------------------------------------------------------------------===//
2153 let hasNewValue = 1, opNewValue = 0 in
2154 class T_M2_mpy < bits<2> LHbits, bit isSat, bit isRnd,
2155 bit hasShift, bit isUnsigned>
2156 : MInst < (outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
2157 "$Rd = "#!if(isUnsigned,"mpyu","mpy")#"($Rs."#!if(LHbits{1},"h","l")
2158 #", $Rt."#!if(LHbits{0},"h)","l)")
2159 #!if(hasShift,":<<1","")
2160 #!if(isRnd,":rnd","")
2161 #!if(isSat,":sat",""),
2162 [], "", M_tc_3x_SLOT23 > {
2167 let IClass = 0b1110;
2169 let Inst{27-24} = 0b1100;
2170 let Inst{23} = hasShift;
2171 let Inst{22} = isUnsigned;
2172 let Inst{21} = isRnd;
2173 let Inst{7} = isSat;
2174 let Inst{6-5} = LHbits;
2176 let Inst{20-16} = Rs;
2177 let Inst{12-8} = Rt;
2180 //Rd=mpy(Rs.[H|L],Rt.[H|L])[:<<1]
2181 def M2_mpy_ll_s1: T_M2_mpy<0b00, 0, 0, 1, 0>;
2182 def M2_mpy_ll_s0: T_M2_mpy<0b00, 0, 0, 0, 0>;
2183 def M2_mpy_lh_s1: T_M2_mpy<0b01, 0, 0, 1, 0>;
2184 def M2_mpy_lh_s0: T_M2_mpy<0b01, 0, 0, 0, 0>;
2185 def M2_mpy_hl_s1: T_M2_mpy<0b10, 0, 0, 1, 0>;
2186 def M2_mpy_hl_s0: T_M2_mpy<0b10, 0, 0, 0, 0>;
2187 def M2_mpy_hh_s1: T_M2_mpy<0b11, 0, 0, 1, 0>;
2188 def M2_mpy_hh_s0: T_M2_mpy<0b11, 0, 0, 0, 0>;
2190 //Rd=mpyu(Rs.[H|L],Rt.[H|L])[:<<1]
2191 def M2_mpyu_ll_s1: T_M2_mpy<0b00, 0, 0, 1, 1>;
2192 def M2_mpyu_ll_s0: T_M2_mpy<0b00, 0, 0, 0, 1>;
2193 def M2_mpyu_lh_s1: T_M2_mpy<0b01, 0, 0, 1, 1>;
2194 def M2_mpyu_lh_s0: T_M2_mpy<0b01, 0, 0, 0, 1>;
2195 def M2_mpyu_hl_s1: T_M2_mpy<0b10, 0, 0, 1, 1>;
2196 def M2_mpyu_hl_s0: T_M2_mpy<0b10, 0, 0, 0, 1>;
2197 def M2_mpyu_hh_s1: T_M2_mpy<0b11, 0, 0, 1, 1>;
2198 def M2_mpyu_hh_s0: T_M2_mpy<0b11, 0, 0, 0, 1>;
2200 //Rd=mpy(Rs.[H|L],Rt.[H|L])[:<<1]:rnd
2201 def M2_mpy_rnd_ll_s1: T_M2_mpy <0b00, 0, 1, 1, 0>;
2202 def M2_mpy_rnd_ll_s0: T_M2_mpy <0b00, 0, 1, 0, 0>;
2203 def M2_mpy_rnd_lh_s1: T_M2_mpy <0b01, 0, 1, 1, 0>;
2204 def M2_mpy_rnd_lh_s0: T_M2_mpy <0b01, 0, 1, 0, 0>;
2205 def M2_mpy_rnd_hl_s1: T_M2_mpy <0b10, 0, 1, 1, 0>;
2206 def M2_mpy_rnd_hl_s0: T_M2_mpy <0b10, 0, 1, 0, 0>;
2207 def M2_mpy_rnd_hh_s1: T_M2_mpy <0b11, 0, 1, 1, 0>;
2208 def M2_mpy_rnd_hh_s0: T_M2_mpy <0b11, 0, 1, 0, 0>;
2210 //Rd=mpy(Rs.[H|L],Rt.[H|L])[:<<1][:sat]
2211 //Rd=mpy(Rs.[H|L],Rt.[H|L])[:<<1][:rnd][:sat]
2212 let Defs = [USR_OVF] in {
2213 def M2_mpy_sat_ll_s1: T_M2_mpy <0b00, 1, 0, 1, 0>;
2214 def M2_mpy_sat_ll_s0: T_M2_mpy <0b00, 1, 0, 0, 0>;
2215 def M2_mpy_sat_lh_s1: T_M2_mpy <0b01, 1, 0, 1, 0>;
2216 def M2_mpy_sat_lh_s0: T_M2_mpy <0b01, 1, 0, 0, 0>;
2217 def M2_mpy_sat_hl_s1: T_M2_mpy <0b10, 1, 0, 1, 0>;
2218 def M2_mpy_sat_hl_s0: T_M2_mpy <0b10, 1, 0, 0, 0>;
2219 def M2_mpy_sat_hh_s1: T_M2_mpy <0b11, 1, 0, 1, 0>;
2220 def M2_mpy_sat_hh_s0: T_M2_mpy <0b11, 1, 0, 0, 0>;
2222 def M2_mpy_sat_rnd_ll_s1: T_M2_mpy <0b00, 1, 1, 1, 0>;
2223 def M2_mpy_sat_rnd_ll_s0: T_M2_mpy <0b00, 1, 1, 0, 0>;
2224 def M2_mpy_sat_rnd_lh_s1: T_M2_mpy <0b01, 1, 1, 1, 0>;
2225 def M2_mpy_sat_rnd_lh_s0: T_M2_mpy <0b01, 1, 1, 0, 0>;
2226 def M2_mpy_sat_rnd_hl_s1: T_M2_mpy <0b10, 1, 1, 1, 0>;
2227 def M2_mpy_sat_rnd_hl_s0: T_M2_mpy <0b10, 1, 1, 0, 0>;
2228 def M2_mpy_sat_rnd_hh_s1: T_M2_mpy <0b11, 1, 1, 1, 0>;
2229 def M2_mpy_sat_rnd_hh_s0: T_M2_mpy <0b11, 1, 1, 0, 0>;
2232 //===----------------------------------------------------------------------===//
2234 // MPYS / Multipy signed/unsigned halfwords and add/subtract the
2235 // result from the accumulator.
2236 //Rx [-+]= mpy[u](Rs.[H|L],Rt.[H|L])[:<<1][:sat]
2237 //===----------------------------------------------------------------------===//
2239 let hasNewValue = 1, opNewValue = 0 in
2240 class T_M2_mpy_acc < bits<2> LHbits, bit isSat, bit isNac,
2241 bit hasShift, bit isUnsigned >
2242 : MInst_acc<(outs IntRegs:$Rx), (ins IntRegs:$dst2, IntRegs:$Rs, IntRegs:$Rt),
2243 "$Rx "#!if(isNac,"-= ","+= ")#!if(isUnsigned,"mpyu","mpy")
2244 #"($Rs."#!if(LHbits{1},"h","l")
2245 #", $Rt."#!if(LHbits{0},"h)","l)")
2246 #!if(hasShift,":<<1","")
2247 #!if(isSat,":sat",""),
2248 [], "$dst2 = $Rx", M_tc_3x_SLOT23 > {
2253 let IClass = 0b1110;
2254 let Inst{27-24} = 0b1110;
2255 let Inst{23} = hasShift;
2256 let Inst{22} = isUnsigned;
2257 let Inst{21} = isNac;
2258 let Inst{7} = isSat;
2259 let Inst{6-5} = LHbits;
2261 let Inst{20-16} = Rs;
2262 let Inst{12-8} = Rt;
2265 //Rx += mpy(Rs.[H|L],Rt.[H|L])[:<<1]
2266 def M2_mpy_acc_ll_s1: T_M2_mpy_acc <0b00, 0, 0, 1, 0>;
2267 def M2_mpy_acc_ll_s0: T_M2_mpy_acc <0b00, 0, 0, 0, 0>;
2268 def M2_mpy_acc_lh_s1: T_M2_mpy_acc <0b01, 0, 0, 1, 0>;
2269 def M2_mpy_acc_lh_s0: T_M2_mpy_acc <0b01, 0, 0, 0, 0>;
2270 def M2_mpy_acc_hl_s1: T_M2_mpy_acc <0b10, 0, 0, 1, 0>;
2271 def M2_mpy_acc_hl_s0: T_M2_mpy_acc <0b10, 0, 0, 0, 0>;
2272 def M2_mpy_acc_hh_s1: T_M2_mpy_acc <0b11, 0, 0, 1, 0>;
2273 def M2_mpy_acc_hh_s0: T_M2_mpy_acc <0b11, 0, 0, 0, 0>;
2275 //Rx += mpyu(Rs.[H|L],Rt.[H|L])[:<<1]
2276 def M2_mpyu_acc_ll_s1: T_M2_mpy_acc <0b00, 0, 0, 1, 1>;
2277 def M2_mpyu_acc_ll_s0: T_M2_mpy_acc <0b00, 0, 0, 0, 1>;
2278 def M2_mpyu_acc_lh_s1: T_M2_mpy_acc <0b01, 0, 0, 1, 1>;
2279 def M2_mpyu_acc_lh_s0: T_M2_mpy_acc <0b01, 0, 0, 0, 1>;
2280 def M2_mpyu_acc_hl_s1: T_M2_mpy_acc <0b10, 0, 0, 1, 1>;
2281 def M2_mpyu_acc_hl_s0: T_M2_mpy_acc <0b10, 0, 0, 0, 1>;
2282 def M2_mpyu_acc_hh_s1: T_M2_mpy_acc <0b11, 0, 0, 1, 1>;
2283 def M2_mpyu_acc_hh_s0: T_M2_mpy_acc <0b11, 0, 0, 0, 1>;
2285 //Rx -= mpy(Rs.[H|L],Rt.[H|L])[:<<1]
2286 def M2_mpy_nac_ll_s1: T_M2_mpy_acc <0b00, 0, 1, 1, 0>;
2287 def M2_mpy_nac_ll_s0: T_M2_mpy_acc <0b00, 0, 1, 0, 0>;
2288 def M2_mpy_nac_lh_s1: T_M2_mpy_acc <0b01, 0, 1, 1, 0>;
2289 def M2_mpy_nac_lh_s0: T_M2_mpy_acc <0b01, 0, 1, 0, 0>;
2290 def M2_mpy_nac_hl_s1: T_M2_mpy_acc <0b10, 0, 1, 1, 0>;
2291 def M2_mpy_nac_hl_s0: T_M2_mpy_acc <0b10, 0, 1, 0, 0>;
2292 def M2_mpy_nac_hh_s1: T_M2_mpy_acc <0b11, 0, 1, 1, 0>;
2293 def M2_mpy_nac_hh_s0: T_M2_mpy_acc <0b11, 0, 1, 0, 0>;
2295 //Rx -= mpyu(Rs.[H|L],Rt.[H|L])[:<<1]
2296 def M2_mpyu_nac_ll_s1: T_M2_mpy_acc <0b00, 0, 1, 1, 1>;
2297 def M2_mpyu_nac_ll_s0: T_M2_mpy_acc <0b00, 0, 1, 0, 1>;
2298 def M2_mpyu_nac_lh_s1: T_M2_mpy_acc <0b01, 0, 1, 1, 1>;
2299 def M2_mpyu_nac_lh_s0: T_M2_mpy_acc <0b01, 0, 1, 0, 1>;
2300 def M2_mpyu_nac_hl_s1: T_M2_mpy_acc <0b10, 0, 1, 1, 1>;
2301 def M2_mpyu_nac_hl_s0: T_M2_mpy_acc <0b10, 0, 1, 0, 1>;
2302 def M2_mpyu_nac_hh_s1: T_M2_mpy_acc <0b11, 0, 1, 1, 1>;
2303 def M2_mpyu_nac_hh_s0: T_M2_mpy_acc <0b11, 0, 1, 0, 1>;
2305 //Rx += mpy(Rs.[H|L],Rt.[H|L])[:<<1]:sat
2306 def M2_mpy_acc_sat_ll_s1: T_M2_mpy_acc <0b00, 1, 0, 1, 0>;
2307 def M2_mpy_acc_sat_ll_s0: T_M2_mpy_acc <0b00, 1, 0, 0, 0>;
2308 def M2_mpy_acc_sat_lh_s1: T_M2_mpy_acc <0b01, 1, 0, 1, 0>;
2309 def M2_mpy_acc_sat_lh_s0: T_M2_mpy_acc <0b01, 1, 0, 0, 0>;
2310 def M2_mpy_acc_sat_hl_s1: T_M2_mpy_acc <0b10, 1, 0, 1, 0>;
2311 def M2_mpy_acc_sat_hl_s0: T_M2_mpy_acc <0b10, 1, 0, 0, 0>;
2312 def M2_mpy_acc_sat_hh_s1: T_M2_mpy_acc <0b11, 1, 0, 1, 0>;
2313 def M2_mpy_acc_sat_hh_s0: T_M2_mpy_acc <0b11, 1, 0, 0, 0>;
2315 //Rx -= mpy(Rs.[H|L],Rt.[H|L])[:<<1]:sat
2316 def M2_mpy_nac_sat_ll_s1: T_M2_mpy_acc <0b00, 1, 1, 1, 0>;
2317 def M2_mpy_nac_sat_ll_s0: T_M2_mpy_acc <0b00, 1, 1, 0, 0>;
2318 def M2_mpy_nac_sat_lh_s1: T_M2_mpy_acc <0b01, 1, 1, 1, 0>;
2319 def M2_mpy_nac_sat_lh_s0: T_M2_mpy_acc <0b01, 1, 1, 0, 0>;
2320 def M2_mpy_nac_sat_hl_s1: T_M2_mpy_acc <0b10, 1, 1, 1, 0>;
2321 def M2_mpy_nac_sat_hl_s0: T_M2_mpy_acc <0b10, 1, 1, 0, 0>;
2322 def M2_mpy_nac_sat_hh_s1: T_M2_mpy_acc <0b11, 1, 1, 1, 0>;
2323 def M2_mpy_nac_sat_hh_s0: T_M2_mpy_acc <0b11, 1, 1, 0, 0>;
2325 //===----------------------------------------------------------------------===//
2327 // MPYS / Multipy signed/unsigned halfwords and add/subtract the
2328 // result from the 64-bit destination register.
2329 //Rxx [-+]= mpy[u](Rs.[H|L],Rt.[H|L])[:<<1][:sat]
2330 //===----------------------------------------------------------------------===//
2332 class T_M2_mpyd_acc < bits<2> LHbits, bit isNac, bit hasShift, bit isUnsigned>
2333 : MInst_acc<(outs DoubleRegs:$Rxx),
2334 (ins DoubleRegs:$dst2, IntRegs:$Rs, IntRegs:$Rt),
2335 "$Rxx "#!if(isNac,"-= ","+= ")#!if(isUnsigned,"mpyu","mpy")
2336 #"($Rs."#!if(LHbits{1},"h","l")
2337 #", $Rt."#!if(LHbits{0},"h)","l)")
2338 #!if(hasShift,":<<1",""),
2339 [], "$dst2 = $Rxx", M_tc_3x_SLOT23 > {
2344 let IClass = 0b1110;
2346 let Inst{27-24} = 0b0110;
2347 let Inst{23} = hasShift;
2348 let Inst{22} = isUnsigned;
2349 let Inst{21} = isNac;
2351 let Inst{6-5} = LHbits;
2352 let Inst{4-0} = Rxx;
2353 let Inst{20-16} = Rs;
2354 let Inst{12-8} = Rt;
2357 def M2_mpyd_acc_hh_s0: T_M2_mpyd_acc <0b11, 0, 0, 0>;
2358 def M2_mpyd_acc_hl_s0: T_M2_mpyd_acc <0b10, 0, 0, 0>;
2359 def M2_mpyd_acc_lh_s0: T_M2_mpyd_acc <0b01, 0, 0, 0>;
2360 def M2_mpyd_acc_ll_s0: T_M2_mpyd_acc <0b00, 0, 0, 0>;
2362 def M2_mpyd_acc_hh_s1: T_M2_mpyd_acc <0b11, 0, 1, 0>;
2363 def M2_mpyd_acc_hl_s1: T_M2_mpyd_acc <0b10, 0, 1, 0>;
2364 def M2_mpyd_acc_lh_s1: T_M2_mpyd_acc <0b01, 0, 1, 0>;
2365 def M2_mpyd_acc_ll_s1: T_M2_mpyd_acc <0b00, 0, 1, 0>;
2367 def M2_mpyd_nac_hh_s0: T_M2_mpyd_acc <0b11, 1, 0, 0>;
2368 def M2_mpyd_nac_hl_s0: T_M2_mpyd_acc <0b10, 1, 0, 0>;
2369 def M2_mpyd_nac_lh_s0: T_M2_mpyd_acc <0b01, 1, 0, 0>;
2370 def M2_mpyd_nac_ll_s0: T_M2_mpyd_acc <0b00, 1, 0, 0>;
2372 def M2_mpyd_nac_hh_s1: T_M2_mpyd_acc <0b11, 1, 1, 0>;
2373 def M2_mpyd_nac_hl_s1: T_M2_mpyd_acc <0b10, 1, 1, 0>;
2374 def M2_mpyd_nac_lh_s1: T_M2_mpyd_acc <0b01, 1, 1, 0>;
2375 def M2_mpyd_nac_ll_s1: T_M2_mpyd_acc <0b00, 1, 1, 0>;
2377 def M2_mpyud_acc_hh_s0: T_M2_mpyd_acc <0b11, 0, 0, 1>;
2378 def M2_mpyud_acc_hl_s0: T_M2_mpyd_acc <0b10, 0, 0, 1>;
2379 def M2_mpyud_acc_lh_s0: T_M2_mpyd_acc <0b01, 0, 0, 1>;
2380 def M2_mpyud_acc_ll_s0: T_M2_mpyd_acc <0b00, 0, 0, 1>;
2382 def M2_mpyud_acc_hh_s1: T_M2_mpyd_acc <0b11, 0, 1, 1>;
2383 def M2_mpyud_acc_hl_s1: T_M2_mpyd_acc <0b10, 0, 1, 1>;
2384 def M2_mpyud_acc_lh_s1: T_M2_mpyd_acc <0b01, 0, 1, 1>;
2385 def M2_mpyud_acc_ll_s1: T_M2_mpyd_acc <0b00, 0, 1, 1>;
2387 def M2_mpyud_nac_hh_s0: T_M2_mpyd_acc <0b11, 1, 0, 1>;
2388 def M2_mpyud_nac_hl_s0: T_M2_mpyd_acc <0b10, 1, 0, 1>;
2389 def M2_mpyud_nac_lh_s0: T_M2_mpyd_acc <0b01, 1, 0, 1>;
2390 def M2_mpyud_nac_ll_s0: T_M2_mpyd_acc <0b00, 1, 0, 1>;
2392 def M2_mpyud_nac_hh_s1: T_M2_mpyd_acc <0b11, 1, 1, 1>;
2393 def M2_mpyud_nac_hl_s1: T_M2_mpyd_acc <0b10, 1, 1, 1>;
2394 def M2_mpyud_nac_lh_s1: T_M2_mpyd_acc <0b01, 1, 1, 1>;
2395 def M2_mpyud_nac_ll_s1: T_M2_mpyd_acc <0b00, 1, 1, 1>;
2397 //===----------------------------------------------------------------------===//
2398 // Template Class -- Vector Multipy
2399 // Used for complex multiply real or imaginary, dual multiply and even halfwords
2400 //===----------------------------------------------------------------------===//
2401 class T_M2_vmpy < string opc, bits<3> MajOp, bits<3> MinOp, bit hasShift,
2402 bit isRnd, bit isSat >
2403 : MInst <(outs DoubleRegs:$Rdd), (ins DoubleRegs:$Rss, DoubleRegs:$Rtt),
2404 "$Rdd = "#opc#"($Rss, $Rtt)"#!if(hasShift,":<<1","")
2405 #!if(isRnd,":rnd","")
2406 #!if(isSat,":sat",""),
2412 let IClass = 0b1110;
2414 let Inst{27-24} = 0b1000;
2415 let Inst{23-21} = MajOp;
2416 let Inst{7-5} = MinOp;
2417 let Inst{4-0} = Rdd;
2418 let Inst{20-16} = Rss;
2419 let Inst{12-8} = Rtt;
2422 // Vector complex multiply imaginary: Rdd=vcmpyi(Rss,Rtt)[:<<1]:sat
2423 let Defs = [USR_OVF] in {
2424 def M2_vcmpy_s1_sat_i: T_M2_vmpy <"vcmpyi", 0b110, 0b110, 1, 0, 1>;
2425 def M2_vcmpy_s0_sat_i: T_M2_vmpy <"vcmpyi", 0b010, 0b110, 0, 0, 1>;
2427 // Vector complex multiply real: Rdd=vcmpyr(Rss,Rtt)[:<<1]:sat
2428 def M2_vcmpy_s1_sat_r: T_M2_vmpy <"vcmpyr", 0b101, 0b110, 1, 0, 1>;
2429 def M2_vcmpy_s0_sat_r: T_M2_vmpy <"vcmpyr", 0b001, 0b110, 0, 0, 1>;
2431 // Vector dual multiply: Rdd=vdmpy(Rss,Rtt)[:<<1]:sat
2432 def M2_vdmpys_s1: T_M2_vmpy <"vdmpy", 0b100, 0b100, 1, 0, 1>;
2433 def M2_vdmpys_s0: T_M2_vmpy <"vdmpy", 0b000, 0b100, 0, 0, 1>;
2435 // Vector multiply even halfwords: Rdd=vmpyeh(Rss,Rtt)[:<<1]:sat
2436 def M2_vmpy2es_s1: T_M2_vmpy <"vmpyeh", 0b100, 0b110, 1, 0, 1>;
2437 def M2_vmpy2es_s0: T_M2_vmpy <"vmpyeh", 0b000, 0b110, 0, 0, 1>;
2439 //Rdd=vmpywoh(Rss,Rtt)[:<<1][:rnd]:sat
2440 def M2_mmpyh_s0: T_M2_vmpy <"vmpywoh", 0b000, 0b111, 0, 0, 1>;
2441 def M2_mmpyh_s1: T_M2_vmpy <"vmpywoh", 0b100, 0b111, 1, 0, 1>;
2442 def M2_mmpyh_rs0: T_M2_vmpy <"vmpywoh", 0b001, 0b111, 0, 1, 1>;
2443 def M2_mmpyh_rs1: T_M2_vmpy <"vmpywoh", 0b101, 0b111, 1, 1, 1>;
2445 //Rdd=vmpyweh(Rss,Rtt)[:<<1][:rnd]:sat
2446 def M2_mmpyl_s0: T_M2_vmpy <"vmpyweh", 0b000, 0b101, 0, 0, 1>;
2447 def M2_mmpyl_s1: T_M2_vmpy <"vmpyweh", 0b100, 0b101, 1, 0, 1>;
2448 def M2_mmpyl_rs0: T_M2_vmpy <"vmpyweh", 0b001, 0b101, 0, 1, 1>;
2449 def M2_mmpyl_rs1: T_M2_vmpy <"vmpyweh", 0b101, 0b101, 1, 1, 1>;
2451 //Rdd=vmpywouh(Rss,Rtt)[:<<1][:rnd]:sat
2452 def M2_mmpyuh_s0: T_M2_vmpy <"vmpywouh", 0b010, 0b111, 0, 0, 1>;
2453 def M2_mmpyuh_s1: T_M2_vmpy <"vmpywouh", 0b110, 0b111, 1, 0, 1>;
2454 def M2_mmpyuh_rs0: T_M2_vmpy <"vmpywouh", 0b011, 0b111, 0, 1, 1>;
2455 def M2_mmpyuh_rs1: T_M2_vmpy <"vmpywouh", 0b111, 0b111, 1, 1, 1>;
2457 //Rdd=vmpyweuh(Rss,Rtt)[:<<1][:rnd]:sat
2458 def M2_mmpyul_s0: T_M2_vmpy <"vmpyweuh", 0b010, 0b101, 0, 0, 1>;
2459 def M2_mmpyul_s1: T_M2_vmpy <"vmpyweuh", 0b110, 0b101, 1, 0, 1>;
2460 def M2_mmpyul_rs0: T_M2_vmpy <"vmpyweuh", 0b011, 0b101, 0, 1, 1>;
2461 def M2_mmpyul_rs1: T_M2_vmpy <"vmpyweuh", 0b111, 0b101, 1, 1, 1>;
2464 let hasNewValue = 1, opNewValue = 0 in
2465 class T_MType_mpy <string mnemonic, bits<4> RegTyBits, RegisterClass RC,
2466 bits<3> MajOp, bits<3> MinOp, bit isSat = 0, bit isRnd = 0,
2467 string op2Suffix = "", bit isRaw = 0, bit isHi = 0 >
2468 : MInst <(outs IntRegs:$dst), (ins RC:$src1, RC:$src2),
2470 #"($src1, $src2"#op2Suffix#")"
2471 #!if(MajOp{2}, ":<<1", "")
2472 #!if(isRnd, ":rnd", "")
2473 #!if(isSat, ":sat", "")
2474 #!if(isRaw, !if(isHi, ":raw:hi", ":raw:lo"), ""), [] > {
2479 let IClass = 0b1110;
2481 let Inst{27-24} = RegTyBits;
2482 let Inst{23-21} = MajOp;
2483 let Inst{20-16} = src1;
2485 let Inst{12-8} = src2;
2486 let Inst{7-5} = MinOp;
2487 let Inst{4-0} = dst;
2490 class T_MType_vrcmpy <string mnemonic, bits<3> MajOp, bits<3> MinOp, bit isHi>
2491 : T_MType_mpy <mnemonic, 0b1001, DoubleRegs, MajOp, MinOp, 1, 1, "", 1, isHi>;
2493 class T_MType_dd <string mnemonic, bits<3> MajOp, bits<3> MinOp,
2494 bit isSat = 0, bit isRnd = 0 >
2495 : T_MType_mpy <mnemonic, 0b1001, DoubleRegs, MajOp, MinOp, isSat, isRnd>;
2497 class T_MType_rr1 <string mnemonic, bits<3> MajOp, bits<3> MinOp,
2498 bit isSat = 0, bit isRnd = 0 >
2499 : T_MType_mpy<mnemonic, 0b1101, IntRegs, MajOp, MinOp, isSat, isRnd>;
2501 class T_MType_rr2 <string mnemonic, bits<3> MajOp, bits<3> MinOp,
2502 bit isSat = 0, bit isRnd = 0, string op2str = "" >
2503 : T_MType_mpy<mnemonic, 0b1101, IntRegs, MajOp, MinOp, isSat, isRnd, op2str>;
2505 def M2_vradduh : T_MType_dd <"vradduh", 0b000, 0b001, 0, 0>;
2506 def M2_vdmpyrs_s0 : T_MType_dd <"vdmpy", 0b000, 0b000, 1, 1>;
2507 def M2_vdmpyrs_s1 : T_MType_dd <"vdmpy", 0b100, 0b000, 1, 1>;
2509 let CextOpcode = "mpyi", InputType = "reg" in
2510 def M2_mpyi : T_MType_rr1 <"mpyi", 0b000, 0b000>, ImmRegRel;
2512 def M2_mpy_up : T_MType_rr1 <"mpy", 0b000, 0b001>;
2513 def M2_mpyu_up : T_MType_rr1 <"mpyu", 0b010, 0b001>;
2515 def M2_dpmpyss_rnd_s0 : T_MType_rr1 <"mpy", 0b001, 0b001, 0, 1>;
2517 def M2_vmpy2s_s0pack : T_MType_rr1 <"vmpyh", 0b001, 0b111, 1, 1>;
2518 def M2_vmpy2s_s1pack : T_MType_rr1 <"vmpyh", 0b101, 0b111, 1, 1>;
2520 def M2_hmmpyh_rs1 : T_MType_rr2 <"mpy", 0b101, 0b100, 1, 1, ".h">;
2521 def M2_hmmpyl_rs1 : T_MType_rr2 <"mpy", 0b111, 0b100, 1, 1, ".l">;
2523 def M2_cmpyrs_s0 : T_MType_rr2 <"cmpy", 0b001, 0b110, 1, 1>;
2524 def M2_cmpyrs_s1 : T_MType_rr2 <"cmpy", 0b101, 0b110, 1, 1>;
2525 def M2_cmpyrsc_s0 : T_MType_rr2 <"cmpy", 0b011, 0b110, 1, 1, "*">;
2526 def M2_cmpyrsc_s1 : T_MType_rr2 <"cmpy", 0b111, 0b110, 1, 1, "*">;
2529 def M2_vraddh : T_MType_dd <"vraddh", 0b001, 0b111, 0>;
2530 def M2_mpysu_up : T_MType_rr1 <"mpysu", 0b011, 0b001, 0>;
2531 def M2_mpy_up_s1 : T_MType_rr1 <"mpy", 0b101, 0b010, 0>;
2532 def M2_mpy_up_s1_sat : T_MType_rr1 <"mpy", 0b111, 0b000, 1>;
2534 def M2_hmmpyh_s1 : T_MType_rr2 <"mpy", 0b101, 0b000, 1, 0, ".h">;
2535 def M2_hmmpyl_s1 : T_MType_rr2 <"mpy", 0b101, 0b001, 1, 0, ".l">;
2537 def: Pat<(i32 (mul I32:$src1, I32:$src2)), (M2_mpyi I32:$src1, I32:$src2)>;
2538 def: Pat<(i32 (mulhs I32:$src1, I32:$src2)), (M2_mpy_up I32:$src1, I32:$src2)>;
2539 def: Pat<(i32 (mulhu I32:$src1, I32:$src2)), (M2_mpyu_up I32:$src1, I32:$src2)>;
2541 let hasNewValue = 1, opNewValue = 0 in
2542 class T_MType_mpy_ri <bit isNeg, Operand ImmOp, list<dag> pattern>
2543 : MInst < (outs IntRegs:$Rd), (ins IntRegs:$Rs, ImmOp:$u8),
2544 "$Rd ="#!if(isNeg, "- ", "+ ")#"mpyi($Rs, #$u8)" ,
2545 pattern, "", M_tc_3x_SLOT23> {
2550 let IClass = 0b1110;
2552 let Inst{27-24} = 0b0000;
2553 let Inst{23} = isNeg;
2556 let Inst{20-16} = Rs;
2557 let Inst{12-5} = u8;
2560 let isExtendable = 1, opExtentBits = 8, opExtendable = 2 in
2561 def M2_mpysip : T_MType_mpy_ri <0, u8Ext,
2562 [(set (i32 IntRegs:$Rd), (mul IntRegs:$Rs, u8ExtPred:$u8))]>;
2564 def M2_mpysin : T_MType_mpy_ri <1, u8Imm,
2565 [(set (i32 IntRegs:$Rd), (ineg (mul IntRegs:$Rs,
2568 // Assember mapped to M2_mpyi
2569 let isAsmParserOnly = 1 in
2570 def M2_mpyui : MInst<(outs IntRegs:$dst),
2571 (ins IntRegs:$src1, IntRegs:$src2),
2572 "$dst = mpyui($src1, $src2)">;
2575 // s9 is NOT the same as m9 - but it works.. so far.
2576 // Assembler maps to either Rd=+mpyi(Rs,#u8) or Rd=-mpyi(Rs,#u8)
2577 // depending on the value of m9. See Arch Spec.
2578 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 9,
2579 CextOpcode = "mpyi", InputType = "imm", hasNewValue = 1,
2580 isAsmParserOnly = 1 in
2581 def M2_mpysmi : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, s9Ext:$src2),
2582 "$dst = mpyi($src1, #$src2)",
2583 [(set (i32 IntRegs:$dst), (mul (i32 IntRegs:$src1),
2584 s9ExtPred:$src2))]>, ImmRegRel;
2586 let hasNewValue = 1, isExtendable = 1, opExtentBits = 8, opExtendable = 3,
2587 InputType = "imm" in
2588 class T_MType_acc_ri <string mnemonic, bits<3> MajOp, Operand ImmOp,
2589 list<dag> pattern = []>
2590 : MInst < (outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, ImmOp:$src3),
2591 "$dst "#mnemonic#"($src2, #$src3)",
2592 pattern, "$src1 = $dst", M_tc_2_SLOT23> {
2597 let IClass = 0b1110;
2599 let Inst{27-26} = 0b00;
2600 let Inst{25-23} = MajOp;
2601 let Inst{20-16} = src2;
2603 let Inst{12-5} = src3;
2604 let Inst{4-0} = dst;
2607 let InputType = "reg", hasNewValue = 1 in
2608 class T_MType_acc_rr <string mnemonic, bits<3> MajOp, bits<3> MinOp,
2609 bit isSwap = 0, list<dag> pattern = [], bit hasNot = 0,
2610 bit isSat = 0, bit isShift = 0>
2611 : MInst < (outs IntRegs:$dst),
2612 (ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3),
2613 "$dst "#mnemonic#"($src2, "#!if(hasNot, "~$src3)","$src3)")
2614 #!if(isShift, ":<<1", "")
2615 #!if(isSat, ":sat", ""),
2616 pattern, "$src1 = $dst", M_tc_2_SLOT23 > {
2621 let IClass = 0b1110;
2623 let Inst{27-24} = 0b1111;
2624 let Inst{23-21} = MajOp;
2625 let Inst{20-16} = !if(isSwap, src3, src2);
2627 let Inst{12-8} = !if(isSwap, src2, src3);
2628 let Inst{7-5} = MinOp;
2629 let Inst{4-0} = dst;
2632 let CextOpcode = "MPYI_acc", Itinerary = M_tc_3x_SLOT23 in {
2633 def M2_macsip : T_MType_acc_ri <"+= mpyi", 0b010, u8Ext,
2634 [(set (i32 IntRegs:$dst),
2635 (add (mul IntRegs:$src2, u8ExtPred:$src3),
2636 IntRegs:$src1))]>, ImmRegRel;
2638 def M2_maci : T_MType_acc_rr <"+= mpyi", 0b000, 0b000, 0,
2639 [(set (i32 IntRegs:$dst),
2640 (add (mul IntRegs:$src2, IntRegs:$src3),
2641 IntRegs:$src1))]>, ImmRegRel;
2644 let CextOpcode = "ADD_acc" in {
2645 let isExtentSigned = 1 in
2646 def M2_accii : T_MType_acc_ri <"+= add", 0b100, s8Ext,
2647 [(set (i32 IntRegs:$dst),
2648 (add (add (i32 IntRegs:$src2), s8_16ExtPred:$src3),
2649 (i32 IntRegs:$src1)))]>, ImmRegRel;
2651 def M2_acci : T_MType_acc_rr <"+= add", 0b000, 0b001, 0,
2652 [(set (i32 IntRegs:$dst),
2653 (add (add (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
2654 (i32 IntRegs:$src1)))]>, ImmRegRel;
2657 let CextOpcode = "SUB_acc" in {
2658 let isExtentSigned = 1 in
2659 def M2_naccii : T_MType_acc_ri <"-= add", 0b101, s8Ext>, ImmRegRel;
2661 def M2_nacci : T_MType_acc_rr <"-= add", 0b100, 0b001, 0>, ImmRegRel;
2664 let Itinerary = M_tc_3x_SLOT23 in
2665 def M2_macsin : T_MType_acc_ri <"-= mpyi", 0b011, u8Ext>;
2667 def M2_xor_xacc : T_MType_acc_rr < "^= xor", 0b100, 0b011, 0>;
2668 def M2_subacc : T_MType_acc_rr <"+= sub", 0b000, 0b011, 1>;
2670 class T_MType_acc_pat1 <InstHexagon MI, SDNode firstOp, SDNode secOp,
2672 : Pat <(secOp IntRegs:$src1, (firstOp IntRegs:$src2, ImmPred:$src3)),
2673 (MI IntRegs:$src1, IntRegs:$src2, ImmPred:$src3)>;
2675 class T_MType_acc_pat2 <InstHexagon MI, SDNode firstOp, SDNode secOp>
2676 : Pat <(i32 (secOp IntRegs:$src1, (firstOp IntRegs:$src2, IntRegs:$src3))),
2677 (MI IntRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
2679 def : T_MType_acc_pat2 <M2_xor_xacc, xor, xor>;
2680 def : T_MType_acc_pat1 <M2_macsin, mul, sub, u8ExtPred>;
2682 def : T_MType_acc_pat1 <M2_naccii, add, sub, s8_16ExtPred>;
2683 def : T_MType_acc_pat2 <M2_nacci, add, sub>;
2685 //===----------------------------------------------------------------------===//
2686 // Template Class -- XType Vector Instructions
2687 //===----------------------------------------------------------------------===//
2688 class T_XTYPE_Vect < string opc, bits<3> MajOp, bits<3> MinOp, bit isConj >
2689 : MInst <(outs DoubleRegs:$Rdd), (ins DoubleRegs:$Rss, DoubleRegs:$Rtt),
2690 "$Rdd = "#opc#"($Rss, $Rtt"#!if(isConj,"*)",")"),
2696 let IClass = 0b1110;
2698 let Inst{27-24} = 0b1000;
2699 let Inst{23-21} = MajOp;
2700 let Inst{7-5} = MinOp;
2701 let Inst{4-0} = Rdd;
2702 let Inst{20-16} = Rss;
2703 let Inst{12-8} = Rtt;
2706 class T_XTYPE_Vect_acc < string opc, bits<3> MajOp, bits<3> MinOp, bit isConj >
2707 : MInst <(outs DoubleRegs:$Rdd),
2708 (ins DoubleRegs:$dst2, DoubleRegs:$Rss, DoubleRegs:$Rtt),
2709 "$Rdd += "#opc#"($Rss, $Rtt"#!if(isConj,"*)",")"),
2710 [], "$dst2 = $Rdd",M_tc_3x_SLOT23 > {
2715 let IClass = 0b1110;
2717 let Inst{27-24} = 0b1010;
2718 let Inst{23-21} = MajOp;
2719 let Inst{7-5} = MinOp;
2720 let Inst{4-0} = Rdd;
2721 let Inst{20-16} = Rss;
2722 let Inst{12-8} = Rtt;
2725 class T_XTYPE_Vect_diff < bits<3> MajOp, string opc >
2726 : MInst <(outs DoubleRegs:$Rdd), (ins DoubleRegs:$Rtt, DoubleRegs:$Rss),
2727 "$Rdd = "#opc#"($Rtt, $Rss)",
2728 [], "",M_tc_2_SLOT23 > {
2733 let IClass = 0b1110;
2735 let Inst{27-24} = 0b1000;
2736 let Inst{23-21} = MajOp;
2737 let Inst{7-5} = 0b000;
2738 let Inst{4-0} = Rdd;
2739 let Inst{20-16} = Rss;
2740 let Inst{12-8} = Rtt;
2743 // Vector reduce add unsigned bytes: Rdd32=vrmpybu(Rss32,Rtt32)
2744 def A2_vraddub: T_XTYPE_Vect <"vraddub", 0b010, 0b001, 0>;
2745 def A2_vraddub_acc: T_XTYPE_Vect_acc <"vraddub", 0b010, 0b001, 0>;
2747 // Vector sum of absolute differences unsigned bytes: Rdd=vrsadub(Rss,Rtt)
2748 def A2_vrsadub: T_XTYPE_Vect <"vrsadub", 0b010, 0b010, 0>;
2749 def A2_vrsadub_acc: T_XTYPE_Vect_acc <"vrsadub", 0b010, 0b010, 0>;
2751 // Vector absolute difference: Rdd=vabsdiffh(Rtt,Rss)
2752 def M2_vabsdiffh: T_XTYPE_Vect_diff<0b011, "vabsdiffh">;
2754 // Vector absolute difference words: Rdd=vabsdiffw(Rtt,Rss)
2755 def M2_vabsdiffw: T_XTYPE_Vect_diff<0b001, "vabsdiffw">;
2757 // Vector reduce complex multiply real or imaginary:
2758 // Rdd[+]=vrcmpy[ir](Rss,Rtt[*])
2759 def M2_vrcmpyi_s0: T_XTYPE_Vect <"vrcmpyi", 0b000, 0b000, 0>;
2760 def M2_vrcmpyi_s0c: T_XTYPE_Vect <"vrcmpyi", 0b010, 0b000, 1>;
2761 def M2_vrcmaci_s0: T_XTYPE_Vect_acc <"vrcmpyi", 0b000, 0b000, 0>;
2762 def M2_vrcmaci_s0c: T_XTYPE_Vect_acc <"vrcmpyi", 0b010, 0b000, 1>;
2764 def M2_vrcmpyr_s0: T_XTYPE_Vect <"vrcmpyr", 0b000, 0b001, 0>;
2765 def M2_vrcmpyr_s0c: T_XTYPE_Vect <"vrcmpyr", 0b011, 0b001, 1>;
2766 def M2_vrcmacr_s0: T_XTYPE_Vect_acc <"vrcmpyr", 0b000, 0b001, 0>;
2767 def M2_vrcmacr_s0c: T_XTYPE_Vect_acc <"vrcmpyr", 0b011, 0b001, 1>;
2769 // Vector reduce halfwords:
2770 // Rdd[+]=vrmpyh(Rss,Rtt)
2771 def M2_vrmpy_s0: T_XTYPE_Vect <"vrmpyh", 0b000, 0b010, 0>;
2772 def M2_vrmac_s0: T_XTYPE_Vect_acc <"vrmpyh", 0b000, 0b010, 0>;
2774 //===----------------------------------------------------------------------===//
2775 // Template Class -- Vector Multipy with accumulation.
2776 // Used for complex multiply real or imaginary, dual multiply and even halfwords
2777 //===----------------------------------------------------------------------===//
2778 let Defs = [USR_OVF] in
2779 class T_M2_vmpy_acc_sat < string opc, bits<3> MajOp, bits<3> MinOp,
2780 bit hasShift, bit isRnd >
2781 : MInst <(outs DoubleRegs:$Rxx),
2782 (ins DoubleRegs:$dst2, DoubleRegs:$Rss, DoubleRegs:$Rtt),
2783 "$Rxx += "#opc#"($Rss, $Rtt)"#!if(hasShift,":<<1","")
2784 #!if(isRnd,":rnd","")#":sat",
2785 [], "$dst2 = $Rxx",M_tc_3x_SLOT23 > {
2790 let IClass = 0b1110;
2792 let Inst{27-24} = 0b1010;
2793 let Inst{23-21} = MajOp;
2794 let Inst{7-5} = MinOp;
2795 let Inst{4-0} = Rxx;
2796 let Inst{20-16} = Rss;
2797 let Inst{12-8} = Rtt;
2800 class T_M2_vmpy_acc < string opc, bits<3> MajOp, bits<3> MinOp,
2801 bit hasShift, bit isRnd >
2802 : MInst <(outs DoubleRegs:$Rxx),
2803 (ins DoubleRegs:$dst2, DoubleRegs:$Rss, DoubleRegs:$Rtt),
2804 "$Rxx += "#opc#"($Rss, $Rtt)"#!if(hasShift,":<<1","")
2805 #!if(isRnd,":rnd",""),
2806 [], "$dst2 = $Rxx",M_tc_3x_SLOT23 > {
2811 let IClass = 0b1110;
2813 let Inst{27-24} = 0b1010;
2814 let Inst{23-21} = MajOp;
2815 let Inst{7-5} = MinOp;
2816 let Inst{4-0} = Rxx;
2817 let Inst{20-16} = Rss;
2818 let Inst{12-8} = Rtt;
2821 // Vector multiply word by signed half with accumulation
2822 // Rxx+=vmpyw[eo]h(Rss,Rtt)[:<<1][:rnd]:sat
2823 def M2_mmacls_s1: T_M2_vmpy_acc_sat <"vmpyweh", 0b100, 0b101, 1, 0>;
2824 def M2_mmacls_s0: T_M2_vmpy_acc_sat <"vmpyweh", 0b000, 0b101, 0, 0>;
2825 def M2_mmacls_rs1: T_M2_vmpy_acc_sat <"vmpyweh", 0b101, 0b101, 1, 1>;
2826 def M2_mmacls_rs0: T_M2_vmpy_acc_sat <"vmpyweh", 0b001, 0b101, 0, 1>;
2828 def M2_mmachs_s1: T_M2_vmpy_acc_sat <"vmpywoh", 0b100, 0b111, 1, 0>;
2829 def M2_mmachs_s0: T_M2_vmpy_acc_sat <"vmpywoh", 0b000, 0b111, 0, 0>;
2830 def M2_mmachs_rs1: T_M2_vmpy_acc_sat <"vmpywoh", 0b101, 0b111, 1, 1>;
2831 def M2_mmachs_rs0: T_M2_vmpy_acc_sat <"vmpywoh", 0b001, 0b111, 0, 1>;
2833 // Vector multiply word by unsigned half with accumulation
2834 // Rxx+=vmpyw[eo]uh(Rss,Rtt)[:<<1][:rnd]:sat
2835 def M2_mmaculs_s1: T_M2_vmpy_acc_sat <"vmpyweuh", 0b110, 0b101, 1, 0>;
2836 def M2_mmaculs_s0: T_M2_vmpy_acc_sat <"vmpyweuh", 0b010, 0b101, 0, 0>;
2837 def M2_mmaculs_rs1: T_M2_vmpy_acc_sat <"vmpyweuh", 0b111, 0b101, 1, 1>;
2838 def M2_mmaculs_rs0: T_M2_vmpy_acc_sat <"vmpyweuh", 0b011, 0b101, 0, 1>;
2840 def M2_mmacuhs_s1: T_M2_vmpy_acc_sat <"vmpywouh", 0b110, 0b111, 1, 0>;
2841 def M2_mmacuhs_s0: T_M2_vmpy_acc_sat <"vmpywouh", 0b010, 0b111, 0, 0>;
2842 def M2_mmacuhs_rs1: T_M2_vmpy_acc_sat <"vmpywouh", 0b111, 0b111, 1, 1>;
2843 def M2_mmacuhs_rs0: T_M2_vmpy_acc_sat <"vmpywouh", 0b011, 0b111, 0, 1>;
2845 // Vector multiply even halfwords with accumulation
2846 // Rxx+=vmpyeh(Rss,Rtt)[:<<1][:sat]
2847 def M2_vmac2es: T_M2_vmpy_acc <"vmpyeh", 0b001, 0b010, 0, 0>;
2848 def M2_vmac2es_s1: T_M2_vmpy_acc_sat <"vmpyeh", 0b100, 0b110, 1, 0>;
2849 def M2_vmac2es_s0: T_M2_vmpy_acc_sat <"vmpyeh", 0b000, 0b110, 0, 0>;
2851 // Vector dual multiply with accumulation
2852 // Rxx+=vdmpy(Rss,Rtt)[:sat]
2853 def M2_vdmacs_s1: T_M2_vmpy_acc_sat <"vdmpy", 0b100, 0b100, 1, 0>;
2854 def M2_vdmacs_s0: T_M2_vmpy_acc_sat <"vdmpy", 0b000, 0b100, 0, 0>;
2856 // Vector complex multiply real or imaginary with accumulation
2857 // Rxx+=vcmpy[ir](Rss,Rtt):sat
2858 def M2_vcmac_s0_sat_r: T_M2_vmpy_acc_sat <"vcmpyr", 0b001, 0b100, 0, 0>;
2859 def M2_vcmac_s0_sat_i: T_M2_vmpy_acc_sat <"vcmpyi", 0b010, 0b100, 0, 0>;
2861 //===----------------------------------------------------------------------===//
2862 // Template Class -- Multiply signed/unsigned halfwords with and without
2863 // saturation and rounding
2864 //===----------------------------------------------------------------------===//
2865 class T_M2_mpyd < bits<2> LHbits, bit isRnd, bit hasShift, bit isUnsigned >
2866 : MInst < (outs DoubleRegs:$Rdd), (ins IntRegs:$Rs, IntRegs:$Rt),
2867 "$Rdd = "#!if(isUnsigned,"mpyu","mpy")#"($Rs."#!if(LHbits{1},"h","l")
2868 #", $Rt."#!if(LHbits{0},"h)","l)")
2869 #!if(hasShift,":<<1","")
2870 #!if(isRnd,":rnd",""),
2876 let IClass = 0b1110;
2878 let Inst{27-24} = 0b0100;
2879 let Inst{23} = hasShift;
2880 let Inst{22} = isUnsigned;
2881 let Inst{21} = isRnd;
2882 let Inst{6-5} = LHbits;
2883 let Inst{4-0} = Rdd;
2884 let Inst{20-16} = Rs;
2885 let Inst{12-8} = Rt;
2888 def M2_mpyd_hh_s0: T_M2_mpyd<0b11, 0, 0, 0>;
2889 def M2_mpyd_hl_s0: T_M2_mpyd<0b10, 0, 0, 0>;
2890 def M2_mpyd_lh_s0: T_M2_mpyd<0b01, 0, 0, 0>;
2891 def M2_mpyd_ll_s0: T_M2_mpyd<0b00, 0, 0, 0>;
2893 def M2_mpyd_hh_s1: T_M2_mpyd<0b11, 0, 1, 0>;
2894 def M2_mpyd_hl_s1: T_M2_mpyd<0b10, 0, 1, 0>;
2895 def M2_mpyd_lh_s1: T_M2_mpyd<0b01, 0, 1, 0>;
2896 def M2_mpyd_ll_s1: T_M2_mpyd<0b00, 0, 1, 0>;
2898 def M2_mpyd_rnd_hh_s0: T_M2_mpyd<0b11, 1, 0, 0>;
2899 def M2_mpyd_rnd_hl_s0: T_M2_mpyd<0b10, 1, 0, 0>;
2900 def M2_mpyd_rnd_lh_s0: T_M2_mpyd<0b01, 1, 0, 0>;
2901 def M2_mpyd_rnd_ll_s0: T_M2_mpyd<0b00, 1, 0, 0>;
2903 def M2_mpyd_rnd_hh_s1: T_M2_mpyd<0b11, 1, 1, 0>;
2904 def M2_mpyd_rnd_hl_s1: T_M2_mpyd<0b10, 1, 1, 0>;
2905 def M2_mpyd_rnd_lh_s1: T_M2_mpyd<0b01, 1, 1, 0>;
2906 def M2_mpyd_rnd_ll_s1: T_M2_mpyd<0b00, 1, 1, 0>;
2908 //Rdd=mpyu(Rs.[HL],Rt.[HL])[:<<1]
2909 def M2_mpyud_hh_s0: T_M2_mpyd<0b11, 0, 0, 1>;
2910 def M2_mpyud_hl_s0: T_M2_mpyd<0b10, 0, 0, 1>;
2911 def M2_mpyud_lh_s0: T_M2_mpyd<0b01, 0, 0, 1>;
2912 def M2_mpyud_ll_s0: T_M2_mpyd<0b00, 0, 0, 1>;
2914 def M2_mpyud_hh_s1: T_M2_mpyd<0b11, 0, 1, 1>;
2915 def M2_mpyud_hl_s1: T_M2_mpyd<0b10, 0, 1, 1>;
2916 def M2_mpyud_lh_s1: T_M2_mpyd<0b01, 0, 1, 1>;
2917 def M2_mpyud_ll_s1: T_M2_mpyd<0b00, 0, 1, 1>;
2919 //===----------------------------------------------------------------------===//
2920 // Template Class for xtype mpy:
2923 // multiply 32X32 and use full result
2924 //===----------------------------------------------------------------------===//
2925 let hasSideEffects = 0 in
2926 class T_XTYPE_mpy64 <string mnemonic, bits<3> MajOp, bits<3> MinOp,
2927 bit isSat, bit hasShift, bit isConj>
2928 : MInst <(outs DoubleRegs:$Rdd),
2929 (ins IntRegs:$Rs, IntRegs:$Rt),
2930 "$Rdd = "#mnemonic#"($Rs, $Rt"#!if(isConj,"*)",")")
2931 #!if(hasShift,":<<1","")
2932 #!if(isSat,":sat",""),
2938 let IClass = 0b1110;
2940 let Inst{27-24} = 0b0101;
2941 let Inst{23-21} = MajOp;
2942 let Inst{20-16} = Rs;
2943 let Inst{12-8} = Rt;
2944 let Inst{7-5} = MinOp;
2945 let Inst{4-0} = Rdd;
2948 //===----------------------------------------------------------------------===//
2949 // Template Class for xtype mpy with accumulation into 64-bit:
2952 // multiply 32X32 and use full result
2953 //===----------------------------------------------------------------------===//
2954 class T_XTYPE_mpy64_acc <string op1, string op2, bits<3> MajOp, bits<3> MinOp,
2955 bit isSat, bit hasShift, bit isConj>
2956 : MInst <(outs DoubleRegs:$Rxx),
2957 (ins DoubleRegs:$dst2, IntRegs:$Rs, IntRegs:$Rt),
2958 "$Rxx "#op2#"= "#op1#"($Rs, $Rt"#!if(isConj,"*)",")")
2959 #!if(hasShift,":<<1","")
2960 #!if(isSat,":sat",""),
2962 [] , "$dst2 = $Rxx" > {
2967 let IClass = 0b1110;
2969 let Inst{27-24} = 0b0111;
2970 let Inst{23-21} = MajOp;
2971 let Inst{20-16} = Rs;
2972 let Inst{12-8} = Rt;
2973 let Inst{7-5} = MinOp;
2974 let Inst{4-0} = Rxx;
2977 // MPY - Multiply and use full result
2978 // Rdd = mpy[u](Rs,Rt)
2979 def M2_dpmpyss_s0 : T_XTYPE_mpy64 < "mpy", 0b000, 0b000, 0, 0, 0>;
2980 def M2_dpmpyuu_s0 : T_XTYPE_mpy64 < "mpyu", 0b010, 0b000, 0, 0, 0>;
2982 // Rxx[+-]= mpy[u](Rs,Rt)
2983 def M2_dpmpyss_acc_s0 : T_XTYPE_mpy64_acc < "mpy", "+", 0b000, 0b000, 0, 0, 0>;
2984 def M2_dpmpyss_nac_s0 : T_XTYPE_mpy64_acc < "mpy", "-", 0b001, 0b000, 0, 0, 0>;
2985 def M2_dpmpyuu_acc_s0 : T_XTYPE_mpy64_acc < "mpyu", "+", 0b010, 0b000, 0, 0, 0>;
2986 def M2_dpmpyuu_nac_s0 : T_XTYPE_mpy64_acc < "mpyu", "-", 0b011, 0b000, 0, 0, 0>;
2988 // Complex multiply real or imaginary
2989 // Rxx=cmpy[ir](Rs,Rt)
2990 def M2_cmpyi_s0 : T_XTYPE_mpy64 < "cmpyi", 0b000, 0b001, 0, 0, 0>;
2991 def M2_cmpyr_s0 : T_XTYPE_mpy64 < "cmpyr", 0b000, 0b010, 0, 0, 0>;
2993 // Rxx+=cmpy[ir](Rs,Rt)
2994 def M2_cmaci_s0 : T_XTYPE_mpy64_acc < "cmpyi", "+", 0b000, 0b001, 0, 0, 0>;
2995 def M2_cmacr_s0 : T_XTYPE_mpy64_acc < "cmpyr", "+", 0b000, 0b010, 0, 0, 0>;
2998 // Rdd=cmpy(Rs,Rt)[:<<]:sat
2999 def M2_cmpys_s0 : T_XTYPE_mpy64 < "cmpy", 0b000, 0b110, 1, 0, 0>;
3000 def M2_cmpys_s1 : T_XTYPE_mpy64 < "cmpy", 0b100, 0b110, 1, 1, 0>;
3002 // Rdd=cmpy(Rs,Rt*)[:<<]:sat
3003 def M2_cmpysc_s0 : T_XTYPE_mpy64 < "cmpy", 0b010, 0b110, 1, 0, 1>;
3004 def M2_cmpysc_s1 : T_XTYPE_mpy64 < "cmpy", 0b110, 0b110, 1, 1, 1>;
3006 // Rxx[-+]=cmpy(Rs,Rt)[:<<1]:sat
3007 def M2_cmacs_s0 : T_XTYPE_mpy64_acc < "cmpy", "+", 0b000, 0b110, 1, 0, 0>;
3008 def M2_cnacs_s0 : T_XTYPE_mpy64_acc < "cmpy", "-", 0b000, 0b111, 1, 0, 0>;
3009 def M2_cmacs_s1 : T_XTYPE_mpy64_acc < "cmpy", "+", 0b100, 0b110, 1, 1, 0>;
3010 def M2_cnacs_s1 : T_XTYPE_mpy64_acc < "cmpy", "-", 0b100, 0b111, 1, 1, 0>;
3012 // Rxx[-+]=cmpy(Rs,Rt*)[:<<1]:sat
3013 def M2_cmacsc_s0 : T_XTYPE_mpy64_acc < "cmpy", "+", 0b010, 0b110, 1, 0, 1>;
3014 def M2_cnacsc_s0 : T_XTYPE_mpy64_acc < "cmpy", "-", 0b010, 0b111, 1, 0, 1>;
3015 def M2_cmacsc_s1 : T_XTYPE_mpy64_acc < "cmpy", "+", 0b110, 0b110, 1, 1, 1>;
3016 def M2_cnacsc_s1 : T_XTYPE_mpy64_acc < "cmpy", "-", 0b110, 0b111, 1, 1, 1>;
3018 // Vector multiply halfwords
3019 // Rdd=vmpyh(Rs,Rt)[:<<]:sat
3020 //let Defs = [USR_OVF] in {
3021 def M2_vmpy2s_s1 : T_XTYPE_mpy64 < "vmpyh", 0b100, 0b101, 1, 1, 0>;
3022 def M2_vmpy2s_s0 : T_XTYPE_mpy64 < "vmpyh", 0b000, 0b101, 1, 0, 0>;
3025 // Rxx+=vmpyh(Rs,Rt)[:<<1][:sat]
3026 def M2_vmac2 : T_XTYPE_mpy64_acc < "vmpyh", "+", 0b001, 0b001, 0, 0, 0>;
3027 def M2_vmac2s_s1 : T_XTYPE_mpy64_acc < "vmpyh", "+", 0b100, 0b101, 1, 1, 0>;
3028 def M2_vmac2s_s0 : T_XTYPE_mpy64_acc < "vmpyh", "+", 0b000, 0b101, 1, 0, 0>;
3030 def: Pat<(i64 (mul (i64 (anyext (i32 IntRegs:$src1))),
3031 (i64 (anyext (i32 IntRegs:$src2))))),
3032 (M2_dpmpyuu_s0 IntRegs:$src1, IntRegs:$src2)>;
3034 def: Pat<(i64 (mul (i64 (sext (i32 IntRegs:$src1))),
3035 (i64 (sext (i32 IntRegs:$src2))))),
3036 (M2_dpmpyss_s0 IntRegs:$src1, IntRegs:$src2)>;
3038 def: Pat<(i64 (mul (is_sext_i32:$src1),
3039 (is_sext_i32:$src2))),
3040 (M2_dpmpyss_s0 (LoReg DoubleRegs:$src1), (LoReg DoubleRegs:$src2))>;
3042 // Multiply and accumulate, use full result.
3043 // Rxx[+-]=mpy(Rs,Rt)
3045 def: Pat<(i64 (add (i64 DoubleRegs:$src1),
3046 (mul (i64 (sext (i32 IntRegs:$src2))),
3047 (i64 (sext (i32 IntRegs:$src3)))))),
3048 (M2_dpmpyss_acc_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
3050 def: Pat<(i64 (sub (i64 DoubleRegs:$src1),
3051 (mul (i64 (sext (i32 IntRegs:$src2))),
3052 (i64 (sext (i32 IntRegs:$src3)))))),
3053 (M2_dpmpyss_nac_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
3055 def: Pat<(i64 (add (i64 DoubleRegs:$src1),
3056 (mul (i64 (anyext (i32 IntRegs:$src2))),
3057 (i64 (anyext (i32 IntRegs:$src3)))))),
3058 (M2_dpmpyuu_acc_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
3060 def: Pat<(i64 (add (i64 DoubleRegs:$src1),
3061 (mul (i64 (zext (i32 IntRegs:$src2))),
3062 (i64 (zext (i32 IntRegs:$src3)))))),
3063 (M2_dpmpyuu_acc_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
3065 def: Pat<(i64 (sub (i64 DoubleRegs:$src1),
3066 (mul (i64 (anyext (i32 IntRegs:$src2))),
3067 (i64 (anyext (i32 IntRegs:$src3)))))),
3068 (M2_dpmpyuu_nac_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
3070 def: Pat<(i64 (sub (i64 DoubleRegs:$src1),
3071 (mul (i64 (zext (i32 IntRegs:$src2))),
3072 (i64 (zext (i32 IntRegs:$src3)))))),
3073 (M2_dpmpyuu_nac_s0 DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3)>;
3075 //===----------------------------------------------------------------------===//
3077 //===----------------------------------------------------------------------===//
3079 //===----------------------------------------------------------------------===//
3081 //===----------------------------------------------------------------------===//
3082 //===----------------------------------------------------------------------===//
3084 //===----------------------------------------------------------------------===//
3086 //===----------------------------------------------------------------------===//
3088 //===----------------------------------------------------------------------===//
3089 //===----------------------------------------------------------------------===//
3091 //===----------------------------------------------------------------------===//
3093 //===----------------------------------------------------------------------===//
3095 //===----------------------------------------------------------------------===//
3096 //===----------------------------------------------------------------------===//
3098 //===----------------------------------------------------------------------===//
3100 //===----------------------------------------------------------------------===//
3102 //===----------------------------------------------------------------------===//
3104 // Store doubleword.
3105 //===----------------------------------------------------------------------===//
3106 // Template class for non-predicated post increment stores with immediate offset
3107 //===----------------------------------------------------------------------===//
3108 let isPredicable = 1, hasSideEffects = 0, addrMode = PostInc in
3109 class T_store_pi <string mnemonic, RegisterClass RC, Operand ImmOp,
3110 bits<4> MajOp, bit isHalf >
3111 : STInst <(outs IntRegs:$_dst_),
3112 (ins IntRegs:$src1, ImmOp:$offset, RC:$src2),
3113 mnemonic#"($src1++#$offset) = $src2"#!if(isHalf, ".h", ""),
3114 [], "$src1 = $_dst_" >,
3121 string ImmOpStr = !cast<string>(ImmOp);
3122 let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
3123 !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
3124 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
3125 /* s4_0Imm */ offset{3-0})));
3126 let isNVStorable = !if (!eq(ImmOpStr, "s4_3Imm"), 0, 1);
3128 let IClass = 0b1010;
3130 let Inst{27-25} = 0b101;
3131 let Inst{24-21} = MajOp;
3132 let Inst{20-16} = src1;
3134 let Inst{12-8} = src2;
3136 let Inst{6-3} = offsetBits;
3140 //===----------------------------------------------------------------------===//
3141 // Template class for predicated post increment stores with immediate offset
3142 //===----------------------------------------------------------------------===//
3143 let isPredicated = 1, hasSideEffects = 0, addrMode = PostInc in
3144 class T_pstore_pi <string mnemonic, RegisterClass RC, Operand ImmOp,
3145 bits<4> MajOp, bit isHalf, bit isPredNot, bit isPredNew >
3146 : STInst <(outs IntRegs:$_dst_),
3147 (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset, RC:$src3),
3148 !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
3149 ") ")#mnemonic#"($src2++#$offset) = $src3"#!if(isHalf, ".h", ""),
3150 [], "$src2 = $_dst_" >,
3158 string ImmOpStr = !cast<string>(ImmOp);
3159 let offsetBits = !if (!eq(ImmOpStr, "s4_3Imm"), offset{6-3},
3160 !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
3161 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
3162 /* s4_0Imm */ offset{3-0})));
3164 let isNVStorable = !if (!eq(ImmOpStr, "s4_3Imm"), 0, 1);
3165 let isPredicatedNew = isPredNew;
3166 let isPredicatedFalse = isPredNot;
3168 let IClass = 0b1010;
3170 let Inst{27-25} = 0b101;
3171 let Inst{24-21} = MajOp;
3172 let Inst{20-16} = src2;
3174 let Inst{12-8} = src3;
3175 let Inst{7} = isPredNew;
3176 let Inst{6-3} = offsetBits;
3177 let Inst{2} = isPredNot;
3178 let Inst{1-0} = src1;
3181 multiclass ST_PostInc<string mnemonic, string BaseOp, RegisterClass RC,
3182 Operand ImmOp, bits<4> MajOp, bit isHalf = 0 > {
3184 let BaseOpcode = "POST_"#BaseOp in {
3185 def S2_#NAME#_pi : T_store_pi <mnemonic, RC, ImmOp, MajOp, isHalf>;
3188 def S2_p#NAME#t_pi : T_pstore_pi <mnemonic, RC, ImmOp, MajOp, isHalf, 0, 0>;
3189 def S2_p#NAME#f_pi : T_pstore_pi <mnemonic, RC, ImmOp, MajOp, isHalf, 1, 0>;
3192 def S2_p#NAME#tnew_pi : T_pstore_pi <mnemonic, RC, ImmOp, MajOp,
3194 def S2_p#NAME#fnew_pi : T_pstore_pi <mnemonic, RC, ImmOp, MajOp,
3199 let accessSize = ByteAccess in
3200 defm storerb: ST_PostInc <"memb", "STrib", IntRegs, s4_0Imm, 0b1000>;
3202 let accessSize = HalfWordAccess in
3203 defm storerh: ST_PostInc <"memh", "STrih", IntRegs, s4_1Imm, 0b1010>;
3205 let accessSize = WordAccess in
3206 defm storeri: ST_PostInc <"memw", "STriw", IntRegs, s4_2Imm, 0b1100>;
3208 let accessSize = DoubleWordAccess in
3209 defm storerd: ST_PostInc <"memd", "STrid", DoubleRegs, s4_3Imm, 0b1110>;
3211 let accessSize = HalfWordAccess, isNVStorable = 0 in
3212 defm storerf: ST_PostInc <"memh", "STrih_H", IntRegs, s4_1Imm, 0b1011, 1>;
3214 class Storepi_pat<PatFrag Store, PatFrag Value, PatFrag Offset,
3216 : Pat<(Store Value:$src1, I32:$src2, Offset:$offset),
3217 (MI I32:$src2, imm:$offset, Value:$src1)>;
3219 def: Storepi_pat<post_truncsti8, I32, s4_0ImmPred, S2_storerb_pi>;
3220 def: Storepi_pat<post_truncsti16, I32, s4_1ImmPred, S2_storerh_pi>;
3221 def: Storepi_pat<post_store, I32, s4_2ImmPred, S2_storeri_pi>;
3222 def: Storepi_pat<post_store, I64, s4_3ImmPred, S2_storerd_pi>;
3224 //===----------------------------------------------------------------------===//
3225 // Template class for post increment stores with register offset.
3226 //===----------------------------------------------------------------------===//
3227 let isNVStorable = 1 in
3228 class T_store_pr <string mnemonic, RegisterClass RC, bits<3> MajOp,
3229 MemAccessSize AccessSz, bit isHalf = 0>
3230 : STInst <(outs IntRegs:$_dst_),
3231 (ins IntRegs:$src1, ModRegs:$src2, RC:$src3),
3232 mnemonic#"($src1++$src2) = $src3"#!if(isHalf, ".h", ""),
3233 [], "$src1 = $_dst_" > {
3237 let accessSize = AccessSz;
3239 let IClass = 0b1010;
3241 let Inst{27-24} = 0b1101;
3242 let Inst{23-21} = MajOp;
3243 let Inst{20-16} = src1;
3244 let Inst{13} = src2;
3245 let Inst{12-8} = src3;
3249 def S2_storerb_pr : T_store_pr<"memb", IntRegs, 0b000, ByteAccess>;
3250 def S2_storerh_pr : T_store_pr<"memh", IntRegs, 0b010, HalfWordAccess>;
3251 def S2_storeri_pr : T_store_pr<"memw", IntRegs, 0b100, WordAccess>;
3252 def S2_storerd_pr : T_store_pr<"memd", DoubleRegs, 0b110, DoubleWordAccess>;
3254 def S2_storerf_pr : T_store_pr<"memh", IntRegs, 0b011, HalfWordAccess, 1>;
3256 let opExtendable = 1, isExtentSigned = 1, isPredicable = 1 in
3257 class T_store_io <string mnemonic, RegisterClass RC, Operand ImmOp,
3258 bits<3>MajOp, bit isH = 0>
3260 (ins IntRegs:$src1, ImmOp:$src2, RC:$src3),
3261 mnemonic#"($src1+#$src2) = $src3"#!if(isH,".h","")>,
3262 AddrModeRel, ImmRegRel {
3264 bits<14> src2; // Actual address offset
3266 bits<11> offsetBits; // Represents offset encoding
3268 string ImmOpStr = !cast<string>(ImmOp);
3270 let opExtentBits = !if (!eq(ImmOpStr, "s11_3Ext"), 14,
3271 !if (!eq(ImmOpStr, "s11_2Ext"), 13,
3272 !if (!eq(ImmOpStr, "s11_1Ext"), 12,
3273 /* s11_0Ext */ 11)));
3274 let offsetBits = !if (!eq(ImmOpStr, "s11_3Ext"), src2{13-3},
3275 !if (!eq(ImmOpStr, "s11_2Ext"), src2{12-2},
3276 !if (!eq(ImmOpStr, "s11_1Ext"), src2{11-1},
3277 /* s11_0Ext */ src2{10-0})));
3278 let IClass = 0b1010;
3281 let Inst{26-25} = offsetBits{10-9};
3283 let Inst{23-21} = MajOp;
3284 let Inst{20-16} = src1;
3285 let Inst{13} = offsetBits{8};
3286 let Inst{12-8} = src3;
3287 let Inst{7-0} = offsetBits{7-0};
3290 let opExtendable = 2, isPredicated = 1 in
3291 class T_pstore_io <string mnemonic, RegisterClass RC, Operand ImmOp,
3292 bits<3>MajOp, bit PredNot, bit isPredNew, bit isH = 0>
3294 (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$src3, RC:$src4),
3295 !if(PredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
3296 ") ")#mnemonic#"($src2+#$src3) = $src4"#!if(isH,".h",""),
3297 [],"",V2LDST_tc_st_SLOT01 >,
3298 AddrModeRel, ImmRegRel {
3301 bits<9> src3; // Actual address offset
3303 bits<6> offsetBits; // Represents offset encoding
3305 let isPredicatedNew = isPredNew;
3306 let isPredicatedFalse = PredNot;
3308 string ImmOpStr = !cast<string>(ImmOp);
3309 let opExtentBits = !if (!eq(ImmOpStr, "u6_3Ext"), 9,
3310 !if (!eq(ImmOpStr, "u6_2Ext"), 8,
3311 !if (!eq(ImmOpStr, "u6_1Ext"), 7,
3313 let offsetBits = !if (!eq(ImmOpStr, "u6_3Ext"), src3{8-3},
3314 !if (!eq(ImmOpStr, "u6_2Ext"), src3{7-2},
3315 !if (!eq(ImmOpStr, "u6_1Ext"), src3{6-1},
3316 /* u6_0Ext */ src3{5-0})));
3317 let IClass = 0b0100;
3320 let Inst{26} = PredNot;
3321 let Inst{25} = isPredNew;
3323 let Inst{23-21} = MajOp;
3324 let Inst{20-16} = src2;
3325 let Inst{13} = offsetBits{5};
3326 let Inst{12-8} = src4;
3327 let Inst{7-3} = offsetBits{4-0};
3328 let Inst{1-0} = src1;
3331 let isExtendable = 1, isNVStorable = 1, hasSideEffects = 0 in
3332 multiclass ST_Idxd<string mnemonic, string CextOp, RegisterClass RC,
3333 Operand ImmOp, Operand predImmOp, bits<3> MajOp, bit isH = 0> {
3334 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
3335 def S2_#NAME#_io : T_store_io <mnemonic, RC, ImmOp, MajOp, isH>;
3338 def S2_p#NAME#t_io : T_pstore_io<mnemonic, RC, predImmOp, MajOp, 0, 0, isH>;
3339 def S2_p#NAME#f_io : T_pstore_io<mnemonic, RC, predImmOp, MajOp, 1, 0, isH>;
3342 def S4_p#NAME#tnew_io : T_pstore_io <mnemonic, RC, predImmOp,
3344 def S4_p#NAME#fnew_io : T_pstore_io <mnemonic, RC, predImmOp,
3349 let addrMode = BaseImmOffset, InputType = "imm" in {
3350 let accessSize = ByteAccess in
3351 defm storerb: ST_Idxd < "memb", "STrib", IntRegs, s11_0Ext, u6_0Ext, 0b000>;
3353 let accessSize = HalfWordAccess, opExtentAlign = 1 in
3354 defm storerh: ST_Idxd < "memh", "STrih", IntRegs, s11_1Ext, u6_1Ext, 0b010>;
3356 let accessSize = WordAccess, opExtentAlign = 2 in
3357 defm storeri: ST_Idxd < "memw", "STriw", IntRegs, s11_2Ext, u6_2Ext, 0b100>;
3359 let accessSize = DoubleWordAccess, isNVStorable = 0, opExtentAlign = 3 in
3360 defm storerd: ST_Idxd < "memd", "STrid", DoubleRegs, s11_3Ext,
3363 let accessSize = HalfWordAccess, opExtentAlign = 1 in
3364 defm storerf: ST_Idxd < "memh", "STrif", IntRegs, s11_1Ext,
3368 // Patterns for generating stores, where the address takes different forms:
3371 // - simple (base address without offset).
3372 // These would usually be used together (via Storex_pat defined below), but
3373 // in some cases one may want to apply different properties (such as
3374 // AddedComplexity) to the individual patterns.
3375 class Storex_fi_pat<PatFrag Store, PatFrag Value, InstHexagon MI>
3376 : Pat<(Store Value:$Rs, AddrFI:$fi), (MI AddrFI:$fi, 0, Value:$Rs)>;
3377 class Storex_add_pat<PatFrag Store, PatFrag Value, PatFrag ImmPred,
3379 : Pat<(Store Value:$Rt, (add (i32 IntRegs:$Rs), ImmPred:$Off)),
3380 (MI IntRegs:$Rs, imm:$Off, Value:$Rt)>;
3381 class Storex_simple_pat<PatFrag Store, PatFrag Value, InstHexagon MI>
3382 : Pat<(Store Value:$Rt, (i32 IntRegs:$Rs)),
3383 (MI IntRegs:$Rs, 0, Value:$Rt)>;
3385 // Patterns for generating stores, where the address takes different forms,
3386 // and where the value being stored is transformed through the value modifier
3387 // ValueMod. The address forms are same as above.
3388 class Storexm_fi_pat<PatFrag Store, PatFrag Value, PatFrag ValueMod,
3390 : Pat<(Store Value:$Rs, AddrFI:$fi),
3391 (MI AddrFI:$fi, 0, (ValueMod Value:$Rs))>;
3392 class Storexm_add_pat<PatFrag Store, PatFrag Value, PatFrag ImmPred,
3393 PatFrag ValueMod, InstHexagon MI>
3394 : Pat<(Store Value:$Rt, (add (i32 IntRegs:$Rs), ImmPred:$Off)),
3395 (MI IntRegs:$Rs, imm:$Off, (ValueMod Value:$Rt))>;
3396 class Storexm_simple_pat<PatFrag Store, PatFrag Value, PatFrag ValueMod,
3398 : Pat<(Store Value:$Rt, (i32 IntRegs:$Rs)),
3399 (MI IntRegs:$Rs, 0, (ValueMod Value:$Rt))>;
3401 multiclass Storex_pat<PatFrag Store, PatFrag Value, PatLeaf ImmPred,
3403 def: Storex_fi_pat <Store, Value, MI>;
3404 def: Storex_add_pat <Store, Value, ImmPred, MI>;
3407 multiclass Storexm_pat<PatFrag Store, PatFrag Value, PatLeaf ImmPred,
3408 PatFrag ValueMod, InstHexagon MI> {
3409 def: Storexm_fi_pat <Store, Value, ValueMod, MI>;
3410 def: Storexm_add_pat <Store, Value, ImmPred, ValueMod, MI>;
3413 // Regular stores in the DAG have two operands: value and address.
3414 // Atomic stores also have two, but they are reversed: address, value.
3415 // To use atomic stores with the patterns, they need to have their operands
3416 // swapped. This relies on the knowledge that the F.Fragment uses names
3418 class SwapSt<PatFrag F>
3419 : PatFrag<(ops node:$val, node:$ptr), F.Fragment>;
3421 let AddedComplexity = 20 in {
3422 defm: Storex_pat<truncstorei8, I32, s11_0ExtPred, S2_storerb_io>;
3423 defm: Storex_pat<truncstorei16, I32, s11_1ExtPred, S2_storerh_io>;
3424 defm: Storex_pat<store, I32, s11_2ExtPred, S2_storeri_io>;
3425 defm: Storex_pat<store, I64, s11_3ExtPred, S2_storerd_io>;
3427 defm: Storex_pat<SwapSt<atomic_store_8>, I32, s11_0ExtPred, S2_storerb_io>;
3428 defm: Storex_pat<SwapSt<atomic_store_16>, I32, s11_1ExtPred, S2_storerh_io>;
3429 defm: Storex_pat<SwapSt<atomic_store_32>, I32, s11_2ExtPred, S2_storeri_io>;
3430 defm: Storex_pat<SwapSt<atomic_store_64>, I64, s11_3ExtPred, S2_storerd_io>;
3433 // Simple patterns should be tried with the least priority.
3434 def: Storex_simple_pat<truncstorei8, I32, S2_storerb_io>;
3435 def: Storex_simple_pat<truncstorei16, I32, S2_storerh_io>;
3436 def: Storex_simple_pat<store, I32, S2_storeri_io>;
3437 def: Storex_simple_pat<store, I64, S2_storerd_io>;
3439 def: Storex_simple_pat<SwapSt<atomic_store_8>, I32, S2_storerb_io>;
3440 def: Storex_simple_pat<SwapSt<atomic_store_16>, I32, S2_storerh_io>;
3441 def: Storex_simple_pat<SwapSt<atomic_store_32>, I32, S2_storeri_io>;
3442 def: Storex_simple_pat<SwapSt<atomic_store_64>, I64, S2_storerd_io>;
3444 let AddedComplexity = 20 in {
3445 defm: Storexm_pat<truncstorei8, I64, s11_0ExtPred, LoReg, S2_storerb_io>;
3446 defm: Storexm_pat<truncstorei16, I64, s11_1ExtPred, LoReg, S2_storerh_io>;
3447 defm: Storexm_pat<truncstorei32, I64, s11_2ExtPred, LoReg, S2_storeri_io>;
3450 def: Storexm_simple_pat<truncstorei8, I64, LoReg, S2_storerb_io>;
3451 def: Storexm_simple_pat<truncstorei16, I64, LoReg, S2_storerh_io>;
3452 def: Storexm_simple_pat<truncstorei32, I64, LoReg, S2_storeri_io>;
3455 let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 13,
3456 isCodeGenOnly = 1, isPseudo = 1, hasSideEffects = 0 in
3457 def STriw_pred : STInst<(outs),
3458 (ins IntRegs:$addr, s11_2Ext:$off, PredRegs:$src1),
3459 ".error \"should not emit\"", []>;
3461 // S2_allocframe: Allocate stack frame.
3462 let Defs = [R29, R30], Uses = [R29, R31, R30],
3463 hasSideEffects = 0, accessSize = DoubleWordAccess in
3464 def S2_allocframe: ST0Inst <
3465 (outs), (ins u11_3Imm:$u11_3),
3466 "allocframe(#$u11_3)" > {
3469 let IClass = 0b1010;
3470 let Inst{27-16} = 0b000010011101;
3471 let Inst{13-11} = 0b000;
3472 let Inst{10-0} = u11_3{13-3};
3475 // S2_storer[bhwdf]_pci: Store byte/half/word/double.
3476 // S2_storer[bhwdf]_pci -> S2_storerbnew_pci
3477 let Uses = [CS], isNVStorable = 1 in
3478 class T_store_pci <string mnemonic, RegisterClass RC,
3479 Operand Imm, bits<4>MajOp,
3480 MemAccessSize AlignSize, string RegSrc = "Rt">
3481 : STInst <(outs IntRegs:$_dst_),
3482 (ins IntRegs:$Rz, Imm:$offset, ModRegs:$Mu, RC:$Rt),
3483 #mnemonic#"($Rz ++ #$offset:circ($Mu)) = $"#RegSrc#"",
3490 let accessSize = AlignSize;
3492 let IClass = 0b1010;
3493 let Inst{27-25} = 0b100;
3494 let Inst{24-21} = MajOp;
3495 let Inst{20-16} = Rz;
3497 let Inst{12-8} = Rt;
3500 !if (!eq(!cast<string>(AlignSize), "DoubleWordAccess"), offset{6-3},
3501 !if (!eq(!cast<string>(AlignSize), "WordAccess"), offset{5-2},
3502 !if (!eq(!cast<string>(AlignSize), "HalfWordAccess"), offset{4-1},
3503 /* ByteAccess */ offset{3-0})));
3507 def S2_storerb_pci : T_store_pci<"memb", IntRegs, s4_0Imm, 0b1000,
3509 def S2_storerh_pci : T_store_pci<"memh", IntRegs, s4_1Imm, 0b1010,
3511 def S2_storerf_pci : T_store_pci<"memh", IntRegs, s4_1Imm, 0b1011,
3512 HalfWordAccess, "Rt.h">;
3513 def S2_storeri_pci : T_store_pci<"memw", IntRegs, s4_2Imm, 0b1100,
3515 def S2_storerd_pci : T_store_pci<"memd", DoubleRegs, s4_3Imm, 0b1110,
3518 let Uses = [CS], isNewValue = 1, mayStore = 1, isNVStore = 1, opNewValue = 4 in
3519 class T_storenew_pci <string mnemonic, Operand Imm,
3520 bits<2>MajOp, MemAccessSize AlignSize>
3521 : NVInst < (outs IntRegs:$_dst_),
3522 (ins IntRegs:$Rz, Imm:$offset, ModRegs:$Mu, IntRegs:$Nt),
3523 #mnemonic#"($Rz ++ #$offset:circ($Mu)) = $Nt.new",
3531 let accessSize = AlignSize;
3533 let IClass = 0b1010;
3534 let Inst{27-21} = 0b1001101;
3535 let Inst{20-16} = Rz;
3537 let Inst{12-11} = MajOp;
3538 let Inst{10-8} = Nt;
3541 !if (!eq(!cast<string>(AlignSize), "WordAccess"), offset{5-2},
3542 !if (!eq(!cast<string>(AlignSize), "HalfWordAccess"), offset{4-1},
3543 /* ByteAccess */ offset{3-0}));
3547 def S2_storerbnew_pci : T_storenew_pci <"memb", s4_0Imm, 0b00, ByteAccess>;
3548 def S2_storerhnew_pci : T_storenew_pci <"memh", s4_1Imm, 0b01, HalfWordAccess>;
3549 def S2_storerinew_pci : T_storenew_pci <"memw", s4_2Imm, 0b10, WordAccess>;
3551 //===----------------------------------------------------------------------===//
3552 // Circular stores with auto-increment register
3553 //===----------------------------------------------------------------------===//
3554 let Uses = [CS], isNVStorable = 1 in
3555 class T_store_pcr <string mnemonic, RegisterClass RC, bits<4>MajOp,
3556 MemAccessSize AlignSize, string RegSrc = "Rt">
3557 : STInst <(outs IntRegs:$_dst_),
3558 (ins IntRegs:$Rz, ModRegs:$Mu, RC:$Rt),
3559 #mnemonic#"($Rz ++ I:circ($Mu)) = $"#RegSrc#"",
3566 let accessSize = AlignSize;
3568 let IClass = 0b1010;
3569 let Inst{27-25} = 0b100;
3570 let Inst{24-21} = MajOp;
3571 let Inst{20-16} = Rz;
3573 let Inst{12-8} = Rt;
3578 def S2_storerb_pcr : T_store_pcr<"memb", IntRegs, 0b1000, ByteAccess>;
3579 def S2_storerh_pcr : T_store_pcr<"memh", IntRegs, 0b1010, HalfWordAccess>;
3580 def S2_storeri_pcr : T_store_pcr<"memw", IntRegs, 0b1100, WordAccess>;
3581 def S2_storerd_pcr : T_store_pcr<"memd", DoubleRegs, 0b1110, DoubleWordAccess>;
3582 def S2_storerf_pcr : T_store_pcr<"memh", IntRegs, 0b1011,
3583 HalfWordAccess, "Rt.h">;
3585 //===----------------------------------------------------------------------===//
3586 // Circular .new stores with auto-increment register
3587 //===----------------------------------------------------------------------===//
3588 let Uses = [CS], isNewValue = 1, mayStore = 1, isNVStore = 1, opNewValue = 3 in
3589 class T_storenew_pcr <string mnemonic, bits<2>MajOp,
3590 MemAccessSize AlignSize>
3591 : NVInst <(outs IntRegs:$_dst_),
3592 (ins IntRegs:$Rz, ModRegs:$Mu, IntRegs:$Nt),
3593 #mnemonic#"($Rz ++ I:circ($Mu)) = $Nt.new" ,
3600 let accessSize = AlignSize;
3602 let IClass = 0b1010;
3603 let Inst{27-21} = 0b1001101;
3604 let Inst{20-16} = Rz;
3606 let Inst{12-11} = MajOp;
3607 let Inst{10-8} = Nt;
3612 def S2_storerbnew_pcr : T_storenew_pcr <"memb", 0b00, ByteAccess>;
3613 def S2_storerhnew_pcr : T_storenew_pcr <"memh", 0b01, HalfWordAccess>;
3614 def S2_storerinew_pcr : T_storenew_pcr <"memw", 0b10, WordAccess>;
3616 //===----------------------------------------------------------------------===//
3617 // Bit-reversed stores with auto-increment register
3618 //===----------------------------------------------------------------------===//
3619 let hasSideEffects = 0 in
3620 class T_store_pbr<string mnemonic, RegisterClass RC,
3621 MemAccessSize addrSize, bits<3> majOp,
3624 <(outs IntRegs:$_dst_),
3625 (ins IntRegs:$Rz, ModRegs:$Mu, RC:$src),
3626 #mnemonic#"($Rz ++ $Mu:brev) = $src"#!if (!eq(isHalf, 1), ".h", ""),
3627 [], "$Rz = $_dst_" > {
3629 let accessSize = addrSize;
3635 let IClass = 0b1010;
3637 let Inst{27-24} = 0b1111;
3638 let Inst{23-21} = majOp;
3640 let Inst{20-16} = Rz;
3642 let Inst{12-8} = src;
3645 let isNVStorable = 1 in {
3646 let BaseOpcode = "S2_storerb_pbr" in
3647 def S2_storerb_pbr : T_store_pbr<"memb", IntRegs, ByteAccess,
3648 0b000>, NewValueRel;
3649 let BaseOpcode = "S2_storerh_pbr" in
3650 def S2_storerh_pbr : T_store_pbr<"memh", IntRegs, HalfWordAccess,
3651 0b010>, NewValueRel;
3652 let BaseOpcode = "S2_storeri_pbr" in
3653 def S2_storeri_pbr : T_store_pbr<"memw", IntRegs, WordAccess,
3654 0b100>, NewValueRel;
3657 def S2_storerf_pbr : T_store_pbr<"memh", IntRegs, HalfWordAccess, 0b011, 1>;
3658 def S2_storerd_pbr : T_store_pbr<"memd", DoubleRegs, DoubleWordAccess, 0b110>;
3660 //===----------------------------------------------------------------------===//
3661 // Bit-reversed .new stores with auto-increment register
3662 //===----------------------------------------------------------------------===//
3663 let isNewValue = 1, mayStore = 1, isNVStore = 1, opNewValue = 3,
3664 hasSideEffects = 0 in
3665 class T_storenew_pbr<string mnemonic, MemAccessSize addrSize, bits<2> majOp>
3666 : NVInst <(outs IntRegs:$_dst_),
3667 (ins IntRegs:$Rz, ModRegs:$Mu, IntRegs:$Nt),
3668 #mnemonic#"($Rz ++ $Mu:brev) = $Nt.new", [],
3669 "$Rz = $_dst_">, NewValueRel {
3670 let accessSize = addrSize;
3675 let IClass = 0b1010;
3677 let Inst{27-21} = 0b1111101;
3678 let Inst{12-11} = majOp;
3680 let Inst{20-16} = Rz;
3682 let Inst{10-8} = Nt;
3685 let BaseOpcode = "S2_storerb_pbr" in
3686 def S2_storerbnew_pbr : T_storenew_pbr<"memb", ByteAccess, 0b00>;
3688 let BaseOpcode = "S2_storerh_pbr" in
3689 def S2_storerhnew_pbr : T_storenew_pbr<"memh", HalfWordAccess, 0b01>;
3691 let BaseOpcode = "S2_storeri_pbr" in
3692 def S2_storerinew_pbr : T_storenew_pbr<"memw", WordAccess, 0b10>;
3694 //===----------------------------------------------------------------------===//
3696 //===----------------------------------------------------------------------===//
3698 //===----------------------------------------------------------------------===//
3699 // Template class for S_2op instructions.
3700 //===----------------------------------------------------------------------===//
3701 let hasSideEffects = 0 in
3702 class T_S2op_1 <string mnemonic, bits<4> RegTyBits, RegisterClass RCOut,
3703 RegisterClass RCIn, bits<2> MajOp, bits<3> MinOp, bit isSat>
3704 : SInst <(outs RCOut:$dst), (ins RCIn:$src),
3705 "$dst = "#mnemonic#"($src)"#!if(isSat, ":sat", ""),
3706 [], "", S_2op_tc_1_SLOT23 > {
3710 let IClass = 0b1000;
3712 let Inst{27-24} = RegTyBits;
3713 let Inst{23-22} = MajOp;
3715 let Inst{20-16} = src;
3716 let Inst{7-5} = MinOp;
3717 let Inst{4-0} = dst;
3720 class T_S2op_1_di <string mnemonic, bits<2> MajOp, bits<3> MinOp>
3721 : T_S2op_1 <mnemonic, 0b0100, DoubleRegs, IntRegs, MajOp, MinOp, 0>;
3723 let hasNewValue = 1 in
3724 class T_S2op_1_id <string mnemonic, bits<2> MajOp, bits<3> MinOp, bit isSat = 0>
3725 : T_S2op_1 <mnemonic, 0b1000, IntRegs, DoubleRegs, MajOp, MinOp, isSat>;
3727 let hasNewValue = 1 in
3728 class T_S2op_1_ii <string mnemonic, bits<2> MajOp, bits<3> MinOp, bit isSat = 0>
3729 : T_S2op_1 <mnemonic, 0b1100, IntRegs, IntRegs, MajOp, MinOp, isSat>;
3731 // Vector sign/zero extend
3732 let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
3733 def S2_vsxtbh : T_S2op_1_di <"vsxtbh", 0b00, 0b000>;
3734 def S2_vsxthw : T_S2op_1_di <"vsxthw", 0b00, 0b100>;
3735 def S2_vzxtbh : T_S2op_1_di <"vzxtbh", 0b00, 0b010>;
3736 def S2_vzxthw : T_S2op_1_di <"vzxthw", 0b00, 0b110>;
3739 // Vector splat bytes/halfwords
3740 let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
3741 def S2_vsplatrb : T_S2op_1_ii <"vsplatb", 0b01, 0b111>;
3742 def S2_vsplatrh : T_S2op_1_di <"vsplath", 0b01, 0b010>;
3745 // Sign extend word to doubleword
3746 def A2_sxtw : T_S2op_1_di <"sxtw", 0b01, 0b000>;
3748 def: Pat <(i64 (sext I32:$src)), (A2_sxtw I32:$src)>;
3750 // Vector saturate and pack
3751 let Defs = [USR_OVF] in {
3752 def S2_svsathb : T_S2op_1_ii <"vsathb", 0b10, 0b000>;
3753 def S2_svsathub : T_S2op_1_ii <"vsathub", 0b10, 0b010>;
3754 def S2_vsathb : T_S2op_1_id <"vsathb", 0b00, 0b110>;
3755 def S2_vsathub : T_S2op_1_id <"vsathub", 0b00, 0b000>;
3756 def S2_vsatwh : T_S2op_1_id <"vsatwh", 0b00, 0b010>;
3757 def S2_vsatwuh : T_S2op_1_id <"vsatwuh", 0b00, 0b100>;
3761 def S2_vtrunohb : T_S2op_1_id <"vtrunohb", 0b10, 0b000>;
3762 def S2_vtrunehb : T_S2op_1_id <"vtrunehb", 0b10, 0b010>;
3764 // Swizzle the bytes of a word
3765 def A2_swiz : T_S2op_1_ii <"swiz", 0b10, 0b111>;
3768 let Defs = [USR_OVF] in {
3769 def A2_sat : T_S2op_1_id <"sat", 0b11, 0b000>;
3770 def A2_satb : T_S2op_1_ii <"satb", 0b11, 0b111>;
3771 def A2_satub : T_S2op_1_ii <"satub", 0b11, 0b110>;
3772 def A2_sath : T_S2op_1_ii <"sath", 0b11, 0b100>;
3773 def A2_satuh : T_S2op_1_ii <"satuh", 0b11, 0b101>;
3774 def A2_roundsat : T_S2op_1_id <"round", 0b11, 0b001, 0b1>;
3777 let Itinerary = S_2op_tc_2_SLOT23 in {
3778 // Vector round and pack
3779 def S2_vrndpackwh : T_S2op_1_id <"vrndwh", 0b10, 0b100>;
3781 let Defs = [USR_OVF] in
3782 def S2_vrndpackwhs : T_S2op_1_id <"vrndwh", 0b10, 0b110, 1>;
3785 def S2_brev : T_S2op_1_ii <"brev", 0b01, 0b110>;
3787 // Absolute value word
3788 def A2_abs : T_S2op_1_ii <"abs", 0b10, 0b100>;
3790 let Defs = [USR_OVF] in
3791 def A2_abssat : T_S2op_1_ii <"abs", 0b10, 0b101, 1>;
3793 // Negate with saturation
3794 let Defs = [USR_OVF] in
3795 def A2_negsat : T_S2op_1_ii <"neg", 0b10, 0b110, 1>;
3798 def: Pat<(i32 (select (i1 (setlt (i32 IntRegs:$src), 0)),
3799 (i32 (sub 0, (i32 IntRegs:$src))),
3800 (i32 IntRegs:$src))),
3801 (A2_abs IntRegs:$src)>;
3803 let AddedComplexity = 50 in
3804 def: Pat<(i32 (xor (add (sra (i32 IntRegs:$src), (i32 31)),
3805 (i32 IntRegs:$src)),
3806 (sra (i32 IntRegs:$src), (i32 31)))),
3807 (A2_abs IntRegs:$src)>;
3809 class T_S2op_2 <string mnemonic, bits<4> RegTyBits, RegisterClass RCOut,
3810 RegisterClass RCIn, bits<3> MajOp, bits<3> MinOp,
3811 bit isSat, bit isRnd, list<dag> pattern = []>
3812 : SInst <(outs RCOut:$dst),
3813 (ins RCIn:$src, u5Imm:$u5),
3814 "$dst = "#mnemonic#"($src, #$u5)"#!if(isSat, ":sat", "")
3815 #!if(isRnd, ":rnd", ""),
3816 pattern, "", S_2op_tc_2_SLOT23> {
3821 let IClass = 0b1000;
3823 let Inst{27-24} = RegTyBits;
3824 let Inst{23-21} = MajOp;
3825 let Inst{20-16} = src;
3827 let Inst{12-8} = u5;
3828 let Inst{7-5} = MinOp;
3829 let Inst{4-0} = dst;
3832 class T_S2op_2_di <string mnemonic, bits<3> MajOp, bits<3> MinOp>
3833 : T_S2op_2 <mnemonic, 0b1000, DoubleRegs, IntRegs, MajOp, MinOp, 0, 0>;
3835 let hasNewValue = 1 in
3836 class T_S2op_2_id <string mnemonic, bits<3> MajOp, bits<3> MinOp>
3837 : T_S2op_2 <mnemonic, 0b1000, IntRegs, DoubleRegs, MajOp, MinOp, 0, 0>;
3839 let hasNewValue = 1 in
3840 class T_S2op_2_ii <string mnemonic, bits<3> MajOp, bits<3> MinOp,
3841 bit isSat = 0, bit isRnd = 0, list<dag> pattern = []>
3842 : T_S2op_2 <mnemonic, 0b1100, IntRegs, IntRegs, MajOp, MinOp,
3843 isSat, isRnd, pattern>;
3845 class T_S2op_shift <string mnemonic, bits<3> MajOp, bits<3> MinOp, SDNode OpNd>
3846 : T_S2op_2_ii <mnemonic, MajOp, MinOp, 0, 0,
3847 [(set (i32 IntRegs:$dst), (OpNd (i32 IntRegs:$src),
3848 (u5ImmPred:$u5)))]>;
3850 // Vector arithmetic shift right by immediate with truncate and pack
3851 def S2_asr_i_svw_trun : T_S2op_2_id <"vasrw", 0b110, 0b010>;
3853 // Arithmetic/logical shift right/left by immediate
3854 let Itinerary = S_2op_tc_1_SLOT23 in {
3855 def S2_asr_i_r : T_S2op_shift <"asr", 0b000, 0b000, sra>;
3856 def S2_lsr_i_r : T_S2op_shift <"lsr", 0b000, 0b001, srl>;
3857 def S2_asl_i_r : T_S2op_shift <"asl", 0b000, 0b010, shl>;
3860 // Shift left by immediate with saturation
3861 let Defs = [USR_OVF] in
3862 def S2_asl_i_r_sat : T_S2op_2_ii <"asl", 0b010, 0b010, 1>;
3864 // Shift right with round
3865 def S2_asr_i_r_rnd : T_S2op_2_ii <"asr", 0b010, 0b000, 0, 1>;
3867 let isAsmParserOnly = 1 in
3868 def S2_asr_i_r_rnd_goodsyntax
3869 : SInst <(outs IntRegs:$dst), (ins IntRegs:$src, u5Imm:$u5),
3870 "$dst = asrrnd($src, #$u5)",
3871 [], "", S_2op_tc_1_SLOT23>;
3873 def: Pat<(i32 (sra (i32 (add (i32 (sra I32:$src1, u5ImmPred:$src2)),
3876 (S2_asr_i_r_rnd IntRegs:$src1, u5ImmPred:$src2)>;
3878 class T_S2op_3<string opc, bits<2>MajOp, bits<3>minOp, bits<1> sat = 0>
3879 : SInst<(outs DoubleRegs:$Rdd), (ins DoubleRegs:$Rss),
3880 "$Rdd = "#opc#"($Rss)"#!if(!eq(sat, 1),":sat","")> {
3883 let IClass = 0b1000;
3884 let Inst{27-24} = 0;
3885 let Inst{23-22} = MajOp;
3886 let Inst{20-16} = Rss;
3887 let Inst{7-5} = minOp;
3888 let Inst{4-0} = Rdd;
3891 def A2_absp : T_S2op_3 <"abs", 0b10, 0b110>;
3892 def A2_negp : T_S2op_3 <"neg", 0b10, 0b101>;
3893 def A2_notp : T_S2op_3 <"not", 0b10, 0b100>;
3895 // Innterleave/deinterleave
3896 def S2_interleave : T_S2op_3 <"interleave", 0b11, 0b101>;
3897 def S2_deinterleave : T_S2op_3 <"deinterleave", 0b11, 0b100>;
3899 // Vector Complex conjugate
3900 def A2_vconj : T_S2op_3 <"vconj", 0b10, 0b111, 1>;
3902 // Vector saturate without pack
3903 def S2_vsathb_nopack : T_S2op_3 <"vsathb", 0b00, 0b111>;
3904 def S2_vsathub_nopack : T_S2op_3 <"vsathub", 0b00, 0b100>;
3905 def S2_vsatwh_nopack : T_S2op_3 <"vsatwh", 0b00, 0b110>;
3906 def S2_vsatwuh_nopack : T_S2op_3 <"vsatwuh", 0b00, 0b101>;
3908 // Vector absolute value halfwords with and without saturation
3909 // Rdd64=vabsh(Rss64)[:sat]
3910 def A2_vabsh : T_S2op_3 <"vabsh", 0b01, 0b100>;
3911 def A2_vabshsat : T_S2op_3 <"vabsh", 0b01, 0b101, 1>;
3913 // Vector absolute value words with and without saturation
3914 def A2_vabsw : T_S2op_3 <"vabsw", 0b01, 0b110>;
3915 def A2_vabswsat : T_S2op_3 <"vabsw", 0b01, 0b111, 1>;
3917 //===----------------------------------------------------------------------===//
3919 //===----------------------------------------------------------------------===//
3922 let hasSideEffects = 0, hasNewValue = 1 in
3923 class T_COUNT_LEADING<string MnOp, bits<3> MajOp, bits<3> MinOp, bit Is32,
3925 : SInst<Out, Inp, "$Rd = "#MnOp#"($Rs)", [], "", S_2op_tc_1_SLOT23> {
3928 let IClass = 0b1000;
3930 let Inst{26} = Is32;
3931 let Inst{25-24} = 0b00;
3932 let Inst{23-21} = MajOp;
3933 let Inst{20-16} = Rs;
3934 let Inst{7-5} = MinOp;
3938 class T_COUNT_LEADING_32<string MnOp, bits<3> MajOp, bits<3> MinOp>
3939 : T_COUNT_LEADING<MnOp, MajOp, MinOp, 0b1,
3940 (outs IntRegs:$Rd), (ins IntRegs:$Rs)>;
3942 class T_COUNT_LEADING_64<string MnOp, bits<3> MajOp, bits<3> MinOp>
3943 : T_COUNT_LEADING<MnOp, MajOp, MinOp, 0b0,
3944 (outs IntRegs:$Rd), (ins DoubleRegs:$Rs)>;
3946 def S2_cl0 : T_COUNT_LEADING_32<"cl0", 0b000, 0b101>;
3947 def S2_cl1 : T_COUNT_LEADING_32<"cl1", 0b000, 0b110>;
3948 def S2_ct0 : T_COUNT_LEADING_32<"ct0", 0b010, 0b100>;
3949 def S2_ct1 : T_COUNT_LEADING_32<"ct1", 0b010, 0b101>;
3950 def S2_cl0p : T_COUNT_LEADING_64<"cl0", 0b010, 0b010>;
3951 def S2_cl1p : T_COUNT_LEADING_64<"cl1", 0b010, 0b100>;
3952 def S2_clb : T_COUNT_LEADING_32<"clb", 0b000, 0b100>;
3953 def S2_clbp : T_COUNT_LEADING_64<"clb", 0b010, 0b000>;
3954 def S2_clbnorm : T_COUNT_LEADING_32<"normamt", 0b000, 0b111>;
3956 def: Pat<(i32 (ctlz I32:$Rs)), (S2_cl0 I32:$Rs)>;
3957 def: Pat<(i32 (ctlz (not I32:$Rs))), (S2_cl1 I32:$Rs)>;
3958 def: Pat<(i32 (cttz I32:$Rs)), (S2_ct0 I32:$Rs)>;
3959 def: Pat<(i32 (cttz (not I32:$Rs))), (S2_ct1 I32:$Rs)>;
3960 def: Pat<(i32 (trunc (ctlz I64:$Rss))), (S2_cl0p I64:$Rss)>;
3961 def: Pat<(i32 (trunc (ctlz (not I64:$Rss)))), (S2_cl1p I64:$Rss)>;
3963 // Bit set/clear/toggle
3965 let hasSideEffects = 0, hasNewValue = 1 in
3966 class T_SCT_BIT_IMM<string MnOp, bits<3> MinOp>
3967 : SInst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, u5Imm:$u5),
3968 "$Rd = "#MnOp#"($Rs, #$u5)", [], "", S_2op_tc_1_SLOT23> {
3972 let IClass = 0b1000;
3973 let Inst{27-21} = 0b1100110;
3974 let Inst{20-16} = Rs;
3976 let Inst{12-8} = u5;
3977 let Inst{7-5} = MinOp;
3981 let hasSideEffects = 0, hasNewValue = 1 in
3982 class T_SCT_BIT_REG<string MnOp, bits<2> MinOp>
3983 : SInst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
3984 "$Rd = "#MnOp#"($Rs, $Rt)", [], "", S_3op_tc_1_SLOT23> {
3988 let IClass = 0b1100;
3989 let Inst{27-22} = 0b011010;
3990 let Inst{20-16} = Rs;
3991 let Inst{12-8} = Rt;
3992 let Inst{7-6} = MinOp;
3996 def S2_clrbit_i : T_SCT_BIT_IMM<"clrbit", 0b001>;
3997 def S2_setbit_i : T_SCT_BIT_IMM<"setbit", 0b000>;
3998 def S2_togglebit_i : T_SCT_BIT_IMM<"togglebit", 0b010>;
3999 def S2_clrbit_r : T_SCT_BIT_REG<"clrbit", 0b01>;
4000 def S2_setbit_r : T_SCT_BIT_REG<"setbit", 0b00>;
4001 def S2_togglebit_r : T_SCT_BIT_REG<"togglebit", 0b10>;
4003 def: Pat<(i32 (and (i32 IntRegs:$Rs), (not (shl 1, u5ImmPred:$u5)))),
4004 (S2_clrbit_i IntRegs:$Rs, u5ImmPred:$u5)>;
4005 def: Pat<(i32 (or (i32 IntRegs:$Rs), (shl 1, u5ImmPred:$u5))),
4006 (S2_setbit_i IntRegs:$Rs, u5ImmPred:$u5)>;
4007 def: Pat<(i32 (xor (i32 IntRegs:$Rs), (shl 1, u5ImmPred:$u5))),
4008 (S2_togglebit_i IntRegs:$Rs, u5ImmPred:$u5)>;
4009 def: Pat<(i32 (and (i32 IntRegs:$Rs), (not (shl 1, (i32 IntRegs:$Rt))))),
4010 (S2_clrbit_r IntRegs:$Rs, IntRegs:$Rt)>;
4011 def: Pat<(i32 (or (i32 IntRegs:$Rs), (shl 1, (i32 IntRegs:$Rt)))),
4012 (S2_setbit_r IntRegs:$Rs, IntRegs:$Rt)>;
4013 def: Pat<(i32 (xor (i32 IntRegs:$Rs), (shl 1, (i32 IntRegs:$Rt)))),
4014 (S2_togglebit_r IntRegs:$Rs, IntRegs:$Rt)>;
4018 let hasSideEffects = 0 in
4019 class T_TEST_BIT_IMM<string MnOp, bits<3> MajOp>
4020 : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, u5Imm:$u5),
4021 "$Pd = "#MnOp#"($Rs, #$u5)",
4022 [], "", S_2op_tc_2early_SLOT23> {
4026 let IClass = 0b1000;
4027 let Inst{27-24} = 0b0101;
4028 let Inst{23-21} = MajOp;
4029 let Inst{20-16} = Rs;
4031 let Inst{12-8} = u5;
4035 let hasSideEffects = 0 in
4036 class T_TEST_BIT_REG<string MnOp, bit IsNeg>
4037 : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt),
4038 "$Pd = "#MnOp#"($Rs, $Rt)",
4039 [], "", S_3op_tc_2early_SLOT23> {
4043 let IClass = 0b1100;
4044 let Inst{27-22} = 0b011100;
4045 let Inst{21} = IsNeg;
4046 let Inst{20-16} = Rs;
4047 let Inst{12-8} = Rt;
4051 def S2_tstbit_i : T_TEST_BIT_IMM<"tstbit", 0b000>;
4052 def S2_tstbit_r : T_TEST_BIT_REG<"tstbit", 0>;
4054 let AddedComplexity = 20 in { // Complexity greater than cmp reg-imm.
4055 def: Pat<(i1 (setne (and (shl 1, u5ImmPred:$u5), (i32 IntRegs:$Rs)), 0)),
4056 (S2_tstbit_i IntRegs:$Rs, u5ImmPred:$u5)>;
4057 def: Pat<(i1 (setne (and (shl 1, (i32 IntRegs:$Rt)), (i32 IntRegs:$Rs)), 0)),
4058 (S2_tstbit_r IntRegs:$Rs, IntRegs:$Rt)>;
4059 def: Pat<(i1 (trunc (i32 IntRegs:$Rs))),
4060 (S2_tstbit_i IntRegs:$Rs, 0)>;
4061 def: Pat<(i1 (trunc (i64 DoubleRegs:$Rs))),
4062 (S2_tstbit_i (LoReg DoubleRegs:$Rs), 0)>;
4065 let hasSideEffects = 0 in
4066 class T_TEST_BITS_IMM<string MnOp, bits<2> MajOp, bit IsNeg>
4067 : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, u6Imm:$u6),
4068 "$Pd = "#MnOp#"($Rs, #$u6)",
4069 [], "", S_2op_tc_2early_SLOT23> {
4073 let IClass = 0b1000;
4074 let Inst{27-24} = 0b0101;
4075 let Inst{23-22} = MajOp;
4076 let Inst{21} = IsNeg;
4077 let Inst{20-16} = Rs;
4078 let Inst{13-8} = u6;
4082 let hasSideEffects = 0 in
4083 class T_TEST_BITS_REG<string MnOp, bits<2> MajOp, bit IsNeg>
4084 : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt),
4085 "$Pd = "#MnOp#"($Rs, $Rt)",
4086 [], "", S_3op_tc_2early_SLOT23> {
4090 let IClass = 0b1100;
4091 let Inst{27-24} = 0b0111;
4092 let Inst{23-22} = MajOp;
4093 let Inst{21} = IsNeg;
4094 let Inst{20-16} = Rs;
4095 let Inst{12-8} = Rt;
4099 def C2_bitsclri : T_TEST_BITS_IMM<"bitsclr", 0b10, 0>;
4100 def C2_bitsclr : T_TEST_BITS_REG<"bitsclr", 0b10, 0>;
4101 def C2_bitsset : T_TEST_BITS_REG<"bitsset", 0b01, 0>;
4103 let AddedComplexity = 20 in { // Complexity greater than compare reg-imm.
4104 def: Pat<(i1 (seteq (and (i32 IntRegs:$Rs), u6ImmPred:$u6), 0)),
4105 (C2_bitsclri IntRegs:$Rs, u6ImmPred:$u6)>;
4106 def: Pat<(i1 (seteq (and (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), 0)),
4107 (C2_bitsclr IntRegs:$Rs, IntRegs:$Rt)>;
4110 let AddedComplexity = 10 in // Complexity greater than compare reg-reg.
4111 def: Pat<(i1 (seteq (and (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), IntRegs:$Rt)),
4112 (C2_bitsset IntRegs:$Rs, IntRegs:$Rt)>;
4114 //===----------------------------------------------------------------------===//
4116 //===----------------------------------------------------------------------===//
4118 //===----------------------------------------------------------------------===//
4120 //===----------------------------------------------------------------------===//
4121 //===----------------------------------------------------------------------===//
4123 //===----------------------------------------------------------------------===//
4125 //===----------------------------------------------------------------------===//
4127 //===----------------------------------------------------------------------===//
4129 //===----------------------------------------------------------------------===//
4131 //===----------------------------------------------------------------------===//
4133 //===----------------------------------------------------------------------===//
4135 //===----------------------------------------------------------------------===//
4137 // Predicate transfer.
4138 let hasSideEffects = 0, hasNewValue = 1 in
4139 def C2_tfrpr : SInst<(outs IntRegs:$Rd), (ins PredRegs:$Ps),
4140 "$Rd = $Ps", [], "", S_2op_tc_1_SLOT23> {
4144 let IClass = 0b1000;
4145 let Inst{27-24} = 0b1001;
4147 let Inst{17-16} = Ps;
4151 // Transfer general register to predicate.
4152 let hasSideEffects = 0 in
4153 def C2_tfrrp: SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs),
4154 "$Pd = $Rs", [], "", S_2op_tc_2early_SLOT23> {
4158 let IClass = 0b1000;
4159 let Inst{27-21} = 0b0101010;
4160 let Inst{20-16} = Rs;
4165 //===----------------------------------------------------------------------===//
4167 //===----------------------------------------------------------------------===//
4169 //===----------------------------------------------------------------------===//
4171 //===----------------------------------------------------------------------===//
4172 class S_2OpInstImm<string Mnemonic, bits<3>MajOp, bits<3>MinOp,
4173 Operand Imm, list<dag> pattern = [], bit isRnd = 0>
4174 : SInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, Imm:$src2),
4175 "$dst = "#Mnemonic#"($src1, #$src2)"#!if(isRnd, ":rnd", ""),
4179 let IClass = 0b1000;
4180 let Inst{27-24} = 0;
4181 let Inst{23-21} = MajOp;
4182 let Inst{20-16} = src1;
4183 let Inst{7-5} = MinOp;
4184 let Inst{4-0} = dst;
4187 class S_2OpInstImmI6<string Mnemonic, SDNode OpNode, bits<3>MinOp>
4188 : S_2OpInstImm<Mnemonic, 0b000, MinOp, u6Imm,
4189 [(set (i64 DoubleRegs:$dst), (OpNode (i64 DoubleRegs:$src1),
4190 u6ImmPred:$src2))]> {
4192 let Inst{13-8} = src2;
4195 // Shift by immediate.
4196 def S2_asr_i_p : S_2OpInstImmI6<"asr", sra, 0b000>;
4197 def S2_asl_i_p : S_2OpInstImmI6<"asl", shl, 0b010>;
4198 def S2_lsr_i_p : S_2OpInstImmI6<"lsr", srl, 0b001>;
4200 // Shift left by small amount and add.
4201 let AddedComplexity = 100, hasNewValue = 1, hasSideEffects = 0 in
4202 def S2_addasl_rrri: SInst <(outs IntRegs:$Rd),
4203 (ins IntRegs:$Rt, IntRegs:$Rs, u3Imm:$u3),
4204 "$Rd = addasl($Rt, $Rs, #$u3)" ,
4205 [(set (i32 IntRegs:$Rd), (add (i32 IntRegs:$Rt),
4206 (shl (i32 IntRegs:$Rs), u3ImmPred:$u3)))],
4207 "", S_3op_tc_2_SLOT23> {
4213 let IClass = 0b1100;
4215 let Inst{27-21} = 0b0100000;
4216 let Inst{20-16} = Rs;
4218 let Inst{12-8} = Rt;
4223 //===----------------------------------------------------------------------===//
4225 //===----------------------------------------------------------------------===//
4227 //===----------------------------------------------------------------------===//
4229 //===----------------------------------------------------------------------===//
4230 //===----------------------------------------------------------------------===//
4232 //===----------------------------------------------------------------------===//
4234 //===----------------------------------------------------------------------===//
4236 //===----------------------------------------------------------------------===//
4237 //===----------------------------------------------------------------------===//
4239 //===----------------------------------------------------------------------===//
4241 //===----------------------------------------------------------------------===//
4243 //===----------------------------------------------------------------------===//
4245 //===----------------------------------------------------------------------===//
4247 //===----------------------------------------------------------------------===//
4248 def HexagonBARRIER: SDNode<"HexagonISD::BARRIER", SDTNone, [SDNPHasChain]>;
4250 let hasSideEffects = 1, isSoloAX = 1 in
4251 def Y2_barrier : SYSInst<(outs), (ins),
4253 [(HexagonBARRIER)],"",ST_tc_st_SLOT0> {
4254 let Inst{31-28} = 0b1010;
4255 let Inst{27-21} = 0b1000000;
4258 //===----------------------------------------------------------------------===//
4260 //===----------------------------------------------------------------------===//
4261 //===----------------------------------------------------------------------===//
4263 //===----------------------------------------------------------------------===//
4265 let isExtendable = 1, isExtentSigned = 1, opExtentBits = 9, opExtentAlign = 2,
4266 opExtendable = 0, hasSideEffects = 0 in
4267 class LOOP_iBase<string mnemonic, Operand brOp, bit mustExtend = 0>
4268 : CRInst<(outs), (ins brOp:$offset, u10Imm:$src2),
4269 #mnemonic#"($offset, #$src2)",
4270 [], "" , CR_tc_3x_SLOT3> {
4274 let IClass = 0b0110;
4276 let Inst{27-22} = 0b100100;
4277 let Inst{21} = !if (!eq(mnemonic, "loop0"), 0b0, 0b1);
4278 let Inst{20-16} = src2{9-5};
4279 let Inst{12-8} = offset{8-4};
4280 let Inst{7-5} = src2{4-2};
4281 let Inst{4-3} = offset{3-2};
4282 let Inst{1-0} = src2{1-0};
4285 let isExtendable = 1, isExtentSigned = 1, opExtentBits = 9, opExtentAlign = 2,
4286 opExtendable = 0, hasSideEffects = 0 in
4287 class LOOP_rBase<string mnemonic, Operand brOp, bit mustExtend = 0>
4288 : CRInst<(outs), (ins brOp:$offset, IntRegs:$src2),
4289 #mnemonic#"($offset, $src2)",
4290 [], "" ,CR_tc_3x_SLOT3> {
4294 let IClass = 0b0110;
4296 let Inst{27-22} = 0b000000;
4297 let Inst{21} = !if (!eq(mnemonic, "loop0"), 0b0, 0b1);
4298 let Inst{20-16} = src2;
4299 let Inst{12-8} = offset{8-4};
4300 let Inst{4-3} = offset{3-2};
4303 multiclass LOOP_ri<string mnemonic> {
4304 def i : LOOP_iBase<mnemonic, brtarget>;
4305 def r : LOOP_rBase<mnemonic, brtarget>;
4309 let Defs = [SA0, LC0, USR] in
4310 defm J2_loop0 : LOOP_ri<"loop0">;
4312 // Interestingly only loop0's appear to set usr.lpcfg
4313 let Defs = [SA1, LC1] in
4314 defm J2_loop1 : LOOP_ri<"loop1">;
4316 let isBranch = 1, isTerminator = 1, hasSideEffects = 0,
4317 Defs = [PC, LC0], Uses = [SA0, LC0] in {
4318 def ENDLOOP0 : Endloop<(outs), (ins brtarget:$offset),
4323 let isBranch = 1, isTerminator = 1, hasSideEffects = 0,
4324 Defs = [PC, LC1], Uses = [SA1, LC1] in {
4325 def ENDLOOP1 : Endloop<(outs), (ins brtarget:$offset),
4330 // Pipelined loop instructions, sp[123]loop0
4331 let Defs = [LC0, SA0, P3, USR], hasSideEffects = 0,
4332 isExtentSigned = 1, isExtendable = 1, opExtentBits = 9, opExtentAlign = 2,
4333 opExtendable = 0, isPredicateLate = 1 in
4334 class SPLOOP_iBase<string SP, bits<2> op>
4335 : CRInst <(outs), (ins brtarget:$r7_2, u10Imm:$U10),
4336 "p3 = sp"#SP#"loop0($r7_2, #$U10)" > {
4340 let IClass = 0b0110;
4342 let Inst{22-21} = op;
4343 let Inst{27-23} = 0b10011;
4344 let Inst{20-16} = U10{9-5};
4345 let Inst{12-8} = r7_2{8-4};
4346 let Inst{7-5} = U10{4-2};
4347 let Inst{4-3} = r7_2{3-2};
4348 let Inst{1-0} = U10{1-0};
4351 let Defs = [LC0, SA0, P3, USR], hasSideEffects = 0,
4352 isExtentSigned = 1, isExtendable = 1, opExtentBits = 9, opExtentAlign = 2,
4353 opExtendable = 0, isPredicateLate = 1 in
4354 class SPLOOP_rBase<string SP, bits<2> op>
4355 : CRInst <(outs), (ins brtarget:$r7_2, IntRegs:$Rs),
4356 "p3 = sp"#SP#"loop0($r7_2, $Rs)" > {
4360 let IClass = 0b0110;
4362 let Inst{22-21} = op;
4363 let Inst{27-23} = 0b00001;
4364 let Inst{20-16} = Rs;
4365 let Inst{12-8} = r7_2{8-4};
4366 let Inst{4-3} = r7_2{3-2};
4369 multiclass SPLOOP_ri<string mnemonic, bits<2> op> {
4370 def i : SPLOOP_iBase<mnemonic, op>;
4371 def r : SPLOOP_rBase<mnemonic, op>;
4374 defm J2_ploop1s : SPLOOP_ri<"1", 0b01>;
4375 defm J2_ploop2s : SPLOOP_ri<"2", 0b10>;
4376 defm J2_ploop3s : SPLOOP_ri<"3", 0b11>;
4378 // if (Rs[!>=<]=#0) jump:[t/nt]
4379 let Defs = [PC], isPredicated = 1, isBranch = 1, hasSideEffects = 0,
4380 hasSideEffects = 0 in
4381 class J2_jump_0_Base<string compare, bit isTak, bits<2> op>
4382 : CRInst <(outs), (ins IntRegs:$Rs, brtarget:$r13_2),
4383 "if ($Rs"#compare#"#0) jump"#!if(isTak, ":t", ":nt")#" $r13_2" > {
4387 let IClass = 0b0110;
4389 let Inst{27-24} = 0b0001;
4390 let Inst{23-22} = op;
4391 let Inst{12} = isTak;
4392 let Inst{21} = r13_2{14};
4393 let Inst{20-16} = Rs;
4394 let Inst{11-1} = r13_2{12-2};
4395 let Inst{13} = r13_2{13};
4398 multiclass J2_jump_compare_0<string compare, bits<2> op> {
4399 def NAME : J2_jump_0_Base<compare, 0, op>;
4400 def NAME#pt : J2_jump_0_Base<compare, 1, op>;
4403 defm J2_jumprz : J2_jump_compare_0<"!=", 0b00>;
4404 defm J2_jumprgtez : J2_jump_compare_0<">=", 0b01>;
4405 defm J2_jumprnz : J2_jump_compare_0<"==", 0b10>;
4406 defm J2_jumprltez : J2_jump_compare_0<"<=", 0b11>;
4408 // Transfer to/from Control/GPR Guest/GPR
4409 let hasSideEffects = 0 in
4410 class TFR_CR_RS_base<RegisterClass CTRC, RegisterClass RC, bit isDouble>
4411 : CRInst <(outs CTRC:$dst), (ins RC:$src),
4412 "$dst = $src", [], "", CR_tc_3x_SLOT3> {
4416 let IClass = 0b0110;
4418 let Inst{27-25} = 0b001;
4419 let Inst{24} = isDouble;
4420 let Inst{23-21} = 0b001;
4421 let Inst{20-16} = src;
4422 let Inst{4-0} = dst;
4425 def A2_tfrrcr : TFR_CR_RS_base<CtrRegs, IntRegs, 0b0>;
4426 def : InstAlias<"m0 = $Rs", (A2_tfrrcr C6, IntRegs:$Rs)>;
4427 def : InstAlias<"m1 = $Rs", (A2_tfrrcr C7, IntRegs:$Rs)>;
4429 let hasSideEffects = 0 in
4430 class TFR_RD_CR_base<RegisterClass RC, RegisterClass CTRC, bit isSingle>
4431 : CRInst <(outs RC:$dst), (ins CTRC:$src),
4432 "$dst = $src", [], "", CR_tc_3x_SLOT3> {
4436 let IClass = 0b0110;
4438 let Inst{27-26} = 0b10;
4439 let Inst{25} = isSingle;
4440 let Inst{24-21} = 0b0000;
4441 let Inst{20-16} = src;
4442 let Inst{4-0} = dst;
4445 let hasNewValue = 1, opNewValue = 0 in
4446 def A2_tfrcrr : TFR_RD_CR_base<IntRegs, CtrRegs, 1>;
4447 def : InstAlias<"$Rd = m0", (A2_tfrcrr IntRegs:$Rd, C6)>;
4448 def : InstAlias<"$Rd = m1", (A2_tfrcrr IntRegs:$Rd, C7)>;
4450 // Y4_trace: Send value to etm trace.
4451 let isSoloAX = 1, hasSideEffects = 0 in
4452 def Y4_trace: CRInst <(outs), (ins IntRegs:$Rs),
4456 let IClass = 0b0110;
4457 let Inst{27-21} = 0b0010010;
4458 let Inst{20-16} = Rs;
4461 let AddedComplexity = 100, isPredicated = 1, isCodeGenOnly = 1 in
4462 def TFR_condset_ri : ALU32_rr<(outs IntRegs:$dst),
4463 (ins PredRegs:$src1, IntRegs:$src2, s12Imm:$src3),
4464 "Error; should not emit",
4465 [(set (i32 IntRegs:$dst),
4466 (i32 (select (i1 PredRegs:$src1), (i32 IntRegs:$src2),
4467 s12ImmPred:$src3)))]>;
4469 let AddedComplexity = 100, isPredicated = 1, isCodeGenOnly = 1 in
4470 def TFR_condset_ir : ALU32_rr<(outs IntRegs:$dst),
4471 (ins PredRegs:$src1, s12Imm:$src2, IntRegs:$src3),
4472 "Error; should not emit",
4473 [(set (i32 IntRegs:$dst),
4474 (i32 (select (i1 PredRegs:$src1), s12ImmPred:$src2,
4475 (i32 IntRegs:$src3))))]>;
4477 let AddedComplexity = 100, isPredicated = 1, isCodeGenOnly = 1 in
4478 def TFR_condset_ii : ALU32_rr<(outs IntRegs:$dst),
4479 (ins PredRegs:$src1, s12Imm:$src2, s12Imm:$src3),
4480 "Error; should not emit",
4481 [(set (i32 IntRegs:$dst),
4482 (i32 (select (i1 PredRegs:$src1), s12ImmPred:$src2,
4483 s12ImmPred:$src3)))]>;
4485 // Generate frameindex addresses.
4486 let isReMaterializable = 1, isCodeGenOnly = 1 in
4487 def TFR_FI : ALU32_ri<(outs IntRegs:$dst), (ins FrameIndex:$src1),
4488 "$dst = add($src1)",
4489 [(set (i32 IntRegs:$dst), ADDRri:$src1)]>;
4491 // Support for generating global address.
4492 // Taken from X86InstrInfo.td.
4493 def SDTHexagonCONST32 : SDTypeProfile<1, 1, [SDTCisVT<0, i32>,
4496 def HexagonCONST32 : SDNode<"HexagonISD::CONST32", SDTHexagonCONST32>;
4497 def HexagonCONST32_GP : SDNode<"HexagonISD::CONST32_GP", SDTHexagonCONST32>;
4499 // HI/LO Instructions
4500 let isReMaterializable = 1, isMoveImm = 1, hasSideEffects = 0,
4501 hasNewValue = 1, opNewValue = 0 in
4502 class REG_IMMED<string RegHalf, string Op, bit Rs, bits<3> MajOp, bit MinOp>
4503 : ALU32_ri<(outs IntRegs:$dst),
4504 (ins i32imm:$imm_value),
4505 "$dst"#RegHalf#" = #"#Op#"($imm_value)", []> {
4508 let IClass = 0b0111;
4511 let Inst{26-24} = MajOp;
4512 let Inst{21} = MinOp;
4513 let Inst{20-16} = dst;
4514 let Inst{23-22} = !if (!eq(Op, "LO"), imm_value{15-14}, imm_value{31-30});
4515 let Inst{13-0} = !if (!eq(Op, "LO"), imm_value{13-0}, imm_value{29-16});
4518 let isAsmParserOnly = 1 in {
4519 def LO : REG_IMMED<".l", "LO", 0b0, 0b001, 0b1>;
4520 def LO_H : REG_IMMED<".l", "HI", 0b0, 0b001, 0b1>;
4521 def HI : REG_IMMED<".h", "HI", 0b0, 0b010, 0b1>;
4522 def HI_L : REG_IMMED<".h", "LO", 0b0, 0b010, 0b1>;
4525 let isMoveImm = 1, isCodeGenOnly = 1 in
4526 def LO_PIC : ALU32_ri<(outs IntRegs:$dst), (ins bblabel:$label),
4527 "$dst.l = #LO($label@GOTREL)",
4530 let isMoveImm = 1, isCodeGenOnly = 1 in
4531 def HI_PIC : ALU32_ri<(outs IntRegs:$dst), (ins bblabel:$label),
4532 "$dst.h = #HI($label@GOTREL)",
4535 let isReMaterializable = 1, isMoveImm = 1, hasSideEffects = 0,
4536 isAsmParserOnly = 1 in
4537 def LOi : ALU32_ri<(outs IntRegs:$dst), (ins i32imm:$imm_value),
4538 "$dst.l = #LO($imm_value)",
4542 let isReMaterializable = 1, isMoveImm = 1, hasSideEffects = 0,
4543 isAsmParserOnly = 1 in
4544 def HIi : ALU32_ri<(outs IntRegs:$dst), (ins i32imm:$imm_value),
4545 "$dst.h = #HI($imm_value)",
4548 let isReMaterializable = 1, isMoveImm = 1, hasSideEffects = 0,
4549 isAsmParserOnly = 1 in
4550 def LO_jt : ALU32_ri<(outs IntRegs:$dst), (ins jumptablebase:$jt),
4551 "$dst.l = #LO($jt)",
4554 let isReMaterializable = 1, isMoveImm = 1, hasSideEffects = 0,
4555 isAsmParserOnly = 1 in
4556 def HI_jt : ALU32_ri<(outs IntRegs:$dst), (ins jumptablebase:$jt),
4557 "$dst.h = #HI($jt)",
4560 // This pattern is incorrect. When we add small data, we should change
4561 // this pattern to use memw(#foo).
4562 // This is for sdata.
4563 let isMoveImm = 1, isAsmParserOnly = 1 in
4564 def CONST32 : CONSTLDInst<(outs IntRegs:$dst), (ins globaladdress:$global),
4565 "$dst = CONST32(#$global)",
4566 [(set (i32 IntRegs:$dst),
4567 (load (HexagonCONST32 tglobaltlsaddr:$global)))]>;
4569 let isReMaterializable = 1, isMoveImm = 1 in
4570 def CONST32_set : LDInst2<(outs IntRegs:$dst), (ins globaladdress:$global),
4571 "$dst = CONST32(#$global)",
4572 [(set (i32 IntRegs:$dst),
4573 (HexagonCONST32 tglobaladdr:$global))]>;
4575 let isReMaterializable = 1, isMoveImm = 1, isAsmParserOnly = 1 in
4576 def CONST32_set_jt : CONSTLDInst<(outs IntRegs:$dst), (ins jumptablebase:$jt),
4577 "$dst = CONST32(#$jt)",
4578 [(set (i32 IntRegs:$dst),
4579 (HexagonCONST32 tjumptable:$jt))]>;
4581 let isReMaterializable = 1, isMoveImm = 1, isAsmParserOnly = 1 in
4582 def CONST32GP_set : LDInst2<(outs IntRegs:$dst), (ins globaladdress:$global),
4583 "$dst = CONST32(#$global)",
4584 [(set (i32 IntRegs:$dst),
4585 (HexagonCONST32_GP tglobaladdr:$global))]>;
4587 let isReMaterializable = 1, isMoveImm = 1, isAsmParserOnly = 1 in
4588 def CONST32_Int_Real : CONSTLDInst<(outs IntRegs:$dst), (ins i32imm:$global),
4589 "$dst = CONST32(#$global)",
4590 [(set (i32 IntRegs:$dst), imm:$global) ]>;
4592 // Map BlockAddress lowering to CONST32_Int_Real
4593 def : Pat<(HexagonCONST32_GP tblockaddress:$addr),
4594 (CONST32_Int_Real tblockaddress:$addr)>;
4596 let isReMaterializable = 1, isMoveImm = 1, isAsmParserOnly = 1 in
4597 def CONST32_Label : LDInst2<(outs IntRegs:$dst), (ins bblabel:$label),
4598 "$dst = CONST32($label)",
4599 [(set (i32 IntRegs:$dst), (HexagonCONST32 bbl:$label))]>;
4601 let isReMaterializable = 1, isMoveImm = 1, isAsmParserOnly = 1 in
4602 def CONST64_Int_Real : CONSTLDInst<(outs DoubleRegs:$dst), (ins i64imm:$global),
4603 "$dst = CONST64(#$global)",
4604 [(set (i64 DoubleRegs:$dst), imm:$global)]>;
4606 let isCodeGenOnly = 1 in
4607 def TFR_PdFalse : SInst<(outs PredRegs:$dst), (ins),
4608 "$dst = xor($dst, $dst)",
4609 [(set (i1 PredRegs:$dst), 0)]>;
4611 // Pseudo instructions.
4612 def SDT_SPCallSeqStart : SDCallSeqStart<[ SDTCisVT<0, i32> ]>;
4613 def SDT_SPCallSeqEnd : SDCallSeqEnd<[ SDTCisVT<0, i32>,
4614 SDTCisVT<1, i32> ]>;
4616 def callseq_start : SDNode<"ISD::CALLSEQ_START", SDT_SPCallSeqStart,
4617 [SDNPHasChain, SDNPOutGlue]>;
4618 def callseq_end : SDNode<"ISD::CALLSEQ_END", SDT_SPCallSeqEnd,
4619 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
4621 def SDT_SPCall : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
4623 // For tailcalls a HexagonTCRet SDNode has 3 SDNode Properties - a chain,
4624 // Optional Flag and Variable Arguments.
4625 // Its 1 Operand has pointer type.
4626 def HexagonTCRet : SDNode<"HexagonISD::TC_RETURN", SDT_SPCall,
4627 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
4629 let Defs = [R29, R30], Uses = [R31, R30, R29], isPseudo = 1 in
4630 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
4631 ".error \"should not emit\" ",
4632 [(callseq_start timm:$amt)]>;
4634 let Defs = [R29, R30, R31], Uses = [R29], isPseudo = 1 in
4635 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
4636 ".error \"should not emit\" ",
4637 [(callseq_end timm:$amt1, timm:$amt2)]>;
4639 // Call subroutine indirectly.
4640 let Defs = VolatileV3.Regs in
4641 def J2_callr : JUMPR_MISC_CALLR<0, 1>;
4643 // Indirect tail-call.
4644 let isPseudo = 1, isCall = 1, isReturn = 1, isBarrier = 1, isPredicable = 0,
4645 isTerminator = 1, isCodeGenOnly = 1 in
4646 def TCRETURNr : T_JMPr;
4648 // Direct tail-calls.
4649 let isCall = 1, isReturn = 1, isBarrier = 1, isPredicable = 0,
4650 isTerminator = 1, isCodeGenOnly = 1 in {
4651 def TCRETURNtg : JInst<(outs), (ins calltarget:$dst), "jump $dst",
4652 [], "", J_tc_2early_SLOT23>;
4653 def TCRETURNtext : JInst<(outs), (ins calltarget:$dst), "jump $dst",
4654 [], "", J_tc_2early_SLOT23>;
4658 def: Pat<(HexagonTCRet tglobaladdr:$dst),
4659 (TCRETURNtg tglobaladdr:$dst)>;
4660 def: Pat<(HexagonTCRet texternalsym:$dst),
4661 (TCRETURNtext texternalsym:$dst)>;
4662 def: Pat<(HexagonTCRet (i32 IntRegs:$dst)),
4663 (TCRETURNr (i32 IntRegs:$dst))>;
4665 // Map from r0 = and(r1, 65535) to r0 = zxth(r1)
4666 def: Pat<(and (i32 IntRegs:$src1), 65535),
4667 (A2_zxth IntRegs:$src1)>;
4669 // Map from r0 = and(r1, 255) to r0 = zxtb(r1).
4670 def: Pat<(and (i32 IntRegs:$src1), 255),
4671 (A2_zxtb IntRegs:$src1)>;
4673 // Map Add(p1, true) to p1 = not(p1).
4674 // Add(p1, false) should never be produced,
4675 // if it does, it got to be mapped to NOOP.
4676 def: Pat<(add (i1 PredRegs:$src1), -1),
4677 (C2_not PredRegs:$src1)>;
4679 // Map from p0 = pnot(p0); r0 = mux(p0, #i, #j) => r0 = mux(p0, #j, #i).
4680 def: Pat<(select (not (i1 PredRegs:$src1)), s8ImmPred:$src2, s8ExtPred:$src3),
4681 (C2_muxii PredRegs:$src1, s8ExtPred:$src3, s8ImmPred:$src2)>;
4683 // Map from p0 = pnot(p0); r0 = select(p0, #i, r1)
4684 // => r0 = C2_muxir(p0, r1, #i)
4685 def: Pat<(select (not (i1 PredRegs:$src1)), s8ExtPred:$src2,
4686 (i32 IntRegs:$src3)),
4687 (C2_muxir PredRegs:$src1, IntRegs:$src3, s8ExtPred:$src2)>;
4689 // Map from p0 = pnot(p0); r0 = mux(p0, r1, #i)
4690 // => r0 = C2_muxri (p0, #i, r1)
4691 def: Pat<(select (not (i1 PredRegs:$src1)), IntRegs:$src2, s8ExtPred:$src3),
4692 (C2_muxri PredRegs:$src1, s8ExtPred:$src3, IntRegs:$src2)>;
4694 // Map from p0 = pnot(p0); if (p0) jump => if (!p0) jump.
4695 def: Pat<(brcond (not (i1 PredRegs:$src1)), bb:$offset),
4696 (J2_jumpf PredRegs:$src1, bb:$offset)>;
4698 // Map from Rdd = sign_extend_inreg(Rss, i32) -> Rdd = A2_sxtw(Rss.lo).
4699 def: Pat<(i64 (sext_inreg (i64 DoubleRegs:$src1), i32)),
4700 (A2_sxtw (LoReg DoubleRegs:$src1))>;
4702 // Map from Rdd = sign_extend_inreg(Rss, i16) -> Rdd = A2_sxtw(A2_sxth(Rss.lo)).
4703 def: Pat<(i64 (sext_inreg (i64 DoubleRegs:$src1), i16)),
4704 (A2_sxtw (A2_sxth (LoReg DoubleRegs:$src1)))>;
4706 // Map from Rdd = sign_extend_inreg(Rss, i8) -> Rdd = A2_sxtw(A2_sxtb(Rss.lo)).
4707 def: Pat<(i64 (sext_inreg (i64 DoubleRegs:$src1), i8)),
4708 (A2_sxtw (A2_sxtb (LoReg DoubleRegs:$src1)))>;
4710 // We want to prevent emitting pnot's as much as possible.
4711 // Map brcond with an unsupported setcc to a J2_jumpf.
4712 def : Pat <(brcond (i1 (setne (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4714 (J2_jumpf (C2_cmpeq (i32 IntRegs:$src1), (i32 IntRegs:$src2)),
4717 def : Pat <(brcond (i1 (setne (i32 IntRegs:$src1), s10ImmPred:$src2)),
4719 (J2_jumpf (C2_cmpeqi (i32 IntRegs:$src1), s10ImmPred:$src2), bb:$offset)>;
4721 def: Pat<(brcond (i1 (setne (i1 PredRegs:$src1), (i1 -1))), bb:$offset),
4722 (J2_jumpf PredRegs:$src1, bb:$offset)>;
4724 def: Pat<(brcond (i1 (setne (i1 PredRegs:$src1), (i1 0))), bb:$offset),
4725 (J2_jumpt PredRegs:$src1, bb:$offset)>;
4727 // cmp.lt(Rs, Imm) -> !cmp.ge(Rs, Imm) -> !cmp.gt(Rs, Imm-1)
4728 def: Pat<(brcond (i1 (setlt (i32 IntRegs:$src1), s8ImmPred:$src2)), bb:$offset),
4729 (J2_jumpf (C2_cmpgti IntRegs:$src1, (DEC_CONST_SIGNED s8ImmPred:$src2)),
4732 // cmp.lt(r0, r1) -> cmp.gt(r1, r0)
4733 def : Pat <(brcond (i1 (setlt (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4735 (J2_jumpt (C2_cmpgt (i32 IntRegs:$src2), (i32 IntRegs:$src1)), bb:$offset)>;
4737 def : Pat <(brcond (i1 (setuge (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4739 (J2_jumpf (C2_cmpgtup (i64 DoubleRegs:$src2), (i64 DoubleRegs:$src1)),
4742 def : Pat <(brcond (i1 (setule (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4744 (J2_jumpf (C2_cmpgtu (i32 IntRegs:$src1), (i32 IntRegs:$src2)),
4747 def : Pat <(brcond (i1 (setule (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4749 (J2_jumpf (C2_cmpgtup (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2)),
4752 // Map from a 64-bit select to an emulated 64-bit mux.
4753 // Hexagon does not support 64-bit MUXes; so emulate with combines.
4754 def: Pat<(select (i1 PredRegs:$src1), (i64 DoubleRegs:$src2),
4755 (i64 DoubleRegs:$src3)),
4756 (A2_combinew (C2_mux PredRegs:$src1, (HiReg DoubleRegs:$src2),
4757 (HiReg DoubleRegs:$src3)),
4758 (C2_mux PredRegs:$src1, (LoReg DoubleRegs:$src2),
4759 (LoReg DoubleRegs:$src3)))>;
4761 // Map from a 1-bit select to logical ops.
4762 // From LegalizeDAG.cpp: (B1 ? B2 : B3) <=> (B1 & B2)|(!B1&B3).
4763 def: Pat<(select (i1 PredRegs:$src1), (i1 PredRegs:$src2), (i1 PredRegs:$src3)),
4764 (C2_or (C2_and PredRegs:$src1, PredRegs:$src2),
4765 (C2_and (C2_not PredRegs:$src1), PredRegs:$src3))>;
4767 // Map Pd = load(addr) -> Rs = load(addr); Pd = Rs.
4768 def : Pat<(i1 (load ADDRriS11_2:$addr)),
4769 (i1 (C2_tfrrp (i32 (L2_loadrb_io AddrFI:$addr, 0))))>;
4771 // Map for truncating from 64 immediates to 32 bit immediates.
4772 def: Pat<(i32 (trunc (i64 DoubleRegs:$src))),
4773 (LoReg DoubleRegs:$src)>;
4775 // Map for truncating from i64 immediates to i1 bit immediates.
4776 def: Pat<(i1 (trunc (i64 DoubleRegs:$src))),
4777 (C2_tfrrp (LoReg DoubleRegs:$src))>;
4779 // Map memb(Rs) = Rdd -> memb(Rs) = Rt.
4780 def : Pat<(truncstorei8 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
4781 (S2_storerb_io AddrFI:$addr, 0, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
4784 // Map memh(Rs) = Rdd -> memh(Rs) = Rt.
4785 def : Pat<(truncstorei16 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
4786 (S2_storerh_io AddrFI:$addr, 0, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
4788 // Map memw(Rs) = Rdd -> memw(Rs) = Rt
4789 def : Pat<(truncstorei32 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
4790 (S2_storeri_io AddrFI:$addr, 0, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
4793 // Map memw(Rs) = Rdd -> memw(Rs) = Rt.
4794 def : Pat<(truncstorei32 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
4795 (S2_storeri_io AddrFI:$addr, 0, (i32 (EXTRACT_SUBREG (i64 DoubleRegs:$src),
4798 // Map from i1 = constant<-1>; memw(addr) = i1 -> r0 = 1; memw(addr) = r0.
4799 def : Pat<(store (i1 -1), ADDRriS11_2:$addr),
4800 (S2_storerb_io AddrFI:$addr, 0, (A2_tfrsi 1))>;
4803 // Map from i1 = constant<-1>; store i1 -> r0 = 1; store r0.
4804 def : Pat<(store (i1 -1), ADDRriS11_2:$addr),
4805 (S2_storerb_io AddrFI:$addr, 0, (A2_tfrsi 1))>;
4807 // Map from memb(Rs) = Pd -> Rt = mux(Pd, #0, #1); store Rt.
4808 def : Pat<(store (i1 PredRegs:$src1), ADDRriS11_2:$addr),
4809 (S2_storerb_io AddrFI:$addr, 0, (i32 (C2_muxii (i1 PredRegs:$src1), 1, 0)) )>;
4811 // rs <= rt -> !(rs > rt).
4812 let AddedComplexity = 30 in
4813 def: Pat<(i1 (setle (i32 IntRegs:$src1), s10ExtPred:$src2)),
4814 (C2_not (C2_cmpgti IntRegs:$src1, s10ExtPred:$src2))>;
4816 // rs <= rt -> !(rs > rt).
4817 def : Pat<(i1 (setle (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4818 (i1 (C2_not (C2_cmpgt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))>;
4820 // Rss <= Rtt -> !(Rss > Rtt).
4821 def: Pat<(i1 (setle (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4822 (C2_not (C2_cmpgtp DoubleRegs:$src1, DoubleRegs:$src2))>;
4824 // Map cmpne -> cmpeq.
4825 // Hexagon_TODO: We should improve on this.
4826 // rs != rt -> !(rs == rt).
4827 let AddedComplexity = 30 in
4828 def: Pat<(i1 (setne (i32 IntRegs:$src1), s10ExtPred:$src2)),
4829 (C2_not (C2_cmpeqi IntRegs:$src1, s10ExtPred:$src2))>;
4831 // Map cmpne(Rs) -> !cmpeqe(Rs).
4832 // rs != rt -> !(rs == rt).
4833 def : Pat <(i1 (setne (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4834 (i1 (C2_not (i1 (C2_cmpeq (i32 IntRegs:$src1), (i32 IntRegs:$src2)))))>;
4836 // Convert setne back to xor for hexagon since we compute w/ pred registers.
4837 def: Pat<(i1 (setne (i1 PredRegs:$src1), (i1 PredRegs:$src2))),
4838 (C2_xor PredRegs:$src1, PredRegs:$src2)>;
4840 // Map cmpne(Rss) -> !cmpew(Rss).
4841 // rs != rt -> !(rs == rt).
4842 def: Pat<(i1 (setne (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4843 (C2_not (C2_cmpeqp DoubleRegs:$src1, DoubleRegs:$src2))>;
4845 // Map cmpge(Rs, Rt) -> !(cmpgt(Rs, Rt).
4846 // rs >= rt -> !(rt > rs).
4847 def : Pat <(i1 (setge (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
4848 (i1 (C2_not (i1 (C2_cmpgt (i32 IntRegs:$src2), (i32 IntRegs:$src1)))))>;
4850 // cmpge(Rs, Imm) -> cmpgt(Rs, Imm-1)
4851 let AddedComplexity = 30 in
4852 def: Pat<(i1 (setge (i32 IntRegs:$src1), s8ExtPred:$src2)),
4853 (C2_cmpgti IntRegs:$src1, (DEC_CONST_SIGNED s8ExtPred:$src2))>;
4855 // Map cmpge(Rss, Rtt) -> !cmpgt(Rtt, Rss).
4856 // rss >= rtt -> !(rtt > rss).
4857 def: Pat<(i1 (setge (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4858 (C2_not (C2_cmpgtp DoubleRegs:$src2, DoubleRegs:$src1))>;
4860 // Map cmplt(Rs, Imm) -> !cmpge(Rs, Imm).
4861 // !cmpge(Rs, Imm) -> !cmpgt(Rs, Imm-1).
4862 // rs < rt -> !(rs >= rt).
4863 let AddedComplexity = 30 in
4864 def: Pat<(i1 (setlt (i32 IntRegs:$src1), s8ExtPred:$src2)),
4865 (C2_not (C2_cmpgti IntRegs:$src1, (DEC_CONST_SIGNED s8ExtPred:$src2)))>;
4867 // Generate cmpgeu(Rs, #0) -> cmpeq(Rs, Rs)
4868 def: Pat<(i1 (setuge (i32 IntRegs:$src1), 0)),
4869 (C2_cmpeq IntRegs:$src1, IntRegs:$src1)>;
4871 // Generate cmpgeu(Rs, #u8) -> cmpgtu(Rs, #u8 -1)
4872 def: Pat<(i1 (setuge (i32 IntRegs:$src1), u8ExtPred:$src2)),
4873 (C2_cmpgtui IntRegs:$src1, (DEC_CONST_UNSIGNED u8ExtPred:$src2))>;
4875 // Generate cmpgtu(Rs, #u9)
4876 def: Pat<(i1 (setugt (i32 IntRegs:$src1), u9ExtPred:$src2)),
4877 (C2_cmpgtui IntRegs:$src1, u9ExtPred:$src2)>;
4879 // Map from Rs >= Rt -> !(Rt > Rs).
4880 // rs >= rt -> !(rt > rs).
4881 def: Pat<(i1 (setuge (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4882 (C2_not (C2_cmpgtup DoubleRegs:$src2, DoubleRegs:$src1))>;
4884 // Map from cmpleu(Rss, Rtt) -> !cmpgtu(Rss, Rtt-1).
4885 // Map from (Rs <= Rt) -> !(Rs > Rt).
4886 def: Pat<(i1 (setule (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2))),
4887 (C2_not (C2_cmpgtup DoubleRegs:$src1, DoubleRegs:$src2))>;
4891 def: Pat<(i32 (sext (i1 PredRegs:$src1))),
4892 (C2_muxii PredRegs:$src1, -1, 0)>;
4895 def: Pat<(i64 (sext (i1 PredRegs:$src1))),
4896 (A2_combinew (A2_tfrsi -1), (C2_muxii PredRegs:$src1, -1, 0))>;
4898 // Convert sign-extended load back to load and sign extend.
4900 def: Pat <(i64 (sextloadi32 ADDRriS11_2:$src1)),
4901 (i64 (A2_sxtw (L2_loadri_io AddrFI:$src1, 0)))>;
4905 def: Pat<(i32 (zext (i1 PredRegs:$src1))),
4906 (C2_muxii PredRegs:$src1, 1, 0)>;
4908 // Map from Rs = Pd to Pd = mux(Pd, #1, #0)
4909 def: Pat<(i32 (anyext (i1 PredRegs:$src1))),
4910 (C2_muxii PredRegs:$src1, 1, 0)>;
4912 // Map from Rss = Pd to Rdd = sxtw (mux(Pd, #1, #0))
4913 def: Pat<(i64 (anyext (i1 PredRegs:$src1))),
4914 (A2_sxtw (C2_muxii PredRegs:$src1, 1, 0))>;
4916 def: Pat<(i64 (or (i64 (shl (i64 DoubleRegs:$srcHigh),
4918 (i64 (zextloadi32 ADDRriS11_2:$srcLow)))),
4919 (i64 (A2_combinew (EXTRACT_SUBREG (i64 DoubleRegs:$srcHigh), subreg_loreg),
4920 (L2_loadri_io AddrFI:$srcLow, 0)))>;
4922 // Multiply 64-bit unsigned and use upper result.
4923 def : Pat <(mulhu (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2)),
4929 (S2_lsr_i_p (M2_dpmpyuu_s0 (LoReg $src1), (LoReg $src2)), 32),
4932 (A2_combinew (A2_tfrsi 0),
4933 (LoReg (M2_dpmpyuu_s0 (LoReg $src1), (HiReg $src2))))),
4937 (S2_lsr_i_p (M2_dpmpyuu_s0 (LoReg $src1), (HiReg $src2)), 32)
4940 // Hexagon specific ISD nodes.
4941 def SDTHexagonADJDYNALLOC : SDTypeProfile<1, 2, [SDTCisVT<0, i32>,
4943 def SDTHexagonARGEXTEND : SDTypeProfile<1, 1, [SDTCisVT<0, i32>]>;
4945 def Hexagon_ADJDYNALLOC : SDNode<"HexagonISD::ADJDYNALLOC",
4946 SDTHexagonADJDYNALLOC>;
4947 def Hexagon_ARGEXTEND : SDNode<"HexagonISD::ARGEXTEND", SDTHexagonARGEXTEND>;
4949 // Needed to tag these instructions for stack layout.
4950 let isCodeGenOnly = 1, usesCustomInserter = 1 in
4951 def ADJDYNALLOC : T_Addri<s6Imm>;
4953 def: Pat<(Hexagon_ADJDYNALLOC I32:$Rs, s16ImmPred:$s16),
4954 (ADJDYNALLOC I32:$Rs, imm:$s16)>;
4956 let isCodeGenOnly = 1 in
4957 def ARGEXTEND : ALU32_rr <(outs IntRegs:$dst), (ins IntRegs:$src1),
4959 [(set (i32 IntRegs:$dst),
4960 (Hexagon_ARGEXTEND (i32 IntRegs:$src1)))]>;
4962 let AddedComplexity = 100 in
4963 def: Pat<(i32 (sext_inreg (Hexagon_ARGEXTEND (i32 IntRegs:$src1)), i16)),
4964 (i32 IntRegs:$src1)>;
4966 def HexagonWrapperJT: SDNode<"HexagonISD::WrapperJT", SDTIntUnaryOp>;
4968 def : Pat<(HexagonWrapperJT tjumptable:$dst),
4969 (i32 (CONST32_set_jt tjumptable:$dst))>;
4973 //===----------------------------------------------------------------------===//
4975 // Shift by immediate/register and accumulate/logical
4976 //===----------------------------------------------------------------------===//
4978 // Rx[+-&|]=asr(Rs,#u5)
4979 // Rx[+-&|^]=lsr(Rs,#u5)
4980 // Rx[+-&|^]=asl(Rs,#u5)
4982 let hasNewValue = 1, opNewValue = 0 in
4983 class T_shift_imm_acc_r <string opc1, string opc2, SDNode OpNode1,
4984 SDNode OpNode2, bits<3> majOp, bits<2> minOp>
4985 : SInst_acc<(outs IntRegs:$Rx),
4986 (ins IntRegs:$src1, IntRegs:$Rs, u5Imm:$u5),
4987 "$Rx "#opc2#opc1#"($Rs, #$u5)",
4988 [(set (i32 IntRegs:$Rx),
4989 (OpNode2 (i32 IntRegs:$src1),
4990 (OpNode1 (i32 IntRegs:$Rs), u5ImmPred:$u5)))],
4991 "$src1 = $Rx", S_2op_tc_2_SLOT23> {
4996 let IClass = 0b1000;
4998 let Inst{27-24} = 0b1110;
4999 let Inst{23-22} = majOp{2-1};
5001 let Inst{7} = majOp{0};
5002 let Inst{6-5} = minOp;
5004 let Inst{20-16} = Rs;
5005 let Inst{12-8} = u5;
5008 // Rx[+-&|]=asr(Rs,Rt)
5009 // Rx[+-&|^]=lsr(Rs,Rt)
5010 // Rx[+-&|^]=asl(Rs,Rt)
5012 let hasNewValue = 1, opNewValue = 0 in
5013 class T_shift_reg_acc_r <string opc1, string opc2, SDNode OpNode1,
5014 SDNode OpNode2, bits<2> majOp, bits<2> minOp>
5015 : SInst_acc<(outs IntRegs:$Rx),
5016 (ins IntRegs:$src1, IntRegs:$Rs, IntRegs:$Rt),
5017 "$Rx "#opc2#opc1#"($Rs, $Rt)",
5018 [(set (i32 IntRegs:$Rx),
5019 (OpNode2 (i32 IntRegs:$src1),
5020 (OpNode1 (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))],
5021 "$src1 = $Rx", S_3op_tc_2_SLOT23 > {
5026 let IClass = 0b1100;
5028 let Inst{27-24} = 0b1100;
5029 let Inst{23-22} = majOp;
5030 let Inst{7-6} = minOp;
5032 let Inst{20-16} = Rs;
5033 let Inst{12-8} = Rt;
5036 // Rxx[+-&|]=asr(Rss,#u6)
5037 // Rxx[+-&|^]=lsr(Rss,#u6)
5038 // Rxx[+-&|^]=asl(Rss,#u6)
5040 class T_shift_imm_acc_p <string opc1, string opc2, SDNode OpNode1,
5041 SDNode OpNode2, bits<3> majOp, bits<2> minOp>
5042 : SInst_acc<(outs DoubleRegs:$Rxx),
5043 (ins DoubleRegs:$src1, DoubleRegs:$Rss, u6Imm:$u6),
5044 "$Rxx "#opc2#opc1#"($Rss, #$u6)",
5045 [(set (i64 DoubleRegs:$Rxx),
5046 (OpNode2 (i64 DoubleRegs:$src1),
5047 (OpNode1 (i64 DoubleRegs:$Rss), u6ImmPred:$u6)))],
5048 "$src1 = $Rxx", S_2op_tc_2_SLOT23> {
5053 let IClass = 0b1000;
5055 let Inst{27-24} = 0b0010;
5056 let Inst{23-22} = majOp{2-1};
5057 let Inst{7} = majOp{0};
5058 let Inst{6-5} = minOp;
5059 let Inst{4-0} = Rxx;
5060 let Inst{20-16} = Rss;
5061 let Inst{13-8} = u6;
5065 // Rxx[+-&|]=asr(Rss,Rt)
5066 // Rxx[+-&|^]=lsr(Rss,Rt)
5067 // Rxx[+-&|^]=asl(Rss,Rt)
5068 // Rxx[+-&|^]=lsl(Rss,Rt)
5070 class T_shift_reg_acc_p <string opc1, string opc2, SDNode OpNode1,
5071 SDNode OpNode2, bits<3> majOp, bits<2> minOp>
5072 : SInst_acc<(outs DoubleRegs:$Rxx),
5073 (ins DoubleRegs:$src1, DoubleRegs:$Rss, IntRegs:$Rt),
5074 "$Rxx "#opc2#opc1#"($Rss, $Rt)",
5075 [(set (i64 DoubleRegs:$Rxx),
5076 (OpNode2 (i64 DoubleRegs:$src1),
5077 (OpNode1 (i64 DoubleRegs:$Rss), (i32 IntRegs:$Rt))))],
5078 "$src1 = $Rxx", S_3op_tc_2_SLOT23> {
5083 let IClass = 0b1100;
5085 let Inst{27-24} = 0b1011;
5086 let Inst{23-21} = majOp;
5087 let Inst{20-16} = Rss;
5088 let Inst{12-8} = Rt;
5089 let Inst{7-6} = minOp;
5090 let Inst{4-0} = Rxx;
5093 //===----------------------------------------------------------------------===//
5094 // Multi-class for the shift instructions with logical/arithmetic operators.
5095 //===----------------------------------------------------------------------===//
5097 multiclass xtype_imm_base<string OpcStr1, string OpcStr2, SDNode OpNode1,
5098 SDNode OpNode2, bits<3> majOp, bits<2> minOp > {
5099 def _i_r#NAME : T_shift_imm_acc_r< OpcStr1, OpcStr2, OpNode1,
5100 OpNode2, majOp, minOp >;
5101 def _i_p#NAME : T_shift_imm_acc_p< OpcStr1, OpcStr2, OpNode1,
5102 OpNode2, majOp, minOp >;
5105 multiclass xtype_imm_acc<string opc1, SDNode OpNode, bits<2>minOp> {
5106 let AddedComplexity = 100 in
5107 defm _acc : xtype_imm_base< opc1, "+= ", OpNode, add, 0b001, minOp>;
5109 defm _nac : xtype_imm_base< opc1, "-= ", OpNode, sub, 0b000, minOp>;
5110 defm _and : xtype_imm_base< opc1, "&= ", OpNode, and, 0b010, minOp>;
5111 defm _or : xtype_imm_base< opc1, "|= ", OpNode, or, 0b011, minOp>;
5114 multiclass xtype_xor_imm_acc<string opc1, SDNode OpNode, bits<2>minOp> {
5115 let AddedComplexity = 100 in
5116 defm _xacc : xtype_imm_base< opc1, "^= ", OpNode, xor, 0b100, minOp>;
5119 defm S2_asr : xtype_imm_acc<"asr", sra, 0b00>;
5121 defm S2_lsr : xtype_imm_acc<"lsr", srl, 0b01>,
5122 xtype_xor_imm_acc<"lsr", srl, 0b01>;
5124 defm S2_asl : xtype_imm_acc<"asl", shl, 0b10>,
5125 xtype_xor_imm_acc<"asl", shl, 0b10>;
5127 multiclass xtype_reg_acc_r<string opc1, SDNode OpNode, bits<2>minOp> {
5128 let AddedComplexity = 100 in
5129 def _acc : T_shift_reg_acc_r <opc1, "+= ", OpNode, add, 0b11, minOp>;
5131 def _nac : T_shift_reg_acc_r <opc1, "-= ", OpNode, sub, 0b10, minOp>;
5132 def _and : T_shift_reg_acc_r <opc1, "&= ", OpNode, and, 0b01, minOp>;
5133 def _or : T_shift_reg_acc_r <opc1, "|= ", OpNode, or, 0b00, minOp>;
5136 multiclass xtype_reg_acc_p<string opc1, SDNode OpNode, bits<2>minOp> {
5137 let AddedComplexity = 100 in
5138 def _acc : T_shift_reg_acc_p <opc1, "+= ", OpNode, add, 0b110, minOp>;
5140 def _nac : T_shift_reg_acc_p <opc1, "-= ", OpNode, sub, 0b100, minOp>;
5141 def _and : T_shift_reg_acc_p <opc1, "&= ", OpNode, and, 0b010, minOp>;
5142 def _or : T_shift_reg_acc_p <opc1, "|= ", OpNode, or, 0b000, minOp>;
5143 def _xor : T_shift_reg_acc_p <opc1, "^= ", OpNode, xor, 0b011, minOp>;
5146 multiclass xtype_reg_acc<string OpcStr, SDNode OpNode, bits<2> minOp > {
5147 defm _r_r : xtype_reg_acc_r <OpcStr, OpNode, minOp>;
5148 defm _r_p : xtype_reg_acc_p <OpcStr, OpNode, minOp>;
5151 defm S2_asl : xtype_reg_acc<"asl", shl, 0b10>;
5152 defm S2_asr : xtype_reg_acc<"asr", sra, 0b00>;
5153 defm S2_lsr : xtype_reg_acc<"lsr", srl, 0b01>;
5154 defm S2_lsl : xtype_reg_acc<"lsl", shl, 0b11>;
5156 //===----------------------------------------------------------------------===//
5157 let hasSideEffects = 0 in
5158 class T_S3op_1 <string mnemonic, RegisterClass RC, bits<2> MajOp, bits<3> MinOp,
5159 bit SwapOps, bit isSat = 0, bit isRnd = 0, bit hasShift = 0>
5160 : SInst <(outs RC:$dst),
5161 (ins DoubleRegs:$src1, DoubleRegs:$src2),
5162 "$dst = "#mnemonic#"($src1, $src2)"#!if(isRnd, ":rnd", "")
5163 #!if(hasShift,":>>1","")
5164 #!if(isSat, ":sat", ""),
5165 [], "", S_3op_tc_2_SLOT23 > {
5170 let IClass = 0b1100;
5172 let Inst{27-24} = 0b0001;
5173 let Inst{23-22} = MajOp;
5174 let Inst{20-16} = !if (SwapOps, src2, src1);
5175 let Inst{12-8} = !if (SwapOps, src1, src2);
5176 let Inst{7-5} = MinOp;
5177 let Inst{4-0} = dst;
5180 class T_S3op_64 <string mnemonic, bits<2> MajOp, bits<3> MinOp, bit SwapOps,
5181 bit isSat = 0, bit isRnd = 0, bit hasShift = 0 >
5182 : T_S3op_1 <mnemonic, DoubleRegs, MajOp, MinOp, SwapOps,
5183 isSat, isRnd, hasShift>;
5185 let Itinerary = S_3op_tc_1_SLOT23 in {
5186 def S2_shuffeb : T_S3op_64 < "shuffeb", 0b00, 0b010, 0>;
5187 def S2_shuffeh : T_S3op_64 < "shuffeh", 0b00, 0b110, 0>;
5188 def S2_shuffob : T_S3op_64 < "shuffob", 0b00, 0b100, 1>;
5189 def S2_shuffoh : T_S3op_64 < "shuffoh", 0b10, 0b000, 1>;
5191 def S2_vtrunewh : T_S3op_64 < "vtrunewh", 0b10, 0b010, 0>;
5192 def S2_vtrunowh : T_S3op_64 < "vtrunowh", 0b10, 0b100, 0>;
5195 def S2_lfsp : T_S3op_64 < "lfs", 0b10, 0b110, 0>;
5197 let hasSideEffects = 0 in
5198 class T_S3op_2 <string mnemonic, bits<3> MajOp, bit SwapOps>
5199 : SInst < (outs DoubleRegs:$Rdd),
5200 (ins DoubleRegs:$Rss, DoubleRegs:$Rtt, PredRegs:$Pu),
5201 "$Rdd = "#mnemonic#"($Rss, $Rtt, $Pu)",
5202 [], "", S_3op_tc_1_SLOT23 > {
5208 let IClass = 0b1100;
5210 let Inst{27-24} = 0b0010;
5211 let Inst{23-21} = MajOp;
5212 let Inst{20-16} = !if (SwapOps, Rtt, Rss);
5213 let Inst{12-8} = !if (SwapOps, Rss, Rtt);
5215 let Inst{4-0} = Rdd;
5218 def S2_valignrb : T_S3op_2 < "valignb", 0b000, 1>;
5219 def S2_vsplicerb : T_S3op_2 < "vspliceb", 0b100, 0>;
5221 //===----------------------------------------------------------------------===//
5222 // Template class used by vector shift, vector rotate, vector neg,
5223 // 32-bit shift, 64-bit shifts, etc.
5224 //===----------------------------------------------------------------------===//
5226 let hasSideEffects = 0 in
5227 class T_S3op_3 <string mnemonic, RegisterClass RC, bits<2> MajOp,
5228 bits<2> MinOp, bit isSat = 0, list<dag> pattern = [] >
5229 : SInst <(outs RC:$dst),
5230 (ins RC:$src1, IntRegs:$src2),
5231 "$dst = "#mnemonic#"($src1, $src2)"#!if(isSat, ":sat", ""),
5232 pattern, "", S_3op_tc_1_SLOT23> {
5237 let IClass = 0b1100;
5239 let Inst{27-24} = !if(!eq(!cast<string>(RC), "IntRegs"), 0b0110, 0b0011);
5240 let Inst{23-22} = MajOp;
5241 let Inst{20-16} = src1;
5242 let Inst{12-8} = src2;
5243 let Inst{7-6} = MinOp;
5244 let Inst{4-0} = dst;
5247 let hasNewValue = 1 in
5248 class T_S3op_shift32 <string mnemonic, SDNode OpNode, bits<2> MinOp>
5249 : T_S3op_3 <mnemonic, IntRegs, 0b01, MinOp, 0,
5250 [(set (i32 IntRegs:$dst), (OpNode (i32 IntRegs:$src1),
5251 (i32 IntRegs:$src2)))]>;
5253 let hasNewValue = 1, Itinerary = S_3op_tc_2_SLOT23 in
5254 class T_S3op_shift32_Sat <string mnemonic, bits<2> MinOp>
5255 : T_S3op_3 <mnemonic, IntRegs, 0b00, MinOp, 1, []>;
5258 class T_S3op_shift64 <string mnemonic, SDNode OpNode, bits<2> MinOp>
5259 : T_S3op_3 <mnemonic, DoubleRegs, 0b10, MinOp, 0,
5260 [(set (i64 DoubleRegs:$dst), (OpNode (i64 DoubleRegs:$src1),
5261 (i32 IntRegs:$src2)))]>;
5264 class T_S3op_shiftVect <string mnemonic, bits<2> MajOp, bits<2> MinOp>
5265 : T_S3op_3 <mnemonic, DoubleRegs, MajOp, MinOp, 0, []>;
5268 // Shift by register
5269 // Rdd=[asr|lsr|asl|lsl](Rss,Rt)
5271 def S2_asr_r_p : T_S3op_shift64 < "asr", sra, 0b00>;
5272 def S2_lsr_r_p : T_S3op_shift64 < "lsr", srl, 0b01>;
5273 def S2_asl_r_p : T_S3op_shift64 < "asl", shl, 0b10>;
5274 def S2_lsl_r_p : T_S3op_shift64 < "lsl", shl, 0b11>;
5276 // Rd=[asr|lsr|asl|lsl](Rs,Rt)
5278 def S2_asr_r_r : T_S3op_shift32<"asr", sra, 0b00>;
5279 def S2_lsr_r_r : T_S3op_shift32<"lsr", srl, 0b01>;
5280 def S2_asl_r_r : T_S3op_shift32<"asl", shl, 0b10>;
5281 def S2_lsl_r_r : T_S3op_shift32<"lsl", shl, 0b11>;
5283 // Shift by register with saturation
5284 // Rd=asr(Rs,Rt):sat
5285 // Rd=asl(Rs,Rt):sat
5287 let Defs = [USR_OVF] in {
5288 def S2_asr_r_r_sat : T_S3op_shift32_Sat<"asr", 0b00>;
5289 def S2_asl_r_r_sat : T_S3op_shift32_Sat<"asl", 0b10>;
5292 let hasNewValue = 1, hasSideEffects = 0 in
5293 class T_S3op_8 <string opc, bits<3> MinOp, bit isSat, bit isRnd, bit hasShift, bit hasSplat = 0>
5294 : SInst < (outs IntRegs:$Rd),
5295 (ins DoubleRegs:$Rss, IntRegs:$Rt),
5296 "$Rd = "#opc#"($Rss, $Rt"#!if(hasSplat, "*", "")#")"
5297 #!if(hasShift, ":<<1", "")
5298 #!if(isRnd, ":rnd", "")
5299 #!if(isSat, ":sat", ""),
5300 [], "", S_3op_tc_1_SLOT23 > {
5305 let IClass = 0b1100;
5307 let Inst{27-24} = 0b0101;
5308 let Inst{20-16} = Rss;
5309 let Inst{12-8} = Rt;
5310 let Inst{7-5} = MinOp;
5314 def S2_asr_r_svw_trun : T_S3op_8<"vasrw", 0b010, 0, 0, 0>;
5316 let Defs = [USR_OVF], Itinerary = S_3op_tc_2_SLOT23 in
5317 def S2_vcrotate : T_S3op_shiftVect < "vcrotate", 0b11, 0b00>;
5319 let hasSideEffects = 0 in
5320 class T_S3op_7 <string mnemonic, bit MajOp >
5321 : SInst <(outs DoubleRegs:$Rdd),
5322 (ins DoubleRegs:$Rss, DoubleRegs:$Rtt, u3Imm:$u3),
5323 "$Rdd = "#mnemonic#"($Rss, $Rtt, #$u3)" ,
5324 [], "", S_3op_tc_1_SLOT23 > {
5330 let IClass = 0b1100;
5332 let Inst{27-24} = 0b0000;
5333 let Inst{23} = MajOp;
5334 let Inst{20-16} = !if(MajOp, Rss, Rtt);
5335 let Inst{12-8} = !if(MajOp, Rtt, Rss);
5337 let Inst{4-0} = Rdd;
5340 def S2_valignib : T_S3op_7 < "valignb", 0>;
5341 def S2_vspliceib : T_S3op_7 < "vspliceb", 1>;
5343 //===----------------------------------------------------------------------===//
5344 // Template class for 'insert bitfield' instructions
5345 //===----------------------------------------------------------------------===//
5346 let hasSideEffects = 0 in
5347 class T_S3op_insert <string mnemonic, RegisterClass RC>
5348 : SInst <(outs RC:$dst),
5349 (ins RC:$src1, RC:$src2, DoubleRegs:$src3),
5350 "$dst = "#mnemonic#"($src2, $src3)" ,
5351 [], "$src1 = $dst", S_3op_tc_1_SLOT23 > {
5356 let IClass = 0b1100;
5358 let Inst{27-26} = 0b10;
5359 let Inst{25-24} = !if(!eq(!cast<string>(RC), "IntRegs"), 0b00, 0b10);
5361 let Inst{20-16} = src2;
5362 let Inst{12-8} = src3;
5363 let Inst{4-0} = dst;
5366 let hasSideEffects = 0 in
5367 class T_S2op_insert <bits<4> RegTyBits, RegisterClass RC, Operand ImmOp>
5368 : SInst <(outs RC:$dst), (ins RC:$dst2, RC:$src1, ImmOp:$src2, ImmOp:$src3),
5369 "$dst = insert($src1, #$src2, #$src3)",
5370 [], "$dst2 = $dst", S_2op_tc_2_SLOT23> {
5377 string ImmOpStr = !cast<string>(ImmOp);
5379 let bit23 = !if (!eq(ImmOpStr, "u6Imm"), src3{5}, 0);
5380 let bit13 = !if (!eq(ImmOpStr, "u6Imm"), src2{5}, 0);
5382 let IClass = 0b1000;
5384 let Inst{27-24} = RegTyBits;
5385 let Inst{23} = bit23;
5386 let Inst{22-21} = src3{4-3};
5387 let Inst{20-16} = src1;
5388 let Inst{13} = bit13;
5389 let Inst{12-8} = src2{4-0};
5390 let Inst{7-5} = src3{2-0};
5391 let Inst{4-0} = dst;
5394 // Rx=insert(Rs,Rtt)
5395 // Rx=insert(Rs,#u5,#U5)
5396 let hasNewValue = 1 in {
5397 def S2_insert_rp : T_S3op_insert <"insert", IntRegs>;
5398 def S2_insert : T_S2op_insert <0b1111, IntRegs, u5Imm>;
5401 // Rxx=insert(Rss,Rtt)
5402 // Rxx=insert(Rss,#u6,#U6)
5403 def S2_insertp_rp : T_S3op_insert<"insert", DoubleRegs>;
5404 def S2_insertp : T_S2op_insert <0b0011, DoubleRegs, u6Imm>;
5406 //===----------------------------------------------------------------------===//
5407 // Template class for 'extract bitfield' instructions
5408 //===----------------------------------------------------------------------===//
5409 let hasNewValue = 1, hasSideEffects = 0 in
5410 class T_S3op_extract <string mnemonic, bits<2> MinOp>
5411 : SInst <(outs IntRegs:$Rd), (ins IntRegs:$Rs, DoubleRegs:$Rtt),
5412 "$Rd = "#mnemonic#"($Rs, $Rtt)",
5413 [], "", S_3op_tc_2_SLOT23 > {
5418 let IClass = 0b1100;
5420 let Inst{27-22} = 0b100100;
5421 let Inst{20-16} = Rs;
5422 let Inst{12-8} = Rtt;
5423 let Inst{7-6} = MinOp;
5427 let hasSideEffects = 0 in
5428 class T_S2op_extract <string mnemonic, bits<4> RegTyBits,
5429 RegisterClass RC, Operand ImmOp>
5430 : SInst <(outs RC:$dst), (ins RC:$src1, ImmOp:$src2, ImmOp:$src3),
5431 "$dst = "#mnemonic#"($src1, #$src2, #$src3)",
5432 [], "", S_2op_tc_2_SLOT23> {
5439 string ImmOpStr = !cast<string>(ImmOp);
5441 let bit23 = !if (!eq(ImmOpStr, "u6Imm"), src3{5},
5442 !if (!eq(mnemonic, "extractu"), 0, 1));
5444 let bit13 = !if (!eq(ImmOpStr, "u6Imm"), src2{5}, 0);
5446 let IClass = 0b1000;
5448 let Inst{27-24} = RegTyBits;
5449 let Inst{23} = bit23;
5450 let Inst{22-21} = src3{4-3};
5451 let Inst{20-16} = src1;
5452 let Inst{13} = bit13;
5453 let Inst{12-8} = src2{4-0};
5454 let Inst{7-5} = src3{2-0};
5455 let Inst{4-0} = dst;
5460 // Rdd=extractu(Rss,Rtt)
5461 // Rdd=extractu(Rss,#u6,#U6)
5462 def S2_extractup_rp : T_S3op_64 < "extractu", 0b00, 0b000, 0>;
5463 def S2_extractup : T_S2op_extract <"extractu", 0b0001, DoubleRegs, u6Imm>;
5465 // Rd=extractu(Rs,Rtt)
5466 // Rd=extractu(Rs,#u5,#U5)
5467 let hasNewValue = 1 in {
5468 def S2_extractu_rp : T_S3op_extract<"extractu", 0b00>;
5469 def S2_extractu : T_S2op_extract <"extractu", 0b1101, IntRegs, u5Imm>;
5472 // Change the sign of the immediate for Rd=-mpyi(Rs,#u8)
5473 def: Pat<(mul (i32 IntRegs:$src1), (ineg n8ImmPred:$src2)),
5474 (M2_mpysin IntRegs:$src1, u8ImmPred:$src2)>;
5476 //===----------------------------------------------------------------------===//
5477 // :raw for of tableindx[bdhw] insns
5478 //===----------------------------------------------------------------------===//
5480 let hasSideEffects = 0, hasNewValue = 1, opNewValue = 0 in
5481 class tableidxRaw<string OpStr, bits<2>MinOp>
5482 : SInst <(outs IntRegs:$Rx),
5483 (ins IntRegs:$_dst_, IntRegs:$Rs, u4Imm:$u4, s6Imm:$S6),
5484 "$Rx = "#OpStr#"($Rs, #$u4, #$S6):raw",
5485 [], "$Rx = $_dst_" > {
5491 let IClass = 0b1000;
5493 let Inst{27-24} = 0b0111;
5494 let Inst{23-22} = MinOp;
5495 let Inst{21} = u4{3};
5496 let Inst{20-16} = Rs;
5497 let Inst{13-8} = S6;
5498 let Inst{7-5} = u4{2-0};
5502 def S2_tableidxb : tableidxRaw<"tableidxb", 0b00>;
5503 def S2_tableidxh : tableidxRaw<"tableidxh", 0b01>;
5504 def S2_tableidxw : tableidxRaw<"tableidxw", 0b10>;
5505 def S2_tableidxd : tableidxRaw<"tableidxd", 0b11>;
5507 //===----------------------------------------------------------------------===//
5508 // V3 Instructions +
5509 //===----------------------------------------------------------------------===//
5511 include "HexagonInstrInfoV3.td"
5513 //===----------------------------------------------------------------------===//
5514 // V3 Instructions -
5515 //===----------------------------------------------------------------------===//
5517 //===----------------------------------------------------------------------===//
5518 // V4 Instructions +
5519 //===----------------------------------------------------------------------===//
5521 include "HexagonInstrInfoV4.td"
5523 //===----------------------------------------------------------------------===//
5524 // V4 Instructions -
5525 //===----------------------------------------------------------------------===//
5527 //===----------------------------------------------------------------------===//
5528 // V5 Instructions +
5529 //===----------------------------------------------------------------------===//
5531 include "HexagonInstrInfoV5.td"
5533 //===----------------------------------------------------------------------===//
5534 // V5 Instructions -
5535 //===----------------------------------------------------------------------===//
5537 //===----------------------------------------------------------------------===//
5538 // ALU32/64/Vector +
5539 //===----------------------------------------------------------------------===///
5541 include "HexagonInstrInfoVector.td"