1 //=- HexagonInstrInfoV4.td - Target Desc. for Hexagon Target -*- 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 V4 instructions in TableGen format.
12 //===----------------------------------------------------------------------===//
14 let hasSideEffects = 0 in
15 class T_Immext<Operand ImmType>
16 : EXTENDERInst<(outs), (ins ImmType:$imm),
17 "immext(#$imm)", []> {
21 let Inst{27-16} = imm{31-20};
22 let Inst{13-0} = imm{19-6};
25 def A4_ext : T_Immext<u26_6Imm>;
26 let isCodeGenOnly = 1 in {
28 def A4_ext_b : T_Immext<brtarget>;
30 def A4_ext_c : T_Immext<calltarget>;
31 def A4_ext_g : T_Immext<globaladdress>;
34 // Fold (add (CONST32 tglobaladdr:$addr) <offset>) into a global address.
35 def FoldGlobalAddr : ComplexPattern<i32, 1, "foldGlobalAddress", [], []>;
37 // Fold (add (CONST32_GP tglobaladdr:$addr) <offset>) into a global address.
38 def FoldGlobalAddrGP : ComplexPattern<i32, 1, "foldGlobalAddressGP", [], []>;
40 def NumUsesBelowThresCONST32 : PatFrag<(ops node:$addr),
41 (HexagonCONST32 node:$addr), [{
42 return hasNumUsesBelowThresGA(N->getOperand(0).getNode());
45 // Hexagon V4 Architecture spec defines 8 instruction classes:
46 // LD ST ALU32 XTYPE J JR MEMOP NV CR SYSTEM(system is not implemented in the
50 // ========================================
51 // Loads (8/16/32/64 bit)
55 // ========================================
56 // Stores (8/16/32/64 bit)
59 // ALU32 Instructions:
60 // ========================================
61 // Arithmetic / Logical (32 bit)
64 // XTYPE Instructions (32/64 bit):
65 // ========================================
66 // Arithmetic, Logical, Bit Manipulation
67 // Multiply (Integer, Fractional, Complex)
68 // Permute / Vector Permute Operations
69 // Predicate Operations
70 // Shift / Shift with Add/Sub/Logical
72 // Vector Halfword (ALU, Shift, Multiply)
73 // Vector Word (ALU, Shift)
76 // ========================================
77 // Jump/Call PC-relative
80 // ========================================
83 // MEMOP Instructions:
84 // ========================================
85 // Operation on memory (8/16/32 bit)
88 // ========================================
93 // ========================================
94 // Control-Register Transfers
95 // Hardware Loop Setup
96 // Predicate Logicals & Reductions
98 // SYSTEM Instructions (not implemented in the compiler):
99 // ========================================
105 //===----------------------------------------------------------------------===//
107 //===----------------------------------------------------------------------===//
109 class T_ALU32_3op_not<string mnemonic, bits<3> MajOp, bits<3> MinOp,
111 : T_ALU32_3op<mnemonic, MajOp, MinOp, OpsRev, 0> {
112 let AsmString = "$Rd = "#mnemonic#"($Rs, ~$Rt)";
115 let BaseOpcode = "andn_rr", CextOpcode = "andn", isCodeGenOnly = 0 in
116 def A4_andn : T_ALU32_3op_not<"and", 0b001, 0b100, 1>;
117 let BaseOpcode = "orn_rr", CextOpcode = "orn", isCodeGenOnly = 0 in
118 def A4_orn : T_ALU32_3op_not<"or", 0b001, 0b101, 1>;
120 let CextOpcode = "rcmp.eq", isCodeGenOnly = 0 in
121 def A4_rcmpeq : T_ALU32_3op<"cmp.eq", 0b011, 0b010, 0, 1>;
122 let CextOpcode = "!rcmp.eq", isCodeGenOnly = 0 in
123 def A4_rcmpneq : T_ALU32_3op<"!cmp.eq", 0b011, 0b011, 0, 1>;
125 let isCodeGenOnly = 0 in {
126 def C4_cmpneq : T_ALU32_3op_cmp<"!cmp.eq", 0b00, 1, 1>;
127 def C4_cmplte : T_ALU32_3op_cmp<"!cmp.gt", 0b10, 1, 0>;
128 def C4_cmplteu : T_ALU32_3op_cmp<"!cmp.gtu", 0b11, 1, 0>;
131 // Pats for instruction selection.
133 // A class to embed the usual comparison patfrags within a zext to i32.
134 // The seteq/setne frags use "lhs" and "rhs" as operands, so use the same
135 // names, or else the frag's "body" won't match the operands.
136 class CmpInReg<PatFrag Op>
137 : PatFrag<(ops node:$lhs, node:$rhs),(i32 (zext (i1 Op.Fragment)))>;
139 def: T_cmp32_rr_pat<A4_rcmpeq, CmpInReg<seteq>, i32>;
140 def: T_cmp32_rr_pat<A4_rcmpneq, CmpInReg<setne>, i32>;
142 class T_CMP_rrbh<string mnemonic, bits<3> MinOp, bit IsComm>
143 : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt),
144 "$Pd = "#mnemonic#"($Rs, $Rt)", [], "", S_3op_tc_2early_SLOT23>,
146 let validSubTargets = HasV4SubT;
147 let InputType = "reg";
148 let CextOpcode = mnemonic;
150 let isCommutable = IsComm;
151 let hasSideEffects = 0;
158 let Inst{27-21} = 0b0111110;
159 let Inst{20-16} = Rs;
161 let Inst{7-5} = MinOp;
165 let isCodeGenOnly = 0 in {
166 def A4_cmpbeq : T_CMP_rrbh<"cmpb.eq", 0b110, 1>;
167 def A4_cmpbgt : T_CMP_rrbh<"cmpb.gt", 0b010, 0>;
168 def A4_cmpbgtu : T_CMP_rrbh<"cmpb.gtu", 0b111, 0>;
169 def A4_cmpheq : T_CMP_rrbh<"cmph.eq", 0b011, 1>;
170 def A4_cmphgt : T_CMP_rrbh<"cmph.gt", 0b100, 0>;
171 def A4_cmphgtu : T_CMP_rrbh<"cmph.gtu", 0b101, 0>;
174 class T_CMP_ribh<string mnemonic, bits<2> MajOp, bit IsHalf, bit IsComm,
175 Operand ImmType, bit IsImmExt, bit IsImmSigned, int ImmBits>
176 : ALU64Inst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, ImmType:$Imm),
177 "$Pd = "#mnemonic#"($Rs, #$Imm)", [], "", ALU64_tc_2early_SLOT23>,
179 let validSubTargets = HasV4SubT;
180 let InputType = "imm";
181 let CextOpcode = mnemonic;
183 let isCommutable = IsComm;
184 let hasSideEffects = 0;
185 let isExtendable = IsImmExt;
186 let opExtendable = !if (IsImmExt, 2, 0);
187 let isExtentSigned = IsImmSigned;
188 let opExtentBits = ImmBits;
195 let Inst{27-24} = 0b1101;
196 let Inst{22-21} = MajOp;
197 let Inst{20-16} = Rs;
198 let Inst{12-5} = Imm;
200 let Inst{3} = IsHalf;
204 let isCodeGenOnly = 0 in {
205 def A4_cmpbeqi : T_CMP_ribh<"cmpb.eq", 0b00, 0, 1, u8Imm, 0, 0, 8>;
206 def A4_cmpbgti : T_CMP_ribh<"cmpb.gt", 0b01, 0, 0, s8Imm, 0, 1, 8>;
207 def A4_cmpbgtui : T_CMP_ribh<"cmpb.gtu", 0b10, 0, 0, u7Ext, 1, 0, 7>;
208 def A4_cmpheqi : T_CMP_ribh<"cmph.eq", 0b00, 1, 1, s8Ext, 1, 1, 8>;
209 def A4_cmphgti : T_CMP_ribh<"cmph.gt", 0b01, 1, 0, s8Ext, 1, 1, 8>;
210 def A4_cmphgtui : T_CMP_ribh<"cmph.gtu", 0b10, 1, 0, u7Ext, 1, 0, 7>;
212 class T_RCMP_EQ_ri<string mnemonic, bit IsNeg>
213 : ALU32_ri<(outs IntRegs:$Rd), (ins IntRegs:$Rs, s8Ext:$s8),
214 "$Rd = "#mnemonic#"($Rs, #$s8)", [], "", ALU32_2op_tc_1_SLOT0123>,
216 let validSubTargets = HasV4SubT;
217 let InputType = "imm";
218 let CextOpcode = !if (IsNeg, "!rcmp.eq", "rcmp.eq");
219 let isExtendable = 1;
220 let opExtendable = 2;
221 let isExtentSigned = 1;
222 let opExtentBits = 8;
230 let Inst{27-24} = 0b0011;
232 let Inst{21} = IsNeg;
233 let Inst{20-16} = Rs;
239 let isCodeGenOnly = 0 in {
240 def A4_rcmpeqi : T_RCMP_EQ_ri<"cmp.eq", 0>;
241 def A4_rcmpneqi : T_RCMP_EQ_ri<"!cmp.eq", 1>;
244 def: Pat<(i32 (zext (i1 (seteq (i32 IntRegs:$Rs), s8ExtPred:$s8)))),
245 (A4_rcmpeqi IntRegs:$Rs, s8ExtPred:$s8)>;
246 def: Pat<(i32 (zext (i1 (setne (i32 IntRegs:$Rs), s8ExtPred:$s8)))),
247 (A4_rcmpneqi IntRegs:$Rs, s8ExtPred:$s8)>;
249 // Preserve the S2_tstbit_r generation
250 def: Pat<(i32 (zext (i1 (setne (i32 (and (i32 (shl 1, (i32 IntRegs:$src2))),
251 (i32 IntRegs:$src1))), 0)))),
252 (C2_muxii (S2_tstbit_r IntRegs:$src1, IntRegs:$src2), 1, 0)>;
255 //===----------------------------------------------------------------------===//
257 //===----------------------------------------------------------------------===//
260 //===----------------------------------------------------------------------===//
262 //===----------------------------------------------------------------------===//
264 // Combine a word and an immediate into a register pair.
265 let hasSideEffects = 0, isExtentSigned = 1, isExtendable = 1,
267 class T_Combine1 <bits<2> MajOp, dag ins, string AsmStr>
268 : ALU32Inst <(outs DoubleRegs:$Rdd), ins, AsmStr> {
274 let Inst{27-24} = 0b0011;
275 let Inst{22-21} = MajOp;
276 let Inst{20-16} = Rs;
282 let opExtendable = 2, isCodeGenOnly = 0 in
283 def A4_combineri : T_Combine1<0b00, (ins IntRegs:$Rs, s8Ext:$s8),
284 "$Rdd = combine($Rs, #$s8)">;
286 let opExtendable = 1, isCodeGenOnly = 0 in
287 def A4_combineir : T_Combine1<0b01, (ins s8Ext:$s8, IntRegs:$Rs),
288 "$Rdd = combine(#$s8, $Rs)">;
290 def HexagonWrapperCombineRI_V4 :
291 SDNode<"HexagonISD::WrapperCombineRI_V4", SDTHexagonI64I32I32>;
292 def HexagonWrapperCombineIR_V4 :
293 SDNode<"HexagonISD::WrapperCombineIR_V4", SDTHexagonI64I32I32>;
295 def : Pat <(HexagonWrapperCombineRI_V4 IntRegs:$r, s8ExtPred:$i),
296 (A4_combineri IntRegs:$r, s8ExtPred:$i)>,
299 def : Pat <(HexagonWrapperCombineIR_V4 s8ExtPred:$i, IntRegs:$r),
300 (A4_combineir s8ExtPred:$i, IntRegs:$r)>,
303 // A4_combineii: Set two small immediates.
304 let hasSideEffects = 0, isExtendable = 1, opExtentBits = 6, opExtendable = 2 in
305 def A4_combineii: ALU32Inst<(outs DoubleRegs:$Rdd), (ins s8Imm:$s8, u6Ext:$U6),
306 "$Rdd = combine(#$s8, #$U6)"> {
312 let Inst{27-23} = 0b11001;
313 let Inst{20-16} = U6{5-1};
314 let Inst{13} = U6{0};
319 //===----------------------------------------------------------------------===//
321 //===----------------------------------------------------------------------===//
323 //===----------------------------------------------------------------------===//
325 //===----------------------------------------------------------------------===//
326 //===----------------------------------------------------------------------===//
327 // Template class for load instructions with Absolute set addressing mode.
328 //===----------------------------------------------------------------------===//
329 let isExtended = 1, opExtendable = 2, hasSideEffects = 0,
330 validSubTargets = HasV4SubT, addrMode = AbsoluteSet in
331 class T_LD_abs_set<string mnemonic, RegisterClass RC>:
332 LDInst2<(outs RC:$dst1, IntRegs:$dst2),
333 (ins u0AlwaysExt:$addr),
334 "$dst1 = "#mnemonic#"($dst2=##$addr)",
338 def LDrid_abs_set_V4 : T_LD_abs_set <"memd", DoubleRegs>;
339 def LDrib_abs_set_V4 : T_LD_abs_set <"memb", IntRegs>;
340 def LDriub_abs_set_V4 : T_LD_abs_set <"memub", IntRegs>;
341 def LDrih_abs_set_V4 : T_LD_abs_set <"memh", IntRegs>;
342 def LDriw_abs_set_V4 : T_LD_abs_set <"memw", IntRegs>;
343 def LDriuh_abs_set_V4 : T_LD_abs_set <"memuh", IntRegs>;
345 //===----------------------------------------------------------------------===//
346 // Template classes for the non-predicated load instructions with
347 // base + register offset addressing mode
348 //===----------------------------------------------------------------------===//
349 class T_load_rr <string mnemonic, RegisterClass RC, bits<3> MajOp>:
350 LDInst<(outs RC:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$u2),
351 "$dst = "#mnemonic#"($src1 + $src2<<#$u2)",
352 [], "", V4LDST_tc_ld_SLOT01>, ImmRegShl, AddrModeRel {
360 let Inst{27-24} = 0b1010;
361 let Inst{23-21} = MajOp;
362 let Inst{20-16} = src1;
363 let Inst{12-8} = src2;
364 let Inst{13} = u2{1};
369 //===----------------------------------------------------------------------===//
370 // Template classes for the predicated load instructions with
371 // base + register offset addressing mode
372 //===----------------------------------------------------------------------===//
373 let isPredicated = 1 in
374 class T_pload_rr <string mnemonic, RegisterClass RC, bits<3> MajOp,
375 bit isNot, bit isPredNew>:
376 LDInst <(outs RC:$dst),
377 (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$u2),
378 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
379 ") ")#"$dst = "#mnemonic#"($src2+$src3<<#$u2)",
380 [], "", V4LDST_tc_ld_SLOT01>, AddrModeRel {
387 let isPredicatedFalse = isNot;
388 let isPredicatedNew = isPredNew;
392 let Inst{27-26} = 0b00;
393 let Inst{25} = isPredNew;
394 let Inst{24} = isNot;
395 let Inst{23-21} = MajOp;
396 let Inst{20-16} = src2;
397 let Inst{12-8} = src3;
398 let Inst{13} = u2{1};
400 let Inst{6-5} = src1;
404 //===----------------------------------------------------------------------===//
405 // multiclass for load instructions with base + register offset
407 //===----------------------------------------------------------------------===//
408 let hasSideEffects = 0, addrMode = BaseRegOffset in
409 multiclass ld_idxd_shl <string mnemonic, string CextOp, RegisterClass RC,
411 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl,
412 InputType = "reg" in {
413 let isPredicable = 1 in
414 def L4_#NAME#_rr : T_load_rr <mnemonic, RC, MajOp>;
417 def L4_p#NAME#t_rr : T_pload_rr <mnemonic, RC, MajOp, 0, 0>;
418 def L4_p#NAME#f_rr : T_pload_rr <mnemonic, RC, MajOp, 1, 0>;
421 def L4_p#NAME#tnew_rr : T_pload_rr <mnemonic, RC, MajOp, 0, 1>;
422 def L4_p#NAME#fnew_rr : T_pload_rr <mnemonic, RC, MajOp, 1, 1>;
426 let hasNewValue = 1, accessSize = ByteAccess, isCodeGenOnly = 0 in {
427 defm loadrb : ld_idxd_shl<"memb", "LDrib", IntRegs, 0b000>;
428 defm loadrub : ld_idxd_shl<"memub", "LDriub", IntRegs, 0b001>;
431 let hasNewValue = 1, accessSize = HalfWordAccess, isCodeGenOnly = 0 in {
432 defm loadrh : ld_idxd_shl<"memh", "LDrih", IntRegs, 0b010>;
433 defm loadruh : ld_idxd_shl<"memuh", "LDriuh", IntRegs, 0b011>;
436 let hasNewValue = 1, accessSize = WordAccess, isCodeGenOnly = 0 in
437 defm loadri : ld_idxd_shl<"memw", "LDriw", IntRegs, 0b100>;
439 let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
440 defm loadrd : ld_idxd_shl<"memd", "LDrid", DoubleRegs, 0b110>;
442 // 'def pats' for load instructions with base + register offset and non-zero
443 // immediate value. Immediate value is used to left-shift the second
445 let AddedComplexity = 40 in {
446 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1,
447 (shl IntRegs:$src2, u2ImmPred:$offset)))),
448 (L4_loadrb_rr IntRegs:$src1,
449 IntRegs:$src2, u2ImmPred:$offset)>,
452 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1,
453 (shl IntRegs:$src2, u2ImmPred:$offset)))),
454 (L4_loadrub_rr IntRegs:$src1,
455 IntRegs:$src2, u2ImmPred:$offset)>,
458 def : Pat <(i32 (extloadi8 (add IntRegs:$src1,
459 (shl IntRegs:$src2, u2ImmPred:$offset)))),
460 (L4_loadrub_rr IntRegs:$src1,
461 IntRegs:$src2, u2ImmPred:$offset)>,
464 def : Pat <(i32 (sextloadi16 (add IntRegs:$src1,
465 (shl IntRegs:$src2, u2ImmPred:$offset)))),
466 (L4_loadrh_rr IntRegs:$src1,
467 IntRegs:$src2, u2ImmPred:$offset)>,
470 def : Pat <(i32 (zextloadi16 (add IntRegs:$src1,
471 (shl IntRegs:$src2, u2ImmPred:$offset)))),
472 (L4_loadruh_rr IntRegs:$src1,
473 IntRegs:$src2, u2ImmPred:$offset)>,
476 def : Pat <(i32 (extloadi16 (add IntRegs:$src1,
477 (shl IntRegs:$src2, u2ImmPred:$offset)))),
478 (L4_loadruh_rr IntRegs:$src1,
479 IntRegs:$src2, u2ImmPred:$offset)>,
482 def : Pat <(i32 (load (add IntRegs:$src1,
483 (shl IntRegs:$src2, u2ImmPred:$offset)))),
484 (L4_loadri_rr IntRegs:$src1,
485 IntRegs:$src2, u2ImmPred:$offset)>,
488 def : Pat <(i64 (load (add IntRegs:$src1,
489 (shl IntRegs:$src2, u2ImmPred:$offset)))),
490 (L4_loadrd_rr IntRegs:$src1,
491 IntRegs:$src2, u2ImmPred:$offset)>,
496 // 'def pats' for load instruction base + register offset and
497 // zero immediate value.
498 let AddedComplexity = 10 in {
499 def : Pat <(i64 (load (add IntRegs:$src1, IntRegs:$src2))),
500 (L4_loadrd_rr IntRegs:$src1, IntRegs:$src2, 0)>,
503 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1, IntRegs:$src2))),
504 (L4_loadrb_rr IntRegs:$src1, IntRegs:$src2, 0)>,
507 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1, IntRegs:$src2))),
508 (L4_loadrub_rr IntRegs:$src1, IntRegs:$src2, 0)>,
511 def : Pat <(i32 (extloadi8 (add IntRegs:$src1, IntRegs:$src2))),
512 (L4_loadrub_rr IntRegs:$src1, IntRegs:$src2, 0)>,
515 def : Pat <(i32 (sextloadi16 (add IntRegs:$src1, IntRegs:$src2))),
516 (L4_loadrh_rr IntRegs:$src1, IntRegs:$src2, 0)>,
519 def : Pat <(i32 (zextloadi16 (add IntRegs:$src1, IntRegs:$src2))),
520 (L4_loadruh_rr IntRegs:$src1, IntRegs:$src2, 0)>,
523 def : Pat <(i32 (extloadi16 (add IntRegs:$src1, IntRegs:$src2))),
524 (L4_loadruh_rr IntRegs:$src1, IntRegs:$src2, 0)>,
527 def : Pat <(i32 (load (add IntRegs:$src1, IntRegs:$src2))),
528 (L4_loadri_rr IntRegs:$src1, IntRegs:$src2, 0)>,
533 def : Pat <(i64 (zext (i1 PredRegs:$src1))),
534 (i64 (A4_combineir 0, (C2_muxii (i1 PredRegs:$src1), 1, 0)))>,
538 def : Pat <(i64 (zext (i32 IntRegs:$src1))),
539 (i64 (A4_combineir 0, (i32 IntRegs:$src1)))>,
542 def: Pat <(i64 (zextloadi8 ADDRriS11_0:$src1)),
543 (i64 (A4_combineir 0, (L2_loadrub_io AddrFI:$src1, 0)))>,
546 let AddedComplexity = 20 in
547 def: Pat <(i64 (zextloadi8 (add (i32 IntRegs:$src1),
548 s11_0ExtPred:$offset))),
549 (i64 (A4_combineir 0, (L2_loadrub_io IntRegs:$src1,
550 s11_0ExtPred:$offset)))>,
554 def: Pat <(i64 (zextloadi1 ADDRriS11_0:$src1)),
555 (i64 (A4_combineir 0, (L2_loadrub_io AddrFI:$src1, 0)))>,
558 let AddedComplexity = 20 in
559 def: Pat <(i64 (zextloadi1 (add (i32 IntRegs:$src1),
560 s11_0ExtPred:$offset))),
561 (i64 (A4_combineir 0, (L2_loadrub_io IntRegs:$src1,
562 s11_0ExtPred:$offset)))>,
566 def: Pat <(i64 (zextloadi16 ADDRriS11_1:$src1)),
567 (i64 (A4_combineir 0, (L2_loadruh_io AddrFI:$src1, 0)))>,
570 let AddedComplexity = 20 in
571 def: Pat <(i64 (zextloadi16 (add (i32 IntRegs:$src1),
572 s11_1ExtPred:$offset))),
573 (i64 (A4_combineir 0, (L2_loadruh_io IntRegs:$src1,
574 s11_1ExtPred:$offset)))>,
578 def: Pat <(i64 (extloadi16 ADDRriS11_2:$src1)),
579 (i64 (A4_combineir 0, (L2_loadrh_io AddrFI:$src1, 0)))>,
582 let AddedComplexity = 20 in
583 def: Pat <(i64 (extloadi16 (add (i32 IntRegs:$src1),
584 s11_1ExtPred:$offset))),
585 (i64 (A4_combineir 0, (L2_loadrh_io IntRegs:$src1,
586 s11_1ExtPred:$offset)))>,
590 def: Pat <(i64 (zextloadi32 ADDRriS11_2:$src1)),
591 (i64 (A4_combineir 0, (L2_loadri_io AddrFI:$src1, 0)))>,
594 let AddedComplexity = 100 in
595 def: Pat <(i64 (zextloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
596 (i64 (A4_combineir 0, (L2_loadri_io IntRegs:$src1,
597 s11_2ExtPred:$offset)))>,
601 def: Pat <(i64 (extloadi32 ADDRriS11_2:$src1)),
602 (i64 (A4_combineir 0, (L2_loadri_io AddrFI:$src1, 0)))>,
605 let AddedComplexity = 100 in
606 def: Pat <(i64 (extloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
607 (i64 (A4_combineir 0, (L2_loadri_io IntRegs:$src1,
608 s11_2ExtPred:$offset)))>,
613 //===----------------------------------------------------------------------===//
615 //===----------------------------------------------------------------------===//
617 //===----------------------------------------------------------------------===//
619 //===----------------------------------------------------------------------===//
621 //===----------------------------------------------------------------------===//
622 // Template class for store instructions with Absolute set addressing mode.
623 //===----------------------------------------------------------------------===//
624 let isExtended = 1, opExtendable = 2, validSubTargets = HasV4SubT,
625 addrMode = AbsoluteSet in
626 class T_ST_abs_set<string mnemonic, RegisterClass RC>:
627 STInst2<(outs IntRegs:$dst1),
628 (ins RC:$src1, u0AlwaysExt:$src2),
629 mnemonic#"($dst1=##$src2) = $src1",
633 def STrid_abs_set_V4 : T_ST_abs_set <"memd", DoubleRegs>;
634 def STrib_abs_set_V4 : T_ST_abs_set <"memb", IntRegs>;
635 def STrih_abs_set_V4 : T_ST_abs_set <"memh", IntRegs>;
636 def STriw_abs_set_V4 : T_ST_abs_set <"memw", IntRegs>;
638 //===----------------------------------------------------------------------===//
639 // Template classes for the non-predicated store instructions with
640 // base + register offset addressing mode
641 //===----------------------------------------------------------------------===//
642 let isPredicable = 1 in
643 class T_store_rr <string mnemonic, RegisterClass RC, bits<3> MajOp, bit isH>
644 : STInst < (outs ), (ins IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, RC:$Rt),
645 mnemonic#"($Rs + $Ru<<#$u2) = $Rt"#!if(isH, ".h",""),
646 [],"",V4LDST_tc_st_SLOT01>, ImmRegShl, AddrModeRel {
655 let Inst{27-24} = 0b1011;
656 let Inst{23-21} = MajOp;
657 let Inst{20-16} = Rs;
659 let Inst{13} = u2{1};
664 //===----------------------------------------------------------------------===//
665 // Template classes for the predicated store instructions with
666 // base + register offset addressing mode
667 //===----------------------------------------------------------------------===//
668 let isPredicated = 1 in
669 class T_pstore_rr <string mnemonic, RegisterClass RC, bits<3> MajOp,
670 bit isNot, bit isPredNew, bit isH>
672 (ins PredRegs:$Pv, IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, RC:$Rt),
674 !if(isNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ",
675 ") ")#mnemonic#"($Rs+$Ru<<#$u2) = $Rt"#!if(isH, ".h",""),
676 [], "", V4LDST_tc_st_SLOT01> , AddrModeRel{
683 let isPredicatedFalse = isNot;
684 let isPredicatedNew = isPredNew;
688 let Inst{27-26} = 0b01;
689 let Inst{25} = isPredNew;
690 let Inst{24} = isNot;
691 let Inst{23-21} = MajOp;
692 let Inst{20-16} = Rs;
694 let Inst{13} = u2{1};
700 //===----------------------------------------------------------------------===//
701 // Template classes for the new-value store instructions with
702 // base + register offset addressing mode
703 //===----------------------------------------------------------------------===//
704 let isPredicable = 1, isNewValue = 1, opNewValue = 3 in
705 class T_store_new_rr <string mnemonic, bits<2> MajOp> :
706 NVInst < (outs ), (ins IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, IntRegs:$Nt),
707 mnemonic#"($Rs + $Ru<<#$u2) = $Nt.new",
708 [],"",V4LDST_tc_st_SLOT0>, ImmRegShl, AddrModeRel {
717 let Inst{27-21} = 0b1011101;
718 let Inst{20-16} = Rs;
720 let Inst{13} = u2{1};
722 let Inst{4-3} = MajOp;
726 //===----------------------------------------------------------------------===//
727 // Template classes for the predicated new-value store instructions with
728 // base + register offset addressing mode
729 //===----------------------------------------------------------------------===//
730 let isPredicated = 1, isNewValue = 1, opNewValue = 4 in
731 class T_pstore_new_rr <string mnemonic, bits<2> MajOp, bit isNot, bit isPredNew>
733 (ins PredRegs:$Pv, IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, IntRegs:$Nt),
734 !if(isNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ",
735 ") ")#mnemonic#"($Rs+$Ru<<#$u2) = $Nt.new",
736 [], "", V4LDST_tc_st_SLOT0>, AddrModeRel {
743 let isPredicatedFalse = isNot;
744 let isPredicatedNew = isPredNew;
747 let Inst{27-26} = 0b01;
748 let Inst{25} = isPredNew;
749 let Inst{24} = isNot;
750 let Inst{23-21} = 0b101;
751 let Inst{20-16} = Rs;
753 let Inst{13} = u2{1};
756 let Inst{4-3} = MajOp;
760 //===----------------------------------------------------------------------===//
761 // multiclass for store instructions with base + register offset addressing
763 //===----------------------------------------------------------------------===//
764 let isNVStorable = 1 in
765 multiclass ST_Idxd_shl<string mnemonic, string CextOp, RegisterClass RC,
766 bits<3> MajOp, bit isH = 0> {
767 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
768 def S4_#NAME#_rr : T_store_rr <mnemonic, RC, MajOp, isH>;
771 def S4_p#NAME#t_rr : T_pstore_rr <mnemonic, RC, MajOp, 0, 0, isH>;
772 def S4_p#NAME#f_rr : T_pstore_rr <mnemonic, RC, MajOp, 1, 0, isH>;
775 def S4_p#NAME#tnew_rr : T_pstore_rr <mnemonic, RC, MajOp, 0, 1, isH>;
776 def S4_p#NAME#fnew_rr : T_pstore_rr <mnemonic, RC, MajOp, 1, 1, isH>;
780 //===----------------------------------------------------------------------===//
781 // multiclass for new-value store instructions with base + register offset
783 //===----------------------------------------------------------------------===//
784 let mayStore = 1, isNVStore = 1 in
785 multiclass ST_Idxd_shl_nv <string mnemonic, string CextOp, RegisterClass RC,
787 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
788 def S4_#NAME#new_rr : T_store_new_rr<mnemonic, MajOp>;
791 def S4_p#NAME#newt_rr : T_pstore_new_rr <mnemonic, MajOp, 0, 0>;
792 def S4_p#NAME#newf_rr : T_pstore_new_rr <mnemonic, MajOp, 1, 0>;
795 def S4_p#NAME#newtnew_rr : T_pstore_new_rr <mnemonic, MajOp, 0, 1>;
796 def S4_p#NAME#newfnew_rr : T_pstore_new_rr <mnemonic, MajOp, 1, 1>;
800 let addrMode = BaseRegOffset, InputType = "reg", hasSideEffects = 0,
801 isCodeGenOnly = 0 in {
802 let accessSize = ByteAccess in
803 defm storerb: ST_Idxd_shl<"memb", "STrib", IntRegs, 0b000>,
804 ST_Idxd_shl_nv<"memb", "STrib", IntRegs, 0b00>;
806 let accessSize = HalfWordAccess in
807 defm storerh: ST_Idxd_shl<"memh", "STrih", IntRegs, 0b010>,
808 ST_Idxd_shl_nv<"memh", "STrih", IntRegs, 0b01>;
810 let accessSize = WordAccess in
811 defm storeri: ST_Idxd_shl<"memw", "STriw", IntRegs, 0b100>,
812 ST_Idxd_shl_nv<"memw", "STriw", IntRegs, 0b10>;
814 let isNVStorable = 0, accessSize = DoubleWordAccess in
815 defm storerd: ST_Idxd_shl<"memd", "STrid", DoubleRegs, 0b110>;
817 let isNVStorable = 0, accessSize = HalfWordAccess in
818 defm storerf: ST_Idxd_shl<"memh", "STrif", IntRegs, 0b011, 1>;
821 let Predicates = [HasV4T], AddedComplexity = 10 in {
822 def : Pat<(truncstorei8 (i32 IntRegs:$src4),
823 (add IntRegs:$src1, (shl IntRegs:$src2,
825 (S4_storerb_rr IntRegs:$src1, IntRegs:$src2,
826 u2ImmPred:$src3, IntRegs:$src4)>;
828 def : Pat<(truncstorei16 (i32 IntRegs:$src4),
829 (add IntRegs:$src1, (shl IntRegs:$src2,
831 (S4_storerh_rr IntRegs:$src1, IntRegs:$src2,
832 u2ImmPred:$src3, IntRegs:$src4)>;
834 def : Pat<(store (i32 IntRegs:$src4),
835 (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))),
836 (S4_storeri_rr IntRegs:$src1, IntRegs:$src2,
837 u2ImmPred:$src3, IntRegs:$src4)>;
839 def : Pat<(store (i64 DoubleRegs:$src4),
840 (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))),
841 (S4_storerd_rr IntRegs:$src1, IntRegs:$src2,
842 u2ImmPred:$src3, DoubleRegs:$src4)>;
845 let isExtended = 1, opExtendable = 2 in
846 class T_ST_LongOff <string mnemonic, PatFrag stOp, RegisterClass RC, ValueType VT> :
848 (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, RC:$src4),
849 mnemonic#"($src1<<#$src2+##$src3) = $src4",
850 [(stOp (VT RC:$src4),
851 (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2),
852 u0AlwaysExtPred:$src3))]>,
855 let isExtended = 1, opExtendable = 2, mayStore = 1, isNVStore = 1 in
856 class T_ST_LongOff_nv <string mnemonic> :
858 (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, IntRegs:$src4),
859 mnemonic#"($src1<<#$src2+##$src3) = $src4.new",
863 multiclass ST_LongOff <string mnemonic, string BaseOp, PatFrag stOp> {
864 let BaseOpcode = BaseOp#"_shl" in {
865 let isNVStorable = 1 in
866 def NAME#_V4 : T_ST_LongOff<mnemonic, stOp, IntRegs, i32>;
868 def NAME#_nv_V4 : T_ST_LongOff_nv<mnemonic>;
872 let AddedComplexity = 10, validSubTargets = HasV4SubT in {
873 def STrid_shl_V4 : T_ST_LongOff<"memd", store, DoubleRegs, i64>;
874 defm STrib_shl : ST_LongOff <"memb", "STrib", truncstorei8>, NewValueRel;
875 defm STrih_shl : ST_LongOff <"memh", "Strih", truncstorei16>, NewValueRel;
876 defm STriw_shl : ST_LongOff <"memw", "STriw", store>, NewValueRel;
879 let AddedComplexity = 40 in
880 multiclass T_ST_LOff_Pats <InstHexagon I, RegisterClass RC, ValueType VT,
882 def : Pat<(stOp (VT RC:$src4),
883 (add (shl IntRegs:$src1, u2ImmPred:$src2),
884 (NumUsesBelowThresCONST32 tglobaladdr:$src3))),
885 (I IntRegs:$src1, u2ImmPred:$src2, tglobaladdr:$src3, RC:$src4)>;
887 def : Pat<(stOp (VT RC:$src4),
889 (NumUsesBelowThresCONST32 tglobaladdr:$src3))),
890 (I IntRegs:$src1, 0, tglobaladdr:$src3, RC:$src4)>;
893 defm : T_ST_LOff_Pats<STrid_shl_V4, DoubleRegs, i64, store>;
894 defm : T_ST_LOff_Pats<STriw_shl_V4, IntRegs, i32, store>;
895 defm : T_ST_LOff_Pats<STrib_shl_V4, IntRegs, i32, truncstorei8>;
896 defm : T_ST_LOff_Pats<STrih_shl_V4, IntRegs, i32, truncstorei16>;
898 // memd(Rx++#s4:3)=Rtt
899 // memd(Rx++#s4:3:circ(Mu))=Rtt
900 // memd(Rx++I:circ(Mu))=Rtt
902 // memd(Rx++Mu:brev)=Rtt
903 // memd(gp+#u16:3)=Rtt
905 // Store doubleword conditionally.
906 // if ([!]Pv[.new]) memd(#u6)=Rtt
907 // TODO: needs to be implemented.
909 //===----------------------------------------------------------------------===//
911 //===----------------------------------------------------------------------===//
912 let isPredicable = 1, isExtendable = 1, isExtentSigned = 1, opExtentBits = 8,
914 class T_StoreImm <string mnemonic, Operand OffsetOp, bits<2> MajOp >
915 : STInst <(outs ), (ins IntRegs:$Rs, OffsetOp:$offset, s8Ext:$S8),
916 mnemonic#"($Rs+#$offset)=#$S8",
917 [], "", V4LDST_tc_st_SLOT01>,
918 ImmRegRel, PredNewRel {
924 string OffsetOpStr = !cast<string>(OffsetOp);
925 let offsetBits = !if (!eq(OffsetOpStr, "u6_2Imm"), offset{7-2},
926 !if (!eq(OffsetOpStr, "u6_1Imm"), offset{6-1},
927 /* u6_0Imm */ offset{5-0}));
931 let Inst{27-25} = 0b110;
932 let Inst{22-21} = MajOp;
933 let Inst{20-16} = Rs;
934 let Inst{12-7} = offsetBits;
935 let Inst{13} = S8{7};
936 let Inst{6-0} = S8{6-0};
939 let isPredicated = 1, isExtendable = 1, isExtentSigned = 1, opExtentBits = 6,
941 class T_StoreImm_pred <string mnemonic, Operand OffsetOp, bits<2> MajOp,
942 bit isPredNot, bit isPredNew >
944 (ins PredRegs:$Pv, IntRegs:$Rs, OffsetOp:$offset, s6Ext:$S6),
945 !if(isPredNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ",
946 ") ")#mnemonic#"($Rs+#$offset)=#$S6",
947 [], "", V4LDST_tc_st_SLOT01>,
948 ImmRegRel, PredNewRel {
955 string OffsetOpStr = !cast<string>(OffsetOp);
956 let offsetBits = !if (!eq(OffsetOpStr, "u6_2Imm"), offset{7-2},
957 !if (!eq(OffsetOpStr, "u6_1Imm"), offset{6-1},
958 /* u6_0Imm */ offset{5-0}));
959 let isPredicatedNew = isPredNew;
960 let isPredicatedFalse = isPredNot;
964 let Inst{27-25} = 0b100;
965 let Inst{24} = isPredNew;
966 let Inst{23} = isPredNot;
967 let Inst{22-21} = MajOp;
968 let Inst{20-16} = Rs;
969 let Inst{13} = S6{5};
970 let Inst{12-7} = offsetBits;
972 let Inst{4-0} = S6{4-0};
976 //===----------------------------------------------------------------------===//
977 // multiclass for store instructions with base + immediate offset
978 // addressing mode and immediate stored value.
979 // mem[bhw](Rx++#s4:3)=#s8
980 // if ([!]Pv[.new]) mem[bhw](Rx++#s4:3)=#s6
981 //===----------------------------------------------------------------------===//
983 multiclass ST_Imm_Pred <string mnemonic, Operand OffsetOp, bits<2> MajOp,
985 def _io : T_StoreImm_pred <mnemonic, OffsetOp, MajOp, PredNot, 0>;
987 def new_io : T_StoreImm_pred <mnemonic, OffsetOp, MajOp, PredNot, 1>;
990 multiclass ST_Imm <string mnemonic, string CextOp, Operand OffsetOp,
992 let CextOpcode = CextOp, BaseOpcode = CextOp#_imm in {
993 def _io : T_StoreImm <mnemonic, OffsetOp, MajOp>;
995 defm t : ST_Imm_Pred <mnemonic, OffsetOp, MajOp, 0>;
996 defm f : ST_Imm_Pred <mnemonic, OffsetOp, MajOp, 1>;
1000 let hasSideEffects = 0, validSubTargets = HasV4SubT, addrMode = BaseImmOffset,
1001 InputType = "imm", isCodeGenOnly = 0 in {
1002 let accessSize = ByteAccess in
1003 defm S4_storeirb : ST_Imm<"memb", "STrib", u6_0Imm, 0b00>;
1005 let accessSize = HalfWordAccess in
1006 defm S4_storeirh : ST_Imm<"memh", "STrih", u6_1Imm, 0b01>;
1008 let accessSize = WordAccess in
1009 defm S4_storeiri : ST_Imm<"memw", "STriw", u6_2Imm, 0b10>;
1012 let Predicates = [HasV4T], AddedComplexity = 10 in {
1013 def: Pat<(truncstorei8 s8ExtPred:$src3, (add IntRegs:$src1, u6_0ImmPred:$src2)),
1014 (S4_storeirb_io IntRegs:$src1, u6_0ImmPred:$src2, s8ExtPred:$src3)>;
1016 def: Pat<(truncstorei16 s8ExtPred:$src3, (add IntRegs:$src1,
1017 u6_1ImmPred:$src2)),
1018 (S4_storeirh_io IntRegs:$src1, u6_1ImmPred:$src2, s8ExtPred:$src3)>;
1020 def: Pat<(store s8ExtPred:$src3, (add IntRegs:$src1, u6_2ImmPred:$src2)),
1021 (S4_storeiri_io IntRegs:$src1, u6_2ImmPred:$src2, s8ExtPred:$src3)>;
1024 let AddedComplexity = 6 in
1025 def : Pat <(truncstorei8 s8ExtPred:$src2, (i32 IntRegs:$src1)),
1026 (S4_storeirb_io IntRegs:$src1, 0, s8ExtPred:$src2)>,
1029 // memb(Rx++#s4:0:circ(Mu))=Rt
1030 // memb(Rx++I:circ(Mu))=Rt
1032 // memb(Rx++Mu:brev)=Rt
1033 // memb(gp+#u16:0)=Rt
1037 // TODO: needs to be implemented
1038 // memh(Re=#U6)=Rt.H
1039 // memh(Rs+#s11:1)=Rt.H
1040 let AddedComplexity = 6 in
1041 def : Pat <(truncstorei16 s8ExtPred:$src2, (i32 IntRegs:$src1)),
1042 (S4_storeirh_io IntRegs:$src1, 0, s8ExtPred:$src2)>,
1045 // memh(Rs+Ru<<#u2)=Rt.H
1046 // TODO: needs to be implemented.
1048 // memh(Ru<<#u2+#U6)=Rt.H
1049 // memh(Rx++#s4:1:circ(Mu))=Rt.H
1050 // memh(Rx++#s4:1:circ(Mu))=Rt
1051 // memh(Rx++I:circ(Mu))=Rt.H
1052 // memh(Rx++I:circ(Mu))=Rt
1053 // memh(Rx++Mu)=Rt.H
1055 // memh(Rx++Mu:brev)=Rt.H
1056 // memh(Rx++Mu:brev)=Rt
1057 // memh(gp+#u16:1)=Rt
1058 // if ([!]Pv[.new]) memh(#u6)=Rt.H
1059 // if ([!]Pv[.new]) memh(#u6)=Rt
1062 // if ([!]Pv[.new]) memh(Rs+#u6:1)=Rt.H
1063 // TODO: needs to be implemented.
1065 // if ([!]Pv[.new]) memh(Rx++#s4:1)=Rt.H
1066 // TODO: Needs to be implemented.
1070 // TODO: Needs to be implemented.
1073 let hasSideEffects = 0 in
1074 def STriw_pred_V4 : STInst2<(outs),
1075 (ins MEMri:$addr, PredRegs:$src1),
1076 "Error; should not emit",
1080 let AddedComplexity = 6 in
1081 def : Pat <(store s8ExtPred:$src2, (i32 IntRegs:$src1)),
1082 (S4_storeiri_io IntRegs:$src1, 0, s8ExtPred:$src2)>,
1085 // memw(Rx++#s4:2)=Rt
1086 // memw(Rx++#s4:2:circ(Mu))=Rt
1087 // memw(Rx++I:circ(Mu))=Rt
1089 // memw(Rx++Mu:brev)=Rt
1091 //===----------------------------------------------------------------------===
1093 //===----------------------------------------------------------------------===
1096 //===----------------------------------------------------------------------===//
1098 //===----------------------------------------------------------------------===//
1100 let opNewValue = 2, opExtendable = 1, isExtentSigned = 1, isPredicable = 1 in
1101 class T_store_io_nv <string mnemonic, RegisterClass RC,
1102 Operand ImmOp, bits<2>MajOp>
1103 : NVInst_V4 <(outs),
1104 (ins IntRegs:$src1, ImmOp:$src2, RC:$src3),
1105 mnemonic#"($src1+#$src2) = $src3.new",
1106 [],"",ST_tc_st_SLOT0> {
1108 bits<13> src2; // Actual address offset
1110 bits<11> offsetBits; // Represents offset encoding
1112 let opExtentBits = !if (!eq(mnemonic, "memb"), 11,
1113 !if (!eq(mnemonic, "memh"), 12,
1114 !if (!eq(mnemonic, "memw"), 13, 0)));
1116 let opExtentAlign = !if (!eq(mnemonic, "memb"), 0,
1117 !if (!eq(mnemonic, "memh"), 1,
1118 !if (!eq(mnemonic, "memw"), 2, 0)));
1120 let offsetBits = !if (!eq(mnemonic, "memb"), src2{10-0},
1121 !if (!eq(mnemonic, "memh"), src2{11-1},
1122 !if (!eq(mnemonic, "memw"), src2{12-2}, 0)));
1124 let IClass = 0b1010;
1127 let Inst{26-25} = offsetBits{10-9};
1128 let Inst{24-21} = 0b1101;
1129 let Inst{20-16} = src1;
1130 let Inst{13} = offsetBits{8};
1131 let Inst{12-11} = MajOp;
1132 let Inst{10-8} = src3;
1133 let Inst{7-0} = offsetBits{7-0};
1136 let opExtendable = 2, opNewValue = 3, isPredicated = 1 in
1137 class T_pstore_io_nv <string mnemonic, RegisterClass RC, Operand predImmOp,
1138 bits<2>MajOp, bit PredNot, bit isPredNew>
1139 : NVInst_V4 <(outs),
1140 (ins PredRegs:$src1, IntRegs:$src2, predImmOp:$src3, RC:$src4),
1141 !if(PredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
1142 ") ")#mnemonic#"($src2+#$src3) = $src4.new",
1143 [],"",V2LDST_tc_st_SLOT0> {
1148 bits<6> offsetBits; // Represents offset encoding
1150 let isPredicatedNew = isPredNew;
1151 let isPredicatedFalse = PredNot;
1152 let opExtentBits = !if (!eq(mnemonic, "memb"), 6,
1153 !if (!eq(mnemonic, "memh"), 7,
1154 !if (!eq(mnemonic, "memw"), 8, 0)));
1156 let opExtentAlign = !if (!eq(mnemonic, "memb"), 0,
1157 !if (!eq(mnemonic, "memh"), 1,
1158 !if (!eq(mnemonic, "memw"), 2, 0)));
1160 let offsetBits = !if (!eq(mnemonic, "memb"), src3{5-0},
1161 !if (!eq(mnemonic, "memh"), src3{6-1},
1162 !if (!eq(mnemonic, "memw"), src3{7-2}, 0)));
1164 let IClass = 0b0100;
1167 let Inst{26} = PredNot;
1168 let Inst{25} = isPredNew;
1169 let Inst{24-21} = 0b0101;
1170 let Inst{20-16} = src2;
1171 let Inst{13} = offsetBits{5};
1172 let Inst{12-11} = MajOp;
1173 let Inst{10-8} = src4;
1174 let Inst{7-3} = offsetBits{4-0};
1176 let Inst{1-0} = src1;
1179 // multiclass for new-value store instructions with base + immediate offset.
1181 let mayStore = 1, isNVStore = 1, isNewValue = 1, hasSideEffects = 0,
1183 multiclass ST_Idxd_nv<string mnemonic, string CextOp, RegisterClass RC,
1184 Operand ImmOp, Operand predImmOp, bits<2> MajOp> {
1186 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
1187 def S2_#NAME#new_io : T_store_io_nv <mnemonic, RC, ImmOp, MajOp>;
1189 def S2_p#NAME#newt_io :T_pstore_io_nv <mnemonic, RC, predImmOp, MajOp, 0, 0>;
1190 def S2_p#NAME#newf_io :T_pstore_io_nv <mnemonic, RC, predImmOp, MajOp, 1, 0>;
1192 def S4_p#NAME#newtnew_io :T_pstore_io_nv <mnemonic, RC, predImmOp,
1194 def S4_p#NAME#newfnew_io :T_pstore_io_nv <mnemonic, RC, predImmOp,
1199 let addrMode = BaseImmOffset, InputType = "imm", isCodeGenOnly = 0 in {
1200 let accessSize = ByteAccess in
1201 defm storerb: ST_Idxd_nv<"memb", "STrib", IntRegs, s11_0Ext,
1202 u6_0Ext, 0b00>, AddrModeRel;
1204 let accessSize = HalfWordAccess, opExtentAlign = 1 in
1205 defm storerh: ST_Idxd_nv<"memh", "STrih", IntRegs, s11_1Ext,
1206 u6_1Ext, 0b01>, AddrModeRel;
1208 let accessSize = WordAccess, opExtentAlign = 2 in
1209 defm storeri: ST_Idxd_nv<"memw", "STriw", IntRegs, s11_2Ext,
1210 u6_2Ext, 0b10>, AddrModeRel;
1213 //===----------------------------------------------------------------------===//
1214 // Template class for non-predicated post increment .new stores
1215 // mem[bhwd](Rx++#s4:[0123])=Nt.new
1216 //===----------------------------------------------------------------------===//
1217 let isPredicable = 1, hasSideEffects = 0, validSubTargets = HasV4SubT,
1218 addrMode = PostInc, isNVStore = 1, isNewValue = 1, opNewValue = 3 in
1219 class T_StorePI_nv <string mnemonic, Operand ImmOp, bits<2> MajOp >
1220 : NVInstPI_V4 <(outs IntRegs:$_dst_),
1221 (ins IntRegs:$src1, ImmOp:$offset, IntRegs:$src2),
1222 mnemonic#"($src1++#$offset) = $src2.new",
1223 [], "$src1 = $_dst_">,
1230 string ImmOpStr = !cast<string>(ImmOp);
1231 let offsetBits = !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
1232 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
1233 /* s4_0Imm */ offset{3-0}));
1234 let IClass = 0b1010;
1236 let Inst{27-21} = 0b1011101;
1237 let Inst{20-16} = src1;
1239 let Inst{12-11} = MajOp;
1240 let Inst{10-8} = src2;
1242 let Inst{6-3} = offsetBits;
1246 //===----------------------------------------------------------------------===//
1247 // Template class for predicated post increment .new stores
1248 // if([!]Pv[.new]) mem[bhwd](Rx++#s4:[0123])=Nt.new
1249 //===----------------------------------------------------------------------===//
1250 let isPredicated = 1, hasSideEffects = 0, validSubTargets = HasV4SubT,
1251 addrMode = PostInc, isNVStore = 1, isNewValue = 1, opNewValue = 4 in
1252 class T_StorePI_nv_pred <string mnemonic, Operand ImmOp,
1253 bits<2> MajOp, bit isPredNot, bit isPredNew >
1254 : NVInstPI_V4 <(outs IntRegs:$_dst_),
1255 (ins PredRegs:$src1, IntRegs:$src2,
1256 ImmOp:$offset, IntRegs:$src3),
1257 !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
1258 ") ")#mnemonic#"($src2++#$offset) = $src3.new",
1259 [], "$src2 = $_dst_">,
1267 string ImmOpStr = !cast<string>(ImmOp);
1268 let offsetBits = !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
1269 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
1270 /* s4_0Imm */ offset{3-0}));
1271 let isPredicatedNew = isPredNew;
1272 let isPredicatedFalse = isPredNot;
1274 let IClass = 0b1010;
1276 let Inst{27-21} = 0b1011101;
1277 let Inst{20-16} = src2;
1279 let Inst{12-11} = MajOp;
1280 let Inst{10-8} = src3;
1281 let Inst{7} = isPredNew;
1282 let Inst{6-3} = offsetBits;
1283 let Inst{2} = isPredNot;
1284 let Inst{1-0} = src1;
1287 multiclass ST_PostInc_Pred_nv<string mnemonic, Operand ImmOp,
1288 bits<2> MajOp, bit PredNot> {
1289 def _pi : T_StorePI_nv_pred <mnemonic, ImmOp, MajOp, PredNot, 0>;
1292 def new_pi : T_StorePI_nv_pred <mnemonic, ImmOp, MajOp, PredNot, 1>;
1295 multiclass ST_PostInc_nv<string mnemonic, string BaseOp, Operand ImmOp,
1297 let BaseOpcode = "POST_"#BaseOp in {
1298 def S2_#NAME#_pi : T_StorePI_nv <mnemonic, ImmOp, MajOp>;
1301 defm S2_p#NAME#t : ST_PostInc_Pred_nv <mnemonic, ImmOp, MajOp, 0>;
1302 defm S2_p#NAME#f : ST_PostInc_Pred_nv <mnemonic, ImmOp, MajOp, 1>;
1306 let accessSize = ByteAccess, isCodeGenOnly = 0 in
1307 defm storerbnew: ST_PostInc_nv <"memb", "STrib", s4_0Imm, 0b00>;
1309 let accessSize = HalfWordAccess, isCodeGenOnly = 0 in
1310 defm storerhnew: ST_PostInc_nv <"memh", "STrih", s4_1Imm, 0b01>;
1312 let accessSize = WordAccess, isCodeGenOnly = 0 in
1313 defm storerinew: ST_PostInc_nv <"memw", "STriw", s4_2Imm, 0b10>;
1315 //===----------------------------------------------------------------------===//
1316 // Template class for post increment .new stores with register offset
1317 //===----------------------------------------------------------------------===//
1318 let isNewValue = 1, mayStore = 1, isNVStore = 1, opNewValue = 3 in
1319 class T_StorePI_RegNV <string mnemonic, bits<2> MajOp, MemAccessSize AccessSz>
1320 : NVInstPI_V4 <(outs IntRegs:$_dst_),
1321 (ins IntRegs:$src1, ModRegs:$src2, IntRegs:$src3),
1322 #mnemonic#"($src1++$src2) = $src3.new",
1323 [], "$src1 = $_dst_"> {
1327 let accessSize = AccessSz;
1329 let IClass = 0b1010;
1331 let Inst{27-21} = 0b1101101;
1332 let Inst{20-16} = src1;
1333 let Inst{13} = src2;
1334 let Inst{12-11} = MajOp;
1335 let Inst{10-8} = src3;
1339 let isCodeGenOnly = 0 in {
1340 def S2_storerbnew_pr : T_StorePI_RegNV<"memb", 0b00, ByteAccess>;
1341 def S2_storerhnew_pr : T_StorePI_RegNV<"memh", 0b01, HalfWordAccess>;
1342 def S2_storerinew_pr : T_StorePI_RegNV<"memw", 0b10, WordAccess>;
1345 // memb(Rx++#s4:0:circ(Mu))=Nt.new
1346 // memb(Rx++I:circ(Mu))=Nt.new
1347 // memb(Rx++Mu)=Nt.new
1348 // memb(Rx++Mu:brev)=Nt.new
1349 // memh(Rx++#s4:1:circ(Mu))=Nt.new
1350 // memh(Rx++I:circ(Mu))=Nt.new
1351 // memh(Rx++Mu)=Nt.new
1352 // memh(Rx++Mu:brev)=Nt.new
1354 // memw(Rx++#s4:2:circ(Mu))=Nt.new
1355 // memw(Rx++I:circ(Mu))=Nt.new
1356 // memw(Rx++Mu)=Nt.new
1357 // memw(Rx++Mu:brev)=Nt.new
1359 //===----------------------------------------------------------------------===//
1361 //===----------------------------------------------------------------------===//
1363 //===----------------------------------------------------------------------===//
1365 //===----------------------------------------------------------------------===//
1367 //===----------------------------------------------------------------------===//
1368 // multiclass/template class for the new-value compare jumps with the register
1370 //===----------------------------------------------------------------------===//
1372 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11,
1373 opExtentAlign = 2 in
1374 class NVJrr_template<string mnemonic, bits<3> majOp, bit NvOpNum,
1375 bit isNegCond, bit isTak>
1377 (ins IntRegs:$src1, IntRegs:$src2, brtarget:$offset),
1378 "if ("#!if(isNegCond, "!","")#mnemonic#
1379 "($src1"#!if(!eq(NvOpNum, 0),".new, ",", ")#
1380 "$src2"#!if(!eq(NvOpNum, 1),".new))","))")#" jump:"
1381 #!if(isTak, "t","nt")#" $offset", []> {
1385 bits<3> Ns; // New-Value Operand
1386 bits<5> RegOp; // Non-New-Value Operand
1389 let isTaken = isTak;
1390 let isPredicatedFalse = isNegCond;
1391 let opNewValue{0} = NvOpNum;
1393 let Ns = !if(!eq(NvOpNum, 0), src1{2-0}, src2{2-0});
1394 let RegOp = !if(!eq(NvOpNum, 0), src2, src1);
1396 let IClass = 0b0010;
1398 let Inst{25-23} = majOp;
1399 let Inst{22} = isNegCond;
1400 let Inst{18-16} = Ns;
1401 let Inst{13} = isTak;
1402 let Inst{12-8} = RegOp;
1403 let Inst{21-20} = offset{10-9};
1404 let Inst{7-1} = offset{8-2};
1408 multiclass NVJrr_cond<string mnemonic, bits<3> majOp, bit NvOpNum,
1410 // Branch not taken:
1411 def _nt_V4: NVJrr_template<mnemonic, majOp, NvOpNum, isNegCond, 0>;
1413 def _t_V4: NVJrr_template<mnemonic, majOp, NvOpNum, isNegCond, 1>;
1416 // NvOpNum = 0 -> First Operand is a new-value Register
1417 // NvOpNum = 1 -> Second Operand is a new-value Register
1419 multiclass NVJrr_base<string mnemonic, string BaseOp, bits<3> majOp,
1421 let BaseOpcode = BaseOp#_NVJ in {
1422 defm _t_Jumpnv : NVJrr_cond<mnemonic, majOp, NvOpNum, 0>; // True cond
1423 defm _f_Jumpnv : NVJrr_cond<mnemonic, majOp, NvOpNum, 1>; // False cond
1427 // if ([!]cmp.eq(Ns.new,Rt)) jump:[n]t #r9:2
1428 // if ([!]cmp.gt(Ns.new,Rt)) jump:[n]t #r9:2
1429 // if ([!]cmp.gtu(Ns.new,Rt)) jump:[n]t #r9:2
1430 // if ([!]cmp.gt(Rt,Ns.new)) jump:[n]t #r9:2
1431 // if ([!]cmp.gtu(Rt,Ns.new)) jump:[n]t #r9:2
1433 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1,
1434 Defs = [PC], hasSideEffects = 0, validSubTargets = HasV4SubT,
1435 isCodeGenOnly = 0 in {
1436 defm CMPEQrr : NVJrr_base<"cmp.eq", "CMPEQ", 0b000, 0>, PredRel;
1437 defm CMPGTrr : NVJrr_base<"cmp.gt", "CMPGT", 0b001, 0>, PredRel;
1438 defm CMPGTUrr : NVJrr_base<"cmp.gtu", "CMPGTU", 0b010, 0>, PredRel;
1439 defm CMPLTrr : NVJrr_base<"cmp.gt", "CMPLT", 0b011, 1>, PredRel;
1440 defm CMPLTUrr : NVJrr_base<"cmp.gtu", "CMPLTU", 0b100, 1>, PredRel;
1443 //===----------------------------------------------------------------------===//
1444 // multiclass/template class for the new-value compare jumps instruction
1445 // with a register and an unsigned immediate (U5) operand.
1446 //===----------------------------------------------------------------------===//
1448 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11,
1449 opExtentAlign = 2 in
1450 class NVJri_template<string mnemonic, bits<3> majOp, bit isNegCond,
1453 (ins IntRegs:$src1, u5Imm:$src2, brtarget:$offset),
1454 "if ("#!if(isNegCond, "!","")#mnemonic#"($src1.new, #$src2)) jump:"
1455 #!if(isTak, "t","nt")#" $offset", []> {
1457 let isTaken = isTak;
1458 let isPredicatedFalse = isNegCond;
1459 let isTaken = isTak;
1465 let IClass = 0b0010;
1467 let Inst{25-23} = majOp;
1468 let Inst{22} = isNegCond;
1469 let Inst{18-16} = src1;
1470 let Inst{13} = isTak;
1471 let Inst{12-8} = src2;
1472 let Inst{21-20} = offset{10-9};
1473 let Inst{7-1} = offset{8-2};
1476 multiclass NVJri_cond<string mnemonic, bits<3> majOp, bit isNegCond> {
1477 // Branch not taken:
1478 def _nt_V4: NVJri_template<mnemonic, majOp, isNegCond, 0>;
1480 def _t_V4: NVJri_template<mnemonic, majOp, isNegCond, 1>;
1483 multiclass NVJri_base<string mnemonic, string BaseOp, bits<3> majOp> {
1484 let BaseOpcode = BaseOp#_NVJri in {
1485 defm _t_Jumpnv : NVJri_cond<mnemonic, majOp, 0>; // True Cond
1486 defm _f_Jumpnv : NVJri_cond<mnemonic, majOp, 1>; // False cond
1490 // if ([!]cmp.eq(Ns.new,#U5)) jump:[n]t #r9:2
1491 // if ([!]cmp.gt(Ns.new,#U5)) jump:[n]t #r9:2
1492 // if ([!]cmp.gtu(Ns.new,#U5)) jump:[n]t #r9:2
1494 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1,
1495 Defs = [PC], hasSideEffects = 0, validSubTargets = HasV4SubT,
1496 isCodeGenOnly = 0 in {
1497 defm CMPEQri : NVJri_base<"cmp.eq", "CMPEQ", 0b000>, PredRel;
1498 defm CMPGTri : NVJri_base<"cmp.gt", "CMPGT", 0b001>, PredRel;
1499 defm CMPGTUri : NVJri_base<"cmp.gtu", "CMPGTU", 0b010>, PredRel;
1502 //===----------------------------------------------------------------------===//
1503 // multiclass/template class for the new-value compare jumps instruction
1504 // with a register and an hardcoded 0/-1 immediate value.
1505 //===----------------------------------------------------------------------===//
1507 let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 11,
1508 opExtentAlign = 2 in
1509 class NVJ_ConstImm_template<string mnemonic, bits<3> majOp, string ImmVal,
1510 bit isNegCond, bit isTak>
1512 (ins IntRegs:$src1, brtarget:$offset),
1513 "if ("#!if(isNegCond, "!","")#mnemonic
1514 #"($src1.new, #"#ImmVal#")) jump:"
1515 #!if(isTak, "t","nt")#" $offset", []> {
1517 let isTaken = isTak;
1518 let isPredicatedFalse = isNegCond;
1519 let isTaken = isTak;
1523 let IClass = 0b0010;
1525 let Inst{25-23} = majOp;
1526 let Inst{22} = isNegCond;
1527 let Inst{18-16} = src1;
1528 let Inst{13} = isTak;
1529 let Inst{21-20} = offset{10-9};
1530 let Inst{7-1} = offset{8-2};
1533 multiclass NVJ_ConstImm_cond<string mnemonic, bits<3> majOp, string ImmVal,
1535 // Branch not taken:
1536 def _nt_V4: NVJ_ConstImm_template<mnemonic, majOp, ImmVal, isNegCond, 0>;
1538 def _t_V4: NVJ_ConstImm_template<mnemonic, majOp, ImmVal, isNegCond, 1>;
1541 multiclass NVJ_ConstImm_base<string mnemonic, string BaseOp, bits<3> majOp,
1543 let BaseOpcode = BaseOp#_NVJ_ConstImm in {
1544 defm _t_Jumpnv : NVJ_ConstImm_cond<mnemonic, majOp, ImmVal, 0>; // True
1545 defm _f_Jumpnv : NVJ_ConstImm_cond<mnemonic, majOp, ImmVal, 1>; // False
1549 // if ([!]tstbit(Ns.new,#0)) jump:[n]t #r9:2
1550 // if ([!]cmp.eq(Ns.new,#-1)) jump:[n]t #r9:2
1551 // if ([!]cmp.gt(Ns.new,#-1)) jump:[n]t #r9:2
1553 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator=1,
1554 Defs = [PC], hasSideEffects = 0, isCodeGenOnly = 0 in {
1555 defm TSTBIT0 : NVJ_ConstImm_base<"tstbit", "TSTBIT", 0b011, "0">, PredRel;
1556 defm CMPEQn1 : NVJ_ConstImm_base<"cmp.eq", "CMPEQ", 0b100, "-1">, PredRel;
1557 defm CMPGTn1 : NVJ_ConstImm_base<"cmp.gt", "CMPGT", 0b101, "-1">, PredRel;
1560 // J4_hintjumpr: Hint indirect conditional jump.
1561 let isBranch = 1, isIndirectBranch = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
1562 def J4_hintjumpr: JRInst <
1567 let IClass = 0b0101;
1568 let Inst{27-21} = 0b0010101;
1569 let Inst{20-16} = Rs;
1572 //===----------------------------------------------------------------------===//
1574 //===----------------------------------------------------------------------===//
1576 //===----------------------------------------------------------------------===//
1578 //===----------------------------------------------------------------------===//
1581 let hasNewValue = 1, isExtendable = 1, opExtendable = 1,
1582 isExtentSigned = 0, opExtentBits = 6, hasSideEffects = 0,
1583 Uses = [PC], validSubTargets = HasV4SubT in
1584 def C4_addipc : CRInst <(outs IntRegs:$Rd), (ins u6Ext:$u6),
1585 "$Rd = add(pc, #$u6)", [], "", CR_tc_2_SLOT3 > {
1589 let IClass = 0b0110;
1590 let Inst{27-16} = 0b101001001001;
1591 let Inst{12-7} = u6;
1597 let hasSideEffects = 0 in
1598 class T_LOGICAL_3OP<string MnOp1, string MnOp2, bits<2> OpBits, bit IsNeg>
1599 : CRInst<(outs PredRegs:$Pd),
1600 (ins PredRegs:$Ps, PredRegs:$Pt, PredRegs:$Pu),
1601 "$Pd = " # MnOp1 # "($Ps, " # MnOp2 # "($Pt, " #
1602 !if (IsNeg,"!","") # "$Pu))",
1603 [], "", CR_tc_2early_SLOT23> {
1609 let IClass = 0b0110;
1610 let Inst{27-24} = 0b1011;
1611 let Inst{23} = IsNeg;
1612 let Inst{22-21} = OpBits;
1614 let Inst{17-16} = Ps;
1621 let isCodeGenOnly = 0 in {
1622 def C4_and_and : T_LOGICAL_3OP<"and", "and", 0b00, 0>;
1623 def C4_and_or : T_LOGICAL_3OP<"and", "or", 0b01, 0>;
1624 def C4_or_and : T_LOGICAL_3OP<"or", "and", 0b10, 0>;
1625 def C4_or_or : T_LOGICAL_3OP<"or", "or", 0b11, 0>;
1626 def C4_and_andn : T_LOGICAL_3OP<"and", "and", 0b00, 1>;
1627 def C4_and_orn : T_LOGICAL_3OP<"and", "or", 0b01, 1>;
1628 def C4_or_andn : T_LOGICAL_3OP<"or", "and", 0b10, 1>;
1629 def C4_or_orn : T_LOGICAL_3OP<"or", "or", 0b11, 1>;
1632 //===----------------------------------------------------------------------===//
1634 //===----------------------------------------------------------------------===//
1636 //===----------------------------------------------------------------------===//
1638 //===----------------------------------------------------------------------===//
1640 // Logical with-not instructions.
1641 let validSubTargets = HasV4SubT, isCodeGenOnly = 0 in {
1642 def A4_andnp : T_ALU64_logical<"and", 0b001, 1, 0, 1>;
1643 def A4_ornp : T_ALU64_logical<"or", 0b011, 1, 0, 1>;
1646 let hasNewValue = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
1647 def S4_parity: ALU64Inst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
1648 "$Rd = parity($Rs, $Rt)", [], "", ALU64_tc_2_SLOT23> {
1653 let IClass = 0b1101;
1654 let Inst{27-21} = 0b0101111;
1655 let Inst{20-16} = Rs;
1656 let Inst{12-8} = Rt;
1659 // Add and accumulate.
1660 // Rd=add(Rs,add(Ru,#s6))
1661 let isExtentSigned = 1, hasNewValue = 1, isExtendable = 1, opExtentBits = 6,
1662 opExtendable = 3, isCodeGenOnly = 0 in
1663 def S4_addaddi : ALU64Inst <(outs IntRegs:$Rd),
1664 (ins IntRegs:$Rs, IntRegs:$Ru, s6Ext:$s6),
1665 "$Rd = add($Rs, add($Ru, #$s6))" ,
1666 [(set (i32 IntRegs:$Rd), (add (i32 IntRegs:$Rs),
1667 (add (i32 IntRegs:$Ru), s6_16ExtPred:$s6)))],
1668 "", ALU64_tc_2_SLOT23> {
1674 let IClass = 0b1101;
1676 let Inst{27-23} = 0b10110;
1677 let Inst{22-21} = s6{5-4};
1678 let Inst{20-16} = Rs;
1679 let Inst{13} = s6{3};
1680 let Inst{12-8} = Rd;
1681 let Inst{7-5} = s6{2-0};
1685 let isExtentSigned = 1, hasSideEffects = 0, hasNewValue = 1, isExtendable = 1,
1686 opExtentBits = 6, opExtendable = 2, isCodeGenOnly = 0 in
1687 def S4_subaddi: ALU64Inst <(outs IntRegs:$Rd),
1688 (ins IntRegs:$Rs, s6Ext:$s6, IntRegs:$Ru),
1689 "$Rd = add($Rs, sub(#$s6, $Ru))",
1690 [], "", ALU64_tc_2_SLOT23> {
1696 let IClass = 0b1101;
1698 let Inst{27-23} = 0b10111;
1699 let Inst{22-21} = s6{5-4};
1700 let Inst{20-16} = Rs;
1701 let Inst{13} = s6{3};
1702 let Inst{12-8} = Rd;
1703 let Inst{7-5} = s6{2-0};
1708 // Rdd=extract(Rss,#u6,#U6)
1709 // Rdd=extract(Rss,Rtt)
1710 // Rd=extract(Rs,Rtt)
1711 // Rd=extract(Rs,#u5,#U5)
1713 let isCodeGenOnly = 0 in {
1714 def S4_extractp_rp : T_S3op_64 < "extract", 0b11, 0b100, 0>;
1715 def S4_extractp : T_S2op_extract <"extract", 0b1010, DoubleRegs, u6Imm>;
1718 let hasNewValue = 1, isCodeGenOnly = 0 in {
1719 def S4_extract_rp : T_S3op_extract<"extract", 0b01>;
1720 def S4_extract : T_S2op_extract <"extract", 0b1101, IntRegs, u5Imm>;
1723 let Itinerary = M_tc_3x_SLOT23, Defs = [USR_OVF], isCodeGenOnly = 0 in {
1724 def M4_mac_up_s1_sat: T_MType_acc_rr<"+= mpy", 0b011, 0b000, 0, [], 0, 1, 1>;
1725 def M4_nac_up_s1_sat: T_MType_acc_rr<"-= mpy", 0b011, 0b001, 0, [], 0, 1, 1>;
1728 // Logical xor with xor accumulation.
1729 // Rxx^=xor(Rss,Rtt)
1730 let hasSideEffects = 0, isCodeGenOnly = 0 in
1732 : SInst <(outs DoubleRegs:$Rxx),
1733 (ins DoubleRegs:$dst2, DoubleRegs:$Rss, DoubleRegs:$Rtt),
1734 "$Rxx ^= xor($Rss, $Rtt)",
1735 [(set (i64 DoubleRegs:$Rxx),
1736 (xor (i64 DoubleRegs:$dst2), (xor (i64 DoubleRegs:$Rss),
1737 (i64 DoubleRegs:$Rtt))))],
1738 "$dst2 = $Rxx", S_3op_tc_1_SLOT23> {
1743 let IClass = 0b1100;
1745 let Inst{27-23} = 0b10101;
1746 let Inst{20-16} = Rss;
1747 let Inst{12-8} = Rtt;
1748 let Inst{4-0} = Rxx;
1752 let isCodeGenOnly = 0 in
1753 def A4_bitspliti : T_S2op_2_di <"bitsplit", 0b110, 0b100>;
1755 // Arithmetic/Convergent round
1756 let isCodeGenOnly = 0 in
1757 def A4_cround_ri : T_S2op_2_ii <"cround", 0b111, 0b000>;
1759 let isCodeGenOnly = 0 in
1760 def A4_round_ri : T_S2op_2_ii <"round", 0b111, 0b100>;
1762 let Defs = [USR_OVF], isCodeGenOnly = 0 in
1763 def A4_round_ri_sat : T_S2op_2_ii <"round", 0b111, 0b110, 1>;
1765 // Logical-logical words.
1766 // Compound or-and -- Rx=or(Ru,and(Rx,#s10))
1767 let isExtentSigned = 1, hasNewValue = 1, isExtendable = 1, opExtentBits = 10,
1768 opExtendable = 3, isCodeGenOnly = 0 in
1770 ALU64Inst<(outs IntRegs:$Rx),
1771 (ins IntRegs:$Ru, IntRegs:$_src_, s10Ext:$s10),
1772 "$Rx = or($Ru, and($_src_, #$s10))" ,
1773 [(set (i32 IntRegs:$Rx),
1774 (or (i32 IntRegs:$Ru), (and (i32 IntRegs:$_src_), s10ExtPred:$s10)))] ,
1775 "$_src_ = $Rx", ALU64_tc_2_SLOT23> {
1780 let IClass = 0b1101;
1782 let Inst{27-22} = 0b101001;
1783 let Inst{20-16} = Rx;
1784 let Inst{21} = s10{9};
1785 let Inst{13-5} = s10{8-0};
1789 // Miscellaneous ALU64 instructions.
1791 let hasNewValue = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
1792 def A4_modwrapu: ALU64Inst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
1793 "$Rd = modwrap($Rs, $Rt)", [], "", ALU64_tc_2_SLOT23> {
1798 let IClass = 0b1101;
1799 let Inst{27-21} = 0b0011111;
1800 let Inst{20-16} = Rs;
1801 let Inst{12-8} = Rt;
1802 let Inst{7-5} = 0b111;
1806 let hasSideEffects = 0, isCodeGenOnly = 0 in
1807 def A4_bitsplit: ALU64Inst<(outs DoubleRegs:$Rd),
1808 (ins IntRegs:$Rs, IntRegs:$Rt),
1809 "$Rd = bitsplit($Rs, $Rt)", [], "", ALU64_tc_1_SLOT23> {
1814 let IClass = 0b1101;
1815 let Inst{27-24} = 0b0100;
1817 let Inst{20-16} = Rs;
1818 let Inst{12-8} = Rt;
1822 let isCodeGenOnly = 0 in {
1823 // Rx[&|]=xor(Rs,Rt)
1824 def M4_or_xor : T_MType_acc_rr < "|= xor", 0b110, 0b001, 0>;
1825 def M4_and_xor : T_MType_acc_rr < "&= xor", 0b010, 0b010, 0>;
1827 // Rx[&|^]=or(Rs,Rt)
1828 def M4_xor_or : T_MType_acc_rr < "^= or", 0b110, 0b011, 0>;
1830 let CextOpcode = "ORr_ORr" in
1831 def M4_or_or : T_MType_acc_rr < "|= or", 0b110, 0b000, 0>;
1832 def M4_and_or : T_MType_acc_rr < "&= or", 0b010, 0b001, 0>;
1834 // Rx[&|^]=and(Rs,Rt)
1835 def M4_xor_and : T_MType_acc_rr < "^= and", 0b110, 0b010, 0>;
1837 let CextOpcode = "ORr_ANDr" in
1838 def M4_or_and : T_MType_acc_rr < "|= and", 0b010, 0b011, 0>;
1839 def M4_and_and : T_MType_acc_rr < "&= and", 0b010, 0b000, 0>;
1841 // Rx[&|^]=and(Rs,~Rt)
1842 def M4_xor_andn : T_MType_acc_rr < "^= and", 0b001, 0b010, 0, [], 1>;
1843 def M4_or_andn : T_MType_acc_rr < "|= and", 0b001, 0b000, 0, [], 1>;
1844 def M4_and_andn : T_MType_acc_rr < "&= and", 0b001, 0b001, 0, [], 1>;
1847 // Compound or-or and or-and
1848 let isExtentSigned = 1, InputType = "imm", hasNewValue = 1, isExtendable = 1,
1849 opExtentBits = 10, opExtendable = 3 in
1850 class T_CompOR <string mnemonic, bits<2> MajOp, SDNode OpNode>
1851 : MInst_acc <(outs IntRegs:$Rx),
1852 (ins IntRegs:$src1, IntRegs:$Rs, s10Ext:$s10),
1853 "$Rx |= "#mnemonic#"($Rs, #$s10)",
1854 [(set (i32 IntRegs:$Rx), (or (i32 IntRegs:$src1),
1855 (OpNode (i32 IntRegs:$Rs), s10ExtPred:$s10)))],
1856 "$src1 = $Rx", ALU64_tc_2_SLOT23>, ImmRegRel {
1861 let IClass = 0b1101;
1863 let Inst{27-24} = 0b1010;
1864 let Inst{23-22} = MajOp;
1865 let Inst{20-16} = Rs;
1866 let Inst{21} = s10{9};
1867 let Inst{13-5} = s10{8-0};
1871 let CextOpcode = "ORr_ANDr", isCodeGenOnly = 0 in
1872 def S4_or_andi : T_CompOR <"and", 0b00, and>;
1874 let CextOpcode = "ORr_ORr", isCodeGenOnly = 0 in
1875 def S4_or_ori : T_CompOR <"or", 0b10, or>;
1877 // Add and accumulate.
1878 // Rd=add(Rs,add(Ru,#s6))
1879 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 6,
1880 validSubTargets = HasV4SubT in
1881 def ADDr_ADDri_V4 : MInst<(outs IntRegs:$dst),
1882 (ins IntRegs:$src1, IntRegs:$src2, s6Ext:$src3),
1883 "$dst = add($src1, add($src2, #$src3))",
1884 [(set (i32 IntRegs:$dst),
1885 (add (i32 IntRegs:$src1), (add (i32 IntRegs:$src2),
1886 s6_16ExtPred:$src3)))]>,
1889 // Rd=add(Rs,sub(#s6,Ru))
1890 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6,
1891 validSubTargets = HasV4SubT in
1892 def ADDr_SUBri_V4 : MInst<(outs IntRegs:$dst),
1893 (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3),
1894 "$dst = add($src1, sub(#$src2, $src3))",
1895 [(set (i32 IntRegs:$dst),
1896 (add (i32 IntRegs:$src1), (sub s6_10ExtPred:$src2,
1897 (i32 IntRegs:$src3))))]>,
1900 // Generates the same instruction as ADDr_SUBri_V4 but matches different
1902 // Rd=add(Rs,sub(#s6,Ru))
1903 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6,
1904 validSubTargets = HasV4SubT in
1905 def ADDri_SUBr_V4 : MInst<(outs IntRegs:$dst),
1906 (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3),
1907 "$dst = add($src1, sub(#$src2, $src3))",
1908 [(set (i32 IntRegs:$dst),
1909 (sub (add (i32 IntRegs:$src1), s6_10ExtPred:$src2),
1910 (i32 IntRegs:$src3)))]>,
1914 // Add or subtract doublewords with carry.
1916 // Rdd=add(Rss,Rtt,Px):carry
1918 // Rdd=sub(Rss,Rtt,Px):carry
1921 // Logical doublewords.
1922 // Rdd=and(Rtt,~Rss)
1923 let validSubTargets = HasV4SubT in
1924 def ANDd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
1925 (ins DoubleRegs:$src1, DoubleRegs:$src2),
1926 "$dst = and($src1, ~$src2)",
1927 [(set (i64 DoubleRegs:$dst), (and (i64 DoubleRegs:$src1),
1928 (not (i64 DoubleRegs:$src2))))]>,
1932 let validSubTargets = HasV4SubT in
1933 def ORd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
1934 (ins DoubleRegs:$src1, DoubleRegs:$src2),
1935 "$dst = or($src1, ~$src2)",
1936 [(set (i64 DoubleRegs:$dst),
1937 (or (i64 DoubleRegs:$src1), (not (i64 DoubleRegs:$src2))))]>,
1941 // Logical-logical doublewords.
1942 // Rxx^=xor(Rss,Rtt)
1943 let validSubTargets = HasV4SubT in
1944 def XORd_XORdd: MInst_acc<(outs DoubleRegs:$dst),
1945 (ins DoubleRegs:$src1, DoubleRegs:$src2, DoubleRegs:$src3),
1946 "$dst ^= xor($src2, $src3)",
1947 [(set (i64 DoubleRegs:$dst),
1948 (xor (i64 DoubleRegs:$src1), (xor (i64 DoubleRegs:$src2),
1949 (i64 DoubleRegs:$src3))))],
1954 // Logical-logical words.
1955 // Rx=or(Ru,and(Rx,#s10))
1956 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
1957 validSubTargets = HasV4SubT in
1958 def ORr_ANDri_V4 : MInst_acc<(outs IntRegs:$dst),
1959 (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
1960 "$dst = or($src1, and($src2, #$src3))",
1961 [(set (i32 IntRegs:$dst),
1962 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1963 s10ExtPred:$src3)))],
1967 // Rx[&|^]=and(Rs,Rt)
1969 let validSubTargets = HasV4SubT in
1970 def ANDr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
1971 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1972 "$dst &= and($src2, $src3)",
1973 [(set (i32 IntRegs:$dst),
1974 (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1975 (i32 IntRegs:$src3))))],
1980 let validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "reg" in
1981 def ORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
1982 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1983 "$dst |= and($src2, $src3)",
1984 [(set (i32 IntRegs:$dst),
1985 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1986 (i32 IntRegs:$src3))))],
1988 Requires<[HasV4T]>, ImmRegRel;
1991 let validSubTargets = HasV4SubT in
1992 def XORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
1993 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1994 "$dst ^= and($src2, $src3)",
1995 [(set (i32 IntRegs:$dst),
1996 (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1997 (i32 IntRegs:$src3))))],
2001 // Rx[&|^]=and(Rs,~Rt)
2003 let validSubTargets = HasV4SubT in
2004 def ANDr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
2005 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
2006 "$dst &= and($src2, ~$src3)",
2007 [(set (i32 IntRegs:$dst),
2008 (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
2009 (not (i32 IntRegs:$src3)))))],
2014 let validSubTargets = HasV4SubT in
2015 def ORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
2016 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
2017 "$dst |= and($src2, ~$src3)",
2018 [(set (i32 IntRegs:$dst),
2019 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
2020 (not (i32 IntRegs:$src3)))))],
2025 let validSubTargets = HasV4SubT in
2026 def XORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
2027 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
2028 "$dst ^= and($src2, ~$src3)",
2029 [(set (i32 IntRegs:$dst),
2030 (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
2031 (not (i32 IntRegs:$src3)))))],
2035 // Rx[&|^]=or(Rs,Rt)
2037 let validSubTargets = HasV4SubT in
2038 def ANDr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
2039 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
2040 "$dst &= or($src2, $src3)",
2041 [(set (i32 IntRegs:$dst),
2042 (and (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
2043 (i32 IntRegs:$src3))))],
2048 let validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "reg" in
2049 def ORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
2050 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
2051 "$dst |= or($src2, $src3)",
2052 [(set (i32 IntRegs:$dst),
2053 (or (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
2054 (i32 IntRegs:$src3))))],
2056 Requires<[HasV4T]>, ImmRegRel;
2059 let validSubTargets = HasV4SubT in
2060 def XORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
2061 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
2062 "$dst ^= or($src2, $src3)",
2063 [(set (i32 IntRegs:$dst),
2064 (xor (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
2065 (i32 IntRegs:$src3))))],
2069 // Rx[&|^]=xor(Rs,Rt)
2071 let validSubTargets = HasV4SubT in
2072 def ANDr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
2073 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
2074 "$dst &= xor($src2, $src3)",
2075 [(set (i32 IntRegs:$dst),
2076 (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
2077 (i32 IntRegs:$src3))))],
2082 let validSubTargets = HasV4SubT in
2083 def ORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
2084 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
2085 "$dst |= xor($src2, $src3)",
2086 [(set (i32 IntRegs:$dst),
2087 (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
2088 (i32 IntRegs:$src3))))],
2093 let validSubTargets = HasV4SubT in
2094 def XORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
2095 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
2096 "$dst ^= xor($src2, $src3)",
2097 [(set (i32 IntRegs:$dst),
2098 (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
2099 (i32 IntRegs:$src3))))],
2104 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
2105 validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "imm" in
2106 def ORr_ANDri2_V4 : MInst_acc<(outs IntRegs:$dst),
2107 (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
2108 "$dst |= and($src2, #$src3)",
2109 [(set (i32 IntRegs:$dst),
2110 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
2111 s10ExtPred:$src3)))],
2113 Requires<[HasV4T]>, ImmRegRel;
2116 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
2117 validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "imm" in
2118 def ORr_ORri_V4 : MInst_acc<(outs IntRegs:$dst),
2119 (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
2120 "$dst |= or($src2, #$src3)",
2121 [(set (i32 IntRegs:$dst),
2122 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
2123 s10ExtPred:$src3)))],
2125 Requires<[HasV4T]>, ImmRegRel;
2129 // Rd=modwrap(Rs,Rt)
2131 // Rd=cround(Rs,#u5)
2133 // Rd=round(Rs,#u5)[:sat]
2134 // Rd=round(Rs,Rt)[:sat]
2135 // Vector reduce add unsigned halfwords
2136 // Rd=vraddh(Rss,Rtt)
2138 // Rdd=vaddb(Rss,Rtt)
2139 // Vector conditional negate
2140 // Rdd=vcnegh(Rss,Rt)
2141 // Rxx+=vrcnegh(Rss,Rt)
2142 // Vector maximum bytes
2143 // Rdd=vmaxb(Rtt,Rss)
2144 // Vector reduce maximum halfwords
2145 // Rxx=vrmaxh(Rss,Ru)
2146 // Rxx=vrmaxuh(Rss,Ru)
2147 // Vector reduce maximum words
2148 // Rxx=vrmaxuw(Rss,Ru)
2149 // Rxx=vrmaxw(Rss,Ru)
2150 // Vector minimum bytes
2151 // Rdd=vminb(Rtt,Rss)
2152 // Vector reduce minimum halfwords
2153 // Rxx=vrminh(Rss,Ru)
2154 // Rxx=vrminuh(Rss,Ru)
2155 // Vector reduce minimum words
2156 // Rxx=vrminuw(Rss,Ru)
2157 // Rxx=vrminw(Rss,Ru)
2158 // Vector subtract bytes
2159 // Rdd=vsubb(Rss,Rtt)
2161 //===----------------------------------------------------------------------===//
2163 //===----------------------------------------------------------------------===//
2166 //===----------------------------------------------------------------------===//
2168 //===----------------------------------------------------------------------===//
2170 // Multiply and user lower result.
2171 // Rd=add(#u6,mpyi(Rs,#U6))
2172 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6,
2173 validSubTargets = HasV4SubT in
2174 def ADDi_MPYri_V4 : MInst<(outs IntRegs:$dst),
2175 (ins u6Ext:$src1, IntRegs:$src2, u6Imm:$src3),
2176 "$dst = add(#$src1, mpyi($src2, #$src3))",
2177 [(set (i32 IntRegs:$dst),
2178 (add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
2179 u6ExtPred:$src1))]>,
2182 // Rd=add(##,mpyi(Rs,#U6))
2183 def : Pat <(add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
2184 (HexagonCONST32 tglobaladdr:$src1)),
2185 (i32 (ADDi_MPYri_V4 tglobaladdr:$src1, IntRegs:$src2,
2188 // Rd=add(#u6,mpyi(Rs,Rt))
2189 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6,
2190 validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in
2191 def ADDi_MPYrr_V4 : MInst<(outs IntRegs:$dst),
2192 (ins u6Ext:$src1, IntRegs:$src2, IntRegs:$src3),
2193 "$dst = add(#$src1, mpyi($src2, $src3))",
2194 [(set (i32 IntRegs:$dst),
2195 (add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
2196 u6ExtPred:$src1))]>,
2197 Requires<[HasV4T]>, ImmRegRel;
2199 // Rd=add(##,mpyi(Rs,Rt))
2200 def : Pat <(add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
2201 (HexagonCONST32 tglobaladdr:$src1)),
2202 (i32 (ADDi_MPYrr_V4 tglobaladdr:$src1, IntRegs:$src2,
2205 // Rd=add(Ru,mpyi(#u6:2,Rs))
2206 let validSubTargets = HasV4SubT in
2207 def ADDr_MPYir_V4 : MInst<(outs IntRegs:$dst),
2208 (ins IntRegs:$src1, u6Imm:$src2, IntRegs:$src3),
2209 "$dst = add($src1, mpyi(#$src2, $src3))",
2210 [(set (i32 IntRegs:$dst),
2211 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src3),
2212 u6_2ImmPred:$src2)))]>,
2215 // Rd=add(Ru,mpyi(Rs,#u6))
2216 let isExtendable = 1, opExtendable = 3, isExtentSigned = 0, opExtentBits = 6,
2217 validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in
2218 def ADDr_MPYri_V4 : MInst<(outs IntRegs:$dst),
2219 (ins IntRegs:$src1, IntRegs:$src2, u6Ext:$src3),
2220 "$dst = add($src1, mpyi($src2, #$src3))",
2221 [(set (i32 IntRegs:$dst),
2222 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
2223 u6ExtPred:$src3)))]>,
2224 Requires<[HasV4T]>, ImmRegRel;
2226 // Rx=add(Ru,mpyi(Rx,Rs))
2227 let validSubTargets = HasV4SubT, InputType = "reg", CextOpcode = "ADD_MPY" in
2228 def ADDr_MPYrr_V4 : MInst_acc<(outs IntRegs:$dst),
2229 (ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3),
2230 "$dst = add($src1, mpyi($src2, $src3))",
2231 [(set (i32 IntRegs:$dst),
2232 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
2233 (i32 IntRegs:$src3))))],
2235 Requires<[HasV4T]>, ImmRegRel;
2238 // Polynomial multiply words
2240 // Rxx^=pmpyw(Rs,Rt)
2242 // Vector reduce multiply word by signed half (32x16)
2243 // Rdd=vrmpyweh(Rss,Rtt)[:<<1]
2244 // Rdd=vrmpywoh(Rss,Rtt)[:<<1]
2245 // Rxx+=vrmpyweh(Rss,Rtt)[:<<1]
2246 // Rxx+=vrmpywoh(Rss,Rtt)[:<<1]
2248 // Multiply and use upper result
2249 // Rd=mpy(Rs,Rt.H):<<1:sat
2250 // Rd=mpy(Rs,Rt.L):<<1:sat
2251 // Rd=mpy(Rs,Rt):<<1
2252 // Rd=mpy(Rs,Rt):<<1:sat
2254 // Rx+=mpy(Rs,Rt):<<1:sat
2255 // Rx-=mpy(Rs,Rt):<<1:sat
2257 // Vector multiply bytes
2258 // Rdd=vmpybsu(Rs,Rt)
2259 // Rdd=vmpybu(Rs,Rt)
2260 // Rxx+=vmpybsu(Rs,Rt)
2261 // Rxx+=vmpybu(Rs,Rt)
2263 // Vector polynomial multiply halfwords
2264 // Rdd=vpmpyh(Rs,Rt)
2265 // Rxx^=vpmpyh(Rs,Rt)
2267 //===----------------------------------------------------------------------===//
2269 //===----------------------------------------------------------------------===//
2272 //===----------------------------------------------------------------------===//
2274 //===----------------------------------------------------------------------===//
2276 // Shift by immediate and accumulate.
2277 // Rx=add(#u8,asl(Rx,#U5))
2278 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2279 validSubTargets = HasV4SubT in
2280 def ADDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
2281 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2282 "$dst = add(#$src1, asl($src2, #$src3))",
2283 [(set (i32 IntRegs:$dst),
2284 (add (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
2289 // Rx=add(#u8,lsr(Rx,#U5))
2290 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2291 validSubTargets = HasV4SubT in
2292 def ADDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
2293 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2294 "$dst = add(#$src1, lsr($src2, #$src3))",
2295 [(set (i32 IntRegs:$dst),
2296 (add (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
2301 // Rx=sub(#u8,asl(Rx,#U5))
2302 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2303 validSubTargets = HasV4SubT in
2304 def SUBi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
2305 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2306 "$dst = sub(#$src1, asl($src2, #$src3))",
2307 [(set (i32 IntRegs:$dst),
2308 (sub (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
2313 // Rx=sub(#u8,lsr(Rx,#U5))
2314 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2315 validSubTargets = HasV4SubT in
2316 def SUBi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
2317 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2318 "$dst = sub(#$src1, lsr($src2, #$src3))",
2319 [(set (i32 IntRegs:$dst),
2320 (sub (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
2326 //Shift by immediate and logical.
2327 //Rx=and(#u8,asl(Rx,#U5))
2328 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2329 validSubTargets = HasV4SubT in
2330 def ANDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
2331 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2332 "$dst = and(#$src1, asl($src2, #$src3))",
2333 [(set (i32 IntRegs:$dst),
2334 (and (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
2339 //Rx=and(#u8,lsr(Rx,#U5))
2340 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2341 validSubTargets = HasV4SubT in
2342 def ANDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
2343 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2344 "$dst = and(#$src1, lsr($src2, #$src3))",
2345 [(set (i32 IntRegs:$dst),
2346 (and (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
2351 //Rx=or(#u8,asl(Rx,#U5))
2352 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2353 AddedComplexity = 30, validSubTargets = HasV4SubT in
2354 def ORi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
2355 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2356 "$dst = or(#$src1, asl($src2, #$src3))",
2357 [(set (i32 IntRegs:$dst),
2358 (or (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
2363 //Rx=or(#u8,lsr(Rx,#U5))
2364 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2365 AddedComplexity = 30, validSubTargets = HasV4SubT in
2366 def ORi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
2367 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2368 "$dst = or(#$src1, lsr($src2, #$src3))",
2369 [(set (i32 IntRegs:$dst),
2370 (or (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
2376 //Shift by register.
2378 let validSubTargets = HasV4SubT in {
2379 def LSLi_V4 : MInst<(outs IntRegs:$dst), (ins s6Imm:$src1, IntRegs:$src2),
2380 "$dst = lsl(#$src1, $src2)",
2381 [(set (i32 IntRegs:$dst), (shl s6ImmPred:$src1,
2382 (i32 IntRegs:$src2)))]>,
2386 //Shift by register and logical.
2388 def ASLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
2389 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
2390 "$dst ^= asl($src2, $src3)",
2391 [(set (i64 DoubleRegs:$dst),
2392 (xor (i64 DoubleRegs:$src1), (shl (i64 DoubleRegs:$src2),
2393 (i32 IntRegs:$src3))))],
2398 def ASRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
2399 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
2400 "$dst ^= asr($src2, $src3)",
2401 [(set (i64 DoubleRegs:$dst),
2402 (xor (i64 DoubleRegs:$src1), (sra (i64 DoubleRegs:$src2),
2403 (i32 IntRegs:$src3))))],
2408 def LSLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
2409 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
2410 "$dst ^= lsl($src2, $src3)",
2411 [(set (i64 DoubleRegs:$dst), (xor (i64 DoubleRegs:$src1),
2412 (shl (i64 DoubleRegs:$src2),
2413 (i32 IntRegs:$src3))))],
2418 def LSRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
2419 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
2420 "$dst ^= lsr($src2, $src3)",
2421 [(set (i64 DoubleRegs:$dst),
2422 (xor (i64 DoubleRegs:$src1), (srl (i64 DoubleRegs:$src2),
2423 (i32 IntRegs:$src3))))],
2428 //===----------------------------------------------------------------------===//
2430 //===----------------------------------------------------------------------===//
2432 //===----------------------------------------------------------------------===//
2433 // MEMOP: Word, Half, Byte
2434 //===----------------------------------------------------------------------===//
2436 def MEMOPIMM : SDNodeXForm<imm, [{
2437 // Call the transformation function XformM5ToU5Imm to get the negative
2438 // immediate's positive counterpart.
2439 int32_t imm = N->getSExtValue();
2440 return XformM5ToU5Imm(imm);
2443 def MEMOPIMM_HALF : SDNodeXForm<imm, [{
2444 // -1 .. -31 represented as 65535..65515
2445 // assigning to a short restores our desired signed value.
2446 // Call the transformation function XformM5ToU5Imm to get the negative
2447 // immediate's positive counterpart.
2448 int16_t imm = N->getSExtValue();
2449 return XformM5ToU5Imm(imm);
2452 def MEMOPIMM_BYTE : SDNodeXForm<imm, [{
2453 // -1 .. -31 represented as 255..235
2454 // assigning to a char restores our desired signed value.
2455 // Call the transformation function XformM5ToU5Imm to get the negative
2456 // immediate's positive counterpart.
2457 int8_t imm = N->getSExtValue();
2458 return XformM5ToU5Imm(imm);
2461 def SETMEMIMM : SDNodeXForm<imm, [{
2462 // Return the bit position we will set [0-31].
2464 int32_t imm = N->getSExtValue();
2465 return XformMskToBitPosU5Imm(imm);
2468 def CLRMEMIMM : SDNodeXForm<imm, [{
2469 // Return the bit position we will clear [0-31].
2471 // we bit negate the value first
2472 int32_t imm = ~(N->getSExtValue());
2473 return XformMskToBitPosU5Imm(imm);
2476 def SETMEMIMM_SHORT : SDNodeXForm<imm, [{
2477 // Return the bit position we will set [0-15].
2479 int16_t imm = N->getSExtValue();
2480 return XformMskToBitPosU4Imm(imm);
2483 def CLRMEMIMM_SHORT : SDNodeXForm<imm, [{
2484 // Return the bit position we will clear [0-15].
2486 // we bit negate the value first
2487 int16_t imm = ~(N->getSExtValue());
2488 return XformMskToBitPosU4Imm(imm);
2491 def SETMEMIMM_BYTE : SDNodeXForm<imm, [{
2492 // Return the bit position we will set [0-7].
2494 int8_t imm = N->getSExtValue();
2495 return XformMskToBitPosU3Imm(imm);
2498 def CLRMEMIMM_BYTE : SDNodeXForm<imm, [{
2499 // Return the bit position we will clear [0-7].
2501 // we bit negate the value first
2502 int8_t imm = ~(N->getSExtValue());
2503 return XformMskToBitPosU3Imm(imm);
2506 //===----------------------------------------------------------------------===//
2507 // Template class for MemOp instructions with the register value.
2508 //===----------------------------------------------------------------------===//
2509 class MemOp_rr_base <string opc, bits<2> opcBits, Operand ImmOp,
2510 string memOp, bits<2> memOpBits> :
2512 (ins IntRegs:$base, ImmOp:$offset, IntRegs:$delta),
2513 opc#"($base+#$offset)"#memOp#"$delta",
2515 Requires<[HasV4T, UseMEMOP]> {
2520 bits<6> offsetBits; // memb - u6:0 , memh - u6:1, memw - u6:2
2522 let offsetBits = !if (!eq(opcBits, 0b00), offset{5-0},
2523 !if (!eq(opcBits, 0b01), offset{6-1},
2524 !if (!eq(opcBits, 0b10), offset{7-2},0)));
2526 let IClass = 0b0011;
2527 let Inst{27-24} = 0b1110;
2528 let Inst{22-21} = opcBits;
2529 let Inst{20-16} = base;
2531 let Inst{12-7} = offsetBits;
2532 let Inst{6-5} = memOpBits;
2533 let Inst{4-0} = delta;
2536 //===----------------------------------------------------------------------===//
2537 // Template class for MemOp instructions with the immediate value.
2538 //===----------------------------------------------------------------------===//
2539 class MemOp_ri_base <string opc, bits<2> opcBits, Operand ImmOp,
2540 string memOp, bits<2> memOpBits> :
2542 (ins IntRegs:$base, ImmOp:$offset, u5Imm:$delta),
2543 opc#"($base+#$offset)"#memOp#"#$delta"
2544 #!if(memOpBits{1},")", ""), // clrbit, setbit - include ')'
2546 Requires<[HasV4T, UseMEMOP]> {
2551 bits<6> offsetBits; // memb - u6:0 , memh - u6:1, memw - u6:2
2553 let offsetBits = !if (!eq(opcBits, 0b00), offset{5-0},
2554 !if (!eq(opcBits, 0b01), offset{6-1},
2555 !if (!eq(opcBits, 0b10), offset{7-2},0)));
2557 let IClass = 0b0011;
2558 let Inst{27-24} = 0b1111;
2559 let Inst{22-21} = opcBits;
2560 let Inst{20-16} = base;
2562 let Inst{12-7} = offsetBits;
2563 let Inst{6-5} = memOpBits;
2564 let Inst{4-0} = delta;
2567 // multiclass to define MemOp instructions with register operand.
2568 multiclass MemOp_rr<string opc, bits<2> opcBits, Operand ImmOp> {
2569 def _ADD#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " += ", 0b00>; // add
2570 def _SUB#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " -= ", 0b01>; // sub
2571 def _AND#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " &= ", 0b10>; // and
2572 def _OR#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " |= ", 0b11>; // or
2575 // multiclass to define MemOp instructions with immediate Operand.
2576 multiclass MemOp_ri<string opc, bits<2> opcBits, Operand ImmOp> {
2577 def _ADD#NAME#_V4 : MemOp_ri_base <opc, opcBits, ImmOp, " += ", 0b00 >;
2578 def _SUB#NAME#_V4 : MemOp_ri_base <opc, opcBits, ImmOp, " -= ", 0b01 >;
2579 def _CLRBIT#NAME#_V4 : MemOp_ri_base<opc, opcBits, ImmOp, " =clrbit(", 0b10>;
2580 def _SETBIT#NAME#_V4 : MemOp_ri_base<opc, opcBits, ImmOp, " =setbit(", 0b11>;
2583 multiclass MemOp_base <string opc, bits<2> opcBits, Operand ImmOp> {
2584 defm r : MemOp_rr <opc, opcBits, ImmOp>;
2585 defm i : MemOp_ri <opc, opcBits, ImmOp>;
2588 // Define MemOp instructions.
2589 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0,
2590 validSubTargets =HasV4SubT in {
2591 let opExtentBits = 6, accessSize = ByteAccess in
2592 defm MemOPb : MemOp_base <"memb", 0b00, u6_0Ext>;
2594 let opExtentBits = 7, accessSize = HalfWordAccess in
2595 defm MemOPh : MemOp_base <"memh", 0b01, u6_1Ext>;
2597 let opExtentBits = 8, accessSize = WordAccess in
2598 defm MemOPw : MemOp_base <"memw", 0b10, u6_2Ext>;
2601 //===----------------------------------------------------------------------===//
2602 // Multiclass to define 'Def Pats' for ALU operations on the memory
2603 // Here value used for the ALU operation is an immediate value.
2604 // mem[bh](Rs+#0) += #U5
2605 // mem[bh](Rs+#u6) += #U5
2606 //===----------------------------------------------------------------------===//
2608 multiclass MemOpi_u5Pats <PatFrag ldOp, PatFrag stOp, PatLeaf ExtPred,
2609 InstHexagon MI, SDNode OpNode> {
2610 let AddedComplexity = 180 in
2611 def : Pat < (stOp (OpNode (ldOp IntRegs:$addr), u5ImmPred:$addend),
2613 (MI IntRegs:$addr, #0, u5ImmPred:$addend )>;
2615 let AddedComplexity = 190 in
2616 def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, ExtPred:$offset)),
2618 (add IntRegs:$base, ExtPred:$offset)),
2619 (MI IntRegs:$base, ExtPred:$offset, u5ImmPred:$addend)>;
2622 multiclass MemOpi_u5ALUOp<PatFrag ldOp, PatFrag stOp, PatLeaf ExtPred,
2623 InstHexagon addMI, InstHexagon subMI> {
2624 defm : MemOpi_u5Pats<ldOp, stOp, ExtPred, addMI, add>;
2625 defm : MemOpi_u5Pats<ldOp, stOp, ExtPred, subMI, sub>;
2628 multiclass MemOpi_u5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2630 defm : MemOpi_u5ALUOp <ldOpHalf, truncstorei16, u6_1ExtPred,
2631 MemOPh_ADDi_V4, MemOPh_SUBi_V4>;
2633 defm : MemOpi_u5ALUOp <ldOpByte, truncstorei8, u6ExtPred,
2634 MemOPb_ADDi_V4, MemOPb_SUBi_V4>;
2637 let Predicates = [HasV4T, UseMEMOP] in {
2638 defm : MemOpi_u5ExtType<zextloadi8, zextloadi16>; // zero extend
2639 defm : MemOpi_u5ExtType<sextloadi8, sextloadi16>; // sign extend
2640 defm : MemOpi_u5ExtType<extloadi8, extloadi16>; // any extend
2643 defm : MemOpi_u5ALUOp <load, store, u6_2ExtPred, MemOPw_ADDi_V4,
2647 //===----------------------------------------------------------------------===//
2648 // multiclass to define 'Def Pats' for ALU operations on the memory.
2649 // Here value used for the ALU operation is a negative value.
2650 // mem[bh](Rs+#0) += #m5
2651 // mem[bh](Rs+#u6) += #m5
2652 //===----------------------------------------------------------------------===//
2654 multiclass MemOpi_m5Pats <PatFrag ldOp, PatFrag stOp, PatLeaf extPred,
2655 PatLeaf immPred, ComplexPattern addrPred,
2656 SDNodeXForm xformFunc, InstHexagon MI> {
2657 let AddedComplexity = 190 in
2658 def : Pat <(stOp (add (ldOp IntRegs:$addr), immPred:$subend),
2660 (MI IntRegs:$addr, #0, (xformFunc immPred:$subend) )>;
2662 let AddedComplexity = 195 in
2663 def : Pat<(stOp (add (ldOp (add IntRegs:$base, extPred:$offset)),
2665 (add IntRegs:$base, extPred:$offset)),
2666 (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$subend))>;
2669 multiclass MemOpi_m5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2671 defm : MemOpi_m5Pats <ldOpHalf, truncstorei16, u6_1ExtPred, m5HImmPred,
2672 ADDRriU6_1, MEMOPIMM_HALF, MemOPh_SUBi_V4>;
2674 defm : MemOpi_m5Pats <ldOpByte, truncstorei8, u6ExtPred, m5BImmPred,
2675 ADDRriU6_0, MEMOPIMM_BYTE, MemOPb_SUBi_V4>;
2678 let Predicates = [HasV4T, UseMEMOP] in {
2679 defm : MemOpi_m5ExtType<zextloadi8, zextloadi16>; // zero extend
2680 defm : MemOpi_m5ExtType<sextloadi8, sextloadi16>; // sign extend
2681 defm : MemOpi_m5ExtType<extloadi8, extloadi16>; // any extend
2684 defm : MemOpi_m5Pats <load, store, u6_2ExtPred, m5ImmPred,
2685 ADDRriU6_2, MEMOPIMM, MemOPw_SUBi_V4>;
2688 //===----------------------------------------------------------------------===//
2689 // Multiclass to define 'def Pats' for bit operations on the memory.
2690 // mem[bhw](Rs+#0) = [clrbit|setbit](#U5)
2691 // mem[bhw](Rs+#u6) = [clrbit|setbit](#U5)
2692 //===----------------------------------------------------------------------===//
2694 multiclass MemOpi_bitPats <PatFrag ldOp, PatFrag stOp, PatLeaf immPred,
2695 PatLeaf extPred, ComplexPattern addrPred,
2696 SDNodeXForm xformFunc, InstHexagon MI, SDNode OpNode> {
2698 // mem[bhw](Rs+#u6:[012]) = [clrbit|setbit](#U5)
2699 let AddedComplexity = 250 in
2700 def : Pat<(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)),
2702 (add IntRegs:$base, extPred:$offset)),
2703 (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$bitend))>;
2705 // mem[bhw](Rs+#0) = [clrbit|setbit](#U5)
2706 let AddedComplexity = 225 in
2707 def : Pat <(stOp (OpNode (ldOp (addrPred IntRegs:$addr, extPred:$offset)),
2709 (addrPred (i32 IntRegs:$addr), extPred:$offset)),
2710 (MI IntRegs:$addr, extPred:$offset, (xformFunc immPred:$bitend))>;
2713 multiclass MemOpi_bitExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2715 defm : MemOpi_bitPats<ldOpByte, truncstorei8, Clr3ImmPred, u6ExtPred,
2716 ADDRriU6_0, CLRMEMIMM_BYTE, MemOPb_CLRBITi_V4, and>;
2718 defm : MemOpi_bitPats<ldOpByte, truncstorei8, Set3ImmPred, u6ExtPred,
2719 ADDRriU6_0, SETMEMIMM_BYTE, MemOPb_SETBITi_V4, or>;
2720 // Half Word - clrbit
2721 defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Clr4ImmPred, u6_1ExtPred,
2722 ADDRriU6_1, CLRMEMIMM_SHORT, MemOPh_CLRBITi_V4, and>;
2723 // Half Word - setbit
2724 defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Set4ImmPred, u6_1ExtPred,
2725 ADDRriU6_1, SETMEMIMM_SHORT, MemOPh_SETBITi_V4, or>;
2728 let Predicates = [HasV4T, UseMEMOP] in {
2729 // mem[bh](Rs+#0) = [clrbit|setbit](#U5)
2730 // mem[bh](Rs+#u6:[01]) = [clrbit|setbit](#U5)
2731 defm : MemOpi_bitExtType<zextloadi8, zextloadi16>; // zero extend
2732 defm : MemOpi_bitExtType<sextloadi8, sextloadi16>; // sign extend
2733 defm : MemOpi_bitExtType<extloadi8, extloadi16>; // any extend
2735 // memw(Rs+#0) = [clrbit|setbit](#U5)
2736 // memw(Rs+#u6:2) = [clrbit|setbit](#U5)
2737 defm : MemOpi_bitPats<load, store, Clr5ImmPred, u6_2ExtPred, ADDRriU6_2,
2738 CLRMEMIMM, MemOPw_CLRBITi_V4, and>;
2739 defm : MemOpi_bitPats<load, store, Set5ImmPred, u6_2ExtPred, ADDRriU6_2,
2740 SETMEMIMM, MemOPw_SETBITi_V4, or>;
2743 //===----------------------------------------------------------------------===//
2744 // Multiclass to define 'def Pats' for ALU operations on the memory
2745 // where addend is a register.
2746 // mem[bhw](Rs+#0) [+-&|]= Rt
2747 // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt
2748 //===----------------------------------------------------------------------===//
2750 multiclass MemOpr_Pats <PatFrag ldOp, PatFrag stOp, ComplexPattern addrPred,
2751 PatLeaf extPred, InstHexagon MI, SDNode OpNode> {
2752 let AddedComplexity = 141 in
2753 // mem[bhw](Rs+#0) [+-&|]= Rt
2754 def : Pat <(stOp (OpNode (ldOp (addrPred IntRegs:$addr, extPred:$offset)),
2755 (i32 IntRegs:$addend)),
2756 (addrPred (i32 IntRegs:$addr), extPred:$offset)),
2757 (MI IntRegs:$addr, extPred:$offset, (i32 IntRegs:$addend) )>;
2759 // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt
2760 let AddedComplexity = 150 in
2761 def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)),
2762 (i32 IntRegs:$orend)),
2763 (add IntRegs:$base, extPred:$offset)),
2764 (MI IntRegs:$base, extPred:$offset, (i32 IntRegs:$orend) )>;
2767 multiclass MemOPr_ALUOp<PatFrag ldOp, PatFrag stOp,
2768 ComplexPattern addrPred, PatLeaf extPred,
2769 InstHexagon addMI, InstHexagon subMI,
2770 InstHexagon andMI, InstHexagon orMI > {
2772 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, addMI, add>;
2773 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, subMI, sub>;
2774 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, andMI, and>;
2775 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, orMI, or>;
2778 multiclass MemOPr_ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2780 defm : MemOPr_ALUOp <ldOpHalf, truncstorei16, ADDRriU6_1, u6_1ExtPred,
2781 MemOPh_ADDr_V4, MemOPh_SUBr_V4,
2782 MemOPh_ANDr_V4, MemOPh_ORr_V4>;
2784 defm : MemOPr_ALUOp <ldOpByte, truncstorei8, ADDRriU6_0, u6ExtPred,
2785 MemOPb_ADDr_V4, MemOPb_SUBr_V4,
2786 MemOPb_ANDr_V4, MemOPb_ORr_V4>;
2789 // Define 'def Pats' for MemOps with register addend.
2790 let Predicates = [HasV4T, UseMEMOP] in {
2792 defm : MemOPr_ExtType<zextloadi8, zextloadi16>; // zero extend
2793 defm : MemOPr_ExtType<sextloadi8, sextloadi16>; // sign extend
2794 defm : MemOPr_ExtType<extloadi8, extloadi16>; // any extend
2796 defm : MemOPr_ALUOp <load, store, ADDRriU6_2, u6_2ExtPred, MemOPw_ADDr_V4,
2797 MemOPw_SUBr_V4, MemOPw_ANDr_V4, MemOPw_ORr_V4 >;
2800 //===----------------------------------------------------------------------===//
2802 //===----------------------------------------------------------------------===//
2804 // Hexagon V4 only supports these flavors of byte/half compare instructions:
2805 // EQ/GT/GTU. Other flavors like GE/GEU/LT/LTU/LE/LEU are not supported by
2806 // hardware. However, compiler can still implement these patterns through
2807 // appropriate patterns combinations based on current implemented patterns.
2808 // The implemented patterns are: EQ/GT/GTU.
2809 // Missing patterns are: GE/GEU/LT/LTU/LE/LEU.
2811 // Following instruction is not being extended as it results into the
2812 // incorrect code for negative numbers.
2813 // Pd=cmpb.eq(Rs,#u8)
2815 let isCompare = 1, isExtendable = 1, opExtendable = 2, hasSideEffects = 0,
2816 validSubTargets = HasV4SubT in
2817 class CMP_NOT_REG_IMM<string OpName, bits<2> op, Operand ImmOp,
2819 : ALU32Inst <(outs PredRegs:$dst), (ins IntRegs:$src1, ImmOp:$src2),
2820 "$dst = !cmp."#OpName#"($src1, #$src2)",
2822 "", ALU32_2op_tc_2early_SLOT0123> {
2827 let IClass = 0b0111;
2828 let Inst{27-24} = 0b0101;
2829 let Inst{23-22} = op;
2830 let Inst{20-16} = src1;
2831 let Inst{21} = !if (!eq(OpName, "gtu"), 0b0, src2{9});
2832 let Inst{13-5} = src2{8-0};
2833 let Inst{4-2} = 0b100;
2834 let Inst{1-0} = dst;
2837 let opExtentBits = 10, isExtentSigned = 1 in {
2838 def C4_cmpneqi : CMP_NOT_REG_IMM <"eq", 0b00, s10Ext, [(set (i1 PredRegs:$dst),
2839 (setne (i32 IntRegs:$src1), s10ExtPred:$src2))]>;
2841 def C4_cmpltei : CMP_NOT_REG_IMM <"gt", 0b01, s10Ext, [(set (i1 PredRegs:$dst),
2842 (not (setgt (i32 IntRegs:$src1), s10ExtPred:$src2)))]>;
2845 let opExtentBits = 9 in
2846 def C4_cmplteui : CMP_NOT_REG_IMM <"gtu", 0b10, u9Ext, [(set (i1 PredRegs:$dst),
2847 (not (setugt (i32 IntRegs:$src1), u9ExtPred:$src2)))]>;
2852 let isCompare = 1, validSubTargets = HasV4SubT in
2853 def CMPnotEQ_rr : ALU32_rr<(outs PredRegs:$dst),
2854 (ins IntRegs:$src1, IntRegs:$src2),
2855 "$dst = !cmp.eq($src1, $src2)",
2856 [(set (i1 PredRegs:$dst),
2857 (setne (i32 IntRegs:$src1), (i32 IntRegs:$src2)))]>,
2861 let isCompare = 1, validSubTargets = HasV4SubT in
2862 def CMPnotGT_rr : ALU32_rr<(outs PredRegs:$dst),
2863 (ins IntRegs:$src1, IntRegs:$src2),
2864 "$dst = !cmp.gt($src1, $src2)",
2865 [(set (i1 PredRegs:$dst),
2866 (not (setgt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))]>,
2870 // p=!cmp.gtu(r1,r2)
2871 let isCompare = 1, validSubTargets = HasV4SubT in
2872 def CMPnotGTU_rr : ALU32_rr<(outs PredRegs:$dst),
2873 (ins IntRegs:$src1, IntRegs:$src2),
2874 "$dst = !cmp.gtu($src1, $src2)",
2875 [(set (i1 PredRegs:$dst),
2876 (not (setugt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))]>,
2879 let isCompare = 1, validSubTargets = HasV4SubT in
2880 def CMPbEQri_V4 : MInst<(outs PredRegs:$dst),
2881 (ins IntRegs:$src1, u8Imm:$src2),
2882 "$dst = cmpb.eq($src1, #$src2)",
2883 [(set (i1 PredRegs:$dst),
2884 (seteq (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2))]>,
2887 def : Pat <(brcond (i1 (setne (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2)),
2889 (J2_jumpf (CMPbEQri_V4 (i32 IntRegs:$src1), u8ImmPred:$src2),
2893 // Pd=cmpb.eq(Rs,Rt)
2894 let isCompare = 1, validSubTargets = HasV4SubT in
2895 def CMPbEQrr_ubub_V4 : MInst<(outs PredRegs:$dst),
2896 (ins IntRegs:$src1, IntRegs:$src2),
2897 "$dst = cmpb.eq($src1, $src2)",
2898 [(set (i1 PredRegs:$dst),
2899 (seteq (and (xor (i32 IntRegs:$src1),
2900 (i32 IntRegs:$src2)), 255), 0))]>,
2903 // Pd=cmpb.eq(Rs,Rt)
2904 let isCompare = 1, validSubTargets = HasV4SubT in
2905 def CMPbEQrr_sbsb_V4 : MInst<(outs PredRegs:$dst),
2906 (ins IntRegs:$src1, IntRegs:$src2),
2907 "$dst = cmpb.eq($src1, $src2)",
2908 [(set (i1 PredRegs:$dst),
2909 (seteq (shl (i32 IntRegs:$src1), (i32 24)),
2910 (shl (i32 IntRegs:$src2), (i32 24))))]>,
2913 // Pd=cmpb.gt(Rs,Rt)
2914 let isCompare = 1, validSubTargets = HasV4SubT in
2915 def CMPbGTrr_V4 : MInst<(outs PredRegs:$dst),
2916 (ins IntRegs:$src1, IntRegs:$src2),
2917 "$dst = cmpb.gt($src1, $src2)",
2918 [(set (i1 PredRegs:$dst),
2919 (setgt (shl (i32 IntRegs:$src1), (i32 24)),
2920 (shl (i32 IntRegs:$src2), (i32 24))))]>,
2923 // Pd=cmpb.gtu(Rs,#u7)
2924 let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7,
2925 isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU", InputType = "imm" in
2926 def CMPbGTUri_V4 : MInst<(outs PredRegs:$dst),
2927 (ins IntRegs:$src1, u7Ext:$src2),
2928 "$dst = cmpb.gtu($src1, #$src2)",
2929 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255),
2930 u7ExtPred:$src2))]>,
2931 Requires<[HasV4T]>, ImmRegRel;
2933 // SDNode for converting immediate C to C-1.
2934 def DEC_CONST_BYTE : SDNodeXForm<imm, [{
2935 // Return the byte immediate const-1 as an SDNode.
2936 int32_t imm = N->getSExtValue();
2937 return XformU7ToU7M1Imm(imm);
2941 // zext( seteq ( and(Rs, 255), u8))
2943 // Pd=cmpb.eq(Rs, #u8)
2944 // if (Pd.new) Rd=#1
2945 // if (!Pd.new) Rd=#0
2946 def : Pat <(i32 (zext (i1 (seteq (i32 (and (i32 IntRegs:$Rs), 255)),
2948 (i32 (TFR_condset_ii (i1 (CMPbEQri_V4 (i32 IntRegs:$Rs),
2954 // zext( setne ( and(Rs, 255), u8))
2956 // Pd=cmpb.eq(Rs, #u8)
2957 // if (Pd.new) Rd=#0
2958 // if (!Pd.new) Rd=#1
2959 def : Pat <(i32 (zext (i1 (setne (i32 (and (i32 IntRegs:$Rs), 255)),
2961 (i32 (TFR_condset_ii (i1 (CMPbEQri_V4 (i32 IntRegs:$Rs),
2967 // zext( seteq (Rs, and(Rt, 255)))
2969 // Pd=cmpb.eq(Rs, Rt)
2970 // if (Pd.new) Rd=#1
2971 // if (!Pd.new) Rd=#0
2972 def : Pat <(i32 (zext (i1 (seteq (i32 IntRegs:$Rt),
2973 (i32 (and (i32 IntRegs:$Rs), 255)))))),
2974 (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs),
2975 (i32 IntRegs:$Rt))),
2980 // zext( setne (Rs, and(Rt, 255)))
2982 // Pd=cmpb.eq(Rs, Rt)
2983 // if (Pd.new) Rd=#0
2984 // if (!Pd.new) Rd=#1
2985 def : Pat <(i32 (zext (i1 (setne (i32 IntRegs:$Rt),
2986 (i32 (and (i32 IntRegs:$Rs), 255)))))),
2987 (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs),
2988 (i32 IntRegs:$Rt))),
2993 // zext( setugt ( and(Rs, 255), u8))
2995 // Pd=cmpb.gtu(Rs, #u8)
2996 // if (Pd.new) Rd=#1
2997 // if (!Pd.new) Rd=#0
2998 def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 255)),
3000 (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs),
3006 // zext( setugt ( and(Rs, 254), u8))
3008 // Pd=cmpb.gtu(Rs, #u8)
3009 // if (Pd.new) Rd=#1
3010 // if (!Pd.new) Rd=#0
3011 def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 254)),
3013 (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs),
3019 // zext( setult ( Rs, Rt))
3021 // Pd=cmp.ltu(Rs, Rt)
3022 // if (Pd.new) Rd=#1
3023 // if (!Pd.new) Rd=#0
3024 // cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs)
3025 def : Pat <(i32 (zext (i1 (setult (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
3026 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rt),
3027 (i32 IntRegs:$Rs))),
3032 // zext( setlt ( Rs, Rt))
3034 // Pd=cmp.lt(Rs, Rt)
3035 // if (Pd.new) Rd=#1
3036 // if (!Pd.new) Rd=#0
3037 // cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs)
3038 def : Pat <(i32 (zext (i1 (setlt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
3039 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rt),
3040 (i32 IntRegs:$Rs))),
3045 // zext( setugt ( Rs, Rt))
3047 // Pd=cmp.gtu(Rs, Rt)
3048 // if (Pd.new) Rd=#1
3049 // if (!Pd.new) Rd=#0
3050 def : Pat <(i32 (zext (i1 (setugt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
3051 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rs),
3052 (i32 IntRegs:$Rt))),
3056 // This pattern interefers with coremark performance, not implementing at this
3059 // zext( setgt ( Rs, Rt))
3061 // Pd=cmp.gt(Rs, Rt)
3062 // if (Pd.new) Rd=#1
3063 // if (!Pd.new) Rd=#0
3066 // zext( setuge ( Rs, Rt))
3068 // Pd=cmp.ltu(Rs, Rt)
3069 // if (Pd.new) Rd=#0
3070 // if (!Pd.new) Rd=#1
3071 // cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs)
3072 def : Pat <(i32 (zext (i1 (setuge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
3073 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rt),
3074 (i32 IntRegs:$Rs))),
3079 // zext( setge ( Rs, Rt))
3081 // Pd=cmp.lt(Rs, Rt)
3082 // if (Pd.new) Rd=#0
3083 // if (!Pd.new) Rd=#1
3084 // cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs)
3085 def : Pat <(i32 (zext (i1 (setge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
3086 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rt),
3087 (i32 IntRegs:$Rs))),
3092 // zext( setule ( Rs, Rt))
3094 // Pd=cmp.gtu(Rs, Rt)
3095 // if (Pd.new) Rd=#0
3096 // if (!Pd.new) Rd=#1
3097 def : Pat <(i32 (zext (i1 (setule (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
3098 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rs),
3099 (i32 IntRegs:$Rt))),
3104 // zext( setle ( Rs, Rt))
3106 // Pd=cmp.gt(Rs, Rt)
3107 // if (Pd.new) Rd=#0
3108 // if (!Pd.new) Rd=#1
3109 def : Pat <(i32 (zext (i1 (setle (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
3110 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rs),
3111 (i32 IntRegs:$Rt))),
3116 // zext( setult ( and(Rs, 255), u8))
3117 // Use the isdigit transformation below
3119 // Generate code of the form 'mux_ii(cmpbgtu(Rdd, C-1),0,1)'
3120 // for C code of the form r = ((c>='0') & (c<='9')) ? 1 : 0;.
3121 // The isdigit transformation relies on two 'clever' aspects:
3122 // 1) The data type is unsigned which allows us to eliminate a zero test after
3123 // biasing the expression by 48. We are depending on the representation of
3124 // the unsigned types, and semantics.
3125 // 2) The front end has converted <= 9 into < 10 on entry to LLVM
3128 // retval = ((c>='0') & (c<='9')) ? 1 : 0;
3129 // The code is transformed upstream of llvm into
3130 // retval = (c-48) < 10 ? 1 : 0;
3131 let AddedComplexity = 139 in
3132 def : Pat <(i32 (zext (i1 (setult (i32 (and (i32 IntRegs:$src1), 255)),
3133 u7StrictPosImmPred:$src2)))),
3134 (i32 (C2_muxii (i1 (CMPbGTUri_V4 (i32 IntRegs:$src1),
3135 (DEC_CONST_BYTE u7StrictPosImmPred:$src2))),
3139 // Pd=cmpb.gtu(Rs,Rt)
3140 let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU",
3141 InputType = "reg" in
3142 def CMPbGTUrr_V4 : MInst<(outs PredRegs:$dst),
3143 (ins IntRegs:$src1, IntRegs:$src2),
3144 "$dst = cmpb.gtu($src1, $src2)",
3145 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255),
3146 (and (i32 IntRegs:$src2), 255)))]>,
3147 Requires<[HasV4T]>, ImmRegRel;
3149 // Following instruction is not being extended as it results into the incorrect
3150 // code for negative numbers.
3152 // Signed half compare(.eq) ri.
3153 // Pd=cmph.eq(Rs,#s8)
3154 let isCompare = 1, validSubTargets = HasV4SubT in
3155 def CMPhEQri_V4 : MInst<(outs PredRegs:$dst),
3156 (ins IntRegs:$src1, s8Imm:$src2),
3157 "$dst = cmph.eq($src1, #$src2)",
3158 [(set (i1 PredRegs:$dst), (seteq (and (i32 IntRegs:$src1), 65535),
3159 s8ImmPred:$src2))]>,
3162 // Signed half compare(.eq) rr.
3163 // Case 1: xor + and, then compare:
3165 // r0=and(r0,#0xffff)
3167 // Pd=cmph.eq(Rs,Rt)
3168 let isCompare = 1, validSubTargets = HasV4SubT in
3169 def CMPhEQrr_xor_V4 : MInst<(outs PredRegs:$dst),
3170 (ins IntRegs:$src1, IntRegs:$src2),
3171 "$dst = cmph.eq($src1, $src2)",
3172 [(set (i1 PredRegs:$dst), (seteq (and (xor (i32 IntRegs:$src1),
3173 (i32 IntRegs:$src2)),
3177 // Signed half compare(.eq) rr.
3178 // Case 2: shift left 16 bits then compare:
3182 // Pd=cmph.eq(Rs,Rt)
3183 let isCompare = 1, validSubTargets = HasV4SubT in
3184 def CMPhEQrr_shl_V4 : MInst<(outs PredRegs:$dst),
3185 (ins IntRegs:$src1, IntRegs:$src2),
3186 "$dst = cmph.eq($src1, $src2)",
3187 [(set (i1 PredRegs:$dst),
3188 (seteq (shl (i32 IntRegs:$src1), (i32 16)),
3189 (shl (i32 IntRegs:$src2), (i32 16))))]>,
3192 /* Incorrect Pattern -- immediate should be right shifted before being
3193 used in the cmph.gt instruction.
3194 // Signed half compare(.gt) ri.
3195 // Pd=cmph.gt(Rs,#s8)
3197 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 8,
3198 isCompare = 1, validSubTargets = HasV4SubT in
3199 def CMPhGTri_V4 : MInst<(outs PredRegs:$dst),
3200 (ins IntRegs:$src1, s8Ext:$src2),
3201 "$dst = cmph.gt($src1, #$src2)",
3202 [(set (i1 PredRegs:$dst),
3203 (setgt (shl (i32 IntRegs:$src1), (i32 16)),
3204 s8ExtPred:$src2))]>,
3208 // Signed half compare(.gt) rr.
3209 // Pd=cmph.gt(Rs,Rt)
3210 let isCompare = 1, validSubTargets = HasV4SubT in
3211 def CMPhGTrr_shl_V4 : MInst<(outs PredRegs:$dst),
3212 (ins IntRegs:$src1, IntRegs:$src2),
3213 "$dst = cmph.gt($src1, $src2)",
3214 [(set (i1 PredRegs:$dst),
3215 (setgt (shl (i32 IntRegs:$src1), (i32 16)),
3216 (shl (i32 IntRegs:$src2), (i32 16))))]>,
3219 // Unsigned half compare rr (.gtu).
3220 // Pd=cmph.gtu(Rs,Rt)
3221 let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU",
3222 InputType = "reg" in
3223 def CMPhGTUrr_V4 : MInst<(outs PredRegs:$dst),
3224 (ins IntRegs:$src1, IntRegs:$src2),
3225 "$dst = cmph.gtu($src1, $src2)",
3226 [(set (i1 PredRegs:$dst),
3227 (setugt (and (i32 IntRegs:$src1), 65535),
3228 (and (i32 IntRegs:$src2), 65535)))]>,
3229 Requires<[HasV4T]>, ImmRegRel;
3231 // Unsigned half compare ri (.gtu).
3232 // Pd=cmph.gtu(Rs,#u7)
3233 let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7,
3234 isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU",
3235 InputType = "imm" in
3236 def CMPhGTUri_V4 : MInst<(outs PredRegs:$dst),
3237 (ins IntRegs:$src1, u7Ext:$src2),
3238 "$dst = cmph.gtu($src1, #$src2)",
3239 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 65535),
3240 u7ExtPred:$src2))]>,
3241 Requires<[HasV4T]>, ImmRegRel;
3243 let validSubTargets = HasV4SubT in
3244 def NTSTBIT_rr : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
3245 "$dst = !tstbit($src1, $src2)",
3246 [(set (i1 PredRegs:$dst),
3247 (seteq (and (shl 1, (i32 IntRegs:$src2)), (i32 IntRegs:$src1)), 0))]>,
3250 let validSubTargets = HasV4SubT in
3251 def NTSTBIT_ri : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
3252 "$dst = !tstbit($src1, $src2)",
3253 [(set (i1 PredRegs:$dst),
3254 (seteq (and (shl 1, u5ImmPred:$src2), (i32 IntRegs:$src1)), 0))]>,
3257 //===----------------------------------------------------------------------===//
3259 //===----------------------------------------------------------------------===//
3261 //Deallocate frame and return.
3263 let isReturn = 1, isTerminator = 1, isBarrier = 1, isPredicable = 1,
3264 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0 in {
3265 let validSubTargets = HasV4SubT in
3266 def DEALLOC_RET_V4 : LD0Inst<(outs), (ins),
3272 // Restore registers and dealloc return function call.
3273 let isCall = 1, isBarrier = 1, isReturn = 1, isTerminator = 1,
3274 Defs = [R29, R30, R31, PC] in {
3275 let validSubTargets = HasV4SubT in
3276 def RESTORE_DEALLOC_RET_JMP_V4 : JInst<(outs),
3277 (ins calltarget:$dst),
3283 // Restore registers and dealloc frame before a tail call.
3284 let isCall = 1, isBarrier = 1,
3285 Defs = [R29, R30, R31, PC] in {
3286 let validSubTargets = HasV4SubT in
3287 def RESTORE_DEALLOC_BEFORE_TAILCALL_V4 : JInst<(outs),
3288 (ins calltarget:$dst),
3294 // Save registers function call.
3295 let isCall = 1, isBarrier = 1,
3296 Uses = [R29, R31] in {
3297 def SAVE_REGISTERS_CALL_V4 : JInst<(outs),
3298 (ins calltarget:$dst),
3299 "call $dst // Save_calle_saved_registers",
3304 // if (Ps) dealloc_return
3305 let isReturn = 1, isTerminator = 1,
3306 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3307 isPredicated = 1 in {
3308 let validSubTargets = HasV4SubT in
3309 def DEALLOC_RET_cPt_V4 : LD0Inst<(outs),
3310 (ins PredRegs:$src1),
3311 "if ($src1) dealloc_return",
3316 // if (!Ps) dealloc_return
3317 let isReturn = 1, isTerminator = 1,
3318 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3319 isPredicated = 1, isPredicatedFalse = 1 in {
3320 let validSubTargets = HasV4SubT in
3321 def DEALLOC_RET_cNotPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
3322 "if (!$src1) dealloc_return",
3327 // if (Ps.new) dealloc_return:nt
3328 let isReturn = 1, isTerminator = 1,
3329 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3330 isPredicated = 1 in {
3331 let validSubTargets = HasV4SubT in
3332 def DEALLOC_RET_cdnPnt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
3333 "if ($src1.new) dealloc_return:nt",
3338 // if (!Ps.new) dealloc_return:nt
3339 let isReturn = 1, isTerminator = 1,
3340 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3341 isPredicated = 1, isPredicatedFalse = 1 in {
3342 let validSubTargets = HasV4SubT in
3343 def DEALLOC_RET_cNotdnPnt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
3344 "if (!$src1.new) dealloc_return:nt",
3349 // if (Ps.new) dealloc_return:t
3350 let isReturn = 1, isTerminator = 1,
3351 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3352 isPredicated = 1 in {
3353 let validSubTargets = HasV4SubT in
3354 def DEALLOC_RET_cdnPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
3355 "if ($src1.new) dealloc_return:t",
3360 // if (!Ps.new) dealloc_return:nt
3361 let isReturn = 1, isTerminator = 1,
3362 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3363 isPredicated = 1, isPredicatedFalse = 1 in {
3364 let validSubTargets = HasV4SubT in
3365 def DEALLOC_RET_cNotdnPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
3366 "if (!$src1.new) dealloc_return:t",
3371 // Load/Store with absolute addressing mode
3374 multiclass ST_Abs_Predbase<string mnemonic, RegisterClass RC, bit isNot,
3376 let isPredicatedNew = isPredNew in
3377 def NAME#_V4 : STInst2<(outs),
3378 (ins PredRegs:$src1, u0AlwaysExt:$absaddr, RC: $src2),
3379 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
3380 ") ")#mnemonic#"(##$absaddr) = $src2",
3385 multiclass ST_Abs_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
3386 let isPredicatedFalse = PredNot in {
3387 defm _c#NAME : ST_Abs_Predbase<mnemonic, RC, PredNot, 0>;
3389 defm _cdn#NAME : ST_Abs_Predbase<mnemonic, RC, PredNot, 1>;
3393 let isNVStorable = 1, isExtended = 1, hasSideEffects = 0 in
3394 multiclass ST_Abs<string mnemonic, string CextOp, RegisterClass RC> {
3395 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
3396 let opExtendable = 0, isPredicable = 1 in
3397 def NAME#_V4 : STInst2<(outs),
3398 (ins u0AlwaysExt:$absaddr, RC:$src),
3399 mnemonic#"(##$absaddr) = $src",
3403 let opExtendable = 1, isPredicated = 1 in {
3404 defm Pt : ST_Abs_Pred<mnemonic, RC, 0>;
3405 defm NotPt : ST_Abs_Pred<mnemonic, RC, 1>;
3410 multiclass ST_Abs_Predbase_nv<string mnemonic, RegisterClass RC, bit isNot,
3412 let isPredicatedNew = isPredNew in
3413 def NAME#_nv_V4 : NVInst_V4<(outs),
3414 (ins PredRegs:$src1, u0AlwaysExt:$absaddr, RC: $src2),
3415 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
3416 ") ")#mnemonic#"(##$absaddr) = $src2.new",
3421 multiclass ST_Abs_Pred_nv<string mnemonic, RegisterClass RC, bit PredNot> {
3422 let isPredicatedFalse = PredNot in {
3423 defm _c#NAME : ST_Abs_Predbase_nv<mnemonic, RC, PredNot, 0>;
3425 defm _cdn#NAME : ST_Abs_Predbase_nv<mnemonic, RC, PredNot, 1>;
3429 let mayStore = 1, isNVStore = 1, isExtended = 1, hasSideEffects = 0 in
3430 multiclass ST_Abs_nv<string mnemonic, string CextOp, RegisterClass RC> {
3431 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
3432 let opExtendable = 0, isPredicable = 1 in
3433 def NAME#_nv_V4 : NVInst_V4<(outs),
3434 (ins u0AlwaysExt:$absaddr, RC:$src),
3435 mnemonic#"(##$absaddr) = $src.new",
3439 let opExtendable = 1, isPredicated = 1 in {
3440 defm Pt : ST_Abs_Pred_nv<mnemonic, RC, 0>;
3441 defm NotPt : ST_Abs_Pred_nv<mnemonic, RC, 1>;
3446 let addrMode = Absolute in {
3447 let accessSize = ByteAccess in
3448 defm STrib_abs : ST_Abs<"memb", "STrib", IntRegs>,
3449 ST_Abs_nv<"memb", "STrib", IntRegs>, AddrModeRel;
3451 let accessSize = HalfWordAccess in
3452 defm STrih_abs : ST_Abs<"memh", "STrih", IntRegs>,
3453 ST_Abs_nv<"memh", "STrih", IntRegs>, AddrModeRel;
3455 let accessSize = WordAccess in
3456 defm STriw_abs : ST_Abs<"memw", "STriw", IntRegs>,
3457 ST_Abs_nv<"memw", "STriw", IntRegs>, AddrModeRel;
3459 let accessSize = DoubleWordAccess, isNVStorable = 0 in
3460 defm STrid_abs : ST_Abs<"memd", "STrid", DoubleRegs>, AddrModeRel;
3463 let Predicates = [HasV4T], AddedComplexity = 30 in {
3464 def : Pat<(truncstorei8 (i32 IntRegs:$src1),
3465 (HexagonCONST32 tglobaladdr:$absaddr)),
3466 (STrib_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
3468 def : Pat<(truncstorei16 (i32 IntRegs:$src1),
3469 (HexagonCONST32 tglobaladdr:$absaddr)),
3470 (STrih_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
3472 def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32 tglobaladdr:$absaddr)),
3473 (STriw_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
3475 def : Pat<(store (i64 DoubleRegs:$src1),
3476 (HexagonCONST32 tglobaladdr:$absaddr)),
3477 (STrid_abs_V4 tglobaladdr: $absaddr, DoubleRegs: $src1)>;
3480 //===----------------------------------------------------------------------===//
3481 // multiclass for store instructions with GP-relative addressing mode.
3482 // mem[bhwd](#global)=Rt
3483 // if ([!]Pv[.new]) mem[bhwd](##global) = Rt
3484 //===----------------------------------------------------------------------===//
3485 let mayStore = 1, isNVStorable = 1 in
3486 multiclass ST_GP<string mnemonic, string BaseOp, RegisterClass RC> {
3487 let BaseOpcode = BaseOp, isPredicable = 1 in
3488 def NAME#_V4 : STInst2<(outs),
3489 (ins globaladdress:$global, RC:$src),
3490 mnemonic#"(#$global) = $src",
3493 // When GP-relative instructions are predicated, their addressing mode is
3494 // changed to absolute and they are always constant extended.
3495 let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1,
3496 isPredicated = 1 in {
3497 defm Pt : ST_Abs_Pred <mnemonic, RC, 0>;
3498 defm NotPt : ST_Abs_Pred <mnemonic, RC, 1>;
3502 let mayStore = 1, isNVStore = 1 in
3503 multiclass ST_GP_nv<string mnemonic, string BaseOp, RegisterClass RC> {
3504 let BaseOpcode = BaseOp, isPredicable = 1 in
3505 def NAME#_nv_V4 : NVInst_V4<(outs),
3506 (ins u0AlwaysExt:$global, RC:$src),
3507 mnemonic#"(#$global) = $src.new",
3511 // When GP-relative instructions are predicated, their addressing mode is
3512 // changed to absolute and they are always constant extended.
3513 let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1,
3514 isPredicated = 1 in {
3515 defm Pt : ST_Abs_Pred_nv<mnemonic, RC, 0>;
3516 defm NotPt : ST_Abs_Pred_nv<mnemonic, RC, 1>;
3520 let validSubTargets = HasV4SubT, hasSideEffects = 0 in {
3521 let isNVStorable = 0 in
3522 defm STd_GP : ST_GP <"memd", "STd_GP", DoubleRegs>, PredNewRel;
3524 defm STb_GP : ST_GP<"memb", "STb_GP", IntRegs>,
3525 ST_GP_nv<"memb", "STb_GP", IntRegs>, NewValueRel;
3526 defm STh_GP : ST_GP<"memh", "STh_GP", IntRegs>,
3527 ST_GP_nv<"memh", "STh_GP", IntRegs>, NewValueRel;
3528 defm STw_GP : ST_GP<"memw", "STw_GP", IntRegs>,
3529 ST_GP_nv<"memw", "STw_GP", IntRegs>, NewValueRel;
3532 // 64 bit atomic store
3533 def : Pat <(atomic_store_64 (HexagonCONST32_GP tglobaladdr:$global),
3534 (i64 DoubleRegs:$src1)),
3535 (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>,
3538 // Map from store(globaladdress) -> memd(#foo)
3539 let AddedComplexity = 100 in
3540 def : Pat <(store (i64 DoubleRegs:$src1),
3541 (HexagonCONST32_GP tglobaladdr:$global)),
3542 (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>;
3544 // 8 bit atomic store
3545 def : Pat < (atomic_store_8 (HexagonCONST32_GP tglobaladdr:$global),
3546 (i32 IntRegs:$src1)),
3547 (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3549 // Map from store(globaladdress) -> memb(#foo)
3550 let AddedComplexity = 100 in
3551 def : Pat<(truncstorei8 (i32 IntRegs:$src1),
3552 (HexagonCONST32_GP tglobaladdr:$global)),
3553 (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3555 // Map from "i1 = constant<-1>; memw(CONST32(#foo)) = i1"
3556 // to "r0 = 1; memw(#foo) = r0"
3557 let AddedComplexity = 100 in
3558 def : Pat<(store (i1 -1), (HexagonCONST32_GP tglobaladdr:$global)),
3559 (STb_GP_V4 tglobaladdr:$global, (A2_tfrsi 1))>;
3561 def : Pat<(atomic_store_16 (HexagonCONST32_GP tglobaladdr:$global),
3562 (i32 IntRegs:$src1)),
3563 (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3565 // Map from store(globaladdress) -> memh(#foo)
3566 let AddedComplexity = 100 in
3567 def : Pat<(truncstorei16 (i32 IntRegs:$src1),
3568 (HexagonCONST32_GP tglobaladdr:$global)),
3569 (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3571 // 32 bit atomic store
3572 def : Pat<(atomic_store_32 (HexagonCONST32_GP tglobaladdr:$global),
3573 (i32 IntRegs:$src1)),
3574 (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3576 // Map from store(globaladdress) -> memw(#foo)
3577 let AddedComplexity = 100 in
3578 def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)),
3579 (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3581 //===----------------------------------------------------------------------===//
3582 // Multiclass for the load instructions with absolute addressing mode.
3583 //===----------------------------------------------------------------------===//
3584 multiclass LD_Abs_Predbase<string mnemonic, RegisterClass RC, bit isNot,
3586 let isPredicatedNew = isPredNew in
3587 def NAME : LDInst2<(outs RC:$dst),
3588 (ins PredRegs:$src1, u0AlwaysExt:$absaddr),
3589 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
3590 ") ")#"$dst = "#mnemonic#"(##$absaddr)",
3595 multiclass LD_Abs_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
3596 let isPredicatedFalse = PredNot in {
3597 defm _c#NAME : LD_Abs_Predbase<mnemonic, RC, PredNot, 0>;
3599 defm _cdn#NAME : LD_Abs_Predbase<mnemonic, RC, PredNot, 1>;
3603 let isExtended = 1, hasSideEffects = 0 in
3604 multiclass LD_Abs<string mnemonic, string CextOp, RegisterClass RC> {
3605 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
3606 let opExtendable = 1, isPredicable = 1 in
3607 def NAME#_V4 : LDInst2<(outs RC:$dst),
3608 (ins u0AlwaysExt:$absaddr),
3609 "$dst = "#mnemonic#"(##$absaddr)",
3613 let opExtendable = 2, isPredicated = 1 in {
3614 defm Pt_V4 : LD_Abs_Pred<mnemonic, RC, 0>;
3615 defm NotPt_V4 : LD_Abs_Pred<mnemonic, RC, 1>;
3620 let addrMode = Absolute in {
3621 let accessSize = ByteAccess in {
3622 defm LDrib_abs : LD_Abs<"memb", "LDrib", IntRegs>, AddrModeRel;
3623 defm LDriub_abs : LD_Abs<"memub", "LDriub", IntRegs>, AddrModeRel;
3625 let accessSize = HalfWordAccess in {
3626 defm LDrih_abs : LD_Abs<"memh", "LDrih", IntRegs>, AddrModeRel;
3627 defm LDriuh_abs : LD_Abs<"memuh", "LDriuh", IntRegs>, AddrModeRel;
3629 let accessSize = WordAccess in
3630 defm LDriw_abs : LD_Abs<"memw", "LDriw", IntRegs>, AddrModeRel;
3632 let accessSize = DoubleWordAccess in
3633 defm LDrid_abs : LD_Abs<"memd", "LDrid", DoubleRegs>, AddrModeRel;
3636 let Predicates = [HasV4T], AddedComplexity = 30 in {
3637 def : Pat<(i32 (load (HexagonCONST32 tglobaladdr:$absaddr))),
3638 (LDriw_abs_V4 tglobaladdr: $absaddr)>;
3640 def : Pat<(i32 (sextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))),
3641 (LDrib_abs_V4 tglobaladdr:$absaddr)>;
3643 def : Pat<(i32 (zextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))),
3644 (LDriub_abs_V4 tglobaladdr:$absaddr)>;
3646 def : Pat<(i32 (sextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
3647 (LDrih_abs_V4 tglobaladdr:$absaddr)>;
3649 def : Pat<(i32 (zextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
3650 (LDriuh_abs_V4 tglobaladdr:$absaddr)>;
3653 //===----------------------------------------------------------------------===//
3654 // multiclass for load instructions with GP-relative addressing mode.
3655 // Rx=mem[bhwd](##global)
3656 // if ([!]Pv[.new]) Rx=mem[bhwd](##global)
3657 //===----------------------------------------------------------------------===//
3658 let hasSideEffects = 0, validSubTargets = HasV4SubT in
3659 multiclass LD_GP<string mnemonic, string BaseOp, RegisterClass RC> {
3660 let BaseOpcode = BaseOp in {
3661 let isPredicable = 1 in
3662 def NAME#_V4 : LDInst2<(outs RC:$dst),
3663 (ins globaladdress:$global),
3664 "$dst = "#mnemonic#"(#$global)",
3667 let isExtended = 1, opExtendable = 2, isPredicated = 1 in {
3668 defm Pt_V4 : LD_Abs_Pred<mnemonic, RC, 0>;
3669 defm NotPt_V4 : LD_Abs_Pred<mnemonic, RC, 1>;
3674 defm LDd_GP : LD_GP<"memd", "LDd_GP", DoubleRegs>, PredNewRel;
3675 defm LDb_GP : LD_GP<"memb", "LDb_GP", IntRegs>, PredNewRel;
3676 defm LDub_GP : LD_GP<"memub", "LDub_GP", IntRegs>, PredNewRel;
3677 defm LDh_GP : LD_GP<"memh", "LDh_GP", IntRegs>, PredNewRel;
3678 defm LDuh_GP : LD_GP<"memuh", "LDuh_GP", IntRegs>, PredNewRel;
3679 defm LDw_GP : LD_GP<"memw", "LDw_GP", IntRegs>, PredNewRel;
3681 def : Pat <(atomic_load_64 (HexagonCONST32_GP tglobaladdr:$global)),
3682 (i64 (LDd_GP_V4 tglobaladdr:$global))>;
3684 def : Pat <(atomic_load_32 (HexagonCONST32_GP tglobaladdr:$global)),
3685 (i32 (LDw_GP_V4 tglobaladdr:$global))>;
3687 def : Pat <(atomic_load_16 (HexagonCONST32_GP tglobaladdr:$global)),
3688 (i32 (LDuh_GP_V4 tglobaladdr:$global))>;
3690 def : Pat <(atomic_load_8 (HexagonCONST32_GP tglobaladdr:$global)),
3691 (i32 (LDub_GP_V4 tglobaladdr:$global))>;
3693 // Map from load(globaladdress) -> memw(#foo + 0)
3694 let AddedComplexity = 100 in
3695 def : Pat <(i64 (load (HexagonCONST32_GP tglobaladdr:$global))),
3696 (i64 (LDd_GP_V4 tglobaladdr:$global))>;
3698 // Map from Pd = load(globaladdress) -> Rd = memb(globaladdress), Pd = Rd
3699 let AddedComplexity = 100 in
3700 def : Pat <(i1 (load (HexagonCONST32_GP tglobaladdr:$global))),
3701 (i1 (C2_tfrrp (i32 (LDb_GP_V4 tglobaladdr:$global))))>;
3703 // When the Interprocedural Global Variable optimizer realizes that a certain
3704 // global variable takes only two constant values, it shrinks the global to
3705 // a boolean. Catch those loads here in the following 3 patterns.
3706 let AddedComplexity = 100 in
3707 def : Pat <(i32 (extloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3708 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3710 let AddedComplexity = 100 in
3711 def : Pat <(i32 (sextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3712 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3714 // Map from load(globaladdress) -> memb(#foo)
3715 let AddedComplexity = 100 in
3716 def : Pat <(i32 (extloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3717 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3719 // Map from load(globaladdress) -> memb(#foo)
3720 let AddedComplexity = 100 in
3721 def : Pat <(i32 (sextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3722 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3724 let AddedComplexity = 100 in
3725 def : Pat <(i32 (zextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3726 (i32 (LDub_GP_V4 tglobaladdr:$global))>;
3728 // Map from load(globaladdress) -> memub(#foo)
3729 let AddedComplexity = 100 in
3730 def : Pat <(i32 (zextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3731 (i32 (LDub_GP_V4 tglobaladdr:$global))>;
3733 // Map from load(globaladdress) -> memh(#foo)
3734 let AddedComplexity = 100 in
3735 def : Pat <(i32 (extloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3736 (i32 (LDh_GP_V4 tglobaladdr:$global))>;
3738 // Map from load(globaladdress) -> memh(#foo)
3739 let AddedComplexity = 100 in
3740 def : Pat <(i32 (sextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3741 (i32 (LDh_GP_V4 tglobaladdr:$global))>;
3743 // Map from load(globaladdress) -> memuh(#foo)
3744 let AddedComplexity = 100 in
3745 def : Pat <(i32 (zextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3746 (i32 (LDuh_GP_V4 tglobaladdr:$global))>;
3748 // Map from load(globaladdress) -> memw(#foo)
3749 let AddedComplexity = 100 in
3750 def : Pat <(i32 (load (HexagonCONST32_GP tglobaladdr:$global))),
3751 (i32 (LDw_GP_V4 tglobaladdr:$global))>;
3754 // Transfer global address into a register
3755 let isExtended = 1, opExtendable = 1, AddedComplexity=50, isMoveImm = 1,
3756 isAsCheapAsAMove = 1, isReMaterializable = 1, validSubTargets = HasV4SubT in
3757 def TFRI_V4 : ALU32_ri<(outs IntRegs:$dst), (ins s16Ext:$src1),
3759 [(set IntRegs:$dst, (HexagonCONST32 tglobaladdr:$src1))]>,
3762 // Transfer a block address into a register
3763 def : Pat<(HexagonCONST32_GP tblockaddress:$src1),
3764 (TFRI_V4 tblockaddress:$src1)>,
3767 let isExtended = 1, opExtendable = 2, AddedComplexity=50,
3768 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3769 def TFRI_cPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3770 (ins PredRegs:$src1, s16Ext:$src2),
3771 "if($src1) $dst = #$src2",
3775 let isExtended = 1, opExtendable = 2, AddedComplexity=50, isPredicatedFalse = 1,
3776 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3777 def TFRI_cNotPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3778 (ins PredRegs:$src1, s16Ext:$src2),
3779 "if(!$src1) $dst = #$src2",
3783 let isExtended = 1, opExtendable = 2, AddedComplexity=50,
3784 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3785 def TFRI_cdnPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3786 (ins PredRegs:$src1, s16Ext:$src2),
3787 "if($src1.new) $dst = #$src2",
3791 let isExtended = 1, opExtendable = 2, AddedComplexity=50, isPredicatedFalse = 1,
3792 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3793 def TFRI_cdnNotPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3794 (ins PredRegs:$src1, s16Ext:$src2),
3795 "if(!$src1.new) $dst = #$src2",
3799 let AddedComplexity = 50, Predicates = [HasV4T] in
3800 def : Pat<(HexagonCONST32_GP tglobaladdr:$src1),
3801 (TFRI_V4 tglobaladdr:$src1)>,
3805 // Load - Indirect with long offset: These instructions take global address
3807 let isExtended = 1, opExtendable = 3, AddedComplexity = 40,
3808 validSubTargets = HasV4SubT in
3809 def LDrid_ind_lo_V4 : LDInst<(outs DoubleRegs:$dst),
3810 (ins IntRegs:$src1, u2Imm:$src2, globaladdressExt:$offset),
3811 "$dst=memd($src1<<#$src2+##$offset)",
3812 [(set (i64 DoubleRegs:$dst),
3813 (load (add (shl IntRegs:$src1, u2ImmPred:$src2),
3814 (HexagonCONST32 tglobaladdr:$offset))))]>,
3817 let AddedComplexity = 40 in
3818 multiclass LD_indirect_lo<string OpcStr, PatFrag OpNode> {
3819 let isExtended = 1, opExtendable = 3, validSubTargets = HasV4SubT in
3820 def _lo_V4 : LDInst<(outs IntRegs:$dst),
3821 (ins IntRegs:$src1, u2Imm:$src2, globaladdressExt:$offset),
3822 !strconcat("$dst = ",
3823 !strconcat(OpcStr, "($src1<<#$src2+##$offset)")),
3825 (i32 (OpNode (add (shl IntRegs:$src1, u2ImmPred:$src2),
3826 (HexagonCONST32 tglobaladdr:$offset)))))]>,
3830 defm LDrib_ind : LD_indirect_lo<"memb", sextloadi8>;
3831 defm LDriub_ind : LD_indirect_lo<"memub", zextloadi8>;
3832 defm LDriub_ind_anyext : LD_indirect_lo<"memub", extloadi8>;
3833 defm LDrih_ind : LD_indirect_lo<"memh", sextloadi16>;
3834 defm LDriuh_ind : LD_indirect_lo<"memuh", zextloadi16>;
3835 defm LDriuh_ind_anyext : LD_indirect_lo<"memuh", extloadi16>;
3836 defm LDriw_ind : LD_indirect_lo<"memw", load>;
3838 let AddedComplexity = 40 in
3839 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1,
3840 (NumUsesBelowThresCONST32 tglobaladdr:$offset)))),
3841 (i32 (LDrib_ind_lo_V4 IntRegs:$src1, 0, tglobaladdr:$offset))>,
3844 let AddedComplexity = 40 in
3845 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1,
3846 (NumUsesBelowThresCONST32 tglobaladdr:$offset)))),
3847 (i32 (LDriub_ind_lo_V4 IntRegs:$src1, 0, tglobaladdr:$offset))>,
3850 let Predicates = [HasV4T], AddedComplexity = 30 in {
3851 def : Pat<(truncstorei8 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3852 (STrib_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3854 def : Pat<(truncstorei16 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3855 (STrih_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3857 def : Pat<(store (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3858 (STriw_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3861 let Predicates = [HasV4T], AddedComplexity = 30 in {
3862 def : Pat<(i32 (load u0AlwaysExtPred:$src)),
3863 (LDriw_abs_V4 u0AlwaysExtPred:$src)>;
3865 def : Pat<(i32 (sextloadi8 u0AlwaysExtPred:$src)),
3866 (LDrib_abs_V4 u0AlwaysExtPred:$src)>;
3868 def : Pat<(i32 (zextloadi8 u0AlwaysExtPred:$src)),
3869 (LDriub_abs_V4 u0AlwaysExtPred:$src)>;
3871 def : Pat<(i32 (sextloadi16 u0AlwaysExtPred:$src)),
3872 (LDrih_abs_V4 u0AlwaysExtPred:$src)>;
3874 def : Pat<(i32 (zextloadi16 u0AlwaysExtPred:$src)),
3875 (LDriuh_abs_V4 u0AlwaysExtPred:$src)>;
3878 // Indexed store word - global address.
3879 // memw(Rs+#u6:2)=#S8
3880 let AddedComplexity = 10 in
3881 def STriw_offset_ext_V4 : STInst<(outs),
3882 (ins IntRegs:$src1, u6_2Imm:$src2, globaladdress:$src3),
3883 "memw($src1+#$src2) = ##$src3",
3884 [(store (HexagonCONST32 tglobaladdr:$src3),
3885 (add IntRegs:$src1, u6_2ImmPred:$src2))]>,
3888 def : Pat<(i64 (ctlz (i64 DoubleRegs:$src1))),
3889 (i64 (A4_combineir (i32 0), (i32 (CTLZ64_rr DoubleRegs:$src1))))>,
3892 def : Pat<(i64 (cttz (i64 DoubleRegs:$src1))),
3893 (i64 (A4_combineir (i32 0), (i32 (CTTZ64_rr DoubleRegs:$src1))))>,
3898 // We need a complexity of 120 here to override preceding handling of
3900 let Predicates = [HasV4T], AddedComplexity = 120 in {
3901 def: Pat <(i64 (extloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3902 (i64 (A4_combineir 0, (LDrib_abs_V4 tglobaladdr:$addr)))>;
3904 def: Pat <(i64 (zextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3905 (i64 (A4_combineir 0, (LDriub_abs_V4 tglobaladdr:$addr)))>;
3907 def: Pat <(i64 (sextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3908 (i64 (A2_sxtw (LDrib_abs_V4 tglobaladdr:$addr)))>;
3910 def: Pat <(i64 (extloadi8 FoldGlobalAddr:$addr)),
3911 (i64 (A4_combineir 0, (LDrib_abs_V4 FoldGlobalAddr:$addr)))>;
3913 def: Pat <(i64 (zextloadi8 FoldGlobalAddr:$addr)),
3914 (i64 (A4_combineir 0, (LDriub_abs_V4 FoldGlobalAddr:$addr)))>;
3916 def: Pat <(i64 (sextloadi8 FoldGlobalAddr:$addr)),
3917 (i64 (A2_sxtw (LDrib_abs_V4 FoldGlobalAddr:$addr)))>;
3920 // We need a complexity of 120 here to override preceding handling of
3922 let AddedComplexity = 120 in {
3923 def: Pat <(i64 (extloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3924 (i64 (A4_combineir 0, (LDrih_abs_V4 tglobaladdr:$addr)))>,
3927 def: Pat <(i64 (zextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3928 (i64 (A4_combineir 0, (LDriuh_abs_V4 tglobaladdr:$addr)))>,
3931 def: Pat <(i64 (sextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3932 (i64 (A2_sxtw (LDrih_abs_V4 tglobaladdr:$addr)))>,
3935 def: Pat <(i64 (extloadi16 FoldGlobalAddr:$addr)),
3936 (i64 (A4_combineir 0, (LDrih_abs_V4 FoldGlobalAddr:$addr)))>,
3939 def: Pat <(i64 (zextloadi16 FoldGlobalAddr:$addr)),
3940 (i64 (A4_combineir 0, (LDriuh_abs_V4 FoldGlobalAddr:$addr)))>,
3943 def: Pat <(i64 (sextloadi16 FoldGlobalAddr:$addr)),
3944 (i64 (A2_sxtw (LDrih_abs_V4 FoldGlobalAddr:$addr)))>,
3948 // We need a complexity of 120 here to override preceding handling of
3950 let AddedComplexity = 120 in {
3951 def: Pat <(i64 (extloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3952 (i64 (A4_combineir 0, (LDriw_abs_V4 tglobaladdr:$addr)))>,
3955 def: Pat <(i64 (zextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3956 (i64 (A4_combineir 0, (LDriw_abs_V4 tglobaladdr:$addr)))>,
3959 def: Pat <(i64 (sextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3960 (i64 (A2_sxtw (LDriw_abs_V4 tglobaladdr:$addr)))>,
3963 def: Pat <(i64 (extloadi32 FoldGlobalAddr:$addr)),
3964 (i64 (A4_combineir 0, (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
3967 def: Pat <(i64 (zextloadi32 FoldGlobalAddr:$addr)),
3968 (i64 (A4_combineir 0, (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
3971 def: Pat <(i64 (sextloadi32 FoldGlobalAddr:$addr)),
3972 (i64 (A2_sxtw (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
3976 // Indexed store double word - global address.
3977 // memw(Rs+#u6:2)=#S8
3978 let AddedComplexity = 10 in
3979 def STrih_offset_ext_V4 : STInst<(outs),
3980 (ins IntRegs:$src1, u6_1Imm:$src2, globaladdress:$src3),
3981 "memh($src1+#$src2) = ##$src3",
3982 [(truncstorei16 (HexagonCONST32 tglobaladdr:$src3),
3983 (add IntRegs:$src1, u6_1ImmPred:$src2))]>,
3985 // Map from store(globaladdress + x) -> memd(#foo + x)
3986 let AddedComplexity = 100 in
3987 def : Pat<(store (i64 DoubleRegs:$src1),
3988 FoldGlobalAddrGP:$addr),
3989 (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
3992 def : Pat<(atomic_store_64 FoldGlobalAddrGP:$addr,
3993 (i64 DoubleRegs:$src1)),
3994 (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
3997 // Map from store(globaladdress + x) -> memb(#foo + x)
3998 let AddedComplexity = 100 in
3999 def : Pat<(truncstorei8 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
4000 (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
4003 def : Pat<(atomic_store_8 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
4004 (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
4007 // Map from store(globaladdress + x) -> memh(#foo + x)
4008 let AddedComplexity = 100 in
4009 def : Pat<(truncstorei16 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
4010 (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
4013 def : Pat<(atomic_store_16 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
4014 (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
4017 // Map from store(globaladdress + x) -> memw(#foo + x)
4018 let AddedComplexity = 100 in
4019 def : Pat<(store (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
4020 (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
4023 def : Pat<(atomic_store_32 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
4024 (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
4027 // Map from load(globaladdress + x) -> memd(#foo + x)
4028 let AddedComplexity = 100 in
4029 def : Pat<(i64 (load FoldGlobalAddrGP:$addr)),
4030 (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>,
4033 def : Pat<(atomic_load_64 FoldGlobalAddrGP:$addr),
4034 (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>,
4037 // Map from load(globaladdress + x) -> memb(#foo + x)
4038 let AddedComplexity = 100 in
4039 def : Pat<(i32 (extloadi8 FoldGlobalAddrGP:$addr)),
4040 (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>,
4043 // Map from load(globaladdress + x) -> memb(#foo + x)
4044 let AddedComplexity = 100 in
4045 def : Pat<(i32 (sextloadi8 FoldGlobalAddrGP:$addr)),
4046 (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>,
4049 //let AddedComplexity = 100 in
4050 let AddedComplexity = 100 in
4051 def : Pat<(i32 (extloadi16 FoldGlobalAddrGP:$addr)),
4052 (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>,
4055 // Map from load(globaladdress + x) -> memh(#foo + x)
4056 let AddedComplexity = 100 in
4057 def : Pat<(i32 (sextloadi16 FoldGlobalAddrGP:$addr)),
4058 (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>,
4061 // Map from load(globaladdress + x) -> memuh(#foo + x)
4062 let AddedComplexity = 100 in
4063 def : Pat<(i32 (zextloadi16 FoldGlobalAddrGP:$addr)),
4064 (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>,
4067 def : Pat<(atomic_load_16 FoldGlobalAddrGP:$addr),
4068 (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>,
4071 // Map from load(globaladdress + x) -> memub(#foo + x)
4072 let AddedComplexity = 100 in
4073 def : Pat<(i32 (zextloadi8 FoldGlobalAddrGP:$addr)),
4074 (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>,
4077 def : Pat<(atomic_load_8 FoldGlobalAddrGP:$addr),
4078 (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>,
4081 // Map from load(globaladdress + x) -> memw(#foo + x)
4082 let AddedComplexity = 100 in
4083 def : Pat<(i32 (load FoldGlobalAddrGP:$addr)),
4084 (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>,
4087 def : Pat<(atomic_load_32 FoldGlobalAddrGP:$addr),
4088 (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>,