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=##$addr)",
338 def LDrid_abs_set_V4 : T_LD_abs_set <"memd", DoubleRegs>;
339 def LDrib_abs_set_V4 : T_LD_abs_set <"memb", IntRegs>;
340 def LDriub_abs_set_V4 : T_LD_abs_set <"memub", IntRegs>;
341 def LDrih_abs_set_V4 : T_LD_abs_set <"memh", IntRegs>;
342 def LDriw_abs_set_V4 : T_LD_abs_set <"memw", IntRegs>;
343 def LDriuh_abs_set_V4 : T_LD_abs_set <"memuh", IntRegs>;
346 // multiclass for load instructions with base + register offset
348 multiclass ld_idxd_shl_pbase<string mnemonic, RegisterClass RC, bit isNot,
350 let isPredicatedNew = isPredNew in
351 def NAME : LDInst2<(outs RC:$dst),
352 (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$offset),
353 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
354 ") ")#"$dst = "#mnemonic#"($src2+$src3<<#$offset)",
355 []>, Requires<[HasV4T]>;
358 multiclass ld_idxd_shl_pred<string mnemonic, RegisterClass RC, bit PredNot> {
359 let isPredicatedFalse = PredNot in {
360 defm _c#NAME : ld_idxd_shl_pbase<mnemonic, RC, PredNot, 0>;
362 defm _cdn#NAME : ld_idxd_shl_pbase<mnemonic, RC, PredNot, 1>;
366 let hasSideEffects = 0 in
367 multiclass ld_idxd_shl<string mnemonic, string CextOp, RegisterClass RC> {
368 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
369 let isPredicable = 1 in
370 def NAME#_V4 : LDInst2<(outs RC:$dst),
371 (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset),
372 "$dst = "#mnemonic#"($src1+$src2<<#$offset)",
373 []>, Requires<[HasV4T]>;
375 let isPredicated = 1 in {
376 defm Pt_V4 : ld_idxd_shl_pred<mnemonic, RC, 0 >;
377 defm NotPt_V4 : ld_idxd_shl_pred<mnemonic, RC, 1>;
382 let addrMode = BaseRegOffset in {
383 let accessSize = ByteAccess in {
384 defm LDrib_indexed_shl: ld_idxd_shl<"memb", "LDrib", IntRegs>,
386 defm LDriub_indexed_shl: ld_idxd_shl<"memub", "LDriub", IntRegs>,
389 let accessSize = HalfWordAccess in {
390 defm LDrih_indexed_shl: ld_idxd_shl<"memh", "LDrih", IntRegs>, AddrModeRel;
391 defm LDriuh_indexed_shl: ld_idxd_shl<"memuh", "LDriuh", IntRegs>,
394 let accessSize = WordAccess in
395 defm LDriw_indexed_shl: ld_idxd_shl<"memw", "LDriw", IntRegs>, AddrModeRel;
397 let accessSize = DoubleWordAccess in
398 defm LDrid_indexed_shl: ld_idxd_shl<"memd", "LDrid", DoubleRegs>,
402 // 'def pats' for load instructions with base + register offset and non-zero
403 // immediate value. Immediate value is used to left-shift the second
405 let AddedComplexity = 40 in {
406 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1,
407 (shl IntRegs:$src2, u2ImmPred:$offset)))),
408 (LDrib_indexed_shl_V4 IntRegs:$src1,
409 IntRegs:$src2, u2ImmPred:$offset)>,
412 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1,
413 (shl IntRegs:$src2, u2ImmPred:$offset)))),
414 (LDriub_indexed_shl_V4 IntRegs:$src1,
415 IntRegs:$src2, u2ImmPred:$offset)>,
418 def : Pat <(i32 (extloadi8 (add IntRegs:$src1,
419 (shl IntRegs:$src2, u2ImmPred:$offset)))),
420 (LDriub_indexed_shl_V4 IntRegs:$src1,
421 IntRegs:$src2, u2ImmPred:$offset)>,
424 def : Pat <(i32 (sextloadi16 (add IntRegs:$src1,
425 (shl IntRegs:$src2, u2ImmPred:$offset)))),
426 (LDrih_indexed_shl_V4 IntRegs:$src1,
427 IntRegs:$src2, u2ImmPred:$offset)>,
430 def : Pat <(i32 (zextloadi16 (add IntRegs:$src1,
431 (shl IntRegs:$src2, u2ImmPred:$offset)))),
432 (LDriuh_indexed_shl_V4 IntRegs:$src1,
433 IntRegs:$src2, u2ImmPred:$offset)>,
436 def : Pat <(i32 (extloadi16 (add IntRegs:$src1,
437 (shl IntRegs:$src2, u2ImmPred:$offset)))),
438 (LDriuh_indexed_shl_V4 IntRegs:$src1,
439 IntRegs:$src2, u2ImmPred:$offset)>,
442 def : Pat <(i32 (load (add IntRegs:$src1,
443 (shl IntRegs:$src2, u2ImmPred:$offset)))),
444 (LDriw_indexed_shl_V4 IntRegs:$src1,
445 IntRegs:$src2, u2ImmPred:$offset)>,
448 def : Pat <(i64 (load (add IntRegs:$src1,
449 (shl IntRegs:$src2, u2ImmPred:$offset)))),
450 (LDrid_indexed_shl_V4 IntRegs:$src1,
451 IntRegs:$src2, u2ImmPred:$offset)>,
456 // 'def pats' for load instruction base + register offset and
457 // zero immediate value.
458 let AddedComplexity = 10 in {
459 def : Pat <(i64 (load (add IntRegs:$src1, IntRegs:$src2))),
460 (LDrid_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
463 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1, IntRegs:$src2))),
464 (LDrib_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
467 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1, IntRegs:$src2))),
468 (LDriub_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
471 def : Pat <(i32 (extloadi8 (add IntRegs:$src1, IntRegs:$src2))),
472 (LDriub_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
475 def : Pat <(i32 (sextloadi16 (add IntRegs:$src1, IntRegs:$src2))),
476 (LDrih_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
479 def : Pat <(i32 (zextloadi16 (add IntRegs:$src1, IntRegs:$src2))),
480 (LDriuh_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
483 def : Pat <(i32 (extloadi16 (add IntRegs:$src1, IntRegs:$src2))),
484 (LDriuh_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
487 def : Pat <(i32 (load (add IntRegs:$src1, IntRegs:$src2))),
488 (LDriw_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
493 def : Pat <(i64 (zext (i1 PredRegs:$src1))),
494 (i64 (A4_combineir 0, (C2_muxii (i1 PredRegs:$src1), 1, 0)))>,
498 def : Pat <(i64 (zext (i32 IntRegs:$src1))),
499 (i64 (A4_combineir 0, (i32 IntRegs:$src1)))>,
502 def: Pat <(i64 (zextloadi8 ADDRriS11_0:$src1)),
503 (i64 (A4_combineir 0, (L2_loadrub_io AddrFI:$src1, 0)))>,
506 let AddedComplexity = 20 in
507 def: Pat <(i64 (zextloadi8 (add (i32 IntRegs:$src1),
508 s11_0ExtPred:$offset))),
509 (i64 (A4_combineir 0, (L2_loadrub_io IntRegs:$src1,
510 s11_0ExtPred:$offset)))>,
514 def: Pat <(i64 (zextloadi1 ADDRriS11_0:$src1)),
515 (i64 (A4_combineir 0, (L2_loadrub_io AddrFI:$src1, 0)))>,
518 let AddedComplexity = 20 in
519 def: Pat <(i64 (zextloadi1 (add (i32 IntRegs:$src1),
520 s11_0ExtPred:$offset))),
521 (i64 (A4_combineir 0, (L2_loadrub_io IntRegs:$src1,
522 s11_0ExtPred:$offset)))>,
526 def: Pat <(i64 (zextloadi16 ADDRriS11_1:$src1)),
527 (i64 (A4_combineir 0, (L2_loadruh_io AddrFI:$src1, 0)))>,
530 let AddedComplexity = 20 in
531 def: Pat <(i64 (zextloadi16 (add (i32 IntRegs:$src1),
532 s11_1ExtPred:$offset))),
533 (i64 (A4_combineir 0, (L2_loadruh_io IntRegs:$src1,
534 s11_1ExtPred:$offset)))>,
538 def: Pat <(i64 (extloadi16 ADDRriS11_2:$src1)),
539 (i64 (A4_combineir 0, (L2_loadrh_io AddrFI:$src1, 0)))>,
542 let AddedComplexity = 20 in
543 def: Pat <(i64 (extloadi16 (add (i32 IntRegs:$src1),
544 s11_1ExtPred:$offset))),
545 (i64 (A4_combineir 0, (L2_loadrh_io IntRegs:$src1,
546 s11_1ExtPred:$offset)))>,
550 def: Pat <(i64 (zextloadi32 ADDRriS11_2:$src1)),
551 (i64 (A4_combineir 0, (L2_loadri_io AddrFI:$src1, 0)))>,
554 let AddedComplexity = 100 in
555 def: Pat <(i64 (zextloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
556 (i64 (A4_combineir 0, (L2_loadri_io IntRegs:$src1,
557 s11_2ExtPred:$offset)))>,
561 def: Pat <(i64 (extloadi32 ADDRriS11_2:$src1)),
562 (i64 (A4_combineir 0, (L2_loadri_io AddrFI:$src1, 0)))>,
565 let AddedComplexity = 100 in
566 def: Pat <(i64 (extloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
567 (i64 (A4_combineir 0, (L2_loadri_io IntRegs:$src1,
568 s11_2ExtPred:$offset)))>,
573 //===----------------------------------------------------------------------===//
575 //===----------------------------------------------------------------------===//
577 //===----------------------------------------------------------------------===//
579 //===----------------------------------------------------------------------===//
581 //===----------------------------------------------------------------------===//
582 // Template class for store instructions with Absolute set addressing mode.
583 //===----------------------------------------------------------------------===//
584 let isExtended = 1, opExtendable = 2, validSubTargets = HasV4SubT,
585 addrMode = AbsoluteSet in
586 class T_ST_abs_set<string mnemonic, RegisterClass RC>:
587 STInst2<(outs IntRegs:$dst1),
588 (ins RC:$src1, u0AlwaysExt:$src2),
589 mnemonic#"($dst1=##$src2) = $src1",
593 def STrid_abs_set_V4 : T_ST_abs_set <"memd", DoubleRegs>;
594 def STrib_abs_set_V4 : T_ST_abs_set <"memb", IntRegs>;
595 def STrih_abs_set_V4 : T_ST_abs_set <"memh", IntRegs>;
596 def STriw_abs_set_V4 : T_ST_abs_set <"memw", IntRegs>;
598 //===----------------------------------------------------------------------===//
599 // multiclass for store instructions with base + register offset addressing
601 //===----------------------------------------------------------------------===//
602 multiclass ST_Idxd_shl_Pbase<string mnemonic, RegisterClass RC, bit isNot,
604 let isPredicatedNew = isPredNew in
605 def NAME : STInst2<(outs),
606 (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
608 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
609 ") ")#mnemonic#"($src2+$src3<<#$src4) = $src5",
614 multiclass ST_Idxd_shl_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
615 let isPredicatedFalse = PredNot in {
616 defm _c#NAME : ST_Idxd_shl_Pbase<mnemonic, RC, PredNot, 0>;
618 defm _cdn#NAME : ST_Idxd_shl_Pbase<mnemonic, RC, PredNot, 1>;
622 let isNVStorable = 1 in
623 multiclass ST_Idxd_shl<string mnemonic, string CextOp, RegisterClass RC> {
624 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
625 let isPredicable = 1 in
626 def NAME#_V4 : STInst2<(outs),
627 (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, RC:$src4),
628 mnemonic#"($src1+$src2<<#$src3) = $src4",
632 let isPredicated = 1 in {
633 defm Pt_V4 : ST_Idxd_shl_Pred<mnemonic, RC, 0 >;
634 defm NotPt_V4 : ST_Idxd_shl_Pred<mnemonic, RC, 1>;
639 // multiclass for new-value store instructions with base + register offset
641 multiclass ST_Idxd_shl_Pbase_nv<string mnemonic, RegisterClass RC, bit isNot,
643 let isPredicatedNew = isPredNew in
644 def NAME#_nv_V4 : NVInst_V4<(outs),
645 (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
647 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
648 ") ")#mnemonic#"($src2+$src3<<#$src4) = $src5.new",
653 multiclass ST_Idxd_shl_Pred_nv<string mnemonic, RegisterClass RC, bit PredNot> {
654 let isPredicatedFalse = PredNot in {
655 defm _c#NAME : ST_Idxd_shl_Pbase_nv<mnemonic, RC, PredNot, 0>;
657 defm _cdn#NAME : ST_Idxd_shl_Pbase_nv<mnemonic, RC, PredNot, 1>;
661 let mayStore = 1, isNVStore = 1 in
662 multiclass ST_Idxd_shl_nv<string mnemonic, string CextOp, RegisterClass RC> {
663 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
664 let isPredicable = 1 in
665 def NAME#_nv_V4 : NVInst_V4<(outs),
666 (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, RC:$src4),
667 mnemonic#"($src1+$src2<<#$src3) = $src4.new",
671 let isPredicated = 1 in {
672 defm Pt : ST_Idxd_shl_Pred_nv<mnemonic, RC, 0 >;
673 defm NotPt : ST_Idxd_shl_Pred_nv<mnemonic, RC, 1>;
678 let addrMode = BaseRegOffset, hasSideEffects = 0,
679 validSubTargets = HasV4SubT in {
680 let accessSize = ByteAccess in
681 defm STrib_indexed_shl: ST_Idxd_shl<"memb", "STrib", IntRegs>,
682 ST_Idxd_shl_nv<"memb", "STrib", IntRegs>, AddrModeRel;
684 let accessSize = HalfWordAccess in
685 defm STrih_indexed_shl: ST_Idxd_shl<"memh", "STrih", IntRegs>,
686 ST_Idxd_shl_nv<"memh", "STrih", IntRegs>, AddrModeRel;
688 let accessSize = WordAccess in
689 defm STriw_indexed_shl: ST_Idxd_shl<"memw", "STriw", IntRegs>,
690 ST_Idxd_shl_nv<"memw", "STriw", IntRegs>, AddrModeRel;
692 let isNVStorable = 0, accessSize = DoubleWordAccess in
693 defm STrid_indexed_shl: ST_Idxd_shl<"memd", "STrid", DoubleRegs>, AddrModeRel;
696 let Predicates = [HasV4T], AddedComplexity = 10 in {
697 def : Pat<(truncstorei8 (i32 IntRegs:$src4),
698 (add IntRegs:$src1, (shl IntRegs:$src2,
700 (STrib_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2,
701 u2ImmPred:$src3, IntRegs:$src4)>;
703 def : Pat<(truncstorei16 (i32 IntRegs:$src4),
704 (add IntRegs:$src1, (shl IntRegs:$src2,
706 (STrih_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2,
707 u2ImmPred:$src3, IntRegs:$src4)>;
709 def : Pat<(store (i32 IntRegs:$src4),
710 (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))),
711 (STriw_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2,
712 u2ImmPred:$src3, IntRegs:$src4)>;
714 def : Pat<(store (i64 DoubleRegs:$src4),
715 (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))),
716 (STrid_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2,
717 u2ImmPred:$src3, DoubleRegs:$src4)>;
720 let isExtended = 1, opExtendable = 2 in
721 class T_ST_LongOff <string mnemonic, PatFrag stOp, RegisterClass RC, ValueType VT> :
723 (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, RC:$src4),
724 mnemonic#"($src1<<#$src2+##$src3) = $src4",
725 [(stOp (VT RC:$src4),
726 (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2),
727 u0AlwaysExtPred:$src3))]>,
730 let isExtended = 1, opExtendable = 2, mayStore = 1, isNVStore = 1 in
731 class T_ST_LongOff_nv <string mnemonic> :
733 (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, IntRegs:$src4),
734 mnemonic#"($src1<<#$src2+##$src3) = $src4.new",
738 multiclass ST_LongOff <string mnemonic, string BaseOp, PatFrag stOp> {
739 let BaseOpcode = BaseOp#"_shl" in {
740 let isNVStorable = 1 in
741 def NAME#_V4 : T_ST_LongOff<mnemonic, stOp, IntRegs, i32>;
743 def NAME#_nv_V4 : T_ST_LongOff_nv<mnemonic>;
747 let AddedComplexity = 10, validSubTargets = HasV4SubT in {
748 def STrid_shl_V4 : T_ST_LongOff<"memd", store, DoubleRegs, i64>;
749 defm STrib_shl : ST_LongOff <"memb", "STrib", truncstorei8>, NewValueRel;
750 defm STrih_shl : ST_LongOff <"memh", "Strih", truncstorei16>, NewValueRel;
751 defm STriw_shl : ST_LongOff <"memw", "STriw", store>, NewValueRel;
754 let AddedComplexity = 40 in
755 multiclass T_ST_LOff_Pats <InstHexagon I, RegisterClass RC, ValueType VT,
757 def : Pat<(stOp (VT RC:$src4),
758 (add (shl IntRegs:$src1, u2ImmPred:$src2),
759 (NumUsesBelowThresCONST32 tglobaladdr:$src3))),
760 (I IntRegs:$src1, u2ImmPred:$src2, tglobaladdr:$src3, RC:$src4)>;
762 def : Pat<(stOp (VT RC:$src4),
764 (NumUsesBelowThresCONST32 tglobaladdr:$src3))),
765 (I IntRegs:$src1, 0, tglobaladdr:$src3, RC:$src4)>;
768 defm : T_ST_LOff_Pats<STrid_shl_V4, DoubleRegs, i64, store>;
769 defm : T_ST_LOff_Pats<STriw_shl_V4, IntRegs, i32, store>;
770 defm : T_ST_LOff_Pats<STrib_shl_V4, IntRegs, i32, truncstorei8>;
771 defm : T_ST_LOff_Pats<STrih_shl_V4, IntRegs, i32, truncstorei16>;
773 // memd(Rx++#s4:3)=Rtt
774 // memd(Rx++#s4:3:circ(Mu))=Rtt
775 // memd(Rx++I:circ(Mu))=Rtt
777 // memd(Rx++Mu:brev)=Rtt
778 // memd(gp+#u16:3)=Rtt
780 // Store doubleword conditionally.
781 // if ([!]Pv[.new]) memd(#u6)=Rtt
782 // TODO: needs to be implemented.
784 //===----------------------------------------------------------------------===//
785 // multiclass for store instructions with base + immediate offset
786 // addressing mode and immediate stored value.
787 // mem[bhw](Rx++#s4:3)=#s8
788 // if ([!]Pv[.new]) mem[bhw](Rx++#s4:3)=#s6
789 //===----------------------------------------------------------------------===//
790 multiclass ST_Imm_Pbase<string mnemonic, Operand OffsetOp, bit isNot,
792 let isPredicatedNew = isPredNew in
793 def NAME : STInst2<(outs),
794 (ins PredRegs:$src1, IntRegs:$src2, OffsetOp:$src3, s6Ext:$src4),
795 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
796 ") ")#mnemonic#"($src2+#$src3) = #$src4",
801 multiclass ST_Imm_Pred<string mnemonic, Operand OffsetOp, bit PredNot> {
802 let isPredicatedFalse = PredNot in {
803 defm _c#NAME : ST_Imm_Pbase<mnemonic, OffsetOp, PredNot, 0>;
805 defm _cdn#NAME : ST_Imm_Pbase<mnemonic, OffsetOp, PredNot, 1>;
809 let isExtendable = 1, isExtentSigned = 1, hasSideEffects = 0 in
810 multiclass ST_Imm<string mnemonic, string CextOp, Operand OffsetOp> {
811 let CextOpcode = CextOp, BaseOpcode = CextOp#_imm in {
812 let opExtendable = 2, opExtentBits = 8, isPredicable = 1 in
813 def NAME#_V4 : STInst2<(outs),
814 (ins IntRegs:$src1, OffsetOp:$src2, s8Ext:$src3),
815 mnemonic#"($src1+#$src2) = #$src3",
819 let opExtendable = 3, opExtentBits = 6, isPredicated = 1 in {
820 defm Pt_V4 : ST_Imm_Pred<mnemonic, OffsetOp, 0>;
821 defm NotPt_V4 : ST_Imm_Pred<mnemonic, OffsetOp, 1 >;
826 let addrMode = BaseImmOffset, InputType = "imm",
827 validSubTargets = HasV4SubT in {
828 let accessSize = ByteAccess in
829 defm STrib_imm : ST_Imm<"memb", "STrib", u6_0Imm>, ImmRegRel, PredNewRel;
831 let accessSize = HalfWordAccess in
832 defm STrih_imm : ST_Imm<"memh", "STrih", u6_1Imm>, ImmRegRel, PredNewRel;
834 let accessSize = WordAccess in
835 defm STriw_imm : ST_Imm<"memw", "STriw", u6_2Imm>, ImmRegRel, PredNewRel;
838 let Predicates = [HasV4T], AddedComplexity = 10 in {
839 def: Pat<(truncstorei8 s8ExtPred:$src3, (add IntRegs:$src1, u6_0ImmPred:$src2)),
840 (STrib_imm_V4 IntRegs:$src1, u6_0ImmPred:$src2, s8ExtPred:$src3)>;
842 def: Pat<(truncstorei16 s8ExtPred:$src3, (add IntRegs:$src1,
844 (STrih_imm_V4 IntRegs:$src1, u6_1ImmPred:$src2, s8ExtPred:$src3)>;
846 def: Pat<(store s8ExtPred:$src3, (add IntRegs:$src1, u6_2ImmPred:$src2)),
847 (STriw_imm_V4 IntRegs:$src1, u6_2ImmPred:$src2, s8ExtPred:$src3)>;
850 let AddedComplexity = 6 in
851 def : Pat <(truncstorei8 s8ExtPred:$src2, (i32 IntRegs:$src1)),
852 (STrib_imm_V4 IntRegs:$src1, 0, s8ExtPred:$src2)>,
855 // memb(Rx++#s4:0:circ(Mu))=Rt
856 // memb(Rx++I:circ(Mu))=Rt
858 // memb(Rx++Mu:brev)=Rt
859 // memb(gp+#u16:0)=Rt
863 // TODO: needs to be implemented
865 // memh(Rs+#s11:1)=Rt.H
866 let AddedComplexity = 6 in
867 def : Pat <(truncstorei16 s8ExtPred:$src2, (i32 IntRegs:$src1)),
868 (STrih_imm_V4 IntRegs:$src1, 0, s8ExtPred:$src2)>,
871 // memh(Rs+Ru<<#u2)=Rt.H
872 // TODO: needs to be implemented.
874 // memh(Ru<<#u2+#U6)=Rt.H
875 // memh(Rx++#s4:1:circ(Mu))=Rt.H
876 // memh(Rx++#s4:1:circ(Mu))=Rt
877 // memh(Rx++I:circ(Mu))=Rt.H
878 // memh(Rx++I:circ(Mu))=Rt
881 // memh(Rx++Mu:brev)=Rt.H
882 // memh(Rx++Mu:brev)=Rt
883 // memh(gp+#u16:1)=Rt
884 // if ([!]Pv[.new]) memh(#u6)=Rt.H
885 // if ([!]Pv[.new]) memh(#u6)=Rt
888 // if ([!]Pv[.new]) memh(Rs+#u6:1)=Rt.H
889 // TODO: needs to be implemented.
891 // if ([!]Pv[.new]) memh(Rx++#s4:1)=Rt.H
892 // TODO: Needs to be implemented.
896 // TODO: Needs to be implemented.
899 let hasSideEffects = 0 in
900 def STriw_pred_V4 : STInst2<(outs),
901 (ins MEMri:$addr, PredRegs:$src1),
902 "Error; should not emit",
906 let AddedComplexity = 6 in
907 def : Pat <(store s8ExtPred:$src2, (i32 IntRegs:$src1)),
908 (STriw_imm_V4 IntRegs:$src1, 0, s8ExtPred:$src2)>,
911 // memw(Rx++#s4:2)=Rt
912 // memw(Rx++#s4:2:circ(Mu))=Rt
913 // memw(Rx++I:circ(Mu))=Rt
915 // memw(Rx++Mu:brev)=Rt
917 //===----------------------------------------------------------------------===
919 //===----------------------------------------------------------------------===
922 //===----------------------------------------------------------------------===//
924 //===----------------------------------------------------------------------===//
926 // multiclass for new-value store instructions with base + immediate offset.
928 multiclass ST_Idxd_Pbase_nv<string mnemonic, RegisterClass RC,
929 Operand predImmOp, bit isNot, bit isPredNew> {
930 let isPredicatedNew = isPredNew in
931 def NAME#_nv_V4 : NVInst_V4<(outs),
932 (ins PredRegs:$src1, IntRegs:$src2, predImmOp:$src3, RC: $src4),
933 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
934 ") ")#mnemonic#"($src2+#$src3) = $src4.new",
939 multiclass ST_Idxd_Pred_nv<string mnemonic, RegisterClass RC, Operand predImmOp,
941 let isPredicatedFalse = PredNot in {
942 defm _c#NAME : ST_Idxd_Pbase_nv<mnemonic, RC, predImmOp, PredNot, 0>;
944 defm _cdn#NAME : ST_Idxd_Pbase_nv<mnemonic, RC, predImmOp, PredNot, 1>;
948 let mayStore = 1, isNVStore = 1, hasSideEffects = 0, isExtendable = 1 in
949 multiclass ST_Idxd_nv<string mnemonic, string CextOp, RegisterClass RC,
950 Operand ImmOp, Operand predImmOp, bits<5> ImmBits,
951 bits<5> PredImmBits> {
953 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
954 let opExtendable = 1, isExtentSigned = 1, opExtentBits = ImmBits,
956 def NAME#_nv_V4 : NVInst_V4<(outs),
957 (ins IntRegs:$src1, ImmOp:$src2, RC:$src3),
958 mnemonic#"($src1+#$src2) = $src3.new",
962 let opExtendable = 2, isExtentSigned = 0, opExtentBits = PredImmBits,
963 isPredicated = 1 in {
964 defm Pt : ST_Idxd_Pred_nv<mnemonic, RC, predImmOp, 0>;
965 defm NotPt : ST_Idxd_Pred_nv<mnemonic, RC, predImmOp, 1>;
970 let addrMode = BaseImmOffset, validSubTargets = HasV4SubT in {
971 let accessSize = ByteAccess in
972 defm STrib_indexed: ST_Idxd_nv<"memb", "STrib", IntRegs, s11_0Ext,
973 u6_0Ext, 11, 6>, AddrModeRel;
975 let accessSize = HalfWordAccess in
976 defm STrih_indexed: ST_Idxd_nv<"memh", "STrih", IntRegs, s11_1Ext,
977 u6_1Ext, 12, 7>, AddrModeRel;
979 let accessSize = WordAccess in
980 defm STriw_indexed: ST_Idxd_nv<"memw", "STriw", IntRegs, s11_2Ext,
981 u6_2Ext, 13, 8>, AddrModeRel;
984 // multiclass for new-value store instructions with base + immediate offset.
985 // and MEMri operand.
986 multiclass ST_MEMri_Pbase_nv<string mnemonic, RegisterClass RC, bit isNot,
988 let isPredicatedNew = isPredNew in
989 def NAME#_nv_V4 : NVInst_V4<(outs),
990 (ins PredRegs:$src1, MEMri:$addr, RC: $src2),
991 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
992 ") ")#mnemonic#"($addr) = $src2.new",
997 multiclass ST_MEMri_Pred_nv<string mnemonic, RegisterClass RC, bit PredNot> {
998 let isPredicatedFalse = PredNot in {
999 defm _c#NAME : ST_MEMri_Pbase_nv<mnemonic, RC, PredNot, 0>;
1002 defm _cdn#NAME : ST_MEMri_Pbase_nv<mnemonic, RC, PredNot, 1>;
1006 let mayStore = 1, isNVStore = 1, isExtendable = 1, hasSideEffects = 0 in
1007 multiclass ST_MEMri_nv<string mnemonic, string CextOp, RegisterClass RC,
1008 bits<5> ImmBits, bits<5> PredImmBits> {
1010 let CextOpcode = CextOp, BaseOpcode = CextOp in {
1011 let opExtendable = 1, isExtentSigned = 1, opExtentBits = ImmBits,
1013 def NAME#_nv_V4 : NVInst_V4<(outs),
1014 (ins MEMri:$addr, RC:$src),
1015 mnemonic#"($addr) = $src.new",
1019 let opExtendable = 2, isExtentSigned = 0, opExtentBits = PredImmBits,
1020 hasSideEffects = 0, isPredicated = 1 in {
1021 defm Pt : ST_MEMri_Pred_nv<mnemonic, RC, 0>;
1022 defm NotPt : ST_MEMri_Pred_nv<mnemonic, RC, 1>;
1027 let addrMode = BaseImmOffset, isMEMri = "true", validSubTargets = HasV4SubT,
1029 let accessSize = ByteAccess in
1030 defm STrib: ST_MEMri_nv<"memb", "STrib", IntRegs, 11, 6>, AddrModeRel;
1032 let accessSize = HalfWordAccess in
1033 defm STrih: ST_MEMri_nv<"memh", "STrih", IntRegs, 12, 7>, AddrModeRel;
1035 let accessSize = WordAccess in
1036 defm STriw: ST_MEMri_nv<"memw", "STriw", IntRegs, 13, 8>, AddrModeRel;
1039 //===----------------------------------------------------------------------===//
1040 // Post increment store
1041 // mem[bhwd](Rx++#s4:[0123])=Nt.new
1042 //===----------------------------------------------------------------------===//
1044 multiclass ST_PostInc_Pbase_nv<string mnemonic, RegisterClass RC, Operand ImmOp,
1045 bit isNot, bit isPredNew> {
1046 let isPredicatedNew = isPredNew in
1047 def NAME#_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
1048 (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset, RC:$src3),
1049 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
1050 ") ")#mnemonic#"($src2++#$offset) = $src3.new",
1056 multiclass ST_PostInc_Pred_nv<string mnemonic, RegisterClass RC,
1057 Operand ImmOp, bit PredNot> {
1058 let isPredicatedFalse = PredNot in {
1059 defm _c#NAME : ST_PostInc_Pbase_nv<mnemonic, RC, ImmOp, PredNot, 0>;
1061 let Predicates = [HasV4T], validSubTargets = HasV4SubT in
1062 defm _cdn#NAME : ST_PostInc_Pbase_nv<mnemonic, RC, ImmOp, PredNot, 1>;
1066 let hasCtrlDep = 1, isNVStore = 1, hasSideEffects = 0 in
1067 multiclass ST_PostInc_nv<string mnemonic, string BaseOp, RegisterClass RC,
1070 let BaseOpcode = "POST_"#BaseOp in {
1071 let isPredicable = 1 in
1072 def NAME#_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
1073 (ins IntRegs:$src1, ImmOp:$offset, RC:$src2),
1074 mnemonic#"($src1++#$offset) = $src2.new",
1079 let isPredicated = 1 in {
1080 defm Pt : ST_PostInc_Pred_nv<mnemonic, RC, ImmOp, 0 >;
1081 defm NotPt : ST_PostInc_Pred_nv<mnemonic, RC, ImmOp, 1 >;
1086 let addrMode = PostInc, validSubTargets = HasV4SubT in {
1087 defm POST_STbri: ST_PostInc_nv <"memb", "STrib", IntRegs, s4_0Imm>, AddrModeRel;
1088 defm POST_SThri: ST_PostInc_nv <"memh", "STrih", IntRegs, s4_1Imm>, AddrModeRel;
1089 defm POST_STwri: ST_PostInc_nv <"memw", "STriw", IntRegs, s4_2Imm>, AddrModeRel;
1092 // memb(Rx++#s4:0:circ(Mu))=Nt.new
1093 // memb(Rx++I:circ(Mu))=Nt.new
1094 // memb(Rx++Mu)=Nt.new
1095 // memb(Rx++Mu:brev)=Nt.new
1096 // memh(Rx++#s4:1:circ(Mu))=Nt.new
1097 // memh(Rx++I:circ(Mu))=Nt.new
1098 // memh(Rx++Mu)=Nt.new
1099 // memh(Rx++Mu:brev)=Nt.new
1101 // memw(Rx++#s4:2:circ(Mu))=Nt.new
1102 // memw(Rx++I:circ(Mu))=Nt.new
1103 // memw(Rx++Mu)=Nt.new
1104 // memw(Rx++Mu:brev)=Nt.new
1106 //===----------------------------------------------------------------------===//
1108 //===----------------------------------------------------------------------===//
1110 //===----------------------------------------------------------------------===//
1112 //===----------------------------------------------------------------------===//
1114 //===----------------------------------------------------------------------===//
1115 // multiclass/template class for the new-value compare jumps with the register
1117 //===----------------------------------------------------------------------===//
1119 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11 in
1120 class NVJrr_template<string mnemonic, bits<3> majOp, bit NvOpNum,
1121 bit isNegCond, bit isTak>
1123 (ins IntRegs:$src1, IntRegs:$src2, brtarget:$offset),
1124 "if ("#!if(isNegCond, "!","")#mnemonic#
1125 "($src1"#!if(!eq(NvOpNum, 0),".new, ",", ")#
1126 "$src2"#!if(!eq(NvOpNum, 1),".new))","))")#" jump:"
1127 #!if(isTak, "t","nt")#" $offset",
1128 []>, Requires<[HasV4T]> {
1132 bits<3> Ns; // New-Value Operand
1133 bits<5> RegOp; // Non-New-Value Operand
1136 let isTaken = isTak;
1137 let isBrTaken = !if(isTaken, "true", "false");
1138 let isPredicatedFalse = isNegCond;
1140 let Ns = !if(!eq(NvOpNum, 0), src1{2-0}, src2{2-0});
1141 let RegOp = !if(!eq(NvOpNum, 0), src2, src1);
1143 let IClass = 0b0010;
1145 let Inst{25-23} = majOp;
1146 let Inst{22} = isNegCond;
1147 let Inst{18-16} = Ns;
1148 let Inst{13} = isTak;
1149 let Inst{12-8} = RegOp;
1150 let Inst{21-20} = offset{10-9};
1151 let Inst{7-1} = offset{8-2};
1155 multiclass NVJrr_cond<string mnemonic, bits<3> majOp, bit NvOpNum,
1157 // Branch not taken:
1158 def _nt_V4: NVJrr_template<mnemonic, majOp, NvOpNum, isNegCond, 0>;
1160 def _t_V4: NVJrr_template<mnemonic, majOp, NvOpNum, isNegCond, 1>;
1163 // NvOpNum = 0 -> First Operand is a new-value Register
1164 // NvOpNum = 1 -> Second Operand is a new-value Register
1166 multiclass NVJrr_base<string mnemonic, string BaseOp, bits<3> majOp,
1168 let BaseOpcode = BaseOp#_NVJ in {
1169 defm _t_Jumpnv : NVJrr_cond<mnemonic, majOp, NvOpNum, 0>; // True cond
1170 defm _f_Jumpnv : NVJrr_cond<mnemonic, majOp, NvOpNum, 1>; // False cond
1174 // if ([!]cmp.eq(Ns.new,Rt)) jump:[n]t #r9:2
1175 // if ([!]cmp.gt(Ns.new,Rt)) jump:[n]t #r9:2
1176 // if ([!]cmp.gtu(Ns.new,Rt)) jump:[n]t #r9:2
1177 // if ([!]cmp.gt(Rt,Ns.new)) jump:[n]t #r9:2
1178 // if ([!]cmp.gtu(Rt,Ns.new)) jump:[n]t #r9:2
1180 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1,
1181 Defs = [PC], hasSideEffects = 0, validSubTargets = HasV4SubT in {
1182 defm CMPEQrr : NVJrr_base<"cmp.eq", "CMPEQ", 0b000, 0>, PredRel;
1183 defm CMPGTrr : NVJrr_base<"cmp.gt", "CMPGT", 0b001, 0>, PredRel;
1184 defm CMPGTUrr : NVJrr_base<"cmp.gtu", "CMPGTU", 0b010, 0>, PredRel;
1185 defm CMPLTrr : NVJrr_base<"cmp.gt", "CMPLT", 0b011, 1>, PredRel;
1186 defm CMPLTUrr : NVJrr_base<"cmp.gtu", "CMPLTU", 0b100, 1>, PredRel;
1189 //===----------------------------------------------------------------------===//
1190 // multiclass/template class for the new-value compare jumps instruction
1191 // with a register and an unsigned immediate (U5) operand.
1192 //===----------------------------------------------------------------------===//
1194 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11 in
1195 class NVJri_template<string mnemonic, bits<3> majOp, bit isNegCond,
1198 (ins IntRegs:$src1, u5Imm:$src2, brtarget:$offset),
1199 "if ("#!if(isNegCond, "!","")#mnemonic#"($src1.new, #$src2)) jump:"
1200 #!if(isTak, "t","nt")#" $offset",
1201 []>, Requires<[HasV4T]> {
1203 let isTaken = isTak;
1204 let isPredicatedFalse = isNegCond;
1205 let isBrTaken = !if(isTaken, "true", "false");
1211 let IClass = 0b0010;
1213 let Inst{25-23} = majOp;
1214 let Inst{22} = isNegCond;
1215 let Inst{18-16} = src1;
1216 let Inst{13} = isTak;
1217 let Inst{12-8} = src2;
1218 let Inst{21-20} = offset{10-9};
1219 let Inst{7-1} = offset{8-2};
1222 multiclass NVJri_cond<string mnemonic, bits<3> majOp, bit isNegCond> {
1223 // Branch not taken:
1224 def _nt_V4: NVJri_template<mnemonic, majOp, isNegCond, 0>;
1226 def _t_V4: NVJri_template<mnemonic, majOp, isNegCond, 1>;
1229 multiclass NVJri_base<string mnemonic, string BaseOp, bits<3> majOp> {
1230 let BaseOpcode = BaseOp#_NVJri in {
1231 defm _t_Jumpnv : NVJri_cond<mnemonic, majOp, 0>; // True Cond
1232 defm _f_Jumpnv : NVJri_cond<mnemonic, majOp, 1>; // False cond
1236 // if ([!]cmp.eq(Ns.new,#U5)) jump:[n]t #r9:2
1237 // if ([!]cmp.gt(Ns.new,#U5)) jump:[n]t #r9:2
1238 // if ([!]cmp.gtu(Ns.new,#U5)) jump:[n]t #r9:2
1240 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1,
1241 Defs = [PC], hasSideEffects = 0, validSubTargets = HasV4SubT in {
1242 defm CMPEQri : NVJri_base<"cmp.eq", "CMPEQ", 0b000>, PredRel;
1243 defm CMPGTri : NVJri_base<"cmp.gt", "CMPGT", 0b001>, PredRel;
1244 defm CMPGTUri : NVJri_base<"cmp.gtu", "CMPGTU", 0b010>, PredRel;
1247 //===----------------------------------------------------------------------===//
1248 // multiclass/template class for the new-value compare jumps instruction
1249 // with a register and an hardcoded 0/-1 immediate value.
1250 //===----------------------------------------------------------------------===//
1252 let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 11 in
1253 class NVJ_ConstImm_template<string mnemonic, bits<3> majOp, string ImmVal,
1254 bit isNegCond, bit isTak>
1256 (ins IntRegs:$src1, brtarget:$offset),
1257 "if ("#!if(isNegCond, "!","")#mnemonic
1258 #"($src1.new, #"#ImmVal#")) jump:"
1259 #!if(isTak, "t","nt")#" $offset",
1260 []>, Requires<[HasV4T]> {
1262 let isTaken = isTak;
1263 let isPredicatedFalse = isNegCond;
1264 let isBrTaken = !if(isTaken, "true", "false");
1268 let IClass = 0b0010;
1270 let Inst{25-23} = majOp;
1271 let Inst{22} = isNegCond;
1272 let Inst{18-16} = src1;
1273 let Inst{13} = isTak;
1274 let Inst{21-20} = offset{10-9};
1275 let Inst{7-1} = offset{8-2};
1278 multiclass NVJ_ConstImm_cond<string mnemonic, bits<3> majOp, string ImmVal,
1280 // Branch not taken:
1281 def _nt_V4: NVJ_ConstImm_template<mnemonic, majOp, ImmVal, isNegCond, 0>;
1283 def _t_V4: NVJ_ConstImm_template<mnemonic, majOp, ImmVal, isNegCond, 1>;
1286 multiclass NVJ_ConstImm_base<string mnemonic, string BaseOp, bits<3> majOp,
1288 let BaseOpcode = BaseOp#_NVJ_ConstImm in {
1289 defm _t_Jumpnv : NVJ_ConstImm_cond<mnemonic, majOp, ImmVal, 0>; // True cond
1290 defm _f_Jumpnv : NVJ_ConstImm_cond<mnemonic, majOp, ImmVal, 1>; // False Cond
1294 // if ([!]tstbit(Ns.new,#0)) jump:[n]t #r9:2
1295 // if ([!]cmp.eq(Ns.new,#-1)) jump:[n]t #r9:2
1296 // if ([!]cmp.gt(Ns.new,#-1)) jump:[n]t #r9:2
1298 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator=1,
1299 Defs = [PC], hasSideEffects = 0 in {
1300 defm TSTBIT0 : NVJ_ConstImm_base<"tstbit", "TSTBIT", 0b011, "0">, PredRel;
1301 defm CMPEQn1 : NVJ_ConstImm_base<"cmp.eq", "CMPEQ", 0b100, "-1">, PredRel;
1302 defm CMPGTn1 : NVJ_ConstImm_base<"cmp.gt", "CMPGT", 0b101, "-1">, PredRel;
1305 //===----------------------------------------------------------------------===//
1307 //===----------------------------------------------------------------------===//
1309 // Add and accumulate.
1310 // Rd=add(Rs,add(Ru,#s6))
1311 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 6,
1312 validSubTargets = HasV4SubT in
1313 def ADDr_ADDri_V4 : MInst<(outs IntRegs:$dst),
1314 (ins IntRegs:$src1, IntRegs:$src2, s6Ext:$src3),
1315 "$dst = add($src1, add($src2, #$src3))",
1316 [(set (i32 IntRegs:$dst),
1317 (add (i32 IntRegs:$src1), (add (i32 IntRegs:$src2),
1318 s6_16ExtPred:$src3)))]>,
1321 // Rd=add(Rs,sub(#s6,Ru))
1322 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6,
1323 validSubTargets = HasV4SubT in
1324 def ADDr_SUBri_V4 : MInst<(outs IntRegs:$dst),
1325 (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3),
1326 "$dst = add($src1, sub(#$src2, $src3))",
1327 [(set (i32 IntRegs:$dst),
1328 (add (i32 IntRegs:$src1), (sub s6_10ExtPred:$src2,
1329 (i32 IntRegs:$src3))))]>,
1332 // Generates the same instruction as ADDr_SUBri_V4 but matches different
1334 // Rd=add(Rs,sub(#s6,Ru))
1335 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6,
1336 validSubTargets = HasV4SubT in
1337 def ADDri_SUBr_V4 : MInst<(outs IntRegs:$dst),
1338 (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3),
1339 "$dst = add($src1, sub(#$src2, $src3))",
1340 [(set (i32 IntRegs:$dst),
1341 (sub (add (i32 IntRegs:$src1), s6_10ExtPred:$src2),
1342 (i32 IntRegs:$src3)))]>,
1346 // Add or subtract doublewords with carry.
1348 // Rdd=add(Rss,Rtt,Px):carry
1350 // Rdd=sub(Rss,Rtt,Px):carry
1353 // Logical doublewords.
1354 // Rdd=and(Rtt,~Rss)
1355 let validSubTargets = HasV4SubT in
1356 def ANDd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
1357 (ins DoubleRegs:$src1, DoubleRegs:$src2),
1358 "$dst = and($src1, ~$src2)",
1359 [(set (i64 DoubleRegs:$dst), (and (i64 DoubleRegs:$src1),
1360 (not (i64 DoubleRegs:$src2))))]>,
1364 let validSubTargets = HasV4SubT in
1365 def ORd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
1366 (ins DoubleRegs:$src1, DoubleRegs:$src2),
1367 "$dst = or($src1, ~$src2)",
1368 [(set (i64 DoubleRegs:$dst),
1369 (or (i64 DoubleRegs:$src1), (not (i64 DoubleRegs:$src2))))]>,
1373 // Logical-logical doublewords.
1374 // Rxx^=xor(Rss,Rtt)
1375 let validSubTargets = HasV4SubT in
1376 def XORd_XORdd: MInst_acc<(outs DoubleRegs:$dst),
1377 (ins DoubleRegs:$src1, DoubleRegs:$src2, DoubleRegs:$src3),
1378 "$dst ^= xor($src2, $src3)",
1379 [(set (i64 DoubleRegs:$dst),
1380 (xor (i64 DoubleRegs:$src1), (xor (i64 DoubleRegs:$src2),
1381 (i64 DoubleRegs:$src3))))],
1386 // Logical-logical words.
1387 // Rx=or(Ru,and(Rx,#s10))
1388 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
1389 validSubTargets = HasV4SubT in
1390 def ORr_ANDri_V4 : MInst_acc<(outs IntRegs:$dst),
1391 (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
1392 "$dst = or($src1, and($src2, #$src3))",
1393 [(set (i32 IntRegs:$dst),
1394 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1395 s10ExtPred:$src3)))],
1399 // Rx[&|^]=and(Rs,Rt)
1401 let validSubTargets = HasV4SubT in
1402 def ANDr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
1403 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1404 "$dst &= and($src2, $src3)",
1405 [(set (i32 IntRegs:$dst),
1406 (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1407 (i32 IntRegs:$src3))))],
1412 let validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "reg" in
1413 def ORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
1414 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1415 "$dst |= and($src2, $src3)",
1416 [(set (i32 IntRegs:$dst),
1417 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1418 (i32 IntRegs:$src3))))],
1420 Requires<[HasV4T]>, ImmRegRel;
1423 let validSubTargets = HasV4SubT in
1424 def XORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
1425 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1426 "$dst ^= and($src2, $src3)",
1427 [(set (i32 IntRegs:$dst),
1428 (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1429 (i32 IntRegs:$src3))))],
1433 // Rx[&|^]=and(Rs,~Rt)
1435 let validSubTargets = HasV4SubT in
1436 def ANDr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
1437 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1438 "$dst &= and($src2, ~$src3)",
1439 [(set (i32 IntRegs:$dst),
1440 (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1441 (not (i32 IntRegs:$src3)))))],
1446 let validSubTargets = HasV4SubT in
1447 def ORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
1448 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1449 "$dst |= and($src2, ~$src3)",
1450 [(set (i32 IntRegs:$dst),
1451 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1452 (not (i32 IntRegs:$src3)))))],
1457 let validSubTargets = HasV4SubT in
1458 def XORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
1459 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1460 "$dst ^= and($src2, ~$src3)",
1461 [(set (i32 IntRegs:$dst),
1462 (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1463 (not (i32 IntRegs:$src3)))))],
1467 // Rx[&|^]=or(Rs,Rt)
1469 let validSubTargets = HasV4SubT in
1470 def ANDr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1471 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1472 "$dst &= or($src2, $src3)",
1473 [(set (i32 IntRegs:$dst),
1474 (and (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
1475 (i32 IntRegs:$src3))))],
1480 let validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "reg" in
1481 def ORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1482 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1483 "$dst |= or($src2, $src3)",
1484 [(set (i32 IntRegs:$dst),
1485 (or (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
1486 (i32 IntRegs:$src3))))],
1488 Requires<[HasV4T]>, ImmRegRel;
1491 let validSubTargets = HasV4SubT in
1492 def XORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1493 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1494 "$dst ^= or($src2, $src3)",
1495 [(set (i32 IntRegs:$dst),
1496 (xor (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
1497 (i32 IntRegs:$src3))))],
1501 // Rx[&|^]=xor(Rs,Rt)
1503 let validSubTargets = HasV4SubT in
1504 def ANDr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1505 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1506 "$dst &= xor($src2, $src3)",
1507 [(set (i32 IntRegs:$dst),
1508 (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
1509 (i32 IntRegs:$src3))))],
1514 let validSubTargets = HasV4SubT in
1515 def ORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1516 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1517 "$dst |= xor($src2, $src3)",
1518 [(set (i32 IntRegs:$dst),
1519 (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
1520 (i32 IntRegs:$src3))))],
1525 let validSubTargets = HasV4SubT in
1526 def XORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1527 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1528 "$dst ^= xor($src2, $src3)",
1529 [(set (i32 IntRegs:$dst),
1530 (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
1531 (i32 IntRegs:$src3))))],
1536 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
1537 validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "imm" in
1538 def ORr_ANDri2_V4 : MInst_acc<(outs IntRegs:$dst),
1539 (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
1540 "$dst |= and($src2, #$src3)",
1541 [(set (i32 IntRegs:$dst),
1542 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1543 s10ExtPred:$src3)))],
1545 Requires<[HasV4T]>, ImmRegRel;
1548 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
1549 validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "imm" in
1550 def ORr_ORri_V4 : MInst_acc<(outs IntRegs:$dst),
1551 (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
1552 "$dst |= or($src2, #$src3)",
1553 [(set (i32 IntRegs:$dst),
1554 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1555 s10ExtPred:$src3)))],
1557 Requires<[HasV4T]>, ImmRegRel;
1561 // Rd=modwrap(Rs,Rt)
1563 // Rd=cround(Rs,#u5)
1565 // Rd=round(Rs,#u5)[:sat]
1566 // Rd=round(Rs,Rt)[:sat]
1567 // Vector reduce add unsigned halfwords
1568 // Rd=vraddh(Rss,Rtt)
1570 // Rdd=vaddb(Rss,Rtt)
1571 // Vector conditional negate
1572 // Rdd=vcnegh(Rss,Rt)
1573 // Rxx+=vrcnegh(Rss,Rt)
1574 // Vector maximum bytes
1575 // Rdd=vmaxb(Rtt,Rss)
1576 // Vector reduce maximum halfwords
1577 // Rxx=vrmaxh(Rss,Ru)
1578 // Rxx=vrmaxuh(Rss,Ru)
1579 // Vector reduce maximum words
1580 // Rxx=vrmaxuw(Rss,Ru)
1581 // Rxx=vrmaxw(Rss,Ru)
1582 // Vector minimum bytes
1583 // Rdd=vminb(Rtt,Rss)
1584 // Vector reduce minimum halfwords
1585 // Rxx=vrminh(Rss,Ru)
1586 // Rxx=vrminuh(Rss,Ru)
1587 // Vector reduce minimum words
1588 // Rxx=vrminuw(Rss,Ru)
1589 // Rxx=vrminw(Rss,Ru)
1590 // Vector subtract bytes
1591 // Rdd=vsubb(Rss,Rtt)
1593 //===----------------------------------------------------------------------===//
1595 //===----------------------------------------------------------------------===//
1598 //===----------------------------------------------------------------------===//
1600 //===----------------------------------------------------------------------===//
1602 // Multiply and user lower result.
1603 // Rd=add(#u6,mpyi(Rs,#U6))
1604 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6,
1605 validSubTargets = HasV4SubT in
1606 def ADDi_MPYri_V4 : MInst<(outs IntRegs:$dst),
1607 (ins u6Ext:$src1, IntRegs:$src2, u6Imm:$src3),
1608 "$dst = add(#$src1, mpyi($src2, #$src3))",
1609 [(set (i32 IntRegs:$dst),
1610 (add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
1611 u6ExtPred:$src1))]>,
1614 // Rd=add(##,mpyi(Rs,#U6))
1615 def : Pat <(add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
1616 (HexagonCONST32 tglobaladdr:$src1)),
1617 (i32 (ADDi_MPYri_V4 tglobaladdr:$src1, IntRegs:$src2,
1620 // Rd=add(#u6,mpyi(Rs,Rt))
1621 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6,
1622 validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in
1623 def ADDi_MPYrr_V4 : MInst<(outs IntRegs:$dst),
1624 (ins u6Ext:$src1, IntRegs:$src2, IntRegs:$src3),
1625 "$dst = add(#$src1, mpyi($src2, $src3))",
1626 [(set (i32 IntRegs:$dst),
1627 (add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
1628 u6ExtPred:$src1))]>,
1629 Requires<[HasV4T]>, ImmRegRel;
1631 // Rd=add(##,mpyi(Rs,Rt))
1632 def : Pat <(add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
1633 (HexagonCONST32 tglobaladdr:$src1)),
1634 (i32 (ADDi_MPYrr_V4 tglobaladdr:$src1, IntRegs:$src2,
1637 // Rd=add(Ru,mpyi(#u6:2,Rs))
1638 let validSubTargets = HasV4SubT in
1639 def ADDr_MPYir_V4 : MInst<(outs IntRegs:$dst),
1640 (ins IntRegs:$src1, u6Imm:$src2, IntRegs:$src3),
1641 "$dst = add($src1, mpyi(#$src2, $src3))",
1642 [(set (i32 IntRegs:$dst),
1643 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src3),
1644 u6_2ImmPred:$src2)))]>,
1647 // Rd=add(Ru,mpyi(Rs,#u6))
1648 let isExtendable = 1, opExtendable = 3, isExtentSigned = 0, opExtentBits = 6,
1649 validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in
1650 def ADDr_MPYri_V4 : MInst<(outs IntRegs:$dst),
1651 (ins IntRegs:$src1, IntRegs:$src2, u6Ext:$src3),
1652 "$dst = add($src1, mpyi($src2, #$src3))",
1653 [(set (i32 IntRegs:$dst),
1654 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
1655 u6ExtPred:$src3)))]>,
1656 Requires<[HasV4T]>, ImmRegRel;
1658 // Rx=add(Ru,mpyi(Rx,Rs))
1659 let validSubTargets = HasV4SubT, InputType = "reg", CextOpcode = "ADD_MPY" in
1660 def ADDr_MPYrr_V4 : MInst_acc<(outs IntRegs:$dst),
1661 (ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3),
1662 "$dst = add($src1, mpyi($src2, $src3))",
1663 [(set (i32 IntRegs:$dst),
1664 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
1665 (i32 IntRegs:$src3))))],
1667 Requires<[HasV4T]>, ImmRegRel;
1670 // Polynomial multiply words
1672 // Rxx^=pmpyw(Rs,Rt)
1674 // Vector reduce multiply word by signed half (32x16)
1675 // Rdd=vrmpyweh(Rss,Rtt)[:<<1]
1676 // Rdd=vrmpywoh(Rss,Rtt)[:<<1]
1677 // Rxx+=vrmpyweh(Rss,Rtt)[:<<1]
1678 // Rxx+=vrmpywoh(Rss,Rtt)[:<<1]
1680 // Multiply and use upper result
1681 // Rd=mpy(Rs,Rt.H):<<1:sat
1682 // Rd=mpy(Rs,Rt.L):<<1:sat
1683 // Rd=mpy(Rs,Rt):<<1
1684 // Rd=mpy(Rs,Rt):<<1:sat
1686 // Rx+=mpy(Rs,Rt):<<1:sat
1687 // Rx-=mpy(Rs,Rt):<<1:sat
1689 // Vector multiply bytes
1690 // Rdd=vmpybsu(Rs,Rt)
1691 // Rdd=vmpybu(Rs,Rt)
1692 // Rxx+=vmpybsu(Rs,Rt)
1693 // Rxx+=vmpybu(Rs,Rt)
1695 // Vector polynomial multiply halfwords
1696 // Rdd=vpmpyh(Rs,Rt)
1697 // Rxx^=vpmpyh(Rs,Rt)
1699 //===----------------------------------------------------------------------===//
1701 //===----------------------------------------------------------------------===//
1704 //===----------------------------------------------------------------------===//
1706 //===----------------------------------------------------------------------===//
1708 // Shift by immediate and accumulate.
1709 // Rx=add(#u8,asl(Rx,#U5))
1710 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1711 validSubTargets = HasV4SubT in
1712 def ADDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
1713 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1714 "$dst = add(#$src1, asl($src2, #$src3))",
1715 [(set (i32 IntRegs:$dst),
1716 (add (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
1721 // Rx=add(#u8,lsr(Rx,#U5))
1722 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1723 validSubTargets = HasV4SubT in
1724 def ADDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
1725 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1726 "$dst = add(#$src1, lsr($src2, #$src3))",
1727 [(set (i32 IntRegs:$dst),
1728 (add (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
1733 // Rx=sub(#u8,asl(Rx,#U5))
1734 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1735 validSubTargets = HasV4SubT in
1736 def SUBi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
1737 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1738 "$dst = sub(#$src1, asl($src2, #$src3))",
1739 [(set (i32 IntRegs:$dst),
1740 (sub (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
1745 // Rx=sub(#u8,lsr(Rx,#U5))
1746 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1747 validSubTargets = HasV4SubT in
1748 def SUBi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
1749 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1750 "$dst = sub(#$src1, lsr($src2, #$src3))",
1751 [(set (i32 IntRegs:$dst),
1752 (sub (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
1758 //Shift by immediate and logical.
1759 //Rx=and(#u8,asl(Rx,#U5))
1760 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1761 validSubTargets = HasV4SubT in
1762 def ANDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
1763 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1764 "$dst = and(#$src1, asl($src2, #$src3))",
1765 [(set (i32 IntRegs:$dst),
1766 (and (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
1771 //Rx=and(#u8,lsr(Rx,#U5))
1772 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1773 validSubTargets = HasV4SubT in
1774 def ANDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
1775 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1776 "$dst = and(#$src1, lsr($src2, #$src3))",
1777 [(set (i32 IntRegs:$dst),
1778 (and (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
1783 //Rx=or(#u8,asl(Rx,#U5))
1784 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1785 AddedComplexity = 30, validSubTargets = HasV4SubT in
1786 def ORi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
1787 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1788 "$dst = or(#$src1, asl($src2, #$src3))",
1789 [(set (i32 IntRegs:$dst),
1790 (or (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
1795 //Rx=or(#u8,lsr(Rx,#U5))
1796 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1797 AddedComplexity = 30, validSubTargets = HasV4SubT in
1798 def ORi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
1799 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1800 "$dst = or(#$src1, lsr($src2, #$src3))",
1801 [(set (i32 IntRegs:$dst),
1802 (or (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
1808 //Shift by register.
1810 let validSubTargets = HasV4SubT in {
1811 def LSLi_V4 : MInst<(outs IntRegs:$dst), (ins s6Imm:$src1, IntRegs:$src2),
1812 "$dst = lsl(#$src1, $src2)",
1813 [(set (i32 IntRegs:$dst), (shl s6ImmPred:$src1,
1814 (i32 IntRegs:$src2)))]>,
1818 //Shift by register and logical.
1820 def ASLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
1821 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
1822 "$dst ^= asl($src2, $src3)",
1823 [(set (i64 DoubleRegs:$dst),
1824 (xor (i64 DoubleRegs:$src1), (shl (i64 DoubleRegs:$src2),
1825 (i32 IntRegs:$src3))))],
1830 def ASRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
1831 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
1832 "$dst ^= asr($src2, $src3)",
1833 [(set (i64 DoubleRegs:$dst),
1834 (xor (i64 DoubleRegs:$src1), (sra (i64 DoubleRegs:$src2),
1835 (i32 IntRegs:$src3))))],
1840 def LSLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
1841 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
1842 "$dst ^= lsl($src2, $src3)",
1843 [(set (i64 DoubleRegs:$dst), (xor (i64 DoubleRegs:$src1),
1844 (shl (i64 DoubleRegs:$src2),
1845 (i32 IntRegs:$src3))))],
1850 def LSRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
1851 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
1852 "$dst ^= lsr($src2, $src3)",
1853 [(set (i64 DoubleRegs:$dst),
1854 (xor (i64 DoubleRegs:$src1), (srl (i64 DoubleRegs:$src2),
1855 (i32 IntRegs:$src3))))],
1860 //===----------------------------------------------------------------------===//
1862 //===----------------------------------------------------------------------===//
1864 //===----------------------------------------------------------------------===//
1865 // MEMOP: Word, Half, Byte
1866 //===----------------------------------------------------------------------===//
1868 def MEMOPIMM : SDNodeXForm<imm, [{
1869 // Call the transformation function XformM5ToU5Imm to get the negative
1870 // immediate's positive counterpart.
1871 int32_t imm = N->getSExtValue();
1872 return XformM5ToU5Imm(imm);
1875 def MEMOPIMM_HALF : SDNodeXForm<imm, [{
1876 // -1 .. -31 represented as 65535..65515
1877 // assigning to a short restores our desired signed value.
1878 // Call the transformation function XformM5ToU5Imm to get the negative
1879 // immediate's positive counterpart.
1880 int16_t imm = N->getSExtValue();
1881 return XformM5ToU5Imm(imm);
1884 def MEMOPIMM_BYTE : SDNodeXForm<imm, [{
1885 // -1 .. -31 represented as 255..235
1886 // assigning to a char restores our desired signed value.
1887 // Call the transformation function XformM5ToU5Imm to get the negative
1888 // immediate's positive counterpart.
1889 int8_t imm = N->getSExtValue();
1890 return XformM5ToU5Imm(imm);
1893 def SETMEMIMM : SDNodeXForm<imm, [{
1894 // Return the bit position we will set [0-31].
1896 int32_t imm = N->getSExtValue();
1897 return XformMskToBitPosU5Imm(imm);
1900 def CLRMEMIMM : SDNodeXForm<imm, [{
1901 // Return the bit position we will clear [0-31].
1903 // we bit negate the value first
1904 int32_t imm = ~(N->getSExtValue());
1905 return XformMskToBitPosU5Imm(imm);
1908 def SETMEMIMM_SHORT : SDNodeXForm<imm, [{
1909 // Return the bit position we will set [0-15].
1911 int16_t imm = N->getSExtValue();
1912 return XformMskToBitPosU4Imm(imm);
1915 def CLRMEMIMM_SHORT : SDNodeXForm<imm, [{
1916 // Return the bit position we will clear [0-15].
1918 // we bit negate the value first
1919 int16_t imm = ~(N->getSExtValue());
1920 return XformMskToBitPosU4Imm(imm);
1923 def SETMEMIMM_BYTE : SDNodeXForm<imm, [{
1924 // Return the bit position we will set [0-7].
1926 int8_t imm = N->getSExtValue();
1927 return XformMskToBitPosU3Imm(imm);
1930 def CLRMEMIMM_BYTE : SDNodeXForm<imm, [{
1931 // Return the bit position we will clear [0-7].
1933 // we bit negate the value first
1934 int8_t imm = ~(N->getSExtValue());
1935 return XformMskToBitPosU3Imm(imm);
1938 //===----------------------------------------------------------------------===//
1939 // Template class for MemOp instructions with the register value.
1940 //===----------------------------------------------------------------------===//
1941 class MemOp_rr_base <string opc, bits<2> opcBits, Operand ImmOp,
1942 string memOp, bits<2> memOpBits> :
1944 (ins IntRegs:$base, ImmOp:$offset, IntRegs:$delta),
1945 opc#"($base+#$offset)"#memOp#"$delta",
1947 Requires<[HasV4T, UseMEMOP]> {
1952 bits<6> offsetBits; // memb - u6:0 , memh - u6:1, memw - u6:2
1954 let offsetBits = !if (!eq(opcBits, 0b00), offset{5-0},
1955 !if (!eq(opcBits, 0b01), offset{6-1},
1956 !if (!eq(opcBits, 0b10), offset{7-2},0)));
1958 let IClass = 0b0011;
1959 let Inst{27-24} = 0b1110;
1960 let Inst{22-21} = opcBits;
1961 let Inst{20-16} = base;
1963 let Inst{12-7} = offsetBits;
1964 let Inst{6-5} = memOpBits;
1965 let Inst{4-0} = delta;
1968 //===----------------------------------------------------------------------===//
1969 // Template class for MemOp instructions with the immediate value.
1970 //===----------------------------------------------------------------------===//
1971 class MemOp_ri_base <string opc, bits<2> opcBits, Operand ImmOp,
1972 string memOp, bits<2> memOpBits> :
1974 (ins IntRegs:$base, ImmOp:$offset, u5Imm:$delta),
1975 opc#"($base+#$offset)"#memOp#"#$delta"
1976 #!if(memOpBits{1},")", ""), // clrbit, setbit - include ')'
1978 Requires<[HasV4T, UseMEMOP]> {
1983 bits<6> offsetBits; // memb - u6:0 , memh - u6:1, memw - u6:2
1985 let offsetBits = !if (!eq(opcBits, 0b00), offset{5-0},
1986 !if (!eq(opcBits, 0b01), offset{6-1},
1987 !if (!eq(opcBits, 0b10), offset{7-2},0)));
1989 let IClass = 0b0011;
1990 let Inst{27-24} = 0b1111;
1991 let Inst{22-21} = opcBits;
1992 let Inst{20-16} = base;
1994 let Inst{12-7} = offsetBits;
1995 let Inst{6-5} = memOpBits;
1996 let Inst{4-0} = delta;
1999 // multiclass to define MemOp instructions with register operand.
2000 multiclass MemOp_rr<string opc, bits<2> opcBits, Operand ImmOp> {
2001 def _ADD#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " += ", 0b00>; // add
2002 def _SUB#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " -= ", 0b01>; // sub
2003 def _AND#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " &= ", 0b10>; // and
2004 def _OR#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " |= ", 0b11>; // or
2007 // multiclass to define MemOp instructions with immediate Operand.
2008 multiclass MemOp_ri<string opc, bits<2> opcBits, Operand ImmOp> {
2009 def _ADD#NAME#_V4 : MemOp_ri_base <opc, opcBits, ImmOp, " += ", 0b00 >;
2010 def _SUB#NAME#_V4 : MemOp_ri_base <opc, opcBits, ImmOp, " -= ", 0b01 >;
2011 def _CLRBIT#NAME#_V4 : MemOp_ri_base<opc, opcBits, ImmOp, " =clrbit(", 0b10>;
2012 def _SETBIT#NAME#_V4 : MemOp_ri_base<opc, opcBits, ImmOp, " =setbit(", 0b11>;
2015 multiclass MemOp_base <string opc, bits<2> opcBits, Operand ImmOp> {
2016 defm r : MemOp_rr <opc, opcBits, ImmOp>;
2017 defm i : MemOp_ri <opc, opcBits, ImmOp>;
2020 // Define MemOp instructions.
2021 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0,
2022 validSubTargets =HasV4SubT in {
2023 let opExtentBits = 6, accessSize = ByteAccess in
2024 defm MemOPb : MemOp_base <"memb", 0b00, u6_0Ext>;
2026 let opExtentBits = 7, accessSize = HalfWordAccess in
2027 defm MemOPh : MemOp_base <"memh", 0b01, u6_1Ext>;
2029 let opExtentBits = 8, accessSize = WordAccess in
2030 defm MemOPw : MemOp_base <"memw", 0b10, u6_2Ext>;
2033 //===----------------------------------------------------------------------===//
2034 // Multiclass to define 'Def Pats' for ALU operations on the memory
2035 // Here value used for the ALU operation is an immediate value.
2036 // mem[bh](Rs+#0) += #U5
2037 // mem[bh](Rs+#u6) += #U5
2038 //===----------------------------------------------------------------------===//
2040 multiclass MemOpi_u5Pats <PatFrag ldOp, PatFrag stOp, PatLeaf ExtPred,
2041 InstHexagon MI, SDNode OpNode> {
2042 let AddedComplexity = 180 in
2043 def : Pat < (stOp (OpNode (ldOp IntRegs:$addr), u5ImmPred:$addend),
2045 (MI IntRegs:$addr, #0, u5ImmPred:$addend )>;
2047 let AddedComplexity = 190 in
2048 def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, ExtPred:$offset)),
2050 (add IntRegs:$base, ExtPred:$offset)),
2051 (MI IntRegs:$base, ExtPred:$offset, u5ImmPred:$addend)>;
2054 multiclass MemOpi_u5ALUOp<PatFrag ldOp, PatFrag stOp, PatLeaf ExtPred,
2055 InstHexagon addMI, InstHexagon subMI> {
2056 defm : MemOpi_u5Pats<ldOp, stOp, ExtPred, addMI, add>;
2057 defm : MemOpi_u5Pats<ldOp, stOp, ExtPred, subMI, sub>;
2060 multiclass MemOpi_u5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2062 defm : MemOpi_u5ALUOp <ldOpHalf, truncstorei16, u6_1ExtPred,
2063 MemOPh_ADDi_V4, MemOPh_SUBi_V4>;
2065 defm : MemOpi_u5ALUOp <ldOpByte, truncstorei8, u6ExtPred,
2066 MemOPb_ADDi_V4, MemOPb_SUBi_V4>;
2069 let Predicates = [HasV4T, UseMEMOP] in {
2070 defm : MemOpi_u5ExtType<zextloadi8, zextloadi16>; // zero extend
2071 defm : MemOpi_u5ExtType<sextloadi8, sextloadi16>; // sign extend
2072 defm : MemOpi_u5ExtType<extloadi8, extloadi16>; // any extend
2075 defm : MemOpi_u5ALUOp <load, store, u6_2ExtPred, MemOPw_ADDi_V4,
2079 //===----------------------------------------------------------------------===//
2080 // multiclass to define 'Def Pats' for ALU operations on the memory.
2081 // Here value used for the ALU operation is a negative value.
2082 // mem[bh](Rs+#0) += #m5
2083 // mem[bh](Rs+#u6) += #m5
2084 //===----------------------------------------------------------------------===//
2086 multiclass MemOpi_m5Pats <PatFrag ldOp, PatFrag stOp, PatLeaf extPred,
2087 PatLeaf immPred, ComplexPattern addrPred,
2088 SDNodeXForm xformFunc, InstHexagon MI> {
2089 let AddedComplexity = 190 in
2090 def : Pat <(stOp (add (ldOp IntRegs:$addr), immPred:$subend),
2092 (MI IntRegs:$addr, #0, (xformFunc immPred:$subend) )>;
2094 let AddedComplexity = 195 in
2095 def : Pat<(stOp (add (ldOp (add IntRegs:$base, extPred:$offset)),
2097 (add IntRegs:$base, extPred:$offset)),
2098 (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$subend))>;
2101 multiclass MemOpi_m5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2103 defm : MemOpi_m5Pats <ldOpHalf, truncstorei16, u6_1ExtPred, m5HImmPred,
2104 ADDRriU6_1, MEMOPIMM_HALF, MemOPh_SUBi_V4>;
2106 defm : MemOpi_m5Pats <ldOpByte, truncstorei8, u6ExtPred, m5BImmPred,
2107 ADDRriU6_0, MEMOPIMM_BYTE, MemOPb_SUBi_V4>;
2110 let Predicates = [HasV4T, UseMEMOP] in {
2111 defm : MemOpi_m5ExtType<zextloadi8, zextloadi16>; // zero extend
2112 defm : MemOpi_m5ExtType<sextloadi8, sextloadi16>; // sign extend
2113 defm : MemOpi_m5ExtType<extloadi8, extloadi16>; // any extend
2116 defm : MemOpi_m5Pats <load, store, u6_2ExtPred, m5ImmPred,
2117 ADDRriU6_2, MEMOPIMM, MemOPw_SUBi_V4>;
2120 //===----------------------------------------------------------------------===//
2121 // Multiclass to define 'def Pats' for bit operations on the memory.
2122 // mem[bhw](Rs+#0) = [clrbit|setbit](#U5)
2123 // mem[bhw](Rs+#u6) = [clrbit|setbit](#U5)
2124 //===----------------------------------------------------------------------===//
2126 multiclass MemOpi_bitPats <PatFrag ldOp, PatFrag stOp, PatLeaf immPred,
2127 PatLeaf extPred, ComplexPattern addrPred,
2128 SDNodeXForm xformFunc, InstHexagon MI, SDNode OpNode> {
2130 // mem[bhw](Rs+#u6:[012]) = [clrbit|setbit](#U5)
2131 let AddedComplexity = 250 in
2132 def : Pat<(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)),
2134 (add IntRegs:$base, extPred:$offset)),
2135 (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$bitend))>;
2137 // mem[bhw](Rs+#0) = [clrbit|setbit](#U5)
2138 let AddedComplexity = 225 in
2139 def : Pat <(stOp (OpNode (ldOp (addrPred IntRegs:$addr, extPred:$offset)),
2141 (addrPred (i32 IntRegs:$addr), extPred:$offset)),
2142 (MI IntRegs:$addr, extPred:$offset, (xformFunc immPred:$bitend))>;
2145 multiclass MemOpi_bitExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2147 defm : MemOpi_bitPats<ldOpByte, truncstorei8, Clr3ImmPred, u6ExtPred,
2148 ADDRriU6_0, CLRMEMIMM_BYTE, MemOPb_CLRBITi_V4, and>;
2150 defm : MemOpi_bitPats<ldOpByte, truncstorei8, Set3ImmPred, u6ExtPred,
2151 ADDRriU6_0, SETMEMIMM_BYTE, MemOPb_SETBITi_V4, or>;
2152 // Half Word - clrbit
2153 defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Clr4ImmPred, u6_1ExtPred,
2154 ADDRriU6_1, CLRMEMIMM_SHORT, MemOPh_CLRBITi_V4, and>;
2155 // Half Word - setbit
2156 defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Set4ImmPred, u6_1ExtPred,
2157 ADDRriU6_1, SETMEMIMM_SHORT, MemOPh_SETBITi_V4, or>;
2160 let Predicates = [HasV4T, UseMEMOP] in {
2161 // mem[bh](Rs+#0) = [clrbit|setbit](#U5)
2162 // mem[bh](Rs+#u6:[01]) = [clrbit|setbit](#U5)
2163 defm : MemOpi_bitExtType<zextloadi8, zextloadi16>; // zero extend
2164 defm : MemOpi_bitExtType<sextloadi8, sextloadi16>; // sign extend
2165 defm : MemOpi_bitExtType<extloadi8, extloadi16>; // any extend
2167 // memw(Rs+#0) = [clrbit|setbit](#U5)
2168 // memw(Rs+#u6:2) = [clrbit|setbit](#U5)
2169 defm : MemOpi_bitPats<load, store, Clr5ImmPred, u6_2ExtPred, ADDRriU6_2,
2170 CLRMEMIMM, MemOPw_CLRBITi_V4, and>;
2171 defm : MemOpi_bitPats<load, store, Set5ImmPred, u6_2ExtPred, ADDRriU6_2,
2172 SETMEMIMM, MemOPw_SETBITi_V4, or>;
2175 //===----------------------------------------------------------------------===//
2176 // Multiclass to define 'def Pats' for ALU operations on the memory
2177 // where addend is a register.
2178 // mem[bhw](Rs+#0) [+-&|]= Rt
2179 // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt
2180 //===----------------------------------------------------------------------===//
2182 multiclass MemOpr_Pats <PatFrag ldOp, PatFrag stOp, ComplexPattern addrPred,
2183 PatLeaf extPred, InstHexagon MI, SDNode OpNode> {
2184 let AddedComplexity = 141 in
2185 // mem[bhw](Rs+#0) [+-&|]= Rt
2186 def : Pat <(stOp (OpNode (ldOp (addrPred IntRegs:$addr, extPred:$offset)),
2187 (i32 IntRegs:$addend)),
2188 (addrPred (i32 IntRegs:$addr), extPred:$offset)),
2189 (MI IntRegs:$addr, extPred:$offset, (i32 IntRegs:$addend) )>;
2191 // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt
2192 let AddedComplexity = 150 in
2193 def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)),
2194 (i32 IntRegs:$orend)),
2195 (add IntRegs:$base, extPred:$offset)),
2196 (MI IntRegs:$base, extPred:$offset, (i32 IntRegs:$orend) )>;
2199 multiclass MemOPr_ALUOp<PatFrag ldOp, PatFrag stOp,
2200 ComplexPattern addrPred, PatLeaf extPred,
2201 InstHexagon addMI, InstHexagon subMI,
2202 InstHexagon andMI, InstHexagon orMI > {
2204 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, addMI, add>;
2205 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, subMI, sub>;
2206 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, andMI, and>;
2207 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, orMI, or>;
2210 multiclass MemOPr_ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2212 defm : MemOPr_ALUOp <ldOpHalf, truncstorei16, ADDRriU6_1, u6_1ExtPred,
2213 MemOPh_ADDr_V4, MemOPh_SUBr_V4,
2214 MemOPh_ANDr_V4, MemOPh_ORr_V4>;
2216 defm : MemOPr_ALUOp <ldOpByte, truncstorei8, ADDRriU6_0, u6ExtPred,
2217 MemOPb_ADDr_V4, MemOPb_SUBr_V4,
2218 MemOPb_ANDr_V4, MemOPb_ORr_V4>;
2221 // Define 'def Pats' for MemOps with register addend.
2222 let Predicates = [HasV4T, UseMEMOP] in {
2224 defm : MemOPr_ExtType<zextloadi8, zextloadi16>; // zero extend
2225 defm : MemOPr_ExtType<sextloadi8, sextloadi16>; // sign extend
2226 defm : MemOPr_ExtType<extloadi8, extloadi16>; // any extend
2228 defm : MemOPr_ALUOp <load, store, ADDRriU6_2, u6_2ExtPred, MemOPw_ADDr_V4,
2229 MemOPw_SUBr_V4, MemOPw_ANDr_V4, MemOPw_ORr_V4 >;
2232 //===----------------------------------------------------------------------===//
2234 //===----------------------------------------------------------------------===//
2236 // Hexagon V4 only supports these flavors of byte/half compare instructions:
2237 // EQ/GT/GTU. Other flavors like GE/GEU/LT/LTU/LE/LEU are not supported by
2238 // hardware. However, compiler can still implement these patterns through
2239 // appropriate patterns combinations based on current implemented patterns.
2240 // The implemented patterns are: EQ/GT/GTU.
2241 // Missing patterns are: GE/GEU/LT/LTU/LE/LEU.
2243 // Following instruction is not being extended as it results into the
2244 // incorrect code for negative numbers.
2245 // Pd=cmpb.eq(Rs,#u8)
2247 let isCompare = 1, isExtendable = 1, opExtendable = 2, hasSideEffects = 0,
2248 validSubTargets = HasV4SubT in
2249 class CMP_NOT_REG_IMM<string OpName, bits<2> op, Operand ImmOp,
2251 : ALU32Inst <(outs PredRegs:$dst), (ins IntRegs:$src1, ImmOp:$src2),
2252 "$dst = !cmp."#OpName#"($src1, #$src2)",
2254 "", ALU32_2op_tc_2early_SLOT0123> {
2259 let IClass = 0b0111;
2260 let Inst{27-24} = 0b0101;
2261 let Inst{23-22} = op;
2262 let Inst{20-16} = src1;
2263 let Inst{21} = !if (!eq(OpName, "gtu"), 0b0, src2{9});
2264 let Inst{13-5} = src2{8-0};
2265 let Inst{4-2} = 0b100;
2266 let Inst{1-0} = dst;
2269 let opExtentBits = 10, isExtentSigned = 1 in {
2270 def C4_cmpneqi : CMP_NOT_REG_IMM <"eq", 0b00, s10Ext, [(set (i1 PredRegs:$dst),
2271 (setne (i32 IntRegs:$src1), s10ExtPred:$src2))]>;
2273 def C4_cmpltei : CMP_NOT_REG_IMM <"gt", 0b01, s10Ext, [(set (i1 PredRegs:$dst),
2274 (not (setgt (i32 IntRegs:$src1), s10ExtPred:$src2)))]>;
2277 let opExtentBits = 9 in
2278 def C4_cmplteui : CMP_NOT_REG_IMM <"gtu", 0b10, u9Ext, [(set (i1 PredRegs:$dst),
2279 (not (setugt (i32 IntRegs:$src1), u9ExtPred:$src2)))]>;
2284 let isCompare = 1, validSubTargets = HasV4SubT in
2285 def CMPnotEQ_rr : ALU32_rr<(outs PredRegs:$dst),
2286 (ins IntRegs:$src1, IntRegs:$src2),
2287 "$dst = !cmp.eq($src1, $src2)",
2288 [(set (i1 PredRegs:$dst),
2289 (setne (i32 IntRegs:$src1), (i32 IntRegs:$src2)))]>,
2293 let isCompare = 1, validSubTargets = HasV4SubT in
2294 def CMPnotGT_rr : ALU32_rr<(outs PredRegs:$dst),
2295 (ins IntRegs:$src1, IntRegs:$src2),
2296 "$dst = !cmp.gt($src1, $src2)",
2297 [(set (i1 PredRegs:$dst),
2298 (not (setgt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))]>,
2302 // p=!cmp.gtu(r1,r2)
2303 let isCompare = 1, validSubTargets = HasV4SubT in
2304 def CMPnotGTU_rr : ALU32_rr<(outs PredRegs:$dst),
2305 (ins IntRegs:$src1, IntRegs:$src2),
2306 "$dst = !cmp.gtu($src1, $src2)",
2307 [(set (i1 PredRegs:$dst),
2308 (not (setugt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))]>,
2311 let isCompare = 1, validSubTargets = HasV4SubT in
2312 def CMPbEQri_V4 : MInst<(outs PredRegs:$dst),
2313 (ins IntRegs:$src1, u8Imm:$src2),
2314 "$dst = cmpb.eq($src1, #$src2)",
2315 [(set (i1 PredRegs:$dst),
2316 (seteq (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2))]>,
2319 def : Pat <(brcond (i1 (setne (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2)),
2321 (J2_jumpf (CMPbEQri_V4 (i32 IntRegs:$src1), u8ImmPred:$src2),
2325 // Pd=cmpb.eq(Rs,Rt)
2326 let isCompare = 1, validSubTargets = HasV4SubT in
2327 def CMPbEQrr_ubub_V4 : MInst<(outs PredRegs:$dst),
2328 (ins IntRegs:$src1, IntRegs:$src2),
2329 "$dst = cmpb.eq($src1, $src2)",
2330 [(set (i1 PredRegs:$dst),
2331 (seteq (and (xor (i32 IntRegs:$src1),
2332 (i32 IntRegs:$src2)), 255), 0))]>,
2335 // Pd=cmpb.eq(Rs,Rt)
2336 let isCompare = 1, validSubTargets = HasV4SubT in
2337 def CMPbEQrr_sbsb_V4 : MInst<(outs PredRegs:$dst),
2338 (ins IntRegs:$src1, IntRegs:$src2),
2339 "$dst = cmpb.eq($src1, $src2)",
2340 [(set (i1 PredRegs:$dst),
2341 (seteq (shl (i32 IntRegs:$src1), (i32 24)),
2342 (shl (i32 IntRegs:$src2), (i32 24))))]>,
2345 // Pd=cmpb.gt(Rs,Rt)
2346 let isCompare = 1, validSubTargets = HasV4SubT in
2347 def CMPbGTrr_V4 : MInst<(outs PredRegs:$dst),
2348 (ins IntRegs:$src1, IntRegs:$src2),
2349 "$dst = cmpb.gt($src1, $src2)",
2350 [(set (i1 PredRegs:$dst),
2351 (setgt (shl (i32 IntRegs:$src1), (i32 24)),
2352 (shl (i32 IntRegs:$src2), (i32 24))))]>,
2355 // Pd=cmpb.gtu(Rs,#u7)
2356 let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7,
2357 isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU", InputType = "imm" in
2358 def CMPbGTUri_V4 : MInst<(outs PredRegs:$dst),
2359 (ins IntRegs:$src1, u7Ext:$src2),
2360 "$dst = cmpb.gtu($src1, #$src2)",
2361 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255),
2362 u7ExtPred:$src2))]>,
2363 Requires<[HasV4T]>, ImmRegRel;
2365 // SDNode for converting immediate C to C-1.
2366 def DEC_CONST_BYTE : SDNodeXForm<imm, [{
2367 // Return the byte immediate const-1 as an SDNode.
2368 int32_t imm = N->getSExtValue();
2369 return XformU7ToU7M1Imm(imm);
2373 // zext( seteq ( and(Rs, 255), u8))
2375 // Pd=cmpb.eq(Rs, #u8)
2376 // if (Pd.new) Rd=#1
2377 // if (!Pd.new) Rd=#0
2378 def : Pat <(i32 (zext (i1 (seteq (i32 (and (i32 IntRegs:$Rs), 255)),
2380 (i32 (TFR_condset_ii (i1 (CMPbEQri_V4 (i32 IntRegs:$Rs),
2386 // zext( setne ( and(Rs, 255), u8))
2388 // Pd=cmpb.eq(Rs, #u8)
2389 // if (Pd.new) Rd=#0
2390 // if (!Pd.new) Rd=#1
2391 def : Pat <(i32 (zext (i1 (setne (i32 (and (i32 IntRegs:$Rs), 255)),
2393 (i32 (TFR_condset_ii (i1 (CMPbEQri_V4 (i32 IntRegs:$Rs),
2399 // zext( seteq (Rs, and(Rt, 255)))
2401 // Pd=cmpb.eq(Rs, Rt)
2402 // if (Pd.new) Rd=#1
2403 // if (!Pd.new) Rd=#0
2404 def : Pat <(i32 (zext (i1 (seteq (i32 IntRegs:$Rt),
2405 (i32 (and (i32 IntRegs:$Rs), 255)))))),
2406 (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs),
2407 (i32 IntRegs:$Rt))),
2412 // zext( setne (Rs, and(Rt, 255)))
2414 // Pd=cmpb.eq(Rs, Rt)
2415 // if (Pd.new) Rd=#0
2416 // if (!Pd.new) Rd=#1
2417 def : Pat <(i32 (zext (i1 (setne (i32 IntRegs:$Rt),
2418 (i32 (and (i32 IntRegs:$Rs), 255)))))),
2419 (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs),
2420 (i32 IntRegs:$Rt))),
2425 // zext( setugt ( and(Rs, 255), u8))
2427 // Pd=cmpb.gtu(Rs, #u8)
2428 // if (Pd.new) Rd=#1
2429 // if (!Pd.new) Rd=#0
2430 def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 255)),
2432 (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs),
2438 // zext( setugt ( and(Rs, 254), u8))
2440 // Pd=cmpb.gtu(Rs, #u8)
2441 // if (Pd.new) Rd=#1
2442 // if (!Pd.new) Rd=#0
2443 def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 254)),
2445 (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs),
2451 // zext( setult ( Rs, Rt))
2453 // Pd=cmp.ltu(Rs, Rt)
2454 // if (Pd.new) Rd=#1
2455 // if (!Pd.new) Rd=#0
2456 // cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs)
2457 def : Pat <(i32 (zext (i1 (setult (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2458 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rt),
2459 (i32 IntRegs:$Rs))),
2464 // zext( setlt ( Rs, Rt))
2466 // Pd=cmp.lt(Rs, Rt)
2467 // if (Pd.new) Rd=#1
2468 // if (!Pd.new) Rd=#0
2469 // cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs)
2470 def : Pat <(i32 (zext (i1 (setlt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2471 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rt),
2472 (i32 IntRegs:$Rs))),
2477 // zext( setugt ( Rs, Rt))
2479 // Pd=cmp.gtu(Rs, Rt)
2480 // if (Pd.new) Rd=#1
2481 // if (!Pd.new) Rd=#0
2482 def : Pat <(i32 (zext (i1 (setugt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2483 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rs),
2484 (i32 IntRegs:$Rt))),
2488 // This pattern interefers with coremark performance, not implementing at this
2491 // zext( setgt ( Rs, Rt))
2493 // Pd=cmp.gt(Rs, Rt)
2494 // if (Pd.new) Rd=#1
2495 // if (!Pd.new) Rd=#0
2498 // zext( setuge ( Rs, Rt))
2500 // Pd=cmp.ltu(Rs, Rt)
2501 // if (Pd.new) Rd=#0
2502 // if (!Pd.new) Rd=#1
2503 // cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs)
2504 def : Pat <(i32 (zext (i1 (setuge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2505 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rt),
2506 (i32 IntRegs:$Rs))),
2511 // zext( setge ( Rs, Rt))
2513 // Pd=cmp.lt(Rs, Rt)
2514 // if (Pd.new) Rd=#0
2515 // if (!Pd.new) Rd=#1
2516 // cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs)
2517 def : Pat <(i32 (zext (i1 (setge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2518 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rt),
2519 (i32 IntRegs:$Rs))),
2524 // zext( setule ( Rs, Rt))
2526 // Pd=cmp.gtu(Rs, Rt)
2527 // if (Pd.new) Rd=#0
2528 // if (!Pd.new) Rd=#1
2529 def : Pat <(i32 (zext (i1 (setule (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2530 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rs),
2531 (i32 IntRegs:$Rt))),
2536 // zext( setle ( Rs, Rt))
2538 // Pd=cmp.gt(Rs, Rt)
2539 // if (Pd.new) Rd=#0
2540 // if (!Pd.new) Rd=#1
2541 def : Pat <(i32 (zext (i1 (setle (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2542 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rs),
2543 (i32 IntRegs:$Rt))),
2548 // zext( setult ( and(Rs, 255), u8))
2549 // Use the isdigit transformation below
2551 // Generate code of the form 'mux_ii(cmpbgtu(Rdd, C-1),0,1)'
2552 // for C code of the form r = ((c>='0') & (c<='9')) ? 1 : 0;.
2553 // The isdigit transformation relies on two 'clever' aspects:
2554 // 1) The data type is unsigned which allows us to eliminate a zero test after
2555 // biasing the expression by 48. We are depending on the representation of
2556 // the unsigned types, and semantics.
2557 // 2) The front end has converted <= 9 into < 10 on entry to LLVM
2560 // retval = ((c>='0') & (c<='9')) ? 1 : 0;
2561 // The code is transformed upstream of llvm into
2562 // retval = (c-48) < 10 ? 1 : 0;
2563 let AddedComplexity = 139 in
2564 def : Pat <(i32 (zext (i1 (setult (i32 (and (i32 IntRegs:$src1), 255)),
2565 u7StrictPosImmPred:$src2)))),
2566 (i32 (C2_muxii (i1 (CMPbGTUri_V4 (i32 IntRegs:$src1),
2567 (DEC_CONST_BYTE u7StrictPosImmPred:$src2))),
2571 // Pd=cmpb.gtu(Rs,Rt)
2572 let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU",
2573 InputType = "reg" in
2574 def CMPbGTUrr_V4 : MInst<(outs PredRegs:$dst),
2575 (ins IntRegs:$src1, IntRegs:$src2),
2576 "$dst = cmpb.gtu($src1, $src2)",
2577 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255),
2578 (and (i32 IntRegs:$src2), 255)))]>,
2579 Requires<[HasV4T]>, ImmRegRel;
2581 // Following instruction is not being extended as it results into the incorrect
2582 // code for negative numbers.
2584 // Signed half compare(.eq) ri.
2585 // Pd=cmph.eq(Rs,#s8)
2586 let isCompare = 1, validSubTargets = HasV4SubT in
2587 def CMPhEQri_V4 : MInst<(outs PredRegs:$dst),
2588 (ins IntRegs:$src1, s8Imm:$src2),
2589 "$dst = cmph.eq($src1, #$src2)",
2590 [(set (i1 PredRegs:$dst), (seteq (and (i32 IntRegs:$src1), 65535),
2591 s8ImmPred:$src2))]>,
2594 // Signed half compare(.eq) rr.
2595 // Case 1: xor + and, then compare:
2597 // r0=and(r0,#0xffff)
2599 // Pd=cmph.eq(Rs,Rt)
2600 let isCompare = 1, validSubTargets = HasV4SubT in
2601 def CMPhEQrr_xor_V4 : MInst<(outs PredRegs:$dst),
2602 (ins IntRegs:$src1, IntRegs:$src2),
2603 "$dst = cmph.eq($src1, $src2)",
2604 [(set (i1 PredRegs:$dst), (seteq (and (xor (i32 IntRegs:$src1),
2605 (i32 IntRegs:$src2)),
2609 // Signed half compare(.eq) rr.
2610 // Case 2: shift left 16 bits then compare:
2614 // Pd=cmph.eq(Rs,Rt)
2615 let isCompare = 1, validSubTargets = HasV4SubT in
2616 def CMPhEQrr_shl_V4 : MInst<(outs PredRegs:$dst),
2617 (ins IntRegs:$src1, IntRegs:$src2),
2618 "$dst = cmph.eq($src1, $src2)",
2619 [(set (i1 PredRegs:$dst),
2620 (seteq (shl (i32 IntRegs:$src1), (i32 16)),
2621 (shl (i32 IntRegs:$src2), (i32 16))))]>,
2624 /* Incorrect Pattern -- immediate should be right shifted before being
2625 used in the cmph.gt instruction.
2626 // Signed half compare(.gt) ri.
2627 // Pd=cmph.gt(Rs,#s8)
2629 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 8,
2630 isCompare = 1, validSubTargets = HasV4SubT in
2631 def CMPhGTri_V4 : MInst<(outs PredRegs:$dst),
2632 (ins IntRegs:$src1, s8Ext:$src2),
2633 "$dst = cmph.gt($src1, #$src2)",
2634 [(set (i1 PredRegs:$dst),
2635 (setgt (shl (i32 IntRegs:$src1), (i32 16)),
2636 s8ExtPred:$src2))]>,
2640 // Signed half compare(.gt) rr.
2641 // Pd=cmph.gt(Rs,Rt)
2642 let isCompare = 1, validSubTargets = HasV4SubT in
2643 def CMPhGTrr_shl_V4 : MInst<(outs PredRegs:$dst),
2644 (ins IntRegs:$src1, IntRegs:$src2),
2645 "$dst = cmph.gt($src1, $src2)",
2646 [(set (i1 PredRegs:$dst),
2647 (setgt (shl (i32 IntRegs:$src1), (i32 16)),
2648 (shl (i32 IntRegs:$src2), (i32 16))))]>,
2651 // Unsigned half compare rr (.gtu).
2652 // Pd=cmph.gtu(Rs,Rt)
2653 let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU",
2654 InputType = "reg" in
2655 def CMPhGTUrr_V4 : MInst<(outs PredRegs:$dst),
2656 (ins IntRegs:$src1, IntRegs:$src2),
2657 "$dst = cmph.gtu($src1, $src2)",
2658 [(set (i1 PredRegs:$dst),
2659 (setugt (and (i32 IntRegs:$src1), 65535),
2660 (and (i32 IntRegs:$src2), 65535)))]>,
2661 Requires<[HasV4T]>, ImmRegRel;
2663 // Unsigned half compare ri (.gtu).
2664 // Pd=cmph.gtu(Rs,#u7)
2665 let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7,
2666 isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU",
2667 InputType = "imm" in
2668 def CMPhGTUri_V4 : MInst<(outs PredRegs:$dst),
2669 (ins IntRegs:$src1, u7Ext:$src2),
2670 "$dst = cmph.gtu($src1, #$src2)",
2671 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 65535),
2672 u7ExtPred:$src2))]>,
2673 Requires<[HasV4T]>, ImmRegRel;
2675 let validSubTargets = HasV4SubT in
2676 def NTSTBIT_rr : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
2677 "$dst = !tstbit($src1, $src2)",
2678 [(set (i1 PredRegs:$dst),
2679 (seteq (and (shl 1, (i32 IntRegs:$src2)), (i32 IntRegs:$src1)), 0))]>,
2682 let validSubTargets = HasV4SubT in
2683 def NTSTBIT_ri : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
2684 "$dst = !tstbit($src1, $src2)",
2685 [(set (i1 PredRegs:$dst),
2686 (seteq (and (shl 1, u5ImmPred:$src2), (i32 IntRegs:$src1)), 0))]>,
2689 //===----------------------------------------------------------------------===//
2691 //===----------------------------------------------------------------------===//
2693 //Deallocate frame and return.
2695 let isReturn = 1, isTerminator = 1, isBarrier = 1, isPredicable = 1,
2696 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0 in {
2697 let validSubTargets = HasV4SubT in
2698 def DEALLOC_RET_V4 : LD0Inst<(outs), (ins),
2704 // Restore registers and dealloc return function call.
2705 let isCall = 1, isBarrier = 1, isReturn = 1, isTerminator = 1,
2706 Defs = [R29, R30, R31, PC] in {
2707 let validSubTargets = HasV4SubT in
2708 def RESTORE_DEALLOC_RET_JMP_V4 : JInst<(outs),
2709 (ins calltarget:$dst),
2715 // Restore registers and dealloc frame before a tail call.
2716 let isCall = 1, isBarrier = 1,
2717 Defs = [R29, R30, R31, PC] in {
2718 let validSubTargets = HasV4SubT in
2719 def RESTORE_DEALLOC_BEFORE_TAILCALL_V4 : JInst<(outs),
2720 (ins calltarget:$dst),
2726 // Save registers function call.
2727 let isCall = 1, isBarrier = 1,
2728 Uses = [R29, R31] in {
2729 def SAVE_REGISTERS_CALL_V4 : JInst<(outs),
2730 (ins calltarget:$dst),
2731 "call $dst // Save_calle_saved_registers",
2736 // if (Ps) dealloc_return
2737 let isReturn = 1, isTerminator = 1,
2738 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
2739 isPredicated = 1 in {
2740 let validSubTargets = HasV4SubT in
2741 def DEALLOC_RET_cPt_V4 : LD0Inst<(outs),
2742 (ins PredRegs:$src1),
2743 "if ($src1) dealloc_return",
2748 // if (!Ps) dealloc_return
2749 let isReturn = 1, isTerminator = 1,
2750 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
2751 isPredicated = 1, isPredicatedFalse = 1 in {
2752 let validSubTargets = HasV4SubT in
2753 def DEALLOC_RET_cNotPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
2754 "if (!$src1) dealloc_return",
2759 // if (Ps.new) dealloc_return:nt
2760 let isReturn = 1, isTerminator = 1,
2761 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
2762 isPredicated = 1 in {
2763 let validSubTargets = HasV4SubT in
2764 def DEALLOC_RET_cdnPnt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
2765 "if ($src1.new) dealloc_return:nt",
2770 // if (!Ps.new) dealloc_return:nt
2771 let isReturn = 1, isTerminator = 1,
2772 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
2773 isPredicated = 1, isPredicatedFalse = 1 in {
2774 let validSubTargets = HasV4SubT in
2775 def DEALLOC_RET_cNotdnPnt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
2776 "if (!$src1.new) dealloc_return:nt",
2781 // if (Ps.new) dealloc_return:t
2782 let isReturn = 1, isTerminator = 1,
2783 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
2784 isPredicated = 1 in {
2785 let validSubTargets = HasV4SubT in
2786 def DEALLOC_RET_cdnPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
2787 "if ($src1.new) dealloc_return:t",
2792 // if (!Ps.new) dealloc_return:nt
2793 let isReturn = 1, isTerminator = 1,
2794 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
2795 isPredicated = 1, isPredicatedFalse = 1 in {
2796 let validSubTargets = HasV4SubT in
2797 def DEALLOC_RET_cNotdnPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
2798 "if (!$src1.new) dealloc_return:t",
2803 // Load/Store with absolute addressing mode
2806 multiclass ST_Abs_Predbase<string mnemonic, RegisterClass RC, bit isNot,
2808 let isPredicatedNew = isPredNew in
2809 def NAME#_V4 : STInst2<(outs),
2810 (ins PredRegs:$src1, u0AlwaysExt:$absaddr, RC: $src2),
2811 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
2812 ") ")#mnemonic#"(##$absaddr) = $src2",
2817 multiclass ST_Abs_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
2818 let isPredicatedFalse = PredNot in {
2819 defm _c#NAME : ST_Abs_Predbase<mnemonic, RC, PredNot, 0>;
2821 defm _cdn#NAME : ST_Abs_Predbase<mnemonic, RC, PredNot, 1>;
2825 let isNVStorable = 1, isExtended = 1, hasSideEffects = 0 in
2826 multiclass ST_Abs<string mnemonic, string CextOp, RegisterClass RC> {
2827 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
2828 let opExtendable = 0, isPredicable = 1 in
2829 def NAME#_V4 : STInst2<(outs),
2830 (ins u0AlwaysExt:$absaddr, RC:$src),
2831 mnemonic#"(##$absaddr) = $src",
2835 let opExtendable = 1, isPredicated = 1 in {
2836 defm Pt : ST_Abs_Pred<mnemonic, RC, 0>;
2837 defm NotPt : ST_Abs_Pred<mnemonic, RC, 1>;
2842 multiclass ST_Abs_Predbase_nv<string mnemonic, RegisterClass RC, bit isNot,
2844 let isPredicatedNew = isPredNew in
2845 def NAME#_nv_V4 : NVInst_V4<(outs),
2846 (ins PredRegs:$src1, u0AlwaysExt:$absaddr, RC: $src2),
2847 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
2848 ") ")#mnemonic#"(##$absaddr) = $src2.new",
2853 multiclass ST_Abs_Pred_nv<string mnemonic, RegisterClass RC, bit PredNot> {
2854 let isPredicatedFalse = PredNot in {
2855 defm _c#NAME : ST_Abs_Predbase_nv<mnemonic, RC, PredNot, 0>;
2857 defm _cdn#NAME : ST_Abs_Predbase_nv<mnemonic, RC, PredNot, 1>;
2861 let mayStore = 1, isNVStore = 1, isExtended = 1, hasSideEffects = 0 in
2862 multiclass ST_Abs_nv<string mnemonic, string CextOp, RegisterClass RC> {
2863 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
2864 let opExtendable = 0, isPredicable = 1 in
2865 def NAME#_nv_V4 : NVInst_V4<(outs),
2866 (ins u0AlwaysExt:$absaddr, RC:$src),
2867 mnemonic#"(##$absaddr) = $src.new",
2871 let opExtendable = 1, isPredicated = 1 in {
2872 defm Pt : ST_Abs_Pred_nv<mnemonic, RC, 0>;
2873 defm NotPt : ST_Abs_Pred_nv<mnemonic, RC, 1>;
2878 let addrMode = Absolute in {
2879 let accessSize = ByteAccess in
2880 defm STrib_abs : ST_Abs<"memb", "STrib", IntRegs>,
2881 ST_Abs_nv<"memb", "STrib", IntRegs>, AddrModeRel;
2883 let accessSize = HalfWordAccess in
2884 defm STrih_abs : ST_Abs<"memh", "STrih", IntRegs>,
2885 ST_Abs_nv<"memh", "STrih", IntRegs>, AddrModeRel;
2887 let accessSize = WordAccess in
2888 defm STriw_abs : ST_Abs<"memw", "STriw", IntRegs>,
2889 ST_Abs_nv<"memw", "STriw", IntRegs>, AddrModeRel;
2891 let accessSize = DoubleWordAccess, isNVStorable = 0 in
2892 defm STrid_abs : ST_Abs<"memd", "STrid", DoubleRegs>, AddrModeRel;
2895 let Predicates = [HasV4T], AddedComplexity = 30 in {
2896 def : Pat<(truncstorei8 (i32 IntRegs:$src1),
2897 (HexagonCONST32 tglobaladdr:$absaddr)),
2898 (STrib_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
2900 def : Pat<(truncstorei16 (i32 IntRegs:$src1),
2901 (HexagonCONST32 tglobaladdr:$absaddr)),
2902 (STrih_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
2904 def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32 tglobaladdr:$absaddr)),
2905 (STriw_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
2907 def : Pat<(store (i64 DoubleRegs:$src1),
2908 (HexagonCONST32 tglobaladdr:$absaddr)),
2909 (STrid_abs_V4 tglobaladdr: $absaddr, DoubleRegs: $src1)>;
2912 //===----------------------------------------------------------------------===//
2913 // multiclass for store instructions with GP-relative addressing mode.
2914 // mem[bhwd](#global)=Rt
2915 // if ([!]Pv[.new]) mem[bhwd](##global) = Rt
2916 //===----------------------------------------------------------------------===//
2917 let mayStore = 1, isNVStorable = 1 in
2918 multiclass ST_GP<string mnemonic, string BaseOp, RegisterClass RC> {
2919 let BaseOpcode = BaseOp, isPredicable = 1 in
2920 def NAME#_V4 : STInst2<(outs),
2921 (ins globaladdress:$global, RC:$src),
2922 mnemonic#"(#$global) = $src",
2925 // When GP-relative instructions are predicated, their addressing mode is
2926 // changed to absolute and they are always constant extended.
2927 let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1,
2928 isPredicated = 1 in {
2929 defm Pt : ST_Abs_Pred <mnemonic, RC, 0>;
2930 defm NotPt : ST_Abs_Pred <mnemonic, RC, 1>;
2934 let mayStore = 1, isNVStore = 1 in
2935 multiclass ST_GP_nv<string mnemonic, string BaseOp, RegisterClass RC> {
2936 let BaseOpcode = BaseOp, isPredicable = 1 in
2937 def NAME#_nv_V4 : NVInst_V4<(outs),
2938 (ins u0AlwaysExt:$global, RC:$src),
2939 mnemonic#"(#$global) = $src.new",
2943 // When GP-relative instructions are predicated, their addressing mode is
2944 // changed to absolute and they are always constant extended.
2945 let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1,
2946 isPredicated = 1 in {
2947 defm Pt : ST_Abs_Pred_nv<mnemonic, RC, 0>;
2948 defm NotPt : ST_Abs_Pred_nv<mnemonic, RC, 1>;
2952 let validSubTargets = HasV4SubT, hasSideEffects = 0 in {
2953 let isNVStorable = 0 in
2954 defm STd_GP : ST_GP <"memd", "STd_GP", DoubleRegs>, PredNewRel;
2956 defm STb_GP : ST_GP<"memb", "STb_GP", IntRegs>,
2957 ST_GP_nv<"memb", "STb_GP", IntRegs>, NewValueRel;
2958 defm STh_GP : ST_GP<"memh", "STh_GP", IntRegs>,
2959 ST_GP_nv<"memh", "STh_GP", IntRegs>, NewValueRel;
2960 defm STw_GP : ST_GP<"memw", "STw_GP", IntRegs>,
2961 ST_GP_nv<"memw", "STw_GP", IntRegs>, NewValueRel;
2964 // 64 bit atomic store
2965 def : Pat <(atomic_store_64 (HexagonCONST32_GP tglobaladdr:$global),
2966 (i64 DoubleRegs:$src1)),
2967 (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>,
2970 // Map from store(globaladdress) -> memd(#foo)
2971 let AddedComplexity = 100 in
2972 def : Pat <(store (i64 DoubleRegs:$src1),
2973 (HexagonCONST32_GP tglobaladdr:$global)),
2974 (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>;
2976 // 8 bit atomic store
2977 def : Pat < (atomic_store_8 (HexagonCONST32_GP tglobaladdr:$global),
2978 (i32 IntRegs:$src1)),
2979 (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
2981 // Map from store(globaladdress) -> memb(#foo)
2982 let AddedComplexity = 100 in
2983 def : Pat<(truncstorei8 (i32 IntRegs:$src1),
2984 (HexagonCONST32_GP tglobaladdr:$global)),
2985 (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
2987 // Map from "i1 = constant<-1>; memw(CONST32(#foo)) = i1"
2988 // to "r0 = 1; memw(#foo) = r0"
2989 let AddedComplexity = 100 in
2990 def : Pat<(store (i1 -1), (HexagonCONST32_GP tglobaladdr:$global)),
2991 (STb_GP_V4 tglobaladdr:$global, (A2_tfrsi 1))>;
2993 def : Pat<(atomic_store_16 (HexagonCONST32_GP tglobaladdr:$global),
2994 (i32 IntRegs:$src1)),
2995 (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
2997 // Map from store(globaladdress) -> memh(#foo)
2998 let AddedComplexity = 100 in
2999 def : Pat<(truncstorei16 (i32 IntRegs:$src1),
3000 (HexagonCONST32_GP tglobaladdr:$global)),
3001 (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3003 // 32 bit atomic store
3004 def : Pat<(atomic_store_32 (HexagonCONST32_GP tglobaladdr:$global),
3005 (i32 IntRegs:$src1)),
3006 (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3008 // Map from store(globaladdress) -> memw(#foo)
3009 let AddedComplexity = 100 in
3010 def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)),
3011 (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3013 //===----------------------------------------------------------------------===//
3014 // Multiclass for the load instructions with absolute addressing mode.
3015 //===----------------------------------------------------------------------===//
3016 multiclass LD_Abs_Predbase<string mnemonic, RegisterClass RC, bit isNot,
3018 let isPredicatedNew = isPredNew in
3019 def NAME : LDInst2<(outs RC:$dst),
3020 (ins PredRegs:$src1, u0AlwaysExt:$absaddr),
3021 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
3022 ") ")#"$dst = "#mnemonic#"(##$absaddr)",
3027 multiclass LD_Abs_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
3028 let isPredicatedFalse = PredNot in {
3029 defm _c#NAME : LD_Abs_Predbase<mnemonic, RC, PredNot, 0>;
3031 defm _cdn#NAME : LD_Abs_Predbase<mnemonic, RC, PredNot, 1>;
3035 let isExtended = 1, hasSideEffects = 0 in
3036 multiclass LD_Abs<string mnemonic, string CextOp, RegisterClass RC> {
3037 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
3038 let opExtendable = 1, isPredicable = 1 in
3039 def NAME#_V4 : LDInst2<(outs RC:$dst),
3040 (ins u0AlwaysExt:$absaddr),
3041 "$dst = "#mnemonic#"(##$absaddr)",
3045 let opExtendable = 2, isPredicated = 1 in {
3046 defm Pt_V4 : LD_Abs_Pred<mnemonic, RC, 0>;
3047 defm NotPt_V4 : LD_Abs_Pred<mnemonic, RC, 1>;
3052 let addrMode = Absolute in {
3053 let accessSize = ByteAccess in {
3054 defm LDrib_abs : LD_Abs<"memb", "LDrib", IntRegs>, AddrModeRel;
3055 defm LDriub_abs : LD_Abs<"memub", "LDriub", IntRegs>, AddrModeRel;
3057 let accessSize = HalfWordAccess in {
3058 defm LDrih_abs : LD_Abs<"memh", "LDrih", IntRegs>, AddrModeRel;
3059 defm LDriuh_abs : LD_Abs<"memuh", "LDriuh", IntRegs>, AddrModeRel;
3061 let accessSize = WordAccess in
3062 defm LDriw_abs : LD_Abs<"memw", "LDriw", IntRegs>, AddrModeRel;
3064 let accessSize = DoubleWordAccess in
3065 defm LDrid_abs : LD_Abs<"memd", "LDrid", DoubleRegs>, AddrModeRel;
3068 let Predicates = [HasV4T], AddedComplexity = 30 in {
3069 def : Pat<(i32 (load (HexagonCONST32 tglobaladdr:$absaddr))),
3070 (LDriw_abs_V4 tglobaladdr: $absaddr)>;
3072 def : Pat<(i32 (sextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))),
3073 (LDrib_abs_V4 tglobaladdr:$absaddr)>;
3075 def : Pat<(i32 (zextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))),
3076 (LDriub_abs_V4 tglobaladdr:$absaddr)>;
3078 def : Pat<(i32 (sextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
3079 (LDrih_abs_V4 tglobaladdr:$absaddr)>;
3081 def : Pat<(i32 (zextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
3082 (LDriuh_abs_V4 tglobaladdr:$absaddr)>;
3085 //===----------------------------------------------------------------------===//
3086 // multiclass for load instructions with GP-relative addressing mode.
3087 // Rx=mem[bhwd](##global)
3088 // if ([!]Pv[.new]) Rx=mem[bhwd](##global)
3089 //===----------------------------------------------------------------------===//
3090 let hasSideEffects = 0, validSubTargets = HasV4SubT in
3091 multiclass LD_GP<string mnemonic, string BaseOp, RegisterClass RC> {
3092 let BaseOpcode = BaseOp in {
3093 let isPredicable = 1 in
3094 def NAME#_V4 : LDInst2<(outs RC:$dst),
3095 (ins globaladdress:$global),
3096 "$dst = "#mnemonic#"(#$global)",
3099 let isExtended = 1, opExtendable = 2, isPredicated = 1 in {
3100 defm Pt_V4 : LD_Abs_Pred<mnemonic, RC, 0>;
3101 defm NotPt_V4 : LD_Abs_Pred<mnemonic, RC, 1>;
3106 defm LDd_GP : LD_GP<"memd", "LDd_GP", DoubleRegs>, PredNewRel;
3107 defm LDb_GP : LD_GP<"memb", "LDb_GP", IntRegs>, PredNewRel;
3108 defm LDub_GP : LD_GP<"memub", "LDub_GP", IntRegs>, PredNewRel;
3109 defm LDh_GP : LD_GP<"memh", "LDh_GP", IntRegs>, PredNewRel;
3110 defm LDuh_GP : LD_GP<"memuh", "LDuh_GP", IntRegs>, PredNewRel;
3111 defm LDw_GP : LD_GP<"memw", "LDw_GP", IntRegs>, PredNewRel;
3113 def : Pat <(atomic_load_64 (HexagonCONST32_GP tglobaladdr:$global)),
3114 (i64 (LDd_GP_V4 tglobaladdr:$global))>;
3116 def : Pat <(atomic_load_32 (HexagonCONST32_GP tglobaladdr:$global)),
3117 (i32 (LDw_GP_V4 tglobaladdr:$global))>;
3119 def : Pat <(atomic_load_16 (HexagonCONST32_GP tglobaladdr:$global)),
3120 (i32 (LDuh_GP_V4 tglobaladdr:$global))>;
3122 def : Pat <(atomic_load_8 (HexagonCONST32_GP tglobaladdr:$global)),
3123 (i32 (LDub_GP_V4 tglobaladdr:$global))>;
3125 // Map from load(globaladdress) -> memw(#foo + 0)
3126 let AddedComplexity = 100 in
3127 def : Pat <(i64 (load (HexagonCONST32_GP tglobaladdr:$global))),
3128 (i64 (LDd_GP_V4 tglobaladdr:$global))>;
3130 // Map from Pd = load(globaladdress) -> Rd = memb(globaladdress), Pd = Rd
3131 let AddedComplexity = 100 in
3132 def : Pat <(i1 (load (HexagonCONST32_GP tglobaladdr:$global))),
3133 (i1 (C2_tfrrp (i32 (LDb_GP_V4 tglobaladdr:$global))))>;
3135 // When the Interprocedural Global Variable optimizer realizes that a certain
3136 // global variable takes only two constant values, it shrinks the global to
3137 // a boolean. Catch those loads here in the following 3 patterns.
3138 let AddedComplexity = 100 in
3139 def : Pat <(i32 (extloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3140 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3142 let AddedComplexity = 100 in
3143 def : Pat <(i32 (sextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3144 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3146 // Map from load(globaladdress) -> memb(#foo)
3147 let AddedComplexity = 100 in
3148 def : Pat <(i32 (extloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3149 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3151 // Map from load(globaladdress) -> memb(#foo)
3152 let AddedComplexity = 100 in
3153 def : Pat <(i32 (sextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3154 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3156 let AddedComplexity = 100 in
3157 def : Pat <(i32 (zextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3158 (i32 (LDub_GP_V4 tglobaladdr:$global))>;
3160 // Map from load(globaladdress) -> memub(#foo)
3161 let AddedComplexity = 100 in
3162 def : Pat <(i32 (zextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3163 (i32 (LDub_GP_V4 tglobaladdr:$global))>;
3165 // Map from load(globaladdress) -> memh(#foo)
3166 let AddedComplexity = 100 in
3167 def : Pat <(i32 (extloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3168 (i32 (LDh_GP_V4 tglobaladdr:$global))>;
3170 // Map from load(globaladdress) -> memh(#foo)
3171 let AddedComplexity = 100 in
3172 def : Pat <(i32 (sextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3173 (i32 (LDh_GP_V4 tglobaladdr:$global))>;
3175 // Map from load(globaladdress) -> memuh(#foo)
3176 let AddedComplexity = 100 in
3177 def : Pat <(i32 (zextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3178 (i32 (LDuh_GP_V4 tglobaladdr:$global))>;
3180 // Map from load(globaladdress) -> memw(#foo)
3181 let AddedComplexity = 100 in
3182 def : Pat <(i32 (load (HexagonCONST32_GP tglobaladdr:$global))),
3183 (i32 (LDw_GP_V4 tglobaladdr:$global))>;
3186 // Transfer global address into a register
3187 let isExtended = 1, opExtendable = 1, AddedComplexity=50, isMoveImm = 1,
3188 isAsCheapAsAMove = 1, isReMaterializable = 1, validSubTargets = HasV4SubT in
3189 def TFRI_V4 : ALU32_ri<(outs IntRegs:$dst), (ins s16Ext:$src1),
3191 [(set IntRegs:$dst, (HexagonCONST32 tglobaladdr:$src1))]>,
3194 // Transfer a block address into a register
3195 def : Pat<(HexagonCONST32_GP tblockaddress:$src1),
3196 (TFRI_V4 tblockaddress:$src1)>,
3199 let isExtended = 1, opExtendable = 2, AddedComplexity=50,
3200 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3201 def TFRI_cPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3202 (ins PredRegs:$src1, s16Ext:$src2),
3203 "if($src1) $dst = #$src2",
3207 let isExtended = 1, opExtendable = 2, AddedComplexity=50, isPredicatedFalse = 1,
3208 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3209 def TFRI_cNotPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3210 (ins PredRegs:$src1, s16Ext:$src2),
3211 "if(!$src1) $dst = #$src2",
3215 let isExtended = 1, opExtendable = 2, AddedComplexity=50,
3216 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3217 def TFRI_cdnPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3218 (ins PredRegs:$src1, s16Ext:$src2),
3219 "if($src1.new) $dst = #$src2",
3223 let isExtended = 1, opExtendable = 2, AddedComplexity=50, isPredicatedFalse = 1,
3224 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3225 def TFRI_cdnNotPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3226 (ins PredRegs:$src1, s16Ext:$src2),
3227 "if(!$src1.new) $dst = #$src2",
3231 let AddedComplexity = 50, Predicates = [HasV4T] in
3232 def : Pat<(HexagonCONST32_GP tglobaladdr:$src1),
3233 (TFRI_V4 tglobaladdr:$src1)>,
3237 // Load - Indirect with long offset: These instructions take global address
3239 let isExtended = 1, opExtendable = 3, AddedComplexity = 40,
3240 validSubTargets = HasV4SubT in
3241 def LDrid_ind_lo_V4 : LDInst<(outs DoubleRegs:$dst),
3242 (ins IntRegs:$src1, u2Imm:$src2, globaladdressExt:$offset),
3243 "$dst=memd($src1<<#$src2+##$offset)",
3244 [(set (i64 DoubleRegs:$dst),
3245 (load (add (shl IntRegs:$src1, u2ImmPred:$src2),
3246 (HexagonCONST32 tglobaladdr:$offset))))]>,
3249 let AddedComplexity = 40 in
3250 multiclass LD_indirect_lo<string OpcStr, PatFrag OpNode> {
3251 let isExtended = 1, opExtendable = 3, validSubTargets = HasV4SubT in
3252 def _lo_V4 : LDInst<(outs IntRegs:$dst),
3253 (ins IntRegs:$src1, u2Imm:$src2, globaladdressExt:$offset),
3254 !strconcat("$dst = ",
3255 !strconcat(OpcStr, "($src1<<#$src2+##$offset)")),
3257 (i32 (OpNode (add (shl IntRegs:$src1, u2ImmPred:$src2),
3258 (HexagonCONST32 tglobaladdr:$offset)))))]>,
3262 defm LDrib_ind : LD_indirect_lo<"memb", sextloadi8>;
3263 defm LDriub_ind : LD_indirect_lo<"memub", zextloadi8>;
3264 defm LDriub_ind_anyext : LD_indirect_lo<"memub", extloadi8>;
3265 defm LDrih_ind : LD_indirect_lo<"memh", sextloadi16>;
3266 defm LDriuh_ind : LD_indirect_lo<"memuh", zextloadi16>;
3267 defm LDriuh_ind_anyext : LD_indirect_lo<"memuh", extloadi16>;
3268 defm LDriw_ind : LD_indirect_lo<"memw", load>;
3270 let AddedComplexity = 40 in
3271 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1,
3272 (NumUsesBelowThresCONST32 tglobaladdr:$offset)))),
3273 (i32 (LDrib_ind_lo_V4 IntRegs:$src1, 0, tglobaladdr:$offset))>,
3276 let AddedComplexity = 40 in
3277 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1,
3278 (NumUsesBelowThresCONST32 tglobaladdr:$offset)))),
3279 (i32 (LDriub_ind_lo_V4 IntRegs:$src1, 0, tglobaladdr:$offset))>,
3282 let Predicates = [HasV4T], AddedComplexity = 30 in {
3283 def : Pat<(truncstorei8 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3284 (STrib_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3286 def : Pat<(truncstorei16 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3287 (STrih_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3289 def : Pat<(store (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3290 (STriw_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3293 let Predicates = [HasV4T], AddedComplexity = 30 in {
3294 def : Pat<(i32 (load u0AlwaysExtPred:$src)),
3295 (LDriw_abs_V4 u0AlwaysExtPred:$src)>;
3297 def : Pat<(i32 (sextloadi8 u0AlwaysExtPred:$src)),
3298 (LDrib_abs_V4 u0AlwaysExtPred:$src)>;
3300 def : Pat<(i32 (zextloadi8 u0AlwaysExtPred:$src)),
3301 (LDriub_abs_V4 u0AlwaysExtPred:$src)>;
3303 def : Pat<(i32 (sextloadi16 u0AlwaysExtPred:$src)),
3304 (LDrih_abs_V4 u0AlwaysExtPred:$src)>;
3306 def : Pat<(i32 (zextloadi16 u0AlwaysExtPred:$src)),
3307 (LDriuh_abs_V4 u0AlwaysExtPred:$src)>;
3310 // Indexed store word - global address.
3311 // memw(Rs+#u6:2)=#S8
3312 let AddedComplexity = 10 in
3313 def STriw_offset_ext_V4 : STInst<(outs),
3314 (ins IntRegs:$src1, u6_2Imm:$src2, globaladdress:$src3),
3315 "memw($src1+#$src2) = ##$src3",
3316 [(store (HexagonCONST32 tglobaladdr:$src3),
3317 (add IntRegs:$src1, u6_2ImmPred:$src2))]>,
3320 def : Pat<(i64 (ctlz (i64 DoubleRegs:$src1))),
3321 (i64 (A4_combineir (i32 0), (i32 (CTLZ64_rr DoubleRegs:$src1))))>,
3324 def : Pat<(i64 (cttz (i64 DoubleRegs:$src1))),
3325 (i64 (A4_combineir (i32 0), (i32 (CTTZ64_rr DoubleRegs:$src1))))>,
3330 // We need a complexity of 120 here to override preceding handling of
3332 let Predicates = [HasV4T], AddedComplexity = 120 in {
3333 def: Pat <(i64 (extloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3334 (i64 (A4_combineir 0, (LDrib_abs_V4 tglobaladdr:$addr)))>;
3336 def: Pat <(i64 (zextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3337 (i64 (A4_combineir 0, (LDriub_abs_V4 tglobaladdr:$addr)))>;
3339 def: Pat <(i64 (sextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3340 (i64 (A2_sxtw (LDrib_abs_V4 tglobaladdr:$addr)))>;
3342 def: Pat <(i64 (extloadi8 FoldGlobalAddr:$addr)),
3343 (i64 (A4_combineir 0, (LDrib_abs_V4 FoldGlobalAddr:$addr)))>;
3345 def: Pat <(i64 (zextloadi8 FoldGlobalAddr:$addr)),
3346 (i64 (A4_combineir 0, (LDriub_abs_V4 FoldGlobalAddr:$addr)))>;
3348 def: Pat <(i64 (sextloadi8 FoldGlobalAddr:$addr)),
3349 (i64 (A2_sxtw (LDrib_abs_V4 FoldGlobalAddr:$addr)))>;
3352 // We need a complexity of 120 here to override preceding handling of
3354 let AddedComplexity = 120 in {
3355 def: Pat <(i64 (extloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3356 (i64 (A4_combineir 0, (LDrih_abs_V4 tglobaladdr:$addr)))>,
3359 def: Pat <(i64 (zextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3360 (i64 (A4_combineir 0, (LDriuh_abs_V4 tglobaladdr:$addr)))>,
3363 def: Pat <(i64 (sextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3364 (i64 (A2_sxtw (LDrih_abs_V4 tglobaladdr:$addr)))>,
3367 def: Pat <(i64 (extloadi16 FoldGlobalAddr:$addr)),
3368 (i64 (A4_combineir 0, (LDrih_abs_V4 FoldGlobalAddr:$addr)))>,
3371 def: Pat <(i64 (zextloadi16 FoldGlobalAddr:$addr)),
3372 (i64 (A4_combineir 0, (LDriuh_abs_V4 FoldGlobalAddr:$addr)))>,
3375 def: Pat <(i64 (sextloadi16 FoldGlobalAddr:$addr)),
3376 (i64 (A2_sxtw (LDrih_abs_V4 FoldGlobalAddr:$addr)))>,
3380 // We need a complexity of 120 here to override preceding handling of
3382 let AddedComplexity = 120 in {
3383 def: Pat <(i64 (extloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3384 (i64 (A4_combineir 0, (LDriw_abs_V4 tglobaladdr:$addr)))>,
3387 def: Pat <(i64 (zextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3388 (i64 (A4_combineir 0, (LDriw_abs_V4 tglobaladdr:$addr)))>,
3391 def: Pat <(i64 (sextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3392 (i64 (A2_sxtw (LDriw_abs_V4 tglobaladdr:$addr)))>,
3395 def: Pat <(i64 (extloadi32 FoldGlobalAddr:$addr)),
3396 (i64 (A4_combineir 0, (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
3399 def: Pat <(i64 (zextloadi32 FoldGlobalAddr:$addr)),
3400 (i64 (A4_combineir 0, (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
3403 def: Pat <(i64 (sextloadi32 FoldGlobalAddr:$addr)),
3404 (i64 (A2_sxtw (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
3408 // Indexed store double word - global address.
3409 // memw(Rs+#u6:2)=#S8
3410 let AddedComplexity = 10 in
3411 def STrih_offset_ext_V4 : STInst<(outs),
3412 (ins IntRegs:$src1, u6_1Imm:$src2, globaladdress:$src3),
3413 "memh($src1+#$src2) = ##$src3",
3414 [(truncstorei16 (HexagonCONST32 tglobaladdr:$src3),
3415 (add IntRegs:$src1, u6_1ImmPred:$src2))]>,
3417 // Map from store(globaladdress + x) -> memd(#foo + x)
3418 let AddedComplexity = 100 in
3419 def : Pat<(store (i64 DoubleRegs:$src1),
3420 FoldGlobalAddrGP:$addr),
3421 (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
3424 def : Pat<(atomic_store_64 FoldGlobalAddrGP:$addr,
3425 (i64 DoubleRegs:$src1)),
3426 (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
3429 // Map from store(globaladdress + x) -> memb(#foo + x)
3430 let AddedComplexity = 100 in
3431 def : Pat<(truncstorei8 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
3432 (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3435 def : Pat<(atomic_store_8 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
3436 (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3439 // Map from store(globaladdress + x) -> memh(#foo + x)
3440 let AddedComplexity = 100 in
3441 def : Pat<(truncstorei16 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
3442 (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3445 def : Pat<(atomic_store_16 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
3446 (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3449 // Map from store(globaladdress + x) -> memw(#foo + x)
3450 let AddedComplexity = 100 in
3451 def : Pat<(store (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
3452 (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3455 def : Pat<(atomic_store_32 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
3456 (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3459 // Map from load(globaladdress + x) -> memd(#foo + x)
3460 let AddedComplexity = 100 in
3461 def : Pat<(i64 (load FoldGlobalAddrGP:$addr)),
3462 (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>,
3465 def : Pat<(atomic_load_64 FoldGlobalAddrGP:$addr),
3466 (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>,
3469 // Map from load(globaladdress + x) -> memb(#foo + x)
3470 let AddedComplexity = 100 in
3471 def : Pat<(i32 (extloadi8 FoldGlobalAddrGP:$addr)),
3472 (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>,
3475 // Map from load(globaladdress + x) -> memb(#foo + x)
3476 let AddedComplexity = 100 in
3477 def : Pat<(i32 (sextloadi8 FoldGlobalAddrGP:$addr)),
3478 (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>,
3481 //let AddedComplexity = 100 in
3482 let AddedComplexity = 100 in
3483 def : Pat<(i32 (extloadi16 FoldGlobalAddrGP:$addr)),
3484 (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>,
3487 // Map from load(globaladdress + x) -> memh(#foo + x)
3488 let AddedComplexity = 100 in
3489 def : Pat<(i32 (sextloadi16 FoldGlobalAddrGP:$addr)),
3490 (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>,
3493 // Map from load(globaladdress + x) -> memuh(#foo + x)
3494 let AddedComplexity = 100 in
3495 def : Pat<(i32 (zextloadi16 FoldGlobalAddrGP:$addr)),
3496 (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>,
3499 def : Pat<(atomic_load_16 FoldGlobalAddrGP:$addr),
3500 (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>,
3503 // Map from load(globaladdress + x) -> memub(#foo + x)
3504 let AddedComplexity = 100 in
3505 def : Pat<(i32 (zextloadi8 FoldGlobalAddrGP:$addr)),
3506 (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>,
3509 def : Pat<(atomic_load_8 FoldGlobalAddrGP:$addr),
3510 (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>,
3513 // Map from load(globaladdress + x) -> memw(#foo + x)
3514 let AddedComplexity = 100 in
3515 def : Pat<(i32 (load FoldGlobalAddrGP:$addr)),
3516 (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>,
3519 def : Pat<(atomic_load_32 FoldGlobalAddrGP:$addr),
3520 (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>,