1 //=- HexagonInstrInfoV4.td - Target Desc. for Hexagon Target -*- tablegen -*-=//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the Hexagon V4 instructions in TableGen format.
12 //===----------------------------------------------------------------------===//
14 let hasSideEffects = 0 in
15 class T_Immext<Operand ImmType>
16 : EXTENDERInst<(outs), (ins ImmType:$imm),
17 "immext(#$imm)", []> {
21 let Inst{27-16} = imm{31-20};
22 let Inst{13-0} = imm{19-6};
25 def A4_ext : T_Immext<u26_6Imm>;
26 let isCodeGenOnly = 1 in {
28 def A4_ext_b : T_Immext<brtarget>;
30 def A4_ext_c : T_Immext<calltarget>;
31 def A4_ext_g : T_Immext<globaladdress>;
34 // Fold (add (CONST32 tglobaladdr:$addr) <offset>) into a global address.
35 def FoldGlobalAddr : ComplexPattern<i32, 1, "foldGlobalAddress", [], []>;
37 // Fold (add (CONST32_GP tglobaladdr:$addr) <offset>) into a global address.
38 def FoldGlobalAddrGP : ComplexPattern<i32, 1, "foldGlobalAddressGP", [], []>;
40 def NumUsesBelowThresCONST32 : PatFrag<(ops node:$addr),
41 (HexagonCONST32 node:$addr), [{
42 return hasNumUsesBelowThresGA(N->getOperand(0).getNode());
45 // Hexagon V4 Architecture spec defines 8 instruction classes:
46 // LD ST ALU32 XTYPE J JR MEMOP NV CR SYSTEM(system is not implemented in the
50 // ========================================
51 // Loads (8/16/32/64 bit)
55 // ========================================
56 // Stores (8/16/32/64 bit)
59 // ALU32 Instructions:
60 // ========================================
61 // Arithmetic / Logical (32 bit)
64 // XTYPE Instructions (32/64 bit):
65 // ========================================
66 // Arithmetic, Logical, Bit Manipulation
67 // Multiply (Integer, Fractional, Complex)
68 // Permute / Vector Permute Operations
69 // Predicate Operations
70 // Shift / Shift with Add/Sub/Logical
72 // Vector Halfword (ALU, Shift, Multiply)
73 // Vector Word (ALU, Shift)
76 // ========================================
77 // Jump/Call PC-relative
80 // ========================================
83 // MEMOP Instructions:
84 // ========================================
85 // Operation on memory (8/16/32 bit)
88 // ========================================
93 // ========================================
94 // Control-Register Transfers
95 // Hardware Loop Setup
96 // Predicate Logicals & Reductions
98 // SYSTEM Instructions (not implemented in the compiler):
99 // ========================================
105 //===----------------------------------------------------------------------===//
107 //===----------------------------------------------------------------------===//
109 class T_ALU32_3op_not<string mnemonic, bits<3> MajOp, bits<3> MinOp,
111 : T_ALU32_3op<mnemonic, MajOp, MinOp, OpsRev, 0> {
112 let AsmString = "$Rd = "#mnemonic#"($Rs, ~$Rt)";
115 let BaseOpcode = "andn_rr", CextOpcode = "andn", isCodeGenOnly = 0 in
116 def A4_andn : T_ALU32_3op_not<"and", 0b001, 0b100, 1>;
117 let BaseOpcode = "orn_rr", CextOpcode = "orn", isCodeGenOnly = 0 in
118 def A4_orn : T_ALU32_3op_not<"or", 0b001, 0b101, 1>;
120 let CextOpcode = "rcmp.eq", isCodeGenOnly = 0 in
121 def A4_rcmpeq : T_ALU32_3op<"cmp.eq", 0b011, 0b010, 0, 1>;
122 let CextOpcode = "!rcmp.eq", isCodeGenOnly = 0 in
123 def A4_rcmpneq : T_ALU32_3op<"!cmp.eq", 0b011, 0b011, 0, 1>;
125 let isCodeGenOnly = 0 in {
126 def C4_cmpneq : T_ALU32_3op_cmp<"!cmp.eq", 0b00, 1, 1>;
127 def C4_cmplte : T_ALU32_3op_cmp<"!cmp.gt", 0b10, 1, 0>;
128 def C4_cmplteu : T_ALU32_3op_cmp<"!cmp.gtu", 0b11, 1, 0>;
131 // Pats for instruction selection.
133 // A class to embed the usual comparison patfrags within a zext to i32.
134 // The seteq/setne frags use "lhs" and "rhs" as operands, so use the same
135 // names, or else the frag's "body" won't match the operands.
136 class CmpInReg<PatFrag Op>
137 : PatFrag<(ops node:$lhs, node:$rhs),(i32 (zext (i1 Op.Fragment)))>;
139 def: T_cmp32_rr_pat<A4_rcmpeq, CmpInReg<seteq>, i32>;
140 def: T_cmp32_rr_pat<A4_rcmpneq, CmpInReg<setne>, i32>;
142 class T_CMP_rrbh<string mnemonic, bits<3> MinOp, bit IsComm>
143 : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt),
144 "$Pd = "#mnemonic#"($Rs, $Rt)", [], "", S_3op_tc_2early_SLOT23>,
146 let validSubTargets = HasV4SubT;
147 let InputType = "reg";
148 let CextOpcode = mnemonic;
150 let isCommutable = IsComm;
151 let hasSideEffects = 0;
158 let Inst{27-21} = 0b0111110;
159 let Inst{20-16} = Rs;
161 let Inst{7-5} = MinOp;
165 let isCodeGenOnly = 0 in {
166 def A4_cmpbeq : T_CMP_rrbh<"cmpb.eq", 0b110, 1>;
167 def A4_cmpbgt : T_CMP_rrbh<"cmpb.gt", 0b010, 0>;
168 def A4_cmpbgtu : T_CMP_rrbh<"cmpb.gtu", 0b111, 0>;
169 def A4_cmpheq : T_CMP_rrbh<"cmph.eq", 0b011, 1>;
170 def A4_cmphgt : T_CMP_rrbh<"cmph.gt", 0b100, 0>;
171 def A4_cmphgtu : T_CMP_rrbh<"cmph.gtu", 0b101, 0>;
174 class T_CMP_ribh<string mnemonic, bits<2> MajOp, bit IsHalf, bit IsComm,
175 Operand ImmType, bit IsImmExt, bit IsImmSigned, int ImmBits>
176 : ALU64Inst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, ImmType:$Imm),
177 "$Pd = "#mnemonic#"($Rs, #$Imm)", [], "", ALU64_tc_2early_SLOT23>,
179 let validSubTargets = HasV4SubT;
180 let InputType = "imm";
181 let CextOpcode = mnemonic;
183 let isCommutable = IsComm;
184 let hasSideEffects = 0;
185 let isExtendable = IsImmExt;
186 let opExtendable = !if (IsImmExt, 2, 0);
187 let isExtentSigned = IsImmSigned;
188 let opExtentBits = ImmBits;
195 let Inst{27-24} = 0b1101;
196 let Inst{22-21} = MajOp;
197 let Inst{20-16} = Rs;
198 let Inst{12-5} = Imm;
200 let Inst{3} = IsHalf;
204 let isCodeGenOnly = 0 in {
205 def A4_cmpbeqi : T_CMP_ribh<"cmpb.eq", 0b00, 0, 1, u8Imm, 0, 0, 8>;
206 def A4_cmpbgti : T_CMP_ribh<"cmpb.gt", 0b01, 0, 0, s8Imm, 0, 1, 8>;
207 def A4_cmpbgtui : T_CMP_ribh<"cmpb.gtu", 0b10, 0, 0, u7Ext, 1, 0, 7>;
208 def A4_cmpheqi : T_CMP_ribh<"cmph.eq", 0b00, 1, 1, s8Ext, 1, 1, 8>;
209 def A4_cmphgti : T_CMP_ribh<"cmph.gt", 0b01, 1, 0, s8Ext, 1, 1, 8>;
210 def A4_cmphgtui : T_CMP_ribh<"cmph.gtu", 0b10, 1, 0, u7Ext, 1, 0, 7>;
212 class T_RCMP_EQ_ri<string mnemonic, bit IsNeg>
213 : ALU32_ri<(outs IntRegs:$Rd), (ins IntRegs:$Rs, s8Ext:$s8),
214 "$Rd = "#mnemonic#"($Rs, #$s8)", [], "", ALU32_2op_tc_1_SLOT0123>,
216 let validSubTargets = HasV4SubT;
217 let InputType = "imm";
218 let CextOpcode = !if (IsNeg, "!rcmp.eq", "rcmp.eq");
219 let isExtendable = 1;
220 let opExtendable = 2;
221 let isExtentSigned = 1;
222 let opExtentBits = 8;
230 let Inst{27-24} = 0b0011;
232 let Inst{21} = IsNeg;
233 let Inst{20-16} = Rs;
239 let isCodeGenOnly = 0 in {
240 def A4_rcmpeqi : T_RCMP_EQ_ri<"cmp.eq", 0>;
241 def A4_rcmpneqi : T_RCMP_EQ_ri<"!cmp.eq", 1>;
244 def: Pat<(i32 (zext (i1 (seteq (i32 IntRegs:$Rs), s8ExtPred:$s8)))),
245 (A4_rcmpeqi IntRegs:$Rs, s8ExtPred:$s8)>;
246 def: Pat<(i32 (zext (i1 (setne (i32 IntRegs:$Rs), s8ExtPred:$s8)))),
247 (A4_rcmpneqi IntRegs:$Rs, s8ExtPred:$s8)>;
249 // Preserve the S2_tstbit_r generation
250 def: Pat<(i32 (zext (i1 (setne (i32 (and (i32 (shl 1, (i32 IntRegs:$src2))),
251 (i32 IntRegs:$src1))), 0)))),
252 (C2_muxii (S2_tstbit_r IntRegs:$src1, IntRegs:$src2), 1, 0)>;
255 //===----------------------------------------------------------------------===//
257 //===----------------------------------------------------------------------===//
260 //===----------------------------------------------------------------------===//
262 //===----------------------------------------------------------------------===//
264 // Combine a word and an immediate into a register pair.
265 let hasSideEffects = 0, isExtentSigned = 1, isExtendable = 1,
267 class T_Combine1 <bits<2> MajOp, dag ins, string AsmStr>
268 : ALU32Inst <(outs DoubleRegs:$Rdd), ins, AsmStr> {
274 let Inst{27-24} = 0b0011;
275 let Inst{22-21} = MajOp;
276 let Inst{20-16} = Rs;
282 let opExtendable = 2, isCodeGenOnly = 0 in
283 def A4_combineri : T_Combine1<0b00, (ins IntRegs:$Rs, s8Ext:$s8),
284 "$Rdd = combine($Rs, #$s8)">;
286 let opExtendable = 1, isCodeGenOnly = 0 in
287 def A4_combineir : T_Combine1<0b01, (ins s8Ext:$s8, IntRegs:$Rs),
288 "$Rdd = combine(#$s8, $Rs)">;
290 def HexagonWrapperCombineRI_V4 :
291 SDNode<"HexagonISD::WrapperCombineRI_V4", SDTHexagonI64I32I32>;
292 def HexagonWrapperCombineIR_V4 :
293 SDNode<"HexagonISD::WrapperCombineIR_V4", SDTHexagonI64I32I32>;
295 def : Pat <(HexagonWrapperCombineRI_V4 IntRegs:$r, s8ExtPred:$i),
296 (A4_combineri IntRegs:$r, s8ExtPred:$i)>,
299 def : Pat <(HexagonWrapperCombineIR_V4 s8ExtPred:$i, IntRegs:$r),
300 (A4_combineir s8ExtPred:$i, IntRegs:$r)>,
303 // A4_combineii: Set two small immediates.
304 let hasSideEffects = 0, isExtendable = 1, opExtentBits = 6, opExtendable = 2 in
305 def A4_combineii: ALU32Inst<(outs DoubleRegs:$Rdd), (ins s8Imm:$s8, u6Ext:$U6),
306 "$Rdd = combine(#$s8, #$U6)"> {
312 let Inst{27-23} = 0b11001;
313 let Inst{20-16} = U6{5-1};
314 let Inst{13} = U6{0};
319 //===----------------------------------------------------------------------===//
321 //===----------------------------------------------------------------------===//
323 //===----------------------------------------------------------------------===//
325 //===----------------------------------------------------------------------===//
326 //===----------------------------------------------------------------------===//
327 // Template class for load instructions with Absolute set addressing mode.
328 //===----------------------------------------------------------------------===//
329 let isExtended = 1, opExtendable = 2, hasSideEffects = 0,
330 validSubTargets = HasV4SubT, addrMode = AbsoluteSet in
331 class T_LD_abs_set<string mnemonic, RegisterClass RC>:
332 LDInst2<(outs RC:$dst1, IntRegs:$dst2),
333 (ins u0AlwaysExt:$addr),
334 "$dst1 = "#mnemonic#"($dst2=##$addr)",
338 def LDrid_abs_set_V4 : T_LD_abs_set <"memd", DoubleRegs>;
339 def LDrib_abs_set_V4 : T_LD_abs_set <"memb", IntRegs>;
340 def LDriub_abs_set_V4 : T_LD_abs_set <"memub", IntRegs>;
341 def LDrih_abs_set_V4 : T_LD_abs_set <"memh", IntRegs>;
342 def LDriw_abs_set_V4 : T_LD_abs_set <"memw", IntRegs>;
343 def LDriuh_abs_set_V4 : T_LD_abs_set <"memuh", IntRegs>;
345 //===----------------------------------------------------------------------===//
346 // Template classes for the non-predicated load instructions with
347 // base + register offset addressing mode
348 //===----------------------------------------------------------------------===//
349 class T_load_rr <string mnemonic, RegisterClass RC, bits<3> MajOp>:
350 LDInst<(outs RC:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$u2),
351 "$dst = "#mnemonic#"($src1 + $src2<<#$u2)",
352 [], "", V4LDST_tc_ld_SLOT01>, ImmRegShl, AddrModeRel {
360 let Inst{27-24} = 0b1010;
361 let Inst{23-21} = MajOp;
362 let Inst{20-16} = src1;
363 let Inst{12-8} = src2;
364 let Inst{13} = u2{1};
369 //===----------------------------------------------------------------------===//
370 // Template classes for the predicated load instructions with
371 // base + register offset addressing mode
372 //===----------------------------------------------------------------------===//
373 let isPredicated = 1 in
374 class T_pload_rr <string mnemonic, RegisterClass RC, bits<3> MajOp,
375 bit isNot, bit isPredNew>:
376 LDInst <(outs RC:$dst),
377 (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$u2),
378 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
379 ") ")#"$dst = "#mnemonic#"($src2+$src3<<#$u2)",
380 [], "", V4LDST_tc_ld_SLOT01>, AddrModeRel {
387 let isPredicatedFalse = isNot;
388 let isPredicatedNew = isPredNew;
392 let Inst{27-26} = 0b00;
393 let Inst{25} = isPredNew;
394 let Inst{24} = isNot;
395 let Inst{23-21} = MajOp;
396 let Inst{20-16} = src2;
397 let Inst{12-8} = src3;
398 let Inst{13} = u2{1};
400 let Inst{6-5} = src1;
404 //===----------------------------------------------------------------------===//
405 // multiclass for load instructions with base + register offset
407 //===----------------------------------------------------------------------===//
408 let hasSideEffects = 0, addrMode = BaseRegOffset in
409 multiclass ld_idxd_shl <string mnemonic, string CextOp, RegisterClass RC,
411 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl,
412 InputType = "reg" in {
413 let isPredicable = 1 in
414 def L4_#NAME#_rr : T_load_rr <mnemonic, RC, MajOp>;
417 def L4_p#NAME#t_rr : T_pload_rr <mnemonic, RC, MajOp, 0, 0>;
418 def L4_p#NAME#f_rr : T_pload_rr <mnemonic, RC, MajOp, 1, 0>;
421 def L4_p#NAME#tnew_rr : T_pload_rr <mnemonic, RC, MajOp, 0, 1>;
422 def L4_p#NAME#fnew_rr : T_pload_rr <mnemonic, RC, MajOp, 1, 1>;
426 let hasNewValue = 1, accessSize = ByteAccess, isCodeGenOnly = 0 in {
427 defm loadrb : ld_idxd_shl<"memb", "LDrib", IntRegs, 0b000>;
428 defm loadrub : ld_idxd_shl<"memub", "LDriub", IntRegs, 0b001>;
431 let hasNewValue = 1, accessSize = HalfWordAccess, isCodeGenOnly = 0 in {
432 defm loadrh : ld_idxd_shl<"memh", "LDrih", IntRegs, 0b010>;
433 defm loadruh : ld_idxd_shl<"memuh", "LDriuh", IntRegs, 0b011>;
436 let hasNewValue = 1, accessSize = WordAccess, isCodeGenOnly = 0 in
437 defm loadri : ld_idxd_shl<"memw", "LDriw", IntRegs, 0b100>;
439 let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
440 defm loadrd : ld_idxd_shl<"memd", "LDrid", DoubleRegs, 0b110>;
442 // 'def pats' for load instructions with base + register offset and non-zero
443 // immediate value. Immediate value is used to left-shift the second
445 let AddedComplexity = 40 in {
446 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1,
447 (shl IntRegs:$src2, u2ImmPred:$offset)))),
448 (L4_loadrb_rr IntRegs:$src1,
449 IntRegs:$src2, u2ImmPred:$offset)>,
452 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1,
453 (shl IntRegs:$src2, u2ImmPred:$offset)))),
454 (L4_loadrub_rr IntRegs:$src1,
455 IntRegs:$src2, u2ImmPred:$offset)>,
458 def : Pat <(i32 (extloadi8 (add IntRegs:$src1,
459 (shl IntRegs:$src2, u2ImmPred:$offset)))),
460 (L4_loadrub_rr IntRegs:$src1,
461 IntRegs:$src2, u2ImmPred:$offset)>,
464 def : Pat <(i32 (sextloadi16 (add IntRegs:$src1,
465 (shl IntRegs:$src2, u2ImmPred:$offset)))),
466 (L4_loadrh_rr IntRegs:$src1,
467 IntRegs:$src2, u2ImmPred:$offset)>,
470 def : Pat <(i32 (zextloadi16 (add IntRegs:$src1,
471 (shl IntRegs:$src2, u2ImmPred:$offset)))),
472 (L4_loadruh_rr IntRegs:$src1,
473 IntRegs:$src2, u2ImmPred:$offset)>,
476 def : Pat <(i32 (extloadi16 (add IntRegs:$src1,
477 (shl IntRegs:$src2, u2ImmPred:$offset)))),
478 (L4_loadruh_rr IntRegs:$src1,
479 IntRegs:$src2, u2ImmPred:$offset)>,
482 def : Pat <(i32 (load (add IntRegs:$src1,
483 (shl IntRegs:$src2, u2ImmPred:$offset)))),
484 (L4_loadri_rr IntRegs:$src1,
485 IntRegs:$src2, u2ImmPred:$offset)>,
488 def : Pat <(i64 (load (add IntRegs:$src1,
489 (shl IntRegs:$src2, u2ImmPred:$offset)))),
490 (L4_loadrd_rr IntRegs:$src1,
491 IntRegs:$src2, u2ImmPred:$offset)>,
496 // 'def pats' for load instruction base + register offset and
497 // zero immediate value.
498 let AddedComplexity = 10 in {
499 def : Pat <(i64 (load (add IntRegs:$src1, IntRegs:$src2))),
500 (L4_loadrd_rr IntRegs:$src1, IntRegs:$src2, 0)>,
503 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1, IntRegs:$src2))),
504 (L4_loadrb_rr IntRegs:$src1, IntRegs:$src2, 0)>,
507 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1, IntRegs:$src2))),
508 (L4_loadrub_rr IntRegs:$src1, IntRegs:$src2, 0)>,
511 def : Pat <(i32 (extloadi8 (add IntRegs:$src1, IntRegs:$src2))),
512 (L4_loadrub_rr IntRegs:$src1, IntRegs:$src2, 0)>,
515 def : Pat <(i32 (sextloadi16 (add IntRegs:$src1, IntRegs:$src2))),
516 (L4_loadrh_rr IntRegs:$src1, IntRegs:$src2, 0)>,
519 def : Pat <(i32 (zextloadi16 (add IntRegs:$src1, IntRegs:$src2))),
520 (L4_loadruh_rr IntRegs:$src1, IntRegs:$src2, 0)>,
523 def : Pat <(i32 (extloadi16 (add IntRegs:$src1, IntRegs:$src2))),
524 (L4_loadruh_rr IntRegs:$src1, IntRegs:$src2, 0)>,
527 def : Pat <(i32 (load (add IntRegs:$src1, IntRegs:$src2))),
528 (L4_loadri_rr IntRegs:$src1, IntRegs:$src2, 0)>,
533 def : Pat <(i64 (zext (i1 PredRegs:$src1))),
534 (i64 (A4_combineir 0, (C2_muxii (i1 PredRegs:$src1), 1, 0)))>,
538 def : Pat <(i64 (zext (i32 IntRegs:$src1))),
539 (i64 (A4_combineir 0, (i32 IntRegs:$src1)))>,
542 def: Pat <(i64 (zextloadi8 ADDRriS11_0:$src1)),
543 (i64 (A4_combineir 0, (L2_loadrub_io AddrFI:$src1, 0)))>,
546 let AddedComplexity = 20 in
547 def: Pat <(i64 (zextloadi8 (add (i32 IntRegs:$src1),
548 s11_0ExtPred:$offset))),
549 (i64 (A4_combineir 0, (L2_loadrub_io IntRegs:$src1,
550 s11_0ExtPred:$offset)))>,
554 def: Pat <(i64 (zextloadi1 ADDRriS11_0:$src1)),
555 (i64 (A4_combineir 0, (L2_loadrub_io AddrFI:$src1, 0)))>,
558 let AddedComplexity = 20 in
559 def: Pat <(i64 (zextloadi1 (add (i32 IntRegs:$src1),
560 s11_0ExtPred:$offset))),
561 (i64 (A4_combineir 0, (L2_loadrub_io IntRegs:$src1,
562 s11_0ExtPred:$offset)))>,
566 def: Pat <(i64 (zextloadi16 ADDRriS11_1:$src1)),
567 (i64 (A4_combineir 0, (L2_loadruh_io AddrFI:$src1, 0)))>,
570 let AddedComplexity = 20 in
571 def: Pat <(i64 (zextloadi16 (add (i32 IntRegs:$src1),
572 s11_1ExtPred:$offset))),
573 (i64 (A4_combineir 0, (L2_loadruh_io IntRegs:$src1,
574 s11_1ExtPred:$offset)))>,
578 def: Pat <(i64 (extloadi16 ADDRriS11_2:$src1)),
579 (i64 (A4_combineir 0, (L2_loadrh_io AddrFI:$src1, 0)))>,
582 let AddedComplexity = 20 in
583 def: Pat <(i64 (extloadi16 (add (i32 IntRegs:$src1),
584 s11_1ExtPred:$offset))),
585 (i64 (A4_combineir 0, (L2_loadrh_io IntRegs:$src1,
586 s11_1ExtPred:$offset)))>,
590 def: Pat <(i64 (zextloadi32 ADDRriS11_2:$src1)),
591 (i64 (A4_combineir 0, (L2_loadri_io AddrFI:$src1, 0)))>,
594 let AddedComplexity = 100 in
595 def: Pat <(i64 (zextloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
596 (i64 (A4_combineir 0, (L2_loadri_io IntRegs:$src1,
597 s11_2ExtPred:$offset)))>,
601 def: Pat <(i64 (extloadi32 ADDRriS11_2:$src1)),
602 (i64 (A4_combineir 0, (L2_loadri_io AddrFI:$src1, 0)))>,
605 let AddedComplexity = 100 in
606 def: Pat <(i64 (extloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
607 (i64 (A4_combineir 0, (L2_loadri_io IntRegs:$src1,
608 s11_2ExtPred:$offset)))>,
613 //===----------------------------------------------------------------------===//
615 //===----------------------------------------------------------------------===//
617 //===----------------------------------------------------------------------===//
619 //===----------------------------------------------------------------------===//
621 //===----------------------------------------------------------------------===//
622 // Template class for store instructions with Absolute set addressing mode.
623 //===----------------------------------------------------------------------===//
624 let isExtended = 1, opExtendable = 2, validSubTargets = HasV4SubT,
625 addrMode = AbsoluteSet in
626 class T_ST_abs_set<string mnemonic, RegisterClass RC>:
627 STInst2<(outs IntRegs:$dst1),
628 (ins RC:$src1, u0AlwaysExt:$src2),
629 mnemonic#"($dst1=##$src2) = $src1",
633 def STrid_abs_set_V4 : T_ST_abs_set <"memd", DoubleRegs>;
634 def STrib_abs_set_V4 : T_ST_abs_set <"memb", IntRegs>;
635 def STrih_abs_set_V4 : T_ST_abs_set <"memh", IntRegs>;
636 def STriw_abs_set_V4 : T_ST_abs_set <"memw", IntRegs>;
638 //===----------------------------------------------------------------------===//
639 // Template classes for the non-predicated store instructions with
640 // base + register offset addressing mode
641 //===----------------------------------------------------------------------===//
642 let isPredicable = 1 in
643 class T_store_rr <string mnemonic, RegisterClass RC, bits<3> MajOp, bit isH>
644 : STInst < (outs ), (ins IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, RC:$Rt),
645 mnemonic#"($Rs + $Ru<<#$u2) = $Rt"#!if(isH, ".h",""),
646 [],"",V4LDST_tc_st_SLOT01>, ImmRegShl, AddrModeRel {
655 let Inst{27-24} = 0b1011;
656 let Inst{23-21} = MajOp;
657 let Inst{20-16} = Rs;
659 let Inst{13} = u2{1};
664 //===----------------------------------------------------------------------===//
665 // Template classes for the predicated store instructions with
666 // base + register offset addressing mode
667 //===----------------------------------------------------------------------===//
668 let isPredicated = 1 in
669 class T_pstore_rr <string mnemonic, RegisterClass RC, bits<3> MajOp,
670 bit isNot, bit isPredNew, bit isH>
672 (ins PredRegs:$Pv, IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, RC:$Rt),
674 !if(isNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ",
675 ") ")#mnemonic#"($Rs+$Ru<<#$u2) = $Rt"#!if(isH, ".h",""),
676 [], "", V4LDST_tc_st_SLOT01> , AddrModeRel{
683 let isPredicatedFalse = isNot;
684 let isPredicatedNew = isPredNew;
688 let Inst{27-26} = 0b01;
689 let Inst{25} = isPredNew;
690 let Inst{24} = isNot;
691 let Inst{23-21} = MajOp;
692 let Inst{20-16} = Rs;
694 let Inst{13} = u2{1};
700 //===----------------------------------------------------------------------===//
701 // Template classes for the new-value store instructions with
702 // base + register offset addressing mode
703 //===----------------------------------------------------------------------===//
704 let isPredicable = 1, isNewValue = 1, opNewValue = 3 in
705 class T_store_new_rr <string mnemonic, bits<2> MajOp> :
706 NVInst < (outs ), (ins IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, IntRegs:$Nt),
707 mnemonic#"($Rs + $Ru<<#$u2) = $Nt.new",
708 [],"",V4LDST_tc_st_SLOT0>, ImmRegShl, AddrModeRel {
717 let Inst{27-21} = 0b1011101;
718 let Inst{20-16} = Rs;
720 let Inst{13} = u2{1};
722 let Inst{4-3} = MajOp;
726 //===----------------------------------------------------------------------===//
727 // Template classes for the predicated new-value store instructions with
728 // base + register offset addressing mode
729 //===----------------------------------------------------------------------===//
730 let isPredicated = 1, isNewValue = 1, opNewValue = 4 in
731 class T_pstore_new_rr <string mnemonic, bits<2> MajOp, bit isNot, bit isPredNew>
733 (ins PredRegs:$Pv, IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, IntRegs:$Nt),
734 !if(isNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ",
735 ") ")#mnemonic#"($Rs+$Ru<<#$u2) = $Nt.new",
736 [], "", V4LDST_tc_st_SLOT0>, AddrModeRel {
743 let isPredicatedFalse = isNot;
744 let isPredicatedNew = isPredNew;
747 let Inst{27-26} = 0b01;
748 let Inst{25} = isPredNew;
749 let Inst{24} = isNot;
750 let Inst{23-21} = 0b101;
751 let Inst{20-16} = Rs;
753 let Inst{13} = u2{1};
756 let Inst{4-3} = MajOp;
760 //===----------------------------------------------------------------------===//
761 // multiclass for store instructions with base + register offset addressing
763 //===----------------------------------------------------------------------===//
764 let isNVStorable = 1 in
765 multiclass ST_Idxd_shl<string mnemonic, string CextOp, RegisterClass RC,
766 bits<3> MajOp, bit isH = 0> {
767 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
768 def S4_#NAME#_rr : T_store_rr <mnemonic, RC, MajOp, isH>;
771 def S4_p#NAME#t_rr : T_pstore_rr <mnemonic, RC, MajOp, 0, 0, isH>;
772 def S4_p#NAME#f_rr : T_pstore_rr <mnemonic, RC, MajOp, 1, 0, isH>;
775 def S4_p#NAME#tnew_rr : T_pstore_rr <mnemonic, RC, MajOp, 0, 1, isH>;
776 def S4_p#NAME#fnew_rr : T_pstore_rr <mnemonic, RC, MajOp, 1, 1, isH>;
780 //===----------------------------------------------------------------------===//
781 // multiclass for new-value store instructions with base + register offset
783 //===----------------------------------------------------------------------===//
784 let mayStore = 1, isNVStore = 1 in
785 multiclass ST_Idxd_shl_nv <string mnemonic, string CextOp, RegisterClass RC,
787 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
788 def S4_#NAME#new_rr : T_store_new_rr<mnemonic, MajOp>;
791 def S4_p#NAME#newt_rr : T_pstore_new_rr <mnemonic, MajOp, 0, 0>;
792 def S4_p#NAME#newf_rr : T_pstore_new_rr <mnemonic, MajOp, 1, 0>;
795 def S4_p#NAME#newtnew_rr : T_pstore_new_rr <mnemonic, MajOp, 0, 1>;
796 def S4_p#NAME#newfnew_rr : T_pstore_new_rr <mnemonic, MajOp, 1, 1>;
800 let addrMode = BaseRegOffset, InputType = "reg", hasSideEffects = 0,
801 isCodeGenOnly = 0 in {
802 let accessSize = ByteAccess in
803 defm storerb: ST_Idxd_shl<"memb", "STrib", IntRegs, 0b000>,
804 ST_Idxd_shl_nv<"memb", "STrib", IntRegs, 0b00>;
806 let accessSize = HalfWordAccess in
807 defm storerh: ST_Idxd_shl<"memh", "STrih", IntRegs, 0b010>,
808 ST_Idxd_shl_nv<"memh", "STrih", IntRegs, 0b01>;
810 let accessSize = WordAccess in
811 defm storeri: ST_Idxd_shl<"memw", "STriw", IntRegs, 0b100>,
812 ST_Idxd_shl_nv<"memw", "STriw", IntRegs, 0b10>;
814 let isNVStorable = 0, accessSize = DoubleWordAccess in
815 defm storerd: ST_Idxd_shl<"memd", "STrid", DoubleRegs, 0b110>;
817 let isNVStorable = 0, accessSize = HalfWordAccess in
818 defm storerf: ST_Idxd_shl<"memh", "STrif", IntRegs, 0b011, 1>;
821 let Predicates = [HasV4T], AddedComplexity = 10 in {
822 def : Pat<(truncstorei8 (i32 IntRegs:$src4),
823 (add IntRegs:$src1, (shl IntRegs:$src2,
825 (S4_storerb_rr IntRegs:$src1, IntRegs:$src2,
826 u2ImmPred:$src3, IntRegs:$src4)>;
828 def : Pat<(truncstorei16 (i32 IntRegs:$src4),
829 (add IntRegs:$src1, (shl IntRegs:$src2,
831 (S4_storerh_rr IntRegs:$src1, IntRegs:$src2,
832 u2ImmPred:$src3, IntRegs:$src4)>;
834 def : Pat<(store (i32 IntRegs:$src4),
835 (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))),
836 (S4_storeri_rr IntRegs:$src1, IntRegs:$src2,
837 u2ImmPred:$src3, IntRegs:$src4)>;
839 def : Pat<(store (i64 DoubleRegs:$src4),
840 (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))),
841 (S4_storerd_rr IntRegs:$src1, IntRegs:$src2,
842 u2ImmPred:$src3, DoubleRegs:$src4)>;
845 let isExtended = 1, opExtendable = 2 in
846 class T_ST_LongOff <string mnemonic, PatFrag stOp, RegisterClass RC, ValueType VT> :
848 (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, RC:$src4),
849 mnemonic#"($src1<<#$src2+##$src3) = $src4",
850 [(stOp (VT RC:$src4),
851 (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2),
852 u0AlwaysExtPred:$src3))]>,
855 let isExtended = 1, opExtendable = 2, mayStore = 1, isNVStore = 1 in
856 class T_ST_LongOff_nv <string mnemonic> :
858 (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, IntRegs:$src4),
859 mnemonic#"($src1<<#$src2+##$src3) = $src4.new",
863 multiclass ST_LongOff <string mnemonic, string BaseOp, PatFrag stOp> {
864 let BaseOpcode = BaseOp#"_shl" in {
865 let isNVStorable = 1 in
866 def NAME#_V4 : T_ST_LongOff<mnemonic, stOp, IntRegs, i32>;
868 def NAME#_nv_V4 : T_ST_LongOff_nv<mnemonic>;
872 let AddedComplexity = 10, validSubTargets = HasV4SubT in {
873 def STrid_shl_V4 : T_ST_LongOff<"memd", store, DoubleRegs, i64>;
874 defm STrib_shl : ST_LongOff <"memb", "STrib", truncstorei8>, NewValueRel;
875 defm STrih_shl : ST_LongOff <"memh", "Strih", truncstorei16>, NewValueRel;
876 defm STriw_shl : ST_LongOff <"memw", "STriw", store>, NewValueRel;
879 let AddedComplexity = 40 in
880 multiclass T_ST_LOff_Pats <InstHexagon I, RegisterClass RC, ValueType VT,
882 def : Pat<(stOp (VT RC:$src4),
883 (add (shl IntRegs:$src1, u2ImmPred:$src2),
884 (NumUsesBelowThresCONST32 tglobaladdr:$src3))),
885 (I IntRegs:$src1, u2ImmPred:$src2, tglobaladdr:$src3, RC:$src4)>;
887 def : Pat<(stOp (VT RC:$src4),
889 (NumUsesBelowThresCONST32 tglobaladdr:$src3))),
890 (I IntRegs:$src1, 0, tglobaladdr:$src3, RC:$src4)>;
893 defm : T_ST_LOff_Pats<STrid_shl_V4, DoubleRegs, i64, store>;
894 defm : T_ST_LOff_Pats<STriw_shl_V4, IntRegs, i32, store>;
895 defm : T_ST_LOff_Pats<STrib_shl_V4, IntRegs, i32, truncstorei8>;
896 defm : T_ST_LOff_Pats<STrih_shl_V4, IntRegs, i32, truncstorei16>;
898 // memd(Rx++#s4:3)=Rtt
899 // memd(Rx++#s4:3:circ(Mu))=Rtt
900 // memd(Rx++I:circ(Mu))=Rtt
902 // memd(Rx++Mu:brev)=Rtt
903 // memd(gp+#u16:3)=Rtt
905 // Store doubleword conditionally.
906 // if ([!]Pv[.new]) memd(#u6)=Rtt
907 // TODO: needs to be implemented.
909 //===----------------------------------------------------------------------===//
910 // multiclass for store instructions with base + immediate offset
911 // addressing mode and immediate stored value.
912 // mem[bhw](Rx++#s4:3)=#s8
913 // if ([!]Pv[.new]) mem[bhw](Rx++#s4:3)=#s6
914 //===----------------------------------------------------------------------===//
915 multiclass ST_Imm_Pbase<string mnemonic, Operand OffsetOp, bit isNot,
917 let isPredicatedNew = isPredNew in
918 def NAME : STInst2<(outs),
919 (ins PredRegs:$src1, IntRegs:$src2, OffsetOp:$src3, s6Ext:$src4),
920 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
921 ") ")#mnemonic#"($src2+#$src3) = #$src4",
926 multiclass ST_Imm_Pred<string mnemonic, Operand OffsetOp, bit PredNot> {
927 let isPredicatedFalse = PredNot in {
928 defm _c#NAME : ST_Imm_Pbase<mnemonic, OffsetOp, PredNot, 0>;
930 defm _cdn#NAME : ST_Imm_Pbase<mnemonic, OffsetOp, PredNot, 1>;
934 let isExtendable = 1, isExtentSigned = 1, hasSideEffects = 0 in
935 multiclass ST_Imm<string mnemonic, string CextOp, Operand OffsetOp> {
936 let CextOpcode = CextOp, BaseOpcode = CextOp#_imm in {
937 let opExtendable = 2, opExtentBits = 8, isPredicable = 1 in
938 def NAME#_V4 : STInst2<(outs),
939 (ins IntRegs:$src1, OffsetOp:$src2, s8Ext:$src3),
940 mnemonic#"($src1+#$src2) = #$src3",
944 let opExtendable = 3, opExtentBits = 6, isPredicated = 1 in {
945 defm Pt_V4 : ST_Imm_Pred<mnemonic, OffsetOp, 0>;
946 defm NotPt_V4 : ST_Imm_Pred<mnemonic, OffsetOp, 1 >;
951 let addrMode = BaseImmOffset, InputType = "imm",
952 validSubTargets = HasV4SubT in {
953 let accessSize = ByteAccess in
954 defm STrib_imm : ST_Imm<"memb", "STrib", u6_0Imm>, ImmRegRel, PredNewRel;
956 let accessSize = HalfWordAccess in
957 defm STrih_imm : ST_Imm<"memh", "STrih", u6_1Imm>, ImmRegRel, PredNewRel;
959 let accessSize = WordAccess in
960 defm STriw_imm : ST_Imm<"memw", "STriw", u6_2Imm>, ImmRegRel, PredNewRel;
963 let Predicates = [HasV4T], AddedComplexity = 10 in {
964 def: Pat<(truncstorei8 s8ExtPred:$src3, (add IntRegs:$src1, u6_0ImmPred:$src2)),
965 (STrib_imm_V4 IntRegs:$src1, u6_0ImmPred:$src2, s8ExtPred:$src3)>;
967 def: Pat<(truncstorei16 s8ExtPred:$src3, (add IntRegs:$src1,
969 (STrih_imm_V4 IntRegs:$src1, u6_1ImmPred:$src2, s8ExtPred:$src3)>;
971 def: Pat<(store s8ExtPred:$src3, (add IntRegs:$src1, u6_2ImmPred:$src2)),
972 (STriw_imm_V4 IntRegs:$src1, u6_2ImmPred:$src2, s8ExtPred:$src3)>;
975 let AddedComplexity = 6 in
976 def : Pat <(truncstorei8 s8ExtPred:$src2, (i32 IntRegs:$src1)),
977 (STrib_imm_V4 IntRegs:$src1, 0, s8ExtPred:$src2)>,
980 // memb(Rx++#s4:0:circ(Mu))=Rt
981 // memb(Rx++I:circ(Mu))=Rt
983 // memb(Rx++Mu:brev)=Rt
984 // memb(gp+#u16:0)=Rt
988 // TODO: needs to be implemented
990 // memh(Rs+#s11:1)=Rt.H
991 let AddedComplexity = 6 in
992 def : Pat <(truncstorei16 s8ExtPred:$src2, (i32 IntRegs:$src1)),
993 (STrih_imm_V4 IntRegs:$src1, 0, s8ExtPred:$src2)>,
996 // memh(Rs+Ru<<#u2)=Rt.H
997 // TODO: needs to be implemented.
999 // memh(Ru<<#u2+#U6)=Rt.H
1000 // memh(Rx++#s4:1:circ(Mu))=Rt.H
1001 // memh(Rx++#s4:1:circ(Mu))=Rt
1002 // memh(Rx++I:circ(Mu))=Rt.H
1003 // memh(Rx++I:circ(Mu))=Rt
1004 // memh(Rx++Mu)=Rt.H
1006 // memh(Rx++Mu:brev)=Rt.H
1007 // memh(Rx++Mu:brev)=Rt
1008 // memh(gp+#u16:1)=Rt
1009 // if ([!]Pv[.new]) memh(#u6)=Rt.H
1010 // if ([!]Pv[.new]) memh(#u6)=Rt
1013 // if ([!]Pv[.new]) memh(Rs+#u6:1)=Rt.H
1014 // TODO: needs to be implemented.
1016 // if ([!]Pv[.new]) memh(Rx++#s4:1)=Rt.H
1017 // TODO: Needs to be implemented.
1021 // TODO: Needs to be implemented.
1024 let hasSideEffects = 0 in
1025 def STriw_pred_V4 : STInst2<(outs),
1026 (ins MEMri:$addr, PredRegs:$src1),
1027 "Error; should not emit",
1031 let AddedComplexity = 6 in
1032 def : Pat <(store s8ExtPred:$src2, (i32 IntRegs:$src1)),
1033 (STriw_imm_V4 IntRegs:$src1, 0, s8ExtPred:$src2)>,
1036 // memw(Rx++#s4:2)=Rt
1037 // memw(Rx++#s4:2:circ(Mu))=Rt
1038 // memw(Rx++I:circ(Mu))=Rt
1040 // memw(Rx++Mu:brev)=Rt
1042 //===----------------------------------------------------------------------===
1044 //===----------------------------------------------------------------------===
1047 //===----------------------------------------------------------------------===//
1049 //===----------------------------------------------------------------------===//
1051 // multiclass for new-value store instructions with base + immediate offset.
1053 multiclass ST_Idxd_Pbase_nv<string mnemonic, RegisterClass RC,
1054 Operand predImmOp, bit isNot, bit isPredNew> {
1055 let isPredicatedNew = isPredNew in
1056 def NAME#_nv_V4 : NVInst_V4<(outs),
1057 (ins PredRegs:$src1, IntRegs:$src2, predImmOp:$src3, RC: $src4),
1058 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
1059 ") ")#mnemonic#"($src2+#$src3) = $src4.new",
1064 multiclass ST_Idxd_Pred_nv<string mnemonic, RegisterClass RC, Operand predImmOp,
1066 let isPredicatedFalse = PredNot in {
1067 defm _c#NAME : ST_Idxd_Pbase_nv<mnemonic, RC, predImmOp, PredNot, 0>;
1069 defm _cdn#NAME : ST_Idxd_Pbase_nv<mnemonic, RC, predImmOp, PredNot, 1>;
1073 let mayStore = 1, isNVStore = 1, hasSideEffects = 0, isExtendable = 1 in
1074 multiclass ST_Idxd_nv<string mnemonic, string CextOp, RegisterClass RC,
1075 Operand ImmOp, Operand predImmOp, bits<5> ImmBits,
1076 bits<5> PredImmBits> {
1078 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
1079 let opExtendable = 1, isExtentSigned = 1, opExtentBits = ImmBits,
1081 def NAME#_nv_V4 : NVInst_V4<(outs),
1082 (ins IntRegs:$src1, ImmOp:$src2, RC:$src3),
1083 mnemonic#"($src1+#$src2) = $src3.new",
1087 let opExtendable = 2, isExtentSigned = 0, opExtentBits = PredImmBits,
1088 isPredicated = 1 in {
1089 defm Pt : ST_Idxd_Pred_nv<mnemonic, RC, predImmOp, 0>;
1090 defm NotPt : ST_Idxd_Pred_nv<mnemonic, RC, predImmOp, 1>;
1095 let addrMode = BaseImmOffset, validSubTargets = HasV4SubT in {
1096 let accessSize = ByteAccess in
1097 defm STrib_indexed: ST_Idxd_nv<"memb", "STrib", IntRegs, s11_0Ext,
1098 u6_0Ext, 11, 6>, AddrModeRel;
1100 let accessSize = HalfWordAccess in
1101 defm STrih_indexed: ST_Idxd_nv<"memh", "STrih", IntRegs, s11_1Ext,
1102 u6_1Ext, 12, 7>, AddrModeRel;
1104 let accessSize = WordAccess in
1105 defm STriw_indexed: ST_Idxd_nv<"memw", "STriw", IntRegs, s11_2Ext,
1106 u6_2Ext, 13, 8>, AddrModeRel;
1109 // multiclass for new-value store instructions with base + immediate offset.
1110 // and MEMri operand.
1111 multiclass ST_MEMri_Pbase_nv<string mnemonic, RegisterClass RC, bit isNot,
1113 let isPredicatedNew = isPredNew in
1114 def NAME#_nv_V4 : NVInst_V4<(outs),
1115 (ins PredRegs:$src1, MEMri:$addr, RC: $src2),
1116 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
1117 ") ")#mnemonic#"($addr) = $src2.new",
1122 multiclass ST_MEMri_Pred_nv<string mnemonic, RegisterClass RC, bit PredNot> {
1123 let isPredicatedFalse = PredNot in {
1124 defm _c#NAME : ST_MEMri_Pbase_nv<mnemonic, RC, PredNot, 0>;
1127 defm _cdn#NAME : ST_MEMri_Pbase_nv<mnemonic, RC, PredNot, 1>;
1131 let mayStore = 1, isNVStore = 1, isExtendable = 1, hasSideEffects = 0 in
1132 multiclass ST_MEMri_nv<string mnemonic, string CextOp, RegisterClass RC,
1133 bits<5> ImmBits, bits<5> PredImmBits> {
1135 let CextOpcode = CextOp, BaseOpcode = CextOp in {
1136 let opExtendable = 1, isExtentSigned = 1, opExtentBits = ImmBits,
1138 def NAME#_nv_V4 : NVInst_V4<(outs),
1139 (ins MEMri:$addr, RC:$src),
1140 mnemonic#"($addr) = $src.new",
1144 let opExtendable = 2, isExtentSigned = 0, opExtentBits = PredImmBits,
1145 hasSideEffects = 0, isPredicated = 1 in {
1146 defm Pt : ST_MEMri_Pred_nv<mnemonic, RC, 0>;
1147 defm NotPt : ST_MEMri_Pred_nv<mnemonic, RC, 1>;
1152 let addrMode = BaseImmOffset, isMEMri = "true", validSubTargets = HasV4SubT,
1154 let accessSize = ByteAccess in
1155 defm STrib: ST_MEMri_nv<"memb", "STrib", IntRegs, 11, 6>, AddrModeRel;
1157 let accessSize = HalfWordAccess in
1158 defm STrih: ST_MEMri_nv<"memh", "STrih", IntRegs, 12, 7>, AddrModeRel;
1160 let accessSize = WordAccess in
1161 defm STriw: ST_MEMri_nv<"memw", "STriw", IntRegs, 13, 8>, AddrModeRel;
1164 //===----------------------------------------------------------------------===//
1165 // Post increment store
1166 // mem[bhwd](Rx++#s4:[0123])=Nt.new
1167 //===----------------------------------------------------------------------===//
1169 multiclass ST_PostInc_Pbase_nv<string mnemonic, RegisterClass RC, Operand ImmOp,
1170 bit isNot, bit isPredNew> {
1171 let isPredicatedNew = isPredNew in
1172 def NAME#_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
1173 (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset, RC:$src3),
1174 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
1175 ") ")#mnemonic#"($src2++#$offset) = $src3.new",
1181 multiclass ST_PostInc_Pred_nv<string mnemonic, RegisterClass RC,
1182 Operand ImmOp, bit PredNot> {
1183 let isPredicatedFalse = PredNot in {
1184 defm _c#NAME : ST_PostInc_Pbase_nv<mnemonic, RC, ImmOp, PredNot, 0>;
1186 let Predicates = [HasV4T], validSubTargets = HasV4SubT in
1187 defm _cdn#NAME : ST_PostInc_Pbase_nv<mnemonic, RC, ImmOp, PredNot, 1>;
1191 let hasCtrlDep = 1, isNVStore = 1, hasSideEffects = 0 in
1192 multiclass ST_PostInc_nv<string mnemonic, string BaseOp, RegisterClass RC,
1195 let BaseOpcode = "POST_"#BaseOp in {
1196 let isPredicable = 1 in
1197 def NAME#_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
1198 (ins IntRegs:$src1, ImmOp:$offset, RC:$src2),
1199 mnemonic#"($src1++#$offset) = $src2.new",
1204 let isPredicated = 1 in {
1205 defm Pt : ST_PostInc_Pred_nv<mnemonic, RC, ImmOp, 0 >;
1206 defm NotPt : ST_PostInc_Pred_nv<mnemonic, RC, ImmOp, 1 >;
1211 let addrMode = PostInc, validSubTargets = HasV4SubT in {
1212 defm POST_STbri: ST_PostInc_nv <"memb", "STrib", IntRegs, s4_0Imm>, AddrModeRel;
1213 defm POST_SThri: ST_PostInc_nv <"memh", "STrih", IntRegs, s4_1Imm>, AddrModeRel;
1214 defm POST_STwri: ST_PostInc_nv <"memw", "STriw", IntRegs, s4_2Imm>, AddrModeRel;
1217 // memb(Rx++#s4:0:circ(Mu))=Nt.new
1218 // memb(Rx++I:circ(Mu))=Nt.new
1219 // memb(Rx++Mu)=Nt.new
1220 // memb(Rx++Mu:brev)=Nt.new
1221 // memh(Rx++#s4:1:circ(Mu))=Nt.new
1222 // memh(Rx++I:circ(Mu))=Nt.new
1223 // memh(Rx++Mu)=Nt.new
1224 // memh(Rx++Mu:brev)=Nt.new
1226 // memw(Rx++#s4:2:circ(Mu))=Nt.new
1227 // memw(Rx++I:circ(Mu))=Nt.new
1228 // memw(Rx++Mu)=Nt.new
1229 // memw(Rx++Mu:brev)=Nt.new
1231 //===----------------------------------------------------------------------===//
1233 //===----------------------------------------------------------------------===//
1235 //===----------------------------------------------------------------------===//
1237 //===----------------------------------------------------------------------===//
1239 //===----------------------------------------------------------------------===//
1240 // multiclass/template class for the new-value compare jumps with the register
1242 //===----------------------------------------------------------------------===//
1244 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11 in
1245 class NVJrr_template<string mnemonic, bits<3> majOp, bit NvOpNum,
1246 bit isNegCond, bit isTak>
1248 (ins IntRegs:$src1, IntRegs:$src2, brtarget:$offset),
1249 "if ("#!if(isNegCond, "!","")#mnemonic#
1250 "($src1"#!if(!eq(NvOpNum, 0),".new, ",", ")#
1251 "$src2"#!if(!eq(NvOpNum, 1),".new))","))")#" jump:"
1252 #!if(isTak, "t","nt")#" $offset",
1253 []>, Requires<[HasV4T]> {
1257 bits<3> Ns; // New-Value Operand
1258 bits<5> RegOp; // Non-New-Value Operand
1261 let isTaken = isTak;
1262 let isBrTaken = !if(isTaken, "true", "false");
1263 let isPredicatedFalse = isNegCond;
1265 let Ns = !if(!eq(NvOpNum, 0), src1{2-0}, src2{2-0});
1266 let RegOp = !if(!eq(NvOpNum, 0), src2, src1);
1268 let IClass = 0b0010;
1270 let Inst{25-23} = majOp;
1271 let Inst{22} = isNegCond;
1272 let Inst{18-16} = Ns;
1273 let Inst{13} = isTak;
1274 let Inst{12-8} = RegOp;
1275 let Inst{21-20} = offset{10-9};
1276 let Inst{7-1} = offset{8-2};
1280 multiclass NVJrr_cond<string mnemonic, bits<3> majOp, bit NvOpNum,
1282 // Branch not taken:
1283 def _nt_V4: NVJrr_template<mnemonic, majOp, NvOpNum, isNegCond, 0>;
1285 def _t_V4: NVJrr_template<mnemonic, majOp, NvOpNum, isNegCond, 1>;
1288 // NvOpNum = 0 -> First Operand is a new-value Register
1289 // NvOpNum = 1 -> Second Operand is a new-value Register
1291 multiclass NVJrr_base<string mnemonic, string BaseOp, bits<3> majOp,
1293 let BaseOpcode = BaseOp#_NVJ in {
1294 defm _t_Jumpnv : NVJrr_cond<mnemonic, majOp, NvOpNum, 0>; // True cond
1295 defm _f_Jumpnv : NVJrr_cond<mnemonic, majOp, NvOpNum, 1>; // False cond
1299 // if ([!]cmp.eq(Ns.new,Rt)) jump:[n]t #r9:2
1300 // if ([!]cmp.gt(Ns.new,Rt)) jump:[n]t #r9:2
1301 // if ([!]cmp.gtu(Ns.new,Rt)) jump:[n]t #r9:2
1302 // if ([!]cmp.gt(Rt,Ns.new)) jump:[n]t #r9:2
1303 // if ([!]cmp.gtu(Rt,Ns.new)) jump:[n]t #r9:2
1305 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1,
1306 Defs = [PC], hasSideEffects = 0, validSubTargets = HasV4SubT in {
1307 defm CMPEQrr : NVJrr_base<"cmp.eq", "CMPEQ", 0b000, 0>, PredRel;
1308 defm CMPGTrr : NVJrr_base<"cmp.gt", "CMPGT", 0b001, 0>, PredRel;
1309 defm CMPGTUrr : NVJrr_base<"cmp.gtu", "CMPGTU", 0b010, 0>, PredRel;
1310 defm CMPLTrr : NVJrr_base<"cmp.gt", "CMPLT", 0b011, 1>, PredRel;
1311 defm CMPLTUrr : NVJrr_base<"cmp.gtu", "CMPLTU", 0b100, 1>, PredRel;
1314 //===----------------------------------------------------------------------===//
1315 // multiclass/template class for the new-value compare jumps instruction
1316 // with a register and an unsigned immediate (U5) operand.
1317 //===----------------------------------------------------------------------===//
1319 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11 in
1320 class NVJri_template<string mnemonic, bits<3> majOp, bit isNegCond,
1323 (ins IntRegs:$src1, u5Imm:$src2, brtarget:$offset),
1324 "if ("#!if(isNegCond, "!","")#mnemonic#"($src1.new, #$src2)) jump:"
1325 #!if(isTak, "t","nt")#" $offset",
1326 []>, Requires<[HasV4T]> {
1328 let isTaken = isTak;
1329 let isPredicatedFalse = isNegCond;
1330 let isBrTaken = !if(isTaken, "true", "false");
1336 let IClass = 0b0010;
1338 let Inst{25-23} = majOp;
1339 let Inst{22} = isNegCond;
1340 let Inst{18-16} = src1;
1341 let Inst{13} = isTak;
1342 let Inst{12-8} = src2;
1343 let Inst{21-20} = offset{10-9};
1344 let Inst{7-1} = offset{8-2};
1347 multiclass NVJri_cond<string mnemonic, bits<3> majOp, bit isNegCond> {
1348 // Branch not taken:
1349 def _nt_V4: NVJri_template<mnemonic, majOp, isNegCond, 0>;
1351 def _t_V4: NVJri_template<mnemonic, majOp, isNegCond, 1>;
1354 multiclass NVJri_base<string mnemonic, string BaseOp, bits<3> majOp> {
1355 let BaseOpcode = BaseOp#_NVJri in {
1356 defm _t_Jumpnv : NVJri_cond<mnemonic, majOp, 0>; // True Cond
1357 defm _f_Jumpnv : NVJri_cond<mnemonic, majOp, 1>; // False cond
1361 // if ([!]cmp.eq(Ns.new,#U5)) jump:[n]t #r9:2
1362 // if ([!]cmp.gt(Ns.new,#U5)) jump:[n]t #r9:2
1363 // if ([!]cmp.gtu(Ns.new,#U5)) jump:[n]t #r9:2
1365 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1,
1366 Defs = [PC], hasSideEffects = 0, validSubTargets = HasV4SubT in {
1367 defm CMPEQri : NVJri_base<"cmp.eq", "CMPEQ", 0b000>, PredRel;
1368 defm CMPGTri : NVJri_base<"cmp.gt", "CMPGT", 0b001>, PredRel;
1369 defm CMPGTUri : NVJri_base<"cmp.gtu", "CMPGTU", 0b010>, PredRel;
1372 //===----------------------------------------------------------------------===//
1373 // multiclass/template class for the new-value compare jumps instruction
1374 // with a register and an hardcoded 0/-1 immediate value.
1375 //===----------------------------------------------------------------------===//
1377 let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 11 in
1378 class NVJ_ConstImm_template<string mnemonic, bits<3> majOp, string ImmVal,
1379 bit isNegCond, bit isTak>
1381 (ins IntRegs:$src1, brtarget:$offset),
1382 "if ("#!if(isNegCond, "!","")#mnemonic
1383 #"($src1.new, #"#ImmVal#")) jump:"
1384 #!if(isTak, "t","nt")#" $offset",
1385 []>, Requires<[HasV4T]> {
1387 let isTaken = isTak;
1388 let isPredicatedFalse = isNegCond;
1389 let isBrTaken = !if(isTaken, "true", "false");
1393 let IClass = 0b0010;
1395 let Inst{25-23} = majOp;
1396 let Inst{22} = isNegCond;
1397 let Inst{18-16} = src1;
1398 let Inst{13} = isTak;
1399 let Inst{21-20} = offset{10-9};
1400 let Inst{7-1} = offset{8-2};
1403 multiclass NVJ_ConstImm_cond<string mnemonic, bits<3> majOp, string ImmVal,
1405 // Branch not taken:
1406 def _nt_V4: NVJ_ConstImm_template<mnemonic, majOp, ImmVal, isNegCond, 0>;
1408 def _t_V4: NVJ_ConstImm_template<mnemonic, majOp, ImmVal, isNegCond, 1>;
1411 multiclass NVJ_ConstImm_base<string mnemonic, string BaseOp, bits<3> majOp,
1413 let BaseOpcode = BaseOp#_NVJ_ConstImm in {
1414 defm _t_Jumpnv : NVJ_ConstImm_cond<mnemonic, majOp, ImmVal, 0>; // True cond
1415 defm _f_Jumpnv : NVJ_ConstImm_cond<mnemonic, majOp, ImmVal, 1>; // False Cond
1419 // if ([!]tstbit(Ns.new,#0)) jump:[n]t #r9:2
1420 // if ([!]cmp.eq(Ns.new,#-1)) jump:[n]t #r9:2
1421 // if ([!]cmp.gt(Ns.new,#-1)) jump:[n]t #r9:2
1423 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator=1,
1424 Defs = [PC], hasSideEffects = 0 in {
1425 defm TSTBIT0 : NVJ_ConstImm_base<"tstbit", "TSTBIT", 0b011, "0">, PredRel;
1426 defm CMPEQn1 : NVJ_ConstImm_base<"cmp.eq", "CMPEQ", 0b100, "-1">, PredRel;
1427 defm CMPGTn1 : NVJ_ConstImm_base<"cmp.gt", "CMPGT", 0b101, "-1">, PredRel;
1430 //===----------------------------------------------------------------------===//
1432 //===----------------------------------------------------------------------===//
1434 // Add and accumulate.
1435 // Rd=add(Rs,add(Ru,#s6))
1436 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 6,
1437 validSubTargets = HasV4SubT in
1438 def ADDr_ADDri_V4 : MInst<(outs IntRegs:$dst),
1439 (ins IntRegs:$src1, IntRegs:$src2, s6Ext:$src3),
1440 "$dst = add($src1, add($src2, #$src3))",
1441 [(set (i32 IntRegs:$dst),
1442 (add (i32 IntRegs:$src1), (add (i32 IntRegs:$src2),
1443 s6_16ExtPred:$src3)))]>,
1446 // Rd=add(Rs,sub(#s6,Ru))
1447 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6,
1448 validSubTargets = HasV4SubT in
1449 def ADDr_SUBri_V4 : MInst<(outs IntRegs:$dst),
1450 (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3),
1451 "$dst = add($src1, sub(#$src2, $src3))",
1452 [(set (i32 IntRegs:$dst),
1453 (add (i32 IntRegs:$src1), (sub s6_10ExtPred:$src2,
1454 (i32 IntRegs:$src3))))]>,
1457 // Generates the same instruction as ADDr_SUBri_V4 but matches different
1459 // Rd=add(Rs,sub(#s6,Ru))
1460 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6,
1461 validSubTargets = HasV4SubT in
1462 def ADDri_SUBr_V4 : MInst<(outs IntRegs:$dst),
1463 (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3),
1464 "$dst = add($src1, sub(#$src2, $src3))",
1465 [(set (i32 IntRegs:$dst),
1466 (sub (add (i32 IntRegs:$src1), s6_10ExtPred:$src2),
1467 (i32 IntRegs:$src3)))]>,
1471 // Add or subtract doublewords with carry.
1473 // Rdd=add(Rss,Rtt,Px):carry
1475 // Rdd=sub(Rss,Rtt,Px):carry
1478 // Logical doublewords.
1479 // Rdd=and(Rtt,~Rss)
1480 let validSubTargets = HasV4SubT in
1481 def ANDd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
1482 (ins DoubleRegs:$src1, DoubleRegs:$src2),
1483 "$dst = and($src1, ~$src2)",
1484 [(set (i64 DoubleRegs:$dst), (and (i64 DoubleRegs:$src1),
1485 (not (i64 DoubleRegs:$src2))))]>,
1489 let validSubTargets = HasV4SubT in
1490 def ORd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
1491 (ins DoubleRegs:$src1, DoubleRegs:$src2),
1492 "$dst = or($src1, ~$src2)",
1493 [(set (i64 DoubleRegs:$dst),
1494 (or (i64 DoubleRegs:$src1), (not (i64 DoubleRegs:$src2))))]>,
1498 // Logical-logical doublewords.
1499 // Rxx^=xor(Rss,Rtt)
1500 let validSubTargets = HasV4SubT in
1501 def XORd_XORdd: MInst_acc<(outs DoubleRegs:$dst),
1502 (ins DoubleRegs:$src1, DoubleRegs:$src2, DoubleRegs:$src3),
1503 "$dst ^= xor($src2, $src3)",
1504 [(set (i64 DoubleRegs:$dst),
1505 (xor (i64 DoubleRegs:$src1), (xor (i64 DoubleRegs:$src2),
1506 (i64 DoubleRegs:$src3))))],
1511 // Logical-logical words.
1512 // Rx=or(Ru,and(Rx,#s10))
1513 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
1514 validSubTargets = HasV4SubT in
1515 def ORr_ANDri_V4 : MInst_acc<(outs IntRegs:$dst),
1516 (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
1517 "$dst = or($src1, and($src2, #$src3))",
1518 [(set (i32 IntRegs:$dst),
1519 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1520 s10ExtPred:$src3)))],
1524 // Rx[&|^]=and(Rs,Rt)
1526 let validSubTargets = HasV4SubT in
1527 def ANDr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
1528 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1529 "$dst &= and($src2, $src3)",
1530 [(set (i32 IntRegs:$dst),
1531 (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1532 (i32 IntRegs:$src3))))],
1537 let validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "reg" in
1538 def ORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
1539 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1540 "$dst |= and($src2, $src3)",
1541 [(set (i32 IntRegs:$dst),
1542 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1543 (i32 IntRegs:$src3))))],
1545 Requires<[HasV4T]>, ImmRegRel;
1548 let validSubTargets = HasV4SubT in
1549 def XORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
1550 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1551 "$dst ^= and($src2, $src3)",
1552 [(set (i32 IntRegs:$dst),
1553 (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1554 (i32 IntRegs:$src3))))],
1558 // Rx[&|^]=and(Rs,~Rt)
1560 let validSubTargets = HasV4SubT in
1561 def ANDr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
1562 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1563 "$dst &= and($src2, ~$src3)",
1564 [(set (i32 IntRegs:$dst),
1565 (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1566 (not (i32 IntRegs:$src3)))))],
1571 let validSubTargets = HasV4SubT in
1572 def ORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
1573 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1574 "$dst |= and($src2, ~$src3)",
1575 [(set (i32 IntRegs:$dst),
1576 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1577 (not (i32 IntRegs:$src3)))))],
1582 let validSubTargets = HasV4SubT in
1583 def XORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
1584 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1585 "$dst ^= and($src2, ~$src3)",
1586 [(set (i32 IntRegs:$dst),
1587 (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1588 (not (i32 IntRegs:$src3)))))],
1592 // Rx[&|^]=or(Rs,Rt)
1594 let validSubTargets = HasV4SubT in
1595 def ANDr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1596 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1597 "$dst &= or($src2, $src3)",
1598 [(set (i32 IntRegs:$dst),
1599 (and (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
1600 (i32 IntRegs:$src3))))],
1605 let validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "reg" in
1606 def ORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1607 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1608 "$dst |= or($src2, $src3)",
1609 [(set (i32 IntRegs:$dst),
1610 (or (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
1611 (i32 IntRegs:$src3))))],
1613 Requires<[HasV4T]>, ImmRegRel;
1616 let validSubTargets = HasV4SubT in
1617 def XORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1618 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1619 "$dst ^= or($src2, $src3)",
1620 [(set (i32 IntRegs:$dst),
1621 (xor (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
1622 (i32 IntRegs:$src3))))],
1626 // Rx[&|^]=xor(Rs,Rt)
1628 let validSubTargets = HasV4SubT in
1629 def ANDr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1630 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1631 "$dst &= xor($src2, $src3)",
1632 [(set (i32 IntRegs:$dst),
1633 (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
1634 (i32 IntRegs:$src3))))],
1639 let validSubTargets = HasV4SubT in
1640 def ORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1641 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1642 "$dst |= xor($src2, $src3)",
1643 [(set (i32 IntRegs:$dst),
1644 (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
1645 (i32 IntRegs:$src3))))],
1650 let validSubTargets = HasV4SubT in
1651 def XORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1652 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1653 "$dst ^= xor($src2, $src3)",
1654 [(set (i32 IntRegs:$dst),
1655 (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
1656 (i32 IntRegs:$src3))))],
1661 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
1662 validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "imm" in
1663 def ORr_ANDri2_V4 : MInst_acc<(outs IntRegs:$dst),
1664 (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
1665 "$dst |= and($src2, #$src3)",
1666 [(set (i32 IntRegs:$dst),
1667 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1668 s10ExtPred:$src3)))],
1670 Requires<[HasV4T]>, ImmRegRel;
1673 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
1674 validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "imm" in
1675 def ORr_ORri_V4 : MInst_acc<(outs IntRegs:$dst),
1676 (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
1677 "$dst |= or($src2, #$src3)",
1678 [(set (i32 IntRegs:$dst),
1679 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1680 s10ExtPred:$src3)))],
1682 Requires<[HasV4T]>, ImmRegRel;
1686 // Rd=modwrap(Rs,Rt)
1688 // Rd=cround(Rs,#u5)
1690 // Rd=round(Rs,#u5)[:sat]
1691 // Rd=round(Rs,Rt)[:sat]
1692 // Vector reduce add unsigned halfwords
1693 // Rd=vraddh(Rss,Rtt)
1695 // Rdd=vaddb(Rss,Rtt)
1696 // Vector conditional negate
1697 // Rdd=vcnegh(Rss,Rt)
1698 // Rxx+=vrcnegh(Rss,Rt)
1699 // Vector maximum bytes
1700 // Rdd=vmaxb(Rtt,Rss)
1701 // Vector reduce maximum halfwords
1702 // Rxx=vrmaxh(Rss,Ru)
1703 // Rxx=vrmaxuh(Rss,Ru)
1704 // Vector reduce maximum words
1705 // Rxx=vrmaxuw(Rss,Ru)
1706 // Rxx=vrmaxw(Rss,Ru)
1707 // Vector minimum bytes
1708 // Rdd=vminb(Rtt,Rss)
1709 // Vector reduce minimum halfwords
1710 // Rxx=vrminh(Rss,Ru)
1711 // Rxx=vrminuh(Rss,Ru)
1712 // Vector reduce minimum words
1713 // Rxx=vrminuw(Rss,Ru)
1714 // Rxx=vrminw(Rss,Ru)
1715 // Vector subtract bytes
1716 // Rdd=vsubb(Rss,Rtt)
1718 //===----------------------------------------------------------------------===//
1720 //===----------------------------------------------------------------------===//
1723 //===----------------------------------------------------------------------===//
1725 //===----------------------------------------------------------------------===//
1727 // Multiply and user lower result.
1728 // Rd=add(#u6,mpyi(Rs,#U6))
1729 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6,
1730 validSubTargets = HasV4SubT in
1731 def ADDi_MPYri_V4 : MInst<(outs IntRegs:$dst),
1732 (ins u6Ext:$src1, IntRegs:$src2, u6Imm:$src3),
1733 "$dst = add(#$src1, mpyi($src2, #$src3))",
1734 [(set (i32 IntRegs:$dst),
1735 (add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
1736 u6ExtPred:$src1))]>,
1739 // Rd=add(##,mpyi(Rs,#U6))
1740 def : Pat <(add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
1741 (HexagonCONST32 tglobaladdr:$src1)),
1742 (i32 (ADDi_MPYri_V4 tglobaladdr:$src1, IntRegs:$src2,
1745 // Rd=add(#u6,mpyi(Rs,Rt))
1746 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6,
1747 validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in
1748 def ADDi_MPYrr_V4 : MInst<(outs IntRegs:$dst),
1749 (ins u6Ext:$src1, IntRegs:$src2, IntRegs:$src3),
1750 "$dst = add(#$src1, mpyi($src2, $src3))",
1751 [(set (i32 IntRegs:$dst),
1752 (add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
1753 u6ExtPred:$src1))]>,
1754 Requires<[HasV4T]>, ImmRegRel;
1756 // Rd=add(##,mpyi(Rs,Rt))
1757 def : Pat <(add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
1758 (HexagonCONST32 tglobaladdr:$src1)),
1759 (i32 (ADDi_MPYrr_V4 tglobaladdr:$src1, IntRegs:$src2,
1762 // Rd=add(Ru,mpyi(#u6:2,Rs))
1763 let validSubTargets = HasV4SubT in
1764 def ADDr_MPYir_V4 : MInst<(outs IntRegs:$dst),
1765 (ins IntRegs:$src1, u6Imm:$src2, IntRegs:$src3),
1766 "$dst = add($src1, mpyi(#$src2, $src3))",
1767 [(set (i32 IntRegs:$dst),
1768 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src3),
1769 u6_2ImmPred:$src2)))]>,
1772 // Rd=add(Ru,mpyi(Rs,#u6))
1773 let isExtendable = 1, opExtendable = 3, isExtentSigned = 0, opExtentBits = 6,
1774 validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in
1775 def ADDr_MPYri_V4 : MInst<(outs IntRegs:$dst),
1776 (ins IntRegs:$src1, IntRegs:$src2, u6Ext:$src3),
1777 "$dst = add($src1, mpyi($src2, #$src3))",
1778 [(set (i32 IntRegs:$dst),
1779 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
1780 u6ExtPred:$src3)))]>,
1781 Requires<[HasV4T]>, ImmRegRel;
1783 // Rx=add(Ru,mpyi(Rx,Rs))
1784 let validSubTargets = HasV4SubT, InputType = "reg", CextOpcode = "ADD_MPY" in
1785 def ADDr_MPYrr_V4 : MInst_acc<(outs IntRegs:$dst),
1786 (ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3),
1787 "$dst = add($src1, mpyi($src2, $src3))",
1788 [(set (i32 IntRegs:$dst),
1789 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
1790 (i32 IntRegs:$src3))))],
1792 Requires<[HasV4T]>, ImmRegRel;
1795 // Polynomial multiply words
1797 // Rxx^=pmpyw(Rs,Rt)
1799 // Vector reduce multiply word by signed half (32x16)
1800 // Rdd=vrmpyweh(Rss,Rtt)[:<<1]
1801 // Rdd=vrmpywoh(Rss,Rtt)[:<<1]
1802 // Rxx+=vrmpyweh(Rss,Rtt)[:<<1]
1803 // Rxx+=vrmpywoh(Rss,Rtt)[:<<1]
1805 // Multiply and use upper result
1806 // Rd=mpy(Rs,Rt.H):<<1:sat
1807 // Rd=mpy(Rs,Rt.L):<<1:sat
1808 // Rd=mpy(Rs,Rt):<<1
1809 // Rd=mpy(Rs,Rt):<<1:sat
1811 // Rx+=mpy(Rs,Rt):<<1:sat
1812 // Rx-=mpy(Rs,Rt):<<1:sat
1814 // Vector multiply bytes
1815 // Rdd=vmpybsu(Rs,Rt)
1816 // Rdd=vmpybu(Rs,Rt)
1817 // Rxx+=vmpybsu(Rs,Rt)
1818 // Rxx+=vmpybu(Rs,Rt)
1820 // Vector polynomial multiply halfwords
1821 // Rdd=vpmpyh(Rs,Rt)
1822 // Rxx^=vpmpyh(Rs,Rt)
1824 //===----------------------------------------------------------------------===//
1826 //===----------------------------------------------------------------------===//
1829 //===----------------------------------------------------------------------===//
1831 //===----------------------------------------------------------------------===//
1833 // Shift by immediate and accumulate.
1834 // Rx=add(#u8,asl(Rx,#U5))
1835 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1836 validSubTargets = HasV4SubT in
1837 def ADDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
1838 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1839 "$dst = add(#$src1, asl($src2, #$src3))",
1840 [(set (i32 IntRegs:$dst),
1841 (add (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
1846 // Rx=add(#u8,lsr(Rx,#U5))
1847 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1848 validSubTargets = HasV4SubT in
1849 def ADDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
1850 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1851 "$dst = add(#$src1, lsr($src2, #$src3))",
1852 [(set (i32 IntRegs:$dst),
1853 (add (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
1858 // Rx=sub(#u8,asl(Rx,#U5))
1859 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1860 validSubTargets = HasV4SubT in
1861 def SUBi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
1862 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1863 "$dst = sub(#$src1, asl($src2, #$src3))",
1864 [(set (i32 IntRegs:$dst),
1865 (sub (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
1870 // Rx=sub(#u8,lsr(Rx,#U5))
1871 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1872 validSubTargets = HasV4SubT in
1873 def SUBi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
1874 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1875 "$dst = sub(#$src1, lsr($src2, #$src3))",
1876 [(set (i32 IntRegs:$dst),
1877 (sub (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
1883 //Shift by immediate and logical.
1884 //Rx=and(#u8,asl(Rx,#U5))
1885 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1886 validSubTargets = HasV4SubT in
1887 def ANDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
1888 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1889 "$dst = and(#$src1, asl($src2, #$src3))",
1890 [(set (i32 IntRegs:$dst),
1891 (and (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
1896 //Rx=and(#u8,lsr(Rx,#U5))
1897 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1898 validSubTargets = HasV4SubT in
1899 def ANDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
1900 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1901 "$dst = and(#$src1, lsr($src2, #$src3))",
1902 [(set (i32 IntRegs:$dst),
1903 (and (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
1908 //Rx=or(#u8,asl(Rx,#U5))
1909 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1910 AddedComplexity = 30, validSubTargets = HasV4SubT in
1911 def ORi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
1912 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1913 "$dst = or(#$src1, asl($src2, #$src3))",
1914 [(set (i32 IntRegs:$dst),
1915 (or (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
1920 //Rx=or(#u8,lsr(Rx,#U5))
1921 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1922 AddedComplexity = 30, validSubTargets = HasV4SubT in
1923 def ORi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
1924 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1925 "$dst = or(#$src1, lsr($src2, #$src3))",
1926 [(set (i32 IntRegs:$dst),
1927 (or (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
1933 //Shift by register.
1935 let validSubTargets = HasV4SubT in {
1936 def LSLi_V4 : MInst<(outs IntRegs:$dst), (ins s6Imm:$src1, IntRegs:$src2),
1937 "$dst = lsl(#$src1, $src2)",
1938 [(set (i32 IntRegs:$dst), (shl s6ImmPred:$src1,
1939 (i32 IntRegs:$src2)))]>,
1943 //Shift by register and logical.
1945 def ASLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
1946 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
1947 "$dst ^= asl($src2, $src3)",
1948 [(set (i64 DoubleRegs:$dst),
1949 (xor (i64 DoubleRegs:$src1), (shl (i64 DoubleRegs:$src2),
1950 (i32 IntRegs:$src3))))],
1955 def ASRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
1956 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
1957 "$dst ^= asr($src2, $src3)",
1958 [(set (i64 DoubleRegs:$dst),
1959 (xor (i64 DoubleRegs:$src1), (sra (i64 DoubleRegs:$src2),
1960 (i32 IntRegs:$src3))))],
1965 def LSLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
1966 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
1967 "$dst ^= lsl($src2, $src3)",
1968 [(set (i64 DoubleRegs:$dst), (xor (i64 DoubleRegs:$src1),
1969 (shl (i64 DoubleRegs:$src2),
1970 (i32 IntRegs:$src3))))],
1975 def LSRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
1976 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
1977 "$dst ^= lsr($src2, $src3)",
1978 [(set (i64 DoubleRegs:$dst),
1979 (xor (i64 DoubleRegs:$src1), (srl (i64 DoubleRegs:$src2),
1980 (i32 IntRegs:$src3))))],
1985 //===----------------------------------------------------------------------===//
1987 //===----------------------------------------------------------------------===//
1989 //===----------------------------------------------------------------------===//
1990 // MEMOP: Word, Half, Byte
1991 //===----------------------------------------------------------------------===//
1993 def MEMOPIMM : SDNodeXForm<imm, [{
1994 // Call the transformation function XformM5ToU5Imm to get the negative
1995 // immediate's positive counterpart.
1996 int32_t imm = N->getSExtValue();
1997 return XformM5ToU5Imm(imm);
2000 def MEMOPIMM_HALF : SDNodeXForm<imm, [{
2001 // -1 .. -31 represented as 65535..65515
2002 // assigning to a short restores our desired signed value.
2003 // Call the transformation function XformM5ToU5Imm to get the negative
2004 // immediate's positive counterpart.
2005 int16_t imm = N->getSExtValue();
2006 return XformM5ToU5Imm(imm);
2009 def MEMOPIMM_BYTE : SDNodeXForm<imm, [{
2010 // -1 .. -31 represented as 255..235
2011 // assigning to a char restores our desired signed value.
2012 // Call the transformation function XformM5ToU5Imm to get the negative
2013 // immediate's positive counterpart.
2014 int8_t imm = N->getSExtValue();
2015 return XformM5ToU5Imm(imm);
2018 def SETMEMIMM : SDNodeXForm<imm, [{
2019 // Return the bit position we will set [0-31].
2021 int32_t imm = N->getSExtValue();
2022 return XformMskToBitPosU5Imm(imm);
2025 def CLRMEMIMM : SDNodeXForm<imm, [{
2026 // Return the bit position we will clear [0-31].
2028 // we bit negate the value first
2029 int32_t imm = ~(N->getSExtValue());
2030 return XformMskToBitPosU5Imm(imm);
2033 def SETMEMIMM_SHORT : SDNodeXForm<imm, [{
2034 // Return the bit position we will set [0-15].
2036 int16_t imm = N->getSExtValue();
2037 return XformMskToBitPosU4Imm(imm);
2040 def CLRMEMIMM_SHORT : SDNodeXForm<imm, [{
2041 // Return the bit position we will clear [0-15].
2043 // we bit negate the value first
2044 int16_t imm = ~(N->getSExtValue());
2045 return XformMskToBitPosU4Imm(imm);
2048 def SETMEMIMM_BYTE : SDNodeXForm<imm, [{
2049 // Return the bit position we will set [0-7].
2051 int8_t imm = N->getSExtValue();
2052 return XformMskToBitPosU3Imm(imm);
2055 def CLRMEMIMM_BYTE : SDNodeXForm<imm, [{
2056 // Return the bit position we will clear [0-7].
2058 // we bit negate the value first
2059 int8_t imm = ~(N->getSExtValue());
2060 return XformMskToBitPosU3Imm(imm);
2063 //===----------------------------------------------------------------------===//
2064 // Template class for MemOp instructions with the register value.
2065 //===----------------------------------------------------------------------===//
2066 class MemOp_rr_base <string opc, bits<2> opcBits, Operand ImmOp,
2067 string memOp, bits<2> memOpBits> :
2069 (ins IntRegs:$base, ImmOp:$offset, IntRegs:$delta),
2070 opc#"($base+#$offset)"#memOp#"$delta",
2072 Requires<[HasV4T, UseMEMOP]> {
2077 bits<6> offsetBits; // memb - u6:0 , memh - u6:1, memw - u6:2
2079 let offsetBits = !if (!eq(opcBits, 0b00), offset{5-0},
2080 !if (!eq(opcBits, 0b01), offset{6-1},
2081 !if (!eq(opcBits, 0b10), offset{7-2},0)));
2083 let IClass = 0b0011;
2084 let Inst{27-24} = 0b1110;
2085 let Inst{22-21} = opcBits;
2086 let Inst{20-16} = base;
2088 let Inst{12-7} = offsetBits;
2089 let Inst{6-5} = memOpBits;
2090 let Inst{4-0} = delta;
2093 //===----------------------------------------------------------------------===//
2094 // Template class for MemOp instructions with the immediate value.
2095 //===----------------------------------------------------------------------===//
2096 class MemOp_ri_base <string opc, bits<2> opcBits, Operand ImmOp,
2097 string memOp, bits<2> memOpBits> :
2099 (ins IntRegs:$base, ImmOp:$offset, u5Imm:$delta),
2100 opc#"($base+#$offset)"#memOp#"#$delta"
2101 #!if(memOpBits{1},")", ""), // clrbit, setbit - include ')'
2103 Requires<[HasV4T, UseMEMOP]> {
2108 bits<6> offsetBits; // memb - u6:0 , memh - u6:1, memw - u6:2
2110 let offsetBits = !if (!eq(opcBits, 0b00), offset{5-0},
2111 !if (!eq(opcBits, 0b01), offset{6-1},
2112 !if (!eq(opcBits, 0b10), offset{7-2},0)));
2114 let IClass = 0b0011;
2115 let Inst{27-24} = 0b1111;
2116 let Inst{22-21} = opcBits;
2117 let Inst{20-16} = base;
2119 let Inst{12-7} = offsetBits;
2120 let Inst{6-5} = memOpBits;
2121 let Inst{4-0} = delta;
2124 // multiclass to define MemOp instructions with register operand.
2125 multiclass MemOp_rr<string opc, bits<2> opcBits, Operand ImmOp> {
2126 def _ADD#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " += ", 0b00>; // add
2127 def _SUB#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " -= ", 0b01>; // sub
2128 def _AND#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " &= ", 0b10>; // and
2129 def _OR#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " |= ", 0b11>; // or
2132 // multiclass to define MemOp instructions with immediate Operand.
2133 multiclass MemOp_ri<string opc, bits<2> opcBits, Operand ImmOp> {
2134 def _ADD#NAME#_V4 : MemOp_ri_base <opc, opcBits, ImmOp, " += ", 0b00 >;
2135 def _SUB#NAME#_V4 : MemOp_ri_base <opc, opcBits, ImmOp, " -= ", 0b01 >;
2136 def _CLRBIT#NAME#_V4 : MemOp_ri_base<opc, opcBits, ImmOp, " =clrbit(", 0b10>;
2137 def _SETBIT#NAME#_V4 : MemOp_ri_base<opc, opcBits, ImmOp, " =setbit(", 0b11>;
2140 multiclass MemOp_base <string opc, bits<2> opcBits, Operand ImmOp> {
2141 defm r : MemOp_rr <opc, opcBits, ImmOp>;
2142 defm i : MemOp_ri <opc, opcBits, ImmOp>;
2145 // Define MemOp instructions.
2146 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0,
2147 validSubTargets =HasV4SubT in {
2148 let opExtentBits = 6, accessSize = ByteAccess in
2149 defm MemOPb : MemOp_base <"memb", 0b00, u6_0Ext>;
2151 let opExtentBits = 7, accessSize = HalfWordAccess in
2152 defm MemOPh : MemOp_base <"memh", 0b01, u6_1Ext>;
2154 let opExtentBits = 8, accessSize = WordAccess in
2155 defm MemOPw : MemOp_base <"memw", 0b10, u6_2Ext>;
2158 //===----------------------------------------------------------------------===//
2159 // Multiclass to define 'Def Pats' for ALU operations on the memory
2160 // Here value used for the ALU operation is an immediate value.
2161 // mem[bh](Rs+#0) += #U5
2162 // mem[bh](Rs+#u6) += #U5
2163 //===----------------------------------------------------------------------===//
2165 multiclass MemOpi_u5Pats <PatFrag ldOp, PatFrag stOp, PatLeaf ExtPred,
2166 InstHexagon MI, SDNode OpNode> {
2167 let AddedComplexity = 180 in
2168 def : Pat < (stOp (OpNode (ldOp IntRegs:$addr), u5ImmPred:$addend),
2170 (MI IntRegs:$addr, #0, u5ImmPred:$addend )>;
2172 let AddedComplexity = 190 in
2173 def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, ExtPred:$offset)),
2175 (add IntRegs:$base, ExtPred:$offset)),
2176 (MI IntRegs:$base, ExtPred:$offset, u5ImmPred:$addend)>;
2179 multiclass MemOpi_u5ALUOp<PatFrag ldOp, PatFrag stOp, PatLeaf ExtPred,
2180 InstHexagon addMI, InstHexagon subMI> {
2181 defm : MemOpi_u5Pats<ldOp, stOp, ExtPred, addMI, add>;
2182 defm : MemOpi_u5Pats<ldOp, stOp, ExtPred, subMI, sub>;
2185 multiclass MemOpi_u5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2187 defm : MemOpi_u5ALUOp <ldOpHalf, truncstorei16, u6_1ExtPred,
2188 MemOPh_ADDi_V4, MemOPh_SUBi_V4>;
2190 defm : MemOpi_u5ALUOp <ldOpByte, truncstorei8, u6ExtPred,
2191 MemOPb_ADDi_V4, MemOPb_SUBi_V4>;
2194 let Predicates = [HasV4T, UseMEMOP] in {
2195 defm : MemOpi_u5ExtType<zextloadi8, zextloadi16>; // zero extend
2196 defm : MemOpi_u5ExtType<sextloadi8, sextloadi16>; // sign extend
2197 defm : MemOpi_u5ExtType<extloadi8, extloadi16>; // any extend
2200 defm : MemOpi_u5ALUOp <load, store, u6_2ExtPred, MemOPw_ADDi_V4,
2204 //===----------------------------------------------------------------------===//
2205 // multiclass to define 'Def Pats' for ALU operations on the memory.
2206 // Here value used for the ALU operation is a negative value.
2207 // mem[bh](Rs+#0) += #m5
2208 // mem[bh](Rs+#u6) += #m5
2209 //===----------------------------------------------------------------------===//
2211 multiclass MemOpi_m5Pats <PatFrag ldOp, PatFrag stOp, PatLeaf extPred,
2212 PatLeaf immPred, ComplexPattern addrPred,
2213 SDNodeXForm xformFunc, InstHexagon MI> {
2214 let AddedComplexity = 190 in
2215 def : Pat <(stOp (add (ldOp IntRegs:$addr), immPred:$subend),
2217 (MI IntRegs:$addr, #0, (xformFunc immPred:$subend) )>;
2219 let AddedComplexity = 195 in
2220 def : Pat<(stOp (add (ldOp (add IntRegs:$base, extPred:$offset)),
2222 (add IntRegs:$base, extPred:$offset)),
2223 (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$subend))>;
2226 multiclass MemOpi_m5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2228 defm : MemOpi_m5Pats <ldOpHalf, truncstorei16, u6_1ExtPred, m5HImmPred,
2229 ADDRriU6_1, MEMOPIMM_HALF, MemOPh_SUBi_V4>;
2231 defm : MemOpi_m5Pats <ldOpByte, truncstorei8, u6ExtPred, m5BImmPred,
2232 ADDRriU6_0, MEMOPIMM_BYTE, MemOPb_SUBi_V4>;
2235 let Predicates = [HasV4T, UseMEMOP] in {
2236 defm : MemOpi_m5ExtType<zextloadi8, zextloadi16>; // zero extend
2237 defm : MemOpi_m5ExtType<sextloadi8, sextloadi16>; // sign extend
2238 defm : MemOpi_m5ExtType<extloadi8, extloadi16>; // any extend
2241 defm : MemOpi_m5Pats <load, store, u6_2ExtPred, m5ImmPred,
2242 ADDRriU6_2, MEMOPIMM, MemOPw_SUBi_V4>;
2245 //===----------------------------------------------------------------------===//
2246 // Multiclass to define 'def Pats' for bit operations on the memory.
2247 // mem[bhw](Rs+#0) = [clrbit|setbit](#U5)
2248 // mem[bhw](Rs+#u6) = [clrbit|setbit](#U5)
2249 //===----------------------------------------------------------------------===//
2251 multiclass MemOpi_bitPats <PatFrag ldOp, PatFrag stOp, PatLeaf immPred,
2252 PatLeaf extPred, ComplexPattern addrPred,
2253 SDNodeXForm xformFunc, InstHexagon MI, SDNode OpNode> {
2255 // mem[bhw](Rs+#u6:[012]) = [clrbit|setbit](#U5)
2256 let AddedComplexity = 250 in
2257 def : Pat<(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)),
2259 (add IntRegs:$base, extPred:$offset)),
2260 (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$bitend))>;
2262 // mem[bhw](Rs+#0) = [clrbit|setbit](#U5)
2263 let AddedComplexity = 225 in
2264 def : Pat <(stOp (OpNode (ldOp (addrPred IntRegs:$addr, extPred:$offset)),
2266 (addrPred (i32 IntRegs:$addr), extPred:$offset)),
2267 (MI IntRegs:$addr, extPred:$offset, (xformFunc immPred:$bitend))>;
2270 multiclass MemOpi_bitExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2272 defm : MemOpi_bitPats<ldOpByte, truncstorei8, Clr3ImmPred, u6ExtPred,
2273 ADDRriU6_0, CLRMEMIMM_BYTE, MemOPb_CLRBITi_V4, and>;
2275 defm : MemOpi_bitPats<ldOpByte, truncstorei8, Set3ImmPred, u6ExtPred,
2276 ADDRriU6_0, SETMEMIMM_BYTE, MemOPb_SETBITi_V4, or>;
2277 // Half Word - clrbit
2278 defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Clr4ImmPred, u6_1ExtPred,
2279 ADDRriU6_1, CLRMEMIMM_SHORT, MemOPh_CLRBITi_V4, and>;
2280 // Half Word - setbit
2281 defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Set4ImmPred, u6_1ExtPred,
2282 ADDRriU6_1, SETMEMIMM_SHORT, MemOPh_SETBITi_V4, or>;
2285 let Predicates = [HasV4T, UseMEMOP] in {
2286 // mem[bh](Rs+#0) = [clrbit|setbit](#U5)
2287 // mem[bh](Rs+#u6:[01]) = [clrbit|setbit](#U5)
2288 defm : MemOpi_bitExtType<zextloadi8, zextloadi16>; // zero extend
2289 defm : MemOpi_bitExtType<sextloadi8, sextloadi16>; // sign extend
2290 defm : MemOpi_bitExtType<extloadi8, extloadi16>; // any extend
2292 // memw(Rs+#0) = [clrbit|setbit](#U5)
2293 // memw(Rs+#u6:2) = [clrbit|setbit](#U5)
2294 defm : MemOpi_bitPats<load, store, Clr5ImmPred, u6_2ExtPred, ADDRriU6_2,
2295 CLRMEMIMM, MemOPw_CLRBITi_V4, and>;
2296 defm : MemOpi_bitPats<load, store, Set5ImmPred, u6_2ExtPred, ADDRriU6_2,
2297 SETMEMIMM, MemOPw_SETBITi_V4, or>;
2300 //===----------------------------------------------------------------------===//
2301 // Multiclass to define 'def Pats' for ALU operations on the memory
2302 // where addend is a register.
2303 // mem[bhw](Rs+#0) [+-&|]= Rt
2304 // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt
2305 //===----------------------------------------------------------------------===//
2307 multiclass MemOpr_Pats <PatFrag ldOp, PatFrag stOp, ComplexPattern addrPred,
2308 PatLeaf extPred, InstHexagon MI, SDNode OpNode> {
2309 let AddedComplexity = 141 in
2310 // mem[bhw](Rs+#0) [+-&|]= Rt
2311 def : Pat <(stOp (OpNode (ldOp (addrPred IntRegs:$addr, extPred:$offset)),
2312 (i32 IntRegs:$addend)),
2313 (addrPred (i32 IntRegs:$addr), extPred:$offset)),
2314 (MI IntRegs:$addr, extPred:$offset, (i32 IntRegs:$addend) )>;
2316 // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt
2317 let AddedComplexity = 150 in
2318 def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)),
2319 (i32 IntRegs:$orend)),
2320 (add IntRegs:$base, extPred:$offset)),
2321 (MI IntRegs:$base, extPred:$offset, (i32 IntRegs:$orend) )>;
2324 multiclass MemOPr_ALUOp<PatFrag ldOp, PatFrag stOp,
2325 ComplexPattern addrPred, PatLeaf extPred,
2326 InstHexagon addMI, InstHexagon subMI,
2327 InstHexagon andMI, InstHexagon orMI > {
2329 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, addMI, add>;
2330 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, subMI, sub>;
2331 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, andMI, and>;
2332 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, orMI, or>;
2335 multiclass MemOPr_ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2337 defm : MemOPr_ALUOp <ldOpHalf, truncstorei16, ADDRriU6_1, u6_1ExtPred,
2338 MemOPh_ADDr_V4, MemOPh_SUBr_V4,
2339 MemOPh_ANDr_V4, MemOPh_ORr_V4>;
2341 defm : MemOPr_ALUOp <ldOpByte, truncstorei8, ADDRriU6_0, u6ExtPred,
2342 MemOPb_ADDr_V4, MemOPb_SUBr_V4,
2343 MemOPb_ANDr_V4, MemOPb_ORr_V4>;
2346 // Define 'def Pats' for MemOps with register addend.
2347 let Predicates = [HasV4T, UseMEMOP] in {
2349 defm : MemOPr_ExtType<zextloadi8, zextloadi16>; // zero extend
2350 defm : MemOPr_ExtType<sextloadi8, sextloadi16>; // sign extend
2351 defm : MemOPr_ExtType<extloadi8, extloadi16>; // any extend
2353 defm : MemOPr_ALUOp <load, store, ADDRriU6_2, u6_2ExtPred, MemOPw_ADDr_V4,
2354 MemOPw_SUBr_V4, MemOPw_ANDr_V4, MemOPw_ORr_V4 >;
2357 //===----------------------------------------------------------------------===//
2359 //===----------------------------------------------------------------------===//
2361 // Hexagon V4 only supports these flavors of byte/half compare instructions:
2362 // EQ/GT/GTU. Other flavors like GE/GEU/LT/LTU/LE/LEU are not supported by
2363 // hardware. However, compiler can still implement these patterns through
2364 // appropriate patterns combinations based on current implemented patterns.
2365 // The implemented patterns are: EQ/GT/GTU.
2366 // Missing patterns are: GE/GEU/LT/LTU/LE/LEU.
2368 // Following instruction is not being extended as it results into the
2369 // incorrect code for negative numbers.
2370 // Pd=cmpb.eq(Rs,#u8)
2372 let isCompare = 1, isExtendable = 1, opExtendable = 2, hasSideEffects = 0,
2373 validSubTargets = HasV4SubT in
2374 class CMP_NOT_REG_IMM<string OpName, bits<2> op, Operand ImmOp,
2376 : ALU32Inst <(outs PredRegs:$dst), (ins IntRegs:$src1, ImmOp:$src2),
2377 "$dst = !cmp."#OpName#"($src1, #$src2)",
2379 "", ALU32_2op_tc_2early_SLOT0123> {
2384 let IClass = 0b0111;
2385 let Inst{27-24} = 0b0101;
2386 let Inst{23-22} = op;
2387 let Inst{20-16} = src1;
2388 let Inst{21} = !if (!eq(OpName, "gtu"), 0b0, src2{9});
2389 let Inst{13-5} = src2{8-0};
2390 let Inst{4-2} = 0b100;
2391 let Inst{1-0} = dst;
2394 let opExtentBits = 10, isExtentSigned = 1 in {
2395 def C4_cmpneqi : CMP_NOT_REG_IMM <"eq", 0b00, s10Ext, [(set (i1 PredRegs:$dst),
2396 (setne (i32 IntRegs:$src1), s10ExtPred:$src2))]>;
2398 def C4_cmpltei : CMP_NOT_REG_IMM <"gt", 0b01, s10Ext, [(set (i1 PredRegs:$dst),
2399 (not (setgt (i32 IntRegs:$src1), s10ExtPred:$src2)))]>;
2402 let opExtentBits = 9 in
2403 def C4_cmplteui : CMP_NOT_REG_IMM <"gtu", 0b10, u9Ext, [(set (i1 PredRegs:$dst),
2404 (not (setugt (i32 IntRegs:$src1), u9ExtPred:$src2)))]>;
2409 let isCompare = 1, validSubTargets = HasV4SubT in
2410 def CMPnotEQ_rr : ALU32_rr<(outs PredRegs:$dst),
2411 (ins IntRegs:$src1, IntRegs:$src2),
2412 "$dst = !cmp.eq($src1, $src2)",
2413 [(set (i1 PredRegs:$dst),
2414 (setne (i32 IntRegs:$src1), (i32 IntRegs:$src2)))]>,
2418 let isCompare = 1, validSubTargets = HasV4SubT in
2419 def CMPnotGT_rr : ALU32_rr<(outs PredRegs:$dst),
2420 (ins IntRegs:$src1, IntRegs:$src2),
2421 "$dst = !cmp.gt($src1, $src2)",
2422 [(set (i1 PredRegs:$dst),
2423 (not (setgt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))]>,
2427 // p=!cmp.gtu(r1,r2)
2428 let isCompare = 1, validSubTargets = HasV4SubT in
2429 def CMPnotGTU_rr : ALU32_rr<(outs PredRegs:$dst),
2430 (ins IntRegs:$src1, IntRegs:$src2),
2431 "$dst = !cmp.gtu($src1, $src2)",
2432 [(set (i1 PredRegs:$dst),
2433 (not (setugt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))]>,
2436 let isCompare = 1, validSubTargets = HasV4SubT in
2437 def CMPbEQri_V4 : MInst<(outs PredRegs:$dst),
2438 (ins IntRegs:$src1, u8Imm:$src2),
2439 "$dst = cmpb.eq($src1, #$src2)",
2440 [(set (i1 PredRegs:$dst),
2441 (seteq (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2))]>,
2444 def : Pat <(brcond (i1 (setne (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2)),
2446 (J2_jumpf (CMPbEQri_V4 (i32 IntRegs:$src1), u8ImmPred:$src2),
2450 // Pd=cmpb.eq(Rs,Rt)
2451 let isCompare = 1, validSubTargets = HasV4SubT in
2452 def CMPbEQrr_ubub_V4 : MInst<(outs PredRegs:$dst),
2453 (ins IntRegs:$src1, IntRegs:$src2),
2454 "$dst = cmpb.eq($src1, $src2)",
2455 [(set (i1 PredRegs:$dst),
2456 (seteq (and (xor (i32 IntRegs:$src1),
2457 (i32 IntRegs:$src2)), 255), 0))]>,
2460 // Pd=cmpb.eq(Rs,Rt)
2461 let isCompare = 1, validSubTargets = HasV4SubT in
2462 def CMPbEQrr_sbsb_V4 : MInst<(outs PredRegs:$dst),
2463 (ins IntRegs:$src1, IntRegs:$src2),
2464 "$dst = cmpb.eq($src1, $src2)",
2465 [(set (i1 PredRegs:$dst),
2466 (seteq (shl (i32 IntRegs:$src1), (i32 24)),
2467 (shl (i32 IntRegs:$src2), (i32 24))))]>,
2470 // Pd=cmpb.gt(Rs,Rt)
2471 let isCompare = 1, validSubTargets = HasV4SubT in
2472 def CMPbGTrr_V4 : MInst<(outs PredRegs:$dst),
2473 (ins IntRegs:$src1, IntRegs:$src2),
2474 "$dst = cmpb.gt($src1, $src2)",
2475 [(set (i1 PredRegs:$dst),
2476 (setgt (shl (i32 IntRegs:$src1), (i32 24)),
2477 (shl (i32 IntRegs:$src2), (i32 24))))]>,
2480 // Pd=cmpb.gtu(Rs,#u7)
2481 let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7,
2482 isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU", InputType = "imm" in
2483 def CMPbGTUri_V4 : MInst<(outs PredRegs:$dst),
2484 (ins IntRegs:$src1, u7Ext:$src2),
2485 "$dst = cmpb.gtu($src1, #$src2)",
2486 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255),
2487 u7ExtPred:$src2))]>,
2488 Requires<[HasV4T]>, ImmRegRel;
2490 // SDNode for converting immediate C to C-1.
2491 def DEC_CONST_BYTE : SDNodeXForm<imm, [{
2492 // Return the byte immediate const-1 as an SDNode.
2493 int32_t imm = N->getSExtValue();
2494 return XformU7ToU7M1Imm(imm);
2498 // zext( seteq ( and(Rs, 255), u8))
2500 // Pd=cmpb.eq(Rs, #u8)
2501 // if (Pd.new) Rd=#1
2502 // if (!Pd.new) Rd=#0
2503 def : Pat <(i32 (zext (i1 (seteq (i32 (and (i32 IntRegs:$Rs), 255)),
2505 (i32 (TFR_condset_ii (i1 (CMPbEQri_V4 (i32 IntRegs:$Rs),
2511 // zext( setne ( and(Rs, 255), u8))
2513 // Pd=cmpb.eq(Rs, #u8)
2514 // if (Pd.new) Rd=#0
2515 // if (!Pd.new) Rd=#1
2516 def : Pat <(i32 (zext (i1 (setne (i32 (and (i32 IntRegs:$Rs), 255)),
2518 (i32 (TFR_condset_ii (i1 (CMPbEQri_V4 (i32 IntRegs:$Rs),
2524 // zext( seteq (Rs, and(Rt, 255)))
2526 // Pd=cmpb.eq(Rs, Rt)
2527 // if (Pd.new) Rd=#1
2528 // if (!Pd.new) Rd=#0
2529 def : Pat <(i32 (zext (i1 (seteq (i32 IntRegs:$Rt),
2530 (i32 (and (i32 IntRegs:$Rs), 255)))))),
2531 (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs),
2532 (i32 IntRegs:$Rt))),
2537 // zext( setne (Rs, and(Rt, 255)))
2539 // Pd=cmpb.eq(Rs, Rt)
2540 // if (Pd.new) Rd=#0
2541 // if (!Pd.new) Rd=#1
2542 def : Pat <(i32 (zext (i1 (setne (i32 IntRegs:$Rt),
2543 (i32 (and (i32 IntRegs:$Rs), 255)))))),
2544 (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs),
2545 (i32 IntRegs:$Rt))),
2550 // zext( setugt ( and(Rs, 255), u8))
2552 // Pd=cmpb.gtu(Rs, #u8)
2553 // if (Pd.new) Rd=#1
2554 // if (!Pd.new) Rd=#0
2555 def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 255)),
2557 (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs),
2563 // zext( setugt ( and(Rs, 254), u8))
2565 // Pd=cmpb.gtu(Rs, #u8)
2566 // if (Pd.new) Rd=#1
2567 // if (!Pd.new) Rd=#0
2568 def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 254)),
2570 (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs),
2576 // zext( setult ( Rs, Rt))
2578 // Pd=cmp.ltu(Rs, Rt)
2579 // if (Pd.new) Rd=#1
2580 // if (!Pd.new) Rd=#0
2581 // cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs)
2582 def : Pat <(i32 (zext (i1 (setult (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2583 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rt),
2584 (i32 IntRegs:$Rs))),
2589 // zext( setlt ( Rs, Rt))
2591 // Pd=cmp.lt(Rs, Rt)
2592 // if (Pd.new) Rd=#1
2593 // if (!Pd.new) Rd=#0
2594 // cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs)
2595 def : Pat <(i32 (zext (i1 (setlt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2596 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rt),
2597 (i32 IntRegs:$Rs))),
2602 // zext( setugt ( Rs, Rt))
2604 // Pd=cmp.gtu(Rs, Rt)
2605 // if (Pd.new) Rd=#1
2606 // if (!Pd.new) Rd=#0
2607 def : Pat <(i32 (zext (i1 (setugt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2608 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rs),
2609 (i32 IntRegs:$Rt))),
2613 // This pattern interefers with coremark performance, not implementing at this
2616 // zext( setgt ( Rs, Rt))
2618 // Pd=cmp.gt(Rs, Rt)
2619 // if (Pd.new) Rd=#1
2620 // if (!Pd.new) Rd=#0
2623 // zext( setuge ( Rs, Rt))
2625 // Pd=cmp.ltu(Rs, Rt)
2626 // if (Pd.new) Rd=#0
2627 // if (!Pd.new) Rd=#1
2628 // cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs)
2629 def : Pat <(i32 (zext (i1 (setuge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2630 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rt),
2631 (i32 IntRegs:$Rs))),
2636 // zext( setge ( Rs, Rt))
2638 // Pd=cmp.lt(Rs, Rt)
2639 // if (Pd.new) Rd=#0
2640 // if (!Pd.new) Rd=#1
2641 // cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs)
2642 def : Pat <(i32 (zext (i1 (setge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2643 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rt),
2644 (i32 IntRegs:$Rs))),
2649 // zext( setule ( Rs, Rt))
2651 // Pd=cmp.gtu(Rs, Rt)
2652 // if (Pd.new) Rd=#0
2653 // if (!Pd.new) Rd=#1
2654 def : Pat <(i32 (zext (i1 (setule (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2655 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rs),
2656 (i32 IntRegs:$Rt))),
2661 // zext( setle ( Rs, Rt))
2663 // Pd=cmp.gt(Rs, Rt)
2664 // if (Pd.new) Rd=#0
2665 // if (!Pd.new) Rd=#1
2666 def : Pat <(i32 (zext (i1 (setle (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2667 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rs),
2668 (i32 IntRegs:$Rt))),
2673 // zext( setult ( and(Rs, 255), u8))
2674 // Use the isdigit transformation below
2676 // Generate code of the form 'mux_ii(cmpbgtu(Rdd, C-1),0,1)'
2677 // for C code of the form r = ((c>='0') & (c<='9')) ? 1 : 0;.
2678 // The isdigit transformation relies on two 'clever' aspects:
2679 // 1) The data type is unsigned which allows us to eliminate a zero test after
2680 // biasing the expression by 48. We are depending on the representation of
2681 // the unsigned types, and semantics.
2682 // 2) The front end has converted <= 9 into < 10 on entry to LLVM
2685 // retval = ((c>='0') & (c<='9')) ? 1 : 0;
2686 // The code is transformed upstream of llvm into
2687 // retval = (c-48) < 10 ? 1 : 0;
2688 let AddedComplexity = 139 in
2689 def : Pat <(i32 (zext (i1 (setult (i32 (and (i32 IntRegs:$src1), 255)),
2690 u7StrictPosImmPred:$src2)))),
2691 (i32 (C2_muxii (i1 (CMPbGTUri_V4 (i32 IntRegs:$src1),
2692 (DEC_CONST_BYTE u7StrictPosImmPred:$src2))),
2696 // Pd=cmpb.gtu(Rs,Rt)
2697 let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU",
2698 InputType = "reg" in
2699 def CMPbGTUrr_V4 : MInst<(outs PredRegs:$dst),
2700 (ins IntRegs:$src1, IntRegs:$src2),
2701 "$dst = cmpb.gtu($src1, $src2)",
2702 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255),
2703 (and (i32 IntRegs:$src2), 255)))]>,
2704 Requires<[HasV4T]>, ImmRegRel;
2706 // Following instruction is not being extended as it results into the incorrect
2707 // code for negative numbers.
2709 // Signed half compare(.eq) ri.
2710 // Pd=cmph.eq(Rs,#s8)
2711 let isCompare = 1, validSubTargets = HasV4SubT in
2712 def CMPhEQri_V4 : MInst<(outs PredRegs:$dst),
2713 (ins IntRegs:$src1, s8Imm:$src2),
2714 "$dst = cmph.eq($src1, #$src2)",
2715 [(set (i1 PredRegs:$dst), (seteq (and (i32 IntRegs:$src1), 65535),
2716 s8ImmPred:$src2))]>,
2719 // Signed half compare(.eq) rr.
2720 // Case 1: xor + and, then compare:
2722 // r0=and(r0,#0xffff)
2724 // Pd=cmph.eq(Rs,Rt)
2725 let isCompare = 1, validSubTargets = HasV4SubT in
2726 def CMPhEQrr_xor_V4 : MInst<(outs PredRegs:$dst),
2727 (ins IntRegs:$src1, IntRegs:$src2),
2728 "$dst = cmph.eq($src1, $src2)",
2729 [(set (i1 PredRegs:$dst), (seteq (and (xor (i32 IntRegs:$src1),
2730 (i32 IntRegs:$src2)),
2734 // Signed half compare(.eq) rr.
2735 // Case 2: shift left 16 bits then compare:
2739 // Pd=cmph.eq(Rs,Rt)
2740 let isCompare = 1, validSubTargets = HasV4SubT in
2741 def CMPhEQrr_shl_V4 : MInst<(outs PredRegs:$dst),
2742 (ins IntRegs:$src1, IntRegs:$src2),
2743 "$dst = cmph.eq($src1, $src2)",
2744 [(set (i1 PredRegs:$dst),
2745 (seteq (shl (i32 IntRegs:$src1), (i32 16)),
2746 (shl (i32 IntRegs:$src2), (i32 16))))]>,
2749 /* Incorrect Pattern -- immediate should be right shifted before being
2750 used in the cmph.gt instruction.
2751 // Signed half compare(.gt) ri.
2752 // Pd=cmph.gt(Rs,#s8)
2754 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 8,
2755 isCompare = 1, validSubTargets = HasV4SubT in
2756 def CMPhGTri_V4 : MInst<(outs PredRegs:$dst),
2757 (ins IntRegs:$src1, s8Ext:$src2),
2758 "$dst = cmph.gt($src1, #$src2)",
2759 [(set (i1 PredRegs:$dst),
2760 (setgt (shl (i32 IntRegs:$src1), (i32 16)),
2761 s8ExtPred:$src2))]>,
2765 // Signed half compare(.gt) rr.
2766 // Pd=cmph.gt(Rs,Rt)
2767 let isCompare = 1, validSubTargets = HasV4SubT in
2768 def CMPhGTrr_shl_V4 : MInst<(outs PredRegs:$dst),
2769 (ins IntRegs:$src1, IntRegs:$src2),
2770 "$dst = cmph.gt($src1, $src2)",
2771 [(set (i1 PredRegs:$dst),
2772 (setgt (shl (i32 IntRegs:$src1), (i32 16)),
2773 (shl (i32 IntRegs:$src2), (i32 16))))]>,
2776 // Unsigned half compare rr (.gtu).
2777 // Pd=cmph.gtu(Rs,Rt)
2778 let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU",
2779 InputType = "reg" in
2780 def CMPhGTUrr_V4 : MInst<(outs PredRegs:$dst),
2781 (ins IntRegs:$src1, IntRegs:$src2),
2782 "$dst = cmph.gtu($src1, $src2)",
2783 [(set (i1 PredRegs:$dst),
2784 (setugt (and (i32 IntRegs:$src1), 65535),
2785 (and (i32 IntRegs:$src2), 65535)))]>,
2786 Requires<[HasV4T]>, ImmRegRel;
2788 // Unsigned half compare ri (.gtu).
2789 // Pd=cmph.gtu(Rs,#u7)
2790 let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7,
2791 isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU",
2792 InputType = "imm" in
2793 def CMPhGTUri_V4 : MInst<(outs PredRegs:$dst),
2794 (ins IntRegs:$src1, u7Ext:$src2),
2795 "$dst = cmph.gtu($src1, #$src2)",
2796 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 65535),
2797 u7ExtPred:$src2))]>,
2798 Requires<[HasV4T]>, ImmRegRel;
2800 let validSubTargets = HasV4SubT in
2801 def NTSTBIT_rr : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
2802 "$dst = !tstbit($src1, $src2)",
2803 [(set (i1 PredRegs:$dst),
2804 (seteq (and (shl 1, (i32 IntRegs:$src2)), (i32 IntRegs:$src1)), 0))]>,
2807 let validSubTargets = HasV4SubT in
2808 def NTSTBIT_ri : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
2809 "$dst = !tstbit($src1, $src2)",
2810 [(set (i1 PredRegs:$dst),
2811 (seteq (and (shl 1, u5ImmPred:$src2), (i32 IntRegs:$src1)), 0))]>,
2814 //===----------------------------------------------------------------------===//
2816 //===----------------------------------------------------------------------===//
2818 //Deallocate frame and return.
2820 let isReturn = 1, isTerminator = 1, isBarrier = 1, isPredicable = 1,
2821 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0 in {
2822 let validSubTargets = HasV4SubT in
2823 def DEALLOC_RET_V4 : LD0Inst<(outs), (ins),
2829 // Restore registers and dealloc return function call.
2830 let isCall = 1, isBarrier = 1, isReturn = 1, isTerminator = 1,
2831 Defs = [R29, R30, R31, PC] in {
2832 let validSubTargets = HasV4SubT in
2833 def RESTORE_DEALLOC_RET_JMP_V4 : JInst<(outs),
2834 (ins calltarget:$dst),
2840 // Restore registers and dealloc frame before a tail call.
2841 let isCall = 1, isBarrier = 1,
2842 Defs = [R29, R30, R31, PC] in {
2843 let validSubTargets = HasV4SubT in
2844 def RESTORE_DEALLOC_BEFORE_TAILCALL_V4 : JInst<(outs),
2845 (ins calltarget:$dst),
2851 // Save registers function call.
2852 let isCall = 1, isBarrier = 1,
2853 Uses = [R29, R31] in {
2854 def SAVE_REGISTERS_CALL_V4 : JInst<(outs),
2855 (ins calltarget:$dst),
2856 "call $dst // Save_calle_saved_registers",
2861 // if (Ps) dealloc_return
2862 let isReturn = 1, isTerminator = 1,
2863 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
2864 isPredicated = 1 in {
2865 let validSubTargets = HasV4SubT in
2866 def DEALLOC_RET_cPt_V4 : LD0Inst<(outs),
2867 (ins PredRegs:$src1),
2868 "if ($src1) dealloc_return",
2873 // if (!Ps) dealloc_return
2874 let isReturn = 1, isTerminator = 1,
2875 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
2876 isPredicated = 1, isPredicatedFalse = 1 in {
2877 let validSubTargets = HasV4SubT in
2878 def DEALLOC_RET_cNotPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
2879 "if (!$src1) dealloc_return",
2884 // if (Ps.new) dealloc_return:nt
2885 let isReturn = 1, isTerminator = 1,
2886 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
2887 isPredicated = 1 in {
2888 let validSubTargets = HasV4SubT in
2889 def DEALLOC_RET_cdnPnt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
2890 "if ($src1.new) dealloc_return:nt",
2895 // if (!Ps.new) dealloc_return:nt
2896 let isReturn = 1, isTerminator = 1,
2897 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
2898 isPredicated = 1, isPredicatedFalse = 1 in {
2899 let validSubTargets = HasV4SubT in
2900 def DEALLOC_RET_cNotdnPnt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
2901 "if (!$src1.new) dealloc_return:nt",
2906 // if (Ps.new) dealloc_return:t
2907 let isReturn = 1, isTerminator = 1,
2908 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
2909 isPredicated = 1 in {
2910 let validSubTargets = HasV4SubT in
2911 def DEALLOC_RET_cdnPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
2912 "if ($src1.new) dealloc_return:t",
2917 // if (!Ps.new) dealloc_return:nt
2918 let isReturn = 1, isTerminator = 1,
2919 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
2920 isPredicated = 1, isPredicatedFalse = 1 in {
2921 let validSubTargets = HasV4SubT in
2922 def DEALLOC_RET_cNotdnPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
2923 "if (!$src1.new) dealloc_return:t",
2928 // Load/Store with absolute addressing mode
2931 multiclass ST_Abs_Predbase<string mnemonic, RegisterClass RC, bit isNot,
2933 let isPredicatedNew = isPredNew in
2934 def NAME#_V4 : STInst2<(outs),
2935 (ins PredRegs:$src1, u0AlwaysExt:$absaddr, RC: $src2),
2936 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
2937 ") ")#mnemonic#"(##$absaddr) = $src2",
2942 multiclass ST_Abs_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
2943 let isPredicatedFalse = PredNot in {
2944 defm _c#NAME : ST_Abs_Predbase<mnemonic, RC, PredNot, 0>;
2946 defm _cdn#NAME : ST_Abs_Predbase<mnemonic, RC, PredNot, 1>;
2950 let isNVStorable = 1, isExtended = 1, hasSideEffects = 0 in
2951 multiclass ST_Abs<string mnemonic, string CextOp, RegisterClass RC> {
2952 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
2953 let opExtendable = 0, isPredicable = 1 in
2954 def NAME#_V4 : STInst2<(outs),
2955 (ins u0AlwaysExt:$absaddr, RC:$src),
2956 mnemonic#"(##$absaddr) = $src",
2960 let opExtendable = 1, isPredicated = 1 in {
2961 defm Pt : ST_Abs_Pred<mnemonic, RC, 0>;
2962 defm NotPt : ST_Abs_Pred<mnemonic, RC, 1>;
2967 multiclass ST_Abs_Predbase_nv<string mnemonic, RegisterClass RC, bit isNot,
2969 let isPredicatedNew = isPredNew in
2970 def NAME#_nv_V4 : NVInst_V4<(outs),
2971 (ins PredRegs:$src1, u0AlwaysExt:$absaddr, RC: $src2),
2972 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
2973 ") ")#mnemonic#"(##$absaddr) = $src2.new",
2978 multiclass ST_Abs_Pred_nv<string mnemonic, RegisterClass RC, bit PredNot> {
2979 let isPredicatedFalse = PredNot in {
2980 defm _c#NAME : ST_Abs_Predbase_nv<mnemonic, RC, PredNot, 0>;
2982 defm _cdn#NAME : ST_Abs_Predbase_nv<mnemonic, RC, PredNot, 1>;
2986 let mayStore = 1, isNVStore = 1, isExtended = 1, hasSideEffects = 0 in
2987 multiclass ST_Abs_nv<string mnemonic, string CextOp, RegisterClass RC> {
2988 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
2989 let opExtendable = 0, isPredicable = 1 in
2990 def NAME#_nv_V4 : NVInst_V4<(outs),
2991 (ins u0AlwaysExt:$absaddr, RC:$src),
2992 mnemonic#"(##$absaddr) = $src.new",
2996 let opExtendable = 1, isPredicated = 1 in {
2997 defm Pt : ST_Abs_Pred_nv<mnemonic, RC, 0>;
2998 defm NotPt : ST_Abs_Pred_nv<mnemonic, RC, 1>;
3003 let addrMode = Absolute in {
3004 let accessSize = ByteAccess in
3005 defm STrib_abs : ST_Abs<"memb", "STrib", IntRegs>,
3006 ST_Abs_nv<"memb", "STrib", IntRegs>, AddrModeRel;
3008 let accessSize = HalfWordAccess in
3009 defm STrih_abs : ST_Abs<"memh", "STrih", IntRegs>,
3010 ST_Abs_nv<"memh", "STrih", IntRegs>, AddrModeRel;
3012 let accessSize = WordAccess in
3013 defm STriw_abs : ST_Abs<"memw", "STriw", IntRegs>,
3014 ST_Abs_nv<"memw", "STriw", IntRegs>, AddrModeRel;
3016 let accessSize = DoubleWordAccess, isNVStorable = 0 in
3017 defm STrid_abs : ST_Abs<"memd", "STrid", DoubleRegs>, AddrModeRel;
3020 let Predicates = [HasV4T], AddedComplexity = 30 in {
3021 def : Pat<(truncstorei8 (i32 IntRegs:$src1),
3022 (HexagonCONST32 tglobaladdr:$absaddr)),
3023 (STrib_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
3025 def : Pat<(truncstorei16 (i32 IntRegs:$src1),
3026 (HexagonCONST32 tglobaladdr:$absaddr)),
3027 (STrih_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
3029 def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32 tglobaladdr:$absaddr)),
3030 (STriw_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
3032 def : Pat<(store (i64 DoubleRegs:$src1),
3033 (HexagonCONST32 tglobaladdr:$absaddr)),
3034 (STrid_abs_V4 tglobaladdr: $absaddr, DoubleRegs: $src1)>;
3037 //===----------------------------------------------------------------------===//
3038 // multiclass for store instructions with GP-relative addressing mode.
3039 // mem[bhwd](#global)=Rt
3040 // if ([!]Pv[.new]) mem[bhwd](##global) = Rt
3041 //===----------------------------------------------------------------------===//
3042 let mayStore = 1, isNVStorable = 1 in
3043 multiclass ST_GP<string mnemonic, string BaseOp, RegisterClass RC> {
3044 let BaseOpcode = BaseOp, isPredicable = 1 in
3045 def NAME#_V4 : STInst2<(outs),
3046 (ins globaladdress:$global, RC:$src),
3047 mnemonic#"(#$global) = $src",
3050 // When GP-relative instructions are predicated, their addressing mode is
3051 // changed to absolute and they are always constant extended.
3052 let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1,
3053 isPredicated = 1 in {
3054 defm Pt : ST_Abs_Pred <mnemonic, RC, 0>;
3055 defm NotPt : ST_Abs_Pred <mnemonic, RC, 1>;
3059 let mayStore = 1, isNVStore = 1 in
3060 multiclass ST_GP_nv<string mnemonic, string BaseOp, RegisterClass RC> {
3061 let BaseOpcode = BaseOp, isPredicable = 1 in
3062 def NAME#_nv_V4 : NVInst_V4<(outs),
3063 (ins u0AlwaysExt:$global, RC:$src),
3064 mnemonic#"(#$global) = $src.new",
3068 // When GP-relative instructions are predicated, their addressing mode is
3069 // changed to absolute and they are always constant extended.
3070 let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1,
3071 isPredicated = 1 in {
3072 defm Pt : ST_Abs_Pred_nv<mnemonic, RC, 0>;
3073 defm NotPt : ST_Abs_Pred_nv<mnemonic, RC, 1>;
3077 let validSubTargets = HasV4SubT, hasSideEffects = 0 in {
3078 let isNVStorable = 0 in
3079 defm STd_GP : ST_GP <"memd", "STd_GP", DoubleRegs>, PredNewRel;
3081 defm STb_GP : ST_GP<"memb", "STb_GP", IntRegs>,
3082 ST_GP_nv<"memb", "STb_GP", IntRegs>, NewValueRel;
3083 defm STh_GP : ST_GP<"memh", "STh_GP", IntRegs>,
3084 ST_GP_nv<"memh", "STh_GP", IntRegs>, NewValueRel;
3085 defm STw_GP : ST_GP<"memw", "STw_GP", IntRegs>,
3086 ST_GP_nv<"memw", "STw_GP", IntRegs>, NewValueRel;
3089 // 64 bit atomic store
3090 def : Pat <(atomic_store_64 (HexagonCONST32_GP tglobaladdr:$global),
3091 (i64 DoubleRegs:$src1)),
3092 (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>,
3095 // Map from store(globaladdress) -> memd(#foo)
3096 let AddedComplexity = 100 in
3097 def : Pat <(store (i64 DoubleRegs:$src1),
3098 (HexagonCONST32_GP tglobaladdr:$global)),
3099 (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>;
3101 // 8 bit atomic store
3102 def : Pat < (atomic_store_8 (HexagonCONST32_GP tglobaladdr:$global),
3103 (i32 IntRegs:$src1)),
3104 (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3106 // Map from store(globaladdress) -> memb(#foo)
3107 let AddedComplexity = 100 in
3108 def : Pat<(truncstorei8 (i32 IntRegs:$src1),
3109 (HexagonCONST32_GP tglobaladdr:$global)),
3110 (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3112 // Map from "i1 = constant<-1>; memw(CONST32(#foo)) = i1"
3113 // to "r0 = 1; memw(#foo) = r0"
3114 let AddedComplexity = 100 in
3115 def : Pat<(store (i1 -1), (HexagonCONST32_GP tglobaladdr:$global)),
3116 (STb_GP_V4 tglobaladdr:$global, (A2_tfrsi 1))>;
3118 def : Pat<(atomic_store_16 (HexagonCONST32_GP tglobaladdr:$global),
3119 (i32 IntRegs:$src1)),
3120 (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3122 // Map from store(globaladdress) -> memh(#foo)
3123 let AddedComplexity = 100 in
3124 def : Pat<(truncstorei16 (i32 IntRegs:$src1),
3125 (HexagonCONST32_GP tglobaladdr:$global)),
3126 (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3128 // 32 bit atomic store
3129 def : Pat<(atomic_store_32 (HexagonCONST32_GP tglobaladdr:$global),
3130 (i32 IntRegs:$src1)),
3131 (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3133 // Map from store(globaladdress) -> memw(#foo)
3134 let AddedComplexity = 100 in
3135 def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)),
3136 (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3138 //===----------------------------------------------------------------------===//
3139 // Multiclass for the load instructions with absolute addressing mode.
3140 //===----------------------------------------------------------------------===//
3141 multiclass LD_Abs_Predbase<string mnemonic, RegisterClass RC, bit isNot,
3143 let isPredicatedNew = isPredNew in
3144 def NAME : LDInst2<(outs RC:$dst),
3145 (ins PredRegs:$src1, u0AlwaysExt:$absaddr),
3146 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
3147 ") ")#"$dst = "#mnemonic#"(##$absaddr)",
3152 multiclass LD_Abs_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
3153 let isPredicatedFalse = PredNot in {
3154 defm _c#NAME : LD_Abs_Predbase<mnemonic, RC, PredNot, 0>;
3156 defm _cdn#NAME : LD_Abs_Predbase<mnemonic, RC, PredNot, 1>;
3160 let isExtended = 1, hasSideEffects = 0 in
3161 multiclass LD_Abs<string mnemonic, string CextOp, RegisterClass RC> {
3162 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
3163 let opExtendable = 1, isPredicable = 1 in
3164 def NAME#_V4 : LDInst2<(outs RC:$dst),
3165 (ins u0AlwaysExt:$absaddr),
3166 "$dst = "#mnemonic#"(##$absaddr)",
3170 let opExtendable = 2, isPredicated = 1 in {
3171 defm Pt_V4 : LD_Abs_Pred<mnemonic, RC, 0>;
3172 defm NotPt_V4 : LD_Abs_Pred<mnemonic, RC, 1>;
3177 let addrMode = Absolute in {
3178 let accessSize = ByteAccess in {
3179 defm LDrib_abs : LD_Abs<"memb", "LDrib", IntRegs>, AddrModeRel;
3180 defm LDriub_abs : LD_Abs<"memub", "LDriub", IntRegs>, AddrModeRel;
3182 let accessSize = HalfWordAccess in {
3183 defm LDrih_abs : LD_Abs<"memh", "LDrih", IntRegs>, AddrModeRel;
3184 defm LDriuh_abs : LD_Abs<"memuh", "LDriuh", IntRegs>, AddrModeRel;
3186 let accessSize = WordAccess in
3187 defm LDriw_abs : LD_Abs<"memw", "LDriw", IntRegs>, AddrModeRel;
3189 let accessSize = DoubleWordAccess in
3190 defm LDrid_abs : LD_Abs<"memd", "LDrid", DoubleRegs>, AddrModeRel;
3193 let Predicates = [HasV4T], AddedComplexity = 30 in {
3194 def : Pat<(i32 (load (HexagonCONST32 tglobaladdr:$absaddr))),
3195 (LDriw_abs_V4 tglobaladdr: $absaddr)>;
3197 def : Pat<(i32 (sextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))),
3198 (LDrib_abs_V4 tglobaladdr:$absaddr)>;
3200 def : Pat<(i32 (zextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))),
3201 (LDriub_abs_V4 tglobaladdr:$absaddr)>;
3203 def : Pat<(i32 (sextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
3204 (LDrih_abs_V4 tglobaladdr:$absaddr)>;
3206 def : Pat<(i32 (zextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
3207 (LDriuh_abs_V4 tglobaladdr:$absaddr)>;
3210 //===----------------------------------------------------------------------===//
3211 // multiclass for load instructions with GP-relative addressing mode.
3212 // Rx=mem[bhwd](##global)
3213 // if ([!]Pv[.new]) Rx=mem[bhwd](##global)
3214 //===----------------------------------------------------------------------===//
3215 let hasSideEffects = 0, validSubTargets = HasV4SubT in
3216 multiclass LD_GP<string mnemonic, string BaseOp, RegisterClass RC> {
3217 let BaseOpcode = BaseOp in {
3218 let isPredicable = 1 in
3219 def NAME#_V4 : LDInst2<(outs RC:$dst),
3220 (ins globaladdress:$global),
3221 "$dst = "#mnemonic#"(#$global)",
3224 let isExtended = 1, opExtendable = 2, isPredicated = 1 in {
3225 defm Pt_V4 : LD_Abs_Pred<mnemonic, RC, 0>;
3226 defm NotPt_V4 : LD_Abs_Pred<mnemonic, RC, 1>;
3231 defm LDd_GP : LD_GP<"memd", "LDd_GP", DoubleRegs>, PredNewRel;
3232 defm LDb_GP : LD_GP<"memb", "LDb_GP", IntRegs>, PredNewRel;
3233 defm LDub_GP : LD_GP<"memub", "LDub_GP", IntRegs>, PredNewRel;
3234 defm LDh_GP : LD_GP<"memh", "LDh_GP", IntRegs>, PredNewRel;
3235 defm LDuh_GP : LD_GP<"memuh", "LDuh_GP", IntRegs>, PredNewRel;
3236 defm LDw_GP : LD_GP<"memw", "LDw_GP", IntRegs>, PredNewRel;
3238 def : Pat <(atomic_load_64 (HexagonCONST32_GP tglobaladdr:$global)),
3239 (i64 (LDd_GP_V4 tglobaladdr:$global))>;
3241 def : Pat <(atomic_load_32 (HexagonCONST32_GP tglobaladdr:$global)),
3242 (i32 (LDw_GP_V4 tglobaladdr:$global))>;
3244 def : Pat <(atomic_load_16 (HexagonCONST32_GP tglobaladdr:$global)),
3245 (i32 (LDuh_GP_V4 tglobaladdr:$global))>;
3247 def : Pat <(atomic_load_8 (HexagonCONST32_GP tglobaladdr:$global)),
3248 (i32 (LDub_GP_V4 tglobaladdr:$global))>;
3250 // Map from load(globaladdress) -> memw(#foo + 0)
3251 let AddedComplexity = 100 in
3252 def : Pat <(i64 (load (HexagonCONST32_GP tglobaladdr:$global))),
3253 (i64 (LDd_GP_V4 tglobaladdr:$global))>;
3255 // Map from Pd = load(globaladdress) -> Rd = memb(globaladdress), Pd = Rd
3256 let AddedComplexity = 100 in
3257 def : Pat <(i1 (load (HexagonCONST32_GP tglobaladdr:$global))),
3258 (i1 (C2_tfrrp (i32 (LDb_GP_V4 tglobaladdr:$global))))>;
3260 // When the Interprocedural Global Variable optimizer realizes that a certain
3261 // global variable takes only two constant values, it shrinks the global to
3262 // a boolean. Catch those loads here in the following 3 patterns.
3263 let AddedComplexity = 100 in
3264 def : Pat <(i32 (extloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3265 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3267 let AddedComplexity = 100 in
3268 def : Pat <(i32 (sextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3269 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3271 // Map from load(globaladdress) -> memb(#foo)
3272 let AddedComplexity = 100 in
3273 def : Pat <(i32 (extloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3274 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3276 // Map from load(globaladdress) -> memb(#foo)
3277 let AddedComplexity = 100 in
3278 def : Pat <(i32 (sextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3279 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3281 let AddedComplexity = 100 in
3282 def : Pat <(i32 (zextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3283 (i32 (LDub_GP_V4 tglobaladdr:$global))>;
3285 // Map from load(globaladdress) -> memub(#foo)
3286 let AddedComplexity = 100 in
3287 def : Pat <(i32 (zextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3288 (i32 (LDub_GP_V4 tglobaladdr:$global))>;
3290 // Map from load(globaladdress) -> memh(#foo)
3291 let AddedComplexity = 100 in
3292 def : Pat <(i32 (extloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3293 (i32 (LDh_GP_V4 tglobaladdr:$global))>;
3295 // Map from load(globaladdress) -> memh(#foo)
3296 let AddedComplexity = 100 in
3297 def : Pat <(i32 (sextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3298 (i32 (LDh_GP_V4 tglobaladdr:$global))>;
3300 // Map from load(globaladdress) -> memuh(#foo)
3301 let AddedComplexity = 100 in
3302 def : Pat <(i32 (zextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3303 (i32 (LDuh_GP_V4 tglobaladdr:$global))>;
3305 // Map from load(globaladdress) -> memw(#foo)
3306 let AddedComplexity = 100 in
3307 def : Pat <(i32 (load (HexagonCONST32_GP tglobaladdr:$global))),
3308 (i32 (LDw_GP_V4 tglobaladdr:$global))>;
3311 // Transfer global address into a register
3312 let isExtended = 1, opExtendable = 1, AddedComplexity=50, isMoveImm = 1,
3313 isAsCheapAsAMove = 1, isReMaterializable = 1, validSubTargets = HasV4SubT in
3314 def TFRI_V4 : ALU32_ri<(outs IntRegs:$dst), (ins s16Ext:$src1),
3316 [(set IntRegs:$dst, (HexagonCONST32 tglobaladdr:$src1))]>,
3319 // Transfer a block address into a register
3320 def : Pat<(HexagonCONST32_GP tblockaddress:$src1),
3321 (TFRI_V4 tblockaddress:$src1)>,
3324 let isExtended = 1, opExtendable = 2, AddedComplexity=50,
3325 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3326 def TFRI_cPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3327 (ins PredRegs:$src1, s16Ext:$src2),
3328 "if($src1) $dst = #$src2",
3332 let isExtended = 1, opExtendable = 2, AddedComplexity=50, isPredicatedFalse = 1,
3333 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3334 def TFRI_cNotPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3335 (ins PredRegs:$src1, s16Ext:$src2),
3336 "if(!$src1) $dst = #$src2",
3340 let isExtended = 1, opExtendable = 2, AddedComplexity=50,
3341 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3342 def TFRI_cdnPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3343 (ins PredRegs:$src1, s16Ext:$src2),
3344 "if($src1.new) $dst = #$src2",
3348 let isExtended = 1, opExtendable = 2, AddedComplexity=50, isPredicatedFalse = 1,
3349 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3350 def TFRI_cdnNotPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3351 (ins PredRegs:$src1, s16Ext:$src2),
3352 "if(!$src1.new) $dst = #$src2",
3356 let AddedComplexity = 50, Predicates = [HasV4T] in
3357 def : Pat<(HexagonCONST32_GP tglobaladdr:$src1),
3358 (TFRI_V4 tglobaladdr:$src1)>,
3362 // Load - Indirect with long offset: These instructions take global address
3364 let isExtended = 1, opExtendable = 3, AddedComplexity = 40,
3365 validSubTargets = HasV4SubT in
3366 def LDrid_ind_lo_V4 : LDInst<(outs DoubleRegs:$dst),
3367 (ins IntRegs:$src1, u2Imm:$src2, globaladdressExt:$offset),
3368 "$dst=memd($src1<<#$src2+##$offset)",
3369 [(set (i64 DoubleRegs:$dst),
3370 (load (add (shl IntRegs:$src1, u2ImmPred:$src2),
3371 (HexagonCONST32 tglobaladdr:$offset))))]>,
3374 let AddedComplexity = 40 in
3375 multiclass LD_indirect_lo<string OpcStr, PatFrag OpNode> {
3376 let isExtended = 1, opExtendable = 3, validSubTargets = HasV4SubT in
3377 def _lo_V4 : LDInst<(outs IntRegs:$dst),
3378 (ins IntRegs:$src1, u2Imm:$src2, globaladdressExt:$offset),
3379 !strconcat("$dst = ",
3380 !strconcat(OpcStr, "($src1<<#$src2+##$offset)")),
3382 (i32 (OpNode (add (shl IntRegs:$src1, u2ImmPred:$src2),
3383 (HexagonCONST32 tglobaladdr:$offset)))))]>,
3387 defm LDrib_ind : LD_indirect_lo<"memb", sextloadi8>;
3388 defm LDriub_ind : LD_indirect_lo<"memub", zextloadi8>;
3389 defm LDriub_ind_anyext : LD_indirect_lo<"memub", extloadi8>;
3390 defm LDrih_ind : LD_indirect_lo<"memh", sextloadi16>;
3391 defm LDriuh_ind : LD_indirect_lo<"memuh", zextloadi16>;
3392 defm LDriuh_ind_anyext : LD_indirect_lo<"memuh", extloadi16>;
3393 defm LDriw_ind : LD_indirect_lo<"memw", load>;
3395 let AddedComplexity = 40 in
3396 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1,
3397 (NumUsesBelowThresCONST32 tglobaladdr:$offset)))),
3398 (i32 (LDrib_ind_lo_V4 IntRegs:$src1, 0, tglobaladdr:$offset))>,
3401 let AddedComplexity = 40 in
3402 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1,
3403 (NumUsesBelowThresCONST32 tglobaladdr:$offset)))),
3404 (i32 (LDriub_ind_lo_V4 IntRegs:$src1, 0, tglobaladdr:$offset))>,
3407 let Predicates = [HasV4T], AddedComplexity = 30 in {
3408 def : Pat<(truncstorei8 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3409 (STrib_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3411 def : Pat<(truncstorei16 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3412 (STrih_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3414 def : Pat<(store (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3415 (STriw_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3418 let Predicates = [HasV4T], AddedComplexity = 30 in {
3419 def : Pat<(i32 (load u0AlwaysExtPred:$src)),
3420 (LDriw_abs_V4 u0AlwaysExtPred:$src)>;
3422 def : Pat<(i32 (sextloadi8 u0AlwaysExtPred:$src)),
3423 (LDrib_abs_V4 u0AlwaysExtPred:$src)>;
3425 def : Pat<(i32 (zextloadi8 u0AlwaysExtPred:$src)),
3426 (LDriub_abs_V4 u0AlwaysExtPred:$src)>;
3428 def : Pat<(i32 (sextloadi16 u0AlwaysExtPred:$src)),
3429 (LDrih_abs_V4 u0AlwaysExtPred:$src)>;
3431 def : Pat<(i32 (zextloadi16 u0AlwaysExtPred:$src)),
3432 (LDriuh_abs_V4 u0AlwaysExtPred:$src)>;
3435 // Indexed store word - global address.
3436 // memw(Rs+#u6:2)=#S8
3437 let AddedComplexity = 10 in
3438 def STriw_offset_ext_V4 : STInst<(outs),
3439 (ins IntRegs:$src1, u6_2Imm:$src2, globaladdress:$src3),
3440 "memw($src1+#$src2) = ##$src3",
3441 [(store (HexagonCONST32 tglobaladdr:$src3),
3442 (add IntRegs:$src1, u6_2ImmPred:$src2))]>,
3445 def : Pat<(i64 (ctlz (i64 DoubleRegs:$src1))),
3446 (i64 (A4_combineir (i32 0), (i32 (CTLZ64_rr DoubleRegs:$src1))))>,
3449 def : Pat<(i64 (cttz (i64 DoubleRegs:$src1))),
3450 (i64 (A4_combineir (i32 0), (i32 (CTTZ64_rr DoubleRegs:$src1))))>,
3455 // We need a complexity of 120 here to override preceding handling of
3457 let Predicates = [HasV4T], AddedComplexity = 120 in {
3458 def: Pat <(i64 (extloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3459 (i64 (A4_combineir 0, (LDrib_abs_V4 tglobaladdr:$addr)))>;
3461 def: Pat <(i64 (zextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3462 (i64 (A4_combineir 0, (LDriub_abs_V4 tglobaladdr:$addr)))>;
3464 def: Pat <(i64 (sextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3465 (i64 (A2_sxtw (LDrib_abs_V4 tglobaladdr:$addr)))>;
3467 def: Pat <(i64 (extloadi8 FoldGlobalAddr:$addr)),
3468 (i64 (A4_combineir 0, (LDrib_abs_V4 FoldGlobalAddr:$addr)))>;
3470 def: Pat <(i64 (zextloadi8 FoldGlobalAddr:$addr)),
3471 (i64 (A4_combineir 0, (LDriub_abs_V4 FoldGlobalAddr:$addr)))>;
3473 def: Pat <(i64 (sextloadi8 FoldGlobalAddr:$addr)),
3474 (i64 (A2_sxtw (LDrib_abs_V4 FoldGlobalAddr:$addr)))>;
3477 // We need a complexity of 120 here to override preceding handling of
3479 let AddedComplexity = 120 in {
3480 def: Pat <(i64 (extloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3481 (i64 (A4_combineir 0, (LDrih_abs_V4 tglobaladdr:$addr)))>,
3484 def: Pat <(i64 (zextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3485 (i64 (A4_combineir 0, (LDriuh_abs_V4 tglobaladdr:$addr)))>,
3488 def: Pat <(i64 (sextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3489 (i64 (A2_sxtw (LDrih_abs_V4 tglobaladdr:$addr)))>,
3492 def: Pat <(i64 (extloadi16 FoldGlobalAddr:$addr)),
3493 (i64 (A4_combineir 0, (LDrih_abs_V4 FoldGlobalAddr:$addr)))>,
3496 def: Pat <(i64 (zextloadi16 FoldGlobalAddr:$addr)),
3497 (i64 (A4_combineir 0, (LDriuh_abs_V4 FoldGlobalAddr:$addr)))>,
3500 def: Pat <(i64 (sextloadi16 FoldGlobalAddr:$addr)),
3501 (i64 (A2_sxtw (LDrih_abs_V4 FoldGlobalAddr:$addr)))>,
3505 // We need a complexity of 120 here to override preceding handling of
3507 let AddedComplexity = 120 in {
3508 def: Pat <(i64 (extloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3509 (i64 (A4_combineir 0, (LDriw_abs_V4 tglobaladdr:$addr)))>,
3512 def: Pat <(i64 (zextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3513 (i64 (A4_combineir 0, (LDriw_abs_V4 tglobaladdr:$addr)))>,
3516 def: Pat <(i64 (sextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3517 (i64 (A2_sxtw (LDriw_abs_V4 tglobaladdr:$addr)))>,
3520 def: Pat <(i64 (extloadi32 FoldGlobalAddr:$addr)),
3521 (i64 (A4_combineir 0, (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
3524 def: Pat <(i64 (zextloadi32 FoldGlobalAddr:$addr)),
3525 (i64 (A4_combineir 0, (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
3528 def: Pat <(i64 (sextloadi32 FoldGlobalAddr:$addr)),
3529 (i64 (A2_sxtw (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
3533 // Indexed store double word - global address.
3534 // memw(Rs+#u6:2)=#S8
3535 let AddedComplexity = 10 in
3536 def STrih_offset_ext_V4 : STInst<(outs),
3537 (ins IntRegs:$src1, u6_1Imm:$src2, globaladdress:$src3),
3538 "memh($src1+#$src2) = ##$src3",
3539 [(truncstorei16 (HexagonCONST32 tglobaladdr:$src3),
3540 (add IntRegs:$src1, u6_1ImmPred:$src2))]>,
3542 // Map from store(globaladdress + x) -> memd(#foo + x)
3543 let AddedComplexity = 100 in
3544 def : Pat<(store (i64 DoubleRegs:$src1),
3545 FoldGlobalAddrGP:$addr),
3546 (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
3549 def : Pat<(atomic_store_64 FoldGlobalAddrGP:$addr,
3550 (i64 DoubleRegs:$src1)),
3551 (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
3554 // Map from store(globaladdress + x) -> memb(#foo + x)
3555 let AddedComplexity = 100 in
3556 def : Pat<(truncstorei8 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
3557 (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3560 def : Pat<(atomic_store_8 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
3561 (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3564 // Map from store(globaladdress + x) -> memh(#foo + x)
3565 let AddedComplexity = 100 in
3566 def : Pat<(truncstorei16 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
3567 (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3570 def : Pat<(atomic_store_16 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
3571 (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3574 // Map from store(globaladdress + x) -> memw(#foo + x)
3575 let AddedComplexity = 100 in
3576 def : Pat<(store (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
3577 (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3580 def : Pat<(atomic_store_32 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
3581 (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3584 // Map from load(globaladdress + x) -> memd(#foo + x)
3585 let AddedComplexity = 100 in
3586 def : Pat<(i64 (load FoldGlobalAddrGP:$addr)),
3587 (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>,
3590 def : Pat<(atomic_load_64 FoldGlobalAddrGP:$addr),
3591 (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>,
3594 // Map from load(globaladdress + x) -> memb(#foo + x)
3595 let AddedComplexity = 100 in
3596 def : Pat<(i32 (extloadi8 FoldGlobalAddrGP:$addr)),
3597 (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>,
3600 // Map from load(globaladdress + x) -> memb(#foo + x)
3601 let AddedComplexity = 100 in
3602 def : Pat<(i32 (sextloadi8 FoldGlobalAddrGP:$addr)),
3603 (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>,
3606 //let AddedComplexity = 100 in
3607 let AddedComplexity = 100 in
3608 def : Pat<(i32 (extloadi16 FoldGlobalAddrGP:$addr)),
3609 (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>,
3612 // Map from load(globaladdress + x) -> memh(#foo + x)
3613 let AddedComplexity = 100 in
3614 def : Pat<(i32 (sextloadi16 FoldGlobalAddrGP:$addr)),
3615 (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>,
3618 // Map from load(globaladdress + x) -> memuh(#foo + x)
3619 let AddedComplexity = 100 in
3620 def : Pat<(i32 (zextloadi16 FoldGlobalAddrGP:$addr)),
3621 (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>,
3624 def : Pat<(atomic_load_16 FoldGlobalAddrGP:$addr),
3625 (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>,
3628 // Map from load(globaladdress + x) -> memub(#foo + x)
3629 let AddedComplexity = 100 in
3630 def : Pat<(i32 (zextloadi8 FoldGlobalAddrGP:$addr)),
3631 (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>,
3634 def : Pat<(atomic_load_8 FoldGlobalAddrGP:$addr),
3635 (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>,
3638 // Map from load(globaladdress + x) -> memw(#foo + x)
3639 let AddedComplexity = 100 in
3640 def : Pat<(i32 (load FoldGlobalAddrGP:$addr)),
3641 (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>,
3644 def : Pat<(atomic_load_32 FoldGlobalAddrGP:$addr),
3645 (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>,