1 //=- HexagonInstrInfoV4.td - Target Desc. for Hexagon Target -*- tablegen -*-=//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the Hexagon V4 instructions in TableGen format.
12 //===----------------------------------------------------------------------===//
14 let hasSideEffects = 0 in
15 class T_Immext<Operand ImmType>
16 : EXTENDERInst<(outs), (ins ImmType:$imm),
17 "immext(#$imm)", []> {
21 let Inst{27-16} = imm{31-20};
22 let Inst{13-0} = imm{19-6};
25 def A4_ext : T_Immext<u26_6Imm>;
26 let isCodeGenOnly = 1 in {
28 def A4_ext_b : T_Immext<brtarget>;
30 def A4_ext_c : T_Immext<calltarget>;
31 def A4_ext_g : T_Immext<globaladdress>;
34 // Fold (add (CONST32 tglobaladdr:$addr) <offset>) into a global address.
35 def FoldGlobalAddr : ComplexPattern<i32, 1, "foldGlobalAddress", [], []>;
37 // Fold (add (CONST32_GP tglobaladdr:$addr) <offset>) into a global address.
38 def FoldGlobalAddrGP : ComplexPattern<i32, 1, "foldGlobalAddressGP", [], []>;
40 def NumUsesBelowThresCONST32 : PatFrag<(ops node:$addr),
41 (HexagonCONST32 node:$addr), [{
42 return hasNumUsesBelowThresGA(N->getOperand(0).getNode());
45 // Hexagon V4 Architecture spec defines 8 instruction classes:
46 // LD ST ALU32 XTYPE J JR MEMOP NV CR SYSTEM(system is not implemented in the
50 // ========================================
51 // Loads (8/16/32/64 bit)
55 // ========================================
56 // Stores (8/16/32/64 bit)
59 // ALU32 Instructions:
60 // ========================================
61 // Arithmetic / Logical (32 bit)
64 // XTYPE Instructions (32/64 bit):
65 // ========================================
66 // Arithmetic, Logical, Bit Manipulation
67 // Multiply (Integer, Fractional, Complex)
68 // Permute / Vector Permute Operations
69 // Predicate Operations
70 // Shift / Shift with Add/Sub/Logical
72 // Vector Halfword (ALU, Shift, Multiply)
73 // Vector Word (ALU, Shift)
76 // ========================================
77 // Jump/Call PC-relative
80 // ========================================
83 // MEMOP Instructions:
84 // ========================================
85 // Operation on memory (8/16/32 bit)
88 // ========================================
93 // ========================================
94 // Control-Register Transfers
95 // Hardware Loop Setup
96 // Predicate Logicals & Reductions
98 // SYSTEM Instructions (not implemented in the compiler):
99 // ========================================
105 //===----------------------------------------------------------------------===//
107 //===----------------------------------------------------------------------===//
109 class T_ALU32_3op_not<string mnemonic, bits<3> MajOp, bits<3> MinOp,
111 : T_ALU32_3op<mnemonic, MajOp, MinOp, OpsRev, 0> {
112 let AsmString = "$Rd = "#mnemonic#"($Rs, ~$Rt)";
115 let BaseOpcode = "andn_rr", CextOpcode = "andn", isCodeGenOnly = 0 in
116 def A4_andn : T_ALU32_3op_not<"and", 0b001, 0b100, 1>;
117 let BaseOpcode = "orn_rr", CextOpcode = "orn", isCodeGenOnly = 0 in
118 def A4_orn : T_ALU32_3op_not<"or", 0b001, 0b101, 1>;
120 let CextOpcode = "rcmp.eq", isCodeGenOnly = 0 in
121 def A4_rcmpeq : T_ALU32_3op<"cmp.eq", 0b011, 0b010, 0, 1>;
122 let CextOpcode = "!rcmp.eq", isCodeGenOnly = 0 in
123 def A4_rcmpneq : T_ALU32_3op<"!cmp.eq", 0b011, 0b011, 0, 1>;
125 let isCodeGenOnly = 0 in {
126 def C4_cmpneq : T_ALU32_3op_cmp<"!cmp.eq", 0b00, 1, 1>;
127 def C4_cmplte : T_ALU32_3op_cmp<"!cmp.gt", 0b10, 1, 0>;
128 def C4_cmplteu : T_ALU32_3op_cmp<"!cmp.gtu", 0b11, 1, 0>;
131 // Pats for instruction selection.
133 // A class to embed the usual comparison patfrags within a zext to i32.
134 // The seteq/setne frags use "lhs" and "rhs" as operands, so use the same
135 // names, or else the frag's "body" won't match the operands.
136 class CmpInReg<PatFrag Op>
137 : PatFrag<(ops node:$lhs, node:$rhs),(i32 (zext (i1 Op.Fragment)))>;
139 def: T_cmp32_rr_pat<A4_rcmpeq, CmpInReg<seteq>, i32>;
140 def: T_cmp32_rr_pat<A4_rcmpneq, CmpInReg<setne>, i32>;
142 class T_CMP_rrbh<string mnemonic, bits<3> MinOp, bit IsComm>
143 : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt),
144 "$Pd = "#mnemonic#"($Rs, $Rt)", [], "", S_3op_tc_2early_SLOT23>,
146 let validSubTargets = HasV4SubT;
147 let InputType = "reg";
148 let CextOpcode = mnemonic;
150 let isCommutable = IsComm;
151 let hasSideEffects = 0;
158 let Inst{27-21} = 0b0111110;
159 let Inst{20-16} = Rs;
161 let Inst{7-5} = MinOp;
165 let isCodeGenOnly = 0 in {
166 def A4_cmpbeq : T_CMP_rrbh<"cmpb.eq", 0b110, 1>;
167 def A4_cmpbgt : T_CMP_rrbh<"cmpb.gt", 0b010, 0>;
168 def A4_cmpbgtu : T_CMP_rrbh<"cmpb.gtu", 0b111, 0>;
169 def A4_cmpheq : T_CMP_rrbh<"cmph.eq", 0b011, 1>;
170 def A4_cmphgt : T_CMP_rrbh<"cmph.gt", 0b100, 0>;
171 def A4_cmphgtu : T_CMP_rrbh<"cmph.gtu", 0b101, 0>;
174 class T_CMP_ribh<string mnemonic, bits<2> MajOp, bit IsHalf, bit IsComm,
175 Operand ImmType, bit IsImmExt, bit IsImmSigned, int ImmBits>
176 : ALU64Inst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, ImmType:$Imm),
177 "$Pd = "#mnemonic#"($Rs, #$Imm)", [], "", ALU64_tc_2early_SLOT23>,
179 let validSubTargets = HasV4SubT;
180 let InputType = "imm";
181 let CextOpcode = mnemonic;
183 let isCommutable = IsComm;
184 let hasSideEffects = 0;
185 let isExtendable = IsImmExt;
186 let opExtendable = !if (IsImmExt, 2, 0);
187 let isExtentSigned = IsImmSigned;
188 let opExtentBits = ImmBits;
195 let Inst{27-24} = 0b1101;
196 let Inst{22-21} = MajOp;
197 let Inst{20-16} = Rs;
198 let Inst{12-5} = Imm;
200 let Inst{3} = IsHalf;
204 let isCodeGenOnly = 0 in {
205 def A4_cmpbeqi : T_CMP_ribh<"cmpb.eq", 0b00, 0, 1, u8Imm, 0, 0, 8>;
206 def A4_cmpbgti : T_CMP_ribh<"cmpb.gt", 0b01, 0, 0, s8Imm, 0, 1, 8>;
207 def A4_cmpbgtui : T_CMP_ribh<"cmpb.gtu", 0b10, 0, 0, u7Ext, 1, 0, 7>;
208 def A4_cmpheqi : T_CMP_ribh<"cmph.eq", 0b00, 1, 1, s8Ext, 1, 1, 8>;
209 def A4_cmphgti : T_CMP_ribh<"cmph.gt", 0b01, 1, 0, s8Ext, 1, 1, 8>;
210 def A4_cmphgtui : T_CMP_ribh<"cmph.gtu", 0b10, 1, 0, u7Ext, 1, 0, 7>;
212 class T_RCMP_EQ_ri<string mnemonic, bit IsNeg>
213 : ALU32_ri<(outs IntRegs:$Rd), (ins IntRegs:$Rs, s8Ext:$s8),
214 "$Rd = "#mnemonic#"($Rs, #$s8)", [], "", ALU32_2op_tc_1_SLOT0123>,
216 let validSubTargets = HasV4SubT;
217 let InputType = "imm";
218 let CextOpcode = !if (IsNeg, "!rcmp.eq", "rcmp.eq");
219 let isExtendable = 1;
220 let opExtendable = 2;
221 let isExtentSigned = 1;
222 let opExtentBits = 8;
230 let Inst{27-24} = 0b0011;
232 let Inst{21} = IsNeg;
233 let Inst{20-16} = Rs;
239 let isCodeGenOnly = 0 in {
240 def A4_rcmpeqi : T_RCMP_EQ_ri<"cmp.eq", 0>;
241 def A4_rcmpneqi : T_RCMP_EQ_ri<"!cmp.eq", 1>;
244 def: Pat<(i32 (zext (i1 (seteq (i32 IntRegs:$Rs), s8ExtPred:$s8)))),
245 (A4_rcmpeqi IntRegs:$Rs, s8ExtPred:$s8)>;
246 def: Pat<(i32 (zext (i1 (setne (i32 IntRegs:$Rs), s8ExtPred:$s8)))),
247 (A4_rcmpneqi IntRegs:$Rs, s8ExtPred:$s8)>;
249 // Preserve the S2_tstbit_r generation
250 def: Pat<(i32 (zext (i1 (setne (i32 (and (i32 (shl 1, (i32 IntRegs:$src2))),
251 (i32 IntRegs:$src1))), 0)))),
252 (C2_muxii (S2_tstbit_r IntRegs:$src1, IntRegs:$src2), 1, 0)>;
255 //===----------------------------------------------------------------------===//
257 //===----------------------------------------------------------------------===//
260 //===----------------------------------------------------------------------===//
262 //===----------------------------------------------------------------------===//
264 // Combine a word and an immediate into a register pair.
265 let hasSideEffects = 0, isExtentSigned = 1, isExtendable = 1,
267 class T_Combine1 <bits<2> MajOp, dag ins, string AsmStr>
268 : ALU32Inst <(outs DoubleRegs:$Rdd), ins, AsmStr> {
274 let Inst{27-24} = 0b0011;
275 let Inst{22-21} = MajOp;
276 let Inst{20-16} = Rs;
282 let opExtendable = 2, isCodeGenOnly = 0 in
283 def A4_combineri : T_Combine1<0b00, (ins IntRegs:$Rs, s8Ext:$s8),
284 "$Rdd = combine($Rs, #$s8)">;
286 let opExtendable = 1, isCodeGenOnly = 0 in
287 def A4_combineir : T_Combine1<0b01, (ins s8Ext:$s8, IntRegs:$Rs),
288 "$Rdd = combine(#$s8, $Rs)">;
290 def HexagonWrapperCombineRI_V4 :
291 SDNode<"HexagonISD::WrapperCombineRI_V4", SDTHexagonI64I32I32>;
292 def HexagonWrapperCombineIR_V4 :
293 SDNode<"HexagonISD::WrapperCombineIR_V4", SDTHexagonI64I32I32>;
295 def : Pat <(HexagonWrapperCombineRI_V4 IntRegs:$r, s8ExtPred:$i),
296 (A4_combineri IntRegs:$r, s8ExtPred:$i)>,
299 def : Pat <(HexagonWrapperCombineIR_V4 s8ExtPred:$i, IntRegs:$r),
300 (A4_combineir s8ExtPred:$i, IntRegs:$r)>,
303 // A4_combineii: Set two small immediates.
304 let hasSideEffects = 0, isExtendable = 1, opExtentBits = 6, opExtendable = 2 in
305 def A4_combineii: ALU32Inst<(outs DoubleRegs:$Rdd), (ins s8Imm:$s8, u6Ext:$U6),
306 "$Rdd = combine(#$s8, #$U6)"> {
312 let Inst{27-23} = 0b11001;
313 let Inst{20-16} = U6{5-1};
314 let Inst{13} = U6{0};
319 //===----------------------------------------------------------------------===//
321 //===----------------------------------------------------------------------===//
323 //===----------------------------------------------------------------------===//
325 //===----------------------------------------------------------------------===//
326 //===----------------------------------------------------------------------===//
327 // Template class for load instructions with Absolute set addressing mode.
328 //===----------------------------------------------------------------------===//
329 let isExtended = 1, opExtendable = 2, hasSideEffects = 0,
330 validSubTargets = HasV4SubT, addrMode = AbsoluteSet in
331 class T_LD_abs_set<string mnemonic, RegisterClass RC>:
332 LDInst2<(outs RC:$dst1, IntRegs:$dst2),
333 (ins u0AlwaysExt:$addr),
334 "$dst1 = "#mnemonic#"($dst2=##$addr)",
338 def LDrid_abs_set_V4 : T_LD_abs_set <"memd", DoubleRegs>;
339 def LDrib_abs_set_V4 : T_LD_abs_set <"memb", IntRegs>;
340 def LDriub_abs_set_V4 : T_LD_abs_set <"memub", IntRegs>;
341 def LDrih_abs_set_V4 : T_LD_abs_set <"memh", IntRegs>;
342 def LDriw_abs_set_V4 : T_LD_abs_set <"memw", IntRegs>;
343 def LDriuh_abs_set_V4 : T_LD_abs_set <"memuh", IntRegs>;
345 //===----------------------------------------------------------------------===//
346 // Template classes for the non-predicated load instructions with
347 // base + register offset addressing mode
348 //===----------------------------------------------------------------------===//
349 class T_load_rr <string mnemonic, RegisterClass RC, bits<3> MajOp>:
350 LDInst<(outs RC:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$u2),
351 "$dst = "#mnemonic#"($src1 + $src2<<#$u2)",
352 [], "", V4LDST_tc_ld_SLOT01>, ImmRegShl, AddrModeRel {
360 let Inst{27-24} = 0b1010;
361 let Inst{23-21} = MajOp;
362 let Inst{20-16} = src1;
363 let Inst{12-8} = src2;
364 let Inst{13} = u2{1};
369 //===----------------------------------------------------------------------===//
370 // Template classes for the predicated load instructions with
371 // base + register offset addressing mode
372 //===----------------------------------------------------------------------===//
373 let isPredicated = 1 in
374 class T_pload_rr <string mnemonic, RegisterClass RC, bits<3> MajOp,
375 bit isNot, bit isPredNew>:
376 LDInst <(outs RC:$dst),
377 (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$u2),
378 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
379 ") ")#"$dst = "#mnemonic#"($src2+$src3<<#$u2)",
380 [], "", V4LDST_tc_ld_SLOT01>, AddrModeRel {
387 let isPredicatedFalse = isNot;
388 let isPredicatedNew = isPredNew;
392 let Inst{27-26} = 0b00;
393 let Inst{25} = isPredNew;
394 let Inst{24} = isNot;
395 let Inst{23-21} = MajOp;
396 let Inst{20-16} = src2;
397 let Inst{12-8} = src3;
398 let Inst{13} = u2{1};
400 let Inst{6-5} = src1;
404 //===----------------------------------------------------------------------===//
405 // multiclass for load instructions with base + register offset
407 //===----------------------------------------------------------------------===//
408 let hasSideEffects = 0, addrMode = BaseRegOffset in
409 multiclass ld_idxd_shl <string mnemonic, string CextOp, RegisterClass RC,
411 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl,
412 InputType = "reg" in {
413 let isPredicable = 1 in
414 def L4_#NAME#_rr : T_load_rr <mnemonic, RC, MajOp>;
417 def L4_p#NAME#t_rr : T_pload_rr <mnemonic, RC, MajOp, 0, 0>;
418 def L4_p#NAME#f_rr : T_pload_rr <mnemonic, RC, MajOp, 1, 0>;
421 def L4_p#NAME#tnew_rr : T_pload_rr <mnemonic, RC, MajOp, 0, 1>;
422 def L4_p#NAME#fnew_rr : T_pload_rr <mnemonic, RC, MajOp, 1, 1>;
426 let hasNewValue = 1, accessSize = ByteAccess, isCodeGenOnly = 0 in {
427 defm loadrb : ld_idxd_shl<"memb", "LDrib", IntRegs, 0b000>;
428 defm loadrub : ld_idxd_shl<"memub", "LDriub", IntRegs, 0b001>;
431 let hasNewValue = 1, accessSize = HalfWordAccess, isCodeGenOnly = 0 in {
432 defm loadrh : ld_idxd_shl<"memh", "LDrih", IntRegs, 0b010>;
433 defm loadruh : ld_idxd_shl<"memuh", "LDriuh", IntRegs, 0b011>;
436 let hasNewValue = 1, accessSize = WordAccess, isCodeGenOnly = 0 in
437 defm loadri : ld_idxd_shl<"memw", "LDriw", IntRegs, 0b100>;
439 let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
440 defm loadrd : ld_idxd_shl<"memd", "LDrid", DoubleRegs, 0b110>;
442 // 'def pats' for load instructions with base + register offset and non-zero
443 // immediate value. Immediate value is used to left-shift the second
445 let AddedComplexity = 40 in {
446 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1,
447 (shl IntRegs:$src2, u2ImmPred:$offset)))),
448 (L4_loadrb_rr IntRegs:$src1,
449 IntRegs:$src2, u2ImmPred:$offset)>,
452 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1,
453 (shl IntRegs:$src2, u2ImmPred:$offset)))),
454 (L4_loadrub_rr IntRegs:$src1,
455 IntRegs:$src2, u2ImmPred:$offset)>,
458 def : Pat <(i32 (extloadi8 (add IntRegs:$src1,
459 (shl IntRegs:$src2, u2ImmPred:$offset)))),
460 (L4_loadrub_rr IntRegs:$src1,
461 IntRegs:$src2, u2ImmPred:$offset)>,
464 def : Pat <(i32 (sextloadi16 (add IntRegs:$src1,
465 (shl IntRegs:$src2, u2ImmPred:$offset)))),
466 (L4_loadrh_rr IntRegs:$src1,
467 IntRegs:$src2, u2ImmPred:$offset)>,
470 def : Pat <(i32 (zextloadi16 (add IntRegs:$src1,
471 (shl IntRegs:$src2, u2ImmPred:$offset)))),
472 (L4_loadruh_rr IntRegs:$src1,
473 IntRegs:$src2, u2ImmPred:$offset)>,
476 def : Pat <(i32 (extloadi16 (add IntRegs:$src1,
477 (shl IntRegs:$src2, u2ImmPred:$offset)))),
478 (L4_loadruh_rr IntRegs:$src1,
479 IntRegs:$src2, u2ImmPred:$offset)>,
482 def : Pat <(i32 (load (add IntRegs:$src1,
483 (shl IntRegs:$src2, u2ImmPred:$offset)))),
484 (L4_loadri_rr IntRegs:$src1,
485 IntRegs:$src2, u2ImmPred:$offset)>,
488 def : Pat <(i64 (load (add IntRegs:$src1,
489 (shl IntRegs:$src2, u2ImmPred:$offset)))),
490 (L4_loadrd_rr IntRegs:$src1,
491 IntRegs:$src2, u2ImmPred:$offset)>,
496 // 'def pats' for load instruction base + register offset and
497 // zero immediate value.
498 let AddedComplexity = 10 in {
499 def : Pat <(i64 (load (add IntRegs:$src1, IntRegs:$src2))),
500 (L4_loadrd_rr IntRegs:$src1, IntRegs:$src2, 0)>,
503 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1, IntRegs:$src2))),
504 (L4_loadrb_rr IntRegs:$src1, IntRegs:$src2, 0)>,
507 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1, IntRegs:$src2))),
508 (L4_loadrub_rr IntRegs:$src1, IntRegs:$src2, 0)>,
511 def : Pat <(i32 (extloadi8 (add IntRegs:$src1, IntRegs:$src2))),
512 (L4_loadrub_rr IntRegs:$src1, IntRegs:$src2, 0)>,
515 def : Pat <(i32 (sextloadi16 (add IntRegs:$src1, IntRegs:$src2))),
516 (L4_loadrh_rr IntRegs:$src1, IntRegs:$src2, 0)>,
519 def : Pat <(i32 (zextloadi16 (add IntRegs:$src1, IntRegs:$src2))),
520 (L4_loadruh_rr IntRegs:$src1, IntRegs:$src2, 0)>,
523 def : Pat <(i32 (extloadi16 (add IntRegs:$src1, IntRegs:$src2))),
524 (L4_loadruh_rr IntRegs:$src1, IntRegs:$src2, 0)>,
527 def : Pat <(i32 (load (add IntRegs:$src1, IntRegs:$src2))),
528 (L4_loadri_rr IntRegs:$src1, IntRegs:$src2, 0)>,
533 def : Pat <(i64 (zext (i1 PredRegs:$src1))),
534 (i64 (A4_combineir 0, (C2_muxii (i1 PredRegs:$src1), 1, 0)))>,
538 def : Pat <(i64 (zext (i32 IntRegs:$src1))),
539 (i64 (A4_combineir 0, (i32 IntRegs:$src1)))>,
542 def: Pat <(i64 (zextloadi8 ADDRriS11_0:$src1)),
543 (i64 (A4_combineir 0, (L2_loadrub_io AddrFI:$src1, 0)))>,
546 let AddedComplexity = 20 in
547 def: Pat <(i64 (zextloadi8 (add (i32 IntRegs:$src1),
548 s11_0ExtPred:$offset))),
549 (i64 (A4_combineir 0, (L2_loadrub_io IntRegs:$src1,
550 s11_0ExtPred:$offset)))>,
554 def: Pat <(i64 (zextloadi1 ADDRriS11_0:$src1)),
555 (i64 (A4_combineir 0, (L2_loadrub_io AddrFI:$src1, 0)))>,
558 let AddedComplexity = 20 in
559 def: Pat <(i64 (zextloadi1 (add (i32 IntRegs:$src1),
560 s11_0ExtPred:$offset))),
561 (i64 (A4_combineir 0, (L2_loadrub_io IntRegs:$src1,
562 s11_0ExtPred:$offset)))>,
566 def: Pat <(i64 (zextloadi16 ADDRriS11_1:$src1)),
567 (i64 (A4_combineir 0, (L2_loadruh_io AddrFI:$src1, 0)))>,
570 let AddedComplexity = 20 in
571 def: Pat <(i64 (zextloadi16 (add (i32 IntRegs:$src1),
572 s11_1ExtPred:$offset))),
573 (i64 (A4_combineir 0, (L2_loadruh_io IntRegs:$src1,
574 s11_1ExtPred:$offset)))>,
578 def: Pat <(i64 (extloadi16 ADDRriS11_2:$src1)),
579 (i64 (A4_combineir 0, (L2_loadrh_io AddrFI:$src1, 0)))>,
582 let AddedComplexity = 20 in
583 def: Pat <(i64 (extloadi16 (add (i32 IntRegs:$src1),
584 s11_1ExtPred:$offset))),
585 (i64 (A4_combineir 0, (L2_loadrh_io IntRegs:$src1,
586 s11_1ExtPred:$offset)))>,
590 def: Pat <(i64 (zextloadi32 ADDRriS11_2:$src1)),
591 (i64 (A4_combineir 0, (L2_loadri_io AddrFI:$src1, 0)))>,
594 let AddedComplexity = 100 in
595 def: Pat <(i64 (zextloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
596 (i64 (A4_combineir 0, (L2_loadri_io IntRegs:$src1,
597 s11_2ExtPred:$offset)))>,
601 def: Pat <(i64 (extloadi32 ADDRriS11_2:$src1)),
602 (i64 (A4_combineir 0, (L2_loadri_io AddrFI:$src1, 0)))>,
605 let AddedComplexity = 100 in
606 def: Pat <(i64 (extloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
607 (i64 (A4_combineir 0, (L2_loadri_io IntRegs:$src1,
608 s11_2ExtPred:$offset)))>,
613 //===----------------------------------------------------------------------===//
615 //===----------------------------------------------------------------------===//
617 //===----------------------------------------------------------------------===//
619 //===----------------------------------------------------------------------===//
621 //===----------------------------------------------------------------------===//
622 // Template class for store instructions with Absolute set addressing mode.
623 //===----------------------------------------------------------------------===//
624 let isExtended = 1, opExtendable = 2, validSubTargets = HasV4SubT,
625 addrMode = AbsoluteSet in
626 class T_ST_abs_set<string mnemonic, RegisterClass RC>:
627 STInst2<(outs IntRegs:$dst1),
628 (ins RC:$src1, u0AlwaysExt:$src2),
629 mnemonic#"($dst1=##$src2) = $src1",
633 def STrid_abs_set_V4 : T_ST_abs_set <"memd", DoubleRegs>;
634 def STrib_abs_set_V4 : T_ST_abs_set <"memb", IntRegs>;
635 def STrih_abs_set_V4 : T_ST_abs_set <"memh", IntRegs>;
636 def STriw_abs_set_V4 : T_ST_abs_set <"memw", IntRegs>;
638 //===----------------------------------------------------------------------===//
639 // Template classes for the non-predicated store instructions with
640 // base + register offset addressing mode
641 //===----------------------------------------------------------------------===//
642 let isPredicable = 1 in
643 class T_store_rr <string mnemonic, RegisterClass RC, bits<3> MajOp, bit isH>
644 : STInst < (outs ), (ins IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, RC:$Rt),
645 mnemonic#"($Rs + $Ru<<#$u2) = $Rt"#!if(isH, ".h",""),
646 [],"",V4LDST_tc_st_SLOT01>, ImmRegShl, AddrModeRel {
655 let Inst{27-24} = 0b1011;
656 let Inst{23-21} = MajOp;
657 let Inst{20-16} = Rs;
659 let Inst{13} = u2{1};
664 //===----------------------------------------------------------------------===//
665 // Template classes for the predicated store instructions with
666 // base + register offset addressing mode
667 //===----------------------------------------------------------------------===//
668 let isPredicated = 1 in
669 class T_pstore_rr <string mnemonic, RegisterClass RC, bits<3> MajOp,
670 bit isNot, bit isPredNew, bit isH>
672 (ins PredRegs:$Pv, IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, RC:$Rt),
674 !if(isNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ",
675 ") ")#mnemonic#"($Rs+$Ru<<#$u2) = $Rt"#!if(isH, ".h",""),
676 [], "", V4LDST_tc_st_SLOT01> , AddrModeRel{
683 let isPredicatedFalse = isNot;
684 let isPredicatedNew = isPredNew;
688 let Inst{27-26} = 0b01;
689 let Inst{25} = isPredNew;
690 let Inst{24} = isNot;
691 let Inst{23-21} = MajOp;
692 let Inst{20-16} = Rs;
694 let Inst{13} = u2{1};
700 //===----------------------------------------------------------------------===//
701 // Template classes for the new-value store instructions with
702 // base + register offset addressing mode
703 //===----------------------------------------------------------------------===//
704 let isPredicable = 1, isNewValue = 1, opNewValue = 3 in
705 class T_store_new_rr <string mnemonic, bits<2> MajOp> :
706 NVInst < (outs ), (ins IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, IntRegs:$Nt),
707 mnemonic#"($Rs + $Ru<<#$u2) = $Nt.new",
708 [],"",V4LDST_tc_st_SLOT0>, ImmRegShl, AddrModeRel {
717 let Inst{27-21} = 0b1011101;
718 let Inst{20-16} = Rs;
720 let Inst{13} = u2{1};
722 let Inst{4-3} = MajOp;
726 //===----------------------------------------------------------------------===//
727 // Template classes for the predicated new-value store instructions with
728 // base + register offset addressing mode
729 //===----------------------------------------------------------------------===//
730 let isPredicated = 1, isNewValue = 1, opNewValue = 4 in
731 class T_pstore_new_rr <string mnemonic, bits<2> MajOp, bit isNot, bit isPredNew>
733 (ins PredRegs:$Pv, IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, IntRegs:$Nt),
734 !if(isNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ",
735 ") ")#mnemonic#"($Rs+$Ru<<#$u2) = $Nt.new",
736 [], "", V4LDST_tc_st_SLOT0>, AddrModeRel {
743 let isPredicatedFalse = isNot;
744 let isPredicatedNew = isPredNew;
747 let Inst{27-26} = 0b01;
748 let Inst{25} = isPredNew;
749 let Inst{24} = isNot;
750 let Inst{23-21} = 0b101;
751 let Inst{20-16} = Rs;
753 let Inst{13} = u2{1};
756 let Inst{4-3} = MajOp;
760 //===----------------------------------------------------------------------===//
761 // multiclass for store instructions with base + register offset addressing
763 //===----------------------------------------------------------------------===//
764 let isNVStorable = 1 in
765 multiclass ST_Idxd_shl<string mnemonic, string CextOp, RegisterClass RC,
766 bits<3> MajOp, bit isH = 0> {
767 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
768 def S4_#NAME#_rr : T_store_rr <mnemonic, RC, MajOp, isH>;
771 def S4_p#NAME#t_rr : T_pstore_rr <mnemonic, RC, MajOp, 0, 0, isH>;
772 def S4_p#NAME#f_rr : T_pstore_rr <mnemonic, RC, MajOp, 1, 0, isH>;
775 def S4_p#NAME#tnew_rr : T_pstore_rr <mnemonic, RC, MajOp, 0, 1, isH>;
776 def S4_p#NAME#fnew_rr : T_pstore_rr <mnemonic, RC, MajOp, 1, 1, isH>;
780 //===----------------------------------------------------------------------===//
781 // multiclass for new-value store instructions with base + register offset
783 //===----------------------------------------------------------------------===//
784 let mayStore = 1, isNVStore = 1 in
785 multiclass ST_Idxd_shl_nv <string mnemonic, string CextOp, RegisterClass RC,
787 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
788 def S4_#NAME#new_rr : T_store_new_rr<mnemonic, MajOp>;
791 def S4_p#NAME#newt_rr : T_pstore_new_rr <mnemonic, MajOp, 0, 0>;
792 def S4_p#NAME#newf_rr : T_pstore_new_rr <mnemonic, MajOp, 1, 0>;
795 def S4_p#NAME#newtnew_rr : T_pstore_new_rr <mnemonic, MajOp, 0, 1>;
796 def S4_p#NAME#newfnew_rr : T_pstore_new_rr <mnemonic, MajOp, 1, 1>;
800 let addrMode = BaseRegOffset, InputType = "reg", hasSideEffects = 0,
801 isCodeGenOnly = 0 in {
802 let accessSize = ByteAccess in
803 defm storerb: ST_Idxd_shl<"memb", "STrib", IntRegs, 0b000>,
804 ST_Idxd_shl_nv<"memb", "STrib", IntRegs, 0b00>;
806 let accessSize = HalfWordAccess in
807 defm storerh: ST_Idxd_shl<"memh", "STrih", IntRegs, 0b010>,
808 ST_Idxd_shl_nv<"memh", "STrih", IntRegs, 0b01>;
810 let accessSize = WordAccess in
811 defm storeri: ST_Idxd_shl<"memw", "STriw", IntRegs, 0b100>,
812 ST_Idxd_shl_nv<"memw", "STriw", IntRegs, 0b10>;
814 let isNVStorable = 0, accessSize = DoubleWordAccess in
815 defm storerd: ST_Idxd_shl<"memd", "STrid", DoubleRegs, 0b110>;
817 let isNVStorable = 0, accessSize = HalfWordAccess in
818 defm storerf: ST_Idxd_shl<"memh", "STrif", IntRegs, 0b011, 1>;
821 let Predicates = [HasV4T], AddedComplexity = 10 in {
822 def : Pat<(truncstorei8 (i32 IntRegs:$src4),
823 (add IntRegs:$src1, (shl IntRegs:$src2,
825 (S4_storerb_rr IntRegs:$src1, IntRegs:$src2,
826 u2ImmPred:$src3, IntRegs:$src4)>;
828 def : Pat<(truncstorei16 (i32 IntRegs:$src4),
829 (add IntRegs:$src1, (shl IntRegs:$src2,
831 (S4_storerh_rr IntRegs:$src1, IntRegs:$src2,
832 u2ImmPred:$src3, IntRegs:$src4)>;
834 def : Pat<(store (i32 IntRegs:$src4),
835 (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))),
836 (S4_storeri_rr IntRegs:$src1, IntRegs:$src2,
837 u2ImmPred:$src3, IntRegs:$src4)>;
839 def : Pat<(store (i64 DoubleRegs:$src4),
840 (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))),
841 (S4_storerd_rr IntRegs:$src1, IntRegs:$src2,
842 u2ImmPred:$src3, DoubleRegs:$src4)>;
845 let isExtended = 1, opExtendable = 2 in
846 class T_ST_LongOff <string mnemonic, PatFrag stOp, RegisterClass RC, ValueType VT> :
848 (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, RC:$src4),
849 mnemonic#"($src1<<#$src2+##$src3) = $src4",
850 [(stOp (VT RC:$src4),
851 (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2),
852 u0AlwaysExtPred:$src3))]>,
855 let isExtended = 1, opExtendable = 2, mayStore = 1, isNVStore = 1 in
856 class T_ST_LongOff_nv <string mnemonic> :
858 (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, IntRegs:$src4),
859 mnemonic#"($src1<<#$src2+##$src3) = $src4.new",
863 multiclass ST_LongOff <string mnemonic, string BaseOp, PatFrag stOp> {
864 let BaseOpcode = BaseOp#"_shl" in {
865 let isNVStorable = 1 in
866 def NAME#_V4 : T_ST_LongOff<mnemonic, stOp, IntRegs, i32>;
868 def NAME#_nv_V4 : T_ST_LongOff_nv<mnemonic>;
872 let AddedComplexity = 10, validSubTargets = HasV4SubT in {
873 def STrid_shl_V4 : T_ST_LongOff<"memd", store, DoubleRegs, i64>;
874 defm STrib_shl : ST_LongOff <"memb", "STrib", truncstorei8>, NewValueRel;
875 defm STrih_shl : ST_LongOff <"memh", "Strih", truncstorei16>, NewValueRel;
876 defm STriw_shl : ST_LongOff <"memw", "STriw", store>, NewValueRel;
879 let AddedComplexity = 40 in
880 multiclass T_ST_LOff_Pats <InstHexagon I, RegisterClass RC, ValueType VT,
882 def : Pat<(stOp (VT RC:$src4),
883 (add (shl IntRegs:$src1, u2ImmPred:$src2),
884 (NumUsesBelowThresCONST32 tglobaladdr:$src3))),
885 (I IntRegs:$src1, u2ImmPred:$src2, tglobaladdr:$src3, RC:$src4)>;
887 def : Pat<(stOp (VT RC:$src4),
889 (NumUsesBelowThresCONST32 tglobaladdr:$src3))),
890 (I IntRegs:$src1, 0, tglobaladdr:$src3, RC:$src4)>;
893 defm : T_ST_LOff_Pats<STrid_shl_V4, DoubleRegs, i64, store>;
894 defm : T_ST_LOff_Pats<STriw_shl_V4, IntRegs, i32, store>;
895 defm : T_ST_LOff_Pats<STrib_shl_V4, IntRegs, i32, truncstorei8>;
896 defm : T_ST_LOff_Pats<STrih_shl_V4, IntRegs, i32, truncstorei16>;
898 // memd(Rx++#s4:3)=Rtt
899 // memd(Rx++#s4:3:circ(Mu))=Rtt
900 // memd(Rx++I:circ(Mu))=Rtt
902 // memd(Rx++Mu:brev)=Rtt
903 // memd(gp+#u16:3)=Rtt
905 // Store doubleword conditionally.
906 // if ([!]Pv[.new]) memd(#u6)=Rtt
907 // TODO: needs to be implemented.
909 //===----------------------------------------------------------------------===//
911 //===----------------------------------------------------------------------===//
912 let isPredicable = 1, isExtendable = 1, isExtentSigned = 1, opExtentBits = 8,
914 class T_StoreImm <string mnemonic, Operand OffsetOp, bits<2> MajOp >
915 : STInst <(outs ), (ins IntRegs:$Rs, OffsetOp:$offset, s8Ext:$S8),
916 mnemonic#"($Rs+#$offset)=#$S8",
917 [], "", V4LDST_tc_st_SLOT01>,
918 ImmRegRel, PredNewRel {
924 string OffsetOpStr = !cast<string>(OffsetOp);
925 let offsetBits = !if (!eq(OffsetOpStr, "u6_2Imm"), offset{7-2},
926 !if (!eq(OffsetOpStr, "u6_1Imm"), offset{6-1},
927 /* u6_0Imm */ offset{5-0}));
931 let Inst{27-25} = 0b110;
932 let Inst{22-21} = MajOp;
933 let Inst{20-16} = Rs;
934 let Inst{12-7} = offsetBits;
935 let Inst{13} = S8{7};
936 let Inst{6-0} = S8{6-0};
939 let isPredicated = 1, isExtendable = 1, isExtentSigned = 1, opExtentBits = 6,
941 class T_StoreImm_pred <string mnemonic, Operand OffsetOp, bits<2> MajOp,
942 bit isPredNot, bit isPredNew >
944 (ins PredRegs:$Pv, IntRegs:$Rs, OffsetOp:$offset, s6Ext:$S6),
945 !if(isPredNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ",
946 ") ")#mnemonic#"($Rs+#$offset)=#$S6",
947 [], "", V4LDST_tc_st_SLOT01>,
948 ImmRegRel, PredNewRel {
955 string OffsetOpStr = !cast<string>(OffsetOp);
956 let offsetBits = !if (!eq(OffsetOpStr, "u6_2Imm"), offset{7-2},
957 !if (!eq(OffsetOpStr, "u6_1Imm"), offset{6-1},
958 /* u6_0Imm */ offset{5-0}));
959 let isPredicatedNew = isPredNew;
960 let isPredicatedFalse = isPredNot;
964 let Inst{27-25} = 0b100;
965 let Inst{24} = isPredNew;
966 let Inst{23} = isPredNot;
967 let Inst{22-21} = MajOp;
968 let Inst{20-16} = Rs;
969 let Inst{13} = S6{5};
970 let Inst{12-7} = offsetBits;
972 let Inst{4-0} = S6{4-0};
976 //===----------------------------------------------------------------------===//
977 // multiclass for store instructions with base + immediate offset
978 // addressing mode and immediate stored value.
979 // mem[bhw](Rx++#s4:3)=#s8
980 // if ([!]Pv[.new]) mem[bhw](Rx++#s4:3)=#s6
981 //===----------------------------------------------------------------------===//
983 multiclass ST_Imm_Pred <string mnemonic, Operand OffsetOp, bits<2> MajOp,
985 def _io : T_StoreImm_pred <mnemonic, OffsetOp, MajOp, PredNot, 0>;
987 def new_io : T_StoreImm_pred <mnemonic, OffsetOp, MajOp, PredNot, 1>;
990 multiclass ST_Imm <string mnemonic, string CextOp, Operand OffsetOp,
992 let CextOpcode = CextOp, BaseOpcode = CextOp#_imm in {
993 def _io : T_StoreImm <mnemonic, OffsetOp, MajOp>;
995 defm t : ST_Imm_Pred <mnemonic, OffsetOp, MajOp, 0>;
996 defm f : ST_Imm_Pred <mnemonic, OffsetOp, MajOp, 1>;
1000 let hasSideEffects = 0, validSubTargets = HasV4SubT, addrMode = BaseImmOffset,
1001 InputType = "imm", isCodeGenOnly = 0 in {
1002 let accessSize = ByteAccess in
1003 defm S4_storeirb : ST_Imm<"memb", "STrib", u6_0Imm, 0b00>;
1005 let accessSize = HalfWordAccess in
1006 defm S4_storeirh : ST_Imm<"memh", "STrih", u6_1Imm, 0b01>;
1008 let accessSize = WordAccess in
1009 defm S4_storeiri : ST_Imm<"memw", "STriw", u6_2Imm, 0b10>;
1012 let Predicates = [HasV4T], AddedComplexity = 10 in {
1013 def: Pat<(truncstorei8 s8ExtPred:$src3, (add IntRegs:$src1, u6_0ImmPred:$src2)),
1014 (S4_storeirb_io IntRegs:$src1, u6_0ImmPred:$src2, s8ExtPred:$src3)>;
1016 def: Pat<(truncstorei16 s8ExtPred:$src3, (add IntRegs:$src1,
1017 u6_1ImmPred:$src2)),
1018 (S4_storeirh_io IntRegs:$src1, u6_1ImmPred:$src2, s8ExtPred:$src3)>;
1020 def: Pat<(store s8ExtPred:$src3, (add IntRegs:$src1, u6_2ImmPred:$src2)),
1021 (S4_storeiri_io IntRegs:$src1, u6_2ImmPred:$src2, s8ExtPred:$src3)>;
1024 let AddedComplexity = 6 in
1025 def : Pat <(truncstorei8 s8ExtPred:$src2, (i32 IntRegs:$src1)),
1026 (S4_storeirb_io IntRegs:$src1, 0, s8ExtPred:$src2)>,
1029 // memb(Rx++#s4:0:circ(Mu))=Rt
1030 // memb(Rx++I:circ(Mu))=Rt
1032 // memb(Rx++Mu:brev)=Rt
1033 // memb(gp+#u16:0)=Rt
1037 // TODO: needs to be implemented
1038 // memh(Re=#U6)=Rt.H
1039 // memh(Rs+#s11:1)=Rt.H
1040 let AddedComplexity = 6 in
1041 def : Pat <(truncstorei16 s8ExtPred:$src2, (i32 IntRegs:$src1)),
1042 (S4_storeirh_io IntRegs:$src1, 0, s8ExtPred:$src2)>,
1045 // memh(Rs+Ru<<#u2)=Rt.H
1046 // TODO: needs to be implemented.
1048 // memh(Ru<<#u2+#U6)=Rt.H
1049 // memh(Rx++#s4:1:circ(Mu))=Rt.H
1050 // memh(Rx++#s4:1:circ(Mu))=Rt
1051 // memh(Rx++I:circ(Mu))=Rt.H
1052 // memh(Rx++I:circ(Mu))=Rt
1053 // memh(Rx++Mu)=Rt.H
1055 // memh(Rx++Mu:brev)=Rt.H
1056 // memh(Rx++Mu:brev)=Rt
1057 // memh(gp+#u16:1)=Rt
1058 // if ([!]Pv[.new]) memh(#u6)=Rt.H
1059 // if ([!]Pv[.new]) memh(#u6)=Rt
1062 // if ([!]Pv[.new]) memh(Rs+#u6:1)=Rt.H
1063 // TODO: needs to be implemented.
1065 // if ([!]Pv[.new]) memh(Rx++#s4:1)=Rt.H
1066 // TODO: Needs to be implemented.
1070 // TODO: Needs to be implemented.
1073 let hasSideEffects = 0 in
1074 def STriw_pred_V4 : STInst2<(outs),
1075 (ins MEMri:$addr, PredRegs:$src1),
1076 "Error; should not emit",
1080 let AddedComplexity = 6 in
1081 def : Pat <(store s8ExtPred:$src2, (i32 IntRegs:$src1)),
1082 (S4_storeiri_io IntRegs:$src1, 0, s8ExtPred:$src2)>,
1085 // memw(Rx++#s4:2)=Rt
1086 // memw(Rx++#s4:2:circ(Mu))=Rt
1087 // memw(Rx++I:circ(Mu))=Rt
1089 // memw(Rx++Mu:brev)=Rt
1091 //===----------------------------------------------------------------------===
1093 //===----------------------------------------------------------------------===
1096 //===----------------------------------------------------------------------===//
1098 //===----------------------------------------------------------------------===//
1100 let opNewValue = 2, opExtendable = 1, isExtentSigned = 1, isPredicable = 1 in
1101 class T_store_io_nv <string mnemonic, RegisterClass RC,
1102 Operand ImmOp, bits<2>MajOp>
1103 : NVInst_V4 <(outs),
1104 (ins IntRegs:$src1, ImmOp:$src2, RC:$src3),
1105 mnemonic#"($src1+#$src2) = $src3.new",
1106 [],"",ST_tc_st_SLOT0> {
1108 bits<13> src2; // Actual address offset
1110 bits<11> offsetBits; // Represents offset encoding
1112 let opExtentBits = !if (!eq(mnemonic, "memb"), 11,
1113 !if (!eq(mnemonic, "memh"), 12,
1114 !if (!eq(mnemonic, "memw"), 13, 0)));
1116 let opExtentAlign = !if (!eq(mnemonic, "memb"), 0,
1117 !if (!eq(mnemonic, "memh"), 1,
1118 !if (!eq(mnemonic, "memw"), 2, 0)));
1120 let offsetBits = !if (!eq(mnemonic, "memb"), src2{10-0},
1121 !if (!eq(mnemonic, "memh"), src2{11-1},
1122 !if (!eq(mnemonic, "memw"), src2{12-2}, 0)));
1124 let IClass = 0b1010;
1127 let Inst{26-25} = offsetBits{10-9};
1128 let Inst{24-21} = 0b1101;
1129 let Inst{20-16} = src1;
1130 let Inst{13} = offsetBits{8};
1131 let Inst{12-11} = MajOp;
1132 let Inst{10-8} = src3;
1133 let Inst{7-0} = offsetBits{7-0};
1136 let opExtendable = 2, opNewValue = 3, isPredicated = 1 in
1137 class T_pstore_io_nv <string mnemonic, RegisterClass RC, Operand predImmOp,
1138 bits<2>MajOp, bit PredNot, bit isPredNew>
1139 : NVInst_V4 <(outs),
1140 (ins PredRegs:$src1, IntRegs:$src2, predImmOp:$src3, RC:$src4),
1141 !if(PredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
1142 ") ")#mnemonic#"($src2+#$src3) = $src4.new",
1143 [],"",V2LDST_tc_st_SLOT0> {
1148 bits<6> offsetBits; // Represents offset encoding
1150 let isPredicatedNew = isPredNew;
1151 let isPredicatedFalse = PredNot;
1152 let opExtentBits = !if (!eq(mnemonic, "memb"), 6,
1153 !if (!eq(mnemonic, "memh"), 7,
1154 !if (!eq(mnemonic, "memw"), 8, 0)));
1156 let opExtentAlign = !if (!eq(mnemonic, "memb"), 0,
1157 !if (!eq(mnemonic, "memh"), 1,
1158 !if (!eq(mnemonic, "memw"), 2, 0)));
1160 let offsetBits = !if (!eq(mnemonic, "memb"), src3{5-0},
1161 !if (!eq(mnemonic, "memh"), src3{6-1},
1162 !if (!eq(mnemonic, "memw"), src3{7-2}, 0)));
1164 let IClass = 0b0100;
1167 let Inst{26} = PredNot;
1168 let Inst{25} = isPredNew;
1169 let Inst{24-21} = 0b0101;
1170 let Inst{20-16} = src2;
1171 let Inst{13} = offsetBits{5};
1172 let Inst{12-11} = MajOp;
1173 let Inst{10-8} = src4;
1174 let Inst{7-3} = offsetBits{4-0};
1176 let Inst{1-0} = src1;
1179 // multiclass for new-value store instructions with base + immediate offset.
1181 let mayStore = 1, isNVStore = 1, isNewValue = 1, hasSideEffects = 0,
1183 multiclass ST_Idxd_nv<string mnemonic, string CextOp, RegisterClass RC,
1184 Operand ImmOp, Operand predImmOp, bits<2> MajOp> {
1186 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
1187 def S2_#NAME#new_io : T_store_io_nv <mnemonic, RC, ImmOp, MajOp>;
1189 def S2_p#NAME#newt_io :T_pstore_io_nv <mnemonic, RC, predImmOp, MajOp, 0, 0>;
1190 def S2_p#NAME#newf_io :T_pstore_io_nv <mnemonic, RC, predImmOp, MajOp, 1, 0>;
1192 def S4_p#NAME#newtnew_io :T_pstore_io_nv <mnemonic, RC, predImmOp,
1194 def S4_p#NAME#newfnew_io :T_pstore_io_nv <mnemonic, RC, predImmOp,
1199 let addrMode = BaseImmOffset, InputType = "imm", isCodeGenOnly = 0 in {
1200 let accessSize = ByteAccess in
1201 defm storerb: ST_Idxd_nv<"memb", "STrib", IntRegs, s11_0Ext,
1202 u6_0Ext, 0b00>, AddrModeRel;
1204 let accessSize = HalfWordAccess, opExtentAlign = 1 in
1205 defm storerh: ST_Idxd_nv<"memh", "STrih", IntRegs, s11_1Ext,
1206 u6_1Ext, 0b01>, AddrModeRel;
1208 let accessSize = WordAccess, opExtentAlign = 2 in
1209 defm storeri: ST_Idxd_nv<"memw", "STriw", IntRegs, s11_2Ext,
1210 u6_2Ext, 0b10>, AddrModeRel;
1213 //===----------------------------------------------------------------------===//
1214 // Template class for non-predicated post increment .new stores
1215 // mem[bhwd](Rx++#s4:[0123])=Nt.new
1216 //===----------------------------------------------------------------------===//
1217 let isPredicable = 1, hasSideEffects = 0, validSubTargets = HasV4SubT,
1218 addrMode = PostInc, isNVStore = 1, isNewValue = 1, opNewValue = 3 in
1219 class T_StorePI_nv <string mnemonic, Operand ImmOp, bits<2> MajOp >
1220 : NVInstPI_V4 <(outs IntRegs:$_dst_),
1221 (ins IntRegs:$src1, ImmOp:$offset, IntRegs:$src2),
1222 mnemonic#"($src1++#$offset) = $src2.new",
1223 [], "$src1 = $_dst_">,
1230 string ImmOpStr = !cast<string>(ImmOp);
1231 let offsetBits = !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
1232 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
1233 /* s4_0Imm */ offset{3-0}));
1234 let IClass = 0b1010;
1236 let Inst{27-21} = 0b1011101;
1237 let Inst{20-16} = src1;
1239 let Inst{12-11} = MajOp;
1240 let Inst{10-8} = src2;
1242 let Inst{6-3} = offsetBits;
1246 //===----------------------------------------------------------------------===//
1247 // Template class for predicated post increment .new stores
1248 // if([!]Pv[.new]) mem[bhwd](Rx++#s4:[0123])=Nt.new
1249 //===----------------------------------------------------------------------===//
1250 let isPredicated = 1, hasSideEffects = 0, validSubTargets = HasV4SubT,
1251 addrMode = PostInc, isNVStore = 1, isNewValue = 1, opNewValue = 4 in
1252 class T_StorePI_nv_pred <string mnemonic, Operand ImmOp,
1253 bits<2> MajOp, bit isPredNot, bit isPredNew >
1254 : NVInstPI_V4 <(outs IntRegs:$_dst_),
1255 (ins PredRegs:$src1, IntRegs:$src2,
1256 ImmOp:$offset, IntRegs:$src3),
1257 !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
1258 ") ")#mnemonic#"($src2++#$offset) = $src3.new",
1259 [], "$src2 = $_dst_">,
1267 string ImmOpStr = !cast<string>(ImmOp);
1268 let offsetBits = !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
1269 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
1270 /* s4_0Imm */ offset{3-0}));
1271 let isPredicatedNew = isPredNew;
1272 let isPredicatedFalse = isPredNot;
1274 let IClass = 0b1010;
1276 let Inst{27-21} = 0b1011101;
1277 let Inst{20-16} = src2;
1279 let Inst{12-11} = MajOp;
1280 let Inst{10-8} = src3;
1281 let Inst{7} = isPredNew;
1282 let Inst{6-3} = offsetBits;
1283 let Inst{2} = isPredNot;
1284 let Inst{1-0} = src1;
1287 multiclass ST_PostInc_Pred_nv<string mnemonic, Operand ImmOp,
1288 bits<2> MajOp, bit PredNot> {
1289 def _pi : T_StorePI_nv_pred <mnemonic, ImmOp, MajOp, PredNot, 0>;
1292 def new_pi : T_StorePI_nv_pred <mnemonic, ImmOp, MajOp, PredNot, 1>;
1295 multiclass ST_PostInc_nv<string mnemonic, string BaseOp, Operand ImmOp,
1297 let BaseOpcode = "POST_"#BaseOp in {
1298 def S2_#NAME#_pi : T_StorePI_nv <mnemonic, ImmOp, MajOp>;
1301 defm S2_p#NAME#t : ST_PostInc_Pred_nv <mnemonic, ImmOp, MajOp, 0>;
1302 defm S2_p#NAME#f : ST_PostInc_Pred_nv <mnemonic, ImmOp, MajOp, 1>;
1306 let accessSize = ByteAccess, isCodeGenOnly = 0 in
1307 defm storerbnew: ST_PostInc_nv <"memb", "STrib", s4_0Imm, 0b00>;
1309 let accessSize = HalfWordAccess, isCodeGenOnly = 0 in
1310 defm storerhnew: ST_PostInc_nv <"memh", "STrih", s4_1Imm, 0b01>;
1312 let accessSize = WordAccess, isCodeGenOnly = 0 in
1313 defm storerinew: ST_PostInc_nv <"memw", "STriw", s4_2Imm, 0b10>;
1315 //===----------------------------------------------------------------------===//
1316 // Template class for post increment .new stores with register offset
1317 //===----------------------------------------------------------------------===//
1318 let isNewValue = 1, mayStore = 1, isNVStore = 1, opNewValue = 3 in
1319 class T_StorePI_RegNV <string mnemonic, bits<2> MajOp, MemAccessSize AccessSz>
1320 : NVInstPI_V4 <(outs IntRegs:$_dst_),
1321 (ins IntRegs:$src1, ModRegs:$src2, IntRegs:$src3),
1322 #mnemonic#"($src1++$src2) = $src3.new",
1323 [], "$src1 = $_dst_"> {
1327 let accessSize = AccessSz;
1329 let IClass = 0b1010;
1331 let Inst{27-21} = 0b1101101;
1332 let Inst{20-16} = src1;
1333 let Inst{13} = src2;
1334 let Inst{12-11} = MajOp;
1335 let Inst{10-8} = src3;
1339 let isCodeGenOnly = 0 in {
1340 def S2_storerbnew_pr : T_StorePI_RegNV<"memb", 0b00, ByteAccess>;
1341 def S2_storerhnew_pr : T_StorePI_RegNV<"memh", 0b01, HalfWordAccess>;
1342 def S2_storerinew_pr : T_StorePI_RegNV<"memw", 0b10, WordAccess>;
1345 // memb(Rx++#s4:0:circ(Mu))=Nt.new
1346 // memb(Rx++I:circ(Mu))=Nt.new
1347 // memb(Rx++Mu)=Nt.new
1348 // memb(Rx++Mu:brev)=Nt.new
1349 // memh(Rx++#s4:1:circ(Mu))=Nt.new
1350 // memh(Rx++I:circ(Mu))=Nt.new
1351 // memh(Rx++Mu)=Nt.new
1352 // memh(Rx++Mu:brev)=Nt.new
1354 // memw(Rx++#s4:2:circ(Mu))=Nt.new
1355 // memw(Rx++I:circ(Mu))=Nt.new
1356 // memw(Rx++Mu)=Nt.new
1357 // memw(Rx++Mu:brev)=Nt.new
1359 //===----------------------------------------------------------------------===//
1361 //===----------------------------------------------------------------------===//
1363 //===----------------------------------------------------------------------===//
1365 //===----------------------------------------------------------------------===//
1367 //===----------------------------------------------------------------------===//
1368 // multiclass/template class for the new-value compare jumps with the register
1370 //===----------------------------------------------------------------------===//
1372 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11,
1373 opExtentAlign = 2 in
1374 class NVJrr_template<string mnemonic, bits<3> majOp, bit NvOpNum,
1375 bit isNegCond, bit isTak>
1377 (ins IntRegs:$src1, IntRegs:$src2, brtarget:$offset),
1378 "if ("#!if(isNegCond, "!","")#mnemonic#
1379 "($src1"#!if(!eq(NvOpNum, 0),".new, ",", ")#
1380 "$src2"#!if(!eq(NvOpNum, 1),".new))","))")#" jump:"
1381 #!if(isTak, "t","nt")#" $offset", []> {
1385 bits<3> Ns; // New-Value Operand
1386 bits<5> RegOp; // Non-New-Value Operand
1389 let isTaken = isTak;
1390 let isPredicatedFalse = isNegCond;
1391 let opNewValue{0} = NvOpNum;
1393 let Ns = !if(!eq(NvOpNum, 0), src1{2-0}, src2{2-0});
1394 let RegOp = !if(!eq(NvOpNum, 0), src2, src1);
1396 let IClass = 0b0010;
1398 let Inst{25-23} = majOp;
1399 let Inst{22} = isNegCond;
1400 let Inst{18-16} = Ns;
1401 let Inst{13} = isTak;
1402 let Inst{12-8} = RegOp;
1403 let Inst{21-20} = offset{10-9};
1404 let Inst{7-1} = offset{8-2};
1408 multiclass NVJrr_cond<string mnemonic, bits<3> majOp, bit NvOpNum,
1410 // Branch not taken:
1411 def _nt_V4: NVJrr_template<mnemonic, majOp, NvOpNum, isNegCond, 0>;
1413 def _t_V4: NVJrr_template<mnemonic, majOp, NvOpNum, isNegCond, 1>;
1416 // NvOpNum = 0 -> First Operand is a new-value Register
1417 // NvOpNum = 1 -> Second Operand is a new-value Register
1419 multiclass NVJrr_base<string mnemonic, string BaseOp, bits<3> majOp,
1421 let BaseOpcode = BaseOp#_NVJ in {
1422 defm _t_Jumpnv : NVJrr_cond<mnemonic, majOp, NvOpNum, 0>; // True cond
1423 defm _f_Jumpnv : NVJrr_cond<mnemonic, majOp, NvOpNum, 1>; // False cond
1427 // if ([!]cmp.eq(Ns.new,Rt)) jump:[n]t #r9:2
1428 // if ([!]cmp.gt(Ns.new,Rt)) jump:[n]t #r9:2
1429 // if ([!]cmp.gtu(Ns.new,Rt)) jump:[n]t #r9:2
1430 // if ([!]cmp.gt(Rt,Ns.new)) jump:[n]t #r9:2
1431 // if ([!]cmp.gtu(Rt,Ns.new)) jump:[n]t #r9:2
1433 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1,
1434 Defs = [PC], hasSideEffects = 0, validSubTargets = HasV4SubT,
1435 isCodeGenOnly = 0 in {
1436 defm CMPEQrr : NVJrr_base<"cmp.eq", "CMPEQ", 0b000, 0>, PredRel;
1437 defm CMPGTrr : NVJrr_base<"cmp.gt", "CMPGT", 0b001, 0>, PredRel;
1438 defm CMPGTUrr : NVJrr_base<"cmp.gtu", "CMPGTU", 0b010, 0>, PredRel;
1439 defm CMPLTrr : NVJrr_base<"cmp.gt", "CMPLT", 0b011, 1>, PredRel;
1440 defm CMPLTUrr : NVJrr_base<"cmp.gtu", "CMPLTU", 0b100, 1>, PredRel;
1443 //===----------------------------------------------------------------------===//
1444 // multiclass/template class for the new-value compare jumps instruction
1445 // with a register and an unsigned immediate (U5) operand.
1446 //===----------------------------------------------------------------------===//
1448 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11,
1449 opExtentAlign = 2 in
1450 class NVJri_template<string mnemonic, bits<3> majOp, bit isNegCond,
1453 (ins IntRegs:$src1, u5Imm:$src2, brtarget:$offset),
1454 "if ("#!if(isNegCond, "!","")#mnemonic#"($src1.new, #$src2)) jump:"
1455 #!if(isTak, "t","nt")#" $offset", []> {
1457 let isTaken = isTak;
1458 let isPredicatedFalse = isNegCond;
1459 let isTaken = isTak;
1465 let IClass = 0b0010;
1467 let Inst{25-23} = majOp;
1468 let Inst{22} = isNegCond;
1469 let Inst{18-16} = src1;
1470 let Inst{13} = isTak;
1471 let Inst{12-8} = src2;
1472 let Inst{21-20} = offset{10-9};
1473 let Inst{7-1} = offset{8-2};
1476 multiclass NVJri_cond<string mnemonic, bits<3> majOp, bit isNegCond> {
1477 // Branch not taken:
1478 def _nt_V4: NVJri_template<mnemonic, majOp, isNegCond, 0>;
1480 def _t_V4: NVJri_template<mnemonic, majOp, isNegCond, 1>;
1483 multiclass NVJri_base<string mnemonic, string BaseOp, bits<3> majOp> {
1484 let BaseOpcode = BaseOp#_NVJri in {
1485 defm _t_Jumpnv : NVJri_cond<mnemonic, majOp, 0>; // True Cond
1486 defm _f_Jumpnv : NVJri_cond<mnemonic, majOp, 1>; // False cond
1490 // if ([!]cmp.eq(Ns.new,#U5)) jump:[n]t #r9:2
1491 // if ([!]cmp.gt(Ns.new,#U5)) jump:[n]t #r9:2
1492 // if ([!]cmp.gtu(Ns.new,#U5)) jump:[n]t #r9:2
1494 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1,
1495 Defs = [PC], hasSideEffects = 0, validSubTargets = HasV4SubT,
1496 isCodeGenOnly = 0 in {
1497 defm CMPEQri : NVJri_base<"cmp.eq", "CMPEQ", 0b000>, PredRel;
1498 defm CMPGTri : NVJri_base<"cmp.gt", "CMPGT", 0b001>, PredRel;
1499 defm CMPGTUri : NVJri_base<"cmp.gtu", "CMPGTU", 0b010>, PredRel;
1502 //===----------------------------------------------------------------------===//
1503 // multiclass/template class for the new-value compare jumps instruction
1504 // with a register and an hardcoded 0/-1 immediate value.
1505 //===----------------------------------------------------------------------===//
1507 let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 11,
1508 opExtentAlign = 2 in
1509 class NVJ_ConstImm_template<string mnemonic, bits<3> majOp, string ImmVal,
1510 bit isNegCond, bit isTak>
1512 (ins IntRegs:$src1, brtarget:$offset),
1513 "if ("#!if(isNegCond, "!","")#mnemonic
1514 #"($src1.new, #"#ImmVal#")) jump:"
1515 #!if(isTak, "t","nt")#" $offset", []> {
1517 let isTaken = isTak;
1518 let isPredicatedFalse = isNegCond;
1519 let isTaken = isTak;
1523 let IClass = 0b0010;
1525 let Inst{25-23} = majOp;
1526 let Inst{22} = isNegCond;
1527 let Inst{18-16} = src1;
1528 let Inst{13} = isTak;
1529 let Inst{21-20} = offset{10-9};
1530 let Inst{7-1} = offset{8-2};
1533 multiclass NVJ_ConstImm_cond<string mnemonic, bits<3> majOp, string ImmVal,
1535 // Branch not taken:
1536 def _nt_V4: NVJ_ConstImm_template<mnemonic, majOp, ImmVal, isNegCond, 0>;
1538 def _t_V4: NVJ_ConstImm_template<mnemonic, majOp, ImmVal, isNegCond, 1>;
1541 multiclass NVJ_ConstImm_base<string mnemonic, string BaseOp, bits<3> majOp,
1543 let BaseOpcode = BaseOp#_NVJ_ConstImm in {
1544 defm _t_Jumpnv : NVJ_ConstImm_cond<mnemonic, majOp, ImmVal, 0>; // True
1545 defm _f_Jumpnv : NVJ_ConstImm_cond<mnemonic, majOp, ImmVal, 1>; // False
1549 // if ([!]tstbit(Ns.new,#0)) jump:[n]t #r9:2
1550 // if ([!]cmp.eq(Ns.new,#-1)) jump:[n]t #r9:2
1551 // if ([!]cmp.gt(Ns.new,#-1)) jump:[n]t #r9:2
1553 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator=1,
1554 Defs = [PC], hasSideEffects = 0, isCodeGenOnly = 0 in {
1555 defm TSTBIT0 : NVJ_ConstImm_base<"tstbit", "TSTBIT", 0b011, "0">, PredRel;
1556 defm CMPEQn1 : NVJ_ConstImm_base<"cmp.eq", "CMPEQ", 0b100, "-1">, PredRel;
1557 defm CMPGTn1 : NVJ_ConstImm_base<"cmp.gt", "CMPGT", 0b101, "-1">, PredRel;
1560 // J4_hintjumpr: Hint indirect conditional jump.
1561 let isBranch = 1, isIndirectBranch = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
1562 def J4_hintjumpr: JRInst <
1567 let IClass = 0b0101;
1568 let Inst{27-21} = 0b0010101;
1569 let Inst{20-16} = Rs;
1572 //===----------------------------------------------------------------------===//
1574 //===----------------------------------------------------------------------===//
1576 //===----------------------------------------------------------------------===//
1578 //===----------------------------------------------------------------------===//
1580 // Add and accumulate.
1581 // Rd=add(Rs,add(Ru,#s6))
1582 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 6,
1583 validSubTargets = HasV4SubT in
1584 def ADDr_ADDri_V4 : MInst<(outs IntRegs:$dst),
1585 (ins IntRegs:$src1, IntRegs:$src2, s6Ext:$src3),
1586 "$dst = add($src1, add($src2, #$src3))",
1587 [(set (i32 IntRegs:$dst),
1588 (add (i32 IntRegs:$src1), (add (i32 IntRegs:$src2),
1589 s6_16ExtPred:$src3)))]>,
1592 // Rd=add(Rs,sub(#s6,Ru))
1593 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6,
1594 validSubTargets = HasV4SubT in
1595 def ADDr_SUBri_V4 : MInst<(outs IntRegs:$dst),
1596 (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3),
1597 "$dst = add($src1, sub(#$src2, $src3))",
1598 [(set (i32 IntRegs:$dst),
1599 (add (i32 IntRegs:$src1), (sub s6_10ExtPred:$src2,
1600 (i32 IntRegs:$src3))))]>,
1603 // Generates the same instruction as ADDr_SUBri_V4 but matches different
1605 // Rd=add(Rs,sub(#s6,Ru))
1606 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6,
1607 validSubTargets = HasV4SubT in
1608 def ADDri_SUBr_V4 : MInst<(outs IntRegs:$dst),
1609 (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3),
1610 "$dst = add($src1, sub(#$src2, $src3))",
1611 [(set (i32 IntRegs:$dst),
1612 (sub (add (i32 IntRegs:$src1), s6_10ExtPred:$src2),
1613 (i32 IntRegs:$src3)))]>,
1617 // Add or subtract doublewords with carry.
1619 // Rdd=add(Rss,Rtt,Px):carry
1621 // Rdd=sub(Rss,Rtt,Px):carry
1624 // Logical doublewords.
1625 // Rdd=and(Rtt,~Rss)
1626 let validSubTargets = HasV4SubT in
1627 def ANDd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
1628 (ins DoubleRegs:$src1, DoubleRegs:$src2),
1629 "$dst = and($src1, ~$src2)",
1630 [(set (i64 DoubleRegs:$dst), (and (i64 DoubleRegs:$src1),
1631 (not (i64 DoubleRegs:$src2))))]>,
1635 let validSubTargets = HasV4SubT in
1636 def ORd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
1637 (ins DoubleRegs:$src1, DoubleRegs:$src2),
1638 "$dst = or($src1, ~$src2)",
1639 [(set (i64 DoubleRegs:$dst),
1640 (or (i64 DoubleRegs:$src1), (not (i64 DoubleRegs:$src2))))]>,
1644 // Logical-logical doublewords.
1645 // Rxx^=xor(Rss,Rtt)
1646 let validSubTargets = HasV4SubT in
1647 def XORd_XORdd: MInst_acc<(outs DoubleRegs:$dst),
1648 (ins DoubleRegs:$src1, DoubleRegs:$src2, DoubleRegs:$src3),
1649 "$dst ^= xor($src2, $src3)",
1650 [(set (i64 DoubleRegs:$dst),
1651 (xor (i64 DoubleRegs:$src1), (xor (i64 DoubleRegs:$src2),
1652 (i64 DoubleRegs:$src3))))],
1657 // Logical-logical words.
1658 // Rx=or(Ru,and(Rx,#s10))
1659 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
1660 validSubTargets = HasV4SubT in
1661 def ORr_ANDri_V4 : MInst_acc<(outs IntRegs:$dst),
1662 (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
1663 "$dst = or($src1, and($src2, #$src3))",
1664 [(set (i32 IntRegs:$dst),
1665 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1666 s10ExtPred:$src3)))],
1670 // Rx[&|^]=and(Rs,Rt)
1672 let validSubTargets = HasV4SubT in
1673 def ANDr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
1674 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1675 "$dst &= and($src2, $src3)",
1676 [(set (i32 IntRegs:$dst),
1677 (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1678 (i32 IntRegs:$src3))))],
1683 let validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "reg" in
1684 def ORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
1685 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1686 "$dst |= and($src2, $src3)",
1687 [(set (i32 IntRegs:$dst),
1688 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1689 (i32 IntRegs:$src3))))],
1691 Requires<[HasV4T]>, ImmRegRel;
1694 let validSubTargets = HasV4SubT in
1695 def XORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
1696 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1697 "$dst ^= and($src2, $src3)",
1698 [(set (i32 IntRegs:$dst),
1699 (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1700 (i32 IntRegs:$src3))))],
1704 // Rx[&|^]=and(Rs,~Rt)
1706 let validSubTargets = HasV4SubT in
1707 def ANDr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
1708 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1709 "$dst &= and($src2, ~$src3)",
1710 [(set (i32 IntRegs:$dst),
1711 (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1712 (not (i32 IntRegs:$src3)))))],
1717 let validSubTargets = HasV4SubT in
1718 def ORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
1719 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1720 "$dst |= and($src2, ~$src3)",
1721 [(set (i32 IntRegs:$dst),
1722 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1723 (not (i32 IntRegs:$src3)))))],
1728 let validSubTargets = HasV4SubT in
1729 def XORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
1730 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1731 "$dst ^= and($src2, ~$src3)",
1732 [(set (i32 IntRegs:$dst),
1733 (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1734 (not (i32 IntRegs:$src3)))))],
1738 // Rx[&|^]=or(Rs,Rt)
1740 let validSubTargets = HasV4SubT in
1741 def ANDr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1742 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1743 "$dst &= or($src2, $src3)",
1744 [(set (i32 IntRegs:$dst),
1745 (and (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
1746 (i32 IntRegs:$src3))))],
1751 let validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "reg" in
1752 def ORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1753 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1754 "$dst |= or($src2, $src3)",
1755 [(set (i32 IntRegs:$dst),
1756 (or (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
1757 (i32 IntRegs:$src3))))],
1759 Requires<[HasV4T]>, ImmRegRel;
1762 let validSubTargets = HasV4SubT in
1763 def XORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1764 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1765 "$dst ^= or($src2, $src3)",
1766 [(set (i32 IntRegs:$dst),
1767 (xor (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
1768 (i32 IntRegs:$src3))))],
1772 // Rx[&|^]=xor(Rs,Rt)
1774 let validSubTargets = HasV4SubT in
1775 def ANDr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1776 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1777 "$dst &= xor($src2, $src3)",
1778 [(set (i32 IntRegs:$dst),
1779 (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
1780 (i32 IntRegs:$src3))))],
1785 let validSubTargets = HasV4SubT in
1786 def ORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1787 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1788 "$dst |= xor($src2, $src3)",
1789 [(set (i32 IntRegs:$dst),
1790 (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
1791 (i32 IntRegs:$src3))))],
1796 let validSubTargets = HasV4SubT in
1797 def XORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
1798 (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
1799 "$dst ^= xor($src2, $src3)",
1800 [(set (i32 IntRegs:$dst),
1801 (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
1802 (i32 IntRegs:$src3))))],
1807 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
1808 validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "imm" in
1809 def ORr_ANDri2_V4 : MInst_acc<(outs IntRegs:$dst),
1810 (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
1811 "$dst |= and($src2, #$src3)",
1812 [(set (i32 IntRegs:$dst),
1813 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1814 s10ExtPred:$src3)))],
1816 Requires<[HasV4T]>, ImmRegRel;
1819 let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
1820 validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "imm" in
1821 def ORr_ORri_V4 : MInst_acc<(outs IntRegs:$dst),
1822 (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
1823 "$dst |= or($src2, #$src3)",
1824 [(set (i32 IntRegs:$dst),
1825 (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
1826 s10ExtPred:$src3)))],
1828 Requires<[HasV4T]>, ImmRegRel;
1832 // Rd=modwrap(Rs,Rt)
1834 // Rd=cround(Rs,#u5)
1836 // Rd=round(Rs,#u5)[:sat]
1837 // Rd=round(Rs,Rt)[:sat]
1838 // Vector reduce add unsigned halfwords
1839 // Rd=vraddh(Rss,Rtt)
1841 // Rdd=vaddb(Rss,Rtt)
1842 // Vector conditional negate
1843 // Rdd=vcnegh(Rss,Rt)
1844 // Rxx+=vrcnegh(Rss,Rt)
1845 // Vector maximum bytes
1846 // Rdd=vmaxb(Rtt,Rss)
1847 // Vector reduce maximum halfwords
1848 // Rxx=vrmaxh(Rss,Ru)
1849 // Rxx=vrmaxuh(Rss,Ru)
1850 // Vector reduce maximum words
1851 // Rxx=vrmaxuw(Rss,Ru)
1852 // Rxx=vrmaxw(Rss,Ru)
1853 // Vector minimum bytes
1854 // Rdd=vminb(Rtt,Rss)
1855 // Vector reduce minimum halfwords
1856 // Rxx=vrminh(Rss,Ru)
1857 // Rxx=vrminuh(Rss,Ru)
1858 // Vector reduce minimum words
1859 // Rxx=vrminuw(Rss,Ru)
1860 // Rxx=vrminw(Rss,Ru)
1861 // Vector subtract bytes
1862 // Rdd=vsubb(Rss,Rtt)
1864 //===----------------------------------------------------------------------===//
1866 //===----------------------------------------------------------------------===//
1869 //===----------------------------------------------------------------------===//
1871 //===----------------------------------------------------------------------===//
1873 // Multiply and user lower result.
1874 // Rd=add(#u6,mpyi(Rs,#U6))
1875 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6,
1876 validSubTargets = HasV4SubT in
1877 def ADDi_MPYri_V4 : MInst<(outs IntRegs:$dst),
1878 (ins u6Ext:$src1, IntRegs:$src2, u6Imm:$src3),
1879 "$dst = add(#$src1, mpyi($src2, #$src3))",
1880 [(set (i32 IntRegs:$dst),
1881 (add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
1882 u6ExtPred:$src1))]>,
1885 // Rd=add(##,mpyi(Rs,#U6))
1886 def : Pat <(add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
1887 (HexagonCONST32 tglobaladdr:$src1)),
1888 (i32 (ADDi_MPYri_V4 tglobaladdr:$src1, IntRegs:$src2,
1891 // Rd=add(#u6,mpyi(Rs,Rt))
1892 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6,
1893 validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in
1894 def ADDi_MPYrr_V4 : MInst<(outs IntRegs:$dst),
1895 (ins u6Ext:$src1, IntRegs:$src2, IntRegs:$src3),
1896 "$dst = add(#$src1, mpyi($src2, $src3))",
1897 [(set (i32 IntRegs:$dst),
1898 (add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
1899 u6ExtPred:$src1))]>,
1900 Requires<[HasV4T]>, ImmRegRel;
1902 // Rd=add(##,mpyi(Rs,Rt))
1903 def : Pat <(add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
1904 (HexagonCONST32 tglobaladdr:$src1)),
1905 (i32 (ADDi_MPYrr_V4 tglobaladdr:$src1, IntRegs:$src2,
1908 // Rd=add(Ru,mpyi(#u6:2,Rs))
1909 let validSubTargets = HasV4SubT in
1910 def ADDr_MPYir_V4 : MInst<(outs IntRegs:$dst),
1911 (ins IntRegs:$src1, u6Imm:$src2, IntRegs:$src3),
1912 "$dst = add($src1, mpyi(#$src2, $src3))",
1913 [(set (i32 IntRegs:$dst),
1914 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src3),
1915 u6_2ImmPred:$src2)))]>,
1918 // Rd=add(Ru,mpyi(Rs,#u6))
1919 let isExtendable = 1, opExtendable = 3, isExtentSigned = 0, opExtentBits = 6,
1920 validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in
1921 def ADDr_MPYri_V4 : MInst<(outs IntRegs:$dst),
1922 (ins IntRegs:$src1, IntRegs:$src2, u6Ext:$src3),
1923 "$dst = add($src1, mpyi($src2, #$src3))",
1924 [(set (i32 IntRegs:$dst),
1925 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
1926 u6ExtPred:$src3)))]>,
1927 Requires<[HasV4T]>, ImmRegRel;
1929 // Rx=add(Ru,mpyi(Rx,Rs))
1930 let validSubTargets = HasV4SubT, InputType = "reg", CextOpcode = "ADD_MPY" in
1931 def ADDr_MPYrr_V4 : MInst_acc<(outs IntRegs:$dst),
1932 (ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3),
1933 "$dst = add($src1, mpyi($src2, $src3))",
1934 [(set (i32 IntRegs:$dst),
1935 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
1936 (i32 IntRegs:$src3))))],
1938 Requires<[HasV4T]>, ImmRegRel;
1941 // Polynomial multiply words
1943 // Rxx^=pmpyw(Rs,Rt)
1945 // Vector reduce multiply word by signed half (32x16)
1946 // Rdd=vrmpyweh(Rss,Rtt)[:<<1]
1947 // Rdd=vrmpywoh(Rss,Rtt)[:<<1]
1948 // Rxx+=vrmpyweh(Rss,Rtt)[:<<1]
1949 // Rxx+=vrmpywoh(Rss,Rtt)[:<<1]
1951 // Multiply and use upper result
1952 // Rd=mpy(Rs,Rt.H):<<1:sat
1953 // Rd=mpy(Rs,Rt.L):<<1:sat
1954 // Rd=mpy(Rs,Rt):<<1
1955 // Rd=mpy(Rs,Rt):<<1:sat
1957 // Rx+=mpy(Rs,Rt):<<1:sat
1958 // Rx-=mpy(Rs,Rt):<<1:sat
1960 // Vector multiply bytes
1961 // Rdd=vmpybsu(Rs,Rt)
1962 // Rdd=vmpybu(Rs,Rt)
1963 // Rxx+=vmpybsu(Rs,Rt)
1964 // Rxx+=vmpybu(Rs,Rt)
1966 // Vector polynomial multiply halfwords
1967 // Rdd=vpmpyh(Rs,Rt)
1968 // Rxx^=vpmpyh(Rs,Rt)
1970 //===----------------------------------------------------------------------===//
1972 //===----------------------------------------------------------------------===//
1975 //===----------------------------------------------------------------------===//
1977 //===----------------------------------------------------------------------===//
1979 // Shift by immediate and accumulate.
1980 // Rx=add(#u8,asl(Rx,#U5))
1981 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1982 validSubTargets = HasV4SubT in
1983 def ADDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
1984 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1985 "$dst = add(#$src1, asl($src2, #$src3))",
1986 [(set (i32 IntRegs:$dst),
1987 (add (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
1992 // Rx=add(#u8,lsr(Rx,#U5))
1993 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
1994 validSubTargets = HasV4SubT in
1995 def ADDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
1996 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
1997 "$dst = add(#$src1, lsr($src2, #$src3))",
1998 [(set (i32 IntRegs:$dst),
1999 (add (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
2004 // Rx=sub(#u8,asl(Rx,#U5))
2005 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2006 validSubTargets = HasV4SubT in
2007 def SUBi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
2008 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2009 "$dst = sub(#$src1, asl($src2, #$src3))",
2010 [(set (i32 IntRegs:$dst),
2011 (sub (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
2016 // Rx=sub(#u8,lsr(Rx,#U5))
2017 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2018 validSubTargets = HasV4SubT in
2019 def SUBi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
2020 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2021 "$dst = sub(#$src1, lsr($src2, #$src3))",
2022 [(set (i32 IntRegs:$dst),
2023 (sub (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
2029 //Shift by immediate and logical.
2030 //Rx=and(#u8,asl(Rx,#U5))
2031 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2032 validSubTargets = HasV4SubT in
2033 def ANDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
2034 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2035 "$dst = and(#$src1, asl($src2, #$src3))",
2036 [(set (i32 IntRegs:$dst),
2037 (and (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
2042 //Rx=and(#u8,lsr(Rx,#U5))
2043 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2044 validSubTargets = HasV4SubT in
2045 def ANDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
2046 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2047 "$dst = and(#$src1, lsr($src2, #$src3))",
2048 [(set (i32 IntRegs:$dst),
2049 (and (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
2054 //Rx=or(#u8,asl(Rx,#U5))
2055 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2056 AddedComplexity = 30, validSubTargets = HasV4SubT in
2057 def ORi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
2058 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2059 "$dst = or(#$src1, asl($src2, #$src3))",
2060 [(set (i32 IntRegs:$dst),
2061 (or (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
2066 //Rx=or(#u8,lsr(Rx,#U5))
2067 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2068 AddedComplexity = 30, validSubTargets = HasV4SubT in
2069 def ORi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
2070 (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
2071 "$dst = or(#$src1, lsr($src2, #$src3))",
2072 [(set (i32 IntRegs:$dst),
2073 (or (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
2079 //Shift by register.
2081 let validSubTargets = HasV4SubT in {
2082 def LSLi_V4 : MInst<(outs IntRegs:$dst), (ins s6Imm:$src1, IntRegs:$src2),
2083 "$dst = lsl(#$src1, $src2)",
2084 [(set (i32 IntRegs:$dst), (shl s6ImmPred:$src1,
2085 (i32 IntRegs:$src2)))]>,
2089 //Shift by register and logical.
2091 def ASLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
2092 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
2093 "$dst ^= asl($src2, $src3)",
2094 [(set (i64 DoubleRegs:$dst),
2095 (xor (i64 DoubleRegs:$src1), (shl (i64 DoubleRegs:$src2),
2096 (i32 IntRegs:$src3))))],
2101 def ASRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
2102 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
2103 "$dst ^= asr($src2, $src3)",
2104 [(set (i64 DoubleRegs:$dst),
2105 (xor (i64 DoubleRegs:$src1), (sra (i64 DoubleRegs:$src2),
2106 (i32 IntRegs:$src3))))],
2111 def LSLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
2112 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
2113 "$dst ^= lsl($src2, $src3)",
2114 [(set (i64 DoubleRegs:$dst), (xor (i64 DoubleRegs:$src1),
2115 (shl (i64 DoubleRegs:$src2),
2116 (i32 IntRegs:$src3))))],
2121 def LSRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
2122 (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
2123 "$dst ^= lsr($src2, $src3)",
2124 [(set (i64 DoubleRegs:$dst),
2125 (xor (i64 DoubleRegs:$src1), (srl (i64 DoubleRegs:$src2),
2126 (i32 IntRegs:$src3))))],
2131 //===----------------------------------------------------------------------===//
2133 //===----------------------------------------------------------------------===//
2135 //===----------------------------------------------------------------------===//
2136 // MEMOP: Word, Half, Byte
2137 //===----------------------------------------------------------------------===//
2139 def MEMOPIMM : SDNodeXForm<imm, [{
2140 // Call the transformation function XformM5ToU5Imm to get the negative
2141 // immediate's positive counterpart.
2142 int32_t imm = N->getSExtValue();
2143 return XformM5ToU5Imm(imm);
2146 def MEMOPIMM_HALF : SDNodeXForm<imm, [{
2147 // -1 .. -31 represented as 65535..65515
2148 // assigning to a short restores our desired signed value.
2149 // Call the transformation function XformM5ToU5Imm to get the negative
2150 // immediate's positive counterpart.
2151 int16_t imm = N->getSExtValue();
2152 return XformM5ToU5Imm(imm);
2155 def MEMOPIMM_BYTE : SDNodeXForm<imm, [{
2156 // -1 .. -31 represented as 255..235
2157 // assigning to a char restores our desired signed value.
2158 // Call the transformation function XformM5ToU5Imm to get the negative
2159 // immediate's positive counterpart.
2160 int8_t imm = N->getSExtValue();
2161 return XformM5ToU5Imm(imm);
2164 def SETMEMIMM : SDNodeXForm<imm, [{
2165 // Return the bit position we will set [0-31].
2167 int32_t imm = N->getSExtValue();
2168 return XformMskToBitPosU5Imm(imm);
2171 def CLRMEMIMM : SDNodeXForm<imm, [{
2172 // Return the bit position we will clear [0-31].
2174 // we bit negate the value first
2175 int32_t imm = ~(N->getSExtValue());
2176 return XformMskToBitPosU5Imm(imm);
2179 def SETMEMIMM_SHORT : SDNodeXForm<imm, [{
2180 // Return the bit position we will set [0-15].
2182 int16_t imm = N->getSExtValue();
2183 return XformMskToBitPosU4Imm(imm);
2186 def CLRMEMIMM_SHORT : SDNodeXForm<imm, [{
2187 // Return the bit position we will clear [0-15].
2189 // we bit negate the value first
2190 int16_t imm = ~(N->getSExtValue());
2191 return XformMskToBitPosU4Imm(imm);
2194 def SETMEMIMM_BYTE : SDNodeXForm<imm, [{
2195 // Return the bit position we will set [0-7].
2197 int8_t imm = N->getSExtValue();
2198 return XformMskToBitPosU3Imm(imm);
2201 def CLRMEMIMM_BYTE : SDNodeXForm<imm, [{
2202 // Return the bit position we will clear [0-7].
2204 // we bit negate the value first
2205 int8_t imm = ~(N->getSExtValue());
2206 return XformMskToBitPosU3Imm(imm);
2209 //===----------------------------------------------------------------------===//
2210 // Template class for MemOp instructions with the register value.
2211 //===----------------------------------------------------------------------===//
2212 class MemOp_rr_base <string opc, bits<2> opcBits, Operand ImmOp,
2213 string memOp, bits<2> memOpBits> :
2215 (ins IntRegs:$base, ImmOp:$offset, IntRegs:$delta),
2216 opc#"($base+#$offset)"#memOp#"$delta",
2218 Requires<[HasV4T, UseMEMOP]> {
2223 bits<6> offsetBits; // memb - u6:0 , memh - u6:1, memw - u6:2
2225 let offsetBits = !if (!eq(opcBits, 0b00), offset{5-0},
2226 !if (!eq(opcBits, 0b01), offset{6-1},
2227 !if (!eq(opcBits, 0b10), offset{7-2},0)));
2229 let IClass = 0b0011;
2230 let Inst{27-24} = 0b1110;
2231 let Inst{22-21} = opcBits;
2232 let Inst{20-16} = base;
2234 let Inst{12-7} = offsetBits;
2235 let Inst{6-5} = memOpBits;
2236 let Inst{4-0} = delta;
2239 //===----------------------------------------------------------------------===//
2240 // Template class for MemOp instructions with the immediate value.
2241 //===----------------------------------------------------------------------===//
2242 class MemOp_ri_base <string opc, bits<2> opcBits, Operand ImmOp,
2243 string memOp, bits<2> memOpBits> :
2245 (ins IntRegs:$base, ImmOp:$offset, u5Imm:$delta),
2246 opc#"($base+#$offset)"#memOp#"#$delta"
2247 #!if(memOpBits{1},")", ""), // clrbit, setbit - include ')'
2249 Requires<[HasV4T, UseMEMOP]> {
2254 bits<6> offsetBits; // memb - u6:0 , memh - u6:1, memw - u6:2
2256 let offsetBits = !if (!eq(opcBits, 0b00), offset{5-0},
2257 !if (!eq(opcBits, 0b01), offset{6-1},
2258 !if (!eq(opcBits, 0b10), offset{7-2},0)));
2260 let IClass = 0b0011;
2261 let Inst{27-24} = 0b1111;
2262 let Inst{22-21} = opcBits;
2263 let Inst{20-16} = base;
2265 let Inst{12-7} = offsetBits;
2266 let Inst{6-5} = memOpBits;
2267 let Inst{4-0} = delta;
2270 // multiclass to define MemOp instructions with register operand.
2271 multiclass MemOp_rr<string opc, bits<2> opcBits, Operand ImmOp> {
2272 def _ADD#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " += ", 0b00>; // add
2273 def _SUB#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " -= ", 0b01>; // sub
2274 def _AND#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " &= ", 0b10>; // and
2275 def _OR#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " |= ", 0b11>; // or
2278 // multiclass to define MemOp instructions with immediate Operand.
2279 multiclass MemOp_ri<string opc, bits<2> opcBits, Operand ImmOp> {
2280 def _ADD#NAME#_V4 : MemOp_ri_base <opc, opcBits, ImmOp, " += ", 0b00 >;
2281 def _SUB#NAME#_V4 : MemOp_ri_base <opc, opcBits, ImmOp, " -= ", 0b01 >;
2282 def _CLRBIT#NAME#_V4 : MemOp_ri_base<opc, opcBits, ImmOp, " =clrbit(", 0b10>;
2283 def _SETBIT#NAME#_V4 : MemOp_ri_base<opc, opcBits, ImmOp, " =setbit(", 0b11>;
2286 multiclass MemOp_base <string opc, bits<2> opcBits, Operand ImmOp> {
2287 defm r : MemOp_rr <opc, opcBits, ImmOp>;
2288 defm i : MemOp_ri <opc, opcBits, ImmOp>;
2291 // Define MemOp instructions.
2292 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0,
2293 validSubTargets =HasV4SubT in {
2294 let opExtentBits = 6, accessSize = ByteAccess in
2295 defm MemOPb : MemOp_base <"memb", 0b00, u6_0Ext>;
2297 let opExtentBits = 7, accessSize = HalfWordAccess in
2298 defm MemOPh : MemOp_base <"memh", 0b01, u6_1Ext>;
2300 let opExtentBits = 8, accessSize = WordAccess in
2301 defm MemOPw : MemOp_base <"memw", 0b10, u6_2Ext>;
2304 //===----------------------------------------------------------------------===//
2305 // Multiclass to define 'Def Pats' for ALU operations on the memory
2306 // Here value used for the ALU operation is an immediate value.
2307 // mem[bh](Rs+#0) += #U5
2308 // mem[bh](Rs+#u6) += #U5
2309 //===----------------------------------------------------------------------===//
2311 multiclass MemOpi_u5Pats <PatFrag ldOp, PatFrag stOp, PatLeaf ExtPred,
2312 InstHexagon MI, SDNode OpNode> {
2313 let AddedComplexity = 180 in
2314 def : Pat < (stOp (OpNode (ldOp IntRegs:$addr), u5ImmPred:$addend),
2316 (MI IntRegs:$addr, #0, u5ImmPred:$addend )>;
2318 let AddedComplexity = 190 in
2319 def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, ExtPred:$offset)),
2321 (add IntRegs:$base, ExtPred:$offset)),
2322 (MI IntRegs:$base, ExtPred:$offset, u5ImmPred:$addend)>;
2325 multiclass MemOpi_u5ALUOp<PatFrag ldOp, PatFrag stOp, PatLeaf ExtPred,
2326 InstHexagon addMI, InstHexagon subMI> {
2327 defm : MemOpi_u5Pats<ldOp, stOp, ExtPred, addMI, add>;
2328 defm : MemOpi_u5Pats<ldOp, stOp, ExtPred, subMI, sub>;
2331 multiclass MemOpi_u5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2333 defm : MemOpi_u5ALUOp <ldOpHalf, truncstorei16, u6_1ExtPred,
2334 MemOPh_ADDi_V4, MemOPh_SUBi_V4>;
2336 defm : MemOpi_u5ALUOp <ldOpByte, truncstorei8, u6ExtPred,
2337 MemOPb_ADDi_V4, MemOPb_SUBi_V4>;
2340 let Predicates = [HasV4T, UseMEMOP] in {
2341 defm : MemOpi_u5ExtType<zextloadi8, zextloadi16>; // zero extend
2342 defm : MemOpi_u5ExtType<sextloadi8, sextloadi16>; // sign extend
2343 defm : MemOpi_u5ExtType<extloadi8, extloadi16>; // any extend
2346 defm : MemOpi_u5ALUOp <load, store, u6_2ExtPred, MemOPw_ADDi_V4,
2350 //===----------------------------------------------------------------------===//
2351 // multiclass to define 'Def Pats' for ALU operations on the memory.
2352 // Here value used for the ALU operation is a negative value.
2353 // mem[bh](Rs+#0) += #m5
2354 // mem[bh](Rs+#u6) += #m5
2355 //===----------------------------------------------------------------------===//
2357 multiclass MemOpi_m5Pats <PatFrag ldOp, PatFrag stOp, PatLeaf extPred,
2358 PatLeaf immPred, ComplexPattern addrPred,
2359 SDNodeXForm xformFunc, InstHexagon MI> {
2360 let AddedComplexity = 190 in
2361 def : Pat <(stOp (add (ldOp IntRegs:$addr), immPred:$subend),
2363 (MI IntRegs:$addr, #0, (xformFunc immPred:$subend) )>;
2365 let AddedComplexity = 195 in
2366 def : Pat<(stOp (add (ldOp (add IntRegs:$base, extPred:$offset)),
2368 (add IntRegs:$base, extPred:$offset)),
2369 (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$subend))>;
2372 multiclass MemOpi_m5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2374 defm : MemOpi_m5Pats <ldOpHalf, truncstorei16, u6_1ExtPred, m5HImmPred,
2375 ADDRriU6_1, MEMOPIMM_HALF, MemOPh_SUBi_V4>;
2377 defm : MemOpi_m5Pats <ldOpByte, truncstorei8, u6ExtPred, m5BImmPred,
2378 ADDRriU6_0, MEMOPIMM_BYTE, MemOPb_SUBi_V4>;
2381 let Predicates = [HasV4T, UseMEMOP] in {
2382 defm : MemOpi_m5ExtType<zextloadi8, zextloadi16>; // zero extend
2383 defm : MemOpi_m5ExtType<sextloadi8, sextloadi16>; // sign extend
2384 defm : MemOpi_m5ExtType<extloadi8, extloadi16>; // any extend
2387 defm : MemOpi_m5Pats <load, store, u6_2ExtPred, m5ImmPred,
2388 ADDRriU6_2, MEMOPIMM, MemOPw_SUBi_V4>;
2391 //===----------------------------------------------------------------------===//
2392 // Multiclass to define 'def Pats' for bit operations on the memory.
2393 // mem[bhw](Rs+#0) = [clrbit|setbit](#U5)
2394 // mem[bhw](Rs+#u6) = [clrbit|setbit](#U5)
2395 //===----------------------------------------------------------------------===//
2397 multiclass MemOpi_bitPats <PatFrag ldOp, PatFrag stOp, PatLeaf immPred,
2398 PatLeaf extPred, ComplexPattern addrPred,
2399 SDNodeXForm xformFunc, InstHexagon MI, SDNode OpNode> {
2401 // mem[bhw](Rs+#u6:[012]) = [clrbit|setbit](#U5)
2402 let AddedComplexity = 250 in
2403 def : Pat<(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)),
2405 (add IntRegs:$base, extPred:$offset)),
2406 (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$bitend))>;
2408 // mem[bhw](Rs+#0) = [clrbit|setbit](#U5)
2409 let AddedComplexity = 225 in
2410 def : Pat <(stOp (OpNode (ldOp (addrPred IntRegs:$addr, extPred:$offset)),
2412 (addrPred (i32 IntRegs:$addr), extPred:$offset)),
2413 (MI IntRegs:$addr, extPred:$offset, (xformFunc immPred:$bitend))>;
2416 multiclass MemOpi_bitExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2418 defm : MemOpi_bitPats<ldOpByte, truncstorei8, Clr3ImmPred, u6ExtPred,
2419 ADDRriU6_0, CLRMEMIMM_BYTE, MemOPb_CLRBITi_V4, and>;
2421 defm : MemOpi_bitPats<ldOpByte, truncstorei8, Set3ImmPred, u6ExtPred,
2422 ADDRriU6_0, SETMEMIMM_BYTE, MemOPb_SETBITi_V4, or>;
2423 // Half Word - clrbit
2424 defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Clr4ImmPred, u6_1ExtPred,
2425 ADDRriU6_1, CLRMEMIMM_SHORT, MemOPh_CLRBITi_V4, and>;
2426 // Half Word - setbit
2427 defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Set4ImmPred, u6_1ExtPred,
2428 ADDRriU6_1, SETMEMIMM_SHORT, MemOPh_SETBITi_V4, or>;
2431 let Predicates = [HasV4T, UseMEMOP] in {
2432 // mem[bh](Rs+#0) = [clrbit|setbit](#U5)
2433 // mem[bh](Rs+#u6:[01]) = [clrbit|setbit](#U5)
2434 defm : MemOpi_bitExtType<zextloadi8, zextloadi16>; // zero extend
2435 defm : MemOpi_bitExtType<sextloadi8, sextloadi16>; // sign extend
2436 defm : MemOpi_bitExtType<extloadi8, extloadi16>; // any extend
2438 // memw(Rs+#0) = [clrbit|setbit](#U5)
2439 // memw(Rs+#u6:2) = [clrbit|setbit](#U5)
2440 defm : MemOpi_bitPats<load, store, Clr5ImmPred, u6_2ExtPred, ADDRriU6_2,
2441 CLRMEMIMM, MemOPw_CLRBITi_V4, and>;
2442 defm : MemOpi_bitPats<load, store, Set5ImmPred, u6_2ExtPred, ADDRriU6_2,
2443 SETMEMIMM, MemOPw_SETBITi_V4, or>;
2446 //===----------------------------------------------------------------------===//
2447 // Multiclass to define 'def Pats' for ALU operations on the memory
2448 // where addend is a register.
2449 // mem[bhw](Rs+#0) [+-&|]= Rt
2450 // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt
2451 //===----------------------------------------------------------------------===//
2453 multiclass MemOpr_Pats <PatFrag ldOp, PatFrag stOp, ComplexPattern addrPred,
2454 PatLeaf extPred, InstHexagon MI, SDNode OpNode> {
2455 let AddedComplexity = 141 in
2456 // mem[bhw](Rs+#0) [+-&|]= Rt
2457 def : Pat <(stOp (OpNode (ldOp (addrPred IntRegs:$addr, extPred:$offset)),
2458 (i32 IntRegs:$addend)),
2459 (addrPred (i32 IntRegs:$addr), extPred:$offset)),
2460 (MI IntRegs:$addr, extPred:$offset, (i32 IntRegs:$addend) )>;
2462 // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt
2463 let AddedComplexity = 150 in
2464 def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)),
2465 (i32 IntRegs:$orend)),
2466 (add IntRegs:$base, extPred:$offset)),
2467 (MI IntRegs:$base, extPred:$offset, (i32 IntRegs:$orend) )>;
2470 multiclass MemOPr_ALUOp<PatFrag ldOp, PatFrag stOp,
2471 ComplexPattern addrPred, PatLeaf extPred,
2472 InstHexagon addMI, InstHexagon subMI,
2473 InstHexagon andMI, InstHexagon orMI > {
2475 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, addMI, add>;
2476 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, subMI, sub>;
2477 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, andMI, and>;
2478 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, orMI, or>;
2481 multiclass MemOPr_ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2483 defm : MemOPr_ALUOp <ldOpHalf, truncstorei16, ADDRriU6_1, u6_1ExtPred,
2484 MemOPh_ADDr_V4, MemOPh_SUBr_V4,
2485 MemOPh_ANDr_V4, MemOPh_ORr_V4>;
2487 defm : MemOPr_ALUOp <ldOpByte, truncstorei8, ADDRriU6_0, u6ExtPred,
2488 MemOPb_ADDr_V4, MemOPb_SUBr_V4,
2489 MemOPb_ANDr_V4, MemOPb_ORr_V4>;
2492 // Define 'def Pats' for MemOps with register addend.
2493 let Predicates = [HasV4T, UseMEMOP] in {
2495 defm : MemOPr_ExtType<zextloadi8, zextloadi16>; // zero extend
2496 defm : MemOPr_ExtType<sextloadi8, sextloadi16>; // sign extend
2497 defm : MemOPr_ExtType<extloadi8, extloadi16>; // any extend
2499 defm : MemOPr_ALUOp <load, store, ADDRriU6_2, u6_2ExtPred, MemOPw_ADDr_V4,
2500 MemOPw_SUBr_V4, MemOPw_ANDr_V4, MemOPw_ORr_V4 >;
2503 //===----------------------------------------------------------------------===//
2505 //===----------------------------------------------------------------------===//
2507 // Hexagon V4 only supports these flavors of byte/half compare instructions:
2508 // EQ/GT/GTU. Other flavors like GE/GEU/LT/LTU/LE/LEU are not supported by
2509 // hardware. However, compiler can still implement these patterns through
2510 // appropriate patterns combinations based on current implemented patterns.
2511 // The implemented patterns are: EQ/GT/GTU.
2512 // Missing patterns are: GE/GEU/LT/LTU/LE/LEU.
2514 // Following instruction is not being extended as it results into the
2515 // incorrect code for negative numbers.
2516 // Pd=cmpb.eq(Rs,#u8)
2518 let isCompare = 1, isExtendable = 1, opExtendable = 2, hasSideEffects = 0,
2519 validSubTargets = HasV4SubT in
2520 class CMP_NOT_REG_IMM<string OpName, bits<2> op, Operand ImmOp,
2522 : ALU32Inst <(outs PredRegs:$dst), (ins IntRegs:$src1, ImmOp:$src2),
2523 "$dst = !cmp."#OpName#"($src1, #$src2)",
2525 "", ALU32_2op_tc_2early_SLOT0123> {
2530 let IClass = 0b0111;
2531 let Inst{27-24} = 0b0101;
2532 let Inst{23-22} = op;
2533 let Inst{20-16} = src1;
2534 let Inst{21} = !if (!eq(OpName, "gtu"), 0b0, src2{9});
2535 let Inst{13-5} = src2{8-0};
2536 let Inst{4-2} = 0b100;
2537 let Inst{1-0} = dst;
2540 let opExtentBits = 10, isExtentSigned = 1 in {
2541 def C4_cmpneqi : CMP_NOT_REG_IMM <"eq", 0b00, s10Ext, [(set (i1 PredRegs:$dst),
2542 (setne (i32 IntRegs:$src1), s10ExtPred:$src2))]>;
2544 def C4_cmpltei : CMP_NOT_REG_IMM <"gt", 0b01, s10Ext, [(set (i1 PredRegs:$dst),
2545 (not (setgt (i32 IntRegs:$src1), s10ExtPred:$src2)))]>;
2548 let opExtentBits = 9 in
2549 def C4_cmplteui : CMP_NOT_REG_IMM <"gtu", 0b10, u9Ext, [(set (i1 PredRegs:$dst),
2550 (not (setugt (i32 IntRegs:$src1), u9ExtPred:$src2)))]>;
2555 let isCompare = 1, validSubTargets = HasV4SubT in
2556 def CMPnotEQ_rr : ALU32_rr<(outs PredRegs:$dst),
2557 (ins IntRegs:$src1, IntRegs:$src2),
2558 "$dst = !cmp.eq($src1, $src2)",
2559 [(set (i1 PredRegs:$dst),
2560 (setne (i32 IntRegs:$src1), (i32 IntRegs:$src2)))]>,
2564 let isCompare = 1, validSubTargets = HasV4SubT in
2565 def CMPnotGT_rr : ALU32_rr<(outs PredRegs:$dst),
2566 (ins IntRegs:$src1, IntRegs:$src2),
2567 "$dst = !cmp.gt($src1, $src2)",
2568 [(set (i1 PredRegs:$dst),
2569 (not (setgt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))]>,
2573 // p=!cmp.gtu(r1,r2)
2574 let isCompare = 1, validSubTargets = HasV4SubT in
2575 def CMPnotGTU_rr : ALU32_rr<(outs PredRegs:$dst),
2576 (ins IntRegs:$src1, IntRegs:$src2),
2577 "$dst = !cmp.gtu($src1, $src2)",
2578 [(set (i1 PredRegs:$dst),
2579 (not (setugt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))]>,
2582 let isCompare = 1, validSubTargets = HasV4SubT in
2583 def CMPbEQri_V4 : MInst<(outs PredRegs:$dst),
2584 (ins IntRegs:$src1, u8Imm:$src2),
2585 "$dst = cmpb.eq($src1, #$src2)",
2586 [(set (i1 PredRegs:$dst),
2587 (seteq (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2))]>,
2590 def : Pat <(brcond (i1 (setne (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2)),
2592 (J2_jumpf (CMPbEQri_V4 (i32 IntRegs:$src1), u8ImmPred:$src2),
2596 // Pd=cmpb.eq(Rs,Rt)
2597 let isCompare = 1, validSubTargets = HasV4SubT in
2598 def CMPbEQrr_ubub_V4 : MInst<(outs PredRegs:$dst),
2599 (ins IntRegs:$src1, IntRegs:$src2),
2600 "$dst = cmpb.eq($src1, $src2)",
2601 [(set (i1 PredRegs:$dst),
2602 (seteq (and (xor (i32 IntRegs:$src1),
2603 (i32 IntRegs:$src2)), 255), 0))]>,
2606 // Pd=cmpb.eq(Rs,Rt)
2607 let isCompare = 1, validSubTargets = HasV4SubT in
2608 def CMPbEQrr_sbsb_V4 : MInst<(outs PredRegs:$dst),
2609 (ins IntRegs:$src1, IntRegs:$src2),
2610 "$dst = cmpb.eq($src1, $src2)",
2611 [(set (i1 PredRegs:$dst),
2612 (seteq (shl (i32 IntRegs:$src1), (i32 24)),
2613 (shl (i32 IntRegs:$src2), (i32 24))))]>,
2616 // Pd=cmpb.gt(Rs,Rt)
2617 let isCompare = 1, validSubTargets = HasV4SubT in
2618 def CMPbGTrr_V4 : MInst<(outs PredRegs:$dst),
2619 (ins IntRegs:$src1, IntRegs:$src2),
2620 "$dst = cmpb.gt($src1, $src2)",
2621 [(set (i1 PredRegs:$dst),
2622 (setgt (shl (i32 IntRegs:$src1), (i32 24)),
2623 (shl (i32 IntRegs:$src2), (i32 24))))]>,
2626 // Pd=cmpb.gtu(Rs,#u7)
2627 let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7,
2628 isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU", InputType = "imm" in
2629 def CMPbGTUri_V4 : MInst<(outs PredRegs:$dst),
2630 (ins IntRegs:$src1, u7Ext:$src2),
2631 "$dst = cmpb.gtu($src1, #$src2)",
2632 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255),
2633 u7ExtPred:$src2))]>,
2634 Requires<[HasV4T]>, ImmRegRel;
2636 // SDNode for converting immediate C to C-1.
2637 def DEC_CONST_BYTE : SDNodeXForm<imm, [{
2638 // Return the byte immediate const-1 as an SDNode.
2639 int32_t imm = N->getSExtValue();
2640 return XformU7ToU7M1Imm(imm);
2644 // zext( seteq ( and(Rs, 255), u8))
2646 // Pd=cmpb.eq(Rs, #u8)
2647 // if (Pd.new) Rd=#1
2648 // if (!Pd.new) Rd=#0
2649 def : Pat <(i32 (zext (i1 (seteq (i32 (and (i32 IntRegs:$Rs), 255)),
2651 (i32 (TFR_condset_ii (i1 (CMPbEQri_V4 (i32 IntRegs:$Rs),
2657 // zext( setne ( and(Rs, 255), u8))
2659 // Pd=cmpb.eq(Rs, #u8)
2660 // if (Pd.new) Rd=#0
2661 // if (!Pd.new) Rd=#1
2662 def : Pat <(i32 (zext (i1 (setne (i32 (and (i32 IntRegs:$Rs), 255)),
2664 (i32 (TFR_condset_ii (i1 (CMPbEQri_V4 (i32 IntRegs:$Rs),
2670 // zext( seteq (Rs, and(Rt, 255)))
2672 // Pd=cmpb.eq(Rs, Rt)
2673 // if (Pd.new) Rd=#1
2674 // if (!Pd.new) Rd=#0
2675 def : Pat <(i32 (zext (i1 (seteq (i32 IntRegs:$Rt),
2676 (i32 (and (i32 IntRegs:$Rs), 255)))))),
2677 (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs),
2678 (i32 IntRegs:$Rt))),
2683 // zext( setne (Rs, and(Rt, 255)))
2685 // Pd=cmpb.eq(Rs, Rt)
2686 // if (Pd.new) Rd=#0
2687 // if (!Pd.new) Rd=#1
2688 def : Pat <(i32 (zext (i1 (setne (i32 IntRegs:$Rt),
2689 (i32 (and (i32 IntRegs:$Rs), 255)))))),
2690 (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs),
2691 (i32 IntRegs:$Rt))),
2696 // zext( setugt ( and(Rs, 255), u8))
2698 // Pd=cmpb.gtu(Rs, #u8)
2699 // if (Pd.new) Rd=#1
2700 // if (!Pd.new) Rd=#0
2701 def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 255)),
2703 (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs),
2709 // zext( setugt ( and(Rs, 254), u8))
2711 // Pd=cmpb.gtu(Rs, #u8)
2712 // if (Pd.new) Rd=#1
2713 // if (!Pd.new) Rd=#0
2714 def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 254)),
2716 (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs),
2722 // zext( setult ( Rs, Rt))
2724 // Pd=cmp.ltu(Rs, Rt)
2725 // if (Pd.new) Rd=#1
2726 // if (!Pd.new) Rd=#0
2727 // cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs)
2728 def : Pat <(i32 (zext (i1 (setult (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2729 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rt),
2730 (i32 IntRegs:$Rs))),
2735 // zext( setlt ( Rs, Rt))
2737 // Pd=cmp.lt(Rs, Rt)
2738 // if (Pd.new) Rd=#1
2739 // if (!Pd.new) Rd=#0
2740 // cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs)
2741 def : Pat <(i32 (zext (i1 (setlt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2742 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rt),
2743 (i32 IntRegs:$Rs))),
2748 // zext( setugt ( Rs, Rt))
2750 // Pd=cmp.gtu(Rs, Rt)
2751 // if (Pd.new) Rd=#1
2752 // if (!Pd.new) Rd=#0
2753 def : Pat <(i32 (zext (i1 (setugt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2754 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rs),
2755 (i32 IntRegs:$Rt))),
2759 // This pattern interefers with coremark performance, not implementing at this
2762 // zext( setgt ( Rs, Rt))
2764 // Pd=cmp.gt(Rs, Rt)
2765 // if (Pd.new) Rd=#1
2766 // if (!Pd.new) Rd=#0
2769 // zext( setuge ( Rs, Rt))
2771 // Pd=cmp.ltu(Rs, Rt)
2772 // if (Pd.new) Rd=#0
2773 // if (!Pd.new) Rd=#1
2774 // cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs)
2775 def : Pat <(i32 (zext (i1 (setuge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2776 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rt),
2777 (i32 IntRegs:$Rs))),
2782 // zext( setge ( Rs, Rt))
2784 // Pd=cmp.lt(Rs, Rt)
2785 // if (Pd.new) Rd=#0
2786 // if (!Pd.new) Rd=#1
2787 // cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs)
2788 def : Pat <(i32 (zext (i1 (setge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2789 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rt),
2790 (i32 IntRegs:$Rs))),
2795 // zext( setule ( Rs, Rt))
2797 // Pd=cmp.gtu(Rs, Rt)
2798 // if (Pd.new) Rd=#0
2799 // if (!Pd.new) Rd=#1
2800 def : Pat <(i32 (zext (i1 (setule (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2801 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rs),
2802 (i32 IntRegs:$Rt))),
2807 // zext( setle ( Rs, Rt))
2809 // Pd=cmp.gt(Rs, Rt)
2810 // if (Pd.new) Rd=#0
2811 // if (!Pd.new) Rd=#1
2812 def : Pat <(i32 (zext (i1 (setle (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2813 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rs),
2814 (i32 IntRegs:$Rt))),
2819 // zext( setult ( and(Rs, 255), u8))
2820 // Use the isdigit transformation below
2822 // Generate code of the form 'mux_ii(cmpbgtu(Rdd, C-1),0,1)'
2823 // for C code of the form r = ((c>='0') & (c<='9')) ? 1 : 0;.
2824 // The isdigit transformation relies on two 'clever' aspects:
2825 // 1) The data type is unsigned which allows us to eliminate a zero test after
2826 // biasing the expression by 48. We are depending on the representation of
2827 // the unsigned types, and semantics.
2828 // 2) The front end has converted <= 9 into < 10 on entry to LLVM
2831 // retval = ((c>='0') & (c<='9')) ? 1 : 0;
2832 // The code is transformed upstream of llvm into
2833 // retval = (c-48) < 10 ? 1 : 0;
2834 let AddedComplexity = 139 in
2835 def : Pat <(i32 (zext (i1 (setult (i32 (and (i32 IntRegs:$src1), 255)),
2836 u7StrictPosImmPred:$src2)))),
2837 (i32 (C2_muxii (i1 (CMPbGTUri_V4 (i32 IntRegs:$src1),
2838 (DEC_CONST_BYTE u7StrictPosImmPred:$src2))),
2842 // Pd=cmpb.gtu(Rs,Rt)
2843 let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU",
2844 InputType = "reg" in
2845 def CMPbGTUrr_V4 : MInst<(outs PredRegs:$dst),
2846 (ins IntRegs:$src1, IntRegs:$src2),
2847 "$dst = cmpb.gtu($src1, $src2)",
2848 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255),
2849 (and (i32 IntRegs:$src2), 255)))]>,
2850 Requires<[HasV4T]>, ImmRegRel;
2852 // Following instruction is not being extended as it results into the incorrect
2853 // code for negative numbers.
2855 // Signed half compare(.eq) ri.
2856 // Pd=cmph.eq(Rs,#s8)
2857 let isCompare = 1, validSubTargets = HasV4SubT in
2858 def CMPhEQri_V4 : MInst<(outs PredRegs:$dst),
2859 (ins IntRegs:$src1, s8Imm:$src2),
2860 "$dst = cmph.eq($src1, #$src2)",
2861 [(set (i1 PredRegs:$dst), (seteq (and (i32 IntRegs:$src1), 65535),
2862 s8ImmPred:$src2))]>,
2865 // Signed half compare(.eq) rr.
2866 // Case 1: xor + and, then compare:
2868 // r0=and(r0,#0xffff)
2870 // Pd=cmph.eq(Rs,Rt)
2871 let isCompare = 1, validSubTargets = HasV4SubT in
2872 def CMPhEQrr_xor_V4 : MInst<(outs PredRegs:$dst),
2873 (ins IntRegs:$src1, IntRegs:$src2),
2874 "$dst = cmph.eq($src1, $src2)",
2875 [(set (i1 PredRegs:$dst), (seteq (and (xor (i32 IntRegs:$src1),
2876 (i32 IntRegs:$src2)),
2880 // Signed half compare(.eq) rr.
2881 // Case 2: shift left 16 bits then compare:
2885 // Pd=cmph.eq(Rs,Rt)
2886 let isCompare = 1, validSubTargets = HasV4SubT in
2887 def CMPhEQrr_shl_V4 : MInst<(outs PredRegs:$dst),
2888 (ins IntRegs:$src1, IntRegs:$src2),
2889 "$dst = cmph.eq($src1, $src2)",
2890 [(set (i1 PredRegs:$dst),
2891 (seteq (shl (i32 IntRegs:$src1), (i32 16)),
2892 (shl (i32 IntRegs:$src2), (i32 16))))]>,
2895 /* Incorrect Pattern -- immediate should be right shifted before being
2896 used in the cmph.gt instruction.
2897 // Signed half compare(.gt) ri.
2898 // Pd=cmph.gt(Rs,#s8)
2900 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 8,
2901 isCompare = 1, validSubTargets = HasV4SubT in
2902 def CMPhGTri_V4 : MInst<(outs PredRegs:$dst),
2903 (ins IntRegs:$src1, s8Ext:$src2),
2904 "$dst = cmph.gt($src1, #$src2)",
2905 [(set (i1 PredRegs:$dst),
2906 (setgt (shl (i32 IntRegs:$src1), (i32 16)),
2907 s8ExtPred:$src2))]>,
2911 // Signed half compare(.gt) rr.
2912 // Pd=cmph.gt(Rs,Rt)
2913 let isCompare = 1, validSubTargets = HasV4SubT in
2914 def CMPhGTrr_shl_V4 : MInst<(outs PredRegs:$dst),
2915 (ins IntRegs:$src1, IntRegs:$src2),
2916 "$dst = cmph.gt($src1, $src2)",
2917 [(set (i1 PredRegs:$dst),
2918 (setgt (shl (i32 IntRegs:$src1), (i32 16)),
2919 (shl (i32 IntRegs:$src2), (i32 16))))]>,
2922 // Unsigned half compare rr (.gtu).
2923 // Pd=cmph.gtu(Rs,Rt)
2924 let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU",
2925 InputType = "reg" in
2926 def CMPhGTUrr_V4 : MInst<(outs PredRegs:$dst),
2927 (ins IntRegs:$src1, IntRegs:$src2),
2928 "$dst = cmph.gtu($src1, $src2)",
2929 [(set (i1 PredRegs:$dst),
2930 (setugt (and (i32 IntRegs:$src1), 65535),
2931 (and (i32 IntRegs:$src2), 65535)))]>,
2932 Requires<[HasV4T]>, ImmRegRel;
2934 // Unsigned half compare ri (.gtu).
2935 // Pd=cmph.gtu(Rs,#u7)
2936 let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7,
2937 isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU",
2938 InputType = "imm" in
2939 def CMPhGTUri_V4 : MInst<(outs PredRegs:$dst),
2940 (ins IntRegs:$src1, u7Ext:$src2),
2941 "$dst = cmph.gtu($src1, #$src2)",
2942 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 65535),
2943 u7ExtPred:$src2))]>,
2944 Requires<[HasV4T]>, ImmRegRel;
2946 let validSubTargets = HasV4SubT in
2947 def NTSTBIT_rr : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
2948 "$dst = !tstbit($src1, $src2)",
2949 [(set (i1 PredRegs:$dst),
2950 (seteq (and (shl 1, (i32 IntRegs:$src2)), (i32 IntRegs:$src1)), 0))]>,
2953 let validSubTargets = HasV4SubT in
2954 def NTSTBIT_ri : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
2955 "$dst = !tstbit($src1, $src2)",
2956 [(set (i1 PredRegs:$dst),
2957 (seteq (and (shl 1, u5ImmPred:$src2), (i32 IntRegs:$src1)), 0))]>,
2960 //===----------------------------------------------------------------------===//
2962 //===----------------------------------------------------------------------===//
2964 //Deallocate frame and return.
2966 let isReturn = 1, isTerminator = 1, isBarrier = 1, isPredicable = 1,
2967 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0 in {
2968 let validSubTargets = HasV4SubT in
2969 def DEALLOC_RET_V4 : LD0Inst<(outs), (ins),
2975 // Restore registers and dealloc return function call.
2976 let isCall = 1, isBarrier = 1, isReturn = 1, isTerminator = 1,
2977 Defs = [R29, R30, R31, PC] in {
2978 let validSubTargets = HasV4SubT in
2979 def RESTORE_DEALLOC_RET_JMP_V4 : JInst<(outs),
2980 (ins calltarget:$dst),
2986 // Restore registers and dealloc frame before a tail call.
2987 let isCall = 1, isBarrier = 1,
2988 Defs = [R29, R30, R31, PC] in {
2989 let validSubTargets = HasV4SubT in
2990 def RESTORE_DEALLOC_BEFORE_TAILCALL_V4 : JInst<(outs),
2991 (ins calltarget:$dst),
2997 // Save registers function call.
2998 let isCall = 1, isBarrier = 1,
2999 Uses = [R29, R31] in {
3000 def SAVE_REGISTERS_CALL_V4 : JInst<(outs),
3001 (ins calltarget:$dst),
3002 "call $dst // Save_calle_saved_registers",
3007 // if (Ps) dealloc_return
3008 let isReturn = 1, isTerminator = 1,
3009 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3010 isPredicated = 1 in {
3011 let validSubTargets = HasV4SubT in
3012 def DEALLOC_RET_cPt_V4 : LD0Inst<(outs),
3013 (ins PredRegs:$src1),
3014 "if ($src1) dealloc_return",
3019 // if (!Ps) dealloc_return
3020 let isReturn = 1, isTerminator = 1,
3021 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3022 isPredicated = 1, isPredicatedFalse = 1 in {
3023 let validSubTargets = HasV4SubT in
3024 def DEALLOC_RET_cNotPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
3025 "if (!$src1) dealloc_return",
3030 // if (Ps.new) dealloc_return:nt
3031 let isReturn = 1, isTerminator = 1,
3032 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3033 isPredicated = 1 in {
3034 let validSubTargets = HasV4SubT in
3035 def DEALLOC_RET_cdnPnt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
3036 "if ($src1.new) dealloc_return:nt",
3041 // if (!Ps.new) dealloc_return:nt
3042 let isReturn = 1, isTerminator = 1,
3043 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3044 isPredicated = 1, isPredicatedFalse = 1 in {
3045 let validSubTargets = HasV4SubT in
3046 def DEALLOC_RET_cNotdnPnt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
3047 "if (!$src1.new) dealloc_return:nt",
3052 // if (Ps.new) dealloc_return:t
3053 let isReturn = 1, isTerminator = 1,
3054 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3055 isPredicated = 1 in {
3056 let validSubTargets = HasV4SubT in
3057 def DEALLOC_RET_cdnPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
3058 "if ($src1.new) dealloc_return:t",
3063 // if (!Ps.new) dealloc_return:nt
3064 let isReturn = 1, isTerminator = 1,
3065 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3066 isPredicated = 1, isPredicatedFalse = 1 in {
3067 let validSubTargets = HasV4SubT in
3068 def DEALLOC_RET_cNotdnPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
3069 "if (!$src1.new) dealloc_return:t",
3074 // Load/Store with absolute addressing mode
3077 multiclass ST_Abs_Predbase<string mnemonic, RegisterClass RC, bit isNot,
3079 let isPredicatedNew = isPredNew in
3080 def NAME#_V4 : STInst2<(outs),
3081 (ins PredRegs:$src1, u0AlwaysExt:$absaddr, RC: $src2),
3082 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
3083 ") ")#mnemonic#"(##$absaddr) = $src2",
3088 multiclass ST_Abs_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
3089 let isPredicatedFalse = PredNot in {
3090 defm _c#NAME : ST_Abs_Predbase<mnemonic, RC, PredNot, 0>;
3092 defm _cdn#NAME : ST_Abs_Predbase<mnemonic, RC, PredNot, 1>;
3096 let isNVStorable = 1, isExtended = 1, hasSideEffects = 0 in
3097 multiclass ST_Abs<string mnemonic, string CextOp, RegisterClass RC> {
3098 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
3099 let opExtendable = 0, isPredicable = 1 in
3100 def NAME#_V4 : STInst2<(outs),
3101 (ins u0AlwaysExt:$absaddr, RC:$src),
3102 mnemonic#"(##$absaddr) = $src",
3106 let opExtendable = 1, isPredicated = 1 in {
3107 defm Pt : ST_Abs_Pred<mnemonic, RC, 0>;
3108 defm NotPt : ST_Abs_Pred<mnemonic, RC, 1>;
3113 multiclass ST_Abs_Predbase_nv<string mnemonic, RegisterClass RC, bit isNot,
3115 let isPredicatedNew = isPredNew in
3116 def NAME#_nv_V4 : NVInst_V4<(outs),
3117 (ins PredRegs:$src1, u0AlwaysExt:$absaddr, RC: $src2),
3118 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
3119 ") ")#mnemonic#"(##$absaddr) = $src2.new",
3124 multiclass ST_Abs_Pred_nv<string mnemonic, RegisterClass RC, bit PredNot> {
3125 let isPredicatedFalse = PredNot in {
3126 defm _c#NAME : ST_Abs_Predbase_nv<mnemonic, RC, PredNot, 0>;
3128 defm _cdn#NAME : ST_Abs_Predbase_nv<mnemonic, RC, PredNot, 1>;
3132 let mayStore = 1, isNVStore = 1, isExtended = 1, hasSideEffects = 0 in
3133 multiclass ST_Abs_nv<string mnemonic, string CextOp, RegisterClass RC> {
3134 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
3135 let opExtendable = 0, isPredicable = 1 in
3136 def NAME#_nv_V4 : NVInst_V4<(outs),
3137 (ins u0AlwaysExt:$absaddr, RC:$src),
3138 mnemonic#"(##$absaddr) = $src.new",
3142 let opExtendable = 1, isPredicated = 1 in {
3143 defm Pt : ST_Abs_Pred_nv<mnemonic, RC, 0>;
3144 defm NotPt : ST_Abs_Pred_nv<mnemonic, RC, 1>;
3149 let addrMode = Absolute in {
3150 let accessSize = ByteAccess in
3151 defm STrib_abs : ST_Abs<"memb", "STrib", IntRegs>,
3152 ST_Abs_nv<"memb", "STrib", IntRegs>, AddrModeRel;
3154 let accessSize = HalfWordAccess in
3155 defm STrih_abs : ST_Abs<"memh", "STrih", IntRegs>,
3156 ST_Abs_nv<"memh", "STrih", IntRegs>, AddrModeRel;
3158 let accessSize = WordAccess in
3159 defm STriw_abs : ST_Abs<"memw", "STriw", IntRegs>,
3160 ST_Abs_nv<"memw", "STriw", IntRegs>, AddrModeRel;
3162 let accessSize = DoubleWordAccess, isNVStorable = 0 in
3163 defm STrid_abs : ST_Abs<"memd", "STrid", DoubleRegs>, AddrModeRel;
3166 let Predicates = [HasV4T], AddedComplexity = 30 in {
3167 def : Pat<(truncstorei8 (i32 IntRegs:$src1),
3168 (HexagonCONST32 tglobaladdr:$absaddr)),
3169 (STrib_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
3171 def : Pat<(truncstorei16 (i32 IntRegs:$src1),
3172 (HexagonCONST32 tglobaladdr:$absaddr)),
3173 (STrih_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
3175 def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32 tglobaladdr:$absaddr)),
3176 (STriw_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
3178 def : Pat<(store (i64 DoubleRegs:$src1),
3179 (HexagonCONST32 tglobaladdr:$absaddr)),
3180 (STrid_abs_V4 tglobaladdr: $absaddr, DoubleRegs: $src1)>;
3183 //===----------------------------------------------------------------------===//
3184 // multiclass for store instructions with GP-relative addressing mode.
3185 // mem[bhwd](#global)=Rt
3186 // if ([!]Pv[.new]) mem[bhwd](##global) = Rt
3187 //===----------------------------------------------------------------------===//
3188 let mayStore = 1, isNVStorable = 1 in
3189 multiclass ST_GP<string mnemonic, string BaseOp, RegisterClass RC> {
3190 let BaseOpcode = BaseOp, isPredicable = 1 in
3191 def NAME#_V4 : STInst2<(outs),
3192 (ins globaladdress:$global, RC:$src),
3193 mnemonic#"(#$global) = $src",
3196 // When GP-relative instructions are predicated, their addressing mode is
3197 // changed to absolute and they are always constant extended.
3198 let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1,
3199 isPredicated = 1 in {
3200 defm Pt : ST_Abs_Pred <mnemonic, RC, 0>;
3201 defm NotPt : ST_Abs_Pred <mnemonic, RC, 1>;
3205 let mayStore = 1, isNVStore = 1 in
3206 multiclass ST_GP_nv<string mnemonic, string BaseOp, RegisterClass RC> {
3207 let BaseOpcode = BaseOp, isPredicable = 1 in
3208 def NAME#_nv_V4 : NVInst_V4<(outs),
3209 (ins u0AlwaysExt:$global, RC:$src),
3210 mnemonic#"(#$global) = $src.new",
3214 // When GP-relative instructions are predicated, their addressing mode is
3215 // changed to absolute and they are always constant extended.
3216 let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1,
3217 isPredicated = 1 in {
3218 defm Pt : ST_Abs_Pred_nv<mnemonic, RC, 0>;
3219 defm NotPt : ST_Abs_Pred_nv<mnemonic, RC, 1>;
3223 let validSubTargets = HasV4SubT, hasSideEffects = 0 in {
3224 let isNVStorable = 0 in
3225 defm STd_GP : ST_GP <"memd", "STd_GP", DoubleRegs>, PredNewRel;
3227 defm STb_GP : ST_GP<"memb", "STb_GP", IntRegs>,
3228 ST_GP_nv<"memb", "STb_GP", IntRegs>, NewValueRel;
3229 defm STh_GP : ST_GP<"memh", "STh_GP", IntRegs>,
3230 ST_GP_nv<"memh", "STh_GP", IntRegs>, NewValueRel;
3231 defm STw_GP : ST_GP<"memw", "STw_GP", IntRegs>,
3232 ST_GP_nv<"memw", "STw_GP", IntRegs>, NewValueRel;
3235 // 64 bit atomic store
3236 def : Pat <(atomic_store_64 (HexagonCONST32_GP tglobaladdr:$global),
3237 (i64 DoubleRegs:$src1)),
3238 (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>,
3241 // Map from store(globaladdress) -> memd(#foo)
3242 let AddedComplexity = 100 in
3243 def : Pat <(store (i64 DoubleRegs:$src1),
3244 (HexagonCONST32_GP tglobaladdr:$global)),
3245 (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>;
3247 // 8 bit atomic store
3248 def : Pat < (atomic_store_8 (HexagonCONST32_GP tglobaladdr:$global),
3249 (i32 IntRegs:$src1)),
3250 (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3252 // Map from store(globaladdress) -> memb(#foo)
3253 let AddedComplexity = 100 in
3254 def : Pat<(truncstorei8 (i32 IntRegs:$src1),
3255 (HexagonCONST32_GP tglobaladdr:$global)),
3256 (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3258 // Map from "i1 = constant<-1>; memw(CONST32(#foo)) = i1"
3259 // to "r0 = 1; memw(#foo) = r0"
3260 let AddedComplexity = 100 in
3261 def : Pat<(store (i1 -1), (HexagonCONST32_GP tglobaladdr:$global)),
3262 (STb_GP_V4 tglobaladdr:$global, (A2_tfrsi 1))>;
3264 def : Pat<(atomic_store_16 (HexagonCONST32_GP tglobaladdr:$global),
3265 (i32 IntRegs:$src1)),
3266 (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3268 // Map from store(globaladdress) -> memh(#foo)
3269 let AddedComplexity = 100 in
3270 def : Pat<(truncstorei16 (i32 IntRegs:$src1),
3271 (HexagonCONST32_GP tglobaladdr:$global)),
3272 (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3274 // 32 bit atomic store
3275 def : Pat<(atomic_store_32 (HexagonCONST32_GP tglobaladdr:$global),
3276 (i32 IntRegs:$src1)),
3277 (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3279 // Map from store(globaladdress) -> memw(#foo)
3280 let AddedComplexity = 100 in
3281 def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)),
3282 (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
3284 //===----------------------------------------------------------------------===//
3285 // Multiclass for the load instructions with absolute addressing mode.
3286 //===----------------------------------------------------------------------===//
3287 multiclass LD_Abs_Predbase<string mnemonic, RegisterClass RC, bit isNot,
3289 let isPredicatedNew = isPredNew in
3290 def NAME : LDInst2<(outs RC:$dst),
3291 (ins PredRegs:$src1, u0AlwaysExt:$absaddr),
3292 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
3293 ") ")#"$dst = "#mnemonic#"(##$absaddr)",
3298 multiclass LD_Abs_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
3299 let isPredicatedFalse = PredNot in {
3300 defm _c#NAME : LD_Abs_Predbase<mnemonic, RC, PredNot, 0>;
3302 defm _cdn#NAME : LD_Abs_Predbase<mnemonic, RC, PredNot, 1>;
3306 let isExtended = 1, hasSideEffects = 0 in
3307 multiclass LD_Abs<string mnemonic, string CextOp, RegisterClass RC> {
3308 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
3309 let opExtendable = 1, isPredicable = 1 in
3310 def NAME#_V4 : LDInst2<(outs RC:$dst),
3311 (ins u0AlwaysExt:$absaddr),
3312 "$dst = "#mnemonic#"(##$absaddr)",
3316 let opExtendable = 2, isPredicated = 1 in {
3317 defm Pt_V4 : LD_Abs_Pred<mnemonic, RC, 0>;
3318 defm NotPt_V4 : LD_Abs_Pred<mnemonic, RC, 1>;
3323 let addrMode = Absolute in {
3324 let accessSize = ByteAccess in {
3325 defm LDrib_abs : LD_Abs<"memb", "LDrib", IntRegs>, AddrModeRel;
3326 defm LDriub_abs : LD_Abs<"memub", "LDriub", IntRegs>, AddrModeRel;
3328 let accessSize = HalfWordAccess in {
3329 defm LDrih_abs : LD_Abs<"memh", "LDrih", IntRegs>, AddrModeRel;
3330 defm LDriuh_abs : LD_Abs<"memuh", "LDriuh", IntRegs>, AddrModeRel;
3332 let accessSize = WordAccess in
3333 defm LDriw_abs : LD_Abs<"memw", "LDriw", IntRegs>, AddrModeRel;
3335 let accessSize = DoubleWordAccess in
3336 defm LDrid_abs : LD_Abs<"memd", "LDrid", DoubleRegs>, AddrModeRel;
3339 let Predicates = [HasV4T], AddedComplexity = 30 in {
3340 def : Pat<(i32 (load (HexagonCONST32 tglobaladdr:$absaddr))),
3341 (LDriw_abs_V4 tglobaladdr: $absaddr)>;
3343 def : Pat<(i32 (sextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))),
3344 (LDrib_abs_V4 tglobaladdr:$absaddr)>;
3346 def : Pat<(i32 (zextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))),
3347 (LDriub_abs_V4 tglobaladdr:$absaddr)>;
3349 def : Pat<(i32 (sextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
3350 (LDrih_abs_V4 tglobaladdr:$absaddr)>;
3352 def : Pat<(i32 (zextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
3353 (LDriuh_abs_V4 tglobaladdr:$absaddr)>;
3356 //===----------------------------------------------------------------------===//
3357 // multiclass for load instructions with GP-relative addressing mode.
3358 // Rx=mem[bhwd](##global)
3359 // if ([!]Pv[.new]) Rx=mem[bhwd](##global)
3360 //===----------------------------------------------------------------------===//
3361 let hasSideEffects = 0, validSubTargets = HasV4SubT in
3362 multiclass LD_GP<string mnemonic, string BaseOp, RegisterClass RC> {
3363 let BaseOpcode = BaseOp in {
3364 let isPredicable = 1 in
3365 def NAME#_V4 : LDInst2<(outs RC:$dst),
3366 (ins globaladdress:$global),
3367 "$dst = "#mnemonic#"(#$global)",
3370 let isExtended = 1, opExtendable = 2, isPredicated = 1 in {
3371 defm Pt_V4 : LD_Abs_Pred<mnemonic, RC, 0>;
3372 defm NotPt_V4 : LD_Abs_Pred<mnemonic, RC, 1>;
3377 defm LDd_GP : LD_GP<"memd", "LDd_GP", DoubleRegs>, PredNewRel;
3378 defm LDb_GP : LD_GP<"memb", "LDb_GP", IntRegs>, PredNewRel;
3379 defm LDub_GP : LD_GP<"memub", "LDub_GP", IntRegs>, PredNewRel;
3380 defm LDh_GP : LD_GP<"memh", "LDh_GP", IntRegs>, PredNewRel;
3381 defm LDuh_GP : LD_GP<"memuh", "LDuh_GP", IntRegs>, PredNewRel;
3382 defm LDw_GP : LD_GP<"memw", "LDw_GP", IntRegs>, PredNewRel;
3384 def : Pat <(atomic_load_64 (HexagonCONST32_GP tglobaladdr:$global)),
3385 (i64 (LDd_GP_V4 tglobaladdr:$global))>;
3387 def : Pat <(atomic_load_32 (HexagonCONST32_GP tglobaladdr:$global)),
3388 (i32 (LDw_GP_V4 tglobaladdr:$global))>;
3390 def : Pat <(atomic_load_16 (HexagonCONST32_GP tglobaladdr:$global)),
3391 (i32 (LDuh_GP_V4 tglobaladdr:$global))>;
3393 def : Pat <(atomic_load_8 (HexagonCONST32_GP tglobaladdr:$global)),
3394 (i32 (LDub_GP_V4 tglobaladdr:$global))>;
3396 // Map from load(globaladdress) -> memw(#foo + 0)
3397 let AddedComplexity = 100 in
3398 def : Pat <(i64 (load (HexagonCONST32_GP tglobaladdr:$global))),
3399 (i64 (LDd_GP_V4 tglobaladdr:$global))>;
3401 // Map from Pd = load(globaladdress) -> Rd = memb(globaladdress), Pd = Rd
3402 let AddedComplexity = 100 in
3403 def : Pat <(i1 (load (HexagonCONST32_GP tglobaladdr:$global))),
3404 (i1 (C2_tfrrp (i32 (LDb_GP_V4 tglobaladdr:$global))))>;
3406 // When the Interprocedural Global Variable optimizer realizes that a certain
3407 // global variable takes only two constant values, it shrinks the global to
3408 // a boolean. Catch those loads here in the following 3 patterns.
3409 let AddedComplexity = 100 in
3410 def : Pat <(i32 (extloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3411 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3413 let AddedComplexity = 100 in
3414 def : Pat <(i32 (sextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3415 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3417 // Map from load(globaladdress) -> memb(#foo)
3418 let AddedComplexity = 100 in
3419 def : Pat <(i32 (extloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3420 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3422 // Map from load(globaladdress) -> memb(#foo)
3423 let AddedComplexity = 100 in
3424 def : Pat <(i32 (sextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3425 (i32 (LDb_GP_V4 tglobaladdr:$global))>;
3427 let AddedComplexity = 100 in
3428 def : Pat <(i32 (zextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3429 (i32 (LDub_GP_V4 tglobaladdr:$global))>;
3431 // Map from load(globaladdress) -> memub(#foo)
3432 let AddedComplexity = 100 in
3433 def : Pat <(i32 (zextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3434 (i32 (LDub_GP_V4 tglobaladdr:$global))>;
3436 // Map from load(globaladdress) -> memh(#foo)
3437 let AddedComplexity = 100 in
3438 def : Pat <(i32 (extloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3439 (i32 (LDh_GP_V4 tglobaladdr:$global))>;
3441 // Map from load(globaladdress) -> memh(#foo)
3442 let AddedComplexity = 100 in
3443 def : Pat <(i32 (sextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3444 (i32 (LDh_GP_V4 tglobaladdr:$global))>;
3446 // Map from load(globaladdress) -> memuh(#foo)
3447 let AddedComplexity = 100 in
3448 def : Pat <(i32 (zextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3449 (i32 (LDuh_GP_V4 tglobaladdr:$global))>;
3451 // Map from load(globaladdress) -> memw(#foo)
3452 let AddedComplexity = 100 in
3453 def : Pat <(i32 (load (HexagonCONST32_GP tglobaladdr:$global))),
3454 (i32 (LDw_GP_V4 tglobaladdr:$global))>;
3457 // Transfer global address into a register
3458 let isExtended = 1, opExtendable = 1, AddedComplexity=50, isMoveImm = 1,
3459 isAsCheapAsAMove = 1, isReMaterializable = 1, validSubTargets = HasV4SubT in
3460 def TFRI_V4 : ALU32_ri<(outs IntRegs:$dst), (ins s16Ext:$src1),
3462 [(set IntRegs:$dst, (HexagonCONST32 tglobaladdr:$src1))]>,
3465 // Transfer a block address into a register
3466 def : Pat<(HexagonCONST32_GP tblockaddress:$src1),
3467 (TFRI_V4 tblockaddress:$src1)>,
3470 let isExtended = 1, opExtendable = 2, AddedComplexity=50,
3471 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3472 def TFRI_cPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3473 (ins PredRegs:$src1, s16Ext:$src2),
3474 "if($src1) $dst = #$src2",
3478 let isExtended = 1, opExtendable = 2, AddedComplexity=50, isPredicatedFalse = 1,
3479 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3480 def TFRI_cNotPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3481 (ins PredRegs:$src1, s16Ext:$src2),
3482 "if(!$src1) $dst = #$src2",
3486 let isExtended = 1, opExtendable = 2, AddedComplexity=50,
3487 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3488 def TFRI_cdnPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3489 (ins PredRegs:$src1, s16Ext:$src2),
3490 "if($src1.new) $dst = #$src2",
3494 let isExtended = 1, opExtendable = 2, AddedComplexity=50, isPredicatedFalse = 1,
3495 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3496 def TFRI_cdnNotPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3497 (ins PredRegs:$src1, s16Ext:$src2),
3498 "if(!$src1.new) $dst = #$src2",
3502 let AddedComplexity = 50, Predicates = [HasV4T] in
3503 def : Pat<(HexagonCONST32_GP tglobaladdr:$src1),
3504 (TFRI_V4 tglobaladdr:$src1)>,
3508 // Load - Indirect with long offset: These instructions take global address
3510 let isExtended = 1, opExtendable = 3, AddedComplexity = 40,
3511 validSubTargets = HasV4SubT in
3512 def LDrid_ind_lo_V4 : LDInst<(outs DoubleRegs:$dst),
3513 (ins IntRegs:$src1, u2Imm:$src2, globaladdressExt:$offset),
3514 "$dst=memd($src1<<#$src2+##$offset)",
3515 [(set (i64 DoubleRegs:$dst),
3516 (load (add (shl IntRegs:$src1, u2ImmPred:$src2),
3517 (HexagonCONST32 tglobaladdr:$offset))))]>,
3520 let AddedComplexity = 40 in
3521 multiclass LD_indirect_lo<string OpcStr, PatFrag OpNode> {
3522 let isExtended = 1, opExtendable = 3, validSubTargets = HasV4SubT in
3523 def _lo_V4 : LDInst<(outs IntRegs:$dst),
3524 (ins IntRegs:$src1, u2Imm:$src2, globaladdressExt:$offset),
3525 !strconcat("$dst = ",
3526 !strconcat(OpcStr, "($src1<<#$src2+##$offset)")),
3528 (i32 (OpNode (add (shl IntRegs:$src1, u2ImmPred:$src2),
3529 (HexagonCONST32 tglobaladdr:$offset)))))]>,
3533 defm LDrib_ind : LD_indirect_lo<"memb", sextloadi8>;
3534 defm LDriub_ind : LD_indirect_lo<"memub", zextloadi8>;
3535 defm LDriub_ind_anyext : LD_indirect_lo<"memub", extloadi8>;
3536 defm LDrih_ind : LD_indirect_lo<"memh", sextloadi16>;
3537 defm LDriuh_ind : LD_indirect_lo<"memuh", zextloadi16>;
3538 defm LDriuh_ind_anyext : LD_indirect_lo<"memuh", extloadi16>;
3539 defm LDriw_ind : LD_indirect_lo<"memw", load>;
3541 let AddedComplexity = 40 in
3542 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1,
3543 (NumUsesBelowThresCONST32 tglobaladdr:$offset)))),
3544 (i32 (LDrib_ind_lo_V4 IntRegs:$src1, 0, tglobaladdr:$offset))>,
3547 let AddedComplexity = 40 in
3548 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1,
3549 (NumUsesBelowThresCONST32 tglobaladdr:$offset)))),
3550 (i32 (LDriub_ind_lo_V4 IntRegs:$src1, 0, tglobaladdr:$offset))>,
3553 let Predicates = [HasV4T], AddedComplexity = 30 in {
3554 def : Pat<(truncstorei8 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3555 (STrib_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3557 def : Pat<(truncstorei16 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3558 (STrih_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3560 def : Pat<(store (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3561 (STriw_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3564 let Predicates = [HasV4T], AddedComplexity = 30 in {
3565 def : Pat<(i32 (load u0AlwaysExtPred:$src)),
3566 (LDriw_abs_V4 u0AlwaysExtPred:$src)>;
3568 def : Pat<(i32 (sextloadi8 u0AlwaysExtPred:$src)),
3569 (LDrib_abs_V4 u0AlwaysExtPred:$src)>;
3571 def : Pat<(i32 (zextloadi8 u0AlwaysExtPred:$src)),
3572 (LDriub_abs_V4 u0AlwaysExtPred:$src)>;
3574 def : Pat<(i32 (sextloadi16 u0AlwaysExtPred:$src)),
3575 (LDrih_abs_V4 u0AlwaysExtPred:$src)>;
3577 def : Pat<(i32 (zextloadi16 u0AlwaysExtPred:$src)),
3578 (LDriuh_abs_V4 u0AlwaysExtPred:$src)>;
3581 // Indexed store word - global address.
3582 // memw(Rs+#u6:2)=#S8
3583 let AddedComplexity = 10 in
3584 def STriw_offset_ext_V4 : STInst<(outs),
3585 (ins IntRegs:$src1, u6_2Imm:$src2, globaladdress:$src3),
3586 "memw($src1+#$src2) = ##$src3",
3587 [(store (HexagonCONST32 tglobaladdr:$src3),
3588 (add IntRegs:$src1, u6_2ImmPred:$src2))]>,
3591 def : Pat<(i64 (ctlz (i64 DoubleRegs:$src1))),
3592 (i64 (A4_combineir (i32 0), (i32 (CTLZ64_rr DoubleRegs:$src1))))>,
3595 def : Pat<(i64 (cttz (i64 DoubleRegs:$src1))),
3596 (i64 (A4_combineir (i32 0), (i32 (CTTZ64_rr DoubleRegs:$src1))))>,
3601 // We need a complexity of 120 here to override preceding handling of
3603 let Predicates = [HasV4T], AddedComplexity = 120 in {
3604 def: Pat <(i64 (extloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3605 (i64 (A4_combineir 0, (LDrib_abs_V4 tglobaladdr:$addr)))>;
3607 def: Pat <(i64 (zextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3608 (i64 (A4_combineir 0, (LDriub_abs_V4 tglobaladdr:$addr)))>;
3610 def: Pat <(i64 (sextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3611 (i64 (A2_sxtw (LDrib_abs_V4 tglobaladdr:$addr)))>;
3613 def: Pat <(i64 (extloadi8 FoldGlobalAddr:$addr)),
3614 (i64 (A4_combineir 0, (LDrib_abs_V4 FoldGlobalAddr:$addr)))>;
3616 def: Pat <(i64 (zextloadi8 FoldGlobalAddr:$addr)),
3617 (i64 (A4_combineir 0, (LDriub_abs_V4 FoldGlobalAddr:$addr)))>;
3619 def: Pat <(i64 (sextloadi8 FoldGlobalAddr:$addr)),
3620 (i64 (A2_sxtw (LDrib_abs_V4 FoldGlobalAddr:$addr)))>;
3623 // We need a complexity of 120 here to override preceding handling of
3625 let AddedComplexity = 120 in {
3626 def: Pat <(i64 (extloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3627 (i64 (A4_combineir 0, (LDrih_abs_V4 tglobaladdr:$addr)))>,
3630 def: Pat <(i64 (zextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3631 (i64 (A4_combineir 0, (LDriuh_abs_V4 tglobaladdr:$addr)))>,
3634 def: Pat <(i64 (sextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3635 (i64 (A2_sxtw (LDrih_abs_V4 tglobaladdr:$addr)))>,
3638 def: Pat <(i64 (extloadi16 FoldGlobalAddr:$addr)),
3639 (i64 (A4_combineir 0, (LDrih_abs_V4 FoldGlobalAddr:$addr)))>,
3642 def: Pat <(i64 (zextloadi16 FoldGlobalAddr:$addr)),
3643 (i64 (A4_combineir 0, (LDriuh_abs_V4 FoldGlobalAddr:$addr)))>,
3646 def: Pat <(i64 (sextloadi16 FoldGlobalAddr:$addr)),
3647 (i64 (A2_sxtw (LDrih_abs_V4 FoldGlobalAddr:$addr)))>,
3651 // We need a complexity of 120 here to override preceding handling of
3653 let AddedComplexity = 120 in {
3654 def: Pat <(i64 (extloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3655 (i64 (A4_combineir 0, (LDriw_abs_V4 tglobaladdr:$addr)))>,
3658 def: Pat <(i64 (zextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3659 (i64 (A4_combineir 0, (LDriw_abs_V4 tglobaladdr:$addr)))>,
3662 def: Pat <(i64 (sextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3663 (i64 (A2_sxtw (LDriw_abs_V4 tglobaladdr:$addr)))>,
3666 def: Pat <(i64 (extloadi32 FoldGlobalAddr:$addr)),
3667 (i64 (A4_combineir 0, (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
3670 def: Pat <(i64 (zextloadi32 FoldGlobalAddr:$addr)),
3671 (i64 (A4_combineir 0, (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
3674 def: Pat <(i64 (sextloadi32 FoldGlobalAddr:$addr)),
3675 (i64 (A2_sxtw (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
3679 // Indexed store double word - global address.
3680 // memw(Rs+#u6:2)=#S8
3681 let AddedComplexity = 10 in
3682 def STrih_offset_ext_V4 : STInst<(outs),
3683 (ins IntRegs:$src1, u6_1Imm:$src2, globaladdress:$src3),
3684 "memh($src1+#$src2) = ##$src3",
3685 [(truncstorei16 (HexagonCONST32 tglobaladdr:$src3),
3686 (add IntRegs:$src1, u6_1ImmPred:$src2))]>,
3688 // Map from store(globaladdress + x) -> memd(#foo + x)
3689 let AddedComplexity = 100 in
3690 def : Pat<(store (i64 DoubleRegs:$src1),
3691 FoldGlobalAddrGP:$addr),
3692 (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
3695 def : Pat<(atomic_store_64 FoldGlobalAddrGP:$addr,
3696 (i64 DoubleRegs:$src1)),
3697 (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
3700 // Map from store(globaladdress + x) -> memb(#foo + x)
3701 let AddedComplexity = 100 in
3702 def : Pat<(truncstorei8 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
3703 (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3706 def : Pat<(atomic_store_8 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
3707 (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3710 // Map from store(globaladdress + x) -> memh(#foo + x)
3711 let AddedComplexity = 100 in
3712 def : Pat<(truncstorei16 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
3713 (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3716 def : Pat<(atomic_store_16 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
3717 (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3720 // Map from store(globaladdress + x) -> memw(#foo + x)
3721 let AddedComplexity = 100 in
3722 def : Pat<(store (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
3723 (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3726 def : Pat<(atomic_store_32 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
3727 (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
3730 // Map from load(globaladdress + x) -> memd(#foo + x)
3731 let AddedComplexity = 100 in
3732 def : Pat<(i64 (load FoldGlobalAddrGP:$addr)),
3733 (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>,
3736 def : Pat<(atomic_load_64 FoldGlobalAddrGP:$addr),
3737 (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>,
3740 // Map from load(globaladdress + x) -> memb(#foo + x)
3741 let AddedComplexity = 100 in
3742 def : Pat<(i32 (extloadi8 FoldGlobalAddrGP:$addr)),
3743 (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>,
3746 // Map from load(globaladdress + x) -> memb(#foo + x)
3747 let AddedComplexity = 100 in
3748 def : Pat<(i32 (sextloadi8 FoldGlobalAddrGP:$addr)),
3749 (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>,
3752 //let AddedComplexity = 100 in
3753 let AddedComplexity = 100 in
3754 def : Pat<(i32 (extloadi16 FoldGlobalAddrGP:$addr)),
3755 (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>,
3758 // Map from load(globaladdress + x) -> memh(#foo + x)
3759 let AddedComplexity = 100 in
3760 def : Pat<(i32 (sextloadi16 FoldGlobalAddrGP:$addr)),
3761 (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>,
3764 // Map from load(globaladdress + x) -> memuh(#foo + x)
3765 let AddedComplexity = 100 in
3766 def : Pat<(i32 (zextloadi16 FoldGlobalAddrGP:$addr)),
3767 (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>,
3770 def : Pat<(atomic_load_16 FoldGlobalAddrGP:$addr),
3771 (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>,
3774 // Map from load(globaladdress + x) -> memub(#foo + x)
3775 let AddedComplexity = 100 in
3776 def : Pat<(i32 (zextloadi8 FoldGlobalAddrGP:$addr)),
3777 (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>,
3780 def : Pat<(atomic_load_8 FoldGlobalAddrGP:$addr),
3781 (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>,
3784 // Map from load(globaladdress + x) -> memw(#foo + x)
3785 let AddedComplexity = 100 in
3786 def : Pat<(i32 (load FoldGlobalAddrGP:$addr)),
3787 (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>,
3790 def : Pat<(atomic_load_32 FoldGlobalAddrGP:$addr),
3791 (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>,