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 def BITPOS32 : SDNodeXForm<imm, [{
35 // Return the bit position we will set [0-31].
37 int32_t imm = N->getSExtValue();
38 return XformMskToBitPosU5Imm(imm);
41 // Fold (add (CONST32 tglobaladdr:$addr) <offset>) into a global address.
42 def FoldGlobalAddr : ComplexPattern<i32, 1, "foldGlobalAddress", [], []>;
44 // Fold (add (CONST32_GP tglobaladdr:$addr) <offset>) into a global address.
45 def FoldGlobalAddrGP : ComplexPattern<i32, 1, "foldGlobalAddressGP", [], []>;
47 def NumUsesBelowThresCONST32 : PatFrag<(ops node:$addr),
48 (HexagonCONST32 node:$addr), [{
49 return hasNumUsesBelowThresGA(N->getOperand(0).getNode());
52 // Hexagon V4 Architecture spec defines 8 instruction classes:
53 // LD ST ALU32 XTYPE J JR MEMOP NV CR SYSTEM(system is not implemented in the
57 // ========================================
58 // Loads (8/16/32/64 bit)
62 // ========================================
63 // Stores (8/16/32/64 bit)
66 // ALU32 Instructions:
67 // ========================================
68 // Arithmetic / Logical (32 bit)
71 // XTYPE Instructions (32/64 bit):
72 // ========================================
73 // Arithmetic, Logical, Bit Manipulation
74 // Multiply (Integer, Fractional, Complex)
75 // Permute / Vector Permute Operations
76 // Predicate Operations
77 // Shift / Shift with Add/Sub/Logical
79 // Vector Halfword (ALU, Shift, Multiply)
80 // Vector Word (ALU, Shift)
83 // ========================================
84 // Jump/Call PC-relative
87 // ========================================
90 // MEMOP Instructions:
91 // ========================================
92 // Operation on memory (8/16/32 bit)
95 // ========================================
100 // ========================================
101 // Control-Register Transfers
102 // Hardware Loop Setup
103 // Predicate Logicals & Reductions
105 // SYSTEM Instructions (not implemented in the compiler):
106 // ========================================
112 //===----------------------------------------------------------------------===//
114 //===----------------------------------------------------------------------===//
116 class T_ALU32_3op_not<string mnemonic, bits<3> MajOp, bits<3> MinOp,
118 : T_ALU32_3op<mnemonic, MajOp, MinOp, OpsRev, 0> {
119 let AsmString = "$Rd = "#mnemonic#"($Rs, ~$Rt)";
122 let BaseOpcode = "andn_rr", CextOpcode = "andn", isCodeGenOnly = 0 in
123 def A4_andn : T_ALU32_3op_not<"and", 0b001, 0b100, 1>;
124 let BaseOpcode = "orn_rr", CextOpcode = "orn", isCodeGenOnly = 0 in
125 def A4_orn : T_ALU32_3op_not<"or", 0b001, 0b101, 1>;
127 let CextOpcode = "rcmp.eq", isCodeGenOnly = 0 in
128 def A4_rcmpeq : T_ALU32_3op<"cmp.eq", 0b011, 0b010, 0, 1>;
129 let CextOpcode = "!rcmp.eq", isCodeGenOnly = 0 in
130 def A4_rcmpneq : T_ALU32_3op<"!cmp.eq", 0b011, 0b011, 0, 1>;
132 let isCodeGenOnly = 0 in {
133 def C4_cmpneq : T_ALU32_3op_cmp<"!cmp.eq", 0b00, 1, 1>;
134 def C4_cmplte : T_ALU32_3op_cmp<"!cmp.gt", 0b10, 1, 0>;
135 def C4_cmplteu : T_ALU32_3op_cmp<"!cmp.gtu", 0b11, 1, 0>;
138 // Pats for instruction selection.
140 // A class to embed the usual comparison patfrags within a zext to i32.
141 // The seteq/setne frags use "lhs" and "rhs" as operands, so use the same
142 // names, or else the frag's "body" won't match the operands.
143 class CmpInReg<PatFrag Op>
144 : PatFrag<(ops node:$lhs, node:$rhs),(i32 (zext (i1 Op.Fragment)))>;
146 def: T_cmp32_rr_pat<A4_rcmpeq, CmpInReg<seteq>, i32>;
147 def: T_cmp32_rr_pat<A4_rcmpneq, CmpInReg<setne>, i32>;
149 def: T_cmp32_rr_pat<C4_cmpneq, setne, i1>;
151 class T_CMP_rrbh<string mnemonic, bits<3> MinOp, bit IsComm>
152 : SInst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, IntRegs:$Rt),
153 "$Pd = "#mnemonic#"($Rs, $Rt)", [], "", S_3op_tc_2early_SLOT23>,
155 let validSubTargets = HasV4SubT;
156 let InputType = "reg";
157 let CextOpcode = mnemonic;
159 let isCommutable = IsComm;
160 let hasSideEffects = 0;
167 let Inst{27-21} = 0b0111110;
168 let Inst{20-16} = Rs;
170 let Inst{7-5} = MinOp;
174 let isCodeGenOnly = 0 in {
175 def A4_cmpbeq : T_CMP_rrbh<"cmpb.eq", 0b110, 1>;
176 def A4_cmpbgt : T_CMP_rrbh<"cmpb.gt", 0b010, 0>;
177 def A4_cmpbgtu : T_CMP_rrbh<"cmpb.gtu", 0b111, 0>;
178 def A4_cmpheq : T_CMP_rrbh<"cmph.eq", 0b011, 1>;
179 def A4_cmphgt : T_CMP_rrbh<"cmph.gt", 0b100, 0>;
180 def A4_cmphgtu : T_CMP_rrbh<"cmph.gtu", 0b101, 0>;
183 class T_CMP_ribh<string mnemonic, bits<2> MajOp, bit IsHalf, bit IsComm,
184 Operand ImmType, bit IsImmExt, bit IsImmSigned, int ImmBits>
185 : ALU64Inst<(outs PredRegs:$Pd), (ins IntRegs:$Rs, ImmType:$Imm),
186 "$Pd = "#mnemonic#"($Rs, #$Imm)", [], "", ALU64_tc_2early_SLOT23>,
188 let validSubTargets = HasV4SubT;
189 let InputType = "imm";
190 let CextOpcode = mnemonic;
192 let isCommutable = IsComm;
193 let hasSideEffects = 0;
194 let isExtendable = IsImmExt;
195 let opExtendable = !if (IsImmExt, 2, 0);
196 let isExtentSigned = IsImmSigned;
197 let opExtentBits = ImmBits;
204 let Inst{27-24} = 0b1101;
205 let Inst{22-21} = MajOp;
206 let Inst{20-16} = Rs;
207 let Inst{12-5} = Imm;
209 let Inst{3} = IsHalf;
213 let isCodeGenOnly = 0 in {
214 def A4_cmpbeqi : T_CMP_ribh<"cmpb.eq", 0b00, 0, 1, u8Imm, 0, 0, 8>;
215 def A4_cmpbgti : T_CMP_ribh<"cmpb.gt", 0b01, 0, 0, s8Imm, 0, 1, 8>;
216 def A4_cmpbgtui : T_CMP_ribh<"cmpb.gtu", 0b10, 0, 0, u7Ext, 1, 0, 7>;
217 def A4_cmpheqi : T_CMP_ribh<"cmph.eq", 0b00, 1, 1, s8Ext, 1, 1, 8>;
218 def A4_cmphgti : T_CMP_ribh<"cmph.gt", 0b01, 1, 0, s8Ext, 1, 1, 8>;
219 def A4_cmphgtui : T_CMP_ribh<"cmph.gtu", 0b10, 1, 0, u7Ext, 1, 0, 7>;
221 class T_RCMP_EQ_ri<string mnemonic, bit IsNeg>
222 : ALU32_ri<(outs IntRegs:$Rd), (ins IntRegs:$Rs, s8Ext:$s8),
223 "$Rd = "#mnemonic#"($Rs, #$s8)", [], "", ALU32_2op_tc_1_SLOT0123>,
225 let validSubTargets = HasV4SubT;
226 let InputType = "imm";
227 let CextOpcode = !if (IsNeg, "!rcmp.eq", "rcmp.eq");
228 let isExtendable = 1;
229 let opExtendable = 2;
230 let isExtentSigned = 1;
231 let opExtentBits = 8;
239 let Inst{27-24} = 0b0011;
241 let Inst{21} = IsNeg;
242 let Inst{20-16} = Rs;
248 let isCodeGenOnly = 0 in {
249 def A4_rcmpeqi : T_RCMP_EQ_ri<"cmp.eq", 0>;
250 def A4_rcmpneqi : T_RCMP_EQ_ri<"!cmp.eq", 1>;
253 def: Pat<(i32 (zext (i1 (seteq (i32 IntRegs:$Rs), s8ExtPred:$s8)))),
254 (A4_rcmpeqi IntRegs:$Rs, s8ExtPred:$s8)>;
255 def: Pat<(i32 (zext (i1 (setne (i32 IntRegs:$Rs), s8ExtPred:$s8)))),
256 (A4_rcmpneqi IntRegs:$Rs, s8ExtPred:$s8)>;
258 // Preserve the S2_tstbit_r generation
259 def: Pat<(i32 (zext (i1 (setne (i32 (and (i32 (shl 1, (i32 IntRegs:$src2))),
260 (i32 IntRegs:$src1))), 0)))),
261 (C2_muxii (S2_tstbit_r IntRegs:$src1, IntRegs:$src2), 1, 0)>;
264 //===----------------------------------------------------------------------===//
266 //===----------------------------------------------------------------------===//
269 //===----------------------------------------------------------------------===//
271 //===----------------------------------------------------------------------===//
273 // Combine a word and an immediate into a register pair.
274 let hasSideEffects = 0, isExtentSigned = 1, isExtendable = 1,
276 class T_Combine1 <bits<2> MajOp, dag ins, string AsmStr>
277 : ALU32Inst <(outs DoubleRegs:$Rdd), ins, AsmStr> {
283 let Inst{27-24} = 0b0011;
284 let Inst{22-21} = MajOp;
285 let Inst{20-16} = Rs;
291 let opExtendable = 2, isCodeGenOnly = 0 in
292 def A4_combineri : T_Combine1<0b00, (ins IntRegs:$Rs, s8Ext:$s8),
293 "$Rdd = combine($Rs, #$s8)">;
295 let opExtendable = 1, isCodeGenOnly = 0 in
296 def A4_combineir : T_Combine1<0b01, (ins s8Ext:$s8, IntRegs:$Rs),
297 "$Rdd = combine(#$s8, $Rs)">;
299 def HexagonWrapperCombineRI_V4 :
300 SDNode<"HexagonISD::WrapperCombineRI_V4", SDTHexagonI64I32I32>;
301 def HexagonWrapperCombineIR_V4 :
302 SDNode<"HexagonISD::WrapperCombineIR_V4", SDTHexagonI64I32I32>;
304 def : Pat <(HexagonWrapperCombineRI_V4 IntRegs:$r, s8ExtPred:$i),
305 (A4_combineri IntRegs:$r, s8ExtPred:$i)>,
308 def : Pat <(HexagonWrapperCombineIR_V4 s8ExtPred:$i, IntRegs:$r),
309 (A4_combineir s8ExtPred:$i, IntRegs:$r)>,
312 // A4_combineii: Set two small immediates.
313 let hasSideEffects = 0, isExtendable = 1, opExtentBits = 6, opExtendable = 2 in
314 def A4_combineii: ALU32Inst<(outs DoubleRegs:$Rdd), (ins s8Imm:$s8, u6Ext:$U6),
315 "$Rdd = combine(#$s8, #$U6)"> {
321 let Inst{27-23} = 0b11001;
322 let Inst{20-16} = U6{5-1};
323 let Inst{13} = U6{0};
328 //===----------------------------------------------------------------------===//
330 //===----------------------------------------------------------------------===//
332 //===----------------------------------------------------------------------===//
334 //===----------------------------------------------------------------------===//
335 //===----------------------------------------------------------------------===//
336 // Template class for load instructions with Absolute set addressing mode.
337 //===----------------------------------------------------------------------===//
338 let isExtended = 1, opExtendable = 2, hasSideEffects = 0,
339 validSubTargets = HasV4SubT, addrMode = AbsoluteSet in
340 class T_LD_abs_set<string mnemonic, RegisterClass RC>:
341 LDInst2<(outs RC:$dst1, IntRegs:$dst2),
342 (ins u0AlwaysExt:$addr),
343 "$dst1 = "#mnemonic#"($dst2=##$addr)",
347 def LDrid_abs_set_V4 : T_LD_abs_set <"memd", DoubleRegs>;
348 def LDrib_abs_set_V4 : T_LD_abs_set <"memb", IntRegs>;
349 def LDriub_abs_set_V4 : T_LD_abs_set <"memub", IntRegs>;
350 def LDrih_abs_set_V4 : T_LD_abs_set <"memh", IntRegs>;
351 def LDriw_abs_set_V4 : T_LD_abs_set <"memw", IntRegs>;
352 def LDriuh_abs_set_V4 : T_LD_abs_set <"memuh", IntRegs>;
354 //===----------------------------------------------------------------------===//
355 // Template classes for the non-predicated load instructions with
356 // base + register offset addressing mode
357 //===----------------------------------------------------------------------===//
358 class T_load_rr <string mnemonic, RegisterClass RC, bits<3> MajOp>:
359 LDInst<(outs RC:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$u2),
360 "$dst = "#mnemonic#"($src1 + $src2<<#$u2)",
361 [], "", V4LDST_tc_ld_SLOT01>, ImmRegShl, AddrModeRel {
369 let Inst{27-24} = 0b1010;
370 let Inst{23-21} = MajOp;
371 let Inst{20-16} = src1;
372 let Inst{12-8} = src2;
373 let Inst{13} = u2{1};
378 //===----------------------------------------------------------------------===//
379 // Template classes for the predicated load instructions with
380 // base + register offset addressing mode
381 //===----------------------------------------------------------------------===//
382 let isPredicated = 1 in
383 class T_pload_rr <string mnemonic, RegisterClass RC, bits<3> MajOp,
384 bit isNot, bit isPredNew>:
385 LDInst <(outs RC:$dst),
386 (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$u2),
387 !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
388 ") ")#"$dst = "#mnemonic#"($src2+$src3<<#$u2)",
389 [], "", V4LDST_tc_ld_SLOT01>, AddrModeRel {
396 let isPredicatedFalse = isNot;
397 let isPredicatedNew = isPredNew;
401 let Inst{27-26} = 0b00;
402 let Inst{25} = isPredNew;
403 let Inst{24} = isNot;
404 let Inst{23-21} = MajOp;
405 let Inst{20-16} = src2;
406 let Inst{12-8} = src3;
407 let Inst{13} = u2{1};
409 let Inst{6-5} = src1;
413 //===----------------------------------------------------------------------===//
414 // multiclass for load instructions with base + register offset
416 //===----------------------------------------------------------------------===//
417 let hasSideEffects = 0, addrMode = BaseRegOffset in
418 multiclass ld_idxd_shl <string mnemonic, string CextOp, RegisterClass RC,
420 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl,
421 InputType = "reg" in {
422 let isPredicable = 1 in
423 def L4_#NAME#_rr : T_load_rr <mnemonic, RC, MajOp>;
426 def L4_p#NAME#t_rr : T_pload_rr <mnemonic, RC, MajOp, 0, 0>;
427 def L4_p#NAME#f_rr : T_pload_rr <mnemonic, RC, MajOp, 1, 0>;
430 def L4_p#NAME#tnew_rr : T_pload_rr <mnemonic, RC, MajOp, 0, 1>;
431 def L4_p#NAME#fnew_rr : T_pload_rr <mnemonic, RC, MajOp, 1, 1>;
435 let hasNewValue = 1, accessSize = ByteAccess, isCodeGenOnly = 0 in {
436 defm loadrb : ld_idxd_shl<"memb", "LDrib", IntRegs, 0b000>;
437 defm loadrub : ld_idxd_shl<"memub", "LDriub", IntRegs, 0b001>;
440 let hasNewValue = 1, accessSize = HalfWordAccess, isCodeGenOnly = 0 in {
441 defm loadrh : ld_idxd_shl<"memh", "LDrih", IntRegs, 0b010>;
442 defm loadruh : ld_idxd_shl<"memuh", "LDriuh", IntRegs, 0b011>;
445 let hasNewValue = 1, accessSize = WordAccess, isCodeGenOnly = 0 in
446 defm loadri : ld_idxd_shl<"memw", "LDriw", IntRegs, 0b100>;
448 let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
449 defm loadrd : ld_idxd_shl<"memd", "LDrid", DoubleRegs, 0b110>;
451 // 'def pats' for load instructions with base + register offset and non-zero
452 // immediate value. Immediate value is used to left-shift the second
454 let AddedComplexity = 40 in {
455 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1,
456 (shl IntRegs:$src2, u2ImmPred:$offset)))),
457 (L4_loadrb_rr IntRegs:$src1,
458 IntRegs:$src2, u2ImmPred:$offset)>,
461 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1,
462 (shl IntRegs:$src2, u2ImmPred:$offset)))),
463 (L4_loadrub_rr IntRegs:$src1,
464 IntRegs:$src2, u2ImmPred:$offset)>,
467 def : Pat <(i32 (extloadi8 (add IntRegs:$src1,
468 (shl IntRegs:$src2, u2ImmPred:$offset)))),
469 (L4_loadrub_rr IntRegs:$src1,
470 IntRegs:$src2, u2ImmPred:$offset)>,
473 def : Pat <(i32 (sextloadi16 (add IntRegs:$src1,
474 (shl IntRegs:$src2, u2ImmPred:$offset)))),
475 (L4_loadrh_rr IntRegs:$src1,
476 IntRegs:$src2, u2ImmPred:$offset)>,
479 def : Pat <(i32 (zextloadi16 (add IntRegs:$src1,
480 (shl IntRegs:$src2, u2ImmPred:$offset)))),
481 (L4_loadruh_rr IntRegs:$src1,
482 IntRegs:$src2, u2ImmPred:$offset)>,
485 def : Pat <(i32 (extloadi16 (add IntRegs:$src1,
486 (shl IntRegs:$src2, u2ImmPred:$offset)))),
487 (L4_loadruh_rr IntRegs:$src1,
488 IntRegs:$src2, u2ImmPred:$offset)>,
491 def : Pat <(i32 (load (add IntRegs:$src1,
492 (shl IntRegs:$src2, u2ImmPred:$offset)))),
493 (L4_loadri_rr IntRegs:$src1,
494 IntRegs:$src2, u2ImmPred:$offset)>,
497 def : Pat <(i64 (load (add IntRegs:$src1,
498 (shl IntRegs:$src2, u2ImmPred:$offset)))),
499 (L4_loadrd_rr IntRegs:$src1,
500 IntRegs:$src2, u2ImmPred:$offset)>,
505 // 'def pats' for load instruction base + register offset and
506 // zero immediate value.
507 let AddedComplexity = 10 in {
508 def : Pat <(i64 (load (add IntRegs:$src1, IntRegs:$src2))),
509 (L4_loadrd_rr IntRegs:$src1, IntRegs:$src2, 0)>,
512 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1, IntRegs:$src2))),
513 (L4_loadrb_rr IntRegs:$src1, IntRegs:$src2, 0)>,
516 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1, IntRegs:$src2))),
517 (L4_loadrub_rr IntRegs:$src1, IntRegs:$src2, 0)>,
520 def : Pat <(i32 (extloadi8 (add IntRegs:$src1, IntRegs:$src2))),
521 (L4_loadrub_rr IntRegs:$src1, IntRegs:$src2, 0)>,
524 def : Pat <(i32 (sextloadi16 (add IntRegs:$src1, IntRegs:$src2))),
525 (L4_loadrh_rr IntRegs:$src1, IntRegs:$src2, 0)>,
528 def : Pat <(i32 (zextloadi16 (add IntRegs:$src1, IntRegs:$src2))),
529 (L4_loadruh_rr IntRegs:$src1, IntRegs:$src2, 0)>,
532 def : Pat <(i32 (extloadi16 (add IntRegs:$src1, IntRegs:$src2))),
533 (L4_loadruh_rr IntRegs:$src1, IntRegs:$src2, 0)>,
536 def : Pat <(i32 (load (add IntRegs:$src1, IntRegs:$src2))),
537 (L4_loadri_rr IntRegs:$src1, IntRegs:$src2, 0)>,
542 def : Pat <(i64 (zext (i1 PredRegs:$src1))),
543 (i64 (A4_combineir 0, (C2_muxii (i1 PredRegs:$src1), 1, 0)))>,
547 def : Pat <(i64 (zext (i32 IntRegs:$src1))),
548 (i64 (A4_combineir 0, (i32 IntRegs:$src1)))>,
551 def: Pat <(i64 (zextloadi8 ADDRriS11_0:$src1)),
552 (i64 (A4_combineir 0, (L2_loadrub_io AddrFI:$src1, 0)))>,
555 let AddedComplexity = 20 in
556 def: Pat <(i64 (zextloadi8 (add (i32 IntRegs:$src1),
557 s11_0ExtPred:$offset))),
558 (i64 (A4_combineir 0, (L2_loadrub_io IntRegs:$src1,
559 s11_0ExtPred:$offset)))>,
563 def: Pat <(i64 (zextloadi1 ADDRriS11_0:$src1)),
564 (i64 (A4_combineir 0, (L2_loadrub_io AddrFI:$src1, 0)))>,
567 let AddedComplexity = 20 in
568 def: Pat <(i64 (zextloadi1 (add (i32 IntRegs:$src1),
569 s11_0ExtPred:$offset))),
570 (i64 (A4_combineir 0, (L2_loadrub_io IntRegs:$src1,
571 s11_0ExtPred:$offset)))>,
575 def: Pat <(i64 (zextloadi16 ADDRriS11_1:$src1)),
576 (i64 (A4_combineir 0, (L2_loadruh_io AddrFI:$src1, 0)))>,
579 let AddedComplexity = 20 in
580 def: Pat <(i64 (zextloadi16 (add (i32 IntRegs:$src1),
581 s11_1ExtPred:$offset))),
582 (i64 (A4_combineir 0, (L2_loadruh_io IntRegs:$src1,
583 s11_1ExtPred:$offset)))>,
587 def: Pat <(i64 (extloadi16 ADDRriS11_2:$src1)),
588 (i64 (A4_combineir 0, (L2_loadrh_io AddrFI:$src1, 0)))>,
591 let AddedComplexity = 20 in
592 def: Pat <(i64 (extloadi16 (add (i32 IntRegs:$src1),
593 s11_1ExtPred:$offset))),
594 (i64 (A4_combineir 0, (L2_loadrh_io IntRegs:$src1,
595 s11_1ExtPred:$offset)))>,
599 def: Pat <(i64 (zextloadi32 ADDRriS11_2:$src1)),
600 (i64 (A4_combineir 0, (L2_loadri_io AddrFI:$src1, 0)))>,
603 let AddedComplexity = 100 in
604 def: Pat <(i64 (zextloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
605 (i64 (A4_combineir 0, (L2_loadri_io IntRegs:$src1,
606 s11_2ExtPred:$offset)))>,
610 def: Pat <(i64 (extloadi32 ADDRriS11_2:$src1)),
611 (i64 (A4_combineir 0, (L2_loadri_io AddrFI:$src1, 0)))>,
614 let AddedComplexity = 100 in
615 def: Pat <(i64 (extloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
616 (i64 (A4_combineir 0, (L2_loadri_io IntRegs:$src1,
617 s11_2ExtPred:$offset)))>,
622 //===----------------------------------------------------------------------===//
624 //===----------------------------------------------------------------------===//
626 //===----------------------------------------------------------------------===//
628 //===----------------------------------------------------------------------===//
630 //===----------------------------------------------------------------------===//
631 // Template class for store instructions with Absolute set addressing mode.
632 //===----------------------------------------------------------------------===//
633 let isExtended = 1, opExtendable = 2, validSubTargets = HasV4SubT,
634 addrMode = AbsoluteSet in
635 class T_ST_abs_set<string mnemonic, RegisterClass RC>:
636 STInst2<(outs IntRegs:$dst1),
637 (ins RC:$src1, u0AlwaysExt:$src2),
638 mnemonic#"($dst1=##$src2) = $src1",
642 def STrid_abs_set_V4 : T_ST_abs_set <"memd", DoubleRegs>;
643 def STrib_abs_set_V4 : T_ST_abs_set <"memb", IntRegs>;
644 def STrih_abs_set_V4 : T_ST_abs_set <"memh", IntRegs>;
645 def STriw_abs_set_V4 : T_ST_abs_set <"memw", IntRegs>;
647 //===----------------------------------------------------------------------===//
648 // Template classes for the non-predicated store instructions with
649 // base + register offset addressing mode
650 //===----------------------------------------------------------------------===//
651 let isPredicable = 1 in
652 class T_store_rr <string mnemonic, RegisterClass RC, bits<3> MajOp, bit isH>
653 : STInst < (outs ), (ins IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, RC:$Rt),
654 mnemonic#"($Rs + $Ru<<#$u2) = $Rt"#!if(isH, ".h",""),
655 [],"",V4LDST_tc_st_SLOT01>, ImmRegShl, AddrModeRel {
664 let Inst{27-24} = 0b1011;
665 let Inst{23-21} = MajOp;
666 let Inst{20-16} = Rs;
668 let Inst{13} = u2{1};
673 //===----------------------------------------------------------------------===//
674 // Template classes for the predicated store instructions with
675 // base + register offset addressing mode
676 //===----------------------------------------------------------------------===//
677 let isPredicated = 1 in
678 class T_pstore_rr <string mnemonic, RegisterClass RC, bits<3> MajOp,
679 bit isNot, bit isPredNew, bit isH>
681 (ins PredRegs:$Pv, IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, RC:$Rt),
683 !if(isNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ",
684 ") ")#mnemonic#"($Rs+$Ru<<#$u2) = $Rt"#!if(isH, ".h",""),
685 [], "", V4LDST_tc_st_SLOT01> , AddrModeRel{
692 let isPredicatedFalse = isNot;
693 let isPredicatedNew = isPredNew;
697 let Inst{27-26} = 0b01;
698 let Inst{25} = isPredNew;
699 let Inst{24} = isNot;
700 let Inst{23-21} = MajOp;
701 let Inst{20-16} = Rs;
703 let Inst{13} = u2{1};
709 //===----------------------------------------------------------------------===//
710 // Template classes for the new-value store instructions with
711 // base + register offset addressing mode
712 //===----------------------------------------------------------------------===//
713 let isPredicable = 1, isNewValue = 1, opNewValue = 3 in
714 class T_store_new_rr <string mnemonic, bits<2> MajOp> :
715 NVInst < (outs ), (ins IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, IntRegs:$Nt),
716 mnemonic#"($Rs + $Ru<<#$u2) = $Nt.new",
717 [],"",V4LDST_tc_st_SLOT0>, ImmRegShl, AddrModeRel {
726 let Inst{27-21} = 0b1011101;
727 let Inst{20-16} = Rs;
729 let Inst{13} = u2{1};
731 let Inst{4-3} = MajOp;
735 //===----------------------------------------------------------------------===//
736 // Template classes for the predicated new-value store instructions with
737 // base + register offset addressing mode
738 //===----------------------------------------------------------------------===//
739 let isPredicated = 1, isNewValue = 1, opNewValue = 4 in
740 class T_pstore_new_rr <string mnemonic, bits<2> MajOp, bit isNot, bit isPredNew>
742 (ins PredRegs:$Pv, IntRegs:$Rs, IntRegs:$Ru, u2Imm:$u2, IntRegs:$Nt),
743 !if(isNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ",
744 ") ")#mnemonic#"($Rs+$Ru<<#$u2) = $Nt.new",
745 [], "", V4LDST_tc_st_SLOT0>, AddrModeRel {
752 let isPredicatedFalse = isNot;
753 let isPredicatedNew = isPredNew;
756 let Inst{27-26} = 0b01;
757 let Inst{25} = isPredNew;
758 let Inst{24} = isNot;
759 let Inst{23-21} = 0b101;
760 let Inst{20-16} = Rs;
762 let Inst{13} = u2{1};
765 let Inst{4-3} = MajOp;
769 //===----------------------------------------------------------------------===//
770 // multiclass for store instructions with base + register offset addressing
772 //===----------------------------------------------------------------------===//
773 let isNVStorable = 1 in
774 multiclass ST_Idxd_shl<string mnemonic, string CextOp, RegisterClass RC,
775 bits<3> MajOp, bit isH = 0> {
776 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
777 def S4_#NAME#_rr : T_store_rr <mnemonic, RC, MajOp, isH>;
780 def S4_p#NAME#t_rr : T_pstore_rr <mnemonic, RC, MajOp, 0, 0, isH>;
781 def S4_p#NAME#f_rr : T_pstore_rr <mnemonic, RC, MajOp, 1, 0, isH>;
784 def S4_p#NAME#tnew_rr : T_pstore_rr <mnemonic, RC, MajOp, 0, 1, isH>;
785 def S4_p#NAME#fnew_rr : T_pstore_rr <mnemonic, RC, MajOp, 1, 1, isH>;
789 //===----------------------------------------------------------------------===//
790 // multiclass for new-value store instructions with base + register offset
792 //===----------------------------------------------------------------------===//
793 let mayStore = 1, isNVStore = 1 in
794 multiclass ST_Idxd_shl_nv <string mnemonic, string CextOp, RegisterClass RC,
796 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
797 def S4_#NAME#new_rr : T_store_new_rr<mnemonic, MajOp>;
800 def S4_p#NAME#newt_rr : T_pstore_new_rr <mnemonic, MajOp, 0, 0>;
801 def S4_p#NAME#newf_rr : T_pstore_new_rr <mnemonic, MajOp, 1, 0>;
804 def S4_p#NAME#newtnew_rr : T_pstore_new_rr <mnemonic, MajOp, 0, 1>;
805 def S4_p#NAME#newfnew_rr : T_pstore_new_rr <mnemonic, MajOp, 1, 1>;
809 let addrMode = BaseRegOffset, InputType = "reg", hasSideEffects = 0,
810 isCodeGenOnly = 0 in {
811 let accessSize = ByteAccess in
812 defm storerb: ST_Idxd_shl<"memb", "STrib", IntRegs, 0b000>,
813 ST_Idxd_shl_nv<"memb", "STrib", IntRegs, 0b00>;
815 let accessSize = HalfWordAccess in
816 defm storerh: ST_Idxd_shl<"memh", "STrih", IntRegs, 0b010>,
817 ST_Idxd_shl_nv<"memh", "STrih", IntRegs, 0b01>;
819 let accessSize = WordAccess in
820 defm storeri: ST_Idxd_shl<"memw", "STriw", IntRegs, 0b100>,
821 ST_Idxd_shl_nv<"memw", "STriw", IntRegs, 0b10>;
823 let isNVStorable = 0, accessSize = DoubleWordAccess in
824 defm storerd: ST_Idxd_shl<"memd", "STrid", DoubleRegs, 0b110>;
826 let isNVStorable = 0, accessSize = HalfWordAccess in
827 defm storerf: ST_Idxd_shl<"memh", "STrif", IntRegs, 0b011, 1>;
830 let Predicates = [HasV4T], AddedComplexity = 10 in {
831 def : Pat<(truncstorei8 (i32 IntRegs:$src4),
832 (add IntRegs:$src1, (shl IntRegs:$src2,
834 (S4_storerb_rr IntRegs:$src1, IntRegs:$src2,
835 u2ImmPred:$src3, IntRegs:$src4)>;
837 def : Pat<(truncstorei16 (i32 IntRegs:$src4),
838 (add IntRegs:$src1, (shl IntRegs:$src2,
840 (S4_storerh_rr IntRegs:$src1, IntRegs:$src2,
841 u2ImmPred:$src3, IntRegs:$src4)>;
843 def : Pat<(store (i32 IntRegs:$src4),
844 (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))),
845 (S4_storeri_rr IntRegs:$src1, IntRegs:$src2,
846 u2ImmPred:$src3, IntRegs:$src4)>;
848 def : Pat<(store (i64 DoubleRegs:$src4),
849 (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))),
850 (S4_storerd_rr IntRegs:$src1, IntRegs:$src2,
851 u2ImmPred:$src3, DoubleRegs:$src4)>;
854 let isExtended = 1, opExtendable = 2 in
855 class T_ST_LongOff <string mnemonic, PatFrag stOp, RegisterClass RC, ValueType VT> :
857 (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, RC:$src4),
858 mnemonic#"($src1<<#$src2+##$src3) = $src4",
859 [(stOp (VT RC:$src4),
860 (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2),
861 u0AlwaysExtPred:$src3))]>,
864 let isExtended = 1, opExtendable = 2, mayStore = 1, isNVStore = 1 in
865 class T_ST_LongOff_nv <string mnemonic> :
867 (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, IntRegs:$src4),
868 mnemonic#"($src1<<#$src2+##$src3) = $src4.new",
872 multiclass ST_LongOff <string mnemonic, string BaseOp, PatFrag stOp> {
873 let BaseOpcode = BaseOp#"_shl" in {
874 let isNVStorable = 1 in
875 def NAME#_V4 : T_ST_LongOff<mnemonic, stOp, IntRegs, i32>;
877 def NAME#_nv_V4 : T_ST_LongOff_nv<mnemonic>;
881 let AddedComplexity = 10, validSubTargets = HasV4SubT in {
882 def STrid_shl_V4 : T_ST_LongOff<"memd", store, DoubleRegs, i64>;
883 defm STrib_shl : ST_LongOff <"memb", "STrib", truncstorei8>, NewValueRel;
884 defm STrih_shl : ST_LongOff <"memh", "Strih", truncstorei16>, NewValueRel;
885 defm STriw_shl : ST_LongOff <"memw", "STriw", store>, NewValueRel;
888 let AddedComplexity = 40 in
889 multiclass T_ST_LOff_Pats <InstHexagon I, RegisterClass RC, ValueType VT,
891 def : Pat<(stOp (VT RC:$src4),
892 (add (shl IntRegs:$src1, u2ImmPred:$src2),
893 (NumUsesBelowThresCONST32 tglobaladdr:$src3))),
894 (I IntRegs:$src1, u2ImmPred:$src2, tglobaladdr:$src3, RC:$src4)>;
896 def : Pat<(stOp (VT RC:$src4),
898 (NumUsesBelowThresCONST32 tglobaladdr:$src3))),
899 (I IntRegs:$src1, 0, tglobaladdr:$src3, RC:$src4)>;
902 defm : T_ST_LOff_Pats<STrid_shl_V4, DoubleRegs, i64, store>;
903 defm : T_ST_LOff_Pats<STriw_shl_V4, IntRegs, i32, store>;
904 defm : T_ST_LOff_Pats<STrib_shl_V4, IntRegs, i32, truncstorei8>;
905 defm : T_ST_LOff_Pats<STrih_shl_V4, IntRegs, i32, truncstorei16>;
907 // memd(Rx++#s4:3)=Rtt
908 // memd(Rx++#s4:3:circ(Mu))=Rtt
909 // memd(Rx++I:circ(Mu))=Rtt
911 // memd(Rx++Mu:brev)=Rtt
912 // memd(gp+#u16:3)=Rtt
914 // Store doubleword conditionally.
915 // if ([!]Pv[.new]) memd(#u6)=Rtt
916 // TODO: needs to be implemented.
918 //===----------------------------------------------------------------------===//
920 //===----------------------------------------------------------------------===//
921 let isPredicable = 1, isExtendable = 1, isExtentSigned = 1, opExtentBits = 8,
923 class T_StoreImm <string mnemonic, Operand OffsetOp, bits<2> MajOp >
924 : STInst <(outs ), (ins IntRegs:$Rs, OffsetOp:$offset, s8Ext:$S8),
925 mnemonic#"($Rs+#$offset)=#$S8",
926 [], "", V4LDST_tc_st_SLOT01>,
927 ImmRegRel, PredNewRel {
933 string OffsetOpStr = !cast<string>(OffsetOp);
934 let offsetBits = !if (!eq(OffsetOpStr, "u6_2Imm"), offset{7-2},
935 !if (!eq(OffsetOpStr, "u6_1Imm"), offset{6-1},
936 /* u6_0Imm */ offset{5-0}));
940 let Inst{27-25} = 0b110;
941 let Inst{22-21} = MajOp;
942 let Inst{20-16} = Rs;
943 let Inst{12-7} = offsetBits;
944 let Inst{13} = S8{7};
945 let Inst{6-0} = S8{6-0};
948 let isPredicated = 1, isExtendable = 1, isExtentSigned = 1, opExtentBits = 6,
950 class T_StoreImm_pred <string mnemonic, Operand OffsetOp, bits<2> MajOp,
951 bit isPredNot, bit isPredNew >
953 (ins PredRegs:$Pv, IntRegs:$Rs, OffsetOp:$offset, s6Ext:$S6),
954 !if(isPredNot, "if (!$Pv", "if ($Pv")#!if(isPredNew, ".new) ",
955 ") ")#mnemonic#"($Rs+#$offset)=#$S6",
956 [], "", V4LDST_tc_st_SLOT01>,
957 ImmRegRel, PredNewRel {
964 string OffsetOpStr = !cast<string>(OffsetOp);
965 let offsetBits = !if (!eq(OffsetOpStr, "u6_2Imm"), offset{7-2},
966 !if (!eq(OffsetOpStr, "u6_1Imm"), offset{6-1},
967 /* u6_0Imm */ offset{5-0}));
968 let isPredicatedNew = isPredNew;
969 let isPredicatedFalse = isPredNot;
973 let Inst{27-25} = 0b100;
974 let Inst{24} = isPredNew;
975 let Inst{23} = isPredNot;
976 let Inst{22-21} = MajOp;
977 let Inst{20-16} = Rs;
978 let Inst{13} = S6{5};
979 let Inst{12-7} = offsetBits;
981 let Inst{4-0} = S6{4-0};
985 //===----------------------------------------------------------------------===//
986 // multiclass for store instructions with base + immediate offset
987 // addressing mode and immediate stored value.
988 // mem[bhw](Rx++#s4:3)=#s8
989 // if ([!]Pv[.new]) mem[bhw](Rx++#s4:3)=#s6
990 //===----------------------------------------------------------------------===//
992 multiclass ST_Imm_Pred <string mnemonic, Operand OffsetOp, bits<2> MajOp,
994 def _io : T_StoreImm_pred <mnemonic, OffsetOp, MajOp, PredNot, 0>;
996 def new_io : T_StoreImm_pred <mnemonic, OffsetOp, MajOp, PredNot, 1>;
999 multiclass ST_Imm <string mnemonic, string CextOp, Operand OffsetOp,
1001 let CextOpcode = CextOp, BaseOpcode = CextOp#_imm in {
1002 def _io : T_StoreImm <mnemonic, OffsetOp, MajOp>;
1004 defm t : ST_Imm_Pred <mnemonic, OffsetOp, MajOp, 0>;
1005 defm f : ST_Imm_Pred <mnemonic, OffsetOp, MajOp, 1>;
1009 let hasSideEffects = 0, validSubTargets = HasV4SubT, addrMode = BaseImmOffset,
1010 InputType = "imm", isCodeGenOnly = 0 in {
1011 let accessSize = ByteAccess in
1012 defm S4_storeirb : ST_Imm<"memb", "STrib", u6_0Imm, 0b00>;
1014 let accessSize = HalfWordAccess in
1015 defm S4_storeirh : ST_Imm<"memh", "STrih", u6_1Imm, 0b01>;
1017 let accessSize = WordAccess in
1018 defm S4_storeiri : ST_Imm<"memw", "STriw", u6_2Imm, 0b10>;
1021 let Predicates = [HasV4T], AddedComplexity = 10 in {
1022 def: Pat<(truncstorei8 s8ExtPred:$src3, (add IntRegs:$src1, u6_0ImmPred:$src2)),
1023 (S4_storeirb_io IntRegs:$src1, u6_0ImmPred:$src2, s8ExtPred:$src3)>;
1025 def: Pat<(truncstorei16 s8ExtPred:$src3, (add IntRegs:$src1,
1026 u6_1ImmPred:$src2)),
1027 (S4_storeirh_io IntRegs:$src1, u6_1ImmPred:$src2, s8ExtPred:$src3)>;
1029 def: Pat<(store s8ExtPred:$src3, (add IntRegs:$src1, u6_2ImmPred:$src2)),
1030 (S4_storeiri_io IntRegs:$src1, u6_2ImmPred:$src2, s8ExtPred:$src3)>;
1033 let AddedComplexity = 6 in
1034 def : Pat <(truncstorei8 s8ExtPred:$src2, (i32 IntRegs:$src1)),
1035 (S4_storeirb_io IntRegs:$src1, 0, s8ExtPred:$src2)>,
1038 // memb(Rx++#s4:0:circ(Mu))=Rt
1039 // memb(Rx++I:circ(Mu))=Rt
1041 // memb(Rx++Mu:brev)=Rt
1042 // memb(gp+#u16:0)=Rt
1046 // TODO: needs to be implemented
1047 // memh(Re=#U6)=Rt.H
1048 // memh(Rs+#s11:1)=Rt.H
1049 let AddedComplexity = 6 in
1050 def : Pat <(truncstorei16 s8ExtPred:$src2, (i32 IntRegs:$src1)),
1051 (S4_storeirh_io IntRegs:$src1, 0, s8ExtPred:$src2)>,
1054 // memh(Rs+Ru<<#u2)=Rt.H
1055 // TODO: needs to be implemented.
1057 // memh(Ru<<#u2+#U6)=Rt.H
1058 // memh(Rx++#s4:1:circ(Mu))=Rt.H
1059 // memh(Rx++#s4:1:circ(Mu))=Rt
1060 // memh(Rx++I:circ(Mu))=Rt.H
1061 // memh(Rx++I:circ(Mu))=Rt
1062 // memh(Rx++Mu)=Rt.H
1064 // memh(Rx++Mu:brev)=Rt.H
1065 // memh(Rx++Mu:brev)=Rt
1066 // memh(gp+#u16:1)=Rt
1067 // if ([!]Pv[.new]) memh(#u6)=Rt.H
1068 // if ([!]Pv[.new]) memh(#u6)=Rt
1071 // if ([!]Pv[.new]) memh(Rs+#u6:1)=Rt.H
1072 // TODO: needs to be implemented.
1074 // if ([!]Pv[.new]) memh(Rx++#s4:1)=Rt.H
1075 // TODO: Needs to be implemented.
1079 // TODO: Needs to be implemented.
1082 let hasSideEffects = 0 in
1083 def STriw_pred_V4 : STInst2<(outs),
1084 (ins MEMri:$addr, PredRegs:$src1),
1085 "Error; should not emit",
1089 let AddedComplexity = 6 in
1090 def : Pat <(store s8ExtPred:$src2, (i32 IntRegs:$src1)),
1091 (S4_storeiri_io IntRegs:$src1, 0, s8ExtPred:$src2)>,
1094 // memw(Rx++#s4:2)=Rt
1095 // memw(Rx++#s4:2:circ(Mu))=Rt
1096 // memw(Rx++I:circ(Mu))=Rt
1098 // memw(Rx++Mu:brev)=Rt
1100 //===----------------------------------------------------------------------===
1102 //===----------------------------------------------------------------------===
1105 //===----------------------------------------------------------------------===//
1107 //===----------------------------------------------------------------------===//
1109 let opNewValue = 2, opExtendable = 1, isExtentSigned = 1, isPredicable = 1 in
1110 class T_store_io_nv <string mnemonic, RegisterClass RC,
1111 Operand ImmOp, bits<2>MajOp>
1112 : NVInst_V4 <(outs),
1113 (ins IntRegs:$src1, ImmOp:$src2, RC:$src3),
1114 mnemonic#"($src1+#$src2) = $src3.new",
1115 [],"",ST_tc_st_SLOT0> {
1117 bits<13> src2; // Actual address offset
1119 bits<11> offsetBits; // Represents offset encoding
1121 let opExtentBits = !if (!eq(mnemonic, "memb"), 11,
1122 !if (!eq(mnemonic, "memh"), 12,
1123 !if (!eq(mnemonic, "memw"), 13, 0)));
1125 let opExtentAlign = !if (!eq(mnemonic, "memb"), 0,
1126 !if (!eq(mnemonic, "memh"), 1,
1127 !if (!eq(mnemonic, "memw"), 2, 0)));
1129 let offsetBits = !if (!eq(mnemonic, "memb"), src2{10-0},
1130 !if (!eq(mnemonic, "memh"), src2{11-1},
1131 !if (!eq(mnemonic, "memw"), src2{12-2}, 0)));
1133 let IClass = 0b1010;
1136 let Inst{26-25} = offsetBits{10-9};
1137 let Inst{24-21} = 0b1101;
1138 let Inst{20-16} = src1;
1139 let Inst{13} = offsetBits{8};
1140 let Inst{12-11} = MajOp;
1141 let Inst{10-8} = src3;
1142 let Inst{7-0} = offsetBits{7-0};
1145 let opExtendable = 2, opNewValue = 3, isPredicated = 1 in
1146 class T_pstore_io_nv <string mnemonic, RegisterClass RC, Operand predImmOp,
1147 bits<2>MajOp, bit PredNot, bit isPredNew>
1148 : NVInst_V4 <(outs),
1149 (ins PredRegs:$src1, IntRegs:$src2, predImmOp:$src3, RC:$src4),
1150 !if(PredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
1151 ") ")#mnemonic#"($src2+#$src3) = $src4.new",
1152 [],"",V2LDST_tc_st_SLOT0> {
1157 bits<6> offsetBits; // Represents offset encoding
1159 let isPredicatedNew = isPredNew;
1160 let isPredicatedFalse = PredNot;
1161 let opExtentBits = !if (!eq(mnemonic, "memb"), 6,
1162 !if (!eq(mnemonic, "memh"), 7,
1163 !if (!eq(mnemonic, "memw"), 8, 0)));
1165 let opExtentAlign = !if (!eq(mnemonic, "memb"), 0,
1166 !if (!eq(mnemonic, "memh"), 1,
1167 !if (!eq(mnemonic, "memw"), 2, 0)));
1169 let offsetBits = !if (!eq(mnemonic, "memb"), src3{5-0},
1170 !if (!eq(mnemonic, "memh"), src3{6-1},
1171 !if (!eq(mnemonic, "memw"), src3{7-2}, 0)));
1173 let IClass = 0b0100;
1176 let Inst{26} = PredNot;
1177 let Inst{25} = isPredNew;
1178 let Inst{24-21} = 0b0101;
1179 let Inst{20-16} = src2;
1180 let Inst{13} = offsetBits{5};
1181 let Inst{12-11} = MajOp;
1182 let Inst{10-8} = src4;
1183 let Inst{7-3} = offsetBits{4-0};
1185 let Inst{1-0} = src1;
1188 // multiclass for new-value store instructions with base + immediate offset.
1190 let mayStore = 1, isNVStore = 1, isNewValue = 1, hasSideEffects = 0,
1192 multiclass ST_Idxd_nv<string mnemonic, string CextOp, RegisterClass RC,
1193 Operand ImmOp, Operand predImmOp, bits<2> MajOp> {
1195 let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
1196 def S2_#NAME#new_io : T_store_io_nv <mnemonic, RC, ImmOp, MajOp>;
1198 def S2_p#NAME#newt_io :T_pstore_io_nv <mnemonic, RC, predImmOp, MajOp, 0, 0>;
1199 def S2_p#NAME#newf_io :T_pstore_io_nv <mnemonic, RC, predImmOp, MajOp, 1, 0>;
1201 def S4_p#NAME#newtnew_io :T_pstore_io_nv <mnemonic, RC, predImmOp,
1203 def S4_p#NAME#newfnew_io :T_pstore_io_nv <mnemonic, RC, predImmOp,
1208 let addrMode = BaseImmOffset, InputType = "imm", isCodeGenOnly = 0 in {
1209 let accessSize = ByteAccess in
1210 defm storerb: ST_Idxd_nv<"memb", "STrib", IntRegs, s11_0Ext,
1211 u6_0Ext, 0b00>, AddrModeRel;
1213 let accessSize = HalfWordAccess, opExtentAlign = 1 in
1214 defm storerh: ST_Idxd_nv<"memh", "STrih", IntRegs, s11_1Ext,
1215 u6_1Ext, 0b01>, AddrModeRel;
1217 let accessSize = WordAccess, opExtentAlign = 2 in
1218 defm storeri: ST_Idxd_nv<"memw", "STriw", IntRegs, s11_2Ext,
1219 u6_2Ext, 0b10>, AddrModeRel;
1222 //===----------------------------------------------------------------------===//
1223 // Template class for non-predicated post increment .new stores
1224 // mem[bhwd](Rx++#s4:[0123])=Nt.new
1225 //===----------------------------------------------------------------------===//
1226 let isPredicable = 1, hasSideEffects = 0, validSubTargets = HasV4SubT,
1227 addrMode = PostInc, isNVStore = 1, isNewValue = 1, opNewValue = 3 in
1228 class T_StorePI_nv <string mnemonic, Operand ImmOp, bits<2> MajOp >
1229 : NVInstPI_V4 <(outs IntRegs:$_dst_),
1230 (ins IntRegs:$src1, ImmOp:$offset, IntRegs:$src2),
1231 mnemonic#"($src1++#$offset) = $src2.new",
1232 [], "$src1 = $_dst_">,
1239 string ImmOpStr = !cast<string>(ImmOp);
1240 let offsetBits = !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
1241 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
1242 /* s4_0Imm */ offset{3-0}));
1243 let IClass = 0b1010;
1245 let Inst{27-21} = 0b1011101;
1246 let Inst{20-16} = src1;
1248 let Inst{12-11} = MajOp;
1249 let Inst{10-8} = src2;
1251 let Inst{6-3} = offsetBits;
1255 //===----------------------------------------------------------------------===//
1256 // Template class for predicated post increment .new stores
1257 // if([!]Pv[.new]) mem[bhwd](Rx++#s4:[0123])=Nt.new
1258 //===----------------------------------------------------------------------===//
1259 let isPredicated = 1, hasSideEffects = 0, validSubTargets = HasV4SubT,
1260 addrMode = PostInc, isNVStore = 1, isNewValue = 1, opNewValue = 4 in
1261 class T_StorePI_nv_pred <string mnemonic, Operand ImmOp,
1262 bits<2> MajOp, bit isPredNot, bit isPredNew >
1263 : NVInstPI_V4 <(outs IntRegs:$_dst_),
1264 (ins PredRegs:$src1, IntRegs:$src2,
1265 ImmOp:$offset, IntRegs:$src3),
1266 !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
1267 ") ")#mnemonic#"($src2++#$offset) = $src3.new",
1268 [], "$src2 = $_dst_">,
1276 string ImmOpStr = !cast<string>(ImmOp);
1277 let offsetBits = !if (!eq(ImmOpStr, "s4_2Imm"), offset{5-2},
1278 !if (!eq(ImmOpStr, "s4_1Imm"), offset{4-1},
1279 /* s4_0Imm */ offset{3-0}));
1280 let isPredicatedNew = isPredNew;
1281 let isPredicatedFalse = isPredNot;
1283 let IClass = 0b1010;
1285 let Inst{27-21} = 0b1011101;
1286 let Inst{20-16} = src2;
1288 let Inst{12-11} = MajOp;
1289 let Inst{10-8} = src3;
1290 let Inst{7} = isPredNew;
1291 let Inst{6-3} = offsetBits;
1292 let Inst{2} = isPredNot;
1293 let Inst{1-0} = src1;
1296 multiclass ST_PostInc_Pred_nv<string mnemonic, Operand ImmOp,
1297 bits<2> MajOp, bit PredNot> {
1298 def _pi : T_StorePI_nv_pred <mnemonic, ImmOp, MajOp, PredNot, 0>;
1301 def new_pi : T_StorePI_nv_pred <mnemonic, ImmOp, MajOp, PredNot, 1>;
1304 multiclass ST_PostInc_nv<string mnemonic, string BaseOp, Operand ImmOp,
1306 let BaseOpcode = "POST_"#BaseOp in {
1307 def S2_#NAME#_pi : T_StorePI_nv <mnemonic, ImmOp, MajOp>;
1310 defm S2_p#NAME#t : ST_PostInc_Pred_nv <mnemonic, ImmOp, MajOp, 0>;
1311 defm S2_p#NAME#f : ST_PostInc_Pred_nv <mnemonic, ImmOp, MajOp, 1>;
1315 let accessSize = ByteAccess, isCodeGenOnly = 0 in
1316 defm storerbnew: ST_PostInc_nv <"memb", "STrib", s4_0Imm, 0b00>;
1318 let accessSize = HalfWordAccess, isCodeGenOnly = 0 in
1319 defm storerhnew: ST_PostInc_nv <"memh", "STrih", s4_1Imm, 0b01>;
1321 let accessSize = WordAccess, isCodeGenOnly = 0 in
1322 defm storerinew: ST_PostInc_nv <"memw", "STriw", s4_2Imm, 0b10>;
1324 //===----------------------------------------------------------------------===//
1325 // Template class for post increment .new stores with register offset
1326 //===----------------------------------------------------------------------===//
1327 let isNewValue = 1, mayStore = 1, isNVStore = 1, opNewValue = 3 in
1328 class T_StorePI_RegNV <string mnemonic, bits<2> MajOp, MemAccessSize AccessSz>
1329 : NVInstPI_V4 <(outs IntRegs:$_dst_),
1330 (ins IntRegs:$src1, ModRegs:$src2, IntRegs:$src3),
1331 #mnemonic#"($src1++$src2) = $src3.new",
1332 [], "$src1 = $_dst_"> {
1336 let accessSize = AccessSz;
1338 let IClass = 0b1010;
1340 let Inst{27-21} = 0b1101101;
1341 let Inst{20-16} = src1;
1342 let Inst{13} = src2;
1343 let Inst{12-11} = MajOp;
1344 let Inst{10-8} = src3;
1348 let isCodeGenOnly = 0 in {
1349 def S2_storerbnew_pr : T_StorePI_RegNV<"memb", 0b00, ByteAccess>;
1350 def S2_storerhnew_pr : T_StorePI_RegNV<"memh", 0b01, HalfWordAccess>;
1351 def S2_storerinew_pr : T_StorePI_RegNV<"memw", 0b10, WordAccess>;
1354 // memb(Rx++#s4:0:circ(Mu))=Nt.new
1355 // memb(Rx++I:circ(Mu))=Nt.new
1356 // memb(Rx++Mu)=Nt.new
1357 // memb(Rx++Mu:brev)=Nt.new
1358 // memh(Rx++#s4:1:circ(Mu))=Nt.new
1359 // memh(Rx++I:circ(Mu))=Nt.new
1360 // memh(Rx++Mu)=Nt.new
1361 // memh(Rx++Mu:brev)=Nt.new
1363 // memw(Rx++#s4:2:circ(Mu))=Nt.new
1364 // memw(Rx++I:circ(Mu))=Nt.new
1365 // memw(Rx++Mu)=Nt.new
1366 // memw(Rx++Mu:brev)=Nt.new
1368 //===----------------------------------------------------------------------===//
1370 //===----------------------------------------------------------------------===//
1372 //===----------------------------------------------------------------------===//
1374 //===----------------------------------------------------------------------===//
1376 //===----------------------------------------------------------------------===//
1377 // multiclass/template class for the new-value compare jumps with the register
1379 //===----------------------------------------------------------------------===//
1381 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11,
1382 opExtentAlign = 2 in
1383 class NVJrr_template<string mnemonic, bits<3> majOp, bit NvOpNum,
1384 bit isNegCond, bit isTak>
1386 (ins IntRegs:$src1, IntRegs:$src2, brtarget:$offset),
1387 "if ("#!if(isNegCond, "!","")#mnemonic#
1388 "($src1"#!if(!eq(NvOpNum, 0),".new, ",", ")#
1389 "$src2"#!if(!eq(NvOpNum, 1),".new))","))")#" jump:"
1390 #!if(isTak, "t","nt")#" $offset", []> {
1394 bits<3> Ns; // New-Value Operand
1395 bits<5> RegOp; // Non-New-Value Operand
1398 let isTaken = isTak;
1399 let isPredicatedFalse = isNegCond;
1400 let opNewValue{0} = NvOpNum;
1402 let Ns = !if(!eq(NvOpNum, 0), src1{2-0}, src2{2-0});
1403 let RegOp = !if(!eq(NvOpNum, 0), src2, src1);
1405 let IClass = 0b0010;
1407 let Inst{25-23} = majOp;
1408 let Inst{22} = isNegCond;
1409 let Inst{18-16} = Ns;
1410 let Inst{13} = isTak;
1411 let Inst{12-8} = RegOp;
1412 let Inst{21-20} = offset{10-9};
1413 let Inst{7-1} = offset{8-2};
1417 multiclass NVJrr_cond<string mnemonic, bits<3> majOp, bit NvOpNum,
1419 // Branch not taken:
1420 def _nt_V4: NVJrr_template<mnemonic, majOp, NvOpNum, isNegCond, 0>;
1422 def _t_V4: NVJrr_template<mnemonic, majOp, NvOpNum, isNegCond, 1>;
1425 // NvOpNum = 0 -> First Operand is a new-value Register
1426 // NvOpNum = 1 -> Second Operand is a new-value Register
1428 multiclass NVJrr_base<string mnemonic, string BaseOp, bits<3> majOp,
1430 let BaseOpcode = BaseOp#_NVJ in {
1431 defm _t_Jumpnv : NVJrr_cond<mnemonic, majOp, NvOpNum, 0>; // True cond
1432 defm _f_Jumpnv : NVJrr_cond<mnemonic, majOp, NvOpNum, 1>; // False cond
1436 // if ([!]cmp.eq(Ns.new,Rt)) jump:[n]t #r9:2
1437 // if ([!]cmp.gt(Ns.new,Rt)) jump:[n]t #r9:2
1438 // if ([!]cmp.gtu(Ns.new,Rt)) jump:[n]t #r9:2
1439 // if ([!]cmp.gt(Rt,Ns.new)) jump:[n]t #r9:2
1440 // if ([!]cmp.gtu(Rt,Ns.new)) jump:[n]t #r9:2
1442 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1,
1443 Defs = [PC], hasSideEffects = 0, validSubTargets = HasV4SubT,
1444 isCodeGenOnly = 0 in {
1445 defm CMPEQrr : NVJrr_base<"cmp.eq", "CMPEQ", 0b000, 0>, PredRel;
1446 defm CMPGTrr : NVJrr_base<"cmp.gt", "CMPGT", 0b001, 0>, PredRel;
1447 defm CMPGTUrr : NVJrr_base<"cmp.gtu", "CMPGTU", 0b010, 0>, PredRel;
1448 defm CMPLTrr : NVJrr_base<"cmp.gt", "CMPLT", 0b011, 1>, PredRel;
1449 defm CMPLTUrr : NVJrr_base<"cmp.gtu", "CMPLTU", 0b100, 1>, PredRel;
1452 //===----------------------------------------------------------------------===//
1453 // multiclass/template class for the new-value compare jumps instruction
1454 // with a register and an unsigned immediate (U5) operand.
1455 //===----------------------------------------------------------------------===//
1457 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 11,
1458 opExtentAlign = 2 in
1459 class NVJri_template<string mnemonic, bits<3> majOp, bit isNegCond,
1462 (ins IntRegs:$src1, u5Imm:$src2, brtarget:$offset),
1463 "if ("#!if(isNegCond, "!","")#mnemonic#"($src1.new, #$src2)) jump:"
1464 #!if(isTak, "t","nt")#" $offset", []> {
1466 let isTaken = isTak;
1467 let isPredicatedFalse = isNegCond;
1468 let isTaken = isTak;
1474 let IClass = 0b0010;
1476 let Inst{25-23} = majOp;
1477 let Inst{22} = isNegCond;
1478 let Inst{18-16} = src1;
1479 let Inst{13} = isTak;
1480 let Inst{12-8} = src2;
1481 let Inst{21-20} = offset{10-9};
1482 let Inst{7-1} = offset{8-2};
1485 multiclass NVJri_cond<string mnemonic, bits<3> majOp, bit isNegCond> {
1486 // Branch not taken:
1487 def _nt_V4: NVJri_template<mnemonic, majOp, isNegCond, 0>;
1489 def _t_V4: NVJri_template<mnemonic, majOp, isNegCond, 1>;
1492 multiclass NVJri_base<string mnemonic, string BaseOp, bits<3> majOp> {
1493 let BaseOpcode = BaseOp#_NVJri in {
1494 defm _t_Jumpnv : NVJri_cond<mnemonic, majOp, 0>; // True Cond
1495 defm _f_Jumpnv : NVJri_cond<mnemonic, majOp, 1>; // False cond
1499 // if ([!]cmp.eq(Ns.new,#U5)) jump:[n]t #r9:2
1500 // if ([!]cmp.gt(Ns.new,#U5)) jump:[n]t #r9:2
1501 // if ([!]cmp.gtu(Ns.new,#U5)) jump:[n]t #r9:2
1503 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator = 1,
1504 Defs = [PC], hasSideEffects = 0, validSubTargets = HasV4SubT,
1505 isCodeGenOnly = 0 in {
1506 defm CMPEQri : NVJri_base<"cmp.eq", "CMPEQ", 0b000>, PredRel;
1507 defm CMPGTri : NVJri_base<"cmp.gt", "CMPGT", 0b001>, PredRel;
1508 defm CMPGTUri : NVJri_base<"cmp.gtu", "CMPGTU", 0b010>, PredRel;
1511 //===----------------------------------------------------------------------===//
1512 // multiclass/template class for the new-value compare jumps instruction
1513 // with a register and an hardcoded 0/-1 immediate value.
1514 //===----------------------------------------------------------------------===//
1516 let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 11,
1517 opExtentAlign = 2 in
1518 class NVJ_ConstImm_template<string mnemonic, bits<3> majOp, string ImmVal,
1519 bit isNegCond, bit isTak>
1521 (ins IntRegs:$src1, brtarget:$offset),
1522 "if ("#!if(isNegCond, "!","")#mnemonic
1523 #"($src1.new, #"#ImmVal#")) jump:"
1524 #!if(isTak, "t","nt")#" $offset", []> {
1526 let isTaken = isTak;
1527 let isPredicatedFalse = isNegCond;
1528 let isTaken = isTak;
1532 let IClass = 0b0010;
1534 let Inst{25-23} = majOp;
1535 let Inst{22} = isNegCond;
1536 let Inst{18-16} = src1;
1537 let Inst{13} = isTak;
1538 let Inst{21-20} = offset{10-9};
1539 let Inst{7-1} = offset{8-2};
1542 multiclass NVJ_ConstImm_cond<string mnemonic, bits<3> majOp, string ImmVal,
1544 // Branch not taken:
1545 def _nt_V4: NVJ_ConstImm_template<mnemonic, majOp, ImmVal, isNegCond, 0>;
1547 def _t_V4: NVJ_ConstImm_template<mnemonic, majOp, ImmVal, isNegCond, 1>;
1550 multiclass NVJ_ConstImm_base<string mnemonic, string BaseOp, bits<3> majOp,
1552 let BaseOpcode = BaseOp#_NVJ_ConstImm in {
1553 defm _t_Jumpnv : NVJ_ConstImm_cond<mnemonic, majOp, ImmVal, 0>; // True
1554 defm _f_Jumpnv : NVJ_ConstImm_cond<mnemonic, majOp, ImmVal, 1>; // False
1558 // if ([!]tstbit(Ns.new,#0)) jump:[n]t #r9:2
1559 // if ([!]cmp.eq(Ns.new,#-1)) jump:[n]t #r9:2
1560 // if ([!]cmp.gt(Ns.new,#-1)) jump:[n]t #r9:2
1562 let isPredicated = 1, isBranch = 1, isNewValue = 1, isTerminator=1,
1563 Defs = [PC], hasSideEffects = 0, isCodeGenOnly = 0 in {
1564 defm TSTBIT0 : NVJ_ConstImm_base<"tstbit", "TSTBIT", 0b011, "0">, PredRel;
1565 defm CMPEQn1 : NVJ_ConstImm_base<"cmp.eq", "CMPEQ", 0b100, "-1">, PredRel;
1566 defm CMPGTn1 : NVJ_ConstImm_base<"cmp.gt", "CMPGT", 0b101, "-1">, PredRel;
1569 // J4_hintjumpr: Hint indirect conditional jump.
1570 let isBranch = 1, isIndirectBranch = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
1571 def J4_hintjumpr: JRInst <
1576 let IClass = 0b0101;
1577 let Inst{27-21} = 0b0010101;
1578 let Inst{20-16} = Rs;
1581 //===----------------------------------------------------------------------===//
1583 //===----------------------------------------------------------------------===//
1585 //===----------------------------------------------------------------------===//
1587 //===----------------------------------------------------------------------===//
1590 let hasNewValue = 1, isExtendable = 1, opExtendable = 1,
1591 isExtentSigned = 0, opExtentBits = 6, hasSideEffects = 0,
1592 Uses = [PC], validSubTargets = HasV4SubT in
1593 def C4_addipc : CRInst <(outs IntRegs:$Rd), (ins u6Ext:$u6),
1594 "$Rd = add(pc, #$u6)", [], "", CR_tc_2_SLOT3 > {
1598 let IClass = 0b0110;
1599 let Inst{27-16} = 0b101001001001;
1600 let Inst{12-7} = u6;
1606 let hasSideEffects = 0 in
1607 class T_LOGICAL_3OP<string MnOp1, string MnOp2, bits<2> OpBits, bit IsNeg>
1608 : CRInst<(outs PredRegs:$Pd),
1609 (ins PredRegs:$Ps, PredRegs:$Pt, PredRegs:$Pu),
1610 "$Pd = " # MnOp1 # "($Ps, " # MnOp2 # "($Pt, " #
1611 !if (IsNeg,"!","") # "$Pu))",
1612 [], "", CR_tc_2early_SLOT23> {
1618 let IClass = 0b0110;
1619 let Inst{27-24} = 0b1011;
1620 let Inst{23} = IsNeg;
1621 let Inst{22-21} = OpBits;
1623 let Inst{17-16} = Ps;
1630 let isCodeGenOnly = 0 in {
1631 def C4_and_and : T_LOGICAL_3OP<"and", "and", 0b00, 0>;
1632 def C4_and_or : T_LOGICAL_3OP<"and", "or", 0b01, 0>;
1633 def C4_or_and : T_LOGICAL_3OP<"or", "and", 0b10, 0>;
1634 def C4_or_or : T_LOGICAL_3OP<"or", "or", 0b11, 0>;
1635 def C4_and_andn : T_LOGICAL_3OP<"and", "and", 0b00, 1>;
1636 def C4_and_orn : T_LOGICAL_3OP<"and", "or", 0b01, 1>;
1637 def C4_or_andn : T_LOGICAL_3OP<"or", "and", 0b10, 1>;
1638 def C4_or_orn : T_LOGICAL_3OP<"or", "or", 0b11, 1>;
1641 //===----------------------------------------------------------------------===//
1643 //===----------------------------------------------------------------------===//
1645 //===----------------------------------------------------------------------===//
1647 //===----------------------------------------------------------------------===//
1649 // Logical with-not instructions.
1650 let validSubTargets = HasV4SubT, isCodeGenOnly = 0 in {
1651 def A4_andnp : T_ALU64_logical<"and", 0b001, 1, 0, 1>;
1652 def A4_ornp : T_ALU64_logical<"or", 0b011, 1, 0, 1>;
1655 let hasNewValue = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
1656 def S4_parity: ALU64Inst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
1657 "$Rd = parity($Rs, $Rt)", [], "", ALU64_tc_2_SLOT23> {
1662 let IClass = 0b1101;
1663 let Inst{27-21} = 0b0101111;
1664 let Inst{20-16} = Rs;
1665 let Inst{12-8} = Rt;
1668 // Add and accumulate.
1669 // Rd=add(Rs,add(Ru,#s6))
1670 let isExtentSigned = 1, hasNewValue = 1, isExtendable = 1, opExtentBits = 6,
1671 opExtendable = 3, isCodeGenOnly = 0 in
1672 def S4_addaddi : ALU64Inst <(outs IntRegs:$Rd),
1673 (ins IntRegs:$Rs, IntRegs:$Ru, s6Ext:$s6),
1674 "$Rd = add($Rs, add($Ru, #$s6))" ,
1675 [(set (i32 IntRegs:$Rd), (add (i32 IntRegs:$Rs),
1676 (add (i32 IntRegs:$Ru), s6_16ExtPred:$s6)))],
1677 "", ALU64_tc_2_SLOT23> {
1683 let IClass = 0b1101;
1685 let Inst{27-23} = 0b10110;
1686 let Inst{22-21} = s6{5-4};
1687 let Inst{20-16} = Rs;
1688 let Inst{13} = s6{3};
1689 let Inst{12-8} = Rd;
1690 let Inst{7-5} = s6{2-0};
1694 let isExtentSigned = 1, hasSideEffects = 0, hasNewValue = 1, isExtendable = 1,
1695 opExtentBits = 6, opExtendable = 2, isCodeGenOnly = 0 in
1696 def S4_subaddi: ALU64Inst <(outs IntRegs:$Rd),
1697 (ins IntRegs:$Rs, s6Ext:$s6, IntRegs:$Ru),
1698 "$Rd = add($Rs, sub(#$s6, $Ru))",
1699 [], "", ALU64_tc_2_SLOT23> {
1705 let IClass = 0b1101;
1707 let Inst{27-23} = 0b10111;
1708 let Inst{22-21} = s6{5-4};
1709 let Inst{20-16} = Rs;
1710 let Inst{13} = s6{3};
1711 let Inst{12-8} = Rd;
1712 let Inst{7-5} = s6{2-0};
1717 // Rdd=extract(Rss,#u6,#U6)
1718 // Rdd=extract(Rss,Rtt)
1719 // Rd=extract(Rs,Rtt)
1720 // Rd=extract(Rs,#u5,#U5)
1722 let isCodeGenOnly = 0 in {
1723 def S4_extractp_rp : T_S3op_64 < "extract", 0b11, 0b100, 0>;
1724 def S4_extractp : T_S2op_extract <"extract", 0b1010, DoubleRegs, u6Imm>;
1727 let hasNewValue = 1, isCodeGenOnly = 0 in {
1728 def S4_extract_rp : T_S3op_extract<"extract", 0b01>;
1729 def S4_extract : T_S2op_extract <"extract", 0b1101, IntRegs, u5Imm>;
1732 let Itinerary = M_tc_3x_SLOT23, Defs = [USR_OVF], isCodeGenOnly = 0 in {
1733 def M4_mac_up_s1_sat: T_MType_acc_rr<"+= mpy", 0b011, 0b000, 0, [], 0, 1, 1>;
1734 def M4_nac_up_s1_sat: T_MType_acc_rr<"-= mpy", 0b011, 0b001, 0, [], 0, 1, 1>;
1737 // Logical xor with xor accumulation.
1738 // Rxx^=xor(Rss,Rtt)
1739 let hasSideEffects = 0, isCodeGenOnly = 0 in
1741 : SInst <(outs DoubleRegs:$Rxx),
1742 (ins DoubleRegs:$dst2, DoubleRegs:$Rss, DoubleRegs:$Rtt),
1743 "$Rxx ^= xor($Rss, $Rtt)",
1744 [(set (i64 DoubleRegs:$Rxx),
1745 (xor (i64 DoubleRegs:$dst2), (xor (i64 DoubleRegs:$Rss),
1746 (i64 DoubleRegs:$Rtt))))],
1747 "$dst2 = $Rxx", S_3op_tc_1_SLOT23> {
1752 let IClass = 0b1100;
1754 let Inst{27-23} = 0b10101;
1755 let Inst{20-16} = Rss;
1756 let Inst{12-8} = Rtt;
1757 let Inst{4-0} = Rxx;
1761 let isCodeGenOnly = 0 in
1762 def A4_bitspliti : T_S2op_2_di <"bitsplit", 0b110, 0b100>;
1764 // Arithmetic/Convergent round
1765 let isCodeGenOnly = 0 in
1766 def A4_cround_ri : T_S2op_2_ii <"cround", 0b111, 0b000>;
1768 let isCodeGenOnly = 0 in
1769 def A4_round_ri : T_S2op_2_ii <"round", 0b111, 0b100>;
1771 let Defs = [USR_OVF], isCodeGenOnly = 0 in
1772 def A4_round_ri_sat : T_S2op_2_ii <"round", 0b111, 0b110, 1>;
1774 // Logical-logical words.
1775 // Compound or-and -- Rx=or(Ru,and(Rx,#s10))
1776 let isExtentSigned = 1, hasNewValue = 1, isExtendable = 1, opExtentBits = 10,
1777 opExtendable = 3, isCodeGenOnly = 0 in
1779 ALU64Inst<(outs IntRegs:$Rx),
1780 (ins IntRegs:$Ru, IntRegs:$_src_, s10Ext:$s10),
1781 "$Rx = or($Ru, and($_src_, #$s10))" ,
1782 [(set (i32 IntRegs:$Rx),
1783 (or (i32 IntRegs:$Ru), (and (i32 IntRegs:$_src_), s10ExtPred:$s10)))] ,
1784 "$_src_ = $Rx", ALU64_tc_2_SLOT23> {
1789 let IClass = 0b1101;
1791 let Inst{27-22} = 0b101001;
1792 let Inst{20-16} = Rx;
1793 let Inst{21} = s10{9};
1794 let Inst{13-5} = s10{8-0};
1798 // Miscellaneous ALU64 instructions.
1800 let hasNewValue = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
1801 def A4_modwrapu: ALU64Inst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, IntRegs:$Rt),
1802 "$Rd = modwrap($Rs, $Rt)", [], "", ALU64_tc_2_SLOT23> {
1807 let IClass = 0b1101;
1808 let Inst{27-21} = 0b0011111;
1809 let Inst{20-16} = Rs;
1810 let Inst{12-8} = Rt;
1811 let Inst{7-5} = 0b111;
1815 let hasSideEffects = 0, isCodeGenOnly = 0 in
1816 def A4_bitsplit: ALU64Inst<(outs DoubleRegs:$Rd),
1817 (ins IntRegs:$Rs, IntRegs:$Rt),
1818 "$Rd = bitsplit($Rs, $Rt)", [], "", ALU64_tc_1_SLOT23> {
1823 let IClass = 0b1101;
1824 let Inst{27-24} = 0b0100;
1826 let Inst{20-16} = Rs;
1827 let Inst{12-8} = Rt;
1831 let isCodeGenOnly = 0 in {
1832 // Rx[&|]=xor(Rs,Rt)
1833 def M4_or_xor : T_MType_acc_rr < "|= xor", 0b110, 0b001, 0>;
1834 def M4_and_xor : T_MType_acc_rr < "&= xor", 0b010, 0b010, 0>;
1836 // Rx[&|^]=or(Rs,Rt)
1837 def M4_xor_or : T_MType_acc_rr < "^= or", 0b110, 0b011, 0>;
1839 let CextOpcode = "ORr_ORr" in
1840 def M4_or_or : T_MType_acc_rr < "|= or", 0b110, 0b000, 0>;
1841 def M4_and_or : T_MType_acc_rr < "&= or", 0b010, 0b001, 0>;
1843 // Rx[&|^]=and(Rs,Rt)
1844 def M4_xor_and : T_MType_acc_rr < "^= and", 0b110, 0b010, 0>;
1846 let CextOpcode = "ORr_ANDr" in
1847 def M4_or_and : T_MType_acc_rr < "|= and", 0b010, 0b011, 0>;
1848 def M4_and_and : T_MType_acc_rr < "&= and", 0b010, 0b000, 0>;
1850 // Rx[&|^]=and(Rs,~Rt)
1851 def M4_xor_andn : T_MType_acc_rr < "^= and", 0b001, 0b010, 0, [], 1>;
1852 def M4_or_andn : T_MType_acc_rr < "|= and", 0b001, 0b000, 0, [], 1>;
1853 def M4_and_andn : T_MType_acc_rr < "&= and", 0b001, 0b001, 0, [], 1>;
1856 // Compound or-or and or-and
1857 let isExtentSigned = 1, InputType = "imm", hasNewValue = 1, isExtendable = 1,
1858 opExtentBits = 10, opExtendable = 3 in
1859 class T_CompOR <string mnemonic, bits<2> MajOp, SDNode OpNode>
1860 : MInst_acc <(outs IntRegs:$Rx),
1861 (ins IntRegs:$src1, IntRegs:$Rs, s10Ext:$s10),
1862 "$Rx |= "#mnemonic#"($Rs, #$s10)",
1863 [(set (i32 IntRegs:$Rx), (or (i32 IntRegs:$src1),
1864 (OpNode (i32 IntRegs:$Rs), s10ExtPred:$s10)))],
1865 "$src1 = $Rx", ALU64_tc_2_SLOT23>, ImmRegRel {
1870 let IClass = 0b1101;
1872 let Inst{27-24} = 0b1010;
1873 let Inst{23-22} = MajOp;
1874 let Inst{20-16} = Rs;
1875 let Inst{21} = s10{9};
1876 let Inst{13-5} = s10{8-0};
1880 let CextOpcode = "ORr_ANDr", isCodeGenOnly = 0 in
1881 def S4_or_andi : T_CompOR <"and", 0b00, and>;
1883 let CextOpcode = "ORr_ORr", isCodeGenOnly = 0 in
1884 def S4_or_ori : T_CompOR <"or", 0b10, or>;
1887 // Rd=modwrap(Rs,Rt)
1889 // Rd=cround(Rs,#u5)
1891 // Rd=round(Rs,#u5)[:sat]
1892 // Rd=round(Rs,Rt)[:sat]
1893 // Vector reduce add unsigned halfwords
1894 // Rd=vraddh(Rss,Rtt)
1896 // Rdd=vaddb(Rss,Rtt)
1897 // Vector conditional negate
1898 // Rdd=vcnegh(Rss,Rt)
1899 // Rxx+=vrcnegh(Rss,Rt)
1900 // Vector maximum bytes
1901 // Rdd=vmaxb(Rtt,Rss)
1902 // Vector reduce maximum halfwords
1903 // Rxx=vrmaxh(Rss,Ru)
1904 // Rxx=vrmaxuh(Rss,Ru)
1905 // Vector reduce maximum words
1906 // Rxx=vrmaxuw(Rss,Ru)
1907 // Rxx=vrmaxw(Rss,Ru)
1908 // Vector minimum bytes
1909 // Rdd=vminb(Rtt,Rss)
1910 // Vector reduce minimum halfwords
1911 // Rxx=vrminh(Rss,Ru)
1912 // Rxx=vrminuh(Rss,Ru)
1913 // Vector reduce minimum words
1914 // Rxx=vrminuw(Rss,Ru)
1915 // Rxx=vrminw(Rss,Ru)
1916 // Vector subtract bytes
1917 // Rdd=vsubb(Rss,Rtt)
1919 //===----------------------------------------------------------------------===//
1921 //===----------------------------------------------------------------------===//
1923 //===----------------------------------------------------------------------===//
1925 //===----------------------------------------------------------------------===//
1928 let isCodeGenOnly = 0 in
1929 def S2_brevp : T_S2op_3 <"brev", 0b11, 0b110>;
1932 let isCodeGenOnly = 0 in {
1933 def S2_ct0p : T_COUNT_LEADING_64<"ct0", 0b111, 0b010>;
1934 def S2_ct1p : T_COUNT_LEADING_64<"ct1", 0b111, 0b100>;
1935 def S4_clbpnorm : T_COUNT_LEADING_64<"normamt", 0b011, 0b000>;
1938 def: Pat<(i32 (trunc (cttz (i64 DoubleRegs:$Rss)))),
1939 (S2_ct0p (i64 DoubleRegs:$Rss))>;
1940 def: Pat<(i32 (trunc (cttz (not (i64 DoubleRegs:$Rss))))),
1941 (S2_ct1p (i64 DoubleRegs:$Rss))>;
1943 let hasSideEffects = 0, hasNewValue = 1, isCodeGenOnly = 0 in
1944 def S4_clbaddi : SInst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, s6Imm:$s6),
1945 "$Rd = add(clb($Rs), #$s6)", [], "", S_2op_tc_2_SLOT23> {
1949 let IClass = 0b1000;
1950 let Inst{27-24} = 0b1100;
1951 let Inst{23-21} = 0b001;
1952 let Inst{20-16} = Rs;
1953 let Inst{13-8} = s6;
1954 let Inst{7-5} = 0b000;
1958 let hasSideEffects = 0, hasNewValue = 1, isCodeGenOnly = 0 in
1959 def S4_clbpaddi : SInst<(outs IntRegs:$Rd), (ins DoubleRegs:$Rs, s6Imm:$s6),
1960 "$Rd = add(clb($Rs), #$s6)", [], "", S_2op_tc_2_SLOT23> {
1964 let IClass = 0b1000;
1965 let Inst{27-24} = 0b1000;
1966 let Inst{23-21} = 0b011;
1967 let Inst{20-16} = Rs;
1968 let Inst{13-8} = s6;
1969 let Inst{7-5} = 0b010;
1974 // Bit test/set/clear
1975 let isCodeGenOnly = 0 in {
1976 def S4_ntstbit_i : T_TEST_BIT_IMM<"!tstbit", 0b001>;
1977 def S4_ntstbit_r : T_TEST_BIT_REG<"!tstbit", 1>;
1980 let AddedComplexity = 20 in { // Complexity greater than cmp reg-imm.
1981 def: Pat<(i1 (seteq (and (shl 1, u5ImmPred:$u5), (i32 IntRegs:$Rs)), 0)),
1982 (S4_ntstbit_i (i32 IntRegs:$Rs), u5ImmPred:$u5)>;
1983 def: Pat<(i1 (seteq (and (shl 1, (i32 IntRegs:$Rt)), (i32 IntRegs:$Rs)), 0)),
1984 (S4_ntstbit_r (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))>;
1987 // Add extra complexity to prefer these instructions over bitsset/bitsclr.
1988 // The reason is that tstbit/ntstbit can be folded into a compound instruction:
1989 // if ([!]tstbit(...)) jump ...
1990 let AddedComplexity = 100 in
1991 def: Pat<(i1 (setne (and (i32 IntRegs:$Rs), (i32 Set5ImmPred:$u5)), (i32 0))),
1992 (S2_tstbit_i (i32 IntRegs:$Rs), (BITPOS32 Set5ImmPred:$u5))>;
1994 let AddedComplexity = 100 in
1995 def: Pat<(i1 (seteq (and (i32 IntRegs:$Rs), (i32 Set5ImmPred:$u5)), (i32 0))),
1996 (S4_ntstbit_i (i32 IntRegs:$Rs), (BITPOS32 Set5ImmPred:$u5))>;
1998 let isCodeGenOnly = 0 in {
1999 def C4_nbitsset : T_TEST_BITS_REG<"!bitsset", 0b01, 1>;
2000 def C4_nbitsclr : T_TEST_BITS_REG<"!bitsclr", 0b10, 1>;
2001 def C4_nbitsclri : T_TEST_BITS_IMM<"!bitsclr", 0b10, 1>;
2004 // Do not increase complexity of these patterns. In the DAG, "cmp i8" may be
2005 // represented as a compare against "value & 0xFF", which is an exact match
2006 // for cmpb (same for cmph). The patterns below do not contain any additional
2007 // complexity that would make them preferable, and if they were actually used
2008 // instead of cmpb/cmph, they would result in a compare against register that
2009 // is loaded with the byte/half mask (i.e. 0xFF or 0xFFFF).
2010 def: Pat<(i1 (setne (and I32:$Rs, u6ImmPred:$u6), 0)),
2011 (C4_nbitsclri I32:$Rs, u6ImmPred:$u6)>;
2012 def: Pat<(i1 (setne (and I32:$Rs, I32:$Rt), 0)),
2013 (C4_nbitsclr I32:$Rs, I32:$Rt)>;
2014 def: Pat<(i1 (setne (and I32:$Rs, I32:$Rt), I32:$Rt)),
2015 (C4_nbitsset I32:$Rs, I32:$Rt)>;
2017 //===----------------------------------------------------------------------===//
2019 //===----------------------------------------------------------------------===//
2021 //===----------------------------------------------------------------------===//
2023 //===----------------------------------------------------------------------===//
2025 // Rd=add(#u6,mpyi(Rs,#U6)) -- Multiply by immed and add immed.
2027 let hasNewValue = 1, isExtendable = 1, opExtentBits = 6, opExtendable = 1,
2028 isCodeGenOnly = 0 in
2029 def M4_mpyri_addi : MInst<(outs IntRegs:$Rd),
2030 (ins u6Ext:$u6, IntRegs:$Rs, u6Imm:$U6),
2031 "$Rd = add(#$u6, mpyi($Rs, #$U6))" ,
2032 [(set (i32 IntRegs:$Rd),
2033 (add (mul (i32 IntRegs:$Rs), u6ImmPred:$U6),
2034 u6ExtPred:$u6))] ,"",ALU64_tc_3x_SLOT23> {
2040 let IClass = 0b1101;
2042 let Inst{27-24} = 0b1000;
2043 let Inst{23} = U6{5};
2044 let Inst{22-21} = u6{5-4};
2045 let Inst{20-16} = Rs;
2046 let Inst{13} = u6{3};
2047 let Inst{12-8} = Rd;
2048 let Inst{7-5} = u6{2-0};
2049 let Inst{4-0} = U6{4-0};
2052 // Rd=add(#u6,mpyi(Rs,Rt))
2053 let CextOpcode = "ADD_MPY", InputType = "imm", hasNewValue = 1,
2054 isExtendable = 1, opExtentBits = 6, opExtendable = 1, isCodeGenOnly = 0 in
2055 def M4_mpyrr_addi : MInst <(outs IntRegs:$Rd),
2056 (ins u6Ext:$u6, IntRegs:$Rs, IntRegs:$Rt),
2057 "$Rd = add(#$u6, mpyi($Rs, $Rt))" ,
2058 [(set (i32 IntRegs:$Rd),
2059 (add (mul (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), u6ExtPred:$u6))],
2060 "", ALU64_tc_3x_SLOT23>, ImmRegRel {
2066 let IClass = 0b1101;
2068 let Inst{27-23} = 0b01110;
2069 let Inst{22-21} = u6{5-4};
2070 let Inst{20-16} = Rs;
2071 let Inst{13} = u6{3};
2072 let Inst{12-8} = Rt;
2073 let Inst{7-5} = u6{2-0};
2077 let hasNewValue = 1 in
2078 class T_AddMpy <bit MajOp, PatLeaf ImmPred, dag ins>
2079 : ALU64Inst <(outs IntRegs:$dst), ins,
2080 "$dst = add($src1, mpyi("#!if(MajOp,"$src3, #$src2))",
2082 [(set (i32 IntRegs:$dst),
2083 (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src3), ImmPred:$src2)))],
2084 "", ALU64_tc_3x_SLOT23> {
2090 let IClass = 0b1101;
2092 bits<6> ImmValue = !if(MajOp, src2{5-0}, src2{7-2});
2094 let Inst{27-24} = 0b1111;
2095 let Inst{23} = MajOp;
2096 let Inst{22-21} = ImmValue{5-4};
2097 let Inst{20-16} = src3;
2098 let Inst{13} = ImmValue{3};
2099 let Inst{12-8} = dst;
2100 let Inst{7-5} = ImmValue{2-0};
2101 let Inst{4-0} = src1;
2104 let isCodeGenOnly = 0 in
2105 def M4_mpyri_addr_u2 : T_AddMpy<0b0, u6_2ImmPred,
2106 (ins IntRegs:$src1, u6_2Imm:$src2, IntRegs:$src3)>;
2108 let isExtendable = 1, opExtentBits = 6, opExtendable = 3,
2109 CextOpcode = "ADD_MPY", InputType = "imm", isCodeGenOnly = 0 in
2110 def M4_mpyri_addr : T_AddMpy<0b1, u6ExtPred,
2111 (ins IntRegs:$src1, IntRegs:$src3, u6Ext:$src2)>, ImmRegRel;
2113 // Rx=add(Ru,mpyi(Rx,Rs))
2114 let validSubTargets = HasV4SubT, CextOpcode = "ADD_MPY", InputType = "reg",
2115 hasNewValue = 1, isCodeGenOnly = 0 in
2116 def M4_mpyrr_addr: MInst_acc <(outs IntRegs:$Rx),
2117 (ins IntRegs:$Ru, IntRegs:$_src_, IntRegs:$Rs),
2118 "$Rx = add($Ru, mpyi($_src_, $Rs))",
2119 [(set (i32 IntRegs:$Rx), (add (i32 IntRegs:$Ru),
2120 (mul (i32 IntRegs:$_src_), (i32 IntRegs:$Rs))))],
2121 "$_src_ = $Rx", M_tc_3x_SLOT23>, ImmRegRel {
2126 let IClass = 0b1110;
2128 let Inst{27-21} = 0b0011000;
2129 let Inst{12-8} = Rx;
2131 let Inst{20-16} = Rs;
2134 // Rd=add(##,mpyi(Rs,#U6))
2135 def : Pat <(add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
2136 (HexagonCONST32 tglobaladdr:$src1)),
2137 (i32 (M4_mpyri_addi tglobaladdr:$src1, IntRegs:$src2,
2140 // Rd=add(##,mpyi(Rs,Rt))
2141 def : Pat <(add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
2142 (HexagonCONST32 tglobaladdr:$src1)),
2143 (i32 (M4_mpyrr_addi tglobaladdr:$src1, IntRegs:$src2,
2146 // Polynomial multiply words
2148 // Rxx^=pmpyw(Rs,Rt)
2150 // Vector reduce multiply word by signed half (32x16)
2151 // Rdd=vrmpyweh(Rss,Rtt)[:<<1]
2152 // Rdd=vrmpywoh(Rss,Rtt)[:<<1]
2153 // Rxx+=vrmpyweh(Rss,Rtt)[:<<1]
2154 // Rxx+=vrmpywoh(Rss,Rtt)[:<<1]
2156 // Multiply and use upper result
2157 // Rd=mpy(Rs,Rt.H):<<1:sat
2158 // Rd=mpy(Rs,Rt.L):<<1:sat
2159 // Rd=mpy(Rs,Rt):<<1
2160 // Rd=mpy(Rs,Rt):<<1:sat
2162 // Rx+=mpy(Rs,Rt):<<1:sat
2163 // Rx-=mpy(Rs,Rt):<<1:sat
2165 // Vector multiply bytes
2166 // Rdd=vmpybsu(Rs,Rt)
2167 // Rdd=vmpybu(Rs,Rt)
2168 // Rxx+=vmpybsu(Rs,Rt)
2169 // Rxx+=vmpybu(Rs,Rt)
2171 // Vector polynomial multiply halfwords
2172 // Rdd=vpmpyh(Rs,Rt)
2173 // Rxx^=vpmpyh(Rs,Rt)
2175 //===----------------------------------------------------------------------===//
2177 //===----------------------------------------------------------------------===//
2180 //===----------------------------------------------------------------------===//
2182 //===----------------------------------------------------------------------===//
2183 // Shift by immediate and accumulate/logical.
2184 // Rx=add(#u8,asl(Rx,#U5)) Rx=add(#u8,lsr(Rx,#U5))
2185 // Rx=sub(#u8,asl(Rx,#U5)) Rx=sub(#u8,lsr(Rx,#U5))
2186 // Rx=and(#u8,asl(Rx,#U5)) Rx=and(#u8,lsr(Rx,#U5))
2187 // Rx=or(#u8,asl(Rx,#U5)) Rx=or(#u8,lsr(Rx,#U5))
2188 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
2189 hasNewValue = 1, opNewValue = 0, validSubTargets = HasV4SubT in
2190 class T_S4_ShiftOperate<string MnOp, string MnSh, SDNode Op, SDNode Sh,
2191 bit asl_lsr, bits<2> MajOp, InstrItinClass Itin>
2192 : MInst_acc<(outs IntRegs:$Rd), (ins u8Ext:$u8, IntRegs:$Rx, u5Imm:$U5),
2193 "$Rd = "#MnOp#"(#$u8, "#MnSh#"($Rx, #$U5))",
2194 [(set (i32 IntRegs:$Rd),
2195 (Op (Sh I32:$Rx, u5ImmPred:$U5), u8ExtPred:$u8))],
2196 "$Rd = $Rx", Itin> {
2203 let IClass = 0b1101;
2204 let Inst{27-24} = 0b1110;
2205 let Inst{23-21} = u8{7-5};
2206 let Inst{20-16} = Rd;
2207 let Inst{13} = u8{4};
2208 let Inst{12-8} = U5;
2209 let Inst{7-5} = u8{3-1};
2210 let Inst{4} = asl_lsr;
2211 let Inst{3} = u8{0};
2212 let Inst{2-1} = MajOp;
2215 multiclass T_ShiftOperate<string mnemonic, SDNode Op, bits<2> MajOp,
2216 InstrItinClass Itin> {
2217 def _asl_ri : T_S4_ShiftOperate<mnemonic, "asl", Op, shl, 0, MajOp, Itin>;
2218 def _lsr_ri : T_S4_ShiftOperate<mnemonic, "lsr", Op, srl, 1, MajOp, Itin>;
2221 let AddedComplexity = 200, isCodeGenOnly = 0 in {
2222 defm S4_addi : T_ShiftOperate<"add", add, 0b10, ALU64_tc_2_SLOT23>;
2223 defm S4_andi : T_ShiftOperate<"and", and, 0b00, ALU64_tc_2_SLOT23>;
2226 let AddedComplexity = 30, isCodeGenOnly = 0 in
2227 defm S4_ori : T_ShiftOperate<"or", or, 0b01, ALU64_tc_1_SLOT23>;
2229 let isCodeGenOnly = 0 in
2230 defm S4_subi : T_ShiftOperate<"sub", sub, 0b11, ALU64_tc_1_SLOT23>;
2233 // Rd=[cround|round](Rs,Rt)
2234 let hasNewValue = 1, Itinerary = S_3op_tc_2_SLOT23, isCodeGenOnly = 0 in {
2235 def A4_cround_rr : T_S3op_3 < "cround", IntRegs, 0b11, 0b00>;
2236 def A4_round_rr : T_S3op_3 < "round", IntRegs, 0b11, 0b10>;
2239 // Rd=round(Rs,Rt):sat
2240 let hasNewValue = 1, Defs = [USR_OVF], Itinerary = S_3op_tc_2_SLOT23,
2241 isCodeGenOnly = 0 in
2242 def A4_round_rr_sat : T_S3op_3 < "round", IntRegs, 0b11, 0b11, 1>;
2244 // Rdd=[add|sub](Rss,Rtt,Px):carry
2245 let isPredicateLate = 1, hasSideEffects = 0 in
2246 class T_S3op_carry <string mnemonic, bits<3> MajOp>
2247 : SInst < (outs DoubleRegs:$Rdd, PredRegs:$Px),
2248 (ins DoubleRegs:$Rss, DoubleRegs:$Rtt, PredRegs:$Pu),
2249 "$Rdd = "#mnemonic#"($Rss, $Rtt, $Pu):carry",
2250 [], "$Px = $Pu", S_3op_tc_1_SLOT23 > {
2256 let IClass = 0b1100;
2258 let Inst{27-24} = 0b0010;
2259 let Inst{23-21} = MajOp;
2260 let Inst{20-16} = Rss;
2261 let Inst{12-8} = Rtt;
2263 let Inst{4-0} = Rdd;
2266 let isCodeGenOnly = 0 in {
2267 def A4_addp_c : T_S3op_carry < "add", 0b110 >;
2268 def A4_subp_c : T_S3op_carry < "sub", 0b111 >;
2271 // Shift an immediate left by register amount.
2272 let hasNewValue = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
2273 def S4_lsli: SInst <(outs IntRegs:$Rd), (ins s6Imm:$s6, IntRegs:$Rt),
2274 "$Rd = lsl(#$s6, $Rt)" ,
2275 [(set (i32 IntRegs:$Rd), (shl s6ImmPred:$s6,
2276 (i32 IntRegs:$Rt)))],
2277 "", S_3op_tc_1_SLOT23> {
2282 let IClass = 0b1100;
2284 let Inst{27-22} = 0b011010;
2285 let Inst{20-16} = s6{5-1};
2286 let Inst{12-8} = Rt;
2287 let Inst{7-6} = 0b11;
2289 let Inst{5} = s6{0};
2292 //===----------------------------------------------------------------------===//
2294 //===----------------------------------------------------------------------===//
2296 //===----------------------------------------------------------------------===//
2297 // MEMOP: Word, Half, Byte
2298 //===----------------------------------------------------------------------===//
2300 def MEMOPIMM : SDNodeXForm<imm, [{
2301 // Call the transformation function XformM5ToU5Imm to get the negative
2302 // immediate's positive counterpart.
2303 int32_t imm = N->getSExtValue();
2304 return XformM5ToU5Imm(imm);
2307 def MEMOPIMM_HALF : SDNodeXForm<imm, [{
2308 // -1 .. -31 represented as 65535..65515
2309 // assigning to a short restores our desired signed value.
2310 // Call the transformation function XformM5ToU5Imm to get the negative
2311 // immediate's positive counterpart.
2312 int16_t imm = N->getSExtValue();
2313 return XformM5ToU5Imm(imm);
2316 def MEMOPIMM_BYTE : SDNodeXForm<imm, [{
2317 // -1 .. -31 represented as 255..235
2318 // assigning to a char restores our desired signed value.
2319 // Call the transformation function XformM5ToU5Imm to get the negative
2320 // immediate's positive counterpart.
2321 int8_t imm = N->getSExtValue();
2322 return XformM5ToU5Imm(imm);
2325 def SETMEMIMM : SDNodeXForm<imm, [{
2326 // Return the bit position we will set [0-31].
2328 int32_t imm = N->getSExtValue();
2329 return XformMskToBitPosU5Imm(imm);
2332 def CLRMEMIMM : SDNodeXForm<imm, [{
2333 // Return the bit position we will clear [0-31].
2335 // we bit negate the value first
2336 int32_t imm = ~(N->getSExtValue());
2337 return XformMskToBitPosU5Imm(imm);
2340 def SETMEMIMM_SHORT : SDNodeXForm<imm, [{
2341 // Return the bit position we will set [0-15].
2343 int16_t imm = N->getSExtValue();
2344 return XformMskToBitPosU4Imm(imm);
2347 def CLRMEMIMM_SHORT : SDNodeXForm<imm, [{
2348 // Return the bit position we will clear [0-15].
2350 // we bit negate the value first
2351 int16_t imm = ~(N->getSExtValue());
2352 return XformMskToBitPosU4Imm(imm);
2355 def SETMEMIMM_BYTE : SDNodeXForm<imm, [{
2356 // Return the bit position we will set [0-7].
2358 int8_t imm = N->getSExtValue();
2359 return XformMskToBitPosU3Imm(imm);
2362 def CLRMEMIMM_BYTE : SDNodeXForm<imm, [{
2363 // Return the bit position we will clear [0-7].
2365 // we bit negate the value first
2366 int8_t imm = ~(N->getSExtValue());
2367 return XformMskToBitPosU3Imm(imm);
2370 //===----------------------------------------------------------------------===//
2371 // Template class for MemOp instructions with the register value.
2372 //===----------------------------------------------------------------------===//
2373 class MemOp_rr_base <string opc, bits<2> opcBits, Operand ImmOp,
2374 string memOp, bits<2> memOpBits> :
2376 (ins IntRegs:$base, ImmOp:$offset, IntRegs:$delta),
2377 opc#"($base+#$offset)"#memOp#"$delta",
2379 Requires<[UseMEMOP]> {
2384 bits<6> offsetBits; // memb - u6:0 , memh - u6:1, memw - u6:2
2386 let offsetBits = !if (!eq(opcBits, 0b00), offset{5-0},
2387 !if (!eq(opcBits, 0b01), offset{6-1},
2388 !if (!eq(opcBits, 0b10), offset{7-2},0)));
2390 let opExtentAlign = opcBits;
2391 let IClass = 0b0011;
2392 let Inst{27-24} = 0b1110;
2393 let Inst{22-21} = opcBits;
2394 let Inst{20-16} = base;
2396 let Inst{12-7} = offsetBits;
2397 let Inst{6-5} = memOpBits;
2398 let Inst{4-0} = delta;
2401 //===----------------------------------------------------------------------===//
2402 // Template class for MemOp instructions with the immediate value.
2403 //===----------------------------------------------------------------------===//
2404 class MemOp_ri_base <string opc, bits<2> opcBits, Operand ImmOp,
2405 string memOp, bits<2> memOpBits> :
2407 (ins IntRegs:$base, ImmOp:$offset, u5Imm:$delta),
2408 opc#"($base+#$offset)"#memOp#"#$delta"
2409 #!if(memOpBits{1},")", ""), // clrbit, setbit - include ')'
2411 Requires<[UseMEMOP]> {
2416 bits<6> offsetBits; // memb - u6:0 , memh - u6:1, memw - u6:2
2418 let offsetBits = !if (!eq(opcBits, 0b00), offset{5-0},
2419 !if (!eq(opcBits, 0b01), offset{6-1},
2420 !if (!eq(opcBits, 0b10), offset{7-2},0)));
2422 let opExtentAlign = opcBits;
2423 let IClass = 0b0011;
2424 let Inst{27-24} = 0b1111;
2425 let Inst{22-21} = opcBits;
2426 let Inst{20-16} = base;
2428 let Inst{12-7} = offsetBits;
2429 let Inst{6-5} = memOpBits;
2430 let Inst{4-0} = delta;
2433 // multiclass to define MemOp instructions with register operand.
2434 multiclass MemOp_rr<string opc, bits<2> opcBits, Operand ImmOp> {
2435 def L4_add#NAME : MemOp_rr_base <opc, opcBits, ImmOp, " += ", 0b00>; // add
2436 def L4_sub#NAME : MemOp_rr_base <opc, opcBits, ImmOp, " -= ", 0b01>; // sub
2437 def L4_and#NAME : MemOp_rr_base <opc, opcBits, ImmOp, " &= ", 0b10>; // and
2438 def L4_or#NAME : MemOp_rr_base <opc, opcBits, ImmOp, " |= ", 0b11>; // or
2441 // multiclass to define MemOp instructions with immediate Operand.
2442 multiclass MemOp_ri<string opc, bits<2> opcBits, Operand ImmOp> {
2443 def L4_iadd#NAME : MemOp_ri_base <opc, opcBits, ImmOp, " += ", 0b00 >;
2444 def L4_isub#NAME : MemOp_ri_base <opc, opcBits, ImmOp, " -= ", 0b01 >;
2445 def L4_iand#NAME : MemOp_ri_base<opc, opcBits, ImmOp, " = clrbit(", 0b10>;
2446 def L4_ior#NAME : MemOp_ri_base<opc, opcBits, ImmOp, " = setbit(", 0b11>;
2449 multiclass MemOp_base <string opc, bits<2> opcBits, Operand ImmOp> {
2450 defm _#NAME : MemOp_rr <opc, opcBits, ImmOp>;
2451 defm _#NAME : MemOp_ri <opc, opcBits, ImmOp>;
2454 // Define MemOp instructions.
2455 let isExtendable = 1, opExtendable = 1, isExtentSigned = 0,
2456 validSubTargets =HasV4SubT in {
2457 let opExtentBits = 6, accessSize = ByteAccess, isCodeGenOnly = 0 in
2458 defm memopb_io : MemOp_base <"memb", 0b00, u6_0Ext>;
2460 let opExtentBits = 7, accessSize = HalfWordAccess, isCodeGenOnly = 0 in
2461 defm memoph_io : MemOp_base <"memh", 0b01, u6_1Ext>;
2463 let opExtentBits = 8, accessSize = WordAccess, isCodeGenOnly = 0 in
2464 defm memopw_io : MemOp_base <"memw", 0b10, u6_2Ext>;
2467 //===----------------------------------------------------------------------===//
2468 // Multiclass to define 'Def Pats' for ALU operations on the memory
2469 // Here value used for the ALU operation is an immediate value.
2470 // mem[bh](Rs+#0) += #U5
2471 // mem[bh](Rs+#u6) += #U5
2472 //===----------------------------------------------------------------------===//
2474 multiclass MemOpi_u5Pats <PatFrag ldOp, PatFrag stOp, PatLeaf ExtPred,
2475 InstHexagon MI, SDNode OpNode> {
2476 let AddedComplexity = 180 in
2477 def : Pat < (stOp (OpNode (ldOp IntRegs:$addr), u5ImmPred:$addend),
2479 (MI IntRegs:$addr, #0, u5ImmPred:$addend )>;
2481 let AddedComplexity = 190 in
2482 def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, ExtPred:$offset)),
2484 (add IntRegs:$base, ExtPred:$offset)),
2485 (MI IntRegs:$base, ExtPred:$offset, u5ImmPred:$addend)>;
2488 multiclass MemOpi_u5ALUOp<PatFrag ldOp, PatFrag stOp, PatLeaf ExtPred,
2489 InstHexagon addMI, InstHexagon subMI> {
2490 defm : MemOpi_u5Pats<ldOp, stOp, ExtPred, addMI, add>;
2491 defm : MemOpi_u5Pats<ldOp, stOp, ExtPred, subMI, sub>;
2494 multiclass MemOpi_u5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2496 defm : MemOpi_u5ALUOp <ldOpHalf, truncstorei16, u6_1ExtPred,
2497 L4_iadd_memoph_io, L4_isub_memoph_io>;
2499 defm : MemOpi_u5ALUOp <ldOpByte, truncstorei8, u6ExtPred,
2500 L4_iadd_memopb_io, L4_isub_memopb_io>;
2503 let Predicates = [HasV4T, UseMEMOP] in {
2504 defm : MemOpi_u5ExtType<zextloadi8, zextloadi16>; // zero extend
2505 defm : MemOpi_u5ExtType<sextloadi8, sextloadi16>; // sign extend
2506 defm : MemOpi_u5ExtType<extloadi8, extloadi16>; // any extend
2509 defm : MemOpi_u5ALUOp <load, store, u6_2ExtPred, L4_iadd_memopw_io,
2513 //===----------------------------------------------------------------------===//
2514 // multiclass to define 'Def Pats' for ALU operations on the memory.
2515 // Here value used for the ALU operation is a negative value.
2516 // mem[bh](Rs+#0) += #m5
2517 // mem[bh](Rs+#u6) += #m5
2518 //===----------------------------------------------------------------------===//
2520 multiclass MemOpi_m5Pats <PatFrag ldOp, PatFrag stOp, PatLeaf extPred,
2521 PatLeaf immPred, ComplexPattern addrPred,
2522 SDNodeXForm xformFunc, InstHexagon MI> {
2523 let AddedComplexity = 190 in
2524 def : Pat <(stOp (add (ldOp IntRegs:$addr), immPred:$subend),
2526 (MI IntRegs:$addr, #0, (xformFunc immPred:$subend) )>;
2528 let AddedComplexity = 195 in
2529 def : Pat<(stOp (add (ldOp (add IntRegs:$base, extPred:$offset)),
2531 (add IntRegs:$base, extPred:$offset)),
2532 (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$subend))>;
2535 multiclass MemOpi_m5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2537 defm : MemOpi_m5Pats <ldOpHalf, truncstorei16, u6_1ExtPred, m5HImmPred,
2538 ADDRriU6_1, MEMOPIMM_HALF, L4_isub_memoph_io>;
2540 defm : MemOpi_m5Pats <ldOpByte, truncstorei8, u6ExtPred, m5BImmPred,
2541 ADDRriU6_0, MEMOPIMM_BYTE, L4_isub_memopb_io>;
2544 let Predicates = [HasV4T, UseMEMOP] in {
2545 defm : MemOpi_m5ExtType<zextloadi8, zextloadi16>; // zero extend
2546 defm : MemOpi_m5ExtType<sextloadi8, sextloadi16>; // sign extend
2547 defm : MemOpi_m5ExtType<extloadi8, extloadi16>; // any extend
2550 defm : MemOpi_m5Pats <load, store, u6_2ExtPred, m5ImmPred,
2551 ADDRriU6_2, MEMOPIMM, L4_isub_memopw_io>;
2554 //===----------------------------------------------------------------------===//
2555 // Multiclass to define 'def Pats' for bit operations on the memory.
2556 // mem[bhw](Rs+#0) = [clrbit|setbit](#U5)
2557 // mem[bhw](Rs+#u6) = [clrbit|setbit](#U5)
2558 //===----------------------------------------------------------------------===//
2560 multiclass MemOpi_bitPats <PatFrag ldOp, PatFrag stOp, PatLeaf immPred,
2561 PatLeaf extPred, ComplexPattern addrPred,
2562 SDNodeXForm xformFunc, InstHexagon MI, SDNode OpNode> {
2564 // mem[bhw](Rs+#u6:[012]) = [clrbit|setbit](#U5)
2565 let AddedComplexity = 250 in
2566 def : Pat<(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)),
2568 (add IntRegs:$base, extPred:$offset)),
2569 (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$bitend))>;
2571 // mem[bhw](Rs+#0) = [clrbit|setbit](#U5)
2572 let AddedComplexity = 225 in
2573 def : Pat <(stOp (OpNode (ldOp (addrPred IntRegs:$addr, extPred:$offset)),
2575 (addrPred (i32 IntRegs:$addr), extPred:$offset)),
2576 (MI IntRegs:$addr, extPred:$offset, (xformFunc immPred:$bitend))>;
2579 multiclass MemOpi_bitExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2581 defm : MemOpi_bitPats<ldOpByte, truncstorei8, Clr3ImmPred, u6ExtPred,
2582 ADDRriU6_0, CLRMEMIMM_BYTE, L4_iand_memopb_io, and>;
2584 defm : MemOpi_bitPats<ldOpByte, truncstorei8, Set3ImmPred, u6ExtPred,
2585 ADDRriU6_0, SETMEMIMM_BYTE, L4_ior_memopb_io, or>;
2586 // Half Word - clrbit
2587 defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Clr4ImmPred, u6_1ExtPred,
2588 ADDRriU6_1, CLRMEMIMM_SHORT, L4_iand_memoph_io, and>;
2589 // Half Word - setbit
2590 defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Set4ImmPred, u6_1ExtPred,
2591 ADDRriU6_1, SETMEMIMM_SHORT, L4_ior_memoph_io, or>;
2594 let Predicates = [HasV4T, UseMEMOP] in {
2595 // mem[bh](Rs+#0) = [clrbit|setbit](#U5)
2596 // mem[bh](Rs+#u6:[01]) = [clrbit|setbit](#U5)
2597 defm : MemOpi_bitExtType<zextloadi8, zextloadi16>; // zero extend
2598 defm : MemOpi_bitExtType<sextloadi8, sextloadi16>; // sign extend
2599 defm : MemOpi_bitExtType<extloadi8, extloadi16>; // any extend
2601 // memw(Rs+#0) = [clrbit|setbit](#U5)
2602 // memw(Rs+#u6:2) = [clrbit|setbit](#U5)
2603 defm : MemOpi_bitPats<load, store, Clr5ImmPred, u6_2ExtPred, ADDRriU6_2,
2604 CLRMEMIMM, L4_iand_memopw_io, and>;
2605 defm : MemOpi_bitPats<load, store, Set5ImmPred, u6_2ExtPred, ADDRriU6_2,
2606 SETMEMIMM, L4_ior_memopw_io, or>;
2609 //===----------------------------------------------------------------------===//
2610 // Multiclass to define 'def Pats' for ALU operations on the memory
2611 // where addend is a register.
2612 // mem[bhw](Rs+#0) [+-&|]= Rt
2613 // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt
2614 //===----------------------------------------------------------------------===//
2616 multiclass MemOpr_Pats <PatFrag ldOp, PatFrag stOp, ComplexPattern addrPred,
2617 PatLeaf extPred, InstHexagon MI, SDNode OpNode> {
2618 let AddedComplexity = 141 in
2619 // mem[bhw](Rs+#0) [+-&|]= Rt
2620 def : Pat <(stOp (OpNode (ldOp (addrPred IntRegs:$addr, extPred:$offset)),
2621 (i32 IntRegs:$addend)),
2622 (addrPred (i32 IntRegs:$addr), extPred:$offset)),
2623 (MI IntRegs:$addr, extPred:$offset, (i32 IntRegs:$addend) )>;
2625 // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt
2626 let AddedComplexity = 150 in
2627 def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)),
2628 (i32 IntRegs:$orend)),
2629 (add IntRegs:$base, extPred:$offset)),
2630 (MI IntRegs:$base, extPred:$offset, (i32 IntRegs:$orend) )>;
2633 multiclass MemOPr_ALUOp<PatFrag ldOp, PatFrag stOp,
2634 ComplexPattern addrPred, PatLeaf extPred,
2635 InstHexagon addMI, InstHexagon subMI,
2636 InstHexagon andMI, InstHexagon orMI > {
2638 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, addMI, add>;
2639 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, subMI, sub>;
2640 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, andMI, and>;
2641 defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, orMI, or>;
2644 multiclass MemOPr_ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
2646 defm : MemOPr_ALUOp <ldOpHalf, truncstorei16, ADDRriU6_1, u6_1ExtPred,
2647 L4_add_memoph_io, L4_sub_memoph_io,
2648 L4_and_memoph_io, L4_or_memoph_io>;
2650 defm : MemOPr_ALUOp <ldOpByte, truncstorei8, ADDRriU6_0, u6ExtPred,
2651 L4_add_memopb_io, L4_sub_memopb_io,
2652 L4_and_memopb_io, L4_or_memopb_io>;
2655 // Define 'def Pats' for MemOps with register addend.
2656 let Predicates = [HasV4T, UseMEMOP] in {
2658 defm : MemOPr_ExtType<zextloadi8, zextloadi16>; // zero extend
2659 defm : MemOPr_ExtType<sextloadi8, sextloadi16>; // sign extend
2660 defm : MemOPr_ExtType<extloadi8, extloadi16>; // any extend
2662 defm : MemOPr_ALUOp <load, store, ADDRriU6_2, u6_2ExtPred, L4_add_memopw_io,
2663 L4_sub_memopw_io, L4_and_memopw_io, L4_or_memopw_io >;
2666 //===----------------------------------------------------------------------===//
2668 //===----------------------------------------------------------------------===//
2670 // Hexagon V4 only supports these flavors of byte/half compare instructions:
2671 // EQ/GT/GTU. Other flavors like GE/GEU/LT/LTU/LE/LEU are not supported by
2672 // hardware. However, compiler can still implement these patterns through
2673 // appropriate patterns combinations based on current implemented patterns.
2674 // The implemented patterns are: EQ/GT/GTU.
2675 // Missing patterns are: GE/GEU/LT/LTU/LE/LEU.
2677 // Following instruction is not being extended as it results into the
2678 // incorrect code for negative numbers.
2679 // Pd=cmpb.eq(Rs,#u8)
2681 let isCompare = 1, isExtendable = 1, opExtendable = 2, hasSideEffects = 0,
2682 validSubTargets = HasV4SubT in
2683 class CMP_NOT_REG_IMM<string OpName, bits<2> op, Operand ImmOp,
2685 : ALU32Inst <(outs PredRegs:$dst), (ins IntRegs:$src1, ImmOp:$src2),
2686 "$dst = !cmp."#OpName#"($src1, #$src2)",
2688 "", ALU32_2op_tc_2early_SLOT0123> {
2693 let IClass = 0b0111;
2694 let Inst{27-24} = 0b0101;
2695 let Inst{23-22} = op;
2696 let Inst{20-16} = src1;
2697 let Inst{21} = !if (!eq(OpName, "gtu"), 0b0, src2{9});
2698 let Inst{13-5} = src2{8-0};
2699 let Inst{4-2} = 0b100;
2700 let Inst{1-0} = dst;
2703 let opExtentBits = 10, isExtentSigned = 1 in {
2704 def C4_cmpneqi : CMP_NOT_REG_IMM <"eq", 0b00, s10Ext, [(set (i1 PredRegs:$dst),
2705 (setne (i32 IntRegs:$src1), s10ExtPred:$src2))]>;
2707 def C4_cmpltei : CMP_NOT_REG_IMM <"gt", 0b01, s10Ext, [(set (i1 PredRegs:$dst),
2708 (not (setgt (i32 IntRegs:$src1), s10ExtPred:$src2)))]>;
2711 let opExtentBits = 9 in
2712 def C4_cmplteui : CMP_NOT_REG_IMM <"gtu", 0b10, u9Ext, [(set (i1 PredRegs:$dst),
2713 (not (setugt (i32 IntRegs:$src1), u9ExtPred:$src2)))]>;
2715 def : Pat <(brcond (i1 (setne (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2)),
2717 (J2_jumpf (A4_cmpbeqi (i32 IntRegs:$src1), u8ImmPred:$src2),
2721 // Pd=cmpb.eq(Rs,Rt)
2722 let isCompare = 1, validSubTargets = HasV4SubT in
2723 def CMPbEQrr_ubub_V4 : MInst<(outs PredRegs:$dst),
2724 (ins IntRegs:$src1, IntRegs:$src2),
2725 "$dst = cmpb.eq($src1, $src2)",
2726 [(set (i1 PredRegs:$dst),
2727 (seteq (and (xor (i32 IntRegs:$src1),
2728 (i32 IntRegs:$src2)), 255), 0))]>,
2731 // Pd=cmpb.eq(Rs,Rt)
2732 let isCompare = 1, validSubTargets = HasV4SubT in
2733 def CMPbEQrr_sbsb_V4 : MInst<(outs PredRegs:$dst),
2734 (ins IntRegs:$src1, IntRegs:$src2),
2735 "$dst = cmpb.eq($src1, $src2)",
2736 [(set (i1 PredRegs:$dst),
2737 (seteq (shl (i32 IntRegs:$src1), (i32 24)),
2738 (shl (i32 IntRegs:$src2), (i32 24))))]>,
2741 // Pd=cmpb.gt(Rs,Rt)
2742 let isCompare = 1, validSubTargets = HasV4SubT in
2743 def CMPbGTrr_V4 : MInst<(outs PredRegs:$dst),
2744 (ins IntRegs:$src1, IntRegs:$src2),
2745 "$dst = cmpb.gt($src1, $src2)",
2746 [(set (i1 PredRegs:$dst),
2747 (setgt (shl (i32 IntRegs:$src1), (i32 24)),
2748 (shl (i32 IntRegs:$src2), (i32 24))))]>,
2751 // Pd=cmpb.gtu(Rs,#u7)
2752 let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7,
2753 isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU", InputType = "imm" in
2754 def CMPbGTUri_V4 : MInst<(outs PredRegs:$dst),
2755 (ins IntRegs:$src1, u7Ext:$src2),
2756 "$dst = cmpb.gtu($src1, #$src2)",
2757 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255),
2758 u7ExtPred:$src2))]>,
2759 Requires<[HasV4T]>, ImmRegRel;
2761 // SDNode for converting immediate C to C-1.
2762 def DEC_CONST_BYTE : SDNodeXForm<imm, [{
2763 // Return the byte immediate const-1 as an SDNode.
2764 int32_t imm = N->getSExtValue();
2765 return XformU7ToU7M1Imm(imm);
2769 // zext( seteq ( and(Rs, 255), u8))
2771 // Pd=cmpb.eq(Rs, #u8)
2772 // if (Pd.new) Rd=#1
2773 // if (!Pd.new) Rd=#0
2774 def : Pat <(i32 (zext (i1 (seteq (i32 (and (i32 IntRegs:$Rs), 255)),
2776 (i32 (TFR_condset_ii (i1 (A4_cmpbeqi (i32 IntRegs:$Rs),
2782 // zext( setne ( and(Rs, 255), u8))
2784 // Pd=cmpb.eq(Rs, #u8)
2785 // if (Pd.new) Rd=#0
2786 // if (!Pd.new) Rd=#1
2787 def : Pat <(i32 (zext (i1 (setne (i32 (and (i32 IntRegs:$Rs), 255)),
2789 (i32 (TFR_condset_ii (i1 (A4_cmpbeqi (i32 IntRegs:$Rs),
2795 // zext( seteq (Rs, and(Rt, 255)))
2797 // Pd=cmpb.eq(Rs, Rt)
2798 // if (Pd.new) Rd=#1
2799 // if (!Pd.new) Rd=#0
2800 def : Pat <(i32 (zext (i1 (seteq (i32 IntRegs:$Rt),
2801 (i32 (and (i32 IntRegs:$Rs), 255)))))),
2802 (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs),
2803 (i32 IntRegs:$Rt))),
2808 // zext( setne (Rs, and(Rt, 255)))
2810 // Pd=cmpb.eq(Rs, Rt)
2811 // if (Pd.new) Rd=#0
2812 // if (!Pd.new) Rd=#1
2813 def : Pat <(i32 (zext (i1 (setne (i32 IntRegs:$Rt),
2814 (i32 (and (i32 IntRegs:$Rs), 255)))))),
2815 (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs),
2816 (i32 IntRegs:$Rt))),
2821 // zext( setugt ( and(Rs, 255), u8))
2823 // Pd=cmpb.gtu(Rs, #u8)
2824 // if (Pd.new) Rd=#1
2825 // if (!Pd.new) Rd=#0
2826 def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 255)),
2828 (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs),
2834 // zext( setugt ( and(Rs, 254), u8))
2836 // Pd=cmpb.gtu(Rs, #u8)
2837 // if (Pd.new) Rd=#1
2838 // if (!Pd.new) Rd=#0
2839 def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 254)),
2841 (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs),
2847 // zext( setult ( Rs, Rt))
2849 // Pd=cmp.ltu(Rs, Rt)
2850 // if (Pd.new) Rd=#1
2851 // if (!Pd.new) Rd=#0
2852 // cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs)
2853 def : Pat <(i32 (zext (i1 (setult (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2854 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rt),
2855 (i32 IntRegs:$Rs))),
2860 // zext( setlt ( Rs, Rt))
2862 // Pd=cmp.lt(Rs, Rt)
2863 // if (Pd.new) Rd=#1
2864 // if (!Pd.new) Rd=#0
2865 // cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs)
2866 def : Pat <(i32 (zext (i1 (setlt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2867 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rt),
2868 (i32 IntRegs:$Rs))),
2873 // zext( setugt ( Rs, Rt))
2875 // Pd=cmp.gtu(Rs, Rt)
2876 // if (Pd.new) Rd=#1
2877 // if (!Pd.new) Rd=#0
2878 def : Pat <(i32 (zext (i1 (setugt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2879 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rs),
2880 (i32 IntRegs:$Rt))),
2884 // This pattern interefers with coremark performance, not implementing at this
2887 // zext( setgt ( Rs, Rt))
2889 // Pd=cmp.gt(Rs, Rt)
2890 // if (Pd.new) Rd=#1
2891 // if (!Pd.new) Rd=#0
2894 // zext( setuge ( Rs, Rt))
2896 // Pd=cmp.ltu(Rs, Rt)
2897 // if (Pd.new) Rd=#0
2898 // if (!Pd.new) Rd=#1
2899 // cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs)
2900 def : Pat <(i32 (zext (i1 (setuge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2901 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rt),
2902 (i32 IntRegs:$Rs))),
2907 // zext( setge ( Rs, Rt))
2909 // Pd=cmp.lt(Rs, Rt)
2910 // if (Pd.new) Rd=#0
2911 // if (!Pd.new) Rd=#1
2912 // cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs)
2913 def : Pat <(i32 (zext (i1 (setge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2914 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rt),
2915 (i32 IntRegs:$Rs))),
2920 // zext( setule ( Rs, Rt))
2922 // Pd=cmp.gtu(Rs, Rt)
2923 // if (Pd.new) Rd=#0
2924 // if (!Pd.new) Rd=#1
2925 def : Pat <(i32 (zext (i1 (setule (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2926 (i32 (TFR_condset_ii (i1 (C2_cmpgtu (i32 IntRegs:$Rs),
2927 (i32 IntRegs:$Rt))),
2932 // zext( setle ( Rs, Rt))
2934 // Pd=cmp.gt(Rs, Rt)
2935 // if (Pd.new) Rd=#0
2936 // if (!Pd.new) Rd=#1
2937 def : Pat <(i32 (zext (i1 (setle (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
2938 (i32 (TFR_condset_ii (i1 (C2_cmpgt (i32 IntRegs:$Rs),
2939 (i32 IntRegs:$Rt))),
2944 // zext( setult ( and(Rs, 255), u8))
2945 // Use the isdigit transformation below
2947 // Generate code of the form 'mux_ii(cmpbgtu(Rdd, C-1),0,1)'
2948 // for C code of the form r = ((c>='0') & (c<='9')) ? 1 : 0;.
2949 // The isdigit transformation relies on two 'clever' aspects:
2950 // 1) The data type is unsigned which allows us to eliminate a zero test after
2951 // biasing the expression by 48. We are depending on the representation of
2952 // the unsigned types, and semantics.
2953 // 2) The front end has converted <= 9 into < 10 on entry to LLVM
2956 // retval = ((c>='0') & (c<='9')) ? 1 : 0;
2957 // The code is transformed upstream of llvm into
2958 // retval = (c-48) < 10 ? 1 : 0;
2959 let AddedComplexity = 139 in
2960 def : Pat <(i32 (zext (i1 (setult (i32 (and (i32 IntRegs:$src1), 255)),
2961 u7StrictPosImmPred:$src2)))),
2962 (i32 (C2_muxii (i1 (CMPbGTUri_V4 (i32 IntRegs:$src1),
2963 (DEC_CONST_BYTE u7StrictPosImmPred:$src2))),
2967 // Pd=cmpb.gtu(Rs,Rt)
2968 let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU",
2969 InputType = "reg" in
2970 def CMPbGTUrr_V4 : MInst<(outs PredRegs:$dst),
2971 (ins IntRegs:$src1, IntRegs:$src2),
2972 "$dst = cmpb.gtu($src1, $src2)",
2973 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255),
2974 (and (i32 IntRegs:$src2), 255)))]>,
2975 Requires<[HasV4T]>, ImmRegRel;
2977 // Following instruction is not being extended as it results into the incorrect
2978 // code for negative numbers.
2980 // Signed half compare(.eq) ri.
2981 // Pd=cmph.eq(Rs,#s8)
2982 let isCompare = 1, validSubTargets = HasV4SubT in
2983 def CMPhEQri_V4 : MInst<(outs PredRegs:$dst),
2984 (ins IntRegs:$src1, s8Imm:$src2),
2985 "$dst = cmph.eq($src1, #$src2)",
2986 [(set (i1 PredRegs:$dst), (seteq (and (i32 IntRegs:$src1), 65535),
2987 s8ImmPred:$src2))]>,
2990 // Signed half compare(.eq) rr.
2991 // Case 1: xor + and, then compare:
2993 // r0=and(r0,#0xffff)
2995 // Pd=cmph.eq(Rs,Rt)
2996 let isCompare = 1, validSubTargets = HasV4SubT in
2997 def CMPhEQrr_xor_V4 : MInst<(outs PredRegs:$dst),
2998 (ins IntRegs:$src1, IntRegs:$src2),
2999 "$dst = cmph.eq($src1, $src2)",
3000 [(set (i1 PredRegs:$dst), (seteq (and (xor (i32 IntRegs:$src1),
3001 (i32 IntRegs:$src2)),
3005 // Signed half compare(.eq) rr.
3006 // Case 2: shift left 16 bits then compare:
3010 // Pd=cmph.eq(Rs,Rt)
3011 let isCompare = 1, validSubTargets = HasV4SubT in
3012 def CMPhEQrr_shl_V4 : MInst<(outs PredRegs:$dst),
3013 (ins IntRegs:$src1, IntRegs:$src2),
3014 "$dst = cmph.eq($src1, $src2)",
3015 [(set (i1 PredRegs:$dst),
3016 (seteq (shl (i32 IntRegs:$src1), (i32 16)),
3017 (shl (i32 IntRegs:$src2), (i32 16))))]>,
3020 /* Incorrect Pattern -- immediate should be right shifted before being
3021 used in the cmph.gt instruction.
3022 // Signed half compare(.gt) ri.
3023 // Pd=cmph.gt(Rs,#s8)
3025 let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 8,
3026 isCompare = 1, validSubTargets = HasV4SubT in
3027 def CMPhGTri_V4 : MInst<(outs PredRegs:$dst),
3028 (ins IntRegs:$src1, s8Ext:$src2),
3029 "$dst = cmph.gt($src1, #$src2)",
3030 [(set (i1 PredRegs:$dst),
3031 (setgt (shl (i32 IntRegs:$src1), (i32 16)),
3032 s8ExtPred:$src2))]>,
3036 // Signed half compare(.gt) rr.
3037 // Pd=cmph.gt(Rs,Rt)
3038 let isCompare = 1, validSubTargets = HasV4SubT in
3039 def CMPhGTrr_shl_V4 : MInst<(outs PredRegs:$dst),
3040 (ins IntRegs:$src1, IntRegs:$src2),
3041 "$dst = cmph.gt($src1, $src2)",
3042 [(set (i1 PredRegs:$dst),
3043 (setgt (shl (i32 IntRegs:$src1), (i32 16)),
3044 (shl (i32 IntRegs:$src2), (i32 16))))]>,
3047 // Unsigned half compare rr (.gtu).
3048 // Pd=cmph.gtu(Rs,Rt)
3049 let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU",
3050 InputType = "reg" in
3051 def CMPhGTUrr_V4 : MInst<(outs PredRegs:$dst),
3052 (ins IntRegs:$src1, IntRegs:$src2),
3053 "$dst = cmph.gtu($src1, $src2)",
3054 [(set (i1 PredRegs:$dst),
3055 (setugt (and (i32 IntRegs:$src1), 65535),
3056 (and (i32 IntRegs:$src2), 65535)))]>,
3057 Requires<[HasV4T]>, ImmRegRel;
3059 // Unsigned half compare ri (.gtu).
3060 // Pd=cmph.gtu(Rs,#u7)
3061 let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7,
3062 isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU",
3063 InputType = "imm" in
3064 def CMPhGTUri_V4 : MInst<(outs PredRegs:$dst),
3065 (ins IntRegs:$src1, u7Ext:$src2),
3066 "$dst = cmph.gtu($src1, #$src2)",
3067 [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 65535),
3068 u7ExtPred:$src2))]>,
3069 Requires<[HasV4T]>, ImmRegRel;
3071 let validSubTargets = HasV4SubT in
3072 def NTSTBIT_rr : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
3073 "$dst = !tstbit($src1, $src2)",
3074 [(set (i1 PredRegs:$dst),
3075 (seteq (and (shl 1, (i32 IntRegs:$src2)), (i32 IntRegs:$src1)), 0))]>,
3078 let validSubTargets = HasV4SubT in
3079 def NTSTBIT_ri : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
3080 "$dst = !tstbit($src1, $src2)",
3081 [(set (i1 PredRegs:$dst),
3082 (seteq (and (shl 1, u5ImmPred:$src2), (i32 IntRegs:$src1)), 0))]>,
3085 //===----------------------------------------------------------------------===//
3087 //===----------------------------------------------------------------------===//
3089 //===----------------------------------------------------------------------===//
3090 // Multiclass for DeallocReturn
3091 //===----------------------------------------------------------------------===//
3092 class L4_RETURN<string mnemonic, bit isNot, bit isPredNew, bit isTak>
3093 : LD0Inst<(outs), (ins PredRegs:$src),
3094 !if(isNot, "if (!$src", "if ($src")#
3095 !if(isPredNew, ".new) ", ") ")#mnemonic#
3096 !if(isPredNew, #!if(isTak,":t", ":nt"),""),
3097 [], "", LD_tc_3or4stall_SLOT0> {
3100 let BaseOpcode = "L4_RETURN";
3101 let isPredicatedFalse = isNot;
3102 let isPredicatedNew = isPredNew;
3103 let isTaken = isTak;
3104 let IClass = 0b1001;
3106 let Inst{27-16} = 0b011000011110;
3108 let Inst{13} = isNot;
3109 let Inst{12} = isTak;
3110 let Inst{11} = isPredNew;
3112 let Inst{9-8} = src;
3113 let Inst{4-0} = 0b11110;
3116 // Produce all predicated forms, p, !p, p.new, !p.new, :t, :nt
3117 multiclass L4_RETURN_PRED<string mnemonic, bit PredNot> {
3118 let isPredicated = 1 in {
3119 def _#NAME# : L4_RETURN <mnemonic, PredNot, 0, 1>;
3120 def _#NAME#new_pnt : L4_RETURN <mnemonic, PredNot, 1, 0>;
3121 def _#NAME#new_pt : L4_RETURN <mnemonic, PredNot, 1, 1>;
3125 multiclass LD_MISC_L4_RETURN<string mnemonic> {
3126 let isBarrier = 1, isPredicable = 1 in
3127 def NAME : LD0Inst <(outs), (ins), mnemonic, [], "",
3128 LD_tc_3or4stall_SLOT0> {
3129 let BaseOpcode = "L4_RETURN";
3130 let IClass = 0b1001;
3131 let Inst{27-16} = 0b011000011110;
3132 let Inst{13-10} = 0b0000;
3133 let Inst{4-0} = 0b11110;
3135 defm t : L4_RETURN_PRED<mnemonic, 0 >;
3136 defm f : L4_RETURN_PRED<mnemonic, 1 >;
3139 let isReturn = 1, isTerminator = 1,
3140 Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
3141 validSubTargets = HasV4SubT, isCodeGenOnly = 0 in
3142 defm L4_return: LD_MISC_L4_RETURN <"dealloc_return">, PredNewRel;
3144 // Restore registers and dealloc return function call.
3145 let isCall = 1, isBarrier = 1, isReturn = 1, isTerminator = 1,
3146 Defs = [R29, R30, R31, PC] in {
3147 let validSubTargets = HasV4SubT in
3148 def RESTORE_DEALLOC_RET_JMP_V4 : JInst<(outs),
3149 (ins calltarget:$dst),
3155 // Restore registers and dealloc frame before a tail call.
3156 let isCall = 1, isBarrier = 1,
3157 Defs = [R29, R30, R31, PC] in {
3158 let validSubTargets = HasV4SubT in
3159 def RESTORE_DEALLOC_BEFORE_TAILCALL_V4 : JInst<(outs),
3160 (ins calltarget:$dst),
3166 // Save registers function call.
3167 let isCall = 1, isBarrier = 1,
3168 Uses = [R29, R31] in {
3169 def SAVE_REGISTERS_CALL_V4 : JInst<(outs),
3170 (ins calltarget:$dst),
3171 "call $dst // Save_calle_saved_registers",
3176 //===----------------------------------------------------------------------===//
3177 // Template class for non predicated store instructions with
3178 // GP-Relative or absolute addressing.
3179 //===----------------------------------------------------------------------===//
3180 let hasSideEffects = 0, isPredicable = 1, isNVStorable = 1 in
3181 class T_StoreAbsGP <string mnemonic, RegisterClass RC, Operand ImmOp,
3182 bits<2>MajOp, Operand AddrOp, bit isAbs, bit isHalf>
3183 : STInst<(outs), (ins AddrOp:$addr, RC:$src),
3184 mnemonic # !if(isAbs, "(##", "(#")#"$addr) = $src"#!if(isHalf, ".h",""),
3185 [], "", V2LDST_tc_st_SLOT01> {
3188 bits<16> offsetBits;
3190 string ImmOpStr = !cast<string>(ImmOp);
3191 let offsetBits = !if (!eq(ImmOpStr, "u16_3Imm"), addr{18-3},
3192 !if (!eq(ImmOpStr, "u16_2Imm"), addr{17-2},
3193 !if (!eq(ImmOpStr, "u16_1Imm"), addr{16-1},
3194 /* u16_0Imm */ addr{15-0})));
3195 let IClass = 0b0100;
3197 let Inst{26-25} = offsetBits{15-14};
3199 let Inst{23-22} = MajOp;
3200 let Inst{21} = isHalf;
3201 let Inst{20-16} = offsetBits{13-9};
3202 let Inst{13} = offsetBits{8};
3203 let Inst{12-8} = src;
3204 let Inst{7-0} = offsetBits{7-0};
3207 //===----------------------------------------------------------------------===//
3208 // Template class for predicated store instructions with
3209 // GP-Relative or absolute addressing.
3210 //===----------------------------------------------------------------------===//
3211 let hasSideEffects = 0, isPredicated = 1, isNVStorable = 1, opExtentBits = 6,
3213 class T_StoreAbs_Pred <string mnemonic, RegisterClass RC, bits<2> MajOp,
3214 bit isHalf, bit isNot, bit isNew>
3215 : STInst<(outs), (ins PredRegs:$src1, u6Ext:$absaddr, RC: $src2),
3216 !if(isNot, "if (!$src1", "if ($src1")#!if(isNew, ".new) ",
3217 ") ")#mnemonic#"(#$absaddr) = $src2"#!if(isHalf, ".h",""),
3218 [], "", ST_tc_st_SLOT01>, AddrModeRel {
3223 let isPredicatedNew = isNew;
3224 let isPredicatedFalse = isNot;
3226 let IClass = 0b1010;
3228 let Inst{27-24} = 0b1111;
3229 let Inst{23-22} = MajOp;
3230 let Inst{21} = isHalf;
3231 let Inst{17-16} = absaddr{5-4};
3232 let Inst{13} = isNew;
3233 let Inst{12-8} = src2;
3235 let Inst{6-3} = absaddr{3-0};
3236 let Inst{2} = isNot;
3237 let Inst{1-0} = src1;
3240 //===----------------------------------------------------------------------===//
3241 // Template class for predicated store instructions with absolute addressing.
3242 //===----------------------------------------------------------------------===//
3243 class T_StoreAbs <string mnemonic, RegisterClass RC, Operand ImmOp,
3244 bits<2> MajOp, bit isHalf>
3245 : T_StoreAbsGP <mnemonic, RC, ImmOp, MajOp, u0AlwaysExt, 1, isHalf>,
3247 string ImmOpStr = !cast<string>(ImmOp);
3248 let opExtentBits = !if (!eq(ImmOpStr, "u16_3Imm"), 19,
3249 !if (!eq(ImmOpStr, "u16_2Imm"), 18,
3250 !if (!eq(ImmOpStr, "u16_1Imm"), 17,
3251 /* u16_0Imm */ 16)));
3253 let opExtentAlign = !if (!eq(ImmOpStr, "u16_3Imm"), 3,
3254 !if (!eq(ImmOpStr, "u16_2Imm"), 2,
3255 !if (!eq(ImmOpStr, "u16_1Imm"), 1,
3256 /* u16_0Imm */ 0)));
3259 //===----------------------------------------------------------------------===//
3260 // Multiclass for store instructions with absolute addressing.
3261 //===----------------------------------------------------------------------===//
3262 let validSubTargets = HasV4SubT, addrMode = Absolute, isExtended = 1 in
3263 multiclass ST_Abs<string mnemonic, string CextOp, RegisterClass RC,
3264 Operand ImmOp, bits<2> MajOp, bit isHalf = 0> {
3265 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
3266 let opExtendable = 0, isPredicable = 1 in
3267 def S2_#NAME#abs : T_StoreAbs <mnemonic, RC, ImmOp, MajOp, isHalf>;
3270 def S4_p#NAME#t_abs : T_StoreAbs_Pred<mnemonic, RC, MajOp, isHalf, 0, 0>;
3271 def S4_p#NAME#f_abs : T_StoreAbs_Pred<mnemonic, RC, MajOp, isHalf, 1, 0>;
3274 def S4_p#NAME#tnew_abs : T_StoreAbs_Pred<mnemonic, RC, MajOp, isHalf, 0, 1>;
3275 def S4_p#NAME#fnew_abs : T_StoreAbs_Pred<mnemonic, RC, MajOp, isHalf, 1, 1>;
3279 //===----------------------------------------------------------------------===//
3280 // Template class for non predicated new-value store instructions with
3281 // GP-Relative or absolute addressing.
3282 //===----------------------------------------------------------------------===//
3283 let hasSideEffects = 0, isPredicable = 1, mayStore = 1, isNVStore = 1,
3284 isNewValue = 1, opNewValue = 1 in
3285 class T_StoreAbsGP_NV <string mnemonic, Operand ImmOp, bits<2>MajOp, bit isAbs>
3286 : NVInst_V4<(outs), (ins u0AlwaysExt:$addr, IntRegs:$src),
3287 mnemonic # !if(isAbs, "(##", "(#")#"$addr) = $src.new",
3288 [], "", V2LDST_tc_st_SLOT0> {
3291 bits<16> offsetBits;
3293 string ImmOpStr = !cast<string>(ImmOp);
3294 let offsetBits = !if (!eq(ImmOpStr, "u16_3Imm"), addr{18-3},
3295 !if (!eq(ImmOpStr, "u16_2Imm"), addr{17-2},
3296 !if (!eq(ImmOpStr, "u16_1Imm"), addr{16-1},
3297 /* u16_0Imm */ addr{15-0})));
3298 let IClass = 0b0100;
3301 let Inst{26-25} = offsetBits{15-14};
3302 let Inst{24-21} = 0b0101;
3303 let Inst{20-16} = offsetBits{13-9};
3304 let Inst{13} = offsetBits{8};
3305 let Inst{12-11} = MajOp;
3306 let Inst{10-8} = src;
3307 let Inst{7-0} = offsetBits{7-0};
3310 //===----------------------------------------------------------------------===//
3311 // Template class for predicated new-value store instructions with
3312 // absolute addressing.
3313 //===----------------------------------------------------------------------===//
3314 let hasSideEffects = 0, isPredicated = 1, mayStore = 1, isNVStore = 1,
3315 isNewValue = 1, opNewValue = 2, opExtentBits = 6, opExtendable = 1 in
3316 class T_StoreAbs_NV_Pred <string mnemonic, bits<2> MajOp, bit isNot, bit isNew>
3317 : NVInst_V4<(outs), (ins PredRegs:$src1, u6Ext:$absaddr, IntRegs:$src2),
3318 !if(isNot, "if (!$src1", "if ($src1")#!if(isNew, ".new) ",
3319 ") ")#mnemonic#"(#$absaddr) = $src2.new",
3320 [], "", ST_tc_st_SLOT0>, AddrModeRel {
3325 let isPredicatedNew = isNew;
3326 let isPredicatedFalse = isNot;
3328 let IClass = 0b1010;
3330 let Inst{27-24} = 0b1111;
3331 let Inst{23-21} = 0b101;
3332 let Inst{17-16} = absaddr{5-4};
3333 let Inst{13} = isNew;
3334 let Inst{12-11} = MajOp;
3335 let Inst{10-8} = src2;
3337 let Inst{6-3} = absaddr{3-0};
3338 let Inst{2} = isNot;
3339 let Inst{1-0} = src1;
3342 //===----------------------------------------------------------------------===//
3343 // Template class for non-predicated new-value store instructions with
3344 // absolute addressing.
3345 //===----------------------------------------------------------------------===//
3346 class T_StoreAbs_NV <string mnemonic, Operand ImmOp, bits<2> MajOp>
3347 : T_StoreAbsGP_NV <mnemonic, ImmOp, MajOp, 1>, AddrModeRel {
3349 string ImmOpStr = !cast<string>(ImmOp);
3350 let opExtentBits = !if (!eq(ImmOpStr, "u16_3Imm"), 19,
3351 !if (!eq(ImmOpStr, "u16_2Imm"), 18,
3352 !if (!eq(ImmOpStr, "u16_1Imm"), 17,
3353 /* u16_0Imm */ 16)));
3355 let opExtentAlign = !if (!eq(ImmOpStr, "u16_3Imm"), 3,
3356 !if (!eq(ImmOpStr, "u16_2Imm"), 2,
3357 !if (!eq(ImmOpStr, "u16_1Imm"), 1,
3358 /* u16_0Imm */ 0)));
3361 //===----------------------------------------------------------------------===//
3362 // Multiclass for new-value store instructions with absolute addressing.
3363 //===----------------------------------------------------------------------===//
3364 let validSubTargets = HasV4SubT, addrMode = Absolute, isExtended = 1 in
3365 multiclass ST_Abs_NV <string mnemonic, string CextOp, Operand ImmOp,
3367 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
3368 let opExtendable = 0, isPredicable = 1 in
3369 def S2_#NAME#newabs : T_StoreAbs_NV <mnemonic, ImmOp, MajOp>;
3372 def S4_p#NAME#newt_abs : T_StoreAbs_NV_Pred <mnemonic, MajOp, 0, 0>;
3373 def S4_p#NAME#newf_abs : T_StoreAbs_NV_Pred <mnemonic, MajOp, 1, 0>;
3376 def S4_p#NAME#newtnew_abs : T_StoreAbs_NV_Pred <mnemonic, MajOp, 0, 1>;
3377 def S4_p#NAME#newfnew_abs : T_StoreAbs_NV_Pred <mnemonic, MajOp, 1, 1>;
3381 //===----------------------------------------------------------------------===//
3382 // Stores with absolute addressing
3383 //===----------------------------------------------------------------------===//
3384 let accessSize = ByteAccess, isCodeGenOnly = 0 in
3385 defm storerb : ST_Abs <"memb", "STrib", IntRegs, u16_0Imm, 0b00>,
3386 ST_Abs_NV <"memb", "STrib", u16_0Imm, 0b00>;
3388 let accessSize = HalfWordAccess, isCodeGenOnly = 0 in
3389 defm storerh : ST_Abs <"memh", "STrih", IntRegs, u16_1Imm, 0b01>,
3390 ST_Abs_NV <"memh", "STrih", u16_1Imm, 0b01>;
3392 let accessSize = WordAccess, isCodeGenOnly = 0 in
3393 defm storeri : ST_Abs <"memw", "STriw", IntRegs, u16_2Imm, 0b10>,
3394 ST_Abs_NV <"memw", "STriw", u16_2Imm, 0b10>;
3396 let isNVStorable = 0, accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
3397 defm storerd : ST_Abs <"memd", "STrid", DoubleRegs, u16_3Imm, 0b11>;
3399 let isNVStorable = 0, accessSize = HalfWordAccess, isCodeGenOnly = 0 in
3400 defm storerf : ST_Abs <"memh", "STrif", IntRegs, u16_1Imm, 0b01, 1>;
3402 //===----------------------------------------------------------------------===//
3403 // GP-relative stores.
3404 // mem[bhwd](#global)=Rt
3405 // Once predicated, these instructions map to absolute addressing mode.
3406 // if ([!]Pv[.new]) mem[bhwd](##global)=Rt
3407 //===----------------------------------------------------------------------===//
3409 let validSubTargets = HasV4SubT in
3410 class T_StoreGP <string mnemonic, string BaseOp, RegisterClass RC,
3411 Operand ImmOp, bits<2> MajOp, bit isHalf = 0>
3412 : T_StoreAbsGP <mnemonic, RC, ImmOp, MajOp, globaladdress, 0, isHalf> {
3413 // Set BaseOpcode same as absolute addressing instructions so that
3414 // non-predicated GP-Rel instructions can have relate with predicated
3415 // Absolute instruction.
3416 let BaseOpcode = BaseOp#_abs;
3419 let validSubTargets = HasV4SubT in
3420 multiclass ST_GP <string mnemonic, string BaseOp, Operand ImmOp,
3421 bits<2> MajOp, bit isHalf = 0> {
3422 // Set BaseOpcode same as absolute addressing instructions so that
3423 // non-predicated GP-Rel instructions can have relate with predicated
3424 // Absolute instruction.
3425 let BaseOpcode = BaseOp#_abs in {
3426 def NAME#gp : T_StoreAbsGP <mnemonic, IntRegs, ImmOp, MajOp,
3427 globaladdress, 0, isHalf>;
3429 def NAME#newgp : T_StoreAbsGP_NV <mnemonic, ImmOp, MajOp, 0> ;
3433 let accessSize = ByteAccess in
3434 defm S2_storerb : ST_GP<"memb", "STrib", u16_0Imm, 0b00>, NewValueRel;
3436 let accessSize = HalfWordAccess in
3437 defm S2_storerh : ST_GP<"memh", "STrih", u16_1Imm, 0b01>, NewValueRel;
3439 let accessSize = WordAccess in
3440 defm S2_storeri : ST_GP<"memw", "STriw", u16_2Imm, 0b10>, NewValueRel;
3442 let isNVStorable = 0, accessSize = DoubleWordAccess in
3443 def S2_storerdgp : T_StoreGP <"memd", "STrid", DoubleRegs,
3444 u16_3Imm, 0b11>, PredNewRel;
3446 let isNVStorable = 0, accessSize = HalfWordAccess in
3447 def S2_storerfgp : T_StoreGP <"memh", "STrif", IntRegs,
3448 u16_1Imm, 0b01, 1>, PredNewRel;
3450 let Predicates = [HasV4T], AddedComplexity = 30 in {
3451 def : Pat<(truncstorei8 (i32 IntRegs:$src1),
3452 (HexagonCONST32 tglobaladdr:$absaddr)),
3453 (S2_storerbabs tglobaladdr: $absaddr, IntRegs: $src1)>;
3455 def : Pat<(truncstorei16 (i32 IntRegs:$src1),
3456 (HexagonCONST32 tglobaladdr:$absaddr)),
3457 (S2_storerhabs tglobaladdr: $absaddr, IntRegs: $src1)>;
3459 def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32 tglobaladdr:$absaddr)),
3460 (S2_storeriabs tglobaladdr: $absaddr, IntRegs: $src1)>;
3462 def : Pat<(store (i64 DoubleRegs:$src1),
3463 (HexagonCONST32 tglobaladdr:$absaddr)),
3464 (S2_storerdabs tglobaladdr: $absaddr, DoubleRegs: $src1)>;
3467 // 64 bit atomic store
3468 def : Pat <(atomic_store_64 (HexagonCONST32_GP tglobaladdr:$global),
3469 (i64 DoubleRegs:$src1)),
3470 (S2_storerdgp tglobaladdr:$global, (i64 DoubleRegs:$src1))>,
3473 // Map from store(globaladdress) -> memd(#foo)
3474 let AddedComplexity = 100 in
3475 def : Pat <(store (i64 DoubleRegs:$src1),
3476 (HexagonCONST32_GP tglobaladdr:$global)),
3477 (S2_storerdgp tglobaladdr:$global, (i64 DoubleRegs:$src1))>;
3479 // 8 bit atomic store
3480 def : Pat < (atomic_store_8 (HexagonCONST32_GP tglobaladdr:$global),
3481 (i32 IntRegs:$src1)),
3482 (S2_storerbgp tglobaladdr:$global, (i32 IntRegs:$src1))>;
3484 // Map from store(globaladdress) -> memb(#foo)
3485 let AddedComplexity = 100 in
3486 def : Pat<(truncstorei8 (i32 IntRegs:$src1),
3487 (HexagonCONST32_GP tglobaladdr:$global)),
3488 (S2_storerbgp tglobaladdr:$global, (i32 IntRegs:$src1))>;
3490 // Map from "i1 = constant<-1>; memw(CONST32(#foo)) = i1"
3491 // to "r0 = 1; memw(#foo) = r0"
3492 let AddedComplexity = 100 in
3493 def : Pat<(store (i1 -1), (HexagonCONST32_GP tglobaladdr:$global)),
3494 (S2_storerbgp tglobaladdr:$global, (A2_tfrsi 1))>;
3496 def : Pat<(atomic_store_16 (HexagonCONST32_GP tglobaladdr:$global),
3497 (i32 IntRegs:$src1)),
3498 (S2_storerhgp tglobaladdr:$global, (i32 IntRegs:$src1))>;
3500 // Map from store(globaladdress) -> memh(#foo)
3501 let AddedComplexity = 100 in
3502 def : Pat<(truncstorei16 (i32 IntRegs:$src1),
3503 (HexagonCONST32_GP tglobaladdr:$global)),
3504 (S2_storerhgp tglobaladdr:$global, (i32 IntRegs:$src1))>;
3506 // 32 bit atomic store
3507 def : Pat<(atomic_store_32 (HexagonCONST32_GP tglobaladdr:$global),
3508 (i32 IntRegs:$src1)),
3509 (S2_storerigp tglobaladdr:$global, (i32 IntRegs:$src1))>;
3511 // Map from store(globaladdress) -> memw(#foo)
3512 let AddedComplexity = 100 in
3513 def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)),
3514 (S2_storerigp tglobaladdr:$global, (i32 IntRegs:$src1))>;
3516 //===----------------------------------------------------------------------===//
3517 // Template class for non predicated load instructions with
3518 // absolute addressing mode.
3519 //===----------------------------------------------------------------------===//
3520 let isPredicable = 1, hasSideEffects = 0, validSubTargets = HasV4SubT in
3521 class T_LoadAbsGP <string mnemonic, RegisterClass RC, Operand ImmOp,
3522 bits<3> MajOp, Operand AddrOp, bit isAbs>
3523 : LDInst <(outs RC:$dst), (ins AddrOp:$addr),
3524 "$dst = "#mnemonic# !if(isAbs, "(##", "(#")#"$addr)",
3525 [], "", V2LDST_tc_ld_SLOT01> {
3528 bits<16> offsetBits;
3530 string ImmOpStr = !cast<string>(ImmOp);
3531 let offsetBits = !if (!eq(ImmOpStr, "u16_3Imm"), addr{18-3},
3532 !if (!eq(ImmOpStr, "u16_2Imm"), addr{17-2},
3533 !if (!eq(ImmOpStr, "u16_1Imm"), addr{16-1},
3534 /* u16_0Imm */ addr{15-0})));
3536 let IClass = 0b0100;
3539 let Inst{26-25} = offsetBits{15-14};
3541 let Inst{23-21} = MajOp;
3542 let Inst{20-16} = offsetBits{13-9};
3543 let Inst{13-5} = offsetBits{8-0};
3544 let Inst{4-0} = dst;
3547 class T_LoadAbs <string mnemonic, RegisterClass RC, Operand ImmOp,
3549 : T_LoadAbsGP <mnemonic, RC, ImmOp, MajOp, u0AlwaysExt, 1>, AddrModeRel {
3551 string ImmOpStr = !cast<string>(ImmOp);
3552 let opExtentBits = !if (!eq(ImmOpStr, "u16_3Imm"), 19,
3553 !if (!eq(ImmOpStr, "u16_2Imm"), 18,
3554 !if (!eq(ImmOpStr, "u16_1Imm"), 17,
3555 /* u16_0Imm */ 16)));
3557 let opExtentAlign = !if (!eq(ImmOpStr, "u16_3Imm"), 3,
3558 !if (!eq(ImmOpStr, "u16_2Imm"), 2,
3559 !if (!eq(ImmOpStr, "u16_1Imm"), 1,
3560 /* u16_0Imm */ 0)));
3562 //===----------------------------------------------------------------------===//
3563 // Template class for predicated load instructions with
3564 // absolute addressing mode.
3565 //===----------------------------------------------------------------------===//
3566 let isPredicated = 1, hasNewValue = 1, opExtentBits = 6, opExtendable = 2 in
3567 class T_LoadAbs_Pred <string mnemonic, RegisterClass RC, bits<3> MajOp,
3568 bit isPredNot, bit isPredNew>
3569 : LDInst <(outs RC:$dst), (ins PredRegs:$src1, u6Ext:$absaddr),
3570 !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
3571 ") ")#"$dst = "#mnemonic#"(#$absaddr)">, AddrModeRel {
3576 let isPredicatedNew = isPredNew;
3577 let isPredicatedFalse = isPredNot;
3579 let IClass = 0b1001;
3581 let Inst{27-24} = 0b1111;
3582 let Inst{23-21} = MajOp;
3583 let Inst{20-16} = absaddr{5-1};
3585 let Inst{12} = isPredNew;
3586 let Inst{11} = isPredNot;
3587 let Inst{10-9} = src1;
3588 let Inst{8} = absaddr{0};
3590 let Inst{4-0} = dst;
3593 //===----------------------------------------------------------------------===//
3594 // Multiclass for the load instructions with absolute addressing mode.
3595 //===----------------------------------------------------------------------===//
3596 multiclass LD_Abs_Pred<string mnemonic, RegisterClass RC, bits<3> MajOp,
3598 def _abs : T_LoadAbs_Pred <mnemonic, RC, MajOp, PredNot, 0>;
3600 def new_abs : T_LoadAbs_Pred <mnemonic, RC, MajOp, PredNot, 1>;
3603 let addrMode = Absolute, isExtended = 1 in
3604 multiclass LD_Abs<string mnemonic, string CextOp, RegisterClass RC,
3605 Operand ImmOp, bits<3> MajOp> {
3606 let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
3607 let opExtendable = 1, isPredicable = 1 in
3608 def L4_#NAME#_abs: T_LoadAbs <mnemonic, RC, ImmOp, MajOp>;
3611 defm L4_p#NAME#t : LD_Abs_Pred<mnemonic, RC, MajOp, 0>;
3612 defm L4_p#NAME#f : LD_Abs_Pred<mnemonic, RC, MajOp, 1>;
3616 let accessSize = ByteAccess, hasNewValue = 1, isCodeGenOnly = 0 in {
3617 defm loadrb : LD_Abs<"memb", "LDrib", IntRegs, u16_0Imm, 0b000>;
3618 defm loadrub : LD_Abs<"memub", "LDriub", IntRegs, u16_0Imm, 0b001>;
3621 let accessSize = HalfWordAccess, hasNewValue = 1, isCodeGenOnly = 0 in {
3622 defm loadrh : LD_Abs<"memh", "LDrih", IntRegs, u16_1Imm, 0b010>;
3623 defm loadruh : LD_Abs<"memuh", "LDriuh", IntRegs, u16_1Imm, 0b011>;
3626 let accessSize = WordAccess, hasNewValue = 1, isCodeGenOnly = 0 in
3627 defm loadri : LD_Abs<"memw", "LDriw", IntRegs, u16_2Imm, 0b100>;
3629 let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
3630 defm loadrd : LD_Abs<"memd", "LDrid", DoubleRegs, u16_3Imm, 0b110>;
3632 //===----------------------------------------------------------------------===//
3633 // multiclass for load instructions with GP-relative addressing mode.
3634 // Rx=mem[bhwd](##global)
3635 // Once predicated, these instructions map to absolute addressing mode.
3636 // if ([!]Pv[.new]) Rx=mem[bhwd](##global)
3637 //===----------------------------------------------------------------------===//
3639 class T_LoadGP <string mnemonic, string BaseOp, RegisterClass RC, Operand ImmOp,
3641 : T_LoadAbsGP <mnemonic, RC, ImmOp, MajOp, globaladdress, 0>, PredNewRel {
3642 let BaseOpcode = BaseOp#_abs;
3645 let accessSize = ByteAccess, hasNewValue = 1 in {
3646 def L2_loadrbgp : T_LoadGP<"memb", "LDrib", IntRegs, u16_0Imm, 0b000>;
3647 def L2_loadrubgp : T_LoadGP<"memub", "LDriub", IntRegs, u16_0Imm, 0b001>;
3650 let accessSize = HalfWordAccess, hasNewValue = 1 in {
3651 def L2_loadrhgp : T_LoadGP<"memh", "LDrih", IntRegs, u16_1Imm, 0b010>;
3652 def L2_loadruhgp : T_LoadGP<"memuh", "LDriuh", IntRegs, u16_1Imm, 0b011>;
3655 let accessSize = WordAccess, hasNewValue = 1 in
3656 def L2_loadrigp : T_LoadGP<"memw", "LDriw", IntRegs, u16_2Imm, 0b100>;
3658 let accessSize = DoubleWordAccess in
3659 def L2_loadrdgp : T_LoadGP<"memd", "LDrid", DoubleRegs, u16_3Imm, 0b110>;
3661 let Predicates = [HasV4T], AddedComplexity = 30 in {
3662 def : Pat<(i32 (load (HexagonCONST32 tglobaladdr:$absaddr))),
3663 (L4_loadri_abs tglobaladdr: $absaddr)>;
3665 def : Pat<(i32 (sextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))),
3666 (L4_loadrb_abs tglobaladdr:$absaddr)>;
3668 def : Pat<(i32 (zextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))),
3669 (L4_loadrub_abs tglobaladdr:$absaddr)>;
3671 def : Pat<(i32 (sextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
3672 (L4_loadrh_abs tglobaladdr:$absaddr)>;
3674 def : Pat<(i32 (zextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
3675 (L4_loadruh_abs tglobaladdr:$absaddr)>;
3678 def : Pat <(atomic_load_64 (HexagonCONST32_GP tglobaladdr:$global)),
3679 (i64 (L2_loadrdgp tglobaladdr:$global))>;
3681 def : Pat <(atomic_load_32 (HexagonCONST32_GP tglobaladdr:$global)),
3682 (i32 (L2_loadrigp tglobaladdr:$global))>;
3684 def : Pat <(atomic_load_16 (HexagonCONST32_GP tglobaladdr:$global)),
3685 (i32 (L2_loadruhgp tglobaladdr:$global))>;
3687 def : Pat <(atomic_load_8 (HexagonCONST32_GP tglobaladdr:$global)),
3688 (i32 (L2_loadrubgp tglobaladdr:$global))>;
3690 // Map from load(globaladdress) -> memw(#foo + 0)
3691 let AddedComplexity = 100 in
3692 def : Pat <(i64 (load (HexagonCONST32_GP tglobaladdr:$global))),
3693 (i64 (L2_loadrdgp tglobaladdr:$global))>;
3695 // Map from Pd = load(globaladdress) -> Rd = memb(globaladdress), Pd = Rd
3696 let AddedComplexity = 100 in
3697 def : Pat <(i1 (load (HexagonCONST32_GP tglobaladdr:$global))),
3698 (i1 (C2_tfrrp (i32 (L2_loadrbgp tglobaladdr:$global))))>;
3700 // When the Interprocedural Global Variable optimizer realizes that a certain
3701 // global variable takes only two constant values, it shrinks the global to
3702 // a boolean. Catch those loads here in the following 3 patterns.
3703 let AddedComplexity = 100 in
3704 def : Pat <(i32 (extloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3705 (i32 (L2_loadrbgp tglobaladdr:$global))>;
3707 let AddedComplexity = 100 in
3708 def : Pat <(i32 (sextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3709 (i32 (L2_loadrbgp tglobaladdr:$global))>;
3711 // Map from load(globaladdress) -> memb(#foo)
3712 let AddedComplexity = 100 in
3713 def : Pat <(i32 (extloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3714 (i32 (L2_loadrbgp tglobaladdr:$global))>;
3716 // Map from load(globaladdress) -> memb(#foo)
3717 let AddedComplexity = 100 in
3718 def : Pat <(i32 (sextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3719 (i32 (L2_loadrbgp tglobaladdr:$global))>;
3721 let AddedComplexity = 100 in
3722 def : Pat <(i32 (zextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
3723 (i32 (L2_loadrubgp tglobaladdr:$global))>;
3725 // Map from load(globaladdress) -> memub(#foo)
3726 let AddedComplexity = 100 in
3727 def : Pat <(i32 (zextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
3728 (i32 (L2_loadrubgp tglobaladdr:$global))>;
3730 // Map from load(globaladdress) -> memh(#foo)
3731 let AddedComplexity = 100 in
3732 def : Pat <(i32 (extloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3733 (i32 (L2_loadrhgp tglobaladdr:$global))>;
3735 // Map from load(globaladdress) -> memh(#foo)
3736 let AddedComplexity = 100 in
3737 def : Pat <(i32 (sextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3738 (i32 (L2_loadrhgp tglobaladdr:$global))>;
3740 // Map from load(globaladdress) -> memuh(#foo)
3741 let AddedComplexity = 100 in
3742 def : Pat <(i32 (zextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
3743 (i32 (L2_loadruhgp tglobaladdr:$global))>;
3745 // Map from load(globaladdress) -> memw(#foo)
3746 let AddedComplexity = 100 in
3747 def : Pat <(i32 (load (HexagonCONST32_GP tglobaladdr:$global))),
3748 (i32 (L2_loadrigp tglobaladdr:$global))>;
3751 // Transfer global address into a register
3752 let isExtended = 1, opExtendable = 1, AddedComplexity=50, isMoveImm = 1,
3753 isAsCheapAsAMove = 1, isReMaterializable = 1, validSubTargets = HasV4SubT in
3754 def TFRI_V4 : ALU32_ri<(outs IntRegs:$dst), (ins s16Ext:$src1),
3756 [(set IntRegs:$dst, (HexagonCONST32 tglobaladdr:$src1))]>,
3759 // Transfer a block address into a register
3760 def : Pat<(HexagonCONST32_GP tblockaddress:$src1),
3761 (TFRI_V4 tblockaddress:$src1)>,
3764 let isExtended = 1, opExtendable = 2, AddedComplexity=50,
3765 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3766 def TFRI_cPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3767 (ins PredRegs:$src1, s16Ext:$src2),
3768 "if($src1) $dst = #$src2",
3772 let isExtended = 1, opExtendable = 2, AddedComplexity=50, isPredicatedFalse = 1,
3773 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3774 def TFRI_cNotPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3775 (ins PredRegs:$src1, s16Ext:$src2),
3776 "if(!$src1) $dst = #$src2",
3780 let isExtended = 1, opExtendable = 2, AddedComplexity=50,
3781 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3782 def TFRI_cdnPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3783 (ins PredRegs:$src1, s16Ext:$src2),
3784 "if($src1.new) $dst = #$src2",
3788 let isExtended = 1, opExtendable = 2, AddedComplexity=50, isPredicatedFalse = 1,
3789 hasSideEffects = 0, isPredicated = 1, validSubTargets = HasV4SubT in
3790 def TFRI_cdnNotPt_V4 : ALU32_ri<(outs IntRegs:$dst),
3791 (ins PredRegs:$src1, s16Ext:$src2),
3792 "if(!$src1.new) $dst = #$src2",
3796 let AddedComplexity = 50, Predicates = [HasV4T] in
3797 def : Pat<(HexagonCONST32_GP tglobaladdr:$src1),
3798 (TFRI_V4 tglobaladdr:$src1)>,
3802 // Load - Indirect with long offset: These instructions take global address
3804 let isExtended = 1, opExtendable = 3, AddedComplexity = 40,
3805 validSubTargets = HasV4SubT in
3806 def LDrid_ind_lo_V4 : LDInst<(outs DoubleRegs:$dst),
3807 (ins IntRegs:$src1, u2Imm:$src2, globaladdressExt:$offset),
3808 "$dst=memd($src1<<#$src2+##$offset)",
3809 [(set (i64 DoubleRegs:$dst),
3810 (load (add (shl IntRegs:$src1, u2ImmPred:$src2),
3811 (HexagonCONST32 tglobaladdr:$offset))))]>,
3814 let AddedComplexity = 40 in
3815 multiclass LD_indirect_lo<string OpcStr, PatFrag OpNode> {
3816 let isExtended = 1, opExtendable = 3, validSubTargets = HasV4SubT in
3817 def _lo_V4 : LDInst<(outs IntRegs:$dst),
3818 (ins IntRegs:$src1, u2Imm:$src2, globaladdressExt:$offset),
3819 !strconcat("$dst = ",
3820 !strconcat(OpcStr, "($src1<<#$src2+##$offset)")),
3822 (i32 (OpNode (add (shl IntRegs:$src1, u2ImmPred:$src2),
3823 (HexagonCONST32 tglobaladdr:$offset)))))]>,
3827 defm LDrib_ind : LD_indirect_lo<"memb", sextloadi8>;
3828 defm LDriub_ind : LD_indirect_lo<"memub", zextloadi8>;
3829 defm LDriub_ind_anyext : LD_indirect_lo<"memub", extloadi8>;
3830 defm LDrih_ind : LD_indirect_lo<"memh", sextloadi16>;
3831 defm LDriuh_ind : LD_indirect_lo<"memuh", zextloadi16>;
3832 defm LDriuh_ind_anyext : LD_indirect_lo<"memuh", extloadi16>;
3833 defm LDriw_ind : LD_indirect_lo<"memw", load>;
3835 let AddedComplexity = 40 in
3836 def : Pat <(i32 (sextloadi8 (add IntRegs:$src1,
3837 (NumUsesBelowThresCONST32 tglobaladdr:$offset)))),
3838 (i32 (LDrib_ind_lo_V4 IntRegs:$src1, 0, tglobaladdr:$offset))>,
3841 let AddedComplexity = 40 in
3842 def : Pat <(i32 (zextloadi8 (add IntRegs:$src1,
3843 (NumUsesBelowThresCONST32 tglobaladdr:$offset)))),
3844 (i32 (LDriub_ind_lo_V4 IntRegs:$src1, 0, tglobaladdr:$offset))>,
3847 let Predicates = [HasV4T], AddedComplexity = 30 in {
3848 def : Pat<(truncstorei8 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3849 (S2_storerbabs u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3851 def : Pat<(truncstorei16 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3852 (S2_storerhabs u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3854 def : Pat<(store (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
3855 (S2_storeriabs u0AlwaysExtPred:$src2, IntRegs: $src1)>;
3858 let Predicates = [HasV4T], AddedComplexity = 30 in {
3859 def : Pat<(i32 (load u0AlwaysExtPred:$src)),
3860 (L4_loadri_abs u0AlwaysExtPred:$src)>;
3862 def : Pat<(i32 (sextloadi8 u0AlwaysExtPred:$src)),
3863 (L4_loadrb_abs u0AlwaysExtPred:$src)>;
3865 def : Pat<(i32 (zextloadi8 u0AlwaysExtPred:$src)),
3866 (L4_loadrub_abs u0AlwaysExtPred:$src)>;
3868 def : Pat<(i32 (sextloadi16 u0AlwaysExtPred:$src)),
3869 (L4_loadrh_abs u0AlwaysExtPred:$src)>;
3871 def : Pat<(i32 (zextloadi16 u0AlwaysExtPred:$src)),
3872 (L4_loadruh_abs u0AlwaysExtPred:$src)>;
3875 // Indexed store word - global address.
3876 // memw(Rs+#u6:2)=#S8
3877 let AddedComplexity = 10 in
3878 def STriw_offset_ext_V4 : STInst<(outs),
3879 (ins IntRegs:$src1, u6_2Imm:$src2, globaladdress:$src3),
3880 "memw($src1+#$src2) = ##$src3",
3881 [(store (HexagonCONST32 tglobaladdr:$src3),
3882 (add IntRegs:$src1, u6_2ImmPred:$src2))]>,
3885 def : Pat<(i64 (ctlz (i64 DoubleRegs:$src1))),
3886 (i64 (A4_combineir (i32 0), (i32 (CTLZ64_rr DoubleRegs:$src1))))>,
3889 def : Pat<(i64 (cttz (i64 DoubleRegs:$src1))),
3890 (i64 (A4_combineir (i32 0), (i32 (CTTZ64_rr DoubleRegs:$src1))))>,
3895 // We need a complexity of 120 here to override preceding handling of
3897 let Predicates = [HasV4T], AddedComplexity = 120 in {
3898 def: Pat <(i64 (extloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3899 (i64 (A4_combineir 0, (L4_loadrb_abs tglobaladdr:$addr)))>;
3901 def: Pat <(i64 (zextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3902 (i64 (A4_combineir 0, (L4_loadrub_abs tglobaladdr:$addr)))>;
3904 def: Pat <(i64 (sextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3905 (i64 (A2_sxtw (L4_loadrb_abs tglobaladdr:$addr)))>;
3907 def: Pat <(i64 (extloadi8 FoldGlobalAddr:$addr)),
3908 (i64 (A4_combineir 0, (L4_loadrb_abs FoldGlobalAddr:$addr)))>;
3910 def: Pat <(i64 (zextloadi8 FoldGlobalAddr:$addr)),
3911 (i64 (A4_combineir 0, (L4_loadrub_abs FoldGlobalAddr:$addr)))>;
3913 def: Pat <(i64 (sextloadi8 FoldGlobalAddr:$addr)),
3914 (i64 (A2_sxtw (L4_loadrb_abs FoldGlobalAddr:$addr)))>;
3917 // We need a complexity of 120 here to override preceding handling of
3919 let AddedComplexity = 120 in {
3920 def: Pat <(i64 (extloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3921 (i64 (A4_combineir 0, (L4_loadrh_abs tglobaladdr:$addr)))>,
3924 def: Pat <(i64 (zextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3925 (i64 (A4_combineir 0, (L4_loadruh_abs tglobaladdr:$addr)))>,
3928 def: Pat <(i64 (sextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3929 (i64 (A2_sxtw (L4_loadrh_abs tglobaladdr:$addr)))>,
3932 def: Pat <(i64 (extloadi16 FoldGlobalAddr:$addr)),
3933 (i64 (A4_combineir 0, (L4_loadrh_abs FoldGlobalAddr:$addr)))>,
3936 def: Pat <(i64 (zextloadi16 FoldGlobalAddr:$addr)),
3937 (i64 (A4_combineir 0, (L4_loadruh_abs FoldGlobalAddr:$addr)))>,
3940 def: Pat <(i64 (sextloadi16 FoldGlobalAddr:$addr)),
3941 (i64 (A2_sxtw (L4_loadrh_abs FoldGlobalAddr:$addr)))>,
3945 // We need a complexity of 120 here to override preceding handling of
3947 let AddedComplexity = 120 in {
3948 def: Pat <(i64 (extloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3949 (i64 (A4_combineir 0, (L4_loadri_abs tglobaladdr:$addr)))>,
3952 def: Pat <(i64 (zextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3953 (i64 (A4_combineir 0, (L4_loadri_abs tglobaladdr:$addr)))>,
3956 def: Pat <(i64 (sextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
3957 (i64 (A2_sxtw (L4_loadri_abs tglobaladdr:$addr)))>,
3960 def: Pat <(i64 (extloadi32 FoldGlobalAddr:$addr)),
3961 (i64 (A4_combineir 0, (L4_loadri_abs FoldGlobalAddr:$addr)))>,
3964 def: Pat <(i64 (zextloadi32 FoldGlobalAddr:$addr)),
3965 (i64 (A4_combineir 0, (L4_loadri_abs FoldGlobalAddr:$addr)))>,
3968 def: Pat <(i64 (sextloadi32 FoldGlobalAddr:$addr)),
3969 (i64 (A2_sxtw (L4_loadri_abs FoldGlobalAddr:$addr)))>,
3973 // Indexed store double word - global address.
3974 // memw(Rs+#u6:2)=#S8
3975 let AddedComplexity = 10 in
3976 def STrih_offset_ext_V4 : STInst<(outs),
3977 (ins IntRegs:$src1, u6_1Imm:$src2, globaladdress:$src3),
3978 "memh($src1+#$src2) = ##$src3",
3979 [(truncstorei16 (HexagonCONST32 tglobaladdr:$src3),
3980 (add IntRegs:$src1, u6_1ImmPred:$src2))]>,
3982 // Map from store(globaladdress + x) -> memd(#foo + x)
3983 let AddedComplexity = 100 in
3984 def : Pat<(store (i64 DoubleRegs:$src1),
3985 FoldGlobalAddrGP:$addr),
3986 (S2_storerdabs FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
3989 def : Pat<(atomic_store_64 FoldGlobalAddrGP:$addr,
3990 (i64 DoubleRegs:$src1)),
3991 (S2_storerdabs FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
3994 // Map from store(globaladdress + x) -> memb(#foo + x)
3995 let AddedComplexity = 100 in
3996 def : Pat<(truncstorei8 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
3997 (S2_storerbabs FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
4000 def : Pat<(atomic_store_8 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
4001 (S2_storerbabs FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
4004 // Map from store(globaladdress + x) -> memh(#foo + x)
4005 let AddedComplexity = 100 in
4006 def : Pat<(truncstorei16 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
4007 (S2_storerhabs FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
4010 def : Pat<(atomic_store_16 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
4011 (S2_storerhabs FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
4014 // Map from store(globaladdress + x) -> memw(#foo + x)
4015 let AddedComplexity = 100 in
4016 def : Pat<(store (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
4017 (S2_storeriabs FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
4020 def : Pat<(atomic_store_32 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
4021 (S2_storeriabs FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
4024 // Map from load(globaladdress + x) -> memd(#foo + x)
4025 let AddedComplexity = 100 in
4026 def : Pat<(i64 (load FoldGlobalAddrGP:$addr)),
4027 (i64 (L4_loadrd_abs FoldGlobalAddrGP:$addr))>,
4030 def : Pat<(atomic_load_64 FoldGlobalAddrGP:$addr),
4031 (i64 (L4_loadrd_abs FoldGlobalAddrGP:$addr))>,
4034 // Map from load(globaladdress + x) -> memb(#foo + x)
4035 let AddedComplexity = 100 in
4036 def : Pat<(i32 (extloadi8 FoldGlobalAddrGP:$addr)),
4037 (i32 (L4_loadrb_abs FoldGlobalAddrGP:$addr))>,
4040 // Map from load(globaladdress + x) -> memb(#foo + x)
4041 let AddedComplexity = 100 in
4042 def : Pat<(i32 (sextloadi8 FoldGlobalAddrGP:$addr)),
4043 (i32 (L4_loadrb_abs FoldGlobalAddrGP:$addr))>,
4046 //let AddedComplexity = 100 in
4047 let AddedComplexity = 100 in
4048 def : Pat<(i32 (extloadi16 FoldGlobalAddrGP:$addr)),
4049 (i32 (L4_loadrh_abs FoldGlobalAddrGP:$addr))>,
4052 // Map from load(globaladdress + x) -> memh(#foo + x)
4053 let AddedComplexity = 100 in
4054 def : Pat<(i32 (sextloadi16 FoldGlobalAddrGP:$addr)),
4055 (i32 (L4_loadrh_abs FoldGlobalAddrGP:$addr))>,
4058 // Map from load(globaladdress + x) -> memuh(#foo + x)
4059 let AddedComplexity = 100 in
4060 def : Pat<(i32 (zextloadi16 FoldGlobalAddrGP:$addr)),
4061 (i32 (L4_loadruh_abs FoldGlobalAddrGP:$addr))>,
4064 def : Pat<(atomic_load_16 FoldGlobalAddrGP:$addr),
4065 (i32 (L4_loadruh_abs FoldGlobalAddrGP:$addr))>,
4068 // Map from load(globaladdress + x) -> memub(#foo + x)
4069 let AddedComplexity = 100 in
4070 def : Pat<(i32 (zextloadi8 FoldGlobalAddrGP:$addr)),
4071 (i32 (L4_loadrub_abs FoldGlobalAddrGP:$addr))>,
4074 def : Pat<(atomic_load_8 FoldGlobalAddrGP:$addr),
4075 (i32 (L4_loadrub_abs FoldGlobalAddrGP:$addr))>,
4078 // Map from load(globaladdress + x) -> memw(#foo + x)
4079 let AddedComplexity = 100 in
4080 def : Pat<(i32 (load FoldGlobalAddrGP:$addr)),
4081 (i32 (L4_loadri_abs FoldGlobalAddrGP:$addr))>,
4084 def : Pat<(atomic_load_32 FoldGlobalAddrGP:$addr),
4085 (i32 (L4_loadri_abs FoldGlobalAddrGP:$addr))>,
4088 //===----------------------------------------------------------------------===//
4089 // :raw for of boundscheck:hi:lo insns
4090 //===----------------------------------------------------------------------===//
4092 // A4_boundscheck_lo: Detect if a register is within bounds.
4093 let hasSideEffects = 0, isCodeGenOnly = 0 in
4094 def A4_boundscheck_lo: ALU64Inst <
4095 (outs PredRegs:$Pd),
4096 (ins DoubleRegs:$Rss, DoubleRegs:$Rtt),
4097 "$Pd = boundscheck($Rss, $Rtt):raw:lo"> {
4102 let IClass = 0b1101;
4104 let Inst{27-23} = 0b00100;
4106 let Inst{7-5} = 0b100;
4108 let Inst{20-16} = Rss;
4109 let Inst{12-8} = Rtt;
4112 // A4_boundscheck_hi: Detect if a register is within bounds.
4113 let hasSideEffects = 0, isCodeGenOnly = 0 in
4114 def A4_boundscheck_hi: ALU64Inst <
4115 (outs PredRegs:$Pd),
4116 (ins DoubleRegs:$Rss, DoubleRegs:$Rtt),
4117 "$Pd = boundscheck($Rss, $Rtt):raw:hi"> {
4122 let IClass = 0b1101;
4124 let Inst{27-23} = 0b00100;
4126 let Inst{7-5} = 0b101;
4128 let Inst{20-16} = Rss;
4129 let Inst{12-8} = Rtt;
4132 let hasSideEffects = 0 in
4133 def A4_boundscheck : MInst <
4134 (outs PredRegs:$Pd), (ins IntRegs:$Rs, DoubleRegs:$Rtt),
4135 "$Pd=boundscheck($Rs,$Rtt)">;
4137 // A4_tlbmatch: Detect if a VA/ASID matches a TLB entry.
4138 let isPredicateLate = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
4139 def A4_tlbmatch : ALU64Inst<(outs PredRegs:$Pd),
4140 (ins DoubleRegs:$Rs, IntRegs:$Rt),
4141 "$Pd = tlbmatch($Rs, $Rt)",
4142 [], "", ALU64_tc_2early_SLOT23> {
4147 let IClass = 0b1101;
4148 let Inst{27-23} = 0b00100;
4149 let Inst{20-16} = Rs;
4151 let Inst{12-8} = Rt;
4152 let Inst{7-5} = 0b011;
4156 // We need custom lowering of ISD::PREFETCH into HexagonISD::DCFETCH
4157 // because the SDNode ISD::PREFETCH has properties MayLoad and MayStore.
4158 // We don't really want either one here.
4159 def SDTHexagonDCFETCH : SDTypeProfile<0, 2, [SDTCisPtrTy<0>,SDTCisInt<1>]>;
4160 def HexagonDCFETCH : SDNode<"HexagonISD::DCFETCH", SDTHexagonDCFETCH,
4163 // Use LD0Inst for dcfetch, but set "mayLoad" to 0 because this doesn't
4164 // really do a load.
4165 let hasSideEffects = 1, mayLoad = 0, isCodeGenOnly = 0 in
4166 def Y2_dcfetchbo : LD0Inst<(outs), (ins IntRegs:$Rs, u11_3Imm:$u11_3),
4167 "dcfetch($Rs + #$u11_3)",
4168 [(HexagonDCFETCH IntRegs:$Rs, u11_3ImmPred:$u11_3)],
4169 "", LD_tc_ld_SLOT0> {
4173 let IClass = 0b1001;
4174 let Inst{27-21} = 0b0100000;
4175 let Inst{20-16} = Rs;
4177 let Inst{10-0} = u11_3{13-3};
4180 //===----------------------------------------------------------------------===//
4181 // Compound instructions
4182 //===----------------------------------------------------------------------===//
4184 let isBranch = 1, hasSideEffects = 0, isExtentSigned = 1,
4185 isPredicated = 1, isPredicatedNew = 1, isExtendable = 1,
4186 opExtentBits = 11, opExtentAlign = 2, opExtendable = 1,
4187 isTerminator = 1, validSubTargets = HasV4SubT in
4188 class CJInst_tstbit_R0<string px, bit np, string tnt>
4189 : InstHexagon<(outs), (ins IntRegs:$Rs, brtarget:$r9_2),
4190 ""#px#" = tstbit($Rs, #0); if ("
4191 #!if(np, "!","")#""#px#".new) jump:"#tnt#" $r9_2",
4192 [], "", COMPOUND, TypeCOMPOUND> {
4197 let isPredicatedFalse = np;
4198 // tnt: Taken/Not Taken
4199 let isBrTaken = !if (!eq(tnt, "t"), "true", "false");
4200 let isTaken = !if (!eq(tnt, "t"), 1, 0);
4202 let IClass = 0b0001;
4203 let Inst{27-26} = 0b00;
4204 let Inst{25} = !if (!eq(px, "!p1"), 1,
4205 !if (!eq(px, "p1"), 1, 0));
4206 let Inst{24-23} = 0b11;
4208 let Inst{21-20} = r9_2{10-9};
4209 let Inst{19-16} = Rs;
4210 let Inst{13} = !if (!eq(tnt, "t"), 1, 0);
4211 let Inst{9-8} = 0b11;
4212 let Inst{7-1} = r9_2{8-2};
4215 let Defs = [PC, P0], Uses = [P0], isCodeGenOnly = 0 in {
4216 def J4_tstbit0_tp0_jump_nt : CJInst_tstbit_R0<"p0", 0, "nt">;
4217 def J4_tstbit0_tp0_jump_t : CJInst_tstbit_R0<"p0", 0, "t">;
4218 def J4_tstbit0_fp0_jump_nt : CJInst_tstbit_R0<"p0", 1, "nt">;
4219 def J4_tstbit0_fp0_jump_t : CJInst_tstbit_R0<"p0", 1, "t">;
4222 let Defs = [PC, P1], Uses = [P1], isCodeGenOnly = 0 in {
4223 def J4_tstbit0_tp1_jump_nt : CJInst_tstbit_R0<"p1", 0, "nt">;
4224 def J4_tstbit0_tp1_jump_t : CJInst_tstbit_R0<"p1", 0, "t">;
4225 def J4_tstbit0_fp1_jump_nt : CJInst_tstbit_R0<"p1", 1, "nt">;
4226 def J4_tstbit0_fp1_jump_t : CJInst_tstbit_R0<"p1", 1, "t">;
4230 let isBranch = 1, hasSideEffects = 0,
4231 isExtentSigned = 1, isPredicated = 1, isPredicatedNew = 1,
4232 isExtendable = 1, opExtentBits = 11, opExtentAlign = 2,
4233 opExtendable = 2, isTerminator = 1, validSubTargets = HasV4SubT in
4234 class CJInst_RR<string px, string op, bit np, string tnt>
4235 : InstHexagon<(outs), (ins IntRegs:$Rs, IntRegs:$Rt, brtarget:$r9_2),
4236 ""#px#" = cmp."#op#"($Rs, $Rt); if ("
4237 #!if(np, "!","")#""#px#".new) jump:"#tnt#" $r9_2",
4238 [], "", COMPOUND, TypeCOMPOUND> {
4244 let isPredicatedFalse = np;
4245 // tnt: Taken/Not Taken
4246 let isBrTaken = !if (!eq(tnt, "t"), "true", "false");
4247 let isTaken = !if (!eq(tnt, "t"), 1, 0);
4249 let IClass = 0b0001;
4250 let Inst{27-23} = !if (!eq(op, "eq"), 0b01000,
4251 !if (!eq(op, "gt"), 0b01001,
4252 !if (!eq(op, "gtu"), 0b01010, 0)));
4254 let Inst{21-20} = r9_2{10-9};
4255 let Inst{19-16} = Rs;
4256 let Inst{13} = !if (!eq(tnt, "t"), 1, 0);
4257 // px: Predicate reg 0/1
4258 let Inst{12} = !if (!eq(px, "!p1"), 1,
4259 !if (!eq(px, "p1"), 1, 0));
4260 let Inst{11-8} = Rt;
4261 let Inst{7-1} = r9_2{8-2};
4264 // P[10] taken/not taken.
4265 multiclass T_tnt_CJInst_RR<string op, bit np> {
4266 let Defs = [PC, P0], Uses = [P0] in {
4267 def NAME#p0_jump_nt : CJInst_RR<"p0", op, np, "nt">;
4268 def NAME#p0_jump_t : CJInst_RR<"p0", op, np, "t">;
4270 let Defs = [PC, P1], Uses = [P1] in {
4271 def NAME#p1_jump_nt : CJInst_RR<"p1", op, np, "nt">;
4272 def NAME#p1_jump_t : CJInst_RR<"p1", op, np, "t">;
4275 // Predicate / !Predicate
4276 multiclass T_pnp_CJInst_RR<string op>{
4277 defm J4_cmp#NAME#_t : T_tnt_CJInst_RR<op, 0>;
4278 defm J4_cmp#NAME#_f : T_tnt_CJInst_RR<op, 1>;
4280 // TypeCJ Instructions compare RR and jump
4281 let isCodeGenOnly = 0 in {
4282 defm eq : T_pnp_CJInst_RR<"eq">;
4283 defm gt : T_pnp_CJInst_RR<"gt">;
4284 defm gtu : T_pnp_CJInst_RR<"gtu">;
4287 let isBranch = 1, hasSideEffects = 0, isExtentSigned = 1,
4288 isPredicated = 1, isPredicatedNew = 1, isExtendable = 1, opExtentBits = 11,
4289 opExtentAlign = 2, opExtendable = 2, isTerminator = 1,
4290 validSubTargets = HasV4SubT in
4291 class CJInst_RU5<string px, string op, bit np, string tnt>
4292 : InstHexagon<(outs), (ins IntRegs:$Rs, u5Imm:$U5, brtarget:$r9_2),
4293 ""#px#" = cmp."#op#"($Rs, #$U5); if ("
4294 #!if(np, "!","")#""#px#".new) jump:"#tnt#" $r9_2",
4295 [], "", COMPOUND, TypeCOMPOUND> {
4301 let isPredicatedFalse = np;
4302 // tnt: Taken/Not Taken
4303 let isBrTaken = !if (!eq(tnt, "t"), "true", "false");
4304 let isTaken = !if (!eq(tnt, "t"), 1, 0);
4306 let IClass = 0b0001;
4307 let Inst{27-26} = 0b00;
4308 // px: Predicate reg 0/1
4309 let Inst{25} = !if (!eq(px, "!p1"), 1,
4310 !if (!eq(px, "p1"), 1, 0));
4311 let Inst{24-23} = !if (!eq(op, "eq"), 0b00,
4312 !if (!eq(op, "gt"), 0b01,
4313 !if (!eq(op, "gtu"), 0b10, 0)));
4315 let Inst{21-20} = r9_2{10-9};
4316 let Inst{19-16} = Rs;
4317 let Inst{13} = !if (!eq(tnt, "t"), 1, 0);
4318 let Inst{12-8} = U5;
4319 let Inst{7-1} = r9_2{8-2};
4321 // P[10] taken/not taken.
4322 multiclass T_tnt_CJInst_RU5<string op, bit np> {
4323 let Defs = [PC, P0], Uses = [P0] in {
4324 def NAME#p0_jump_nt : CJInst_RU5<"p0", op, np, "nt">;
4325 def NAME#p0_jump_t : CJInst_RU5<"p0", op, np, "t">;
4327 let Defs = [PC, P1], Uses = [P1] in {
4328 def NAME#p1_jump_nt : CJInst_RU5<"p1", op, np, "nt">;
4329 def NAME#p1_jump_t : CJInst_RU5<"p1", op, np, "t">;
4332 // Predicate / !Predicate
4333 multiclass T_pnp_CJInst_RU5<string op>{
4334 defm J4_cmp#NAME#i_t : T_tnt_CJInst_RU5<op, 0>;
4335 defm J4_cmp#NAME#i_f : T_tnt_CJInst_RU5<op, 1>;
4337 // TypeCJ Instructions compare RI and jump
4338 let isCodeGenOnly = 0 in {
4339 defm eq : T_pnp_CJInst_RU5<"eq">;
4340 defm gt : T_pnp_CJInst_RU5<"gt">;
4341 defm gtu : T_pnp_CJInst_RU5<"gtu">;
4344 let isBranch = 1, hasSideEffects = 0, isExtentSigned = 1,
4345 isPredicated = 1, isPredicatedFalse = 1, isPredicatedNew = 1,
4346 isExtendable = 1, opExtentBits = 11, opExtentAlign = 2, opExtendable = 1,
4347 isTerminator = 1, validSubTargets = HasV4SubT in
4348 class CJInst_Rn1<string px, string op, bit np, string tnt>
4349 : InstHexagon<(outs), (ins IntRegs:$Rs, brtarget:$r9_2),
4350 ""#px#" = cmp."#op#"($Rs,#-1); if ("
4351 #!if(np, "!","")#""#px#".new) jump:"#tnt#" $r9_2",
4352 [], "", COMPOUND, TypeCOMPOUND> {
4357 let isPredicatedFalse = np;
4358 // tnt: Taken/Not Taken
4359 let isBrTaken = !if (!eq(tnt, "t"), "true", "false");
4360 let isTaken = !if (!eq(tnt, "t"), 1, 0);
4362 let IClass = 0b0001;
4363 let Inst{27-26} = 0b00;
4364 let Inst{25} = !if (!eq(px, "!p1"), 1,
4365 !if (!eq(px, "p1"), 1, 0));
4367 let Inst{24-23} = 0b11;
4369 let Inst{21-20} = r9_2{10-9};
4370 let Inst{19-16} = Rs;
4371 let Inst{13} = !if (!eq(tnt, "t"), 1, 0);
4372 let Inst{9-8} = !if (!eq(op, "eq"), 0b00,
4373 !if (!eq(op, "gt"), 0b01, 0));
4374 let Inst{7-1} = r9_2{8-2};
4377 // P[10] taken/not taken.
4378 multiclass T_tnt_CJInst_Rn1<string op, bit np> {
4379 let Defs = [PC, P0], Uses = [P0] in {
4380 def NAME#p0_jump_nt : CJInst_Rn1<"p0", op, np, "nt">;
4381 def NAME#p0_jump_t : CJInst_Rn1<"p0", op, np, "t">;
4383 let Defs = [PC, P1], Uses = [P1] in {
4384 def NAME#p1_jump_nt : CJInst_Rn1<"p1", op, np, "nt">;
4385 def NAME#p1_jump_t : CJInst_Rn1<"p1", op, np, "t">;
4388 // Predicate / !Predicate
4389 multiclass T_pnp_CJInst_Rn1<string op>{
4390 defm J4_cmp#NAME#n1_t : T_tnt_CJInst_Rn1<op, 0>;
4391 defm J4_cmp#NAME#n1_f : T_tnt_CJInst_Rn1<op, 1>;
4393 // TypeCJ Instructions compare -1 and jump
4394 let isCodeGenOnly = 0 in {
4395 defm eq : T_pnp_CJInst_Rn1<"eq">;
4396 defm gt : T_pnp_CJInst_Rn1<"gt">;
4399 // J4_jumpseti: Direct unconditional jump and set register to immediate.
4400 let Defs = [PC], isBranch = 1, hasSideEffects = 0, hasNewValue = 1,
4401 isExtentSigned = 1, opNewValue = 0, isExtendable = 1, opExtentBits = 11,
4402 opExtentAlign = 2, opExtendable = 2, validSubTargets = HasV4SubT,
4403 isCodeGenOnly = 0 in
4404 def J4_jumpseti: CJInst <
4406 (ins u6Imm:$U6, brtarget:$r9_2),
4407 "$Rd = #$U6 ; jump $r9_2"> {
4412 let IClass = 0b0001;
4413 let Inst{27-24} = 0b0110;
4414 let Inst{21-20} = r9_2{10-9};
4415 let Inst{19-16} = Rd;
4416 let Inst{13-8} = U6;
4417 let Inst{7-1} = r9_2{8-2};
4420 // J4_jumpsetr: Direct unconditional jump and transfer register.
4421 let Defs = [PC], isBranch = 1, hasSideEffects = 0, hasNewValue = 1,
4422 isExtentSigned = 1, opNewValue = 0, isExtendable = 1, opExtentBits = 11,
4423 opExtentAlign = 2, opExtendable = 2, validSubTargets = HasV4SubT,
4424 isCodeGenOnly = 0 in
4425 def J4_jumpsetr: CJInst <
4427 (ins IntRegs:$Rs, brtarget:$r9_2),
4428 "$Rd = $Rs ; jump $r9_2"> {
4433 let IClass = 0b0001;
4434 let Inst{27-24} = 0b0111;
4435 let Inst{21-20} = r9_2{10-9};
4436 let Inst{11-8} = Rd;
4437 let Inst{19-16} = Rs;
4438 let Inst{7-1} = r9_2{8-2};