1 //===-- SparcInstrInfo.td - Target Description for Sparc Target -----------===//
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 Sparc instructions in TableGen format.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // Instruction format superclass
16 //===----------------------------------------------------------------------===//
18 include "SparcInstrFormats.td"
20 //===----------------------------------------------------------------------===//
21 // Feature predicates.
22 //===----------------------------------------------------------------------===//
24 // True when generating 32-bit code.
25 def Is32Bit : Predicate<"!Subtarget.is64Bit()">;
27 // True when generating 64-bit code. This also implies HasV9.
28 def Is64Bit : Predicate<"Subtarget.is64Bit()">;
30 // HasV9 - This predicate is true when the target processor supports V9
31 // instructions. Note that the machine may be running in 32-bit mode.
32 def HasV9 : Predicate<"Subtarget.isV9()">,
33 AssemblerPredicate<"FeatureV9">;
35 // HasNoV9 - This predicate is true when the target doesn't have V9
36 // instructions. Use of this is just a hack for the isel not having proper
37 // costs for V8 instructions that are more expensive than their V9 ones.
38 def HasNoV9 : Predicate<"!Subtarget.isV9()">;
40 // HasVIS - This is true when the target processor has VIS extensions.
41 def HasVIS : Predicate<"Subtarget.isVIS()">;
43 // HasHardQuad - This is true when the target processor supports quad floating
44 // point instructions.
45 def HasHardQuad : Predicate<"Subtarget.hasHardQuad()">;
47 // UseDeprecatedInsts - This predicate is true when the target processor is a
48 // V8, or when it is V9 but the V8 deprecated instructions are efficient enough
49 // to use when appropriate. In either of these cases, the instruction selector
50 // will pick deprecated instructions.
51 def UseDeprecatedInsts : Predicate<"Subtarget.useDeprecatedV8Instructions()">;
53 //===----------------------------------------------------------------------===//
54 // Instruction Pattern Stuff
55 //===----------------------------------------------------------------------===//
57 def simm11 : PatLeaf<(imm), [{ return isInt<11>(N->getSExtValue()); }]>;
59 def simm13 : PatLeaf<(imm), [{ return isInt<13>(N->getSExtValue()); }]>;
61 def LO10 : SDNodeXForm<imm, [{
62 return CurDAG->getTargetConstant((unsigned)N->getZExtValue() & 1023,
66 def HI22 : SDNodeXForm<imm, [{
67 // Transformation function: shift the immediate value down into the low bits.
68 return CurDAG->getTargetConstant((unsigned)N->getZExtValue() >> 10, MVT::i32);
71 def SETHIimm : PatLeaf<(imm), [{
72 return isShiftedUInt<22, 10>(N->getZExtValue());
76 def ADDRrr : ComplexPattern<iPTR, 2, "SelectADDRrr", [], []>;
77 def ADDRri : ComplexPattern<iPTR, 2, "SelectADDRri", [frameindex], []>;
80 def SparcMEMrrAsmOperand : AsmOperandClass {
82 let ParserMethod = "parseMEMOperand";
85 def SparcMEMriAsmOperand : AsmOperandClass {
87 let ParserMethod = "parseMEMOperand";
90 def MEMrr : Operand<iPTR> {
91 let PrintMethod = "printMemOperand";
92 let MIOperandInfo = (ops ptr_rc, ptr_rc);
93 let ParserMatchClass = SparcMEMrrAsmOperand;
95 def MEMri : Operand<iPTR> {
96 let PrintMethod = "printMemOperand";
97 let MIOperandInfo = (ops ptr_rc, i32imm);
98 let ParserMatchClass = SparcMEMriAsmOperand;
101 def TLSSym : Operand<iPTR>;
103 // Branch targets have OtherVT type.
104 def brtarget : Operand<OtherVT> {
105 let EncoderMethod = "getBranchTargetOpValue";
108 def calltarget : Operand<i32> {
109 let EncoderMethod = "getCallTargetOpValue";
110 let DecoderMethod = "DecodeCall";
113 def simm13Op : Operand<i32> {
114 let DecoderMethod = "DecodeSIMM13";
117 // Operand for printing out a condition code.
118 let PrintMethod = "printCCOperand" in
119 def CCOp : Operand<i32>;
122 SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>]>;
124 SDTypeProfile<0, 2, [SDTCisFP<0>, SDTCisSameAs<0, 1>]>;
126 SDTypeProfile<0, 2, [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>]>;
128 SDTypeProfile<1, 3, [SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>, SDTCisVT<3, i32>]>;
130 SDTypeProfile<1, 1, [SDTCisVT<0, f32>, SDTCisFP<1>]>;
132 SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisVT<1, f32>]>;
134 SDTypeProfile<1, 1, [SDTCisVT<0, f64>, SDTCisFP<1>]>;
136 SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisVT<1, f64>]>;
139 SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>, SDTCisPtrTy<2>]>;
141 SDTypeProfile<1, 2, [SDTCisPtrTy<0>, SDTCisPtrTy<1>]>;
143 def SPcmpicc : SDNode<"SPISD::CMPICC", SDTSPcmpicc, [SDNPOutGlue]>;
144 def SPcmpfcc : SDNode<"SPISD::CMPFCC", SDTSPcmpfcc, [SDNPOutGlue]>;
145 def SPbricc : SDNode<"SPISD::BRICC", SDTSPbrcc, [SDNPHasChain, SDNPInGlue]>;
146 def SPbrxcc : SDNode<"SPISD::BRXCC", SDTSPbrcc, [SDNPHasChain, SDNPInGlue]>;
147 def SPbrfcc : SDNode<"SPISD::BRFCC", SDTSPbrcc, [SDNPHasChain, SDNPInGlue]>;
149 def SPhi : SDNode<"SPISD::Hi", SDTIntUnaryOp>;
150 def SPlo : SDNode<"SPISD::Lo", SDTIntUnaryOp>;
152 def SPftoi : SDNode<"SPISD::FTOI", SDTSPFTOI>;
153 def SPitof : SDNode<"SPISD::ITOF", SDTSPITOF>;
154 def SPftox : SDNode<"SPISD::FTOX", SDTSPFTOX>;
155 def SPxtof : SDNode<"SPISD::XTOF", SDTSPXTOF>;
157 def SPselecticc : SDNode<"SPISD::SELECT_ICC", SDTSPselectcc, [SDNPInGlue]>;
158 def SPselectxcc : SDNode<"SPISD::SELECT_XCC", SDTSPselectcc, [SDNPInGlue]>;
159 def SPselectfcc : SDNode<"SPISD::SELECT_FCC", SDTSPselectcc, [SDNPInGlue]>;
161 // These are target-independent nodes, but have target-specific formats.
162 def SDT_SPCallSeqStart : SDCallSeqStart<[ SDTCisVT<0, i32> ]>;
163 def SDT_SPCallSeqEnd : SDCallSeqEnd<[ SDTCisVT<0, i32>,
166 def callseq_start : SDNode<"ISD::CALLSEQ_START", SDT_SPCallSeqStart,
167 [SDNPHasChain, SDNPOutGlue]>;
168 def callseq_end : SDNode<"ISD::CALLSEQ_END", SDT_SPCallSeqEnd,
169 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
171 def SDT_SPCall : SDTypeProfile<0, -1, [SDTCisVT<0, i32>]>;
172 def call : SDNode<"SPISD::CALL", SDT_SPCall,
173 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
176 def SDT_SPRet : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
177 def retflag : SDNode<"SPISD::RET_FLAG", SDT_SPRet,
178 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
180 def flushw : SDNode<"SPISD::FLUSHW", SDTNone,
181 [SDNPHasChain, SDNPSideEffect, SDNPMayStore]>;
183 def tlsadd : SDNode<"SPISD::TLS_ADD", SDTSPtlsadd>;
184 def tlsld : SDNode<"SPISD::TLS_LD", SDTSPtlsld>;
185 def tlscall : SDNode<"SPISD::TLS_CALL", SDT_SPCall,
186 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
189 def getPCX : Operand<iPTR> {
190 let PrintMethod = "printGetPCX";
193 //===----------------------------------------------------------------------===//
194 // SPARC Flag Conditions
195 //===----------------------------------------------------------------------===//
197 // Note that these values must be kept in sync with the CCOp::CondCode enum
199 class ICC_VAL<int N> : PatLeaf<(i32 N)>;
200 def ICC_NE : ICC_VAL< 9>; // Not Equal
201 def ICC_E : ICC_VAL< 1>; // Equal
202 def ICC_G : ICC_VAL<10>; // Greater
203 def ICC_LE : ICC_VAL< 2>; // Less or Equal
204 def ICC_GE : ICC_VAL<11>; // Greater or Equal
205 def ICC_L : ICC_VAL< 3>; // Less
206 def ICC_GU : ICC_VAL<12>; // Greater Unsigned
207 def ICC_LEU : ICC_VAL< 4>; // Less or Equal Unsigned
208 def ICC_CC : ICC_VAL<13>; // Carry Clear/Great or Equal Unsigned
209 def ICC_CS : ICC_VAL< 5>; // Carry Set/Less Unsigned
210 def ICC_POS : ICC_VAL<14>; // Positive
211 def ICC_NEG : ICC_VAL< 6>; // Negative
212 def ICC_VC : ICC_VAL<15>; // Overflow Clear
213 def ICC_VS : ICC_VAL< 7>; // Overflow Set
215 class FCC_VAL<int N> : PatLeaf<(i32 N)>;
216 def FCC_U : FCC_VAL<23>; // Unordered
217 def FCC_G : FCC_VAL<22>; // Greater
218 def FCC_UG : FCC_VAL<21>; // Unordered or Greater
219 def FCC_L : FCC_VAL<20>; // Less
220 def FCC_UL : FCC_VAL<19>; // Unordered or Less
221 def FCC_LG : FCC_VAL<18>; // Less or Greater
222 def FCC_NE : FCC_VAL<17>; // Not Equal
223 def FCC_E : FCC_VAL<25>; // Equal
224 def FCC_UE : FCC_VAL<24>; // Unordered or Equal
225 def FCC_GE : FCC_VAL<25>; // Greater or Equal
226 def FCC_UGE : FCC_VAL<26>; // Unordered or Greater or Equal
227 def FCC_LE : FCC_VAL<27>; // Less or Equal
228 def FCC_ULE : FCC_VAL<28>; // Unordered or Less or Equal
229 def FCC_O : FCC_VAL<29>; // Ordered
231 //===----------------------------------------------------------------------===//
232 // Instruction Class Templates
233 //===----------------------------------------------------------------------===//
235 /// F3_12 multiclass - Define a normal F3_1/F3_2 pattern in one shot.
236 multiclass F3_12<string OpcStr, bits<6> Op3Val, SDNode OpNode,
237 RegisterClass RC, ValueType Ty, Operand immOp> {
238 def rr : F3_1<2, Op3Val,
239 (outs RC:$rd), (ins RC:$rs1, RC:$rs2),
240 !strconcat(OpcStr, " $rs1, $rs2, $rd"),
241 [(set Ty:$rd, (OpNode Ty:$rs1, Ty:$rs2))]>;
242 def ri : F3_2<2, Op3Val,
243 (outs RC:$rd), (ins RC:$rs1, immOp:$simm13),
244 !strconcat(OpcStr, " $rs1, $simm13, $rd"),
245 [(set Ty:$rd, (OpNode Ty:$rs1, (Ty simm13:$simm13)))]>;
248 /// F3_12np multiclass - Define a normal F3_1/F3_2 pattern in one shot, with no
250 multiclass F3_12np<string OpcStr, bits<6> Op3Val> {
251 def rr : F3_1<2, Op3Val,
252 (outs IntRegs:$rd), (ins IntRegs:$rs1, IntRegs:$rs2),
253 !strconcat(OpcStr, " $rs1, $rs2, $rd"), []>;
254 def ri : F3_2<2, Op3Val,
255 (outs IntRegs:$rd), (ins IntRegs:$rs1, simm13Op:$simm13),
256 !strconcat(OpcStr, " $rs1, $simm13, $rd"), []>;
259 // Load multiclass - Define both Reg+Reg/Reg+Imm patterns in one shot.
260 multiclass Load<string OpcStr, bits<6> Op3Val, SDPatternOperator OpNode,
261 RegisterClass RC, ValueType Ty> {
262 def rr : F3_1<3, Op3Val,
263 (outs RC:$dst), (ins MEMrr:$addr),
264 !strconcat(OpcStr, " [$addr], $dst"),
265 [(set Ty:$dst, (OpNode ADDRrr:$addr))]>;
266 def ri : F3_2<3, Op3Val,
267 (outs RC:$dst), (ins MEMri:$addr),
268 !strconcat(OpcStr, " [$addr], $dst"),
269 [(set Ty:$dst, (OpNode ADDRri:$addr))]>;
272 // Store multiclass - Define both Reg+Reg/Reg+Imm patterns in one shot.
273 multiclass Store<string OpcStr, bits<6> Op3Val, SDPatternOperator OpNode,
274 RegisterClass RC, ValueType Ty> {
275 def rr : F3_1<3, Op3Val,
276 (outs), (ins MEMrr:$addr, RC:$rd),
277 !strconcat(OpcStr, " $rd, [$addr]"),
278 [(OpNode Ty:$rd, ADDRrr:$addr)]>;
279 def ri : F3_2<3, Op3Val,
280 (outs), (ins MEMri:$addr, RC:$rd),
281 !strconcat(OpcStr, " $rd, [$addr]"),
282 [(OpNode Ty:$rd, ADDRri:$addr)]>;
285 //===----------------------------------------------------------------------===//
287 //===----------------------------------------------------------------------===//
289 // Pseudo instructions.
290 class Pseudo<dag outs, dag ins, string asmstr, list<dag> pattern>
291 : InstSP<outs, ins, asmstr, pattern> {
292 let isCodeGenOnly = 1;
298 def GETPCX : Pseudo<(outs getPCX:$getpcseq), (ins), "$getpcseq", [] >;
301 let Defs = [O6], Uses = [O6] in {
302 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
303 "!ADJCALLSTACKDOWN $amt",
304 [(callseq_start timm:$amt)]>;
305 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
306 "!ADJCALLSTACKUP $amt1",
307 [(callseq_end timm:$amt1, timm:$amt2)]>;
310 let hasSideEffects = 1, mayStore = 1 in {
311 let rd = 0, rs1 = 0, rs2 = 0 in
312 def FLUSHW : F3_1<0b10, 0b101011, (outs), (ins),
314 [(flushw)]>, Requires<[HasV9]>;
315 let rd = 0, rs1 = 1, simm13 = 3 in
316 def TA3 : F3_2<0b10, 0b111010, (outs), (ins),
321 let isBarrier = 1, isTerminator = 1, rd = 0b1000, rs1 = 0, simm13 = 5 in
322 def TA5 : F3_2<0b10, 0b111010, (outs), (ins), "ta 5", [(trap)]>;
325 def UNIMP : F2_1<0b000, (outs), (ins i32imm:$imm22),
328 // SELECT_CC_* - Used to implement the SELECT_CC DAG operation. Expanded after
329 // instruction selection into a branch sequence. This has to handle all
330 // permutations of selection between i32/f32/f64 on ICC and FCC.
331 // Expanded after instruction selection.
332 let Uses = [ICC], usesCustomInserter = 1 in {
333 def SELECT_CC_Int_ICC
334 : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, IntRegs:$F, i32imm:$Cond),
335 "; SELECT_CC_Int_ICC PSEUDO!",
336 [(set i32:$dst, (SPselecticc i32:$T, i32:$F, imm:$Cond))]>;
338 : Pseudo<(outs FPRegs:$dst), (ins FPRegs:$T, FPRegs:$F, i32imm:$Cond),
339 "; SELECT_CC_FP_ICC PSEUDO!",
340 [(set f32:$dst, (SPselecticc f32:$T, f32:$F, imm:$Cond))]>;
342 def SELECT_CC_DFP_ICC
343 : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$T, DFPRegs:$F, i32imm:$Cond),
344 "; SELECT_CC_DFP_ICC PSEUDO!",
345 [(set f64:$dst, (SPselecticc f64:$T, f64:$F, imm:$Cond))]>;
347 def SELECT_CC_QFP_ICC
348 : Pseudo<(outs QFPRegs:$dst), (ins QFPRegs:$T, QFPRegs:$F, i32imm:$Cond),
349 "; SELECT_CC_QFP_ICC PSEUDO!",
350 [(set f128:$dst, (SPselecticc f128:$T, f128:$F, imm:$Cond))]>;
353 let usesCustomInserter = 1, Uses = [FCC] in {
355 def SELECT_CC_Int_FCC
356 : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, IntRegs:$F, i32imm:$Cond),
357 "; SELECT_CC_Int_FCC PSEUDO!",
358 [(set i32:$dst, (SPselectfcc i32:$T, i32:$F, imm:$Cond))]>;
361 : Pseudo<(outs FPRegs:$dst), (ins FPRegs:$T, FPRegs:$F, i32imm:$Cond),
362 "; SELECT_CC_FP_FCC PSEUDO!",
363 [(set f32:$dst, (SPselectfcc f32:$T, f32:$F, imm:$Cond))]>;
364 def SELECT_CC_DFP_FCC
365 : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$T, DFPRegs:$F, i32imm:$Cond),
366 "; SELECT_CC_DFP_FCC PSEUDO!",
367 [(set f64:$dst, (SPselectfcc f64:$T, f64:$F, imm:$Cond))]>;
368 def SELECT_CC_QFP_FCC
369 : Pseudo<(outs QFPRegs:$dst), (ins QFPRegs:$T, QFPRegs:$F, i32imm:$Cond),
370 "; SELECT_CC_QFP_FCC PSEUDO!",
371 [(set f128:$dst, (SPselectfcc f128:$T, f128:$F, imm:$Cond))]>;
375 let isTerminator = 1, hasDelaySlot = 1, isBarrier = 1 in {
376 def JMPLrr: F3_1<2, 0b111000, (outs IntRegs:$dst), (ins MEMrr:$addr),
377 "jmpl $addr, $dst", []>;
378 def JMPLri: F3_2<2, 0b111000, (outs IntRegs:$dst), (ins MEMri:$addr),
379 "jmpl $addr, $dst", []>;
382 // Section A.3 - Synthetic Instructions, p. 85
383 // special cases of JMPL:
384 let isReturn = 1, isTerminator = 1, hasDelaySlot = 1, isBarrier = 1,
385 isCodeGenOnly = 1 in {
386 let rd = 0, rs1 = 15 in
387 def RETL: F3_2<2, 0b111000, (outs), (ins i32imm:$val),
388 "jmp %o7+$val", [(retflag simm13:$val)]>;
390 let rd = 0, rs1 = 31 in
391 def RET: F3_2<2, 0b111000, (outs), (ins i32imm:$val),
395 // Section B.1 - Load Integer Instructions, p. 90
396 let DecoderMethod = "DecodeLoadInt" in {
397 defm LDSB : Load<"ldsb", 0b001001, sextloadi8, IntRegs, i32>;
398 defm LDSH : Load<"ldsh", 0b001010, sextloadi16, IntRegs, i32>;
399 defm LDUB : Load<"ldub", 0b000001, zextloadi8, IntRegs, i32>;
400 defm LDUH : Load<"lduh", 0b000010, zextloadi16, IntRegs, i32>;
401 defm LD : Load<"ld", 0b000000, load, IntRegs, i32>;
404 // Section B.2 - Load Floating-point Instructions, p. 92
405 let DecoderMethod = "DecodeLoadFP" in
406 defm LDF : Load<"ld", 0b100000, load, FPRegs, f32>;
407 let DecoderMethod = "DecodeLoadDFP" in
408 defm LDDF : Load<"ldd", 0b100011, load, DFPRegs, f64>;
409 let DecoderMethod = "DecodeLoadQFP" in
410 defm LDQF : Load<"ldq", 0b100010, load, QFPRegs, f128>,
411 Requires<[HasV9, HasHardQuad]>;
413 // Section B.4 - Store Integer Instructions, p. 95
414 let DecoderMethod = "DecodeStoreInt" in {
415 defm STB : Store<"stb", 0b000101, truncstorei8, IntRegs, i32>;
416 defm STH : Store<"sth", 0b000110, truncstorei16, IntRegs, i32>;
417 defm ST : Store<"st", 0b000100, store, IntRegs, i32>;
420 // Section B.5 - Store Floating-point Instructions, p. 97
421 let DecoderMethod = "DecodeStoreFP" in
422 defm STF : Store<"st", 0b100100, store, FPRegs, f32>;
423 let DecoderMethod = "DecodeStoreDFP" in
424 defm STDF : Store<"std", 0b100111, store, DFPRegs, f64>;
425 let DecoderMethod = "DecodeStoreQFP" in
426 defm STQF : Store<"stq", 0b100110, store, QFPRegs, f128>,
427 Requires<[HasV9, HasHardQuad]>;
429 // Section B.9 - SETHI Instruction, p. 104
430 def SETHIi: F2_1<0b100,
431 (outs IntRegs:$rd), (ins i32imm:$imm22),
433 [(set i32:$rd, SETHIimm:$imm22)]>;
435 // Section B.10 - NOP Instruction, p. 105
436 // (It's a special case of SETHI)
437 let rd = 0, imm22 = 0 in
438 def NOP : F2_1<0b100, (outs), (ins), "nop", []>;
440 // Section B.11 - Logical Instructions, p. 106
441 defm AND : F3_12<"and", 0b000001, and, IntRegs, i32, simm13Op>;
443 def ANDNrr : F3_1<2, 0b000101,
444 (outs IntRegs:$rd), (ins IntRegs:$rs1, IntRegs:$rs2),
445 "andn $rs1, $rs2, $rd",
446 [(set i32:$rd, (and i32:$rs1, (not i32:$rs2)))]>;
447 def ANDNri : F3_2<2, 0b000101,
448 (outs IntRegs:$rd), (ins IntRegs:$rs1, simm13Op:$simm13),
449 "andn $rs1, $simm13, $rd", []>;
451 defm OR : F3_12<"or", 0b000010, or, IntRegs, i32, simm13Op>;
453 def ORNrr : F3_1<2, 0b000110,
454 (outs IntRegs:$rd), (ins IntRegs:$rs1, IntRegs:$rs2),
455 "orn $rs1, $rs2, $rd",
456 [(set i32:$rd, (or i32:$rs1, (not i32:$rs2)))]>;
457 def ORNri : F3_2<2, 0b000110,
458 (outs IntRegs:$rd), (ins IntRegs:$rs1, simm13Op:$simm13),
459 "orn $rs1, $simm13, $rd", []>;
460 defm XOR : F3_12<"xor", 0b000011, xor, IntRegs, i32, simm13Op>;
462 def XNORrr : F3_1<2, 0b000111,
463 (outs IntRegs:$rd), (ins IntRegs:$rs1, IntRegs:$rs2),
464 "xnor $rs1, $rs2, $rd",
465 [(set i32:$rd, (not (xor i32:$rs1, i32:$rs2)))]>;
466 def XNORri : F3_2<2, 0b000111,
467 (outs IntRegs:$rd), (ins IntRegs:$rs1, simm13Op:$simm13),
468 "xnor $rs1, $simm13, $rd", []>;
470 let Defs = [ICC] in {
471 defm ANDCC : F3_12np<"andcc", 0b010001>;
472 defm ANDNCC : F3_12np<"andncc", 0b010101>;
473 defm ORCC : F3_12np<"orcc", 0b010010>;
474 defm ORNCC : F3_12np<"orncc", 0b010110>;
475 defm XORCC : F3_12np<"xorcc", 0b010011>;
476 defm XNORCC : F3_12np<"xnorcc", 0b010111>;
479 // Section B.12 - Shift Instructions, p. 107
480 defm SLL : F3_12<"sll", 0b100101, shl, IntRegs, i32, simm13Op>;
481 defm SRL : F3_12<"srl", 0b100110, srl, IntRegs, i32, simm13Op>;
482 defm SRA : F3_12<"sra", 0b100111, sra, IntRegs, i32, simm13Op>;
484 // Section B.13 - Add Instructions, p. 108
485 defm ADD : F3_12<"add", 0b000000, add, IntRegs, i32, simm13Op>;
487 // "LEA" forms of add (patterns to make tblgen happy)
488 let Predicates = [Is32Bit], isCodeGenOnly = 1 in
489 def LEA_ADDri : F3_2<2, 0b000000,
490 (outs IntRegs:$dst), (ins MEMri:$addr),
491 "add ${addr:arith}, $dst",
492 [(set iPTR:$dst, ADDRri:$addr)]>;
495 defm ADDCC : F3_12<"addcc", 0b010000, addc, IntRegs, i32, simm13Op>;
498 defm ADDC : F3_12np<"addx", 0b001000>;
500 let Uses = [ICC], Defs = [ICC] in
501 defm ADDE : F3_12<"addxcc", 0b011000, adde, IntRegs, i32, simm13Op>;
503 // Section B.15 - Subtract Instructions, p. 110
504 defm SUB : F3_12 <"sub" , 0b000100, sub, IntRegs, i32, simm13Op>;
505 let Uses = [ICC], Defs = [ICC] in
506 defm SUBE : F3_12 <"subxcc" , 0b011100, sube, IntRegs, i32, simm13Op>;
509 defm SUBCC : F3_12 <"subcc", 0b010100, subc, IntRegs, i32, simm13Op>;
512 defm SUBC : F3_12np <"subx", 0b001100>;
514 let Defs = [ICC], rd = 0 in {
515 def CMPrr : F3_1<2, 0b010100,
516 (outs), (ins IntRegs:$rs1, IntRegs:$rs2),
518 [(SPcmpicc i32:$rs1, i32:$rs2)]>;
519 def CMPri : F3_2<2, 0b010100,
520 (outs), (ins IntRegs:$rs1, simm13Op:$simm13),
522 [(SPcmpicc i32:$rs1, (i32 simm13:$simm13))]>;
525 // Section B.18 - Multiply Instructions, p. 113
527 defm UMUL : F3_12np<"umul", 0b001010>;
528 defm SMUL : F3_12 <"smul", 0b001011, mul, IntRegs, i32, simm13Op>;
531 let Defs = [Y, ICC] in {
532 defm UMULCC : F3_12np<"umulcc", 0b011010>;
533 defm SMULCC : F3_12np<"smulcc", 0b011011>;
536 // Section B.19 - Divide Instructions, p. 115
538 defm UDIV : F3_12np<"udiv", 0b001110>;
539 defm SDIV : F3_12np<"sdiv", 0b001111>;
542 let Defs = [Y, ICC] in {
543 defm UDIVCC : F3_12np<"udivcc", 0b011110>;
544 defm SDIVCC : F3_12np<"sdivcc", 0b011111>;
547 // Section B.20 - SAVE and RESTORE, p. 117
548 defm SAVE : F3_12np<"save" , 0b111100>;
549 defm RESTORE : F3_12np<"restore", 0b111101>;
551 // Section B.21 - Branch on Integer Condition Codes Instructions, p. 119
553 // unconditional branch class.
554 class BranchAlways<dag ins, string asmstr, list<dag> pattern>
555 : F2_2<0b010, 0, (outs), ins, asmstr, pattern> {
557 let isTerminator = 1;
558 let hasDelaySlot = 1;
563 def BA : BranchAlways<(ins brtarget:$imm22), "ba $imm22", [(br bb:$imm22)]>;
565 // conditional branch class:
566 class BranchSP<dag ins, string asmstr, list<dag> pattern>
567 : F2_2<0b010, 0, (outs), ins, asmstr, pattern> {
569 let isTerminator = 1;
570 let hasDelaySlot = 1;
573 // conditional branch with annul class:
574 class BranchSPA<dag ins, string asmstr, list<dag> pattern>
575 : F2_2<0b010, 1, (outs), ins, asmstr, pattern> {
577 let isTerminator = 1;
578 let hasDelaySlot = 1;
581 // Indirect branch instructions.
582 let isTerminator = 1, isBarrier = 1, hasDelaySlot = 1, isBranch =1,
583 isIndirectBranch = 1, rd = 0, isCodeGenOnly = 1 in {
584 def BINDrr : F3_1<2, 0b111000,
585 (outs), (ins MEMrr:$ptr),
587 [(brind ADDRrr:$ptr)]>;
588 def BINDri : F3_2<2, 0b111000,
589 (outs), (ins MEMri:$ptr),
591 [(brind ADDRri:$ptr)]>;
594 let Uses = [ICC] in {
595 def BCOND : BranchSP<(ins brtarget:$imm22, CCOp:$cond),
597 [(SPbricc bb:$imm22, imm:$cond)]>;
598 def BCONDA : BranchSPA<(ins brtarget:$imm22, CCOp:$cond),
599 "b$cond,a $imm22", []>;
602 // Section B.22 - Branch on Floating-point Condition Codes Instructions, p. 121
604 // floating-point conditional branch class:
605 class FPBranchSP<dag ins, string asmstr, list<dag> pattern>
606 : F2_2<0b110, 0, (outs), ins, asmstr, pattern> {
608 let isTerminator = 1;
609 let hasDelaySlot = 1;
612 // floating-point conditional branch with annul class:
613 class FPBranchSPA<dag ins, string asmstr, list<dag> pattern>
614 : F2_2<0b110, 1, (outs), ins, asmstr, pattern> {
616 let isTerminator = 1;
617 let hasDelaySlot = 1;
620 let Uses = [FCC] in {
621 def FBCOND : FPBranchSP<(ins brtarget:$imm22, CCOp:$cond),
623 [(SPbrfcc bb:$imm22, imm:$cond)]>;
624 def FBCONDA : FPBranchSPA<(ins brtarget:$imm22, CCOp:$cond),
625 "fb$cond,a $imm22", []>;
628 // Section B.24 - Call and Link Instruction, p. 125
629 // This is the only Format 1 instruction
631 hasDelaySlot = 1, isCall = 1 in {
632 def CALL : InstSP<(outs), (ins calltarget:$disp, variable_ops),
636 let Inst{29-0} = disp;
639 // indirect calls: special cases of JMPL.
640 let isCodeGenOnly = 1, rd = 15 in {
641 def CALLrr : F3_1<2, 0b111000,
642 (outs), (ins MEMrr:$ptr, variable_ops),
644 [(call ADDRrr:$ptr)]>;
645 def CALLri : F3_2<2, 0b111000,
646 (outs), (ins MEMri:$ptr, variable_ops),
648 [(call ADDRri:$ptr)]>;
652 // Section B.28 - Read State Register Instructions
653 let Uses = [Y], rs1 = 0, rs2 = 0 in
654 def RDY : F3_1<2, 0b101000,
655 (outs IntRegs:$dst), (ins),
658 // Section B.29 - Write State Register Instructions
659 let Defs = [Y], rd = 0 in {
660 def WRYrr : F3_1<2, 0b110000,
661 (outs), (ins IntRegs:$rs1, IntRegs:$rs2),
662 "wr $rs1, $rs2, %y", []>;
663 def WRYri : F3_2<2, 0b110000,
664 (outs), (ins IntRegs:$rs1, simm13Op:$simm13),
665 "wr $rs1, $simm13, %y", []>;
667 // Convert Integer to Floating-point Instructions, p. 141
668 def FITOS : F3_3u<2, 0b110100, 0b011000100,
669 (outs FPRegs:$rd), (ins FPRegs:$rs2),
671 [(set FPRegs:$rd, (SPitof FPRegs:$rs2))]>;
672 def FITOD : F3_3u<2, 0b110100, 0b011001000,
673 (outs DFPRegs:$rd), (ins FPRegs:$rs2),
675 [(set DFPRegs:$rd, (SPitof FPRegs:$rs2))]>;
676 def FITOQ : F3_3u<2, 0b110100, 0b011001100,
677 (outs QFPRegs:$rd), (ins FPRegs:$rs2),
679 [(set QFPRegs:$rd, (SPitof FPRegs:$rs2))]>,
680 Requires<[HasHardQuad]>;
682 // Convert Floating-point to Integer Instructions, p. 142
683 def FSTOI : F3_3u<2, 0b110100, 0b011010001,
684 (outs FPRegs:$rd), (ins FPRegs:$rs2),
686 [(set FPRegs:$rd, (SPftoi FPRegs:$rs2))]>;
687 def FDTOI : F3_3u<2, 0b110100, 0b011010010,
688 (outs FPRegs:$rd), (ins DFPRegs:$rs2),
690 [(set FPRegs:$rd, (SPftoi DFPRegs:$rs2))]>;
691 def FQTOI : F3_3u<2, 0b110100, 0b011010011,
692 (outs FPRegs:$rd), (ins QFPRegs:$rs2),
694 [(set FPRegs:$rd, (SPftoi QFPRegs:$rs2))]>,
695 Requires<[HasHardQuad]>;
697 // Convert between Floating-point Formats Instructions, p. 143
698 def FSTOD : F3_3u<2, 0b110100, 0b011001001,
699 (outs DFPRegs:$rd), (ins FPRegs:$rs2),
701 [(set f64:$rd, (fextend f32:$rs2))]>;
702 def FSTOQ : F3_3u<2, 0b110100, 0b011001101,
703 (outs QFPRegs:$rd), (ins FPRegs:$rs2),
705 [(set f128:$rd, (fextend f32:$rs2))]>,
706 Requires<[HasHardQuad]>;
707 def FDTOS : F3_3u<2, 0b110100, 0b011000110,
708 (outs FPRegs:$rd), (ins DFPRegs:$rs2),
710 [(set f32:$rd, (fround f64:$rs2))]>;
711 def FDTOQ : F3_3u<2, 0b110100, 0b011001110,
712 (outs QFPRegs:$rd), (ins DFPRegs:$rs2),
714 [(set f128:$rd, (fextend f64:$rs2))]>,
715 Requires<[HasHardQuad]>;
716 def FQTOS : F3_3u<2, 0b110100, 0b011000111,
717 (outs FPRegs:$rd), (ins QFPRegs:$rs2),
719 [(set f32:$rd, (fround f128:$rs2))]>,
720 Requires<[HasHardQuad]>;
721 def FQTOD : F3_3u<2, 0b110100, 0b011001011,
722 (outs DFPRegs:$rd), (ins QFPRegs:$rs2),
724 [(set f64:$rd, (fround f128:$rs2))]>,
725 Requires<[HasHardQuad]>;
727 // Floating-point Move Instructions, p. 144
728 def FMOVS : F3_3u<2, 0b110100, 0b000000001,
729 (outs FPRegs:$rd), (ins FPRegs:$rs2),
730 "fmovs $rs2, $rd", []>;
731 def FNEGS : F3_3u<2, 0b110100, 0b000000101,
732 (outs FPRegs:$rd), (ins FPRegs:$rs2),
734 [(set f32:$rd, (fneg f32:$rs2))]>;
735 def FABSS : F3_3u<2, 0b110100, 0b000001001,
736 (outs FPRegs:$rd), (ins FPRegs:$rs2),
738 [(set f32:$rd, (fabs f32:$rs2))]>;
741 // Floating-point Square Root Instructions, p.145
742 def FSQRTS : F3_3u<2, 0b110100, 0b000101001,
743 (outs FPRegs:$rd), (ins FPRegs:$rs2),
745 [(set f32:$rd, (fsqrt f32:$rs2))]>;
746 def FSQRTD : F3_3u<2, 0b110100, 0b000101010,
747 (outs DFPRegs:$rd), (ins DFPRegs:$rs2),
749 [(set f64:$rd, (fsqrt f64:$rs2))]>;
750 def FSQRTQ : F3_3u<2, 0b110100, 0b000101011,
751 (outs QFPRegs:$rd), (ins QFPRegs:$rs2),
753 [(set f128:$rd, (fsqrt f128:$rs2))]>,
754 Requires<[HasHardQuad]>;
758 // Floating-point Add and Subtract Instructions, p. 146
759 def FADDS : F3_3<2, 0b110100, 0b001000001,
760 (outs FPRegs:$rd), (ins FPRegs:$rs1, FPRegs:$rs2),
761 "fadds $rs1, $rs2, $rd",
762 [(set f32:$rd, (fadd f32:$rs1, f32:$rs2))]>;
763 def FADDD : F3_3<2, 0b110100, 0b001000010,
764 (outs DFPRegs:$rd), (ins DFPRegs:$rs1, DFPRegs:$rs2),
765 "faddd $rs1, $rs2, $rd",
766 [(set f64:$rd, (fadd f64:$rs1, f64:$rs2))]>;
767 def FADDQ : F3_3<2, 0b110100, 0b001000011,
768 (outs QFPRegs:$rd), (ins QFPRegs:$rs1, QFPRegs:$rs2),
769 "faddq $rs1, $rs2, $rd",
770 [(set f128:$rd, (fadd f128:$rs1, f128:$rs2))]>,
771 Requires<[HasHardQuad]>;
773 def FSUBS : F3_3<2, 0b110100, 0b001000101,
774 (outs FPRegs:$rd), (ins FPRegs:$rs1, FPRegs:$rs2),
775 "fsubs $rs1, $rs2, $rd",
776 [(set f32:$rd, (fsub f32:$rs1, f32:$rs2))]>;
777 def FSUBD : F3_3<2, 0b110100, 0b001000110,
778 (outs DFPRegs:$rd), (ins DFPRegs:$rs1, DFPRegs:$rs2),
779 "fsubd $rs1, $rs2, $rd",
780 [(set f64:$rd, (fsub f64:$rs1, f64:$rs2))]>;
781 def FSUBQ : F3_3<2, 0b110100, 0b001000111,
782 (outs QFPRegs:$rd), (ins QFPRegs:$rs1, QFPRegs:$rs2),
783 "fsubq $rs1, $rs2, $rd",
784 [(set f128:$rd, (fsub f128:$rs1, f128:$rs2))]>,
785 Requires<[HasHardQuad]>;
788 // Floating-point Multiply and Divide Instructions, p. 147
789 def FMULS : F3_3<2, 0b110100, 0b001001001,
790 (outs FPRegs:$rd), (ins FPRegs:$rs1, FPRegs:$rs2),
791 "fmuls $rs1, $rs2, $rd",
792 [(set f32:$rd, (fmul f32:$rs1, f32:$rs2))]>;
793 def FMULD : F3_3<2, 0b110100, 0b001001010,
794 (outs DFPRegs:$rd), (ins DFPRegs:$rs1, DFPRegs:$rs2),
795 "fmuld $rs1, $rs2, $rd",
796 [(set f64:$rd, (fmul f64:$rs1, f64:$rs2))]>;
797 def FMULQ : F3_3<2, 0b110100, 0b001001011,
798 (outs QFPRegs:$rd), (ins QFPRegs:$rs1, QFPRegs:$rs2),
799 "fmulq $rs1, $rs2, $rd",
800 [(set f128:$rd, (fmul f128:$rs1, f128:$rs2))]>,
801 Requires<[HasHardQuad]>;
803 def FSMULD : F3_3<2, 0b110100, 0b001101001,
804 (outs DFPRegs:$rd), (ins FPRegs:$rs1, FPRegs:$rs2),
805 "fsmuld $rs1, $rs2, $rd",
806 [(set f64:$rd, (fmul (fextend f32:$rs1),
807 (fextend f32:$rs2)))]>;
808 def FDMULQ : F3_3<2, 0b110100, 0b001101110,
809 (outs QFPRegs:$rd), (ins DFPRegs:$rs1, DFPRegs:$rs2),
810 "fdmulq $rs1, $rs2, $rd",
811 [(set f128:$rd, (fmul (fextend f64:$rs1),
812 (fextend f64:$rs2)))]>,
813 Requires<[HasHardQuad]>;
815 def FDIVS : F3_3<2, 0b110100, 0b001001101,
816 (outs FPRegs:$rd), (ins FPRegs:$rs1, FPRegs:$rs2),
817 "fdivs $rs1, $rs2, $rd",
818 [(set f32:$rd, (fdiv f32:$rs1, f32:$rs2))]>;
819 def FDIVD : F3_3<2, 0b110100, 0b001001110,
820 (outs DFPRegs:$rd), (ins DFPRegs:$rs1, DFPRegs:$rs2),
821 "fdivd $rs1, $rs2, $rd",
822 [(set f64:$rd, (fdiv f64:$rs1, f64:$rs2))]>;
823 def FDIVQ : F3_3<2, 0b110100, 0b001001111,
824 (outs QFPRegs:$rd), (ins QFPRegs:$rs1, QFPRegs:$rs2),
825 "fdivq $rs1, $rs2, $rd",
826 [(set f128:$rd, (fdiv f128:$rs1, f128:$rs2))]>,
827 Requires<[HasHardQuad]>;
829 // Floating-point Compare Instructions, p. 148
830 // Note: the 2nd template arg is different for these guys.
831 // Note 2: the result of a FCMP is not available until the 2nd cycle
832 // after the instr is retired, but there is no interlock in Sparc V8.
833 // This behavior is modeled with a forced noop after the instruction in
836 let Defs = [FCC] in {
837 def FCMPS : F3_3c<2, 0b110101, 0b001010001,
838 (outs), (ins FPRegs:$rs1, FPRegs:$rs2),
840 [(SPcmpfcc f32:$rs1, f32:$rs2)]>;
841 def FCMPD : F3_3c<2, 0b110101, 0b001010010,
842 (outs), (ins DFPRegs:$rs1, DFPRegs:$rs2),
844 [(SPcmpfcc f64:$rs1, f64:$rs2)]>;
845 def FCMPQ : F3_3c<2, 0b110101, 0b001010011,
846 (outs), (ins QFPRegs:$rs1, QFPRegs:$rs2),
848 [(SPcmpfcc f128:$rs1, f128:$rs2)]>,
849 Requires<[HasHardQuad]>;
852 //===----------------------------------------------------------------------===//
853 // Instructions for Thread Local Storage(TLS).
854 //===----------------------------------------------------------------------===//
855 let isCodeGenOnly = 1, isAsmParserOnly = 1 in {
856 def TLS_ADDrr : F3_1<2, 0b000000,
858 (ins IntRegs:$rs1, IntRegs:$rs2, TLSSym:$sym),
859 "add $rs1, $rs2, $rd, $sym",
861 (tlsadd i32:$rs1, i32:$rs2, tglobaltlsaddr:$sym))]>;
864 def TLS_LDrr : F3_1<3, 0b000000,
865 (outs IntRegs:$dst), (ins MEMrr:$addr, TLSSym:$sym),
866 "ld [$addr], $dst, $sym",
868 (tlsld ADDRrr:$addr, tglobaltlsaddr:$sym))]>;
870 let Uses = [O6], isCall = 1, hasDelaySlot = 1 in
871 def TLS_CALL : InstSP<(outs),
872 (ins calltarget:$disp, TLSSym:$sym, variable_ops),
874 [(tlscall texternalsym:$disp, tglobaltlsaddr:$sym)]> {
877 let Inst{29-0} = disp;
881 //===----------------------------------------------------------------------===//
883 //===----------------------------------------------------------------------===//
885 // V9 Conditional Moves.
886 let Predicates = [HasV9], Constraints = "$f = $rd" in {
887 // Move Integer Register on Condition (MOVcc) p. 194 of the V9 manual.
888 let Uses = [ICC], cc = 0b100 in {
890 : F4_1<0b101100, (outs IntRegs:$rd),
891 (ins IntRegs:$rs2, IntRegs:$f, CCOp:$cond),
892 "mov$cond %icc, $rs2, $rd",
893 [(set i32:$rd, (SPselecticc i32:$rs2, i32:$f, imm:$cond))]>;
896 : F4_2<0b101100, (outs IntRegs:$rd),
897 (ins i32imm:$simm11, IntRegs:$f, CCOp:$cond),
898 "mov$cond %icc, $simm11, $rd",
900 (SPselecticc simm11:$simm11, i32:$f, imm:$cond))]>;
903 let Uses = [FCC], cc = 0b000 in {
905 : F4_1<0b101100, (outs IntRegs:$rd),
906 (ins IntRegs:$rs2, IntRegs:$f, CCOp:$cond),
907 "mov$cond %fcc0, $rs2, $rd",
908 [(set i32:$rd, (SPselectfcc i32:$rs2, i32:$f, imm:$cond))]>;
910 : F4_2<0b101100, (outs IntRegs:$rd),
911 (ins i32imm:$simm11, IntRegs:$f, CCOp:$cond),
912 "mov$cond %fcc0, $simm11, $rd",
914 (SPselectfcc simm11:$simm11, i32:$f, imm:$cond))]>;
917 let Uses = [ICC], opf_cc = 0b100 in {
919 : F4_3<0b110101, 0b000001, (outs FPRegs:$rd),
920 (ins FPRegs:$rs2, FPRegs:$f, CCOp:$cond),
921 "fmovs$cond %icc, $rs2, $rd",
922 [(set f32:$rd, (SPselecticc f32:$rs2, f32:$f, imm:$cond))]>;
924 : F4_3<0b110101, 0b000010, (outs DFPRegs:$rd),
925 (ins DFPRegs:$rs2, DFPRegs:$f, CCOp:$cond),
926 "fmovd$cond %icc, $rs2, $rd",
927 [(set f64:$rd, (SPselecticc f64:$rs2, f64:$f, imm:$cond))]>;
929 : F4_3<0b110101, 0b000011, (outs QFPRegs:$rd),
930 (ins QFPRegs:$rs2, QFPRegs:$f, CCOp:$cond),
931 "fmovq$cond %icc, $rs2, $rd",
932 [(set f128:$rd, (SPselecticc f128:$rs2, f128:$f, imm:$cond))]>,
933 Requires<[HasHardQuad]>;
936 let Uses = [FCC], opf_cc = 0b000 in {
938 : F4_3<0b110101, 0b000001, (outs FPRegs:$rd),
939 (ins FPRegs:$rs2, FPRegs:$f, CCOp:$cond),
940 "fmovs$cond %fcc0, $rs2, $rd",
941 [(set f32:$rd, (SPselectfcc f32:$rs2, f32:$f, imm:$cond))]>;
943 : F4_3<0b110101, 0b000010, (outs DFPRegs:$rd),
944 (ins DFPRegs:$rs2, DFPRegs:$f, CCOp:$cond),
945 "fmovd$cond %fcc0, $rs2, $rd",
946 [(set f64:$rd, (SPselectfcc f64:$rs2, f64:$f, imm:$cond))]>;
948 : F4_3<0b110101, 0b000011, (outs QFPRegs:$rd),
949 (ins QFPRegs:$rs2, QFPRegs:$f, CCOp:$cond),
950 "fmovq$cond %fcc0, $rs2, $rd",
951 [(set f128:$rd, (SPselectfcc f128:$rs2, f128:$f, imm:$cond))]>,
952 Requires<[HasHardQuad]>;
957 // Floating-Point Move Instructions, p. 164 of the V9 manual.
958 let Predicates = [HasV9] in {
959 def FMOVD : F3_3u<2, 0b110100, 0b000000010,
960 (outs DFPRegs:$rd), (ins DFPRegs:$rs2),
961 "fmovd $rs2, $rd", []>;
962 def FMOVQ : F3_3u<2, 0b110100, 0b000000011,
963 (outs QFPRegs:$rd), (ins QFPRegs:$rs2),
964 "fmovq $rs2, $rd", []>,
965 Requires<[HasHardQuad]>;
966 def FNEGD : F3_3u<2, 0b110100, 0b000000110,
967 (outs DFPRegs:$rd), (ins DFPRegs:$rs2),
969 [(set f64:$rd, (fneg f64:$rs2))]>;
970 def FNEGQ : F3_3u<2, 0b110100, 0b000000111,
971 (outs QFPRegs:$rd), (ins QFPRegs:$rs2),
973 [(set f128:$rd, (fneg f128:$rs2))]>,
974 Requires<[HasHardQuad]>;
975 def FABSD : F3_3u<2, 0b110100, 0b000001010,
976 (outs DFPRegs:$rd), (ins DFPRegs:$rs2),
978 [(set f64:$rd, (fabs f64:$rs2))]>;
979 def FABSQ : F3_3u<2, 0b110100, 0b000001011,
980 (outs QFPRegs:$rd), (ins QFPRegs:$rs2),
982 [(set f128:$rd, (fabs f128:$rs2))]>,
983 Requires<[HasHardQuad]>;
986 // POPCrr - This does a ctpop of a 64-bit register. As such, we have to clear
987 // the top 32-bits before using it. To do this clearing, we use a SRLri X,0.
989 def POPCrr : F3_1<2, 0b101110,
990 (outs IntRegs:$dst), (ins IntRegs:$src),
991 "popc $src, $dst", []>, Requires<[HasV9]>;
992 def : Pat<(ctpop i32:$src),
993 (POPCrr (SRLri $src, 0))>;
996 let hasSideEffects =1, rd = 0, rs1 = 0b01111, rs2 = 0 in
997 def STBAR : F3_1<2, 0b101000, (outs), (ins), "stbar", []>;
999 let Predicates = [HasV9], hasSideEffects = 1, rd = 0, rs1 = 0b01111 in
1000 def MEMBARi : F3_2<2, 0b101000, (outs), (ins simm13Op:$simm13),
1001 "membar $simm13", []>;
1003 let Constraints = "$val = $dst" in {
1004 def SWAPrr : F3_1<3, 0b001111,
1005 (outs IntRegs:$dst), (ins MEMrr:$addr, IntRegs:$val),
1006 "swap [$addr], $dst",
1007 [(set i32:$dst, (atomic_swap_32 ADDRrr:$addr, i32:$val))]>;
1008 def SWAPri : F3_2<3, 0b001111,
1009 (outs IntRegs:$dst), (ins MEMri:$addr, IntRegs:$val),
1010 "swap [$addr], $dst",
1011 [(set i32:$dst, (atomic_swap_32 ADDRri:$addr, i32:$val))]>;
1014 let Predicates = [HasV9], Constraints = "$swap = $rd" in
1015 def CASrr: F3_1_asi<3, 0b111100, 0b10000000,
1016 (outs IntRegs:$rd), (ins IntRegs:$rs1, IntRegs:$rs2,
1018 "cas [$rs1], $rs2, $rd",
1020 (atomic_cmp_swap iPTR:$rs1, i32:$rs2, i32:$swap))]>;
1022 let Defs = [ICC] in {
1023 defm TADDCC : F3_12np<"taddcc", 0b100000>;
1024 defm TSUBCC : F3_12np<"tsubcc", 0b100001>;
1026 let hasSideEffects = 1 in {
1027 defm TADDCCTV : F3_12np<"taddcctv", 0b100010>;
1028 defm TSUBCCTV : F3_12np<"tsubcctv", 0b100011>;
1032 //===----------------------------------------------------------------------===//
1033 // Non-Instruction Patterns
1034 //===----------------------------------------------------------------------===//
1036 // Small immediates.
1037 def : Pat<(i32 simm13:$val),
1038 (ORri (i32 G0), imm:$val)>;
1039 // Arbitrary immediates.
1040 def : Pat<(i32 imm:$val),
1041 (ORri (SETHIi (HI22 imm:$val)), (LO10 imm:$val))>;
1044 // Global addresses, constant pool entries
1045 let Predicates = [Is32Bit] in {
1047 def : Pat<(SPhi tglobaladdr:$in), (SETHIi tglobaladdr:$in)>;
1048 def : Pat<(SPlo tglobaladdr:$in), (ORri (i32 G0), tglobaladdr:$in)>;
1049 def : Pat<(SPhi tconstpool:$in), (SETHIi tconstpool:$in)>;
1050 def : Pat<(SPlo tconstpool:$in), (ORri (i32 G0), tconstpool:$in)>;
1052 // GlobalTLS addresses
1053 def : Pat<(SPhi tglobaltlsaddr:$in), (SETHIi tglobaltlsaddr:$in)>;
1054 def : Pat<(SPlo tglobaltlsaddr:$in), (ORri (i32 G0), tglobaltlsaddr:$in)>;
1055 def : Pat<(add (SPhi tglobaltlsaddr:$in1), (SPlo tglobaltlsaddr:$in2)),
1056 (ADDri (SETHIi tglobaltlsaddr:$in1), (tglobaltlsaddr:$in2))>;
1057 def : Pat<(xor (SPhi tglobaltlsaddr:$in1), (SPlo tglobaltlsaddr:$in2)),
1058 (XORri (SETHIi tglobaltlsaddr:$in1), (tglobaltlsaddr:$in2))>;
1061 def : Pat<(SPhi tblockaddress:$in), (SETHIi tblockaddress:$in)>;
1062 def : Pat<(SPlo tblockaddress:$in), (ORri (i32 G0), tblockaddress:$in)>;
1064 // Add reg, lo. This is used when taking the addr of a global/constpool entry.
1065 def : Pat<(add iPTR:$r, (SPlo tglobaladdr:$in)), (ADDri $r, tglobaladdr:$in)>;
1066 def : Pat<(add iPTR:$r, (SPlo tconstpool:$in)), (ADDri $r, tconstpool:$in)>;
1067 def : Pat<(add iPTR:$r, (SPlo tblockaddress:$in)),
1068 (ADDri $r, tblockaddress:$in)>;
1072 def : Pat<(call tglobaladdr:$dst),
1073 (CALL tglobaladdr:$dst)>;
1074 def : Pat<(call texternalsym:$dst),
1075 (CALL texternalsym:$dst)>;
1077 // Map integer extload's to zextloads.
1078 def : Pat<(i32 (extloadi1 ADDRrr:$src)), (LDUBrr ADDRrr:$src)>;
1079 def : Pat<(i32 (extloadi1 ADDRri:$src)), (LDUBri ADDRri:$src)>;
1080 def : Pat<(i32 (extloadi8 ADDRrr:$src)), (LDUBrr ADDRrr:$src)>;
1081 def : Pat<(i32 (extloadi8 ADDRri:$src)), (LDUBri ADDRri:$src)>;
1082 def : Pat<(i32 (extloadi16 ADDRrr:$src)), (LDUHrr ADDRrr:$src)>;
1083 def : Pat<(i32 (extloadi16 ADDRri:$src)), (LDUHri ADDRri:$src)>;
1085 // zextload bool -> zextload byte
1086 def : Pat<(i32 (zextloadi1 ADDRrr:$src)), (LDUBrr ADDRrr:$src)>;
1087 def : Pat<(i32 (zextloadi1 ADDRri:$src)), (LDUBri ADDRri:$src)>;
1089 // store 0, addr -> store %g0, addr
1090 def : Pat<(store (i32 0), ADDRrr:$dst), (STrr ADDRrr:$dst, (i32 G0))>;
1091 def : Pat<(store (i32 0), ADDRri:$dst), (STri ADDRri:$dst, (i32 G0))>;
1093 // store bar for all atomic_fence in V8.
1094 let Predicates = [HasNoV9] in
1095 def : Pat<(atomic_fence imm, imm), (STBAR)>;
1097 // atomic_load_32 addr -> load addr
1098 def : Pat<(i32 (atomic_load ADDRrr:$src)), (LDrr ADDRrr:$src)>;
1099 def : Pat<(i32 (atomic_load ADDRri:$src)), (LDri ADDRri:$src)>;
1101 // atomic_store_32 val, addr -> store val, addr
1102 def : Pat<(atomic_store ADDRrr:$dst, i32:$val), (STrr ADDRrr:$dst, $val)>;
1103 def : Pat<(atomic_store ADDRri:$dst, i32:$val), (STri ADDRri:$dst, $val)>;
1106 include "SparcInstr64Bit.td"
1107 include "SparcInstrAliases.td"