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()">,
42 AssemblerPredicate<"FeatureVIS">;
43 def HasVIS2 : Predicate<"Subtarget->isVIS2()">,
44 AssemblerPredicate<"FeatureVIS2">;
45 def HasVIS3 : Predicate<"Subtarget->isVIS3()">,
46 AssemblerPredicate<"FeatureVIS3">;
48 // HasHardQuad - This is true when the target processor supports quad floating
49 // point instructions.
50 def HasHardQuad : Predicate<"Subtarget->hasHardQuad()">;
52 // UseDeprecatedInsts - This predicate is true when the target processor is a
53 // V8, or when it is V9 but the V8 deprecated instructions are efficient enough
54 // to use when appropriate. In either of these cases, the instruction selector
55 // will pick deprecated instructions.
56 def UseDeprecatedInsts : Predicate<"Subtarget->useDeprecatedV8Instructions()">;
58 //===----------------------------------------------------------------------===//
59 // Instruction Pattern Stuff
60 //===----------------------------------------------------------------------===//
62 def simm11 : PatLeaf<(imm), [{ return isInt<11>(N->getSExtValue()); }]>;
64 def simm13 : PatLeaf<(imm), [{ return isInt<13>(N->getSExtValue()); }]>;
66 def LO10 : SDNodeXForm<imm, [{
67 return CurDAG->getTargetConstant((unsigned)N->getZExtValue() & 1023, SDLoc(N),
71 def HI22 : SDNodeXForm<imm, [{
72 // Transformation function: shift the immediate value down into the low bits.
73 return CurDAG->getTargetConstant((unsigned)N->getZExtValue() >> 10, SDLoc(N),
77 def SETHIimm : PatLeaf<(imm), [{
78 return isShiftedUInt<22, 10>(N->getZExtValue());
82 def ADDRrr : ComplexPattern<iPTR, 2, "SelectADDRrr", [], []>;
83 def ADDRri : ComplexPattern<iPTR, 2, "SelectADDRri", [frameindex], []>;
86 def SparcMEMrrAsmOperand : AsmOperandClass {
88 let ParserMethod = "parseMEMOperand";
91 def SparcMEMriAsmOperand : AsmOperandClass {
93 let ParserMethod = "parseMEMOperand";
96 def MEMrr : Operand<iPTR> {
97 let PrintMethod = "printMemOperand";
98 let MIOperandInfo = (ops ptr_rc, ptr_rc);
99 let ParserMatchClass = SparcMEMrrAsmOperand;
101 def MEMri : Operand<iPTR> {
102 let PrintMethod = "printMemOperand";
103 let MIOperandInfo = (ops ptr_rc, i32imm);
104 let ParserMatchClass = SparcMEMriAsmOperand;
107 def TLSSym : Operand<iPTR>;
109 // Branch targets have OtherVT type.
110 def brtarget : Operand<OtherVT> {
111 let EncoderMethod = "getBranchTargetOpValue";
114 def bprtarget : Operand<OtherVT> {
115 let EncoderMethod = "getBranchPredTargetOpValue";
118 def bprtarget16 : Operand<OtherVT> {
119 let EncoderMethod = "getBranchOnRegTargetOpValue";
122 def calltarget : Operand<i32> {
123 let EncoderMethod = "getCallTargetOpValue";
124 let DecoderMethod = "DecodeCall";
127 def simm13Op : Operand<i32> {
128 let DecoderMethod = "DecodeSIMM13";
131 // Operand for printing out a condition code.
132 let PrintMethod = "printCCOperand" in
133 def CCOp : Operand<i32>;
136 SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>]>;
138 SDTypeProfile<0, 2, [SDTCisFP<0>, SDTCisSameAs<0, 1>]>;
140 SDTypeProfile<0, 2, [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>]>;
142 SDTypeProfile<1, 3, [SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>, SDTCisVT<3, i32>]>;
144 SDTypeProfile<1, 1, [SDTCisVT<0, f32>, SDTCisFP<1>]>;
146 SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisVT<1, f32>]>;
148 SDTypeProfile<1, 1, [SDTCisVT<0, f64>, SDTCisFP<1>]>;
150 SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisVT<1, f64>]>;
153 SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>, SDTCisPtrTy<2>]>;
155 SDTypeProfile<1, 2, [SDTCisPtrTy<0>, SDTCisPtrTy<1>]>;
157 def SPcmpicc : SDNode<"SPISD::CMPICC", SDTSPcmpicc, [SDNPOutGlue]>;
158 def SPcmpfcc : SDNode<"SPISD::CMPFCC", SDTSPcmpfcc, [SDNPOutGlue]>;
159 def SPbricc : SDNode<"SPISD::BRICC", SDTSPbrcc, [SDNPHasChain, SDNPInGlue]>;
160 def SPbrxcc : SDNode<"SPISD::BRXCC", SDTSPbrcc, [SDNPHasChain, SDNPInGlue]>;
161 def SPbrfcc : SDNode<"SPISD::BRFCC", SDTSPbrcc, [SDNPHasChain, SDNPInGlue]>;
163 def SPhi : SDNode<"SPISD::Hi", SDTIntUnaryOp>;
164 def SPlo : SDNode<"SPISD::Lo", SDTIntUnaryOp>;
166 def SPftoi : SDNode<"SPISD::FTOI", SDTSPFTOI>;
167 def SPitof : SDNode<"SPISD::ITOF", SDTSPITOF>;
168 def SPftox : SDNode<"SPISD::FTOX", SDTSPFTOX>;
169 def SPxtof : SDNode<"SPISD::XTOF", SDTSPXTOF>;
171 def SPselecticc : SDNode<"SPISD::SELECT_ICC", SDTSPselectcc, [SDNPInGlue]>;
172 def SPselectxcc : SDNode<"SPISD::SELECT_XCC", SDTSPselectcc, [SDNPInGlue]>;
173 def SPselectfcc : SDNode<"SPISD::SELECT_FCC", SDTSPselectcc, [SDNPInGlue]>;
175 // These are target-independent nodes, but have target-specific formats.
176 def SDT_SPCallSeqStart : SDCallSeqStart<[ SDTCisVT<0, i32> ]>;
177 def SDT_SPCallSeqEnd : SDCallSeqEnd<[ SDTCisVT<0, i32>,
180 def callseq_start : SDNode<"ISD::CALLSEQ_START", SDT_SPCallSeqStart,
181 [SDNPHasChain, SDNPOutGlue]>;
182 def callseq_end : SDNode<"ISD::CALLSEQ_END", SDT_SPCallSeqEnd,
183 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
185 def SDT_SPCall : SDTypeProfile<0, -1, [SDTCisVT<0, i32>]>;
186 def call : SDNode<"SPISD::CALL", SDT_SPCall,
187 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
190 def SDT_SPRet : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
191 def retflag : SDNode<"SPISD::RET_FLAG", SDT_SPRet,
192 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
194 def flushw : SDNode<"SPISD::FLUSHW", SDTNone,
195 [SDNPHasChain, SDNPSideEffect, SDNPMayStore]>;
197 def tlsadd : SDNode<"SPISD::TLS_ADD", SDTSPtlsadd>;
198 def tlsld : SDNode<"SPISD::TLS_LD", SDTSPtlsld>;
199 def tlscall : SDNode<"SPISD::TLS_CALL", SDT_SPCall,
200 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
203 def getPCX : Operand<iPTR> {
204 let PrintMethod = "printGetPCX";
207 //===----------------------------------------------------------------------===//
208 // SPARC Flag Conditions
209 //===----------------------------------------------------------------------===//
211 // Note that these values must be kept in sync with the CCOp::CondCode enum
213 class ICC_VAL<int N> : PatLeaf<(i32 N)>;
214 def ICC_NE : ICC_VAL< 9>; // Not Equal
215 def ICC_E : ICC_VAL< 1>; // Equal
216 def ICC_G : ICC_VAL<10>; // Greater
217 def ICC_LE : ICC_VAL< 2>; // Less or Equal
218 def ICC_GE : ICC_VAL<11>; // Greater or Equal
219 def ICC_L : ICC_VAL< 3>; // Less
220 def ICC_GU : ICC_VAL<12>; // Greater Unsigned
221 def ICC_LEU : ICC_VAL< 4>; // Less or Equal Unsigned
222 def ICC_CC : ICC_VAL<13>; // Carry Clear/Great or Equal Unsigned
223 def ICC_CS : ICC_VAL< 5>; // Carry Set/Less Unsigned
224 def ICC_POS : ICC_VAL<14>; // Positive
225 def ICC_NEG : ICC_VAL< 6>; // Negative
226 def ICC_VC : ICC_VAL<15>; // Overflow Clear
227 def ICC_VS : ICC_VAL< 7>; // Overflow Set
229 class FCC_VAL<int N> : PatLeaf<(i32 N)>;
230 def FCC_U : FCC_VAL<23>; // Unordered
231 def FCC_G : FCC_VAL<22>; // Greater
232 def FCC_UG : FCC_VAL<21>; // Unordered or Greater
233 def FCC_L : FCC_VAL<20>; // Less
234 def FCC_UL : FCC_VAL<19>; // Unordered or Less
235 def FCC_LG : FCC_VAL<18>; // Less or Greater
236 def FCC_NE : FCC_VAL<17>; // Not Equal
237 def FCC_E : FCC_VAL<25>; // Equal
238 def FCC_UE : FCC_VAL<24>; // Unordered or Equal
239 def FCC_GE : FCC_VAL<25>; // Greater or Equal
240 def FCC_UGE : FCC_VAL<26>; // Unordered or Greater or Equal
241 def FCC_LE : FCC_VAL<27>; // Less or Equal
242 def FCC_ULE : FCC_VAL<28>; // Unordered or Less or Equal
243 def FCC_O : FCC_VAL<29>; // Ordered
245 //===----------------------------------------------------------------------===//
246 // Instruction Class Templates
247 //===----------------------------------------------------------------------===//
249 /// F3_12 multiclass - Define a normal F3_1/F3_2 pattern in one shot.
250 multiclass F3_12<string OpcStr, bits<6> Op3Val, SDNode OpNode,
251 RegisterClass RC, ValueType Ty, Operand immOp> {
252 def rr : F3_1<2, Op3Val,
253 (outs RC:$rd), (ins RC:$rs1, RC:$rs2),
254 !strconcat(OpcStr, " $rs1, $rs2, $rd"),
255 [(set Ty:$rd, (OpNode Ty:$rs1, Ty:$rs2))]>;
256 def ri : F3_2<2, Op3Val,
257 (outs RC:$rd), (ins RC:$rs1, immOp:$simm13),
258 !strconcat(OpcStr, " $rs1, $simm13, $rd"),
259 [(set Ty:$rd, (OpNode Ty:$rs1, (Ty simm13:$simm13)))]>;
262 /// F3_12np multiclass - Define a normal F3_1/F3_2 pattern in one shot, with no
264 multiclass F3_12np<string OpcStr, bits<6> Op3Val> {
265 def rr : F3_1<2, Op3Val,
266 (outs IntRegs:$rd), (ins IntRegs:$rs1, IntRegs:$rs2),
267 !strconcat(OpcStr, " $rs1, $rs2, $rd"), []>;
268 def ri : F3_2<2, Op3Val,
269 (outs IntRegs:$rd), (ins IntRegs:$rs1, simm13Op:$simm13),
270 !strconcat(OpcStr, " $rs1, $simm13, $rd"), []>;
273 // Load multiclass - Define both Reg+Reg/Reg+Imm patterns in one shot.
274 multiclass Load<string OpcStr, bits<6> Op3Val, SDPatternOperator OpNode,
275 RegisterClass RC, ValueType Ty> {
276 def rr : F3_1<3, Op3Val,
277 (outs RC:$dst), (ins MEMrr:$addr),
278 !strconcat(OpcStr, " [$addr], $dst"),
279 [(set Ty:$dst, (OpNode ADDRrr:$addr))]>;
280 def ri : F3_2<3, Op3Val,
281 (outs RC:$dst), (ins MEMri:$addr),
282 !strconcat(OpcStr, " [$addr], $dst"),
283 [(set Ty:$dst, (OpNode ADDRri:$addr))]>;
286 // TODO: Instructions of the LoadASI class are currently asm only; hooking up
287 // CodeGen's address spaces to use these is a future task.
288 class LoadASI<string OpcStr, bits<6> Op3Val, SDPatternOperator OpNode,
289 RegisterClass RC, ValueType Ty> :
290 F3_1_asi<3, Op3Val, (outs RC:$dst), (ins MEMrr:$addr, i8imm:$asi),
291 !strconcat(OpcStr, "a [$addr] $asi, $dst"),
294 // LoadA multiclass - As above, but also define alternate address space variant
295 multiclass LoadA<string OpcStr, bits<6> Op3Val, bits<6> LoadAOp3Val,
296 SDPatternOperator OpNode, RegisterClass RC, ValueType Ty> :
297 Load<OpcStr, Op3Val, OpNode, RC, Ty> {
298 def Arr : LoadASI<OpcStr, LoadAOp3Val, OpNode, RC, Ty>;
301 // The LDSTUB instruction is supported for asm only.
302 // It is unlikely that general-purpose code could make use of it.
303 // CAS is preferred for sparc v9.
304 def LDSTUBrr : F3_1<3, 0b001101, (outs IntRegs:$dst), (ins MEMrr:$addr),
305 "ldstub [$addr], $dst", []>;
306 def LDSTUBri : F3_2<3, 0b001101, (outs IntRegs:$dst), (ins MEMri:$addr),
307 "ldstub [$addr], $dst", []>;
308 def LDSTUBArr : F3_1_asi<3, 0b011101, (outs IntRegs:$dst),
309 (ins MEMrr:$addr, i8imm:$asi),
310 "ldstuba [$addr] $asi, $dst", []>;
312 // Store multiclass - Define both Reg+Reg/Reg+Imm patterns in one shot.
313 multiclass Store<string OpcStr, bits<6> Op3Val, SDPatternOperator OpNode,
314 RegisterClass RC, ValueType Ty> {
315 def rr : F3_1<3, Op3Val,
316 (outs), (ins MEMrr:$addr, RC:$rd),
317 !strconcat(OpcStr, " $rd, [$addr]"),
318 [(OpNode Ty:$rd, ADDRrr:$addr)]>;
319 def ri : F3_2<3, Op3Val,
320 (outs), (ins MEMri:$addr, RC:$rd),
321 !strconcat(OpcStr, " $rd, [$addr]"),
322 [(OpNode Ty:$rd, ADDRri:$addr)]>;
325 // TODO: Instructions of the StoreASI class are currently asm only; hooking up
326 // CodeGen's address spaces to use these is a future task.
327 class StoreASI<string OpcStr, bits<6> Op3Val,
328 SDPatternOperator OpNode, RegisterClass RC, ValueType Ty> :
329 F3_1_asi<3, Op3Val, (outs), (ins MEMrr:$addr, RC:$rd, i8imm:$asi),
330 !strconcat(OpcStr, "a $rd, [$addr] $asi"),
333 multiclass StoreA<string OpcStr, bits<6> Op3Val, bits<6> StoreAOp3Val,
334 SDPatternOperator OpNode, RegisterClass RC, ValueType Ty> :
335 Store<OpcStr, Op3Val, OpNode, RC, Ty> {
336 def Arr : StoreASI<OpcStr, StoreAOp3Val, OpNode, RC, Ty>;
339 //===----------------------------------------------------------------------===//
341 //===----------------------------------------------------------------------===//
343 // Pseudo instructions.
344 class Pseudo<dag outs, dag ins, string asmstr, list<dag> pattern>
345 : InstSP<outs, ins, asmstr, pattern> {
346 let isCodeGenOnly = 1;
352 def GETPCX : Pseudo<(outs getPCX:$getpcseq), (ins), "$getpcseq", [] >;
355 let Defs = [O6], Uses = [O6] in {
356 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
357 "!ADJCALLSTACKDOWN $amt",
358 [(callseq_start timm:$amt)]>;
359 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
360 "!ADJCALLSTACKUP $amt1",
361 [(callseq_end timm:$amt1, timm:$amt2)]>;
364 let hasSideEffects = 1, mayStore = 1 in {
365 let rd = 0, rs1 = 0, rs2 = 0 in
366 def FLUSHW : F3_1<0b10, 0b101011, (outs), (ins),
368 [(flushw)]>, Requires<[HasV9]>;
369 let rd = 0, rs1 = 1, simm13 = 3 in
370 def TA3 : F3_2<0b10, 0b111010, (outs), (ins),
375 // SELECT_CC_* - Used to implement the SELECT_CC DAG operation. Expanded after
376 // instruction selection into a branch sequence. This has to handle all
377 // permutations of selection between i32/f32/f64 on ICC and FCC.
378 // Expanded after instruction selection.
379 let Uses = [ICC], usesCustomInserter = 1 in {
380 def SELECT_CC_Int_ICC
381 : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, IntRegs:$F, i32imm:$Cond),
382 "; SELECT_CC_Int_ICC PSEUDO!",
383 [(set i32:$dst, (SPselecticc i32:$T, i32:$F, imm:$Cond))]>;
385 : Pseudo<(outs FPRegs:$dst), (ins FPRegs:$T, FPRegs:$F, i32imm:$Cond),
386 "; SELECT_CC_FP_ICC PSEUDO!",
387 [(set f32:$dst, (SPselecticc f32:$T, f32:$F, imm:$Cond))]>;
389 def SELECT_CC_DFP_ICC
390 : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$T, DFPRegs:$F, i32imm:$Cond),
391 "; SELECT_CC_DFP_ICC PSEUDO!",
392 [(set f64:$dst, (SPselecticc f64:$T, f64:$F, imm:$Cond))]>;
394 def SELECT_CC_QFP_ICC
395 : Pseudo<(outs QFPRegs:$dst), (ins QFPRegs:$T, QFPRegs:$F, i32imm:$Cond),
396 "; SELECT_CC_QFP_ICC PSEUDO!",
397 [(set f128:$dst, (SPselecticc f128:$T, f128:$F, imm:$Cond))]>;
400 let usesCustomInserter = 1, Uses = [FCC0] in {
402 def SELECT_CC_Int_FCC
403 : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, IntRegs:$F, i32imm:$Cond),
404 "; SELECT_CC_Int_FCC PSEUDO!",
405 [(set i32:$dst, (SPselectfcc i32:$T, i32:$F, imm:$Cond))]>;
408 : Pseudo<(outs FPRegs:$dst), (ins FPRegs:$T, FPRegs:$F, i32imm:$Cond),
409 "; SELECT_CC_FP_FCC PSEUDO!",
410 [(set f32:$dst, (SPselectfcc f32:$T, f32:$F, imm:$Cond))]>;
411 def SELECT_CC_DFP_FCC
412 : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$T, DFPRegs:$F, i32imm:$Cond),
413 "; SELECT_CC_DFP_FCC PSEUDO!",
414 [(set f64:$dst, (SPselectfcc f64:$T, f64:$F, imm:$Cond))]>;
415 def SELECT_CC_QFP_FCC
416 : Pseudo<(outs QFPRegs:$dst), (ins QFPRegs:$T, QFPRegs:$F, i32imm:$Cond),
417 "; SELECT_CC_QFP_FCC PSEUDO!",
418 [(set f128:$dst, (SPselectfcc f128:$T, f128:$F, imm:$Cond))]>;
421 // Section B.1 - Load Integer Instructions, p. 90
422 let DecoderMethod = "DecodeLoadInt" in {
423 defm LDSB : LoadA<"ldsb", 0b001001, 0b011001, sextloadi8, IntRegs, i32>;
424 defm LDSH : LoadA<"ldsh", 0b001010, 0b011010, sextloadi16, IntRegs, i32>;
425 defm LDUB : LoadA<"ldub", 0b000001, 0b010001, zextloadi8, IntRegs, i32>;
426 defm LDUH : LoadA<"lduh", 0b000010, 0b010010, zextloadi16, IntRegs, i32>;
427 defm LD : LoadA<"ld", 0b000000, 0b010000, load, IntRegs, i32>;
430 let DecoderMethod = "DecodeLoadIntPair" in
431 defm LDD : LoadA<"ldd", 0b000011, 0b010011, load, IntPair, v2i32>;
433 // Section B.2 - Load Floating-point Instructions, p. 92
434 let DecoderMethod = "DecodeLoadFP" in {
435 defm LDF : Load<"ld", 0b100000, load, FPRegs, f32>;
436 def LDFArr : LoadASI<"ld", 0b110000, load, FPRegs, f32>,
439 let DecoderMethod = "DecodeLoadDFP" in {
440 defm LDDF : Load<"ldd", 0b100011, load, DFPRegs, f64>;
441 def LDDFArr : LoadASI<"ldd", 0b110011, load, DFPRegs, f64>,
444 let DecoderMethod = "DecodeLoadQFP" in
445 defm LDQF : LoadA<"ldq", 0b100010, 0b110010, load, QFPRegs, f128>,
446 Requires<[HasV9, HasHardQuad]>;
448 let DecoderMethod = "DecodeLoadFP" in
449 let Defs = [FSR] in {
451 def LDFSRrr : F3_1<3, 0b100001, (outs), (ins MEMrr:$addr),
452 "ld [$addr], %fsr", []>;
453 def LDFSRri : F3_2<3, 0b100001, (outs), (ins MEMri:$addr),
454 "ld [$addr], %fsr", []>;
457 def LDXFSRrr : F3_1<3, 0b100001, (outs), (ins MEMrr:$addr),
458 "ldx [$addr], %fsr", []>, Requires<[HasV9]>;
459 def LDXFSRri : F3_2<3, 0b100001, (outs), (ins MEMri:$addr),
460 "ldx [$addr], %fsr", []>, Requires<[HasV9]>;
464 // Section B.4 - Store Integer Instructions, p. 95
465 let DecoderMethod = "DecodeStoreInt" in {
466 defm STB : StoreA<"stb", 0b000101, 0b010101, truncstorei8, IntRegs, i32>;
467 defm STH : StoreA<"sth", 0b000110, 0b010110, truncstorei16, IntRegs, i32>;
468 defm ST : StoreA<"st", 0b000100, 0b010100, store, IntRegs, i32>;
471 let DecoderMethod = "DecodeStoreIntPair" in
472 defm STD : StoreA<"std", 0b000111, 0b010111, store, IntPair, v2i32>;
474 // Section B.5 - Store Floating-point Instructions, p. 97
475 let DecoderMethod = "DecodeStoreFP" in {
476 defm STF : Store<"st", 0b100100, store, FPRegs, f32>;
477 def STFArr : StoreASI<"st", 0b110100, store, FPRegs, f32>,
480 let DecoderMethod = "DecodeStoreDFP" in {
481 defm STDF : Store<"std", 0b100111, store, DFPRegs, f64>;
482 def STDFArr : StoreASI<"std", 0b110111, store, DFPRegs, f64>,
485 let DecoderMethod = "DecodeStoreQFP" in
486 defm STQF : StoreA<"stq", 0b100110, 0b110110, store, QFPRegs, f128>,
487 Requires<[HasV9, HasHardQuad]>;
489 // Section B.8 - SWAP Register with Memory Instruction
491 let Constraints = "$val = $dst", DecoderMethod = "DecodeSWAP" in {
492 def SWAPrr : F3_1<3, 0b001111,
493 (outs IntRegs:$dst), (ins MEMrr:$addr, IntRegs:$val),
494 "swap [$addr], $dst",
495 [(set i32:$dst, (atomic_swap_32 ADDRrr:$addr, i32:$val))]>;
496 def SWAPri : F3_2<3, 0b001111,
497 (outs IntRegs:$dst), (ins MEMri:$addr, IntRegs:$val),
498 "swap [$addr], $dst",
499 [(set i32:$dst, (atomic_swap_32 ADDRri:$addr, i32:$val))]>;
500 def SWAPArr : F3_1_asi<3, 0b011111,
501 (outs IntRegs:$dst), (ins MEMrr:$addr, i8imm:$asi, IntRegs:$val),
502 "swapa [$addr] $asi, $dst",
503 [/*FIXME: pattern?*/]>;
507 // Section B.9 - SETHI Instruction, p. 104
508 def SETHIi: F2_1<0b100,
509 (outs IntRegs:$rd), (ins i32imm:$imm22),
511 [(set i32:$rd, SETHIimm:$imm22)]>;
513 // Section B.10 - NOP Instruction, p. 105
514 // (It's a special case of SETHI)
515 let rd = 0, imm22 = 0 in
516 def NOP : F2_1<0b100, (outs), (ins), "nop", []>;
518 // Section B.11 - Logical Instructions, p. 106
519 defm AND : F3_12<"and", 0b000001, and, IntRegs, i32, simm13Op>;
521 def ANDNrr : F3_1<2, 0b000101,
522 (outs IntRegs:$rd), (ins IntRegs:$rs1, IntRegs:$rs2),
523 "andn $rs1, $rs2, $rd",
524 [(set i32:$rd, (and i32:$rs1, (not i32:$rs2)))]>;
525 def ANDNri : F3_2<2, 0b000101,
526 (outs IntRegs:$rd), (ins IntRegs:$rs1, simm13Op:$simm13),
527 "andn $rs1, $simm13, $rd", []>;
529 defm OR : F3_12<"or", 0b000010, or, IntRegs, i32, simm13Op>;
531 def ORNrr : F3_1<2, 0b000110,
532 (outs IntRegs:$rd), (ins IntRegs:$rs1, IntRegs:$rs2),
533 "orn $rs1, $rs2, $rd",
534 [(set i32:$rd, (or i32:$rs1, (not i32:$rs2)))]>;
535 def ORNri : F3_2<2, 0b000110,
536 (outs IntRegs:$rd), (ins IntRegs:$rs1, simm13Op:$simm13),
537 "orn $rs1, $simm13, $rd", []>;
538 defm XOR : F3_12<"xor", 0b000011, xor, IntRegs, i32, simm13Op>;
540 def XNORrr : F3_1<2, 0b000111,
541 (outs IntRegs:$rd), (ins IntRegs:$rs1, IntRegs:$rs2),
542 "xnor $rs1, $rs2, $rd",
543 [(set i32:$rd, (not (xor i32:$rs1, i32:$rs2)))]>;
544 def XNORri : F3_2<2, 0b000111,
545 (outs IntRegs:$rd), (ins IntRegs:$rs1, simm13Op:$simm13),
546 "xnor $rs1, $simm13, $rd", []>;
548 let Defs = [ICC] in {
549 defm ANDCC : F3_12np<"andcc", 0b010001>;
550 defm ANDNCC : F3_12np<"andncc", 0b010101>;
551 defm ORCC : F3_12np<"orcc", 0b010010>;
552 defm ORNCC : F3_12np<"orncc", 0b010110>;
553 defm XORCC : F3_12np<"xorcc", 0b010011>;
554 defm XNORCC : F3_12np<"xnorcc", 0b010111>;
557 // Section B.12 - Shift Instructions, p. 107
558 defm SLL : F3_12<"sll", 0b100101, shl, IntRegs, i32, simm13Op>;
559 defm SRL : F3_12<"srl", 0b100110, srl, IntRegs, i32, simm13Op>;
560 defm SRA : F3_12<"sra", 0b100111, sra, IntRegs, i32, simm13Op>;
562 // Section B.13 - Add Instructions, p. 108
563 defm ADD : F3_12<"add", 0b000000, add, IntRegs, i32, simm13Op>;
565 // "LEA" forms of add (patterns to make tblgen happy)
566 let Predicates = [Is32Bit], isCodeGenOnly = 1 in
567 def LEA_ADDri : F3_2<2, 0b000000,
568 (outs IntRegs:$dst), (ins MEMri:$addr),
569 "add ${addr:arith}, $dst",
570 [(set iPTR:$dst, ADDRri:$addr)]>;
573 defm ADDCC : F3_12<"addcc", 0b010000, addc, IntRegs, i32, simm13Op>;
576 defm ADDC : F3_12np<"addx", 0b001000>;
578 let Uses = [ICC], Defs = [ICC] in
579 defm ADDE : F3_12<"addxcc", 0b011000, adde, IntRegs, i32, simm13Op>;
581 // Section B.15 - Subtract Instructions, p. 110
582 defm SUB : F3_12 <"sub" , 0b000100, sub, IntRegs, i32, simm13Op>;
583 let Uses = [ICC], Defs = [ICC] in
584 defm SUBE : F3_12 <"subxcc" , 0b011100, sube, IntRegs, i32, simm13Op>;
587 defm SUBCC : F3_12 <"subcc", 0b010100, subc, IntRegs, i32, simm13Op>;
590 defm SUBC : F3_12np <"subx", 0b001100>;
592 // cmp (from Section A.3) is a specialized alias for subcc
593 let Defs = [ICC], rd = 0 in {
594 def CMPrr : F3_1<2, 0b010100,
595 (outs), (ins IntRegs:$rs1, IntRegs:$rs2),
597 [(SPcmpicc i32:$rs1, i32:$rs2)]>;
598 def CMPri : F3_2<2, 0b010100,
599 (outs), (ins IntRegs:$rs1, simm13Op:$simm13),
601 [(SPcmpicc i32:$rs1, (i32 simm13:$simm13))]>;
604 // Section B.18 - Multiply Instructions, p. 113
606 defm UMUL : F3_12np<"umul", 0b001010>;
607 defm SMUL : F3_12 <"smul", 0b001011, mul, IntRegs, i32, simm13Op>;
610 let Defs = [Y, ICC] in {
611 defm UMULCC : F3_12np<"umulcc", 0b011010>;
612 defm SMULCC : F3_12np<"smulcc", 0b011011>;
615 // Section B.19 - Divide Instructions, p. 115
616 let Uses = [Y], Defs = [Y] in {
617 defm UDIV : F3_12np<"udiv", 0b001110>;
618 defm SDIV : F3_12np<"sdiv", 0b001111>;
621 let Uses = [Y], Defs = [Y, ICC] in {
622 defm UDIVCC : F3_12np<"udivcc", 0b011110>;
623 defm SDIVCC : F3_12np<"sdivcc", 0b011111>;
626 // Section B.20 - SAVE and RESTORE, p. 117
627 defm SAVE : F3_12np<"save" , 0b111100>;
628 defm RESTORE : F3_12np<"restore", 0b111101>;
630 // Section B.21 - Branch on Integer Condition Codes Instructions, p. 119
632 // unconditional branch class.
633 class BranchAlways<dag ins, string asmstr, list<dag> pattern>
634 : F2_2<0b010, 0, (outs), ins, asmstr, pattern> {
636 let isTerminator = 1;
637 let hasDelaySlot = 1;
642 def BA : BranchAlways<(ins brtarget:$imm22), "ba $imm22", [(br bb:$imm22)]>;
645 let isBranch = 1, isTerminator = 1, hasDelaySlot = 1 in {
647 // conditional branch class:
648 class BranchSP<dag ins, string asmstr, list<dag> pattern>
649 : F2_2<0b010, 0, (outs), ins, asmstr, pattern>;
651 // conditional branch with annul class:
652 class BranchSPA<dag ins, string asmstr, list<dag> pattern>
653 : F2_2<0b010, 1, (outs), ins, asmstr, pattern>;
655 // Conditional branch class on %icc|%xcc with predication:
656 multiclass IPredBranch<string regstr, list<dag> CCPattern> {
657 def CC : F2_3<0b001, 0, 1, (outs), (ins bprtarget:$imm19, CCOp:$cond),
658 !strconcat("b$cond ", !strconcat(regstr, ", $imm19")),
660 def CCA : F2_3<0b001, 1, 1, (outs), (ins bprtarget:$imm19, CCOp:$cond),
661 !strconcat("b$cond,a ", !strconcat(regstr, ", $imm19")),
663 def CCNT : F2_3<0b001, 0, 0, (outs), (ins bprtarget:$imm19, CCOp:$cond),
664 !strconcat("b$cond,pn ", !strconcat(regstr, ", $imm19")),
666 def CCANT : F2_3<0b001, 1, 0, (outs), (ins bprtarget:$imm19, CCOp:$cond),
667 !strconcat("b$cond,a,pn ", !strconcat(regstr, ", $imm19")),
671 } // let isBranch = 1, isTerminator = 1, hasDelaySlot = 1
674 // Indirect branch instructions.
675 let isTerminator = 1, isBarrier = 1, hasDelaySlot = 1, isBranch =1,
676 isIndirectBranch = 1, rd = 0, isCodeGenOnly = 1 in {
677 def BINDrr : F3_1<2, 0b111000,
678 (outs), (ins MEMrr:$ptr),
680 [(brind ADDRrr:$ptr)]>;
681 def BINDri : F3_2<2, 0b111000,
682 (outs), (ins MEMri:$ptr),
684 [(brind ADDRri:$ptr)]>;
687 let Uses = [ICC] in {
688 def BCOND : BranchSP<(ins brtarget:$imm22, CCOp:$cond),
690 [(SPbricc bb:$imm22, imm:$cond)]>;
691 def BCONDA : BranchSPA<(ins brtarget:$imm22, CCOp:$cond),
692 "b$cond,a $imm22", []>;
694 let Predicates = [HasV9], cc = 0b00 in
695 defm BPI : IPredBranch<"%icc", []>;
698 // Section B.22 - Branch on Floating-point Condition Codes Instructions, p. 121
700 let isBranch = 1, isTerminator = 1, hasDelaySlot = 1 in {
702 // floating-point conditional branch class:
703 class FPBranchSP<dag ins, string asmstr, list<dag> pattern>
704 : F2_2<0b110, 0, (outs), ins, asmstr, pattern>;
706 // floating-point conditional branch with annul class:
707 class FPBranchSPA<dag ins, string asmstr, list<dag> pattern>
708 : F2_2<0b110, 1, (outs), ins, asmstr, pattern>;
710 // Conditional branch class on %fcc0-%fcc3 with predication:
711 multiclass FPredBranch {
712 def CC : F2_3<0b101, 0, 1, (outs), (ins bprtarget:$imm19, CCOp:$cond,
714 "fb$cond $cc, $imm19", []>;
715 def CCA : F2_3<0b101, 1, 1, (outs), (ins bprtarget:$imm19, CCOp:$cond,
717 "fb$cond,a $cc, $imm19", []>;
718 def CCNT : F2_3<0b101, 0, 0, (outs), (ins bprtarget:$imm19, CCOp:$cond,
720 "fb$cond,pn $cc, $imm19", []>;
721 def CCANT : F2_3<0b101, 1, 0, (outs), (ins bprtarget:$imm19, CCOp:$cond,
723 "fb$cond,a,pn $cc, $imm19", []>;
725 } // let isBranch = 1, isTerminator = 1, hasDelaySlot = 1
727 let Uses = [FCC0] in {
728 def FBCOND : FPBranchSP<(ins brtarget:$imm22, CCOp:$cond),
730 [(SPbrfcc bb:$imm22, imm:$cond)]>;
731 def FBCONDA : FPBranchSPA<(ins brtarget:$imm22, CCOp:$cond),
732 "fb$cond,a $imm22", []>;
735 let Predicates = [HasV9] in
736 defm BPF : FPredBranch;
739 // Section B.24 - Call and Link Instruction, p. 125
740 // This is the only Format 1 instruction
742 hasDelaySlot = 1, isCall = 1 in {
743 def CALL : InstSP<(outs), (ins calltarget:$disp, variable_ops),
747 let Inst{29-0} = disp;
750 // indirect calls: special cases of JMPL.
751 let isCodeGenOnly = 1, rd = 15 in {
752 def CALLrr : F3_1<2, 0b111000,
753 (outs), (ins MEMrr:$ptr, variable_ops),
755 [(call ADDRrr:$ptr)]>;
756 def CALLri : F3_2<2, 0b111000,
757 (outs), (ins MEMri:$ptr, variable_ops),
759 [(call ADDRri:$ptr)]>;
763 // Section B.25 - Jump and Link Instruction
766 let isTerminator = 1, hasDelaySlot = 1, isBarrier = 1,
767 DecoderMethod = "DecodeJMPL" in {
768 def JMPLrr: F3_1<2, 0b111000, (outs IntRegs:$dst), (ins MEMrr:$addr),
769 "jmpl $addr, $dst", []>;
770 def JMPLri: F3_2<2, 0b111000, (outs IntRegs:$dst), (ins MEMri:$addr),
771 "jmpl $addr, $dst", []>;
774 // Section A.3 - Synthetic Instructions, p. 85
775 // special cases of JMPL:
776 let isReturn = 1, isTerminator = 1, hasDelaySlot = 1, isBarrier = 1,
777 isCodeGenOnly = 1 in {
778 let rd = 0, rs1 = 15 in
779 def RETL: F3_2<2, 0b111000, (outs), (ins i32imm:$val),
780 "jmp %o7+$val", [(retflag simm13:$val)]>;
782 let rd = 0, rs1 = 31 in
783 def RET: F3_2<2, 0b111000, (outs), (ins i32imm:$val),
787 // Section B.26 - Return from Trap Instruction
788 let isReturn = 1, isTerminator = 1, hasDelaySlot = 1,
789 isBarrier = 1, rd = 0, DecoderMethod = "DecodeReturn" in {
790 def RETTrr : F3_1<2, 0b111001, (outs), (ins MEMrr:$addr),
792 def RETTri : F3_2<2, 0b111001, (outs), (ins MEMri:$addr),
797 // Section B.27 - Trap on Integer Condition Codes Instruction
798 multiclass TRAP<string regStr> {
799 def rr : TRAPSPrr<0b111010, (outs), (ins IntRegs:$rs1, IntRegs:$rs2,
801 !strconcat(!strconcat("t$cond ", regStr), ", $rs1 + $rs2"), []>;
802 def ri : TRAPSPri<0b111010, (outs), (ins IntRegs:$rs1, i32imm:$imm,
804 !strconcat(!strconcat("t$cond ", regStr), ", $rs1 + $imm"), []>;
807 let hasSideEffects = 1, Uses = [ICC], cc = 0b00 in
808 defm TICC : TRAP<"%icc">;
810 let isBarrier = 1, isTerminator = 1, rd = 0b01000, rs1 = 0, simm13 = 5 in
811 def TA5 : F3_2<0b10, 0b111010, (outs), (ins), "ta 5", [(trap)]>;
813 // Section B.28 - Read State Register Instructions
815 def RDASR : F3_1<2, 0b101000,
816 (outs IntRegs:$rd), (ins ASRRegs:$rs1),
819 // PSR, WIM, and TBR don't exist on the SparcV9, only the V8.
820 let Predicates = [HasNoV9] in {
821 let rs2 = 0, rs1 = 0, Uses=[PSR] in
822 def RDPSR : F3_1<2, 0b101001,
823 (outs IntRegs:$rd), (ins),
826 let rs2 = 0, rs1 = 0, Uses=[WIM] in
827 def RDWIM : F3_1<2, 0b101010,
828 (outs IntRegs:$rd), (ins),
831 let rs2 = 0, rs1 = 0, Uses=[TBR] in
832 def RDTBR : F3_1<2, 0b101011,
833 (outs IntRegs:$rd), (ins),
837 // Section B.29 - Write State Register Instructions
838 def WRASRrr : F3_1<2, 0b110000,
839 (outs ASRRegs:$rd), (ins IntRegs:$rs1, IntRegs:$rs2),
840 "wr $rs1, $rs2, $rd", []>;
841 def WRASRri : F3_2<2, 0b110000,
842 (outs ASRRegs:$rd), (ins IntRegs:$rs1, simm13Op:$simm13),
843 "wr $rs1, $simm13, $rd", []>;
845 // PSR, WIM, and TBR don't exist on the SparcV9, only the V8.
846 let Predicates = [HasNoV9] in {
847 let Defs = [PSR], rd=0 in {
848 def WRPSRrr : F3_1<2, 0b110001,
849 (outs), (ins IntRegs:$rs1, IntRegs:$rs2),
850 "wr $rs1, $rs2, %psr", []>;
851 def WRPSRri : F3_2<2, 0b110001,
852 (outs), (ins IntRegs:$rs1, simm13Op:$simm13),
853 "wr $rs1, $simm13, %psr", []>;
856 let Defs = [WIM], rd=0 in {
857 def WRWIMrr : F3_1<2, 0b110010,
858 (outs), (ins IntRegs:$rs1, IntRegs:$rs2),
859 "wr $rs1, $rs2, %wim", []>;
860 def WRWIMri : F3_2<2, 0b110010,
861 (outs), (ins IntRegs:$rs1, simm13Op:$simm13),
862 "wr $rs1, $simm13, %wim", []>;
865 let Defs = [TBR], rd=0 in {
866 def WRTBRrr : F3_1<2, 0b110011,
867 (outs), (ins IntRegs:$rs1, IntRegs:$rs2),
868 "wr $rs1, $rs2, %tbr", []>;
869 def WRTBRri : F3_2<2, 0b110011,
870 (outs), (ins IntRegs:$rs1, simm13Op:$simm13),
871 "wr $rs1, $simm13, %tbr", []>;
875 // Section B.30 - STBAR Instruction
876 let hasSideEffects = 1, rd = 0, rs1 = 0b01111, rs2 = 0 in
877 def STBAR : F3_1<2, 0b101000, (outs), (ins), "stbar", []>;
880 // Section B.31 - Unimplmented Instruction
882 def UNIMP : F2_1<0b000, (outs), (ins i32imm:$imm22),
885 // Section B.32 - Flush Instruction Memory
887 def FLUSHrr : F3_1<2, 0b111011, (outs), (ins MEMrr:$addr),
889 def FLUSHri : F3_2<2, 0b111011, (outs), (ins MEMri:$addr),
892 // The no-arg FLUSH is only here for the benefit of the InstAlias
893 // "flush", which cannot seem to use FLUSHrr, due to the inability
894 // to construct a MEMrr with fixed G0 registers.
895 let rs1 = 0, rs2 = 0 in
896 def FLUSH : F3_1<2, 0b111011, (outs), (ins), "flush %g0", []>;
899 // Section B.33 - Floating-point Operate (FPop) Instructions
901 // Convert Integer to Floating-point Instructions, p. 141
902 def FITOS : F3_3u<2, 0b110100, 0b011000100,
903 (outs FPRegs:$rd), (ins FPRegs:$rs2),
905 [(set FPRegs:$rd, (SPitof FPRegs:$rs2))]>;
906 def FITOD : F3_3u<2, 0b110100, 0b011001000,
907 (outs DFPRegs:$rd), (ins FPRegs:$rs2),
909 [(set DFPRegs:$rd, (SPitof FPRegs:$rs2))]>;
910 def FITOQ : F3_3u<2, 0b110100, 0b011001100,
911 (outs QFPRegs:$rd), (ins FPRegs:$rs2),
913 [(set QFPRegs:$rd, (SPitof FPRegs:$rs2))]>,
914 Requires<[HasHardQuad]>;
916 // Convert Floating-point to Integer Instructions, p. 142
917 def FSTOI : F3_3u<2, 0b110100, 0b011010001,
918 (outs FPRegs:$rd), (ins FPRegs:$rs2),
920 [(set FPRegs:$rd, (SPftoi FPRegs:$rs2))]>;
921 def FDTOI : F3_3u<2, 0b110100, 0b011010010,
922 (outs FPRegs:$rd), (ins DFPRegs:$rs2),
924 [(set FPRegs:$rd, (SPftoi DFPRegs:$rs2))]>;
925 def FQTOI : F3_3u<2, 0b110100, 0b011010011,
926 (outs FPRegs:$rd), (ins QFPRegs:$rs2),
928 [(set FPRegs:$rd, (SPftoi QFPRegs:$rs2))]>,
929 Requires<[HasHardQuad]>;
931 // Convert between Floating-point Formats Instructions, p. 143
932 def FSTOD : F3_3u<2, 0b110100, 0b011001001,
933 (outs DFPRegs:$rd), (ins FPRegs:$rs2),
935 [(set f64:$rd, (fextend f32:$rs2))]>;
936 def FSTOQ : F3_3u<2, 0b110100, 0b011001101,
937 (outs QFPRegs:$rd), (ins FPRegs:$rs2),
939 [(set f128:$rd, (fextend f32:$rs2))]>,
940 Requires<[HasHardQuad]>;
941 def FDTOS : F3_3u<2, 0b110100, 0b011000110,
942 (outs FPRegs:$rd), (ins DFPRegs:$rs2),
944 [(set f32:$rd, (fround f64:$rs2))]>;
945 def FDTOQ : F3_3u<2, 0b110100, 0b011001110,
946 (outs QFPRegs:$rd), (ins DFPRegs:$rs2),
948 [(set f128:$rd, (fextend f64:$rs2))]>,
949 Requires<[HasHardQuad]>;
950 def FQTOS : F3_3u<2, 0b110100, 0b011000111,
951 (outs FPRegs:$rd), (ins QFPRegs:$rs2),
953 [(set f32:$rd, (fround f128:$rs2))]>,
954 Requires<[HasHardQuad]>;
955 def FQTOD : F3_3u<2, 0b110100, 0b011001011,
956 (outs DFPRegs:$rd), (ins QFPRegs:$rs2),
958 [(set f64:$rd, (fround f128:$rs2))]>,
959 Requires<[HasHardQuad]>;
961 // Floating-point Move Instructions, p. 144
962 def FMOVS : F3_3u<2, 0b110100, 0b000000001,
963 (outs FPRegs:$rd), (ins FPRegs:$rs2),
964 "fmovs $rs2, $rd", []>;
965 def FNEGS : F3_3u<2, 0b110100, 0b000000101,
966 (outs FPRegs:$rd), (ins FPRegs:$rs2),
968 [(set f32:$rd, (fneg f32:$rs2))]>;
969 def FABSS : F3_3u<2, 0b110100, 0b000001001,
970 (outs FPRegs:$rd), (ins FPRegs:$rs2),
972 [(set f32:$rd, (fabs f32:$rs2))]>;
975 // Floating-point Square Root Instructions, p.145
976 def FSQRTS : F3_3u<2, 0b110100, 0b000101001,
977 (outs FPRegs:$rd), (ins FPRegs:$rs2),
979 [(set f32:$rd, (fsqrt f32:$rs2))]>;
980 def FSQRTD : F3_3u<2, 0b110100, 0b000101010,
981 (outs DFPRegs:$rd), (ins DFPRegs:$rs2),
983 [(set f64:$rd, (fsqrt f64:$rs2))]>;
984 def FSQRTQ : F3_3u<2, 0b110100, 0b000101011,
985 (outs QFPRegs:$rd), (ins QFPRegs:$rs2),
987 [(set f128:$rd, (fsqrt f128:$rs2))]>,
988 Requires<[HasHardQuad]>;
992 // Floating-point Add and Subtract Instructions, p. 146
993 def FADDS : F3_3<2, 0b110100, 0b001000001,
994 (outs FPRegs:$rd), (ins FPRegs:$rs1, FPRegs:$rs2),
995 "fadds $rs1, $rs2, $rd",
996 [(set f32:$rd, (fadd f32:$rs1, f32:$rs2))]>;
997 def FADDD : F3_3<2, 0b110100, 0b001000010,
998 (outs DFPRegs:$rd), (ins DFPRegs:$rs1, DFPRegs:$rs2),
999 "faddd $rs1, $rs2, $rd",
1000 [(set f64:$rd, (fadd f64:$rs1, f64:$rs2))]>;
1001 def FADDQ : F3_3<2, 0b110100, 0b001000011,
1002 (outs QFPRegs:$rd), (ins QFPRegs:$rs1, QFPRegs:$rs2),
1003 "faddq $rs1, $rs2, $rd",
1004 [(set f128:$rd, (fadd f128:$rs1, f128:$rs2))]>,
1005 Requires<[HasHardQuad]>;
1007 def FSUBS : F3_3<2, 0b110100, 0b001000101,
1008 (outs FPRegs:$rd), (ins FPRegs:$rs1, FPRegs:$rs2),
1009 "fsubs $rs1, $rs2, $rd",
1010 [(set f32:$rd, (fsub f32:$rs1, f32:$rs2))]>;
1011 def FSUBD : F3_3<2, 0b110100, 0b001000110,
1012 (outs DFPRegs:$rd), (ins DFPRegs:$rs1, DFPRegs:$rs2),
1013 "fsubd $rs1, $rs2, $rd",
1014 [(set f64:$rd, (fsub f64:$rs1, f64:$rs2))]>;
1015 def FSUBQ : F3_3<2, 0b110100, 0b001000111,
1016 (outs QFPRegs:$rd), (ins QFPRegs:$rs1, QFPRegs:$rs2),
1017 "fsubq $rs1, $rs2, $rd",
1018 [(set f128:$rd, (fsub f128:$rs1, f128:$rs2))]>,
1019 Requires<[HasHardQuad]>;
1022 // Floating-point Multiply and Divide Instructions, p. 147
1023 def FMULS : F3_3<2, 0b110100, 0b001001001,
1024 (outs FPRegs:$rd), (ins FPRegs:$rs1, FPRegs:$rs2),
1025 "fmuls $rs1, $rs2, $rd",
1026 [(set f32:$rd, (fmul f32:$rs1, f32:$rs2))]>;
1027 def FMULD : F3_3<2, 0b110100, 0b001001010,
1028 (outs DFPRegs:$rd), (ins DFPRegs:$rs1, DFPRegs:$rs2),
1029 "fmuld $rs1, $rs2, $rd",
1030 [(set f64:$rd, (fmul f64:$rs1, f64:$rs2))]>;
1031 def FMULQ : F3_3<2, 0b110100, 0b001001011,
1032 (outs QFPRegs:$rd), (ins QFPRegs:$rs1, QFPRegs:$rs2),
1033 "fmulq $rs1, $rs2, $rd",
1034 [(set f128:$rd, (fmul f128:$rs1, f128:$rs2))]>,
1035 Requires<[HasHardQuad]>;
1037 def FSMULD : F3_3<2, 0b110100, 0b001101001,
1038 (outs DFPRegs:$rd), (ins FPRegs:$rs1, FPRegs:$rs2),
1039 "fsmuld $rs1, $rs2, $rd",
1040 [(set f64:$rd, (fmul (fextend f32:$rs1),
1041 (fextend f32:$rs2)))]>;
1042 def FDMULQ : F3_3<2, 0b110100, 0b001101110,
1043 (outs QFPRegs:$rd), (ins DFPRegs:$rs1, DFPRegs:$rs2),
1044 "fdmulq $rs1, $rs2, $rd",
1045 [(set f128:$rd, (fmul (fextend f64:$rs1),
1046 (fextend f64:$rs2)))]>,
1047 Requires<[HasHardQuad]>;
1049 def FDIVS : F3_3<2, 0b110100, 0b001001101,
1050 (outs FPRegs:$rd), (ins FPRegs:$rs1, FPRegs:$rs2),
1051 "fdivs $rs1, $rs2, $rd",
1052 [(set f32:$rd, (fdiv f32:$rs1, f32:$rs2))]>;
1053 def FDIVD : F3_3<2, 0b110100, 0b001001110,
1054 (outs DFPRegs:$rd), (ins DFPRegs:$rs1, DFPRegs:$rs2),
1055 "fdivd $rs1, $rs2, $rd",
1056 [(set f64:$rd, (fdiv f64:$rs1, f64:$rs2))]>;
1057 def FDIVQ : F3_3<2, 0b110100, 0b001001111,
1058 (outs QFPRegs:$rd), (ins QFPRegs:$rs1, QFPRegs:$rs2),
1059 "fdivq $rs1, $rs2, $rd",
1060 [(set f128:$rd, (fdiv f128:$rs1, f128:$rs2))]>,
1061 Requires<[HasHardQuad]>;
1063 // Floating-point Compare Instructions, p. 148
1064 // Note: the 2nd template arg is different for these guys.
1065 // Note 2: the result of a FCMP is not available until the 2nd cycle
1066 // after the instr is retired, but there is no interlock in Sparc V8.
1067 // This behavior is modeled with a forced noop after the instruction in
1070 let Defs = [FCC0], rd = 0, isCodeGenOnly = 1 in {
1071 def FCMPS : F3_3c<2, 0b110101, 0b001010001,
1072 (outs), (ins FPRegs:$rs1, FPRegs:$rs2),
1074 [(SPcmpfcc f32:$rs1, f32:$rs2)]>;
1075 def FCMPD : F3_3c<2, 0b110101, 0b001010010,
1076 (outs), (ins DFPRegs:$rs1, DFPRegs:$rs2),
1078 [(SPcmpfcc f64:$rs1, f64:$rs2)]>;
1079 def FCMPQ : F3_3c<2, 0b110101, 0b001010011,
1080 (outs), (ins QFPRegs:$rs1, QFPRegs:$rs2),
1082 [(SPcmpfcc f128:$rs1, f128:$rs2)]>,
1083 Requires<[HasHardQuad]>;
1086 //===----------------------------------------------------------------------===//
1087 // Instructions for Thread Local Storage(TLS).
1088 //===----------------------------------------------------------------------===//
1089 let isCodeGenOnly = 1, isAsmParserOnly = 1 in {
1090 def TLS_ADDrr : F3_1<2, 0b000000,
1092 (ins IntRegs:$rs1, IntRegs:$rs2, TLSSym:$sym),
1093 "add $rs1, $rs2, $rd, $sym",
1095 (tlsadd i32:$rs1, i32:$rs2, tglobaltlsaddr:$sym))]>;
1098 def TLS_LDrr : F3_1<3, 0b000000,
1099 (outs IntRegs:$dst), (ins MEMrr:$addr, TLSSym:$sym),
1100 "ld [$addr], $dst, $sym",
1102 (tlsld ADDRrr:$addr, tglobaltlsaddr:$sym))]>;
1104 let Uses = [O6], isCall = 1, hasDelaySlot = 1 in
1105 def TLS_CALL : InstSP<(outs),
1106 (ins calltarget:$disp, TLSSym:$sym, variable_ops),
1108 [(tlscall texternalsym:$disp, tglobaltlsaddr:$sym)]> {
1111 let Inst{29-0} = disp;
1115 //===----------------------------------------------------------------------===//
1117 //===----------------------------------------------------------------------===//
1119 // V9 Conditional Moves.
1120 let Predicates = [HasV9], Constraints = "$f = $rd" in {
1121 // Move Integer Register on Condition (MOVcc) p. 194 of the V9 manual.
1122 let Uses = [ICC], intcc = 1, cc = 0b00 in {
1124 : F4_1<0b101100, (outs IntRegs:$rd),
1125 (ins IntRegs:$rs2, IntRegs:$f, CCOp:$cond),
1126 "mov$cond %icc, $rs2, $rd",
1127 [(set i32:$rd, (SPselecticc i32:$rs2, i32:$f, imm:$cond))]>;
1130 : F4_2<0b101100, (outs IntRegs:$rd),
1131 (ins i32imm:$simm11, IntRegs:$f, CCOp:$cond),
1132 "mov$cond %icc, $simm11, $rd",
1134 (SPselecticc simm11:$simm11, i32:$f, imm:$cond))]>;
1137 let Uses = [FCC0], intcc = 0, cc = 0b00 in {
1139 : F4_1<0b101100, (outs IntRegs:$rd),
1140 (ins IntRegs:$rs2, IntRegs:$f, CCOp:$cond),
1141 "mov$cond %fcc0, $rs2, $rd",
1142 [(set i32:$rd, (SPselectfcc i32:$rs2, i32:$f, imm:$cond))]>;
1144 : F4_2<0b101100, (outs IntRegs:$rd),
1145 (ins i32imm:$simm11, IntRegs:$f, CCOp:$cond),
1146 "mov$cond %fcc0, $simm11, $rd",
1148 (SPselectfcc simm11:$simm11, i32:$f, imm:$cond))]>;
1151 let Uses = [ICC], intcc = 1, opf_cc = 0b00 in {
1153 : F4_3<0b110101, 0b000001, (outs FPRegs:$rd),
1154 (ins FPRegs:$rs2, FPRegs:$f, CCOp:$cond),
1155 "fmovs$cond %icc, $rs2, $rd",
1156 [(set f32:$rd, (SPselecticc f32:$rs2, f32:$f, imm:$cond))]>;
1158 : F4_3<0b110101, 0b000010, (outs DFPRegs:$rd),
1159 (ins DFPRegs:$rs2, DFPRegs:$f, CCOp:$cond),
1160 "fmovd$cond %icc, $rs2, $rd",
1161 [(set f64:$rd, (SPselecticc f64:$rs2, f64:$f, imm:$cond))]>;
1163 : F4_3<0b110101, 0b000011, (outs QFPRegs:$rd),
1164 (ins QFPRegs:$rs2, QFPRegs:$f, CCOp:$cond),
1165 "fmovq$cond %icc, $rs2, $rd",
1166 [(set f128:$rd, (SPselecticc f128:$rs2, f128:$f, imm:$cond))]>,
1167 Requires<[HasHardQuad]>;
1170 let Uses = [FCC0], intcc = 0, opf_cc = 0b00 in {
1172 : F4_3<0b110101, 0b000001, (outs FPRegs:$rd),
1173 (ins FPRegs:$rs2, FPRegs:$f, CCOp:$cond),
1174 "fmovs$cond %fcc0, $rs2, $rd",
1175 [(set f32:$rd, (SPselectfcc f32:$rs2, f32:$f, imm:$cond))]>;
1177 : F4_3<0b110101, 0b000010, (outs DFPRegs:$rd),
1178 (ins DFPRegs:$rs2, DFPRegs:$f, CCOp:$cond),
1179 "fmovd$cond %fcc0, $rs2, $rd",
1180 [(set f64:$rd, (SPselectfcc f64:$rs2, f64:$f, imm:$cond))]>;
1182 : F4_3<0b110101, 0b000011, (outs QFPRegs:$rd),
1183 (ins QFPRegs:$rs2, QFPRegs:$f, CCOp:$cond),
1184 "fmovq$cond %fcc0, $rs2, $rd",
1185 [(set f128:$rd, (SPselectfcc f128:$rs2, f128:$f, imm:$cond))]>,
1186 Requires<[HasHardQuad]>;
1191 // Floating-Point Move Instructions, p. 164 of the V9 manual.
1192 let Predicates = [HasV9] in {
1193 def FMOVD : F3_3u<2, 0b110100, 0b000000010,
1194 (outs DFPRegs:$rd), (ins DFPRegs:$rs2),
1195 "fmovd $rs2, $rd", []>;
1196 def FMOVQ : F3_3u<2, 0b110100, 0b000000011,
1197 (outs QFPRegs:$rd), (ins QFPRegs:$rs2),
1198 "fmovq $rs2, $rd", []>,
1199 Requires<[HasHardQuad]>;
1200 def FNEGD : F3_3u<2, 0b110100, 0b000000110,
1201 (outs DFPRegs:$rd), (ins DFPRegs:$rs2),
1203 [(set f64:$rd, (fneg f64:$rs2))]>;
1204 def FNEGQ : F3_3u<2, 0b110100, 0b000000111,
1205 (outs QFPRegs:$rd), (ins QFPRegs:$rs2),
1207 [(set f128:$rd, (fneg f128:$rs2))]>,
1208 Requires<[HasHardQuad]>;
1209 def FABSD : F3_3u<2, 0b110100, 0b000001010,
1210 (outs DFPRegs:$rd), (ins DFPRegs:$rs2),
1212 [(set f64:$rd, (fabs f64:$rs2))]>;
1213 def FABSQ : F3_3u<2, 0b110100, 0b000001011,
1214 (outs QFPRegs:$rd), (ins QFPRegs:$rs2),
1216 [(set f128:$rd, (fabs f128:$rs2))]>,
1217 Requires<[HasHardQuad]>;
1220 // Floating-point compare instruction with %fcc0-%fcc3.
1221 def V9FCMPS : F3_3c<2, 0b110101, 0b001010001,
1222 (outs FCCRegs:$rd), (ins FPRegs:$rs1, FPRegs:$rs2),
1223 "fcmps $rd, $rs1, $rs2", []>;
1224 def V9FCMPD : F3_3c<2, 0b110101, 0b001010010,
1225 (outs FCCRegs:$rd), (ins DFPRegs:$rs1, DFPRegs:$rs2),
1226 "fcmpd $rd, $rs1, $rs2", []>;
1227 def V9FCMPQ : F3_3c<2, 0b110101, 0b001010011,
1228 (outs FCCRegs:$rd), (ins QFPRegs:$rs1, QFPRegs:$rs2),
1229 "fcmpq $rd, $rs1, $rs2", []>,
1230 Requires<[HasHardQuad]>;
1232 let hasSideEffects = 1 in {
1233 def V9FCMPES : F3_3c<2, 0b110101, 0b001010101,
1234 (outs FCCRegs:$rd), (ins FPRegs:$rs1, FPRegs:$rs2),
1235 "fcmpes $rd, $rs1, $rs2", []>;
1236 def V9FCMPED : F3_3c<2, 0b110101, 0b001010110,
1237 (outs FCCRegs:$rd), (ins DFPRegs:$rs1, DFPRegs:$rs2),
1238 "fcmped $rd, $rs1, $rs2", []>;
1239 def V9FCMPEQ : F3_3c<2, 0b110101, 0b001010111,
1240 (outs FCCRegs:$rd), (ins QFPRegs:$rs1, QFPRegs:$rs2),
1241 "fcmpeq $rd, $rs1, $rs2", []>,
1242 Requires<[HasHardQuad]>;
1245 // Floating point conditional move instrucitons with %fcc0-%fcc3.
1246 let Predicates = [HasV9] in {
1247 let Constraints = "$f = $rd", intcc = 0 in {
1249 : F4_1<0b101100, (outs IntRegs:$rd),
1250 (ins FCCRegs:$cc, IntRegs:$rs2, IntRegs:$f, CCOp:$cond),
1251 "mov$cond $cc, $rs2, $rd", []>;
1253 : F4_2<0b101100, (outs IntRegs:$rd),
1254 (ins FCCRegs:$cc, i32imm:$simm11, IntRegs:$f, CCOp:$cond),
1255 "mov$cond $cc, $simm11, $rd", []>;
1257 : F4_3<0b110101, 0b000001, (outs FPRegs:$rd),
1258 (ins FCCRegs:$opf_cc, FPRegs:$rs2, FPRegs:$f, CCOp:$cond),
1259 "fmovs$cond $opf_cc, $rs2, $rd", []>;
1261 : F4_3<0b110101, 0b000010, (outs DFPRegs:$rd),
1262 (ins FCCRegs:$opf_cc, DFPRegs:$rs2, DFPRegs:$f, CCOp:$cond),
1263 "fmovd$cond $opf_cc, $rs2, $rd", []>;
1265 : F4_3<0b110101, 0b000011, (outs QFPRegs:$rd),
1266 (ins FCCRegs:$opf_cc, QFPRegs:$rs2, QFPRegs:$f, CCOp:$cond),
1267 "fmovq$cond $opf_cc, $rs2, $rd", []>,
1268 Requires<[HasHardQuad]>;
1269 } // Constraints = "$f = $rd", ...
1270 } // let Predicates = [hasV9]
1273 // POPCrr - This does a ctpop of a 64-bit register. As such, we have to clear
1274 // the top 32-bits before using it. To do this clearing, we use a SRLri X,0.
1276 def POPCrr : F3_1<2, 0b101110,
1277 (outs IntRegs:$rd), (ins IntRegs:$rs2),
1278 "popc $rs2, $rd", []>, Requires<[HasV9]>;
1279 def : Pat<(ctpop i32:$src),
1280 (POPCrr (SRLri $src, 0))>;
1282 let Predicates = [HasV9], hasSideEffects = 1, rd = 0, rs1 = 0b01111 in
1283 def MEMBARi : F3_2<2, 0b101000, (outs), (ins simm13Op:$simm13),
1284 "membar $simm13", []>;
1286 // TODO: Should add a CASArr variant. In fact, the CAS instruction,
1287 // unlike other instructions, only comes in a form which requires an
1288 // ASI be provided. The ASI value hardcoded here is ASI_PRIMARY, the
1289 // default unprivileged ASI for SparcV9. (Also of note: some modern
1290 // SparcV8 implementations provide CASA as an extension, but require
1291 // the use of SparcV8's default ASI, 0xA ("User Data") instead.)
1292 let Predicates = [HasV9], Constraints = "$swap = $rd", asi = 0b10000000 in
1293 def CASrr: F3_1_asi<3, 0b111100,
1294 (outs IntRegs:$rd), (ins IntRegs:$rs1, IntRegs:$rs2,
1296 "cas [$rs1], $rs2, $rd",
1298 (atomic_cmp_swap iPTR:$rs1, i32:$rs2, i32:$swap))]>;
1300 let Defs = [ICC] in {
1301 defm TADDCC : F3_12np<"taddcc", 0b100000>;
1302 defm TSUBCC : F3_12np<"tsubcc", 0b100001>;
1304 let hasSideEffects = 1 in {
1305 defm TADDCCTV : F3_12np<"taddcctv", 0b100010>;
1306 defm TSUBCCTV : F3_12np<"tsubcctv", 0b100011>;
1310 //===----------------------------------------------------------------------===//
1311 // Non-Instruction Patterns
1312 //===----------------------------------------------------------------------===//
1314 // Small immediates.
1315 def : Pat<(i32 simm13:$val),
1316 (ORri (i32 G0), imm:$val)>;
1317 // Arbitrary immediates.
1318 def : Pat<(i32 imm:$val),
1319 (ORri (SETHIi (HI22 imm:$val)), (LO10 imm:$val))>;
1322 // Global addresses, constant pool entries
1323 let Predicates = [Is32Bit] in {
1325 def : Pat<(SPhi tglobaladdr:$in), (SETHIi tglobaladdr:$in)>;
1326 def : Pat<(SPlo tglobaladdr:$in), (ORri (i32 G0), tglobaladdr:$in)>;
1327 def : Pat<(SPhi tconstpool:$in), (SETHIi tconstpool:$in)>;
1328 def : Pat<(SPlo tconstpool:$in), (ORri (i32 G0), tconstpool:$in)>;
1330 // GlobalTLS addresses
1331 def : Pat<(SPhi tglobaltlsaddr:$in), (SETHIi tglobaltlsaddr:$in)>;
1332 def : Pat<(SPlo tglobaltlsaddr:$in), (ORri (i32 G0), tglobaltlsaddr:$in)>;
1333 def : Pat<(add (SPhi tglobaltlsaddr:$in1), (SPlo tglobaltlsaddr:$in2)),
1334 (ADDri (SETHIi tglobaltlsaddr:$in1), (tglobaltlsaddr:$in2))>;
1335 def : Pat<(xor (SPhi tglobaltlsaddr:$in1), (SPlo tglobaltlsaddr:$in2)),
1336 (XORri (SETHIi tglobaltlsaddr:$in1), (tglobaltlsaddr:$in2))>;
1339 def : Pat<(SPhi tblockaddress:$in), (SETHIi tblockaddress:$in)>;
1340 def : Pat<(SPlo tblockaddress:$in), (ORri (i32 G0), tblockaddress:$in)>;
1342 // Add reg, lo. This is used when taking the addr of a global/constpool entry.
1343 def : Pat<(add iPTR:$r, (SPlo tglobaladdr:$in)), (ADDri $r, tglobaladdr:$in)>;
1344 def : Pat<(add iPTR:$r, (SPlo tconstpool:$in)), (ADDri $r, tconstpool:$in)>;
1345 def : Pat<(add iPTR:$r, (SPlo tblockaddress:$in)),
1346 (ADDri $r, tblockaddress:$in)>;
1350 def : Pat<(call tglobaladdr:$dst),
1351 (CALL tglobaladdr:$dst)>;
1352 def : Pat<(call texternalsym:$dst),
1353 (CALL texternalsym:$dst)>;
1355 // Map integer extload's to zextloads.
1356 def : Pat<(i32 (extloadi1 ADDRrr:$src)), (LDUBrr ADDRrr:$src)>;
1357 def : Pat<(i32 (extloadi1 ADDRri:$src)), (LDUBri ADDRri:$src)>;
1358 def : Pat<(i32 (extloadi8 ADDRrr:$src)), (LDUBrr ADDRrr:$src)>;
1359 def : Pat<(i32 (extloadi8 ADDRri:$src)), (LDUBri ADDRri:$src)>;
1360 def : Pat<(i32 (extloadi16 ADDRrr:$src)), (LDUHrr ADDRrr:$src)>;
1361 def : Pat<(i32 (extloadi16 ADDRri:$src)), (LDUHri ADDRri:$src)>;
1363 // zextload bool -> zextload byte
1364 def : Pat<(i32 (zextloadi1 ADDRrr:$src)), (LDUBrr ADDRrr:$src)>;
1365 def : Pat<(i32 (zextloadi1 ADDRri:$src)), (LDUBri ADDRri:$src)>;
1367 // store 0, addr -> store %g0, addr
1368 def : Pat<(store (i32 0), ADDRrr:$dst), (STrr ADDRrr:$dst, (i32 G0))>;
1369 def : Pat<(store (i32 0), ADDRri:$dst), (STri ADDRri:$dst, (i32 G0))>;
1371 // store bar for all atomic_fence in V8.
1372 let Predicates = [HasNoV9] in
1373 def : Pat<(atomic_fence imm, imm), (STBAR)>;
1375 // atomic_load_32 addr -> load addr
1376 def : Pat<(i32 (atomic_load ADDRrr:$src)), (LDrr ADDRrr:$src)>;
1377 def : Pat<(i32 (atomic_load ADDRri:$src)), (LDri ADDRri:$src)>;
1379 // atomic_store_32 val, addr -> store val, addr
1380 def : Pat<(atomic_store ADDRrr:$dst, i32:$val), (STrr ADDRrr:$dst, $val)>;
1381 def : Pat<(atomic_store ADDRri:$dst, i32:$val), (STri ADDRri:$dst, $val)>;
1384 def : Pat<(vector_extract (v2i32 IntPair:$Rn), 0),
1385 (i32 (EXTRACT_SUBREG IntPair:$Rn, sub_even))>;
1386 def : Pat<(vector_extract (v2i32 IntPair:$Rn), 1),
1387 (i32 (EXTRACT_SUBREG IntPair:$Rn, sub_odd))>;
1390 def : Pat<(build_vector (i32 IntRegs:$a1), (i32 IntRegs:$a2)),
1392 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)), (i32 IntRegs:$a1), sub_even),
1393 (i32 IntRegs:$a2), sub_odd)>;
1396 include "SparcInstr64Bit.td"
1397 include "SparcInstrVIS.td"
1398 include "SparcInstrAliases.td"