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()">;
34 // HasNoV9 - This predicate is true when the target doesn't have V9
35 // instructions. Use of this is just a hack for the isel not having proper
36 // costs for V8 instructions that are more expensive than their V9 ones.
37 def HasNoV9 : Predicate<"!Subtarget.isV9()">;
39 // HasVIS - This is true when the target processor has VIS extensions.
40 def HasVIS : Predicate<"Subtarget.isVIS()">;
42 // UseDeprecatedInsts - This predicate is true when the target processor is a
43 // V8, or when it is V9 but the V8 deprecated instructions are efficient enough
44 // to use when appropriate. In either of these cases, the instruction selector
45 // will pick deprecated instructions.
46 def UseDeprecatedInsts : Predicate<"Subtarget.useDeprecatedV8Instructions()">;
48 //===----------------------------------------------------------------------===//
49 // Instruction Pattern Stuff
50 //===----------------------------------------------------------------------===//
52 def simm11 : PatLeaf<(imm), [{ return isInt<11>(N->getSExtValue()); }]>;
54 def simm13 : PatLeaf<(imm), [{ return isInt<13>(N->getSExtValue()); }]>;
56 def LO10 : SDNodeXForm<imm, [{
57 return CurDAG->getTargetConstant((unsigned)N->getZExtValue() & 1023,
61 def HI22 : SDNodeXForm<imm, [{
62 // Transformation function: shift the immediate value down into the low bits.
63 return CurDAG->getTargetConstant((unsigned)N->getZExtValue() >> 10, MVT::i32);
66 def SETHIimm : PatLeaf<(imm), [{
67 return (((unsigned)N->getZExtValue() >> 10) << 10) ==
68 (unsigned)N->getZExtValue();
72 def ADDRrr : ComplexPattern<iPTR, 2, "SelectADDRrr", [], []>;
73 def ADDRri : ComplexPattern<iPTR, 2, "SelectADDRri", [frameindex], []>;
76 def MEMrr : Operand<iPTR> {
77 let PrintMethod = "printMemOperand";
78 let MIOperandInfo = (ops ptr_rc, ptr_rc);
80 def MEMri : Operand<iPTR> {
81 let PrintMethod = "printMemOperand";
82 let MIOperandInfo = (ops ptr_rc, i32imm);
85 // Branch targets have OtherVT type.
86 def brtarget : Operand<OtherVT>;
87 def calltarget : Operand<i32>;
89 // Operand for printing out a condition code.
90 let PrintMethod = "printCCOperand" in
91 def CCOp : Operand<i32>;
94 SDTypeProfile<0, 2, [SDTCisFP<0>, SDTCisSameAs<0, 1>]>;
96 SDTypeProfile<0, 2, [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>]>;
98 SDTypeProfile<1, 3, [SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>, SDTCisVT<3, i32>]>;
100 SDTypeProfile<1, 1, [SDTCisVT<0, f32>, SDTCisFP<1>]>;
102 SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisVT<1, f32>]>;
104 def SPcmpicc : SDNode<"SPISD::CMPICC", SDTIntBinOp, [SDNPOutGlue]>;
105 def SPcmpfcc : SDNode<"SPISD::CMPFCC", SDTSPcmpfcc, [SDNPOutGlue]>;
106 def SPbricc : SDNode<"SPISD::BRICC", SDTSPbrcc, [SDNPHasChain, SDNPInGlue]>;
107 def SPbrxcc : SDNode<"SPISD::BRXCC", SDTSPbrcc, [SDNPHasChain, SDNPInGlue]>;
108 def SPbrfcc : SDNode<"SPISD::BRFCC", SDTSPbrcc, [SDNPHasChain, SDNPInGlue]>;
110 def SPhi : SDNode<"SPISD::Hi", SDTIntUnaryOp>;
111 def SPlo : SDNode<"SPISD::Lo", SDTIntUnaryOp>;
113 def SPftoi : SDNode<"SPISD::FTOI", SDTSPFTOI>;
114 def SPitof : SDNode<"SPISD::ITOF", SDTSPITOF>;
116 def SPselecticc : SDNode<"SPISD::SELECT_ICC", SDTSPselectcc, [SDNPInGlue]>;
117 def SPselectxcc : SDNode<"SPISD::SELECT_XCC", SDTSPselectcc, [SDNPInGlue]>;
118 def SPselectfcc : SDNode<"SPISD::SELECT_FCC", SDTSPselectcc, [SDNPInGlue]>;
120 // These are target-independent nodes, but have target-specific formats.
121 def SDT_SPCallSeqStart : SDCallSeqStart<[ SDTCisVT<0, i32> ]>;
122 def SDT_SPCallSeqEnd : SDCallSeqEnd<[ SDTCisVT<0, i32>,
125 def callseq_start : SDNode<"ISD::CALLSEQ_START", SDT_SPCallSeqStart,
126 [SDNPHasChain, SDNPOutGlue]>;
127 def callseq_end : SDNode<"ISD::CALLSEQ_END", SDT_SPCallSeqEnd,
128 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
130 def SDT_SPCall : SDTypeProfile<0, -1, [SDTCisVT<0, i32>]>;
131 def call : SDNode<"SPISD::CALL", SDT_SPCall,
132 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
135 def SDT_SPRet : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
136 def retflag : SDNode<"SPISD::RET_FLAG", SDT_SPRet,
137 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
139 def flushw : SDNode<"SPISD::FLUSHW", SDTNone,
140 [SDNPHasChain, SDNPSideEffect, SDNPMayStore]>;
142 def getPCX : Operand<i32> {
143 let PrintMethod = "printGetPCX";
146 //===----------------------------------------------------------------------===//
147 // SPARC Flag Conditions
148 //===----------------------------------------------------------------------===//
150 // Note that these values must be kept in sync with the CCOp::CondCode enum
152 class ICC_VAL<int N> : PatLeaf<(i32 N)>;
153 def ICC_NE : ICC_VAL< 9>; // Not Equal
154 def ICC_E : ICC_VAL< 1>; // Equal
155 def ICC_G : ICC_VAL<10>; // Greater
156 def ICC_LE : ICC_VAL< 2>; // Less or Equal
157 def ICC_GE : ICC_VAL<11>; // Greater or Equal
158 def ICC_L : ICC_VAL< 3>; // Less
159 def ICC_GU : ICC_VAL<12>; // Greater Unsigned
160 def ICC_LEU : ICC_VAL< 4>; // Less or Equal Unsigned
161 def ICC_CC : ICC_VAL<13>; // Carry Clear/Great or Equal Unsigned
162 def ICC_CS : ICC_VAL< 5>; // Carry Set/Less Unsigned
163 def ICC_POS : ICC_VAL<14>; // Positive
164 def ICC_NEG : ICC_VAL< 6>; // Negative
165 def ICC_VC : ICC_VAL<15>; // Overflow Clear
166 def ICC_VS : ICC_VAL< 7>; // Overflow Set
168 class FCC_VAL<int N> : PatLeaf<(i32 N)>;
169 def FCC_U : FCC_VAL<23>; // Unordered
170 def FCC_G : FCC_VAL<22>; // Greater
171 def FCC_UG : FCC_VAL<21>; // Unordered or Greater
172 def FCC_L : FCC_VAL<20>; // Less
173 def FCC_UL : FCC_VAL<19>; // Unordered or Less
174 def FCC_LG : FCC_VAL<18>; // Less or Greater
175 def FCC_NE : FCC_VAL<17>; // Not Equal
176 def FCC_E : FCC_VAL<25>; // Equal
177 def FCC_UE : FCC_VAL<24>; // Unordered or Equal
178 def FCC_GE : FCC_VAL<25>; // Greater or Equal
179 def FCC_UGE : FCC_VAL<26>; // Unordered or Greater or Equal
180 def FCC_LE : FCC_VAL<27>; // Less or Equal
181 def FCC_ULE : FCC_VAL<28>; // Unordered or Less or Equal
182 def FCC_O : FCC_VAL<29>; // Ordered
184 //===----------------------------------------------------------------------===//
185 // Instruction Class Templates
186 //===----------------------------------------------------------------------===//
188 /// F3_12 multiclass - Define a normal F3_1/F3_2 pattern in one shot.
189 multiclass F3_12<string OpcStr, bits<6> Op3Val, SDNode OpNode> {
190 def rr : F3_1<2, Op3Val,
191 (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
192 !strconcat(OpcStr, " $b, $c, $dst"),
193 [(set i32:$dst, (OpNode i32:$b, i32:$c))]>;
194 def ri : F3_2<2, Op3Val,
195 (outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
196 !strconcat(OpcStr, " $b, $c, $dst"),
197 [(set i32:$dst, (OpNode i32:$b, (i32 simm13:$c)))]>;
200 /// F3_12np multiclass - Define a normal F3_1/F3_2 pattern in one shot, with no
202 multiclass F3_12np<string OpcStr, bits<6> Op3Val> {
203 def rr : F3_1<2, Op3Val,
204 (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
205 !strconcat(OpcStr, " $b, $c, $dst"), []>;
206 def ri : F3_2<2, Op3Val,
207 (outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
208 !strconcat(OpcStr, " $b, $c, $dst"), []>;
211 //===----------------------------------------------------------------------===//
213 //===----------------------------------------------------------------------===//
215 // Pseudo instructions.
216 class Pseudo<dag outs, dag ins, string asmstr, list<dag> pattern>
217 : InstSP<outs, ins, asmstr, pattern>;
221 def GETPCX : Pseudo<(outs getPCX:$getpcseq), (ins), "$getpcseq", [] >;
224 let Defs = [O6], Uses = [O6] in {
225 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
226 "!ADJCALLSTACKDOWN $amt",
227 [(callseq_start timm:$amt)]>;
228 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
229 "!ADJCALLSTACKUP $amt1",
230 [(callseq_end timm:$amt1, timm:$amt2)]>;
233 let hasSideEffects = 1, mayStore = 1 in {
234 let rd = 0, rs1 = 0, rs2 = 0 in
235 def FLUSHW : F3_1<0b10, 0b101011, (outs), (ins),
237 [(flushw)]>, Requires<[HasV9]>;
238 let rd = 0, rs1 = 1, simm13 = 3 in
239 def TA3 : F3_2<0b10, 0b111010, (outs), (ins),
244 def UNIMP : F2_1<0b000, (outs), (ins i32imm:$val),
247 // FpMOVD/FpNEGD/FpABSD - These are lowered to single-precision ops by the
249 let Predicates = [HasNoV9] in { // Only emit these in V8 mode.
250 def FpMOVD : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$src),
251 "!FpMOVD $src, $dst", []>;
252 def FpNEGD : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$src),
253 "!FpNEGD $src, $dst",
254 [(set f64:$dst, (fneg f64:$src))]>;
255 def FpABSD : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$src),
256 "!FpABSD $src, $dst",
257 [(set f64:$dst, (fabs f64:$src))]>;
260 // SELECT_CC_* - Used to implement the SELECT_CC DAG operation. Expanded after
261 // instruction selection into a branch sequence. This has to handle all
262 // permutations of selection between i32/f32/f64 on ICC and FCC.
263 // Expanded after instruction selection.
264 let Uses = [ICC], usesCustomInserter = 1 in {
265 def SELECT_CC_Int_ICC
266 : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, IntRegs:$F, i32imm:$Cond),
267 "; SELECT_CC_Int_ICC PSEUDO!",
268 [(set i32:$dst, (SPselecticc i32:$T, i32:$F, imm:$Cond))]>;
270 : Pseudo<(outs FPRegs:$dst), (ins FPRegs:$T, FPRegs:$F, i32imm:$Cond),
271 "; SELECT_CC_FP_ICC PSEUDO!",
272 [(set f32:$dst, (SPselecticc f32:$T, f32:$F, imm:$Cond))]>;
274 def SELECT_CC_DFP_ICC
275 : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$T, DFPRegs:$F, i32imm:$Cond),
276 "; SELECT_CC_DFP_ICC PSEUDO!",
277 [(set f64:$dst, (SPselecticc f64:$T, f64:$F, imm:$Cond))]>;
280 let usesCustomInserter = 1, Uses = [FCC] in {
282 def SELECT_CC_Int_FCC
283 : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, IntRegs:$F, i32imm:$Cond),
284 "; SELECT_CC_Int_FCC PSEUDO!",
285 [(set i32:$dst, (SPselectfcc i32:$T, i32:$F, imm:$Cond))]>;
288 : Pseudo<(outs FPRegs:$dst), (ins FPRegs:$T, FPRegs:$F, i32imm:$Cond),
289 "; SELECT_CC_FP_FCC PSEUDO!",
290 [(set f32:$dst, (SPselectfcc f32:$T, f32:$F, imm:$Cond))]>;
291 def SELECT_CC_DFP_FCC
292 : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$T, DFPRegs:$F, i32imm:$Cond),
293 "; SELECT_CC_DFP_FCC PSEUDO!",
294 [(set f64:$dst, (SPselectfcc f64:$T, f64:$F, imm:$Cond))]>;
298 // Section A.3 - Synthetic Instructions, p. 85
299 // special cases of JMPL:
300 let isReturn = 1, isTerminator = 1, hasDelaySlot = 1, isBarrier = 1 in {
301 let rd = O7.Num, rs1 = G0.Num in
302 def RETL: F3_2<2, 0b111000, (outs), (ins i32imm:$val),
303 "jmp %o7+$val", [(retflag simm13:$val)]>;
305 let rd = I7.Num, rs1 = G0.Num in
306 def RET: F3_2<2, 0b111000, (outs), (ins i32imm:$val),
310 // Section B.1 - Load Integer Instructions, p. 90
311 def LDSBrr : F3_1<3, 0b001001,
312 (outs IntRegs:$dst), (ins MEMrr:$addr),
313 "ldsb [$addr], $dst",
314 [(set i32:$dst, (sextloadi8 ADDRrr:$addr))]>;
315 def LDSBri : F3_2<3, 0b001001,
316 (outs IntRegs:$dst), (ins MEMri:$addr),
317 "ldsb [$addr], $dst",
318 [(set i32:$dst, (sextloadi8 ADDRri:$addr))]>;
319 def LDSHrr : F3_1<3, 0b001010,
320 (outs IntRegs:$dst), (ins MEMrr:$addr),
321 "ldsh [$addr], $dst",
322 [(set i32:$dst, (sextloadi16 ADDRrr:$addr))]>;
323 def LDSHri : F3_2<3, 0b001010,
324 (outs IntRegs:$dst), (ins MEMri:$addr),
325 "ldsh [$addr], $dst",
326 [(set i32:$dst, (sextloadi16 ADDRri:$addr))]>;
327 def LDUBrr : F3_1<3, 0b000001,
328 (outs IntRegs:$dst), (ins MEMrr:$addr),
329 "ldub [$addr], $dst",
330 [(set i32:$dst, (zextloadi8 ADDRrr:$addr))]>;
331 def LDUBri : F3_2<3, 0b000001,
332 (outs IntRegs:$dst), (ins MEMri:$addr),
333 "ldub [$addr], $dst",
334 [(set i32:$dst, (zextloadi8 ADDRri:$addr))]>;
335 def LDUHrr : F3_1<3, 0b000010,
336 (outs IntRegs:$dst), (ins MEMrr:$addr),
337 "lduh [$addr], $dst",
338 [(set i32:$dst, (zextloadi16 ADDRrr:$addr))]>;
339 def LDUHri : F3_2<3, 0b000010,
340 (outs IntRegs:$dst), (ins MEMri:$addr),
341 "lduh [$addr], $dst",
342 [(set i32:$dst, (zextloadi16 ADDRri:$addr))]>;
343 def LDrr : F3_1<3, 0b000000,
344 (outs IntRegs:$dst), (ins MEMrr:$addr),
346 [(set i32:$dst, (load ADDRrr:$addr))]>;
347 def LDri : F3_2<3, 0b000000,
348 (outs IntRegs:$dst), (ins MEMri:$addr),
350 [(set i32:$dst, (load ADDRri:$addr))]>;
352 // Section B.2 - Load Floating-point Instructions, p. 92
353 def LDFrr : F3_1<3, 0b100000,
354 (outs FPRegs:$dst), (ins MEMrr:$addr),
356 [(set f32:$dst, (load ADDRrr:$addr))]>;
357 def LDFri : F3_2<3, 0b100000,
358 (outs FPRegs:$dst), (ins MEMri:$addr),
360 [(set f32:$dst, (load ADDRri:$addr))]>;
361 def LDDFrr : F3_1<3, 0b100011,
362 (outs DFPRegs:$dst), (ins MEMrr:$addr),
364 [(set f64:$dst, (load ADDRrr:$addr))]>;
365 def LDDFri : F3_2<3, 0b100011,
366 (outs DFPRegs:$dst), (ins MEMri:$addr),
368 [(set f64:$dst, (load ADDRri:$addr))]>;
370 // Section B.4 - Store Integer Instructions, p. 95
371 def STBrr : F3_1<3, 0b000101,
372 (outs), (ins MEMrr:$addr, IntRegs:$src),
374 [(truncstorei8 i32:$src, ADDRrr:$addr)]>;
375 def STBri : F3_2<3, 0b000101,
376 (outs), (ins MEMri:$addr, IntRegs:$src),
378 [(truncstorei8 i32:$src, ADDRri:$addr)]>;
379 def STHrr : F3_1<3, 0b000110,
380 (outs), (ins MEMrr:$addr, IntRegs:$src),
382 [(truncstorei16 i32:$src, ADDRrr:$addr)]>;
383 def STHri : F3_2<3, 0b000110,
384 (outs), (ins MEMri:$addr, IntRegs:$src),
386 [(truncstorei16 i32:$src, ADDRri:$addr)]>;
387 def STrr : F3_1<3, 0b000100,
388 (outs), (ins MEMrr:$addr, IntRegs:$src),
390 [(store i32:$src, ADDRrr:$addr)]>;
391 def STri : F3_2<3, 0b000100,
392 (outs), (ins MEMri:$addr, IntRegs:$src),
394 [(store i32:$src, ADDRri:$addr)]>;
396 // Section B.5 - Store Floating-point Instructions, p. 97
397 def STFrr : F3_1<3, 0b100100,
398 (outs), (ins MEMrr:$addr, FPRegs:$src),
400 [(store f32:$src, ADDRrr:$addr)]>;
401 def STFri : F3_2<3, 0b100100,
402 (outs), (ins MEMri:$addr, FPRegs:$src),
404 [(store f32:$src, ADDRri:$addr)]>;
405 def STDFrr : F3_1<3, 0b100111,
406 (outs), (ins MEMrr:$addr, DFPRegs:$src),
408 [(store f64:$src, ADDRrr:$addr)]>;
409 def STDFri : F3_2<3, 0b100111,
410 (outs), (ins MEMri:$addr, DFPRegs:$src),
412 [(store f64:$src, ADDRri:$addr)]>;
414 // Section B.9 - SETHI Instruction, p. 104
415 def SETHIi: F2_1<0b100,
416 (outs IntRegs:$dst), (ins i32imm:$src),
418 [(set i32:$dst, SETHIimm:$src)]>;
420 // Section B.10 - NOP Instruction, p. 105
421 // (It's a special case of SETHI)
422 let rd = 0, imm22 = 0 in
423 def NOP : F2_1<0b100, (outs), (ins), "nop", []>;
425 // Section B.11 - Logical Instructions, p. 106
426 defm AND : F3_12<"and", 0b000001, and>;
428 def ANDNrr : F3_1<2, 0b000101,
429 (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
431 [(set i32:$dst, (and i32:$b, (not i32:$c)))]>;
432 def ANDNri : F3_2<2, 0b000101,
433 (outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
434 "andn $b, $c, $dst", []>;
436 defm OR : F3_12<"or", 0b000010, or>;
438 def ORNrr : F3_1<2, 0b000110,
439 (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
441 [(set i32:$dst, (or i32:$b, (not i32:$c)))]>;
442 def ORNri : F3_2<2, 0b000110,
443 (outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
444 "orn $b, $c, $dst", []>;
445 defm XOR : F3_12<"xor", 0b000011, xor>;
447 def XNORrr : F3_1<2, 0b000111,
448 (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
450 [(set i32:$dst, (not (xor i32:$b, i32:$c)))]>;
451 def XNORri : F3_2<2, 0b000111,
452 (outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
453 "xnor $b, $c, $dst", []>;
455 // Section B.12 - Shift Instructions, p. 107
456 defm SLL : F3_12<"sll", 0b100101, shl>;
457 defm SRL : F3_12<"srl", 0b100110, srl>;
458 defm SRA : F3_12<"sra", 0b100111, sra>;
460 // Section B.13 - Add Instructions, p. 108
461 defm ADD : F3_12<"add", 0b000000, add>;
463 // "LEA" forms of add (patterns to make tblgen happy)
464 def LEA_ADDri : F3_2<2, 0b000000,
465 (outs IntRegs:$dst), (ins MEMri:$addr),
466 "add ${addr:arith}, $dst",
467 [(set i32:$dst, ADDRri:$addr)]>;
470 defm ADDCC : F3_12<"addcc", 0b010000, addc>;
473 defm ADDX : F3_12<"addx", 0b001000, adde>;
475 // Section B.15 - Subtract Instructions, p. 110
476 defm SUB : F3_12 <"sub" , 0b000100, sub>;
478 defm SUBX : F3_12 <"subx" , 0b001100, sube>;
481 defm SUBCC : F3_12 <"subcc", 0b010100, SPcmpicc>;
483 let Uses = [ICC], Defs = [ICC] in
484 def SUBXCCrr: F3_1<2, 0b011100,
485 (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
486 "subxcc $b, $c, $dst", []>;
489 // Section B.18 - Multiply Instructions, p. 113
491 defm UMUL : F3_12np<"umul", 0b001010>;
492 defm SMUL : F3_12 <"smul", 0b001011, mul>;
495 // Section B.19 - Divide Instructions, p. 115
497 defm UDIV : F3_12np<"udiv", 0b001110>;
498 defm SDIV : F3_12np<"sdiv", 0b001111>;
501 // Section B.20 - SAVE and RESTORE, p. 117
502 defm SAVE : F3_12np<"save" , 0b111100>;
503 defm RESTORE : F3_12np<"restore", 0b111101>;
505 // Section B.21 - Branch on Integer Condition Codes Instructions, p. 119
507 // conditional branch class:
508 class BranchSP<bits<4> cc, dag ins, string asmstr, list<dag> pattern>
509 : F2_2<cc, 0b010, (outs), ins, asmstr, pattern> {
511 let isTerminator = 1;
512 let hasDelaySlot = 1;
516 def BA : BranchSP<0b1000, (ins brtarget:$dst),
520 // FIXME: the encoding for the JIT should look at the condition field.
522 def BCOND : BranchSP<0, (ins brtarget:$dst, CCOp:$cc),
524 [(SPbricc bb:$dst, imm:$cc)]>;
527 // Section B.22 - Branch on Floating-point Condition Codes Instructions, p. 121
529 // floating-point conditional branch class:
530 class FPBranchSP<bits<4> cc, dag ins, string asmstr, list<dag> pattern>
531 : F2_2<cc, 0b110, (outs), ins, asmstr, pattern> {
533 let isTerminator = 1;
534 let hasDelaySlot = 1;
537 // FIXME: the encoding for the JIT should look at the condition field.
539 def FBCOND : FPBranchSP<0, (ins brtarget:$dst, CCOp:$cc),
541 [(SPbrfcc bb:$dst, imm:$cc)]>;
544 // Section B.24 - Call and Link Instruction, p. 125
545 // This is the only Format 1 instruction
547 hasDelaySlot = 1, isCall = 1,
548 Defs = [O0, O1, O2, O3, O4, O5, O7, G1, G2, G3, G4, G5, G6, G7,
549 D0, D1, D2, D3, D4, D5, D6, D7, D8, D9, D10, D11, D12, D13, D14, D15,
551 def CALL : InstSP<(outs), (ins calltarget:$dst, variable_ops),
555 let Inst{29-0} = disp;
559 def JMPLrr : F3_1<2, 0b111000,
560 (outs), (ins MEMrr:$ptr, variable_ops),
562 [(call ADDRrr:$ptr)]>;
563 def JMPLri : F3_2<2, 0b111000,
564 (outs), (ins MEMri:$ptr, variable_ops),
566 [(call ADDRri:$ptr)]>;
569 // Section B.28 - Read State Register Instructions
571 def RDY : F3_1<2, 0b101000,
572 (outs IntRegs:$dst), (ins),
575 // Section B.29 - Write State Register Instructions
577 def WRYrr : F3_1<2, 0b110000,
578 (outs), (ins IntRegs:$b, IntRegs:$c),
579 "wr $b, $c, %y", []>;
580 def WRYri : F3_2<2, 0b110000,
581 (outs), (ins IntRegs:$b, i32imm:$c),
582 "wr $b, $c, %y", []>;
584 // Convert Integer to Floating-point Instructions, p. 141
585 def FITOS : F3_3<2, 0b110100, 0b011000100,
586 (outs FPRegs:$dst), (ins FPRegs:$src),
588 [(set FPRegs:$dst, (SPitof FPRegs:$src))]>;
589 def FITOD : F3_3<2, 0b110100, 0b011001000,
590 (outs DFPRegs:$dst), (ins FPRegs:$src),
592 [(set DFPRegs:$dst, (SPitof FPRegs:$src))]>;
594 // Convert Floating-point to Integer Instructions, p. 142
595 def FSTOI : F3_3<2, 0b110100, 0b011010001,
596 (outs FPRegs:$dst), (ins FPRegs:$src),
598 [(set FPRegs:$dst, (SPftoi FPRegs:$src))]>;
599 def FDTOI : F3_3<2, 0b110100, 0b011010010,
600 (outs FPRegs:$dst), (ins DFPRegs:$src),
602 [(set FPRegs:$dst, (SPftoi DFPRegs:$src))]>;
604 // Convert between Floating-point Formats Instructions, p. 143
605 def FSTOD : F3_3<2, 0b110100, 0b011001001,
606 (outs DFPRegs:$dst), (ins FPRegs:$src),
608 [(set f64:$dst, (fextend f32:$src))]>;
609 def FDTOS : F3_3<2, 0b110100, 0b011000110,
610 (outs FPRegs:$dst), (ins DFPRegs:$src),
612 [(set f32:$dst, (fround f64:$src))]>;
614 // Floating-point Move Instructions, p. 144
615 def FMOVS : F3_3<2, 0b110100, 0b000000001,
616 (outs FPRegs:$dst), (ins FPRegs:$src),
617 "fmovs $src, $dst", []>;
618 def FNEGS : F3_3<2, 0b110100, 0b000000101,
619 (outs FPRegs:$dst), (ins FPRegs:$src),
621 [(set f32:$dst, (fneg f32:$src))]>;
622 def FABSS : F3_3<2, 0b110100, 0b000001001,
623 (outs FPRegs:$dst), (ins FPRegs:$src),
625 [(set f32:$dst, (fabs f32:$src))]>;
628 // Floating-point Square Root Instructions, p.145
629 def FSQRTS : F3_3<2, 0b110100, 0b000101001,
630 (outs FPRegs:$dst), (ins FPRegs:$src),
632 [(set f32:$dst, (fsqrt f32:$src))]>;
633 def FSQRTD : F3_3<2, 0b110100, 0b000101010,
634 (outs DFPRegs:$dst), (ins DFPRegs:$src),
636 [(set f64:$dst, (fsqrt f64:$src))]>;
640 // Floating-point Add and Subtract Instructions, p. 146
641 def FADDS : F3_3<2, 0b110100, 0b001000001,
642 (outs FPRegs:$dst), (ins FPRegs:$src1, FPRegs:$src2),
643 "fadds $src1, $src2, $dst",
644 [(set f32:$dst, (fadd f32:$src1, f32:$src2))]>;
645 def FADDD : F3_3<2, 0b110100, 0b001000010,
646 (outs DFPRegs:$dst), (ins DFPRegs:$src1, DFPRegs:$src2),
647 "faddd $src1, $src2, $dst",
648 [(set f64:$dst, (fadd f64:$src1, f64:$src2))]>;
649 def FSUBS : F3_3<2, 0b110100, 0b001000101,
650 (outs FPRegs:$dst), (ins FPRegs:$src1, FPRegs:$src2),
651 "fsubs $src1, $src2, $dst",
652 [(set f32:$dst, (fsub f32:$src1, f32:$src2))]>;
653 def FSUBD : F3_3<2, 0b110100, 0b001000110,
654 (outs DFPRegs:$dst), (ins DFPRegs:$src1, DFPRegs:$src2),
655 "fsubd $src1, $src2, $dst",
656 [(set f64:$dst, (fsub f64:$src1, f64:$src2))]>;
658 // Floating-point Multiply and Divide Instructions, p. 147
659 def FMULS : F3_3<2, 0b110100, 0b001001001,
660 (outs FPRegs:$dst), (ins FPRegs:$src1, FPRegs:$src2),
661 "fmuls $src1, $src2, $dst",
662 [(set f32:$dst, (fmul f32:$src1, f32:$src2))]>;
663 def FMULD : F3_3<2, 0b110100, 0b001001010,
664 (outs DFPRegs:$dst), (ins DFPRegs:$src1, DFPRegs:$src2),
665 "fmuld $src1, $src2, $dst",
666 [(set f64:$dst, (fmul f64:$src1, f64:$src2))]>;
667 def FSMULD : F3_3<2, 0b110100, 0b001101001,
668 (outs DFPRegs:$dst), (ins FPRegs:$src1, FPRegs:$src2),
669 "fsmuld $src1, $src2, $dst",
670 [(set f64:$dst, (fmul (fextend f32:$src1),
671 (fextend f32:$src2)))]>;
672 def FDIVS : F3_3<2, 0b110100, 0b001001101,
673 (outs FPRegs:$dst), (ins FPRegs:$src1, FPRegs:$src2),
674 "fdivs $src1, $src2, $dst",
675 [(set f32:$dst, (fdiv f32:$src1, f32:$src2))]>;
676 def FDIVD : F3_3<2, 0b110100, 0b001001110,
677 (outs DFPRegs:$dst), (ins DFPRegs:$src1, DFPRegs:$src2),
678 "fdivd $src1, $src2, $dst",
679 [(set f64:$dst, (fdiv f64:$src1, f64:$src2))]>;
681 // Floating-point Compare Instructions, p. 148
682 // Note: the 2nd template arg is different for these guys.
683 // Note 2: the result of a FCMP is not available until the 2nd cycle
684 // after the instr is retired, but there is no interlock. This behavior
685 // is modelled with a forced noop after the instruction.
686 let Defs = [FCC] in {
687 def FCMPS : F3_3<2, 0b110101, 0b001010001,
688 (outs), (ins FPRegs:$src1, FPRegs:$src2),
689 "fcmps $src1, $src2\n\tnop",
690 [(SPcmpfcc f32:$src1, f32:$src2)]>;
691 def FCMPD : F3_3<2, 0b110101, 0b001010010,
692 (outs), (ins DFPRegs:$src1, DFPRegs:$src2),
693 "fcmpd $src1, $src2\n\tnop",
694 [(SPcmpfcc f64:$src1, f64:$src2)]>;
697 //===----------------------------------------------------------------------===//
699 //===----------------------------------------------------------------------===//
701 // V9 Conditional Moves.
702 let Predicates = [HasV9], Constraints = "$T = $dst" in {
703 // Move Integer Register on Condition (MOVcc) p. 194 of the V9 manual.
704 // FIXME: Add instruction encodings for the JIT some day.
705 let Uses = [ICC] in {
707 : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, IntRegs:$F, CCOp:$cc),
708 "mov$cc %icc, $F, $dst",
709 [(set i32:$dst, (SPselecticc i32:$F, i32:$T, imm:$cc))]>;
711 : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, i32imm:$F, CCOp:$cc),
712 "mov$cc %icc, $F, $dst",
713 [(set i32:$dst, (SPselecticc simm11:$F, i32:$T, imm:$cc))]>;
716 let Uses = [FCC] in {
718 : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, IntRegs:$F, CCOp:$cc),
719 "mov$cc %fcc0, $F, $dst",
720 [(set i32:$dst, (SPselectfcc i32:$F, i32:$T, imm:$cc))]>;
722 : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, i32imm:$F, CCOp:$cc),
723 "mov$cc %fcc0, $F, $dst",
724 [(set i32:$dst, (SPselectfcc simm11:$F, i32:$T, imm:$cc))]>;
727 let Uses = [ICC] in {
729 : Pseudo<(outs FPRegs:$dst), (ins FPRegs:$T, FPRegs:$F, CCOp:$cc),
730 "fmovs$cc %icc, $F, $dst",
732 (SPselecticc f32:$F, f32:$T, imm:$cc))]>;
734 : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$T, DFPRegs:$F, CCOp:$cc),
735 "fmovd$cc %icc, $F, $dst",
736 [(set f64:$dst, (SPselecticc f64:$F, f64:$T, imm:$cc))]>;
739 let Uses = [FCC] in {
741 : Pseudo<(outs FPRegs:$dst), (ins FPRegs:$T, FPRegs:$F, CCOp:$cc),
742 "fmovs$cc %fcc0, $F, $dst",
743 [(set f32:$dst, (SPselectfcc f32:$F, f32:$T, imm:$cc))]>;
745 : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$T, DFPRegs:$F, CCOp:$cc),
746 "fmovd$cc %fcc0, $F, $dst",
747 [(set f64:$dst, (SPselectfcc f64:$F, f64:$T, imm:$cc))]>;
752 // Floating-Point Move Instructions, p. 164 of the V9 manual.
753 let Predicates = [HasV9] in {
754 def FMOVD : F3_3<2, 0b110100, 0b000000010,
755 (outs DFPRegs:$dst), (ins DFPRegs:$src),
756 "fmovd $src, $dst", []>;
757 def FNEGD : F3_3<2, 0b110100, 0b000000110,
758 (outs DFPRegs:$dst), (ins DFPRegs:$src),
760 [(set f64:$dst, (fneg f64:$src))]>;
761 def FABSD : F3_3<2, 0b110100, 0b000001010,
762 (outs DFPRegs:$dst), (ins DFPRegs:$src),
764 [(set f64:$dst, (fabs f64:$src))]>;
767 // POPCrr - This does a ctpop of a 64-bit register. As such, we have to clear
768 // the top 32-bits before using it. To do this clearing, we use a SLLri X,0.
769 def POPCrr : F3_1<2, 0b101110,
770 (outs IntRegs:$dst), (ins IntRegs:$src),
771 "popc $src, $dst", []>, Requires<[HasV9]>;
772 def : Pat<(ctpop i32:$src),
773 (POPCrr (SLLri $src, 0))>;
775 //===----------------------------------------------------------------------===//
776 // Non-Instruction Patterns
777 //===----------------------------------------------------------------------===//
780 def : Pat<(i32 simm13:$val),
781 (ORri (i32 G0), imm:$val)>;
782 // Arbitrary immediates.
783 def : Pat<(i32 imm:$val),
784 (ORri (SETHIi (HI22 imm:$val)), (LO10 imm:$val))>;
787 def : Pat<(subc i32:$b, i32:$c),
789 def : Pat<(subc i32:$b, simm13:$val),
790 (SUBCCri $b, imm:$val)>;
792 // Global addresses, constant pool entries
793 def : Pat<(SPhi tglobaladdr:$in), (SETHIi tglobaladdr:$in)>;
794 def : Pat<(SPlo tglobaladdr:$in), (ORri (i32 G0), tglobaladdr:$in)>;
795 def : Pat<(SPhi tconstpool:$in), (SETHIi tconstpool:$in)>;
796 def : Pat<(SPlo tconstpool:$in), (ORri (i32 G0), tconstpool:$in)>;
798 // Add reg, lo. This is used when taking the addr of a global/constpool entry.
799 def : Pat<(add iPTR:$r, (SPlo tglobaladdr:$in)), (ADDri $r, tglobaladdr:$in)>;
800 def : Pat<(add iPTR:$r, (SPlo tconstpool:$in)), (ADDri $r, tconstpool:$in)>;
803 def : Pat<(call tglobaladdr:$dst),
804 (CALL tglobaladdr:$dst)>;
805 def : Pat<(call texternalsym:$dst),
806 (CALL texternalsym:$dst)>;
808 // Map integer extload's to zextloads.
809 def : Pat<(i32 (extloadi1 ADDRrr:$src)), (LDUBrr ADDRrr:$src)>;
810 def : Pat<(i32 (extloadi1 ADDRri:$src)), (LDUBri ADDRri:$src)>;
811 def : Pat<(i32 (extloadi8 ADDRrr:$src)), (LDUBrr ADDRrr:$src)>;
812 def : Pat<(i32 (extloadi8 ADDRri:$src)), (LDUBri ADDRri:$src)>;
813 def : Pat<(i32 (extloadi16 ADDRrr:$src)), (LDUHrr ADDRrr:$src)>;
814 def : Pat<(i32 (extloadi16 ADDRri:$src)), (LDUHri ADDRri:$src)>;
816 // zextload bool -> zextload byte
817 def : Pat<(i32 (zextloadi1 ADDRrr:$src)), (LDUBrr ADDRrr:$src)>;
818 def : Pat<(i32 (zextloadi1 ADDRri:$src)), (LDUBri ADDRri:$src)>;
820 include "SparcInstr64Bit.td"