include "SparcV8InstrFormats.td"
+//===----------------------------------------------------------------------===//
+// Feature predicates.
+//===----------------------------------------------------------------------===//
+
+// HasV9 - This predicate is true when the target processor supports V9
+// instructions. Note that the machine may be running in 32-bit mode.
+def HasV9 : Predicate<"Subtarget.isV9()">;
+
+// HasNoV9 - This predicate is true when the target doesn't have V9
+// instructions. Use of this is just a hack for the isel not having proper
+// costs for V8 instructions that are more expensive than their V9 ones.
+def HasNoV9 : Predicate<"!Subtarget.isV9()">;
+
+// HasVIS - This is true when the target processor has VIS extensions.
+def HasVIS : Predicate<"Subtarget.isVIS()">;
+
+// UseDeprecatedInsts - This predicate is true when the target processor is a
+// V8, or when it is V9 but the V8 deprecated instructions are efficient enough
+// to use when appropriate. In either of these cases, the instruction selector
+// will pick deprecated instructions.
+def UseDeprecatedInsts : Predicate<"Subtarget.useDeprecatedV8Instructions()">;
+
//===----------------------------------------------------------------------===//
// Instruction Pattern Stuff
//===----------------------------------------------------------------------===//
+def simm11 : PatLeaf<(imm), [{
+ // simm11 predicate - True if the imm fits in a 11-bit sign extended field.
+ return (((int)N->getValue() << (32-11)) >> (32-11)) == (int)N->getValue();
+}]>;
+
def simm13 : PatLeaf<(imm), [{
// simm13 predicate - True if the imm fits in a 13-bit sign extended field.
return (((int)N->getValue() << (32-13)) >> (32-13)) == (int)N->getValue();
def brtarget : Operand<OtherVT>;
def calltarget : Operand<i32>;
+// Operand for printing out a condition code.
+let PrintMethod = "printV8CCOperand" in
+ def V8CC : Operand<i32>;
+
def SDTV8cmpfcc :
SDTypeProfile<1, 2, [SDTCisVT<0, FlagVT>, SDTCisFP<1>, SDTCisSameAs<1, 2>]>;
def SDTV8brcc :
-SDTypeProfile<0, 3, [SDTCisVT<0, OtherVT>, SDTCisVT<1, OtherVT>,
+SDTypeProfile<0, 3, [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
SDTCisVT<2, FlagVT>]>;
def SDTV8selectcc :
SDTypeProfile<1, 4, [SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>,
def SDTV8ITOF :
SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisVT<1, f32>]>;
-def V8cmpicc : SDNode<"V8ISD::CMPICC", SDTIntBinOp,
- [SDNPCommutative, SDNPOutFlag]>;
-def V8cmpfcc : SDNode<"V8ISD::CMPFCC", SDTV8cmpfcc>;
+def V8cmpicc : SDNode<"V8ISD::CMPICC", SDTIntBinOp, [SDNPOutFlag]>;
+def V8cmpfcc : SDNode<"V8ISD::CMPFCC", SDTV8cmpfcc, [SDNPOutFlag]>;
def V8bricc : SDNode<"V8ISD::BRICC", SDTV8brcc, [SDNPHasChain]>;
def V8brfcc : SDNode<"V8ISD::BRFCC", SDTV8brcc, [SDNPHasChain]>;
def callseq_end : SDNode<"ISD::CALLSEQ_END", SDT_V8CallSeq, [SDNPHasChain]>;
def SDT_V8Call : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
-def call : SDNode<"ISD::CALL", SDT_V8Call,
+def call : SDNode<"V8ISD::CALL", SDT_V8Call,
[SDNPHasChain, SDNPOptInFlag, SDNPOutFlag]>;
def SDT_V8RetFlag : SDTypeProfile<0, 0, []>;
def retflag : SDNode<"V8ISD::RET_FLAG", SDT_V8RetFlag,
[SDNPHasChain, SDNPOptInFlag]>;
+//===----------------------------------------------------------------------===//
+// SPARC Flag Conditions
+//===----------------------------------------------------------------------===//
+
+// Note that these values must be kept in sync with the V8CC::CondCode enum
+// values.
+class ICC_VAL<int N> : PatLeaf<(i32 N)>;
+def ICC_NE : ICC_VAL< 9>; // Not Equal
+def ICC_E : ICC_VAL< 1>; // Equal
+def ICC_G : ICC_VAL<10>; // Greater
+def ICC_LE : ICC_VAL< 2>; // Less or Equal
+def ICC_GE : ICC_VAL<11>; // Greater or Equal
+def ICC_L : ICC_VAL< 3>; // Less
+def ICC_GU : ICC_VAL<12>; // Greater Unsigned
+def ICC_LEU : ICC_VAL< 4>; // Less or Equal Unsigned
+def ICC_CC : ICC_VAL<13>; // Carry Clear/Great or Equal Unsigned
+def ICC_CS : ICC_VAL< 5>; // Carry Set/Less Unsigned
+def ICC_POS : ICC_VAL<14>; // Positive
+def ICC_NEG : ICC_VAL< 6>; // Negative
+def ICC_VC : ICC_VAL<15>; // Overflow Clear
+def ICC_VS : ICC_VAL< 7>; // Overflow Set
+
+class FCC_VAL<int N> : PatLeaf<(i32 N)>;
+def FCC_U : FCC_VAL<23>; // Unordered
+def FCC_G : FCC_VAL<22>; // Greater
+def FCC_UG : FCC_VAL<21>; // Unordered or Greater
+def FCC_L : FCC_VAL<20>; // Less
+def FCC_UL : FCC_VAL<19>; // Unordered or Less
+def FCC_LG : FCC_VAL<18>; // Less or Greater
+def FCC_NE : FCC_VAL<17>; // Not Equal
+def FCC_E : FCC_VAL<25>; // Equal
+def FCC_UE : FCC_VAL<24>; // Unordered or Equal
+def FCC_GE : FCC_VAL<25>; // Greater or Equal
+def FCC_UGE : FCC_VAL<26>; // Unordered or Greater or Equal
+def FCC_LE : FCC_VAL<27>; // Less or Equal
+def FCC_ULE : FCC_VAL<28>; // Unordered or Less or Equal
+def FCC_O : FCC_VAL<29>; // Ordered
+
+
//===----------------------------------------------------------------------===//
// Instructions
//===----------------------------------------------------------------------===//
class Pseudo<dag ops, string asmstr, list<dag> pattern>
: InstV8<ops, asmstr, pattern>;
-def PHI : Pseudo<(ops variable_ops), "PHI", []>;
def ADJCALLSTACKDOWN : Pseudo<(ops i32imm:$amt),
"!ADJCALLSTACKDOWN $amt",
[(callseq_start imm:$amt)]>;
// FpMOVD/FpNEGD/FpABSD - These are lowered to single-precision ops by the
// fpmover pass.
-def FpMOVD : Pseudo<(ops DFPRegs:$dst, DFPRegs:$src),
- "!FpMOVD $src, $dst", []>; // pseudo 64-bit double move
-def FpNEGD : Pseudo<(ops DFPRegs:$dst, DFPRegs:$src),
- "!FpNEGD $src, $dst",
- [(set DFPRegs:$dst, (fneg DFPRegs:$src))]>;
-def FpABSD : Pseudo<(ops DFPRegs:$dst, DFPRegs:$src),
- "!FpABSD $src, $dst",
- [(set DFPRegs:$dst, (fabs DFPRegs:$src))]>;
+let Predicates = [HasNoV9] in { // Only emit these in V8 mode.
+ def FpMOVD : Pseudo<(ops DFPRegs:$dst, DFPRegs:$src),
+ "!FpMOVD $src, $dst", []>;
+ def FpNEGD : Pseudo<(ops DFPRegs:$dst, DFPRegs:$src),
+ "!FpNEGD $src, $dst",
+ [(set DFPRegs:$dst, (fneg DFPRegs:$src))]>;
+ def FpABSD : Pseudo<(ops DFPRegs:$dst, DFPRegs:$src),
+ "!FpABSD $src, $dst",
+ [(set DFPRegs:$dst, (fabs DFPRegs:$src))]>;
+}
// SELECT_CC_* - Used to implement the SELECT_CC DAG operation. Expanded by the
// scheduler into a branch sequence. This has to handle all permutations of
// selection between i32/f32/f64 on ICC and FCC.
-let usesCustomDAGSchedInserter = 1 in { // Expanded by the scheduler.
+let usesCustomDAGSchedInserter = 1, // Expanded by the scheduler.
+ Predicates = [HasNoV9] in { // V9 has conditional moves
def SELECT_CC_Int_ICC
: Pseudo<(ops IntRegs:$dst, IntRegs:$T, IntRegs:$F, i32imm:$Cond),
"; SELECT_CC_Int_ICC PSEUDO!",
imm:$Cond, FCC))]>;
}
+
// Section A.3 - Synthetic Instructions, p. 85
// special cases of JMPL:
let isReturn = 1, isTerminator = 1, hasDelaySlot = 1, noResults = 1 in {
def BA : BranchV8<0b1000, (ops brtarget:$dst),
"ba $dst",
[(br bb:$dst)]>;
-def BNE : BranchV8<0b1001, (ops brtarget:$dst),
- "bne $dst",
- [(V8bricc bb:$dst, SETNE, ICC)]>;
-def BE : BranchV8<0b0001, (ops brtarget:$dst),
- "be $dst",
- [(V8bricc bb:$dst, SETEQ, ICC)]>;
-def BG : BranchV8<0b1010, (ops brtarget:$dst),
- "bg $dst",
- [(V8bricc bb:$dst, SETGT, ICC)]>;
-def BLE : BranchV8<0b0010, (ops brtarget:$dst),
- "ble $dst",
- [(V8bricc bb:$dst, SETLE, ICC)]>;
-def BGE : BranchV8<0b1011, (ops brtarget:$dst),
- "bge $dst",
- [(V8bricc bb:$dst, SETGE, ICC)]>;
-def BL : BranchV8<0b0011, (ops brtarget:$dst),
- "bl $dst",
- [(V8bricc bb:$dst, SETLT, ICC)]>;
-def BGU : BranchV8<0b1100, (ops brtarget:$dst),
- "bgu $dst",
- [(V8bricc bb:$dst, SETUGT, ICC)]>;
-def BLEU : BranchV8<0b0100, (ops brtarget:$dst),
- "bleu $dst",
- [(V8bricc bb:$dst, SETULE, ICC)]>;
-def BCC : BranchV8<0b1101, (ops brtarget:$dst),
- "bcc $dst",
- [(V8bricc bb:$dst, SETUGE, ICC)]>;
-def BCS : BranchV8<0b0101, (ops brtarget:$dst),
- "bcs $dst",
- [(V8bricc bb:$dst, SETULT, ICC)]>;
+
+// FIXME: the encoding for the JIT should look at the condition field.
+def BCOND : BranchV8<0, (ops brtarget:$dst, V8CC:$cc),
+ "b$cc $dst",
+ [(V8bricc bb:$dst, imm:$cc, ICC)]>;
+
// Section B.22 - Branch on Floating-point Condition Codes Instructions, p. 121
let noResults = 1;
}
-def FBU : FPBranchV8<0b0111, (ops brtarget:$dst),
- "fbu $dst",
- [(V8brfcc bb:$dst, SETUO, FCC)]>;
-def FBG : FPBranchV8<0b0110, (ops brtarget:$dst),
- "fbg $dst",
- [(V8brfcc bb:$dst, SETGT, FCC)]>;
-def FBUG : FPBranchV8<0b0101, (ops brtarget:$dst),
- "fbug $dst",
- [(V8brfcc bb:$dst, SETUGT, FCC)]>;
-def FBL : FPBranchV8<0b0100, (ops brtarget:$dst),
- "fbl $dst",
- [(V8brfcc bb:$dst, SETLT, FCC)]>;
-def FBUL : FPBranchV8<0b0011, (ops brtarget:$dst),
- "fbul $dst",
- [(V8brfcc bb:$dst, SETULT, FCC)]>;
-def FBLG : FPBranchV8<0b0010, (ops brtarget:$dst),
- "fblg $dst",
- [(V8brfcc bb:$dst, SETONE, FCC)]>;
-def FBNE : FPBranchV8<0b0001, (ops brtarget:$dst),
- "fbne $dst",
- [(V8brfcc bb:$dst, SETNE, FCC)]>;
-def FBE : FPBranchV8<0b1001, (ops brtarget:$dst),
- "fbe $dst",
- [(V8brfcc bb:$dst, SETEQ, FCC)]>;
-def FBUE : FPBranchV8<0b1010, (ops brtarget:$dst),
- "fbue $dst",
- [(V8brfcc bb:$dst, SETUEQ, FCC)]>;
-def FBGE : FPBranchV8<0b1011, (ops brtarget:$dst),
- "fbge $dst",
- [(V8brfcc bb:$dst, SETGE, FCC)]>;
-def FBUGE: FPBranchV8<0b1100, (ops brtarget:$dst),
- "fbuge $dst",
- [(V8brfcc bb:$dst, SETUGE, FCC)]>;
-def FBLE : FPBranchV8<0b1101, (ops brtarget:$dst),
- "fble $dst",
- [(V8brfcc bb:$dst, SETLE, FCC)]>;
-def FBULE: FPBranchV8<0b1110, (ops brtarget:$dst),
- "fbule $dst",
- [(V8brfcc bb:$dst, SETULE, FCC)]>;
-def FBO : FPBranchV8<0b1111, (ops brtarget:$dst),
- "fbo $dst",
- [(V8brfcc bb:$dst, SETO, FCC)]>;
-
+// FIXME: the encoding for the JIT should look at the condition field.
+def FBCOND : FPBranchV8<0, (ops brtarget:$dst, V8CC:$cc),
+ "fb$cc $dst",
+ [(V8brfcc bb:$dst, imm:$cc, FCC)]>;
// Section B.24 - Call and Link Instruction, p. 125
"fcmpd $src1, $src2\n\tnop",
[(set FCC, (V8cmpfcc DFPRegs:$src1, DFPRegs:$src2))]>;
+
+//===----------------------------------------------------------------------===//
+// V9 Instructions
+//===----------------------------------------------------------------------===//
+
+// V9 Conditional Moves.
+let Predicates = [HasV9], isTwoAddress = 1 in {
+ // Move Integer Register on Condition (MOVcc) p. 194 of the V9 manual.
+ // FIXME: Add instruction encodings for the JIT some day.
+ def MOVICCrr
+ : Pseudo<(ops IntRegs:$dst, IntRegs:$T, IntRegs:$F, V8CC:$cc),
+ "mov$cc %icc, $F, $dst",
+ [(set IntRegs:$dst,
+ (V8selecticc IntRegs:$F, IntRegs:$T, imm:$cc, ICC))]>;
+ def MOVICCri
+ : Pseudo<(ops IntRegs:$dst, IntRegs:$T, i32imm:$F, V8CC:$cc),
+ "mov$cc %icc, $F, $dst",
+ [(set IntRegs:$dst,
+ (V8selecticc simm11:$F, IntRegs:$T, imm:$cc, ICC))]>;
+
+ def MOVFCCrr
+ : Pseudo<(ops IntRegs:$dst, IntRegs:$T, IntRegs:$F, V8CC:$cc),
+ "movf$cc %fcc, $F, $dst",
+ [(set IntRegs:$dst,
+ (V8selectfcc IntRegs:$F, IntRegs:$T, imm:$cc, FCC))]>;
+ def MOVFCCri
+ : Pseudo<(ops IntRegs:$dst, IntRegs:$T, i32imm:$F, V8CC:$cc),
+ "movf$cc %fcc, $F, $dst",
+ [(set IntRegs:$dst,
+ (V8selectfcc simm11:$F, IntRegs:$T, imm:$cc, FCC))]>;
+
+ def FMOVS_ICC
+ : Pseudo<(ops FPRegs:$dst, FPRegs:$T, FPRegs:$F, V8CC:$cc),
+ "fmovs$cc %icc, $F, $dst",
+ [(set FPRegs:$dst,
+ (V8selecticc FPRegs:$F, FPRegs:$T, imm:$cc, ICC))]>;
+ def FMOVD_ICC
+ : Pseudo<(ops DFPRegs:$dst, DFPRegs:$T, DFPRegs:$F, V8CC:$cc),
+ "fmovd$cc %icc, $F, $dst",
+ [(set DFPRegs:$dst,
+ (V8selecticc DFPRegs:$F, DFPRegs:$T, imm:$cc, ICC))]>;
+ def FMOVS_FCC
+ : Pseudo<(ops FPRegs:$dst, FPRegs:$T, FPRegs:$F, V8CC:$cc),
+ "fmovs$cc %fcc, $F, $dst",
+ [(set FPRegs:$dst,
+ (V8selectfcc FPRegs:$F, FPRegs:$T, imm:$cc, FCC))]>;
+ def FMOVD_FCC
+ : Pseudo<(ops DFPRegs:$dst, DFPRegs:$T, DFPRegs:$F, V8CC:$cc),
+ "fmovd$cc %fcc, $F, $dst",
+ [(set DFPRegs:$dst,
+ (V8selectfcc DFPRegs:$F, DFPRegs:$T, imm:$cc, FCC))]>;
+
+}
+
+// Floating-Point Move Instructions, p. 164 of the V9 manual.
+let Predicates = [HasV9] in {
+ def FMOVD : F3_3<2, 0b110100, 0b000000010,
+ (ops DFPRegs:$dst, DFPRegs:$src),
+ "fmovd $src, $dst", []>;
+ def FNEGD : F3_3<2, 0b110100, 0b000000110,
+ (ops DFPRegs:$dst, DFPRegs:$src),
+ "fnegd $src, $dst",
+ [(set DFPRegs:$dst, (fneg DFPRegs:$src))]>;
+ def FABSD : F3_3<2, 0b110100, 0b000001010,
+ (ops DFPRegs:$dst, DFPRegs:$src),
+ "fabsd $src, $dst",
+ [(set DFPRegs:$dst, (fabs DFPRegs:$src))]>;
+}
+
+// POPCrr - This does a ctpop of a 64-bit register. As such, we have to clear
+// the top 32-bits before using it. To do this clearing, we use a SLLri X,0.
+def POPCrr : F3_1<2, 0b101110,
+ (ops IntRegs:$dst, IntRegs:$src),
+ "popc $src, $dst", []>, Requires<[HasV9]>;
+def : Pat<(ctpop IntRegs:$src),
+ (POPCrr (SLLri IntRegs:$src, 0))>;
+
//===----------------------------------------------------------------------===//
// Non-Instruction Patterns
//===----------------------------------------------------------------------===//
def : Pat<(V8hi tconstpool:$in), (SETHIi tconstpool:$in)>;
def : Pat<(V8lo tconstpool:$in), (ORri G0, tconstpool:$in)>;
+// Add reg, lo. This is used when taking the addr of a global/constpool entry.
+def : Pat<(add IntRegs:$r, (V8lo tglobaladdr:$in)),
+ (ADDri IntRegs:$r, tglobaladdr:$in)>;
+def : Pat<(add IntRegs:$r, (V8lo tconstpool:$in)),
+ (ADDri IntRegs:$r, tconstpool:$in)>;
+
+
// Calls:
def : Pat<(call tglobaladdr:$dst),
(CALL tglobaladdr:$dst)>;
projects / oota-llvm.git / blobdiff