//===----------------------------------------------------------------------===//
// EH Pseudo Instructions
//
+let SchedRW = [WriteSystem] in {
let isTerminator = 1, isReturn = 1, isBarrier = 1,
hasCtrlDep = 1, isCodeGenOnly = 1 in {
def EH_RETURN : I<0xC3, RawFrm, (outs), (ins GR32:$addr),
"ret\t#eh_return, addr: $addr",
- [(X86ehret GR32:$addr)], IIC_RET>;
+ [(X86ehret GR32:$addr)], IIC_RET>, Sched<[WriteJumpLd]>;
}
hasCtrlDep = 1, isCodeGenOnly = 1 in {
def EH_RETURN64 : I<0xC3, RawFrm, (outs), (ins GR64:$addr),
"ret\t#eh_return, addr: $addr",
- [(X86ehret GR64:$addr)], IIC_RET>;
+ [(X86ehret GR64:$addr)], IIC_RET>, Sched<[WriteJumpLd]>;
}
Requires<[In64BitMode]>;
}
}
+} // SchedRW
let isBranch = 1, isTerminator = 1, isCodeGenOnly = 1 in {
def EH_SjLj_Setup : I<0, Pseudo, (outs), (ins brtarget:$dst),
// Alias Instructions
//===----------------------------------------------------------------------===//
-// Alias instructions that map movr0 to xor.
+// Alias instruction mapping movr0 to xor.
// FIXME: remove when we can teach regalloc that xor reg, reg is ok.
// FIXME: Set encoding to pseudo.
let Defs = [EFLAGS], isReMaterializable = 1, isAsCheapAsAMove = 1,
- isCodeGenOnly = 1 in {
-def MOV8r0 : I<0x30, MRMInitReg, (outs GR8 :$dst), (ins), "",
- [(set GR8:$dst, 0)], IIC_ALU_NONMEM>;
-
-// We want to rewrite MOV16r0 in terms of MOV32r0, because it's a smaller
-// encoding and avoids a partial-register update sometimes, but doing so
-// at isel time interferes with rematerialization in the current register
-// allocator. For now, this is rewritten when the instruction is lowered
-// to an MCInst.
-def MOV16r0 : I<0x31, MRMInitReg, (outs GR16:$dst), (ins),
- "",
- [(set GR16:$dst, 0)], IIC_ALU_NONMEM>, OpSize;
-
-// FIXME: Set encoding to pseudo.
+ isCodeGenOnly = 1 in
def MOV32r0 : I<0x31, MRMInitReg, (outs GR32:$dst), (ins), "",
- [(set GR32:$dst, 0)], IIC_ALU_NONMEM>;
+ [(set GR32:$dst, 0)], IIC_ALU_NONMEM>, Sched<[WriteZero]>;
+
+// Other widths can also make use of the 32-bit xor, which may have a smaller
+// encoding and avoid partial register updates.
+def : Pat<(i8 0), (EXTRACT_SUBREG (MOV32r0), sub_8bit)>;
+def : Pat<(i16 0), (EXTRACT_SUBREG (MOV32r0), sub_16bit)>;
+def : Pat<(i64 0), (SUBREG_TO_REG (i64 0), (MOV32r0), sub_32bit)> {
+ let AddedComplexity = 20;
}
-// We want to rewrite MOV64r0 in terms of MOV32r0, because it's sometimes a
-// smaller encoding, but doing so at isel time interferes with rematerialization
-// in the current register allocator. For now, this is rewritten when the
-// instruction is lowered to an MCInst.
-// FIXME: AddedComplexity gives this a higher priority than MOV64ri32. Remove
-// when we have a better way to specify isel priority.
-let Defs = [EFLAGS], isCodeGenOnly=1,
- AddedComplexity = 1, isReMaterializable = 1, isAsCheapAsAMove = 1 in
-def MOV64r0 : I<0x31, MRMInitReg, (outs GR64:$dst), (ins), "",
- [(set GR64:$dst, 0)], IIC_ALU_NONMEM>;
-
// Materialize i64 constant where top 32-bits are zero. This could theoretically
// use MOV32ri with a SUBREG_TO_REG to represent the zero-extension, however
// that would make it more difficult to rematerialize.
let AddedComplexity = 1, isReMaterializable = 1, isAsCheapAsAMove = 1,
- isCodeGenOnly = 1 in
-def MOV64ri64i32 : Ii32<0xB8, AddRegFrm, (outs GR64:$dst), (ins i64i32imm:$src),
- "", [(set GR64:$dst, i64immZExt32:$src)],
- IIC_ALU_NONMEM>;
+ isCodeGenOnly = 1, neverHasSideEffects = 1 in
+def MOV32ri64 : Ii32<0xb8, AddRegFrm, (outs GR32:$dst), (ins i64i32imm:$src),
+ "", [], IIC_ALU_NONMEM>, Sched<[WriteALU]>;
+
+// This 64-bit pseudo-move can be used for both a 64-bit constant that is
+// actually the zero-extension of a 32-bit constant, and for labels in the
+// x86-64 small code model.
+def mov64imm32 : ComplexPattern<i64, 1, "SelectMOV64Imm32", [imm, X86Wrapper]>;
+
+let AddedComplexity = 1 in
+def : Pat<(i64 mov64imm32:$src),
+ (SUBREG_TO_REG (i64 0), (MOV32ri64 mov64imm32:$src), sub_32bit)>;
// Use sbb to materialize carry bit.
-let Uses = [EFLAGS], Defs = [EFLAGS], isPseudo = 1 in {
+let Uses = [EFLAGS], Defs = [EFLAGS], isPseudo = 1, SchedRW = [WriteALU] in {
// FIXME: These are pseudo ops that should be replaced with Pat<> patterns.
// However, Pat<> can't replicate the destination reg into the inputs of the
// result.
//===----------------------------------------------------------------------===//
// String Pseudo Instructions
//
+let SchedRW = [WriteMicrocoded] in {
let Defs = [ECX,EDI,ESI], Uses = [ECX,EDI,ESI], isCodeGenOnly = 1 in {
def REP_MOVSB_32 : I<0xA4, RawFrm, (outs), (ins), "{rep;movsb|rep movsb}",
[(X86rep_movs i8)], IIC_REP_MOVS>, REP,
[(X86rep_stos i64)], IIC_REP_STOS>, REP,
Requires<[In64BitMode]>;
}
+} // SchedRW
//===----------------------------------------------------------------------===//
// Thread Local Storage Instructions
multiclass PSEUDO_ATOMIC_LOAD_BINOP<string mnemonic> {
let usesCustomInserter = 1, mayLoad = 1, mayStore = 1 in {
+ let Defs = [EFLAGS, AL] in
def NAME#8 : I<0, Pseudo, (outs GR8:$dst),
(ins i8mem:$ptr, GR8:$val),
!strconcat(mnemonic, "8 PSEUDO!"), []>;
+ let Defs = [EFLAGS, AX] in
def NAME#16 : I<0, Pseudo,(outs GR16:$dst),
(ins i16mem:$ptr, GR16:$val),
!strconcat(mnemonic, "16 PSEUDO!"), []>;
+ let Defs = [EFLAGS, EAX] in
def NAME#32 : I<0, Pseudo, (outs GR32:$dst),
(ins i32mem:$ptr, GR32:$val),
!strconcat(mnemonic, "32 PSEUDO!"), []>;
+ let Defs = [EFLAGS, RAX] in
def NAME#64 : I<0, Pseudo, (outs GR64:$dst),
(ins i64mem:$ptr, GR64:$val),
!strconcat(mnemonic, "64 PSEUDO!"), []>;
defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMUMIN", "atomic_load_umin">;
multiclass PSEUDO_ATOMIC_LOAD_BINOP6432<string mnemonic> {
- let usesCustomInserter = 1, mayLoad = 1, mayStore = 1, hasSideEffects = 0 in
+ let usesCustomInserter = 1, Defs = [EFLAGS, EAX, EDX],
+ mayLoad = 1, mayStore = 1, hasSideEffects = 0 in
def NAME#6432 : I<0, Pseudo, (outs GR32:$dst1, GR32:$dst2),
(ins i64mem:$ptr, GR32:$val1, GR32:$val2),
!strconcat(mnemonic, "6432 PSEUDO!"), []>;
let isCodeGenOnly = 1, Defs = [EFLAGS] in
def OR32mrLocked : I<0x09, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$zero),
"or{l}\t{$zero, $dst|$dst, $zero}",
- [], IIC_ALU_MEM>, Requires<[In32BitMode]>, LOCK;
+ [], IIC_ALU_MEM>, Requires<[In32BitMode]>, LOCK,
+ Sched<[WriteALULd, WriteRMW]>;
let hasSideEffects = 1 in
def Int_MemBarrier : I<0, Pseudo, (outs), (ins),
"#MEMBARRIER",
- [(X86MemBarrier)]>;
+ [(X86MemBarrier)]>, Sched<[WriteLoad]>;
// RegOpc corresponds to the mr version of the instruction
// ImmOpc corresponds to the mi version of the instruction
// ImmMod corresponds to the instruction format of the mi and mi8 versions
multiclass LOCK_ArithBinOp<bits<8> RegOpc, bits<8> ImmOpc, bits<8> ImmOpc8,
Format ImmMod, string mnemonic> {
-let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1 in {
+let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1,
+ SchedRW = [WriteALULd, WriteRMW] in {
def NAME#8mr : I<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4},
RegOpc{3}, RegOpc{2}, RegOpc{1}, 0 },
// Optimized codegen when the non-memory output is not used.
multiclass LOCK_ArithUnOp<bits<8> Opc8, bits<8> Opc, Format Form,
string mnemonic> {
-let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1 in {
+let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1,
+ SchedRW = [WriteALULd, WriteRMW] in {
def NAME#8m : I<Opc8, Form, (outs), (ins i8mem :$dst),
!strconcat(mnemonic, "{b}\t$dst"),
multiclass LCMPXCHG_BinOp<bits<8> Opc8, bits<8> Opc, Format Form,
string mnemonic, SDPatternOperator frag,
InstrItinClass itin8, InstrItinClass itin> {
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, SchedRW = [WriteALULd, WriteRMW] in {
let Defs = [AL, EFLAGS], Uses = [AL] in
def NAME#8 : I<Opc8, Form, (outs), (ins i8mem:$ptr, GR8:$swap),
!strconcat(mnemonic, "{b}\t{$swap, $ptr|$ptr, $swap}"),
}
}
-let Defs = [EAX, EDX, EFLAGS], Uses = [EAX, EBX, ECX, EDX] in {
+let Defs = [EAX, EDX, EFLAGS], Uses = [EAX, EBX, ECX, EDX],
+ SchedRW = [WriteALULd, WriteRMW] in {
defm LCMPXCHG8B : LCMPXCHG_UnOp<0xC7, MRM1m, "cmpxchg8b",
X86cas8, i64mem,
IIC_CMPX_LOCK_8B>;
}
let Defs = [RAX, RDX, EFLAGS], Uses = [RAX, RBX, RCX, RDX],
- Predicates = [HasCmpxchg16b] in {
+ Predicates = [HasCmpxchg16b], SchedRW = [WriteALULd, WriteRMW] in {
defm LCMPXCHG16B : LCMPXCHG_UnOp<0xC7, MRM1m, "cmpxchg16b",
X86cas16, i128mem,
IIC_CMPX_LOCK_16B>, REX_W;
multiclass ATOMIC_LOAD_BINOP<bits<8> opc8, bits<8> opc, string mnemonic,
string frag,
InstrItinClass itin8, InstrItinClass itin> {
- let Constraints = "$val = $dst", Defs = [EFLAGS], isCodeGenOnly = 1 in {
+ let Constraints = "$val = $dst", Defs = [EFLAGS], isCodeGenOnly = 1,
+ SchedRW = [WriteALULd, WriteRMW] in {
def NAME#8 : I<opc8, MRMSrcMem, (outs GR8:$dst),
(ins GR8:$val, i8mem:$ptr),
!strconcat(mnemonic, "{b}\t{$val, $ptr|$ptr, $val}"),
def : Pat<(i64 (X86Wrapper tblockaddress:$dst)),
(MOV64ri tblockaddress:$dst)>, Requires<[FarData]>;
-// In static codegen with small code model, we can get the address of a label
-// into a register with 'movl'. FIXME: This is a hack, the 'imm' predicate of
-// the MOV64ri64i32 should accept these.
-def : Pat<(i64 (X86Wrapper tconstpool :$dst)),
- (MOV64ri64i32 tconstpool :$dst)>, Requires<[SmallCode]>;
-def : Pat<(i64 (X86Wrapper tjumptable :$dst)),
- (MOV64ri64i32 tjumptable :$dst)>, Requires<[SmallCode]>;
-def : Pat<(i64 (X86Wrapper tglobaladdr :$dst)),
- (MOV64ri64i32 tglobaladdr :$dst)>, Requires<[SmallCode]>;
-def : Pat<(i64 (X86Wrapper texternalsym:$dst)),
- (MOV64ri64i32 texternalsym:$dst)>, Requires<[SmallCode]>;
-def : Pat<(i64 (X86Wrapper tblockaddress:$dst)),
- (MOV64ri64i32 tblockaddress:$dst)>, Requires<[SmallCode]>;
-
// In kernel code model, we can get the address of a label
// into a register with 'movq'. FIXME: This is a hack, the 'imm' predicate of
// the MOV64ri32 should accept these.
// This corresponds to add $foo@tpoff, %rax
def : Pat<(add GR64:$src1, (X86Wrapper tglobaltlsaddr :$dst)),
(ADD64ri32 GR64:$src1, tglobaltlsaddr :$dst)>;
-// This corresponds to mov foo@tpoff(%rbx), %eax
-def : Pat<(load (i64 (X86Wrapper tglobaltlsaddr :$dst))),
- (MOV64rm tglobaltlsaddr :$dst)>;
// Direct PC relative function call for small code model. 32-bit displacement
// inverted.
multiclass CMOVmr<PatLeaf InvertedCond, Instruction Inst16, Instruction Inst32,
Instruction Inst64> {
- def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, InvertedCond, EFLAGS),
- (Inst16 GR16:$src2, addr:$src1)>;
- def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, InvertedCond, EFLAGS),
- (Inst32 GR32:$src2, addr:$src1)>;
- def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, InvertedCond, EFLAGS),
- (Inst64 GR64:$src2, addr:$src1)>;
+ let Predicates = [HasCMov] in {
+ def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, InvertedCond, EFLAGS),
+ (Inst16 GR16:$src2, addr:$src1)>;
+ def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, InvertedCond, EFLAGS),
+ (Inst32 GR32:$src2, addr:$src1)>;
+ def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, InvertedCond, EFLAGS),
+ (Inst64 GR64:$src2, addr:$src1)>;
+ }
}
defm : CMOVmr<X86_COND_B , CMOVAE16rm, CMOVAE32rm, CMOVAE64rm>;
def : Pat<(zextloadi8i1 addr:$src), (MOV8rm addr:$src)>;
def : Pat<(zextloadi16i1 addr:$src), (MOVZX16rm8 addr:$src)>;
def : Pat<(zextloadi32i1 addr:$src), (MOVZX32rm8 addr:$src)>;
-def : Pat<(zextloadi64i1 addr:$src), (MOVZX64rm8 addr:$src)>;
+def : Pat<(zextloadi64i1 addr:$src),
+ (SUBREG_TO_REG (i64 0), (MOVZX32rm8 addr:$src), sub_32bit)>;
// extload bool -> extload byte
// When extloading from 16-bit and smaller memory locations into 64-bit
def : Pat<(extloadi32i8 addr:$src), (MOVZX32rm8 addr:$src)>;
def : Pat<(extloadi32i16 addr:$src), (MOVZX32rm16 addr:$src)>;
-def : Pat<(extloadi64i1 addr:$src), (MOVZX64rm8 addr:$src)>;
-def : Pat<(extloadi64i8 addr:$src), (MOVZX64rm8 addr:$src)>;
-def : Pat<(extloadi64i16 addr:$src), (MOVZX64rm16 addr:$src)>;
// For other extloads, use subregs, since the high contents of the register are
// defined after an extload.
+def : Pat<(extloadi64i1 addr:$src),
+ (SUBREG_TO_REG (i64 0), (MOVZX32rm8 addr:$src), sub_32bit)>;
+def : Pat<(extloadi64i8 addr:$src),
+ (SUBREG_TO_REG (i64 0), (MOVZX32rm8 addr:$src), sub_32bit)>;
+def : Pat<(extloadi64i16 addr:$src),
+ (SUBREG_TO_REG (i64 0), (MOVZX32rm16 addr:$src), sub_32bit)>;
def : Pat<(extloadi64i32 addr:$src),
- (SUBREG_TO_REG (i64 0), (MOV32rm addr:$src),
- sub_32bit)>;
+ (SUBREG_TO_REG (i64 0), (MOV32rm addr:$src), sub_32bit)>;
// anyext. Define these to do an explicit zero-extend to
// avoid partial-register updates.
def : Pat<(i32 (anyext GR16:$src)),
(INSERT_SUBREG (i32 (IMPLICIT_DEF)), GR16:$src, sub_16bit)>;
-def : Pat<(i64 (anyext GR8 :$src)), (MOVZX64rr8 GR8 :$src)>;
-def : Pat<(i64 (anyext GR16:$src)), (MOVZX64rr16 GR16 :$src)>;
+def : Pat<(i64 (anyext GR8 :$src)),
+ (SUBREG_TO_REG (i64 0), (MOVZX32rr8 GR8 :$src), sub_32bit)>;
+def : Pat<(i64 (anyext GR16:$src)),
+ (SUBREG_TO_REG (i64 0), (MOVZX32rr16 GR16 :$src), sub_32bit)>;
def : Pat<(i64 (anyext GR32:$src)),
(SUBREG_TO_REG (i64 0), GR32:$src, sub_32bit)>;
// (or x1, x2) -> (add x1, x2) if two operands are known not to share bits.
-let AddedComplexity = 5 in { // Try this before the selecting to OR
+// Try this before the selecting to OR.
+let AddedComplexity = 5, SchedRW = [WriteALU] in {
let isConvertibleToThreeAddress = 1,
Constraints = "$src1 = $dst", Defs = [EFLAGS] in {
[(set GR64:$dst, (or_is_add GR64:$src1,
i64immSExt32:$src2))]>;
}
-} // AddedComplexity
+} // AddedComplexity, SchedRW
//===----------------------------------------------------------------------===//
// r & (2^32-1) ==> movz
def : Pat<(and GR64:$src, 0x00000000FFFFFFFF),
- (MOVZX64rr32 (EXTRACT_SUBREG GR64:$src, sub_32bit))>;
+ (SUBREG_TO_REG (i64 0),
+ (MOV32rr (EXTRACT_SUBREG GR64:$src, sub_32bit)),
+ sub_32bit)>;
// r & (2^16-1) ==> movz
def : Pat<(and GR64:$src, 0xffff),
- (MOVZX64rr16 (i16 (EXTRACT_SUBREG GR64:$src, sub_16bit)))>;
+ (SUBREG_TO_REG (i64 0),
+ (MOVZX32rr16 (i16 (EXTRACT_SUBREG GR64:$src, sub_16bit))),
+ sub_32bit)>;
// r & (2^8-1) ==> movz
def : Pat<(and GR64:$src, 0xff),
- (MOVZX64rr8 (i8 (EXTRACT_SUBREG GR64:$src, sub_8bit)))>;
+ (SUBREG_TO_REG (i64 0),
+ (MOVZX32rr8 (i8 (EXTRACT_SUBREG GR64:$src, sub_8bit))),
+ sub_32bit)>;
// r & (2^8-1) ==> movz
def : Pat<(and GR32:$src1, 0xff),
(MOVZX32rr8 (EXTRACT_SUBREG GR32:$src1, sub_8bit))>,
projects / oota-llvm.git / blobdiff