def ADJCALLSTACKDOWN32 : I<0, Pseudo, (outs), (ins i32imm:$amt),
"#ADJCALLSTACKDOWN",
[(X86callseq_start timm:$amt)]>,
- Requires<[In32BitMode]>;
+ Requires<[NotLP64]>;
def ADJCALLSTACKUP32 : I<0, Pseudo, (outs), (ins i32imm:$amt1, i32imm:$amt2),
"#ADJCALLSTACKUP",
[(X86callseq_end timm:$amt1, timm:$amt2)]>,
- Requires<[In32BitMode]>;
+ Requires<[NotLP64]>;
}
// ADJCALLSTACKDOWN/UP implicitly use/def RSP because they may be expanded into
def ADJCALLSTACKDOWN64 : I<0, Pseudo, (outs), (ins i32imm:$amt),
"#ADJCALLSTACKDOWN",
[(X86callseq_start timm:$amt)]>,
- Requires<[In64BitMode]>;
+ Requires<[IsLP64]>;
def ADJCALLSTACKUP64 : I<0, Pseudo, (outs), (ins i32imm:$amt1, i32imm:$amt2),
"#ADJCALLSTACKUP",
[(X86callseq_end timm:$amt1, timm:$amt2)]>,
- Requires<[In64BitMode]>;
+ Requires<[IsLP64]>;
}
// x86-64 va_start lowering magic.
-let usesCustomInserter = 1 in {
+let usesCustomInserter = 1, Defs = [EFLAGS] in {
def VASTART_SAVE_XMM_REGS : I<0, Pseudo,
(outs),
(ins GR8:$al,
"#VASTART_SAVE_XMM_REGS $al, $regsavefi, $offset",
[(X86vastart_save_xmm_regs GR8:$al,
imm:$regsavefi,
- imm:$offset)]>;
+ imm:$offset),
+ (implicit EFLAGS)]>;
// The VAARG_64 pseudo-instruction takes the address of the va_list,
// and places the address of the next argument into a register.
// When using segmented stacks these are lowered into instructions which first
// check if the current stacklet has enough free memory. If it does, memory is
-// allocated by bumping the stack pointer. Otherwise memory is allocated from
+// allocated by bumping the stack pointer. Otherwise memory is allocated from
// the heap.
let Defs = [EAX, ESP, EFLAGS], Uses = [ESP] in
"# variable sized alloca for segmented stacks",
[(set GR32:$dst,
(X86SegAlloca GR32:$size))]>,
- Requires<[In32BitMode]>;
+ Requires<[NotLP64]>;
let Defs = [RAX, RSP, EFLAGS], Uses = [RSP] in
def SEG_ALLOCA_64 : I<0, Pseudo, (outs GR64:$dst), (ins GR64:$size),
// The MSVC runtime contains an _ftol2 routine for converting floating-point
// to integer values. It has a strange calling convention: the input is
-// popped from the x87 stack, and the return value is given in EDX:EAX. No
-// other registers (aside from flags) are touched.
+// popped from the x87 stack, and the return value is given in EDX:EAX. ECX is
+// used as a temporary register. No other registers (aside from flags) are
+// touched.
// Microsoft toolchains do not support 80-bit precision, so a WIN_FTOL_80
// variant is unnecessary.
-let Defs = [EAX, EDX, EFLAGS], FPForm = SpecialFP in {
+let Defs = [EAX, EDX, ECX, EFLAGS], FPForm = SpecialFP in {
def WIN_FTOL_32 : I<0, Pseudo, (outs), (ins RFP32:$src),
"# win32 fptoui",
[(X86WinFTOL RFP32:$src)]>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def WIN_FTOL_64 : I<0, Pseudo, (outs), (ins RFP64:$src),
"# win32 fptoui",
[(X86WinFTOL RFP64:$src)]>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
}
//===----------------------------------------------------------------------===//
def EH_SjLj_SetJmp32 : I<0, Pseudo, (outs GR32:$dst), (ins i32mem:$buf),
"#EH_SJLJ_SETJMP32",
[(set GR32:$dst, (X86eh_sjlj_setjmp addr:$buf))]>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def EH_SjLj_SetJmp64 : I<0, Pseudo, (outs GR32:$dst), (ins i64mem:$buf),
"#EH_SJLJ_SETJMP64",
[(set GR32:$dst, (X86eh_sjlj_setjmp addr:$buf))]>,
def EH_SjLj_LongJmp32 : I<0, Pseudo, (outs), (ins i32mem:$buf),
"#EH_SJLJ_LONGJMP32",
[(X86eh_sjlj_longjmp addr:$buf)]>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def EH_SjLj_LongJmp64 : I<0, Pseudo, (outs), (ins i64mem:$buf),
"#EH_SJLJ_LONGJMP64",
[(X86eh_sjlj_longjmp addr:$buf)]>,
"#EH_SjLj_Setup\t$dst", []>;
}
+//===----------------------------------------------------------------------===//
+// Pseudo instructions used by unwind info.
+//
+let isPseudo = 1 in {
+ def SEH_PushReg : I<0, Pseudo, (outs), (ins i32imm:$reg),
+ "#SEH_PushReg $reg", []>;
+ def SEH_SaveReg : I<0, Pseudo, (outs), (ins i32imm:$reg, i32imm:$dst),
+ "#SEH_SaveReg $reg, $dst", []>;
+ def SEH_SaveXMM : I<0, Pseudo, (outs), (ins i32imm:$reg, i32imm:$dst),
+ "#SEH_SaveXMM $reg, $dst", []>;
+ def SEH_StackAlloc : I<0, Pseudo, (outs), (ins i32imm:$size),
+ "#SEH_StackAlloc $size", []>;
+ def SEH_SetFrame : I<0, Pseudo, (outs), (ins i32imm:$reg, i32imm:$offset),
+ "#SEH_SetFrame $reg, $offset", []>;
+ def SEH_PushFrame : I<0, Pseudo, (outs), (ins i1imm:$mode),
+ "#SEH_PushFrame $mode", []>;
+ def SEH_EndPrologue : I<0, Pseudo, (outs), (ins),
+ "#SEH_EndPrologue", []>;
+ def SEH_Epilogue : I<0, Pseudo, (outs), (ins),
+ "#SEH_Epilogue", []>;
+}
+
//===----------------------------------------------------------------------===//
// Pseudo instructions used by segmented stacks.
//
// Alias instruction mapping movr0 to xor.
// FIXME: remove when we can teach regalloc that xor reg, reg is ok.
-// FIXME: Set encoding to pseudo.
let Defs = [EFLAGS], isReMaterializable = 1, isAsCheapAsAMove = 1,
- isCodeGenOnly = 1 in
-def MOV32r0 : I<0x31, MRMInitReg, (outs GR32:$dst), (ins), "",
+ isPseudo = 1 in
+def MOV32r0 : I<0, Pseudo, (outs GR32:$dst), (ins), "",
[(set GR32:$dst, 0)], IIC_ALU_NONMEM>, Sched<[WriteZero]>;
// Other widths can also make use of the 32-bit xor, which may have a smaller
// 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>, Sched<[WriteALU]>;
+ 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, SchedRW = [WriteALU] 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,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def REP_MOVSW_32 : I<0xA5, RawFrm, (outs), (ins), "{rep;movsw|rep movsw}",
- [(X86rep_movs i16)], IIC_REP_MOVS>, REP, OpSize,
- Requires<[In32BitMode]>;
+ [(X86rep_movs i16)], IIC_REP_MOVS>, REP, OpSize16,
+ Requires<[Not64BitMode]>;
def REP_MOVSD_32 : I<0xA5, RawFrm, (outs), (ins), "{rep;movsl|rep movsd}",
- [(X86rep_movs i32)], IIC_REP_MOVS>, REP,
- Requires<[In32BitMode]>;
+ [(X86rep_movs i32)], IIC_REP_MOVS>, REP, OpSize32,
+ Requires<[Not64BitMode]>;
}
let Defs = [RCX,RDI,RSI], Uses = [RCX,RDI,RSI], isCodeGenOnly = 1 in {
[(X86rep_movs i8)], IIC_REP_MOVS>, REP,
Requires<[In64BitMode]>;
def REP_MOVSW_64 : I<0xA5, RawFrm, (outs), (ins), "{rep;movsw|rep movsw}",
- [(X86rep_movs i16)], IIC_REP_MOVS>, REP, OpSize,
+ [(X86rep_movs i16)], IIC_REP_MOVS>, REP, OpSize16,
Requires<[In64BitMode]>;
def REP_MOVSD_64 : I<0xA5, RawFrm, (outs), (ins), "{rep;movsl|rep movsd}",
- [(X86rep_movs i32)], IIC_REP_MOVS>, REP,
+ [(X86rep_movs i32)], IIC_REP_MOVS>, REP, OpSize32,
Requires<[In64BitMode]>;
def REP_MOVSQ_64 : RI<0xA5, RawFrm, (outs), (ins), "{rep;movsq|rep movsq}",
[(X86rep_movs i64)], IIC_REP_MOVS>, REP,
let Uses = [AL,ECX,EDI] in
def REP_STOSB_32 : I<0xAA, RawFrm, (outs), (ins), "{rep;stosb|rep stosb}",
[(X86rep_stos i8)], IIC_REP_STOS>, REP,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
let Uses = [AX,ECX,EDI] in
def REP_STOSW_32 : I<0xAB, RawFrm, (outs), (ins), "{rep;stosw|rep stosw}",
- [(X86rep_stos i16)], IIC_REP_STOS>, REP, OpSize,
- Requires<[In32BitMode]>;
+ [(X86rep_stos i16)], IIC_REP_STOS>, REP, OpSize16,
+ Requires<[Not64BitMode]>;
let Uses = [EAX,ECX,EDI] in
def REP_STOSD_32 : I<0xAB, RawFrm, (outs), (ins), "{rep;stosl|rep stosd}",
- [(X86rep_stos i32)], IIC_REP_STOS>, REP,
- Requires<[In32BitMode]>;
+ [(X86rep_stos i32)], IIC_REP_STOS>, REP, OpSize32,
+ Requires<[Not64BitMode]>;
}
let Defs = [RCX,RDI], isCodeGenOnly = 1 in {
Requires<[In64BitMode]>;
let Uses = [AX,RCX,RDI] in
def REP_STOSW_64 : I<0xAB, RawFrm, (outs), (ins), "{rep;stosw|rep stosw}",
- [(X86rep_stos i16)], IIC_REP_STOS>, REP, OpSize,
+ [(X86rep_stos i16)], IIC_REP_STOS>, REP, OpSize16,
Requires<[In64BitMode]>;
let Uses = [RAX,RCX,RDI] in
def REP_STOSD_64 : I<0xAB, RawFrm, (outs), (ins), "{rep;stosl|rep stosd}",
- [(X86rep_stos i32)], IIC_REP_STOS>, REP,
+ [(X86rep_stos i32)], IIC_REP_STOS>, REP, OpSize32,
Requires<[In64BitMode]>;
-
+
let Uses = [RAX,RCX,RDI] in
def REP_STOSQ_64 : RI<0xAB, RawFrm, (outs), (ins), "{rep;stosq|rep stosq}",
[(X86rep_stos i64)], IIC_REP_STOS>, REP,
// All calls clobber the non-callee saved registers. ESP is marked as
// a use to prevent stack-pointer assignments that appear immediately
// before calls from potentially appearing dead.
-let Defs = [EAX, ECX, EDX, FP0, FP1, FP2, FP3, FP4, FP5, FP6, ST0,
+let Defs = [EAX, ECX, EDX, FP0, FP1, FP2, FP3, FP4, FP5, FP6, FP7,
+ ST0, ST1, ST2, ST3, ST4, ST5, ST6, ST7,
MM0, MM1, MM2, MM3, MM4, MM5, MM6, MM7,
XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7,
XMM8, XMM9, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15, EFLAGS],
def TLS_addr32 : I<0, Pseudo, (outs), (ins i32mem:$sym),
"# TLS_addr32",
[(X86tlsaddr tls32addr:$sym)]>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def TLS_base_addr32 : I<0, Pseudo, (outs), (ins i32mem:$sym),
"# TLS_base_addr32",
[(X86tlsbaseaddr tls32baseaddr:$sym)]>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
}
// All calls clobber the non-callee saved registers. RSP is marked as
// a use to prevent stack-pointer assignments that appear immediately
// before calls from potentially appearing dead.
let Defs = [RAX, RCX, RDX, RSI, RDI, R8, R9, R10, R11,
- FP0, FP1, FP2, FP3, FP4, FP5, FP6, ST0, ST1,
+ FP0, FP1, FP2, FP3, FP4, FP5, FP6, FP7,
+ ST0, ST1, ST2, ST3, ST4, ST5, ST6, ST7,
MM0, MM1, MM2, MM3, MM4, MM5, MM6, MM7,
XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7,
XMM8, XMM9, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15, EFLAGS],
def TLSCall_32 : I<0, Pseudo, (outs), (ins i32mem:$sym),
"# TLSCall_32",
[(X86TLSCall addr:$sym)]>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
// For x86_64, the address of the thunk is passed in %rdi, on return
// the address of the variable is in %rax. All other registers are preserved.
} // UsesCustomInserter = 1, Uses = [EFLAGS]
-//===----------------------------------------------------------------------===//
-// Atomic Instruction Pseudo Instructions
-//===----------------------------------------------------------------------===//
-
-// Pseudo atomic 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!"), []>;
- }
-}
-
-multiclass PSEUDO_ATOMIC_LOAD_BINOP_PATS<string name, string frag> {
- def : Pat<(!cast<PatFrag>(frag # "_8") addr:$ptr, GR8:$val),
- (!cast<Instruction>(name # "8") addr:$ptr, GR8:$val)>;
- def : Pat<(!cast<PatFrag>(frag # "_16") addr:$ptr, GR16:$val),
- (!cast<Instruction>(name # "16") addr:$ptr, GR16:$val)>;
- def : Pat<(!cast<PatFrag>(frag # "_32") addr:$ptr, GR32:$val),
- (!cast<Instruction>(name # "32") addr:$ptr, GR32:$val)>;
- def : Pat<(!cast<PatFrag>(frag # "_64") addr:$ptr, GR64:$val),
- (!cast<Instruction>(name # "64") addr:$ptr, GR64:$val)>;
-}
-
-// Atomic exchange, and, or, xor
-defm ATOMAND : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMAND">;
-defm ATOMOR : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMOR">;
-defm ATOMXOR : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMXOR">;
-defm ATOMNAND : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMNAND">;
-defm ATOMMAX : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMMAX">;
-defm ATOMMIN : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMMIN">;
-defm ATOMUMAX : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMUMAX">;
-defm ATOMUMIN : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMUMIN">;
-
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMAND", "atomic_load_and">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMOR", "atomic_load_or">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMXOR", "atomic_load_xor">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMNAND", "atomic_load_nand">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMMAX", "atomic_load_max">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMMIN", "atomic_load_min">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMUMAX", "atomic_load_umax">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMUMIN", "atomic_load_umin">;
-
-multiclass PSEUDO_ATOMIC_LOAD_BINOP6432<string mnemonic> {
- 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:$val2),
- !strconcat(mnemonic, "6432 PSEUDO!"), []>;
-}
-
-defm ATOMAND : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMAND">;
-defm ATOMOR : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMOR">;
-defm ATOMXOR : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMXOR">;
-defm ATOMNAND : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMNAND">;
-defm ATOMADD : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMADD">;
-defm ATOMSUB : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMSUB">;
-defm ATOMMAX : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMMAX">;
-defm ATOMMIN : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMMIN">;
-defm ATOMUMAX : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMUMAX">;
-defm ATOMUMIN : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMUMIN">;
-defm ATOMSWAP : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMSWAP">;
-
//===----------------------------------------------------------------------===//
// Normal-Instructions-With-Lock-Prefix Pseudo Instructions
//===----------------------------------------------------------------------===//
let isCodeGenOnly = 1, Defs = [EFLAGS] in
def OR32mrLocked : I<0x09, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$zero),
"or{l}\t{$zero, $dst|$dst, $zero}",
- [], IIC_ALU_MEM>, Requires<[In32BitMode]>, LOCK,
+ [], IIC_ALU_MEM>, Requires<[Not64BitMode]>, LOCK,
Sched<[WriteALULd, WriteRMW]>;
let hasSideEffects = 1 in
MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src2),
!strconcat(mnemonic, "{w}\t",
"{$src2, $dst|$dst, $src2}"),
- [], IIC_ALU_NONMEM>, OpSize, LOCK;
+ [], IIC_ALU_NONMEM>, OpSize16, LOCK;
def NAME#32mr : I<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4},
RegOpc{3}, RegOpc{2}, RegOpc{1}, 1 },
MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src2),
!strconcat(mnemonic, "{l}\t",
"{$src2, $dst|$dst, $src2}"),
- [], IIC_ALU_NONMEM>, LOCK;
+ [], IIC_ALU_NONMEM>, OpSize32, LOCK;
def NAME#64mr : RI<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4},
RegOpc{3}, RegOpc{2}, RegOpc{1}, 1 },
MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2),
ImmMod, (outs), (ins i16mem :$dst, i16imm :$src2),
!strconcat(mnemonic, "{w}\t",
"{$src2, $dst|$dst, $src2}"),
- [], IIC_ALU_MEM>, OpSize, LOCK;
+ [], IIC_ALU_MEM>, OpSize16, LOCK;
def NAME#32mi : Ii32<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4},
ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 1 },
ImmMod, (outs), (ins i32mem :$dst, i32imm :$src2),
!strconcat(mnemonic, "{l}\t",
"{$src2, $dst|$dst, $src2}"),
- [], IIC_ALU_MEM>, LOCK;
+ [], IIC_ALU_MEM>, OpSize32, LOCK;
def NAME#64mi32 : RIi32<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4},
ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 1 },
ImmMod, (outs), (ins i16mem :$dst, i16i8imm :$src2),
!strconcat(mnemonic, "{w}\t",
"{$src2, $dst|$dst, $src2}"),
- [], IIC_ALU_MEM>, OpSize, LOCK;
+ [], IIC_ALU_MEM>, OpSize16, LOCK;
def NAME#32mi8 : Ii8<{ImmOpc8{7}, ImmOpc8{6}, ImmOpc8{5}, ImmOpc8{4},
ImmOpc8{3}, ImmOpc8{2}, ImmOpc8{1}, 1 },
ImmMod, (outs), (ins i32mem :$dst, i32i8imm :$src2),
!strconcat(mnemonic, "{l}\t",
"{$src2, $dst|$dst, $src2}"),
- [], IIC_ALU_MEM>, LOCK;
+ [], IIC_ALU_MEM>, OpSize32, LOCK;
def NAME#64mi8 : RIi8<{ImmOpc8{7}, ImmOpc8{6}, ImmOpc8{5}, ImmOpc8{4},
ImmOpc8{3}, ImmOpc8{2}, ImmOpc8{1}, 1 },
ImmMod, (outs), (ins i64mem :$dst, i64i8imm :$src2),
[], IIC_UNARY_MEM>, LOCK;
def NAME#16m : I<Opc, Form, (outs), (ins i16mem:$dst),
!strconcat(mnemonic, "{w}\t$dst"),
- [], IIC_UNARY_MEM>, OpSize, LOCK;
+ [], IIC_UNARY_MEM>, OpSize16, LOCK;
def NAME#32m : I<Opc, Form, (outs), (ins i32mem:$dst),
!strconcat(mnemonic, "{l}\t$dst"),
- [], IIC_UNARY_MEM>, LOCK;
+ [], IIC_UNARY_MEM>, OpSize32, LOCK;
def NAME#64m : RI<Opc, Form, (outs), (ins i64mem:$dst),
!strconcat(mnemonic, "{q}\t$dst"),
[], IIC_UNARY_MEM>, LOCK;
let Defs = [AX, EFLAGS], Uses = [AX] in
def NAME#16 : I<Opc, Form, (outs), (ins i16mem:$ptr, GR16:$swap),
!strconcat(mnemonic, "{w}\t{$swap, $ptr|$ptr, $swap}"),
- [(frag addr:$ptr, GR16:$swap, 2)], itin>, TB, OpSize, LOCK;
+ [(frag addr:$ptr, GR16:$swap, 2)], itin>, TB, OpSize16, LOCK;
let Defs = [EAX, EFLAGS], Uses = [EAX] in
def NAME#32 : I<Opc, Form, (outs), (ins i32mem:$ptr, GR32:$swap),
!strconcat(mnemonic, "{l}\t{$swap, $ptr|$ptr, $swap}"),
- [(frag addr:$ptr, GR32:$swap, 4)], itin>, TB, LOCK;
+ [(frag addr:$ptr, GR32:$swap, 4)], itin>, TB, OpSize32, LOCK;
let Defs = [RAX, EFLAGS], Uses = [RAX] in
def NAME#64 : RI<Opc, Form, (outs), (ins i64mem:$ptr, GR64:$swap),
!strconcat(mnemonic, "{q}\t{$swap, $ptr|$ptr, $swap}"),
[(set
GR16:$dst,
(!cast<PatFrag>(frag # "_16") addr:$ptr, GR16:$val))],
- itin>, OpSize;
+ itin>, OpSize16;
def NAME#32 : I<opc, MRMSrcMem, (outs GR32:$dst),
(ins GR32:$val, i32mem:$ptr),
!strconcat(mnemonic, "{l}\t{$val, $ptr|$ptr, $val}"),
[(set
GR32:$dst,
(!cast<PatFrag>(frag # "_32") addr:$ptr, GR32:$val))],
- itin>;
+ itin>, OpSize32;
def NAME#64 : RI<opc, MRMSrcMem, (outs GR64:$dst),
(ins GR64:$val, i64mem:$ptr),
!strconcat(mnemonic, "{q}\t{$val, $ptr|$ptr, $val}"),
IIC_XADD_LOCK_MEM8, IIC_XADD_LOCK_MEM>,
TB, LOCK;
-def ACQUIRE_MOV8rm : I<0, Pseudo, (outs GR8 :$dst), (ins i8mem :$src),
- "#ACQUIRE_MOV PSEUDO!",
- [(set GR8:$dst, (atomic_load_8 addr:$src))]>;
-def ACQUIRE_MOV16rm : I<0, Pseudo, (outs GR16:$dst), (ins i16mem:$src),
- "#ACQUIRE_MOV PSEUDO!",
- [(set GR16:$dst, (atomic_load_16 addr:$src))]>;
-def ACQUIRE_MOV32rm : I<0, Pseudo, (outs GR32:$dst), (ins i32mem:$src),
- "#ACQUIRE_MOV PSEUDO!",
- [(set GR32:$dst, (atomic_load_32 addr:$src))]>;
-def ACQUIRE_MOV64rm : I<0, Pseudo, (outs GR64:$dst), (ins i64mem:$src),
- "#ACQUIRE_MOV PSEUDO!",
- [(set GR64:$dst, (atomic_load_64 addr:$src))]>;
+/* The following multiclass tries to make sure that in code like
+ * x.store (immediate op x.load(acquire), release)
+ * an operation directly on memory is generated instead of wasting a register.
+ * It is not automatic as atomic_store/load are only lowered to MOV instructions
+ * extremely late to prevent them from being accidentally reordered in the backend
+ * (see below the RELEASE_MOV* / ACQUIRE_MOV* pseudo-instructions)
+ */
+multiclass RELEASE_BINOP_MI<string op> {
+ def NAME#8mi : I<0, Pseudo, (outs), (ins i8mem:$dst, i8imm:$src),
+ "#RELEASE_BINOP PSEUDO!",
+ [(atomic_store_8 addr:$dst, (!cast<PatFrag>(op)
+ (atomic_load_8 addr:$dst), (i8 imm:$src)))]>;
+ // NAME#16 is not generated as 16-bit arithmetic instructions are considered
+ // costly and avoided as far as possible by this backend anyway
+ def NAME#32mi : I<0, Pseudo, (outs), (ins i32mem:$dst, i32imm:$src),
+ "#RELEASE_BINOP PSEUDO!",
+ [(atomic_store_32 addr:$dst, (!cast<PatFrag>(op)
+ (atomic_load_32 addr:$dst), (i32 imm:$src)))]>;
+ def NAME#64mi32 : I<0, Pseudo, (outs), (ins i64mem:$dst, i64i32imm:$src),
+ "#RELEASE_BINOP PSEUDO!",
+ [(atomic_store_64 addr:$dst, (!cast<PatFrag>(op)
+ (atomic_load_64 addr:$dst), (i64immSExt32:$src)))]>;
+}
+defm RELEASE_ADD : RELEASE_BINOP_MI<"add">;
+defm RELEASE_AND : RELEASE_BINOP_MI<"and">;
+defm RELEASE_OR : RELEASE_BINOP_MI<"or">;
+defm RELEASE_XOR : RELEASE_BINOP_MI<"xor">;
+// Note: we don't deal with sub, because substractions of constants are
+// optimized into additions before this code can run
+
+multiclass RELEASE_UNOP<dag dag8, dag dag16, dag dag32, dag dag64> {
+ def NAME#8m : I<0, Pseudo, (outs), (ins i8mem:$dst),
+ "#RELEASE_UNOP PSEUDO!",
+ [(atomic_store_8 addr:$dst, dag8)]>;
+ def NAME#16m : I<0, Pseudo, (outs), (ins i16mem:$dst),
+ "#RELEASE_UNOP PSEUDO!",
+ [(atomic_store_16 addr:$dst, dag16)]>;
+ def NAME#32m : I<0, Pseudo, (outs), (ins i32mem:$dst),
+ "#RELEASE_UNOP PSEUDO!",
+ [(atomic_store_32 addr:$dst, dag32)]>;
+ def NAME#64m : I<0, Pseudo, (outs), (ins i64mem:$dst),
+ "#RELEASE_UNOP PSEUDO!",
+ [(atomic_store_64 addr:$dst, dag64)]>;
+}
+
+defm RELEASE_INC : RELEASE_UNOP<
+ (add (atomic_load_8 addr:$dst), (i8 1)),
+ (add (atomic_load_16 addr:$dst), (i16 1)),
+ (add (atomic_load_32 addr:$dst), (i32 1)),
+ (add (atomic_load_64 addr:$dst), (i64 1))>, Requires<[NotSlowIncDec]>;
+defm RELEASE_DEC : RELEASE_UNOP<
+ (add (atomic_load_8 addr:$dst), (i8 -1)),
+ (add (atomic_load_16 addr:$dst), (i16 -1)),
+ (add (atomic_load_32 addr:$dst), (i32 -1)),
+ (add (atomic_load_64 addr:$dst), (i64 -1))>, Requires<[NotSlowIncDec]>;
+/*
+TODO: These don't work because the type inference of TableGen fails.
+TODO: find a way to fix it.
+defm RELEASE_NEG : RELEASE_UNOP<
+ (ineg (atomic_load_8 addr:$dst)),
+ (ineg (atomic_load_16 addr:$dst)),
+ (ineg (atomic_load_32 addr:$dst)),
+ (ineg (atomic_load_64 addr:$dst))>;
+defm RELEASE_NOT : RELEASE_UNOP<
+ (not (atomic_load_8 addr:$dst)),
+ (not (atomic_load_16 addr:$dst)),
+ (not (atomic_load_32 addr:$dst)),
+ (not (atomic_load_64 addr:$dst))>;
+*/
+
+def RELEASE_MOV8mi : I<0, Pseudo, (outs), (ins i8mem:$dst, i8imm:$src),
+ "#RELEASE_MOV PSEUDO !",
+ [(atomic_store_8 addr:$dst, (i8 imm:$src))]>;
+def RELEASE_MOV16mi : I<0, Pseudo, (outs), (ins i16mem:$dst, i16imm:$src),
+ "#RELEASE_MOV PSEUDO !",
+ [(atomic_store_16 addr:$dst, (i16 imm:$src))]>;
+def RELEASE_MOV32mi : I<0, Pseudo, (outs), (ins i32mem:$dst, i32imm:$src),
+ "#RELEASE_MOV PSEUDO !",
+ [(atomic_store_32 addr:$dst, (i32 imm:$src))]>;
+def RELEASE_MOV64mi32 : I<0, Pseudo, (outs), (ins i64mem:$dst, i64i32imm:$src),
+ "#RELEASE_MOV PSEUDO !",
+ [(atomic_store_64 addr:$dst, i64immSExt32:$src)]>;
def RELEASE_MOV8mr : I<0, Pseudo, (outs), (ins i8mem :$dst, GR8 :$src),
"#RELEASE_MOV PSEUDO!",
"#RELEASE_MOV PSEUDO!",
[(atomic_store_64 addr:$dst, GR64:$src)]>;
+def ACQUIRE_MOV8rm : I<0, Pseudo, (outs GR8 :$dst), (ins i8mem :$src),
+ "#ACQUIRE_MOV PSEUDO!",
+ [(set GR8:$dst, (atomic_load_8 addr:$src))]>;
+def ACQUIRE_MOV16rm : I<0, Pseudo, (outs GR16:$dst), (ins i16mem:$src),
+ "#ACQUIRE_MOV PSEUDO!",
+ [(set GR16:$dst, (atomic_load_16 addr:$src))]>;
+def ACQUIRE_MOV32rm : I<0, Pseudo, (outs GR32:$dst), (ins i32mem:$src),
+ "#ACQUIRE_MOV PSEUDO!",
+ [(set GR32:$dst, (atomic_load_32 addr:$src))]>;
+def ACQUIRE_MOV64rm : I<0, Pseudo, (outs GR64:$dst), (ins i64mem:$src),
+ "#ACQUIRE_MOV PSEUDO!",
+ [(set GR64:$dst, (atomic_load_64 addr:$src))]>;
//===----------------------------------------------------------------------===//
// Conditional Move Pseudo Instructions.
//===----------------------------------------------------------------------===//
-
// CMOV* - Used to implement the SSE SELECT DAG operation. Expanded after
// instruction selection into a branch sequence.
let Uses = [EFLAGS], usesCustomInserter = 1 in {
[(set VR256:$dst,
(v4i64 (X86cmov VR256:$t, VR256:$f, imm:$cond,
EFLAGS)))]>;
+ def CMOV_V8I64 : I<0, Pseudo,
+ (outs VR512:$dst), (ins VR512:$t, VR512:$f, i8imm:$cond),
+ "#CMOV_V8I64 PSEUDO!",
+ [(set VR512:$dst,
+ (v8i64 (X86cmov VR512:$t, VR512:$f, imm:$cond,
+ EFLAGS)))]>;
+ def CMOV_V8F64 : I<0, Pseudo,
+ (outs VR512:$dst), (ins VR512:$t, VR512:$f, i8imm:$cond),
+ "#CMOV_V8F64 PSEUDO!",
+ [(set VR512:$dst,
+ (v8f64 (X86cmov VR512:$t, VR512:$f, imm:$cond,
+ EFLAGS)))]>;
+ def CMOV_V16F32 : I<0, Pseudo,
+ (outs VR512:$dst), (ins VR512:$t, VR512:$f, i8imm:$cond),
+ "#CMOV_V16F32 PSEUDO!",
+ [(set VR512:$dst,
+ (v16f32 (X86cmov VR512:$t, VR512:$f, imm:$cond,
+ EFLAGS)))]>;
}
def : Pat<(store (i32 (X86Wrapper tblockaddress:$src)), addr:$dst),
(MOV32mi addr:$dst, tblockaddress:$src)>;
-
-
// ConstantPool GlobalAddress, ExternalSymbol, and JumpTable when not in small
// code model mode, should use 'movabs'. FIXME: This is really a hack, the
// 'movabs' predicate should handle this sort of thing.
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.
(MOV64mi32 addr:$dst, tblockaddress:$src)>,
Requires<[NearData, IsStatic]>;
-
-
// Calls
// tls has some funny stuff here...
// This corresponds to movabs $foo@tpoff, %rax
def : Pat<(i64 (X86Wrapper tglobaltlsaddr :$dst)),
- (MOV64ri tglobaltlsaddr :$dst)>;
+ (MOV64ri32 tglobaltlsaddr :$dst)>;
// This corresponds to add $foo@tpoff, %rax
def : Pat<(add GR64:$src1, (X86Wrapper tglobaltlsaddr :$dst)),
(ADD64ri32 GR64:$src1, tglobaltlsaddr :$dst)>;
def : Pat<(X86tcret ptr_rc_tailcall:$dst, imm:$off),
(TCRETURNri ptr_rc_tailcall:$dst, imm:$off)>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
// FIXME: This is disabled for 32-bit PIC mode because the global base
// register which is part of the address mode may be assigned a
// callee-saved register.
def : Pat<(X86tcret (load addr:$dst), imm:$off),
(TCRETURNmi addr:$dst, imm:$off)>,
- Requires<[In32BitMode, IsNotPIC]>;
+ Requires<[Not64BitMode, IsNotPIC]>;
def : Pat<(X86tcret (i32 tglobaladdr:$dst), imm:$off),
(TCRETURNdi texternalsym:$dst, imm:$off)>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def : Pat<(X86tcret (i32 texternalsym:$dst), imm:$off),
(TCRETURNdi texternalsym:$dst, imm:$off)>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def : Pat<(X86tcret ptr_rc_tailcall:$dst, imm:$off),
(TCRETURNri64 ptr_rc_tailcall:$dst, imm:$off)>,
return N->getOpcode() != ISD::TRUNCATE &&
N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
N->getOpcode() != ISD::CopyFromReg &&
+ N->getOpcode() != ISD::AssertSext &&
N->getOpcode() != X86ISD::CMOV;
}]>;
return CurDAG->MaskedValueIsZero(N->getOperand(0), CN->getAPIntValue());
APInt KnownZero0, KnownOne0;
- CurDAG->ComputeMaskedBits(N->getOperand(0), KnownZero0, KnownOne0, 0);
+ CurDAG->computeKnownBits(N->getOperand(0), KnownZero0, KnownOne0, 0);
APInt KnownZero1, KnownOne1;
- CurDAG->ComputeMaskedBits(N->getOperand(1), KnownZero1, KnownOne1, 0);
+ CurDAG->computeKnownBits(N->getOperand(1), KnownZero1, KnownOne1, 0);
return (~KnownZero0 & ~KnownZero1) == 0;
}]>;
(MOVZX32rr8 (EXTRACT_SUBREG (i32 (COPY_TO_REGCLASS GR32:$src1,
GR32_ABCD)),
sub_8bit))>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
// r & (2^8-1) ==> movz
def : Pat<(and GR16:$src1, 0xff),
(EXTRACT_SUBREG (MOVZX32rr8 (EXTRACT_SUBREG
(i16 (COPY_TO_REGCLASS GR16:$src1, GR16_ABCD)), sub_8bit)),
sub_16bit)>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
// r & (2^32-1) ==> movz
def : Pat<(and GR64:$src, 0x00000000FFFFFFFF),
(MOVSX32rr8 (EXTRACT_SUBREG (i32 (COPY_TO_REGCLASS GR32:$src,
GR32_ABCD)),
sub_8bit))>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def : Pat<(sext_inreg GR16:$src, i8),
(EXTRACT_SUBREG (i32 (MOVSX32rr8 (EXTRACT_SUBREG
(i32 (COPY_TO_REGCLASS GR16:$src, GR16_ABCD)), sub_8bit))),
sub_16bit)>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def : Pat<(sext_inreg GR64:$src, i32),
(MOVSX64rr32 (EXTRACT_SUBREG GR64:$src, sub_32bit))>;
def : Pat<(i8 (trunc GR32:$src)),
(EXTRACT_SUBREG (i32 (COPY_TO_REGCLASS GR32:$src, GR32_ABCD)),
sub_8bit)>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def : Pat<(i8 (trunc GR16:$src)),
(EXTRACT_SUBREG (i16 (COPY_TO_REGCLASS GR16:$src, GR16_ABCD)),
sub_8bit)>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def : Pat<(i32 (trunc GR64:$src)),
(EXTRACT_SUBREG GR64:$src, sub_32bit)>;
def : Pat<(i16 (trunc GR64:$src)),
def : Pat<(i8 (trunc (srl_su GR16:$src, (i8 8)))),
(EXTRACT_SUBREG (i16 (COPY_TO_REGCLASS GR16:$src, GR16_ABCD)),
sub_8bit_hi)>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def : Pat<(i8 (trunc (srl_su GR32:$src, (i8 8)))),
(EXTRACT_SUBREG (i32 (COPY_TO_REGCLASS GR32:$src, GR32_ABCD)),
sub_8bit_hi)>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def : Pat<(srl GR16:$src, (i8 8)),
(EXTRACT_SUBREG
(MOVZX32rr8
(EXTRACT_SUBREG (i16 (COPY_TO_REGCLASS GR16:$src, GR16_ABCD)),
sub_8bit_hi)),
sub_16bit)>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def : Pat<(i32 (zext (srl_su GR16:$src, (i8 8)))),
(MOVZX32rr8 (EXTRACT_SUBREG (i16 (COPY_TO_REGCLASS GR16:$src,
GR16_ABCD)),
sub_8bit_hi))>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def : Pat<(i32 (anyext (srl_su GR16:$src, (i8 8)))),
(MOVZX32rr8 (EXTRACT_SUBREG (i16 (COPY_TO_REGCLASS GR16:$src,
GR16_ABCD)),
sub_8bit_hi))>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def : Pat<(and (srl_su GR32:$src, (i8 8)), (i32 255)),
(MOVZX32rr8 (EXTRACT_SUBREG (i32 (COPY_TO_REGCLASS GR32:$src,
GR32_ABCD)),
sub_8bit_hi))>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
def : Pat<(srl (and_su GR32:$src, 0xff00), (i8 8)),
(MOVZX32rr8 (EXTRACT_SUBREG (i32 (COPY_TO_REGCLASS GR32:$src,
GR32_ABCD)),
sub_8bit_hi))>,
- Requires<[In32BitMode]>;
+ Requires<[Not64BitMode]>;
// h-register tricks.
// For now, be conservative on x86-64 and use an h-register extract only if the
def immShift32 : ImmLeaf<i8, [{ return CountTrailingOnes_32(Imm) >= 5; }]>;
def immShift64 : ImmLeaf<i8, [{ return CountTrailingOnes_32(Imm) >= 6; }]>;
-// (shl x (and y, 31)) ==> (shl x, y)
-def : Pat<(shl GR8:$src1, (and CL, immShift32)),
- (SHL8rCL GR8:$src1)>;
-def : Pat<(shl GR16:$src1, (and CL, immShift32)),
- (SHL16rCL GR16:$src1)>;
-def : Pat<(shl GR32:$src1, (and CL, immShift32)),
- (SHL32rCL GR32:$src1)>;
-def : Pat<(store (shl (loadi8 addr:$dst), (and CL, immShift32)), addr:$dst),
- (SHL8mCL addr:$dst)>;
-def : Pat<(store (shl (loadi16 addr:$dst), (and CL, immShift32)), addr:$dst),
- (SHL16mCL addr:$dst)>;
-def : Pat<(store (shl (loadi32 addr:$dst), (and CL, immShift32)), addr:$dst),
- (SHL32mCL addr:$dst)>;
-
-def : Pat<(srl GR8:$src1, (and CL, immShift32)),
- (SHR8rCL GR8:$src1)>;
-def : Pat<(srl GR16:$src1, (and CL, immShift32)),
- (SHR16rCL GR16:$src1)>;
-def : Pat<(srl GR32:$src1, (and CL, immShift32)),
- (SHR32rCL GR32:$src1)>;
-def : Pat<(store (srl (loadi8 addr:$dst), (and CL, immShift32)), addr:$dst),
- (SHR8mCL addr:$dst)>;
-def : Pat<(store (srl (loadi16 addr:$dst), (and CL, immShift32)), addr:$dst),
- (SHR16mCL addr:$dst)>;
-def : Pat<(store (srl (loadi32 addr:$dst), (and CL, immShift32)), addr:$dst),
- (SHR32mCL addr:$dst)>;
-
-def : Pat<(sra GR8:$src1, (and CL, immShift32)),
- (SAR8rCL GR8:$src1)>;
-def : Pat<(sra GR16:$src1, (and CL, immShift32)),
- (SAR16rCL GR16:$src1)>;
-def : Pat<(sra GR32:$src1, (and CL, immShift32)),
- (SAR32rCL GR32:$src1)>;
-def : Pat<(store (sra (loadi8 addr:$dst), (and CL, immShift32)), addr:$dst),
- (SAR8mCL addr:$dst)>;
-def : Pat<(store (sra (loadi16 addr:$dst), (and CL, immShift32)), addr:$dst),
- (SAR16mCL addr:$dst)>;
-def : Pat<(store (sra (loadi32 addr:$dst), (and CL, immShift32)), addr:$dst),
- (SAR32mCL addr:$dst)>;
-
-// (shl x (and y, 63)) ==> (shl x, y)
-def : Pat<(shl GR64:$src1, (and CL, immShift64)),
- (SHL64rCL GR64:$src1)>;
-def : Pat<(store (shl (loadi64 addr:$dst), (and CL, 63)), addr:$dst),
- (SHL64mCL addr:$dst)>;
-
-def : Pat<(srl GR64:$src1, (and CL, immShift64)),
- (SHR64rCL GR64:$src1)>;
-def : Pat<(store (srl (loadi64 addr:$dst), (and CL, 63)), addr:$dst),
- (SHR64mCL addr:$dst)>;
-
-def : Pat<(sra GR64:$src1, (and CL, immShift64)),
- (SAR64rCL GR64:$src1)>;
-def : Pat<(store (sra (loadi64 addr:$dst), (and CL, 63)), addr:$dst),
- (SAR64mCL addr:$dst)>;
+// Shift amount is implicitly masked.
+multiclass MaskedShiftAmountPats<SDNode frag, string name> {
+ // (shift x (and y, 31)) ==> (shift x, y)
+ def : Pat<(frag GR8:$src1, (and CL, immShift32)),
+ (!cast<Instruction>(name # "8rCL") GR8:$src1)>;
+ def : Pat<(frag GR16:$src1, (and CL, immShift32)),
+ (!cast<Instruction>(name # "16rCL") GR16:$src1)>;
+ def : Pat<(frag GR32:$src1, (and CL, immShift32)),
+ (!cast<Instruction>(name # "32rCL") GR32:$src1)>;
+ def : Pat<(store (frag (loadi8 addr:$dst), (and CL, immShift32)), addr:$dst),
+ (!cast<Instruction>(name # "8mCL") addr:$dst)>;
+ def : Pat<(store (frag (loadi16 addr:$dst), (and CL, immShift32)), addr:$dst),
+ (!cast<Instruction>(name # "16mCL") addr:$dst)>;
+ def : Pat<(store (frag (loadi32 addr:$dst), (and CL, immShift32)), addr:$dst),
+ (!cast<Instruction>(name # "32mCL") addr:$dst)>;
+
+ // (shift x (and y, 63)) ==> (shift x, y)
+ def : Pat<(frag GR64:$src1, (and CL, immShift64)),
+ (!cast<Instruction>(name # "64rCL") GR64:$src1)>;
+ def : Pat<(store (frag (loadi64 addr:$dst), (and CL, 63)), addr:$dst),
+ (!cast<Instruction>(name # "64mCL") addr:$dst)>;
+}
+defm : MaskedShiftAmountPats<shl, "SHL">;
+defm : MaskedShiftAmountPats<srl, "SHR">;
+defm : MaskedShiftAmountPats<sra, "SAR">;
+defm : MaskedShiftAmountPats<rotl, "ROL">;
+defm : MaskedShiftAmountPats<rotr, "ROR">;
// (anyext (setcc_carry)) -> (setcc_carry)
def : Pat<(i16 (anyext (i8 (X86setcc_c X86_COND_B, EFLAGS)))),
(IMUL64rmi32 addr:$src1, i64immSExt32:$src2)>;
// Increment reg.
-def : Pat<(add GR8 :$src, 1), (INC8r GR8 :$src)>;
-def : Pat<(add GR16:$src, 1), (INC16r GR16:$src)>, Requires<[In32BitMode]>;
-def : Pat<(add GR16:$src, 1), (INC64_16r GR16:$src)>, Requires<[In64BitMode]>;
-def : Pat<(add GR32:$src, 1), (INC32r GR32:$src)>, Requires<[In32BitMode]>;
-def : Pat<(add GR32:$src, 1), (INC64_32r GR32:$src)>, Requires<[In64BitMode]>;
-def : Pat<(add GR64:$src, 1), (INC64r GR64:$src)>;
+// Do not make INC if it is slow
+def : Pat<(add GR8:$src, 1),
+ (INC8r GR8:$src)>, Requires<[NotSlowIncDec]>;
+def : Pat<(add GR16:$src, 1),
+ (INC16r GR16:$src)>, Requires<[NotSlowIncDec, Not64BitMode]>;
+def : Pat<(add GR16:$src, 1),
+ (INC64_16r GR16:$src)>, Requires<[NotSlowIncDec, In64BitMode]>;
+def : Pat<(add GR32:$src, 1),
+ (INC32r GR32:$src)>, Requires<[NotSlowIncDec, Not64BitMode]>;
+def : Pat<(add GR32:$src, 1),
+ (INC64_32r GR32:$src)>, Requires<[NotSlowIncDec, In64BitMode]>;
+def : Pat<(add GR64:$src, 1),
+ (INC64r GR64:$src)>, Requires<[NotSlowIncDec]>;
// Decrement reg.
-def : Pat<(add GR8 :$src, -1), (DEC8r GR8 :$src)>;
-def : Pat<(add GR16:$src, -1), (DEC16r GR16:$src)>, Requires<[In32BitMode]>;
-def : Pat<(add GR16:$src, -1), (DEC64_16r GR16:$src)>, Requires<[In64BitMode]>;
-def : Pat<(add GR32:$src, -1), (DEC32r GR32:$src)>, Requires<[In32BitMode]>;
-def : Pat<(add GR32:$src, -1), (DEC64_32r GR32:$src)>, Requires<[In64BitMode]>;
-def : Pat<(add GR64:$src, -1), (DEC64r GR64:$src)>;
+// Do not make DEC if it is slow
+def : Pat<(add GR8:$src, -1),
+ (DEC8r GR8:$src)>, Requires<[NotSlowIncDec]>;
+def : Pat<(add GR16:$src, -1),
+ (DEC16r GR16:$src)>, Requires<[NotSlowIncDec, Not64BitMode]>;
+def : Pat<(add GR16:$src, -1),
+ (DEC64_16r GR16:$src)>, Requires<[NotSlowIncDec, In64BitMode]>;
+def : Pat<(add GR32:$src, -1),
+ (DEC32r GR32:$src)>, Requires<[NotSlowIncDec, Not64BitMode]>;
+def : Pat<(add GR32:$src, -1),
+ (DEC64_32r GR32:$src)>, Requires<[NotSlowIncDec, In64BitMode]>;
+def : Pat<(add GR64:$src, -1),
+ (DEC64r GR64:$src)>, Requires<[NotSlowIncDec]>;
// or reg/reg.
def : Pat<(or GR8 :$src1, GR8 :$src2), (OR8rr GR8 :$src1, GR8 :$src2)>;
def : Pat<(cttz_zero_undef (loadi16 addr:$src)), (BSF16rm addr:$src)>;
def : Pat<(cttz_zero_undef (loadi32 addr:$src)), (BSF32rm addr:$src)>;
def : Pat<(cttz_zero_undef (loadi64 addr:$src)), (BSF64rm addr:$src)>;
+
+// When HasMOVBE is enabled it is possible to get a non-legalized
+// register-register 16 bit bswap. This maps it to a ROL instruction.
+let Predicates = [HasMOVBE] in {
+ def : Pat<(bswap GR16:$src), (ROL16ri GR16:$src, (i8 8))>;
+}