// 64-bits but only 32 bits are significant.
def i64i32imm : Operand<i64>;
+
+// 64-bits but only 32 bits are significant, and those bits are treated as being
+// pc relative.
+def i64i32imm_pcrel : Operand<i64> {
+ let PrintMethod = "print_pcrel_imm";
+}
+
+
// 64-bits but only 8 bits are significant.
def i64i8imm : Operand<i64>;
def lea64_32mem : Operand<i32> {
let PrintMethod = "printlea64_32mem";
+ let AsmOperandLowerMethod = "lower_lea64_32mem";
let MIOperandInfo = (ops GR32, i8imm, GR32, i32imm);
}
// Complex Pattern Definitions.
//
def lea64addr : ComplexPattern<i64, 4, "SelectLEAAddr",
- [add, mul, X86mul_imm, shl, or, frameindex, X86Wrapper],
+ [add, mul, X86mul_imm, shl, or, frameindex, X86Wrapper,
+ X86WrapperRIP],
[]>;
+def tls64addr : ComplexPattern<i64, 4, "SelectTLSADDRAddr",
+ [tglobaltlsaddr], []>;
+
//===----------------------------------------------------------------------===//
// Pattern fragments.
//
// NOTE: this pattern doesn't match "X86call imm", because we do not know
// that the offset between an arbitrary immediate and the call will fit in
// the 32-bit pcrel field that we have.
- def CALL64pcrel32 : I<0xE8, RawFrm,
- (outs), (ins i64i32imm:$dst, variable_ops),
- "call\t${dst:call}", []>,
+ def CALL64pcrel32 : Ii32<0xE8, RawFrm,
+ (outs), (ins i64i32imm_pcrel:$dst, variable_ops),
+ "call\t$dst", []>,
Requires<[In64BitMode]>;
def CALL64r : I<0xFF, MRM2r, (outs), (ins GR64:$dst, variable_ops),
"call\t{*}$dst", [(X86call GR64:$dst)]>;
(outs), (ins GR64:$reg), "push{q}\t$reg", []>;
}
+let Defs = [RSP], Uses = [RSP], neverHasSideEffects = 1, mayStore = 1 in {
+def PUSH64i8 : Ii8<0x6a, RawFrm, (outs), (ins i8imm:$imm),
+ "push{q}\t$imm", []>;
+def PUSH64i16 : Ii16<0x68, RawFrm, (outs), (ins i16imm:$imm),
+ "push{q}\t$imm", []>;
+def PUSH64i32 : Ii32<0x68, RawFrm, (outs), (ins i32imm:$imm),
+ "push{q}\t$imm", []>;
+}
+
let Defs = [RSP, EFLAGS], Uses = [RSP], mayLoad = 1 in
def POPFQ : I<0x9D, RawFrm, (outs), (ins), "popf", []>, REX_W;
let Defs = [RSP], Uses = [RSP, EFLAGS], mayStore = 1 in
// Thread Local Storage Instructions
//===----------------------------------------------------------------------===//
-let hasSideEffects = 1, Uses = [RDI] in
-def TLS_addr64 : I<0, Pseudo, (outs), (ins i64imm:$sym),
- ".byte\t0x66; leaq\t${sym:mem}(%rip), %rdi; .word\t0x6666; rex64",
- [(X86tlsaddr tglobaltlsaddr:$sym)]>,
+// 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,
+ 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],
+ Uses = [RSP] in
+def TLS_addr64 : I<0, Pseudo, (outs), (ins lea64mem:$sym),
+ ".byte\t0x66; "
+ "leaq\t$sym(%rip), %rdi; "
+ ".word\t0x6666; "
+ "rex64; "
+ "call\t__tls_get_addr@PLT",
+ [(X86tlsaddr tls64addr:$sym)]>,
Requires<[In64BitMode]>;
let AddedComplexity = 5 in
"movq\t%gs:$src, $dst",
[(set GR64:$dst, (gsload addr:$src))]>, SegGS;
+let AddedComplexity = 5 in
+def MOV64FSrm : RI<0x8B, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
+ "movq\t%fs:$src, $dst",
+ [(set GR64:$dst, (fsload addr:$src))]>, SegFS;
+
//===----------------------------------------------------------------------===//
// Atomic Instructions
//===----------------------------------------------------------------------===//
let Defs = [RAX, EFLAGS], Uses = [RAX] in {
def LCMPXCHG64 : RI<0xB1, MRMDestMem, (outs), (ins i64mem:$ptr, GR64:$swap),
- "lock\n\tcmpxchgq\t$swap,$ptr",
+ "lock\n\t"
+ "cmpxchgq\t$swap,$ptr",
[(X86cas addr:$ptr, GR64:$swap, 8)]>, TB, LOCK;
}
let Constraints = "$val = $dst" in {
let Defs = [EFLAGS] in
def LXADD64 : RI<0xC1, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$ptr,GR64:$val),
- "lock\n\txadd\t$val, $ptr",
+ "lock\n\t"
+ "xadd\t$val, $ptr",
[(set GR64:$dst, (atomic_load_add_64 addr:$ptr, GR64:$val))]>,
TB, LOCK;
def XCHG64rm : RI<0x87, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$ptr,GR64:$val),
def : Pat<(i64 (X86Wrapper texternalsym:$dst)),
(MOV64ri texternalsym:$dst)>, Requires<[NotSmallCode]>;
+// If we have small model and -static mode, it is safe to store global addresses
+// directly as immediates. FIXME: This is really a hack, the 'imm' predicate
+// should handle this sort of thing.
def : Pat<(store (i64 (X86Wrapper tconstpool:$src)), addr:$dst),
(MOV64mi32 addr:$dst, tconstpool:$src)>,
Requires<[SmallCode, IsStatic]>;
(MOV64mi32 addr:$dst, texternalsym:$src)>,
Requires<[SmallCode, IsStatic]>;
+// If we have small model and -static mode, it is safe to store global addresses
+// directly as immediates. FIXME: This is really a hack, the 'imm' predicate
+// should handle this sort of thing.
+def : Pat<(store (i64 (X86WrapperRIP tconstpool:$src)), addr:$dst),
+ (MOV64mi32 addr:$dst, tconstpool:$src)>,
+ Requires<[SmallCode, IsStatic]>;
+def : Pat<(store (i64 (X86WrapperRIP tjumptable:$src)), addr:$dst),
+ (MOV64mi32 addr:$dst, tjumptable:$src)>,
+ Requires<[SmallCode, IsStatic]>;
+def : Pat<(store (i64 (X86WrapperRIP tglobaladdr:$src)), addr:$dst),
+ (MOV64mi32 addr:$dst, tglobaladdr:$src)>,
+ Requires<[SmallCode, IsStatic]>;
+def : Pat<(store (i64 (X86WrapperRIP texternalsym:$src)), addr:$dst),
+ (MOV64mi32 addr:$dst, texternalsym:$src)>,
+ Requires<[SmallCode, IsStatic]>;
+
+
// Calls
// Direct PC relative function call for small code model. 32-bit displacement
// sign extended to 64-bit.
Requires<[In64BitMode]>;
// h-register tricks.
-// For now, be conservative and only the extract if the value is immediately
-// zero-extended or stored, which are somewhat common cases. This uses a bunch
-// of code to prevent a register requiring a REX prefix from being allocated in
-// the same instruction as the h register, as there's currently no way to
-// describe this requirement to the register allocator.
+// For now, be conservative on x86-64 and use an h-register extract only if the
+// value is immediately zero-extended or stored, which are somewhat common
+// cases. This uses a bunch of code to prevent a register requiring a REX prefix
+// from being allocated in the same instruction as the h register, as there's
+// currently no way to describe this requirement to the register allocator.
// h-register extract and zero-extend.
def : Pat<(and (srl_su GR64:$src, (i8 8)), (i64 255)),
(SUBREG_TO_REG
(i64 0),
(MOVZX32_NOREXrr8
- (EXTRACT_SUBREG (COPY_TO_REGCLASS GR64:$src, GR64_),
+ (EXTRACT_SUBREG (COPY_TO_REGCLASS GR64:$src, GR64_ABCD),
x86_subreg_8bit_hi)),
x86_subreg_32bit)>;
def : Pat<(and (srl_su GR32:$src, (i8 8)), (i32 255)),
(MOVZX32_NOREXrr8
- (EXTRACT_SUBREG (COPY_TO_REGCLASS GR32:$src, GR32_),
+ (EXTRACT_SUBREG (COPY_TO_REGCLASS GR32:$src, GR32_ABCD),
x86_subreg_8bit_hi))>,
Requires<[In64BitMode]>;
def : Pat<(srl_su GR16:$src, (i8 8)),
(EXTRACT_SUBREG
(MOVZX32_NOREXrr8
- (EXTRACT_SUBREG (COPY_TO_REGCLASS GR16:$src, GR16_),
+ (EXTRACT_SUBREG (COPY_TO_REGCLASS GR16:$src, GR16_ABCD),
x86_subreg_8bit_hi)),
x86_subreg_16bit)>,
Requires<[In64BitMode]>;
+def : Pat<(i32 (zext (srl_su GR16:$src, (i8 8)))),
+ (MOVZX32_NOREXrr8
+ (EXTRACT_SUBREG (COPY_TO_REGCLASS GR16:$src, GR16_ABCD),
+ x86_subreg_8bit_hi))>,
+ Requires<[In64BitMode]>;
+def : Pat<(i64 (zext (srl_su GR16:$src, (i8 8)))),
+ (SUBREG_TO_REG
+ (i64 0),
+ (MOVZX32_NOREXrr8
+ (EXTRACT_SUBREG (COPY_TO_REGCLASS GR16:$src, GR16_ABCD),
+ x86_subreg_8bit_hi)),
+ x86_subreg_32bit)>;
// h-register extract and store.
def : Pat<(store (i8 (trunc_su (srl_su GR64:$src, (i8 8)))), addr:$dst),
(MOV8mr_NOREX
addr:$dst,
- (EXTRACT_SUBREG (COPY_TO_REGCLASS GR64:$src, GR64_),
+ (EXTRACT_SUBREG (COPY_TO_REGCLASS GR64:$src, GR64_ABCD),
x86_subreg_8bit_hi))>;
def : Pat<(store (i8 (trunc_su (srl_su GR32:$src, (i8 8)))), addr:$dst),
(MOV8mr_NOREX
addr:$dst,
- (EXTRACT_SUBREG (COPY_TO_REGCLASS GR32:$src, GR32_),
+ (EXTRACT_SUBREG (COPY_TO_REGCLASS GR32:$src, GR32_ABCD),
x86_subreg_8bit_hi))>,
Requires<[In64BitMode]>;
def : Pat<(store (i8 (trunc_su (srl_su GR16:$src, (i8 8)))), addr:$dst),
(MOV8mr_NOREX
addr:$dst,
- (EXTRACT_SUBREG (COPY_TO_REGCLASS GR16:$src, GR16_),
+ (EXTRACT_SUBREG (COPY_TO_REGCLASS GR16:$src, GR16_ABCD),
x86_subreg_8bit_hi))>,
Requires<[In64BitMode]>;
projects / oota-llvm.git / blobdiff