return false;
}]>;
+def gsload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
+ LoadSDNode *LD = cast<LoadSDNode>(N);
+ const Value *Src = LD->getSrcValue();
+ if (!Src)
+ return false;
+ if (const PointerType *PT = dyn_cast<PointerType>(Src->getType()))
+ return PT->getAddressSpace() == 256;
+ return false;
+}]>;
+
def loadi8 : PatFrag<(ops node:$ptr), (i8 (load node:$ptr))>;
def loadi64 : PatFrag<(ops node:$ptr), (i64 (load node:$ptr))>;
}
} // Defs = [EFLAGS]
+// Match xor -1 to not. Favors these over a move imm + xor to save code size.
+let AddedComplexity = 15 in {
def NOT8r : I<0xF6, MRM2r, (outs GR8 :$dst), (ins GR8 :$src), "not{b}\t$dst",
[(set GR8:$dst, (not GR8:$src))]>;
def NOT16r : I<0xF7, MRM2r, (outs GR16:$dst), (ins GR16:$src), "not{w}\t$dst",
[(set GR16:$dst, (not GR16:$src))]>, OpSize;
def NOT32r : I<0xF7, MRM2r, (outs GR32:$dst), (ins GR32:$src), "not{l}\t$dst",
[(set GR32:$dst, (not GR32:$src))]>;
+}
let isTwoAddress = 0 in {
def NOT8m : I<0xF6, MRM2m, (outs), (ins i8mem :$dst), "not{b}\t$dst",
[(store (not (loadi8 addr:$dst)), addr:$dst)]>;
} // Defs = [EFLAGS]
// Bit tests.
-// TODO: BT with immediate operands
// TODO: BTC, BTR, and BTS
let Defs = [EFLAGS] in {
-def BT16rr : I<0xA3, MRMSrcReg, (outs), (ins GR16:$src1, GR16:$src2),
+def BT16rr : I<0xA3, MRMDestReg, (outs), (ins GR16:$src1, GR16:$src2),
"bt{w}\t{$src2, $src1|$src1, $src2}",
[(X86bt GR16:$src1, GR16:$src2),
(implicit EFLAGS)]>, OpSize, TB;
-def BT32rr : I<0xA3, MRMSrcReg, (outs), (ins GR32:$src1, GR32:$src2),
+def BT32rr : I<0xA3, MRMDestReg, (outs), (ins GR32:$src1, GR32:$src2),
"bt{l}\t{$src2, $src1|$src1, $src2}",
[(X86bt GR32:$src1, GR32:$src2),
(implicit EFLAGS)]>, TB;
-def BT16mr : I<0xA3, MRMDestMem, (outs), (ins i16mem:$src1, GR16:$src2),
- "bt{w}\t{$src2, $src1|$src1, $src2}",
- [(X86bt (loadi16 addr:$src1), GR16:$src2),
- (implicit EFLAGS)]>, OpSize, TB, Requires<[FastBTMem]>;
-def BT32mr : I<0xA3, MRMDestMem, (outs), (ins i32mem:$src1, GR32:$src2),
- "bt{l}\t{$src2, $src1|$src1, $src2}",
- [(X86bt (loadi32 addr:$src1), GR32:$src2),
- (implicit EFLAGS)]>, TB, Requires<[FastBTMem]>;
+
+// Unlike with the register+register form, the memory+register form of the
+// bt instruction does not ignore the high bits of the index. From ISel's
+// perspective, this is pretty bizarre. Disable these instructions for now.
+//def BT16mr : I<0xA3, MRMDestMem, (outs), (ins i16mem:$src1, GR16:$src2),
+// "bt{w}\t{$src2, $src1|$src1, $src2}",
+// [(X86bt (loadi16 addr:$src1), GR16:$src2),
+// (implicit EFLAGS)]>, OpSize, TB, Requires<[FastBTMem]>;
+//def BT32mr : I<0xA3, MRMDestMem, (outs), (ins i32mem:$src1, GR32:$src2),
+// "bt{l}\t{$src2, $src1|$src1, $src2}",
+// [(X86bt (loadi32 addr:$src1), GR32:$src2),
+// (implicit EFLAGS)]>, TB, Requires<[FastBTMem]>;
+
+def BT16ri8 : Ii8<0xBA, MRM4r, (outs), (ins GR16:$src1, i16i8imm:$src2),
+ "bt{w}\t{$src2, $src1|$src1, $src2}",
+ [(X86bt GR16:$src1, i16immSExt8:$src2),
+ (implicit EFLAGS)]>, OpSize, TB;
+def BT32ri8 : Ii8<0xBA, MRM4r, (outs), (ins GR32:$src1, i32i8imm:$src2),
+ "bt{l}\t{$src2, $src1|$src1, $src2}",
+ [(X86bt GR32:$src1, i32immSExt8:$src2),
+ (implicit EFLAGS)]>, TB;
+// Note that these instructions don't need FastBTMem because that
+// only applies when the other operand is in a register. When it's
+// an immediate, bt is still fast.
+def BT16mi8 : Ii8<0xBA, MRM4m, (outs), (ins i16mem:$src1, i16i8imm:$src2),
+ "bt{w}\t{$src2, $src1|$src1, $src2}",
+ [(X86bt (loadi16 addr:$src1), i16immSExt8:$src2),
+ (implicit EFLAGS)]>, OpSize, TB;
+def BT32mi8 : Ii8<0xBA, MRM4m, (outs), (ins i32mem:$src1, i32i8imm:$src2),
+ "bt{l}\t{$src2, $src1|$src1, $src2}",
+ [(X86bt (loadi32 addr:$src1), i32immSExt8:$src2),
+ (implicit EFLAGS)]>, TB;
} // Defs = [EFLAGS]
// Sign/Zero extenders
"movl\t%gs:0, $dst",
[(set GR32:$dst, X86TLStp)]>, SegGS;
+let AddedComplexity = 5 in
+def GS_MOV32rm : I<0x8B, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
+ "movl\t%gs:$src, $dst",
+ [(set GR32:$dst, (gsload addr:$src))]>, SegGS;
+
//===----------------------------------------------------------------------===//
// DWARF Pseudo Instructions
//
def : Pat<(parallel (X86cmp GR32:$src1, 0), (implicit EFLAGS)),
(TEST32rr GR32:$src1, GR32:$src1)>;
+// Conditional moves with folded loads with operands swapped and conditions
+// inverted.
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_B, EFLAGS),
+ (CMOVAE16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_B, EFLAGS),
+ (CMOVAE32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_AE, EFLAGS),
+ (CMOVB16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_AE, EFLAGS),
+ (CMOVB32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_E, EFLAGS),
+ (CMOVNE16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_E, EFLAGS),
+ (CMOVNE32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_NE, EFLAGS),
+ (CMOVE16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_NE, EFLAGS),
+ (CMOVE32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_BE, EFLAGS),
+ (CMOVA16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_BE, EFLAGS),
+ (CMOVA32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_A, EFLAGS),
+ (CMOVBE16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_A, EFLAGS),
+ (CMOVBE32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_L, EFLAGS),
+ (CMOVGE16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_L, EFLAGS),
+ (CMOVGE32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_GE, EFLAGS),
+ (CMOVL16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_GE, EFLAGS),
+ (CMOVL32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_LE, EFLAGS),
+ (CMOVG16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_LE, EFLAGS),
+ (CMOVG32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_G, EFLAGS),
+ (CMOVLE16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_G, EFLAGS),
+ (CMOVLE32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_P, EFLAGS),
+ (CMOVNP16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_P, EFLAGS),
+ (CMOVNP32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_NP, EFLAGS),
+ (CMOVP16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_NP, EFLAGS),
+ (CMOVP32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_S, EFLAGS),
+ (CMOVNS16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_S, EFLAGS),
+ (CMOVNS32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_NS, EFLAGS),
+ (CMOVS16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_NS, EFLAGS),
+ (CMOVS32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_O, EFLAGS),
+ (CMOVNO16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_O, EFLAGS),
+ (CMOVNO32rm GR32:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, X86_COND_NO, EFLAGS),
+ (CMOVO16rm GR16:$src2, addr:$src1)>;
+def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, X86_COND_NO, EFLAGS),
+ (CMOVO32rm GR32:$src2, addr:$src1)>;
+
// zextload bool -> zextload byte
def : Pat<(zextloadi8i1 addr:$src), (MOV8rm addr:$src)>;
def : Pat<(zextloadi16i1 addr:$src), (MOVZX16rm8 addr:$src)>;
(implicit EFLAGS)),
(IMUL32rmi8 addr:$src1, i32immSExt8:$src2)>;
+// Optimize multiple with overflow by 2.
+let AddedComplexity = 2 in {
+def : Pat<(parallel (X86smul_ovf GR16:$src1, 2),
+ (implicit EFLAGS)),
+ (ADD16rr GR16:$src1, GR16:$src1)>;
+
+def : Pat<(parallel (X86smul_ovf GR32:$src1, 2),
+ (implicit EFLAGS)),
+ (ADD32rr GR32:$src1, GR32:$src1)>;
+}
+
//===----------------------------------------------------------------------===//
// Floating Point Stack Support
//===----------------------------------------------------------------------===//