6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2025-01-05 12:31:33 +00:00
Code Issues Projects Releases Wiki Activity

Refactor out code to fast-isel a memcpy operation with a small constant

Browse Source 
length.  (I'm planning to use this to implement byval.)



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130274 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Eli Friedman 2011-04-27 01:45:07 +00:00
parent 9266cc400e
commit d5089a9794
1 changed files with 40 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

72
lib/Target/X86/X86FastISel.cpp
View File

@ -132,6 +132,9 @@ private:
} }
bool isTypeLegal(const Type *Ty, MVT &VT, bool AllowI1 = false); bool isTypeLegal(const Type *Ty, MVT &VT, bool AllowI1 = false);
bool TryEmitSmallMemcpy(X86AddressMode DestAM,
X86AddressMode SrcAM, uint64_t Len);
}; };
} // end anonymous namespace. } // end anonymous namespace.
@ -1323,6 +1326,40 @@ bool X86FastISel::X86SelectExtractValue(const Instruction *I) {
return false; return false;
} }
bool X86FastISel::TryEmitSmallMemcpy(X86AddressMode DestAM,
X86AddressMode SrcAM, uint64_t Len) {
// Make sure we don't bloat code by inlining very large memcpy's.
bool i64Legal = TLI.isTypeLegal(MVT::i64);
if (Len > (i64Legal ? 32 : 16)) return false;
// We don't care about alignment here since we just emit integer accesses.
while (Len) {
MVT VT;
if (Len >= 8 && i64Legal)
VT = MVT::i64;
else if (Len >= 4)
VT = MVT::i32;
else if (Len >= 2)
VT = MVT::i16;
else {
assert(Len == 1);
VT = MVT::i8;
}
unsigned Reg;
bool RV = X86FastEmitLoad(VT, SrcAM, Reg);
RV &= X86FastEmitStore(VT, Reg, DestAM);
assert(RV && "Failed to emit load or store??");
unsigned Size = VT.getSizeInBits()/8;
Len -= Size;
DestAM.Disp += Size;
SrcAM.Disp += Size;
}
return true;
}
bool X86FastISel::X86VisitIntrinsicCall(const IntrinsicInst &I) { bool X86FastISel::X86VisitIntrinsicCall(const IntrinsicInst &I) {
// FIXME: Handle more intrinsics. // FIXME: Handle more intrinsics.
switch (I.getIntrinsicID()) { switch (I.getIntrinsicID()) {
@ -1332,45 +1369,16 @@ bool X86FastISel::X86VisitIntrinsicCall(const IntrinsicInst &I) {
// Don't handle volatile or variable length memcpys. // Don't handle volatile or variable length memcpys.
if (MCI.isVolatile() || !isa<ConstantInt>(MCI.getLength())) if (MCI.isVolatile() || !isa<ConstantInt>(MCI.getLength()))
return false; return false;
// Don't inline super long memcpys. We could lower these to a memcpy call,
// but we might as well bail out.
uint64_t Len = cast<ConstantInt>(MCI.getLength())->getZExtValue(); uint64_t Len = cast<ConstantInt>(MCI.getLength())->getZExtValue();
bool i64Legal = TLI.isTypeLegal(MVT::i64);
if (Len > (i64Legal ? 32 : 16)) return false;
// Get the address of the dest and source addresses. // Get the address of the dest and source addresses.
X86AddressMode DestAM, SrcAM; X86AddressMode DestAM, SrcAM;
if (!X86SelectAddress(MCI.getRawDest(), DestAM) || if (!X86SelectAddress(MCI.getRawDest(), DestAM) ||
!X86SelectAddress(MCI.getRawSource(), SrcAM)) !X86SelectAddress(MCI.getRawSource(), SrcAM))
return false; return false;
// We don't care about alignment here since we just emit integer accesses. return TryEmitSmallMemcpy(DestAM, SrcAM, Len);
while (Len) {
MVT VT;
if (Len >= 8 && i64Legal)
VT = MVT::i64;
else if (Len >= 4)
VT = MVT::i32;
else if (Len >= 2)
VT = MVT::i16;
else {
assert(Len == 1);
VT = MVT::i8;
}
unsigned Reg;
bool RV = X86FastEmitLoad(VT, SrcAM, Reg);
RV &= X86FastEmitStore(VT, Reg, DestAM);
assert(RV && "Failed to emit load or store??");
unsigned Size = VT.getSizeInBits()/8;
Len -= Size;
DestAM.Disp += Size;
SrcAM.Disp += Size;
}
return true;
} }
case Intrinsic::stackprotector: { case Intrinsic::stackprotector: {