replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix swapped BasePtr and Offset in pre-inc memory addresses. 

PPCTargetLowering::getPreIndexedAddressParts currently provides
the base part of a memory address in the offset result, and the
offset part in the base result.  That swap is then undone again
when an MI instruction is generated (in PPCDAGToDAGISel::Select
for loads, and using .md Pat patterns for stores).

This patch reverts this double swap, to make common code and
back-end be in sync as to which part of the address is base
and which is offset.

To avoid performance regressions in certain cases, target code
now checks whether the choice of base register would be rejected
for pre-inc accesses by common code, and attempts to swap base
and offset again in such cases.  (Overall, this means that now
pre-ice accesses are generated *more* frequently than before.)

llvm-svn: 177733
This commit is contained in:
Ulrich Weigand 2013-03-22 14:58:48 +00:00
parent d1b99d350c
commit e90b022468
4 changed files with 40 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
View File

@ -1111,7 +1111,7 @@ SDNode *PPCDAGToDAGISel::Select(SDNode *N) {
SDValue Chain = LD->getChain(); SDValue Chain = LD->getChain();
SDValue Base = LD->getBasePtr(); SDValue Base = LD->getBasePtr();
SDValue Ops[] = { Offset, Base, Chain }; SDValue Ops[] = { Base, Offset, Chain };
return CurDAG->getMachineNode(Opcode, dl, LD->getValueType(0), return CurDAG->getMachineNode(Opcode, dl, LD->getValueType(0),
PPCLowering.getPointerTy(), PPCLowering.getPointerTy(),
MVT::Other, Ops, 3); MVT::Other, Ops, 3);

22
llvm/lib/Target/PowerPC/PPCISelLowering.cpp
View File

@ -1192,6 +1192,7 @@ bool PPCTargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
SelectionDAG &DAG) const { SelectionDAG &DAG) const {
if (DisablePPCPreinc) return false; if (DisablePPCPreinc) return false;
bool isLoad = true;
SDValue Ptr; SDValue Ptr;
EVT VT; EVT VT;
unsigned Alignment; unsigned Alignment;
@ -1203,6 +1204,7 @@ bool PPCTargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
Ptr = ST->getBasePtr(); Ptr = ST->getBasePtr();
VT = ST->getMemoryVT(); VT = ST->getMemoryVT();
Alignment = ST->getAlignment(); Alignment = ST->getAlignment();
isLoad = false;
} else } else
return false; return false;
@ -1210,7 +1212,25 @@ bool PPCTargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
if (VT.isVector()) if (VT.isVector())
return false; return false;
if (SelectAddressRegReg(Ptr, Offset, Base, DAG)) { if (SelectAddressRegReg(Ptr, Base, Offset, DAG)) {
// Common code will reject creating a pre-inc form if the base pointer
// is a frame index, or if N is a store and the base pointer is either
// the same as or a predecessor of the value being stored. Check for
// those situations here, and try with swapped Base/Offset instead.
bool Swap = false;
if (isa<FrameIndexSDNode>(Base) || isa<RegisterSDNode>(Base))
Swap = true;
else if (!isLoad) {
SDValue Val = cast<StoreSDNode>(N)->getValue();
if (Val == Base || Base.getNode()->isPredecessorOf(Val.getNode()))
Swap = true;
}
if (Swap)
std::swap(Base, Offset);
AM = ISD::PRE_INC; AM = ISD::PRE_INC;
return true; return true;
} }

16
llvm/lib/Target/PowerPC/PPCInstr64Bit.td
View File

@ -875,14 +875,14 @@ def : Pat<(pre_truncsti32 G8RC:$rS, ptr_rc_nor0:$ptrreg, iaddroff:$ptroff),
def : Pat<(aligned4pre_store G8RC:$rS, ptr_rc_nor0:$ptrreg, iaddroff:$ptroff), def : Pat<(aligned4pre_store G8RC:$rS, ptr_rc_nor0:$ptrreg, iaddroff:$ptroff),
(STDU G8RC:$rS, iaddroff:$ptroff, ptr_rc_nor0:$ptrreg)>; (STDU G8RC:$rS, iaddroff:$ptroff, ptr_rc_nor0:$ptrreg)>;
def : Pat<(pre_truncsti8 G8RC:$rS, ptr_rc:$ptrreg, ptr_rc_nor0:$ptroff), def : Pat<(pre_truncsti8 G8RC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff),
(STBUX8 G8RC:$rS, ptr_rc_nor0:$ptroff, ptr_rc:$ptrreg)>; (STBUX8 G8RC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff)>;
def : Pat<(pre_truncsti16 G8RC:$rS, ptr_rc:$ptrreg, ptr_rc_nor0:$ptroff), def : Pat<(pre_truncsti16 G8RC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff),
(STHUX8 G8RC:$rS, ptr_rc_nor0:$ptroff, ptr_rc:$ptrreg)>; (STHUX8 G8RC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff)>;
def : Pat<(pre_truncsti32 G8RC:$rS, ptr_rc:$ptrreg, ptr_rc_nor0:$ptroff), def : Pat<(pre_truncsti32 G8RC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff),
(STWUX8 G8RC:$rS, ptr_rc_nor0:$ptroff, ptr_rc:$ptrreg)>; (STWUX8 G8RC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff)>;
def : Pat<(pre_store G8RC:$rS, ptr_rc:$ptrreg, ptr_rc_nor0:$ptroff), def : Pat<(pre_store G8RC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff),
(STDUX G8RC:$rS, ptr_rc_nor0:$ptroff, ptr_rc:$ptrreg)>; (STDUX G8RC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff)>;
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//

20
llvm/lib/Target/PowerPC/PPCInstrInfo.td
View File

@ -1004,16 +1004,16 @@ def STFDUX: XForm_8<31, 759, (outs ptr_rc_nor0:$ea_res), (ins F8RC:$rS, memrr:$d
// Patterns to match the pre-inc stores. We can't put the patterns on // Patterns to match the pre-inc stores. We can't put the patterns on
// the instruction definitions directly as ISel wants the address base // the instruction definitions directly as ISel wants the address base
// and offset to be separate operands, not a single complex operand. // and offset to be separate operands, not a single complex operand.
def : Pat<(pre_truncsti8 GPRC:$rS, ptr_rc:$ptrreg, ptr_rc_nor0:$ptroff), def : Pat<(pre_truncsti8 GPRC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff),
(STBUX GPRC:$rS, ptr_rc_nor0:$ptroff, ptr_rc:$ptrreg)>; (STBUX GPRC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff)>;
def : Pat<(pre_truncsti16 GPRC:$rS, ptr_rc:$ptrreg, ptr_rc_nor0:$ptroff), def : Pat<(pre_truncsti16 GPRC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff),
(STHUX GPRC:$rS, ptr_rc_nor0:$ptroff, ptr_rc:$ptrreg)>; (STHUX GPRC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff)>;
def : Pat<(pre_store GPRC:$rS, ptr_rc:$ptrreg, ptr_rc_nor0:$ptroff), def : Pat<(pre_store GPRC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff),
(STWUX GPRC:$rS, ptr_rc_nor0:$ptroff, ptr_rc:$ptrreg)>; (STWUX GPRC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff)>;
def : Pat<(pre_store F4RC:$rS, ptr_rc:$ptrreg, ptr_rc_nor0:$ptroff), def : Pat<(pre_store F4RC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff),
(STFSUX F4RC:$rS, ptr_rc_nor0:$ptroff, ptr_rc:$ptrreg)>; (STFSUX F4RC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff)>;
def : Pat<(pre_store F8RC:$rS, ptr_rc:$ptrreg, ptr_rc_nor0:$ptroff), def : Pat<(pre_store F8RC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff),
(STFDUX F8RC:$rS, ptr_rc_nor0:$ptroff, ptr_rc:$ptrreg)>; (STFDUX F8RC:$rS, ptr_rc_nor0:$ptrreg, ptr_rc:$ptroff)>;
def SYNC : XForm_24_sync<31, 598, (outs), (ins), def SYNC : XForm_24_sync<31, 598, (outs), (ins),
"sync", LdStSync, "sync", LdStSync,