RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 11:42:57 +01:00
Code Issues Projects Releases Wiki Activity

Fold more shifts into inserts, and update the README

Browse Source 
llvm-svn: 28168
This commit is contained in:
Nate Begeman 2006-05-08 17:38:32 +00:00
parent bbe4393bc4
commit a706539a72
2 changed files with 34 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
lib/Target/PowerPC/PPCISelDAGToDAG.cpp
View File

@ -392,25 +392,25 @@ static bool isIntImmediate(SDOperand N, unsigned& Imm) {
/// SelectBitfieldInsert - turn an or of two masked values into /// SelectBitfieldInsert - turn an or of two masked values into
/// the rotate left word immediate then mask insert (rlwimi) instruction. /// the rotate left word immediate then mask insert (rlwimi) instruction.
SDNode *PPCDAGToDAGISel::SelectBitfieldInsert(SDNode *N) { SDNode *PPCDAGToDAGISel::SelectBitfieldInsert(SDNode *N) {
unsigned TgtMask = 0xFFFFFFFF, InsMask = 0xFFFFFFFF, SH = 0;
unsigned Value;
SDOperand Op0 = N->getOperand(0); SDOperand Op0 = N->getOperand(0);
SDOperand Op1 = N->getOperand(1); SDOperand Op1 = N->getOperand(1);
unsigned Op0Opc = Op0.getOpcode();
unsigned Op1Opc = Op1.getOpcode();
uint64_t LKZ, LKO, RKZ, RKO; uint64_t LKZ, LKO, RKZ, RKO;
TLI.ComputeMaskedBits(Op0, TgtMask, LKZ, LKO); TLI.ComputeMaskedBits(Op0, 0xFFFFFFFFULL, LKZ, LKO);
TLI.ComputeMaskedBits(Op1, TgtMask, RKZ, RKO); TLI.ComputeMaskedBits(Op1, 0xFFFFFFFFULL, RKZ, RKO);
if ((LKZ | RKZ) == 0x00000000FFFFFFFFULL) { unsigned TargetMask = LKZ;
unsigned PInsMask = ~RKZ; unsigned InsertMask = RKZ;
unsigned PTgtMask = ~LKZ;
// If the LHS has a foldable shift, then swap it to the RHS so that we can if ((TargetMask | InsertMask) == 0xFFFFFFFF) {
// fold the shift into the insert. unsigned Op0Opc = Op0.getOpcode();
unsigned Op1Opc = Op1.getOpcode();
unsigned Value, SH = 0;
TargetMask = ~TargetMask;
InsertMask = ~InsertMask;
// If the LHS has a foldable shift and the RHS does not, then swap it to the
// RHS so that we can fold the shift into the insert.
if (Op0Opc == ISD::AND && Op1Opc == ISD::AND) { if (Op0Opc == ISD::AND && Op1Opc == ISD::AND) {
if (Op0.getOperand(0).getOpcode() == ISD::SHL || if (Op0.getOperand(0).getOpcode() == ISD::SHL ||
Op0.getOperand(0).getOpcode() == ISD::SRL) { Op0.getOperand(0).getOpcode() == ISD::SRL) {
@ -418,15 +418,22 @@ SDNode *PPCDAGToDAGISel::SelectBitfieldInsert(SDNode *N) {
Op1.getOperand(0).getOpcode() != ISD::SRL) { Op1.getOperand(0).getOpcode() != ISD::SRL) {
std::swap(Op0, Op1); std::swap(Op0, Op1);
std::swap(Op0Opc, Op1Opc); std::swap(Op0Opc, Op1Opc);
std::swap(PInsMask, PTgtMask); std::swap(TargetMask, InsertMask);
} }
} }
} else if (Op0Opc == ISD::SHL || Op0Opc == ISD::SRL) {
if (Op1Opc == ISD::AND && Op1.getOperand(0).getOpcode() != ISD::SHL &&
Op1.getOperand(0).getOpcode() != ISD::SRL) {
std::swap(Op0, Op1);
std::swap(Op0Opc, Op1Opc);
std::swap(TargetMask, InsertMask);
}
} }
unsigned MB, ME; unsigned MB, ME;
if (isRunOfOnes(PInsMask, MB, ME)) { if (isRunOfOnes(InsertMask, MB, ME)) {
SDOperand Tmp1, Tmp2, Tmp3; SDOperand Tmp1, Tmp2, Tmp3;
bool DisjointMask = (PTgtMask ^ PInsMask) == 0xFFFFFFFF; bool DisjointMask = (TargetMask ^ InsertMask) == 0xFFFFFFFF;
if ((Op1Opc == ISD::SHL || Op1Opc == ISD::SRL) && if ((Op1Opc == ISD::SHL || Op1Opc == ISD::SRL) &&
isIntImmediate(Op1.getOperand(1), Value)) { isIntImmediate(Op1.getOperand(1), Value)) {

15
lib/Target/PowerPC/README.txt
View File

@ -516,10 +516,17 @@ _foo:
srwi r4, r2, 30 srwi r4, r2, 30
srwi r5, r2, 31 srwi r5, r2, 31
or r4, r4, r5 or r4, r4, r5
slwi r4, r4, 31 rlwimi r2, r4, 31, 0, 0
rlwimi r4, r2, 0, 1, 31 stw r2, 0(r3)
stw r4, 0(r3)
blr blr
I *think* that could use another rlwimi. What this code is really doing is ORing bit 0 with bit 1. We could codegen this
as:
_foo:
lwz r2, 0(r3)
slwi r4, r2, 1
rlwinm r4, r4, 0, 0, 0
or r2, r2, r4
stw r2, 0(r3)
blr