void RemoveBarrierPass::_addEntryPoint(
analysis::DataflowGraph::iterator block )
{
analysis::DataflowGraph::iterator entry = _dfg().insert(_dfg().begin());
ir::PTXInstruction move( ir::PTXInstruction::Mov );
move.type = ir::PTXOperand::u32;
move.addressSpace = ir::PTXInstruction::Local;
move.a.identifier = "_Zocelot_resume_point";
move.a.addressMode = ir::PTXOperand::Address;
move.a.type = ir::PTXOperand::u32;
move.d.reg = _tempRegister();
move.d.addressMode = ir::PTXOperand::Register;
move.d.type = ir::PTXOperand::u32;
_dfg().insert( entry, move, 0 );
ir::PTXInstruction load( ir::PTXInstruction::Ld );
load.addressSpace = ir::PTXInstruction::Local;
load.type = ir::PTXOperand::u32;
load.a = move.d;
load.d.reg = _tempRegister();
load.d.addressMode = ir::PTXOperand::Register;
load.d.type = ir::PTXOperand::u32;
_dfg().insert( entry, load, 1 );
ir::PTXInstruction setp( ir::PTXInstruction::SetP );
setp.type = ir::PTXOperand::u32;
setp.comparisonOperator = ir::PTXInstruction::Eq;
setp.d.reg = _tempRegister();
setp.d.addressMode = ir::PTXOperand::Register;
setp.d.type = ir::PTXOperand::pred;
setp.a = load.d;
setp.b.addressMode = ir::PTXOperand::Immediate;
setp.b.type = ir::PTXOperand::u32;
setp.b.imm_uint = block->id();
_dfg().insert( entry, setp, 2 );
ir::PTXInstruction branch( ir::PTXInstruction::Bra );
branch.d.addressMode = ir::PTXOperand::Label;
branch.d.identifier = block->label();
branch.pg = setp.d;
_dfg().insert( entry, branch, 3 );
_dfg().target( entry, block );
}
void RemoveBarrierPass::_addSpillCode( analysis::DataflowGraph::iterator block,
analysis::DataflowGraph::iterator target,
const analysis::DataflowGraph::RegisterSet& alive, bool isBarrier )
{
unsigned int bytes = 0;
ir::PTXInstruction move ( ir::PTXInstruction::Mov );
move.type = ir::PTXOperand::u32;
move.addressSpace = ir::PTXInstruction::Local;
move.a.identifier = "_Zocelot_spill_area";
move.a.addressMode = ir::PTXOperand::Address;
move.a.type = ir::PTXOperand::u32;
move.d.reg = _tempRegister();
move.d.addressMode = ir::PTXOperand::Register;
move.d.type = ir::PTXOperand::u32;
_dfg().insert( block, move, block->instructions().size() - 1 );
report( " Saving " << alive.size() << " Registers" );
for( analysis::DataflowGraph::RegisterSet::const_iterator
reg = alive.begin(); reg != alive.end(); ++reg )
{
report( " r" << reg->id << " ("
<< ir::PTXOperand::bytes( reg->type ) << " bytes)" );
ir::PTXInstruction save( ir::PTXInstruction::St );
save.type = reg->type;
save.addressSpace = ir::PTXInstruction::Local;
save.d.addressMode = ir::PTXOperand::Indirect;
save.d.reg = move.d.reg;
save.d.type = ir::PTXOperand::u32;
save.d.offset = bytes;
bytes += ir::PTXOperand::bytes( save.type );
save.a.addressMode = ir::PTXOperand::Register;
save.a.type = reg->type;
save.a.reg = reg->id;
_dfg().insert( block, save,
block->instructions().size() - 1 );
}
_spillBytes = std::max( bytes, _spillBytes );
move.type = ir::PTXOperand::u32;
move.addressSpace = ir::PTXInstruction::Local;
move.a.identifier = "_Zocelot_resume_point";
move.a.addressMode = ir::PTXOperand::Address;
move.a.type = ir::PTXOperand::u32;
move.d.reg = _tempRegister();
move.d.addressMode = ir::PTXOperand::Register;
move.d.type = ir::PTXOperand::u32;
_dfg().insert( block, move, block->instructions().size() - 1 );
ir::PTXInstruction save( ir::PTXInstruction::St );
save.type = ir::PTXOperand::u32;
save.addressSpace = ir::PTXInstruction::Local;
save.d.addressMode = ir::PTXOperand::Indirect;
save.d.reg = move.d.reg;
save.d.type = ir::PTXOperand::u32;
save.a.addressMode = ir::PTXOperand::Immediate;
save.a.type = ir::PTXOperand::u32;
save.a.imm_uint = target->id();
_dfg().insert( block, save, block->instructions().size() - 1 );
if( isBarrier )
{
move.d.reg = _tempRegister();
move.a.identifier = "_Zocelot_barrier_next_kernel";
_dfg().insert( block, move,
block->instructions().size() - 1 );
save.d.reg = move.d.reg;
save.a.imm_uint = _kernelId;
_dfg().insert( block, save,
block->instructions().size() - 1 );
}
}