void RemoveBarrierPass::_runOnBlock( analysis::DataflowGraph::iterator block )
{
typedef analysis::DataflowGraph::InstructionVector::const_iterator
const_iterator;
for( const_iterator _instruction = block->instructions().begin();
_instruction != block->instructions().end(); ++_instruction )
{
ir::PTXInstruction& instruction = static_cast<
ir::PTXInstruction& >( *_instruction->i );
if( instruction.opcode == ir::PTXInstruction::Bar
|| ( instruction.opcode == ir::PTXInstruction::Call
&& !instruction.tailCall ) )
{
#if 0
if( _externals != 0
&& instruction.opcode == ir::PTXInstruction::Call )
{
if( _externals->find( instruction.a.identifier ) != 0 )
{
report( "Skipping external call "
<< instruction.toString() );
continue;
}
}
#endif
unsigned int bytes = _spillBytes;
_spillBytes = 1;
usesBarriers = true;
_removeBarrier( block, std::distance(
const_iterator( block->instructions().begin() ),
_instruction ) );
_spillBytes = std::max( bytes, _spillBytes );
++_reentryPoint;
_dfg().compute();
break;
}
}
}
void RemoveBarrierPass::_removeBarrier( analysis::DataflowGraph::iterator block,
unsigned int id )
{
typedef analysis::DataflowGraph::RegisterSet RegisterSet;
analysis::DataflowGraph::InstructionVector::const_iterator
_instruction( block->instructions().begin() );
std::advance( _instruction, id );
analysis::DataflowGraph::iterator exitBlock( _dfg().end() );
std::advance( exitBlock, -1 );
ir::PTXInstruction& instruction = static_cast< ir::PTXInstruction& >(
*_instruction->i );
bool isBarrier = instruction.opcode == ir::PTXInstruction::Bar;
if( isBarrier )
{
report( " Converting instruction " << instruction.toString() );
instruction.opcode = ir::PTXInstruction::Call;
instruction.tailCall = true;
instruction.branchTargetInstruction = -1;
instruction.a = ir::PTXOperand(
ir::PTXOperand::FunctionName, "_ZOcelotBarrierKernel");
instruction.d.addressMode = ir::PTXOperand::Invalid;
report( " Converted to " << instruction.toString() );
}
RegisterSet alive = block->alive( _instruction );
analysis::DataflowGraph::iterator bottom = _dfg().split( block,
id + 1, false );
_addSpillCode( block, bottom, alive, isBarrier );
_addRestoreCode( bottom, alive );
_dfg().redirect( block, bottom, exitBlock );
if( !isBarrier && instruction.pg.condition != ir::PTXOperand::PT )
{
_dfg().target( block, bottom, true );
}
_addEntryPoint( bottom );
}
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 );
}
}