Class to fuse certain pattern of subgraph into CircleBCQFullyConnected or CircleBCQGather. More...

#include <FuseBCQPass.h>

Collaboration diagram for luci::FuseBCQPass:

[legend]

Public Member Functions
const char *	name (void) const final

bool	run (luci::Module *m) final

bool	run (loco::Graph *g) final
	Run the pass.

Public Member Functions inherited from logo::Pass
virtual	~Pass ()=default

Detailed Description

Class to fuse certain pattern of subgraph into CircleBCQFullyConnected or CircleBCQGather.

Definition at line 29 of file FuseBCQPass.h.

Member Function Documentation

◆ name()

const char * luci::FuseBCQPass::name ( void ) const

inlinefinalvirtual

Reimplemented from logo::Pass.

Definition at line 31 of file FuseBCQPass.h.

31{ return "luci::FuseBCQPass"; }

◆ run() [1/2]

bool luci::FuseBCQPass::run ( loco::Graph * graph )

finalvirtual

Run the pass.

Returns: false if there was nothing changed

Implements logo::Pass.

Definition at line 692 of file FuseBCQPass.cpp.

{
  // Do nothing for graph
  return false;
}

Referenced by package.infer.session::inference().

◆ run() [2/2]

bool luci::FuseBCQPass::run ( luci::Module * m )

finalvirtual

Implements luci::Pass.

Definition at line 613 of file FuseBCQPass.cpp.

{
  bool changed = false;
 
  const int32_t start_magicnum = -2e9 + 27;
  const int32_t end_magicnum = 2e9 - 27;
 
  loco::Graph *main_graph = m->graph(0);
 
  luci::CircleConst *metadata_node = nullptr;
  for (auto node : loco::output_nodes(main_graph))
  {
    auto output_node = loco::must_cast<luci::CircleOutput *>(node);
 
    // Metadata node should be first output
    if (output_node->index() != 0)
      continue;
 
    // Metadata should be constant and dtype should be S32
    auto const_node = dynamic_cast<luci::CircleConst *>(output_node->from());
    if (const_node == nullptr || const_node->dtype() != loco::DataType::S32)
      continue;
 
    // Metadata has at least four elements
    const auto element_cnt = const_node->size<loco::DataType::S32>();
    if (element_cnt < 4)
      continue;
 
    // Metadata has magic numbers at first and at last
    const auto start_value = const_node->at<loco::DataType::S32>(0);
    const auto end_value = const_node->at<loco::DataType::S32>(element_cnt - 1);
    if (start_value == start_magicnum && end_value == end_magicnum)
    {
      metadata_node = const_node;
      break;
    }
  }
 
  if (metadata_node != nullptr)
  {
    const auto bcq_version = metadata_node->at<loco::DataType::S32>(1);
    const auto original_output_cnt = metadata_node->at<loco::DataType::S32>(2);
 
    if (bcq_version == 1)
    {
      const auto bundle_cnt = metadata_node->at<loco::DataType::S32>(3);
 
      BCQFuser<1> fuser{original_output_cnt, bundle_cnt};
      fuser.register_bcq_info(main_graph);
 
      for (size_t g = 0; g < m->size(); ++g)
        if (fuser.fuseBCQ(m->graph(g)))
          changed = true;
    }
    else
    {
      LOGGER(l);
      WARN(l) << "Not supported BCQ version is found." << std::endl;
    }
 
    // Remove all of BCQ information nodes iff there is no change
    if (changed == false)
    {
      for (auto node : loco::output_nodes(main_graph))
      {
        auto output_node = loco::must_cast<luci::CircleOutput *>(node);
        if (output_node->index() == 0 || (int)output_node->index() > original_output_cnt)
        {
          auto noOp = main_graph->nodes()->create<luci::CircleOutputExclude>();
          output_node->from(noOp);
          changed = true;
        }
      }
    }
  }
 
  return changed;
}

References luci::CircleConst::at(), loco::NodePool::create(), luci::CircleOutput::from(), luci::CircleOutput::index(), LOGGER, m, loco::Graph::nodes(), luci::output_node(), loco::output_nodes(), luci::CircleConst::size(), and WARN.