Parallel query execution

[pmenon]

This is a PR for parallel query execution in codegen.

• Supported parallel operators:
  • [x] Sequential scans
  • [x] Sorting
  • [x] Hash joins
  • [ ] Aggregations
  • [ ] Inserts (purposely disabled until we have COPY)
  • [ ] Updates
  • [ ] Deletes
• Parallelism is done in fork-join manner.
• Settings added to toggle parallel execution, and set the minimum table size threshold before parallel execution kicks in. This logic should be modified.
  • @chenboy Check plan_generator to see if the logic looks good.
  • I've split up RuntimeState into QueryState and PipelineState.
  • QueryState exists for the lifetime of the query. It is initialized in the init() function, and torn down in the tearDown() function.
  • PipelineState exists for only one pipeline. It is created on entry to the pipeline function, and is torn down upon exit.
  • I've cleaned up proxies to simplify loading/store struct member variables using names rather than index positions.
  • What used to be codegen.CreateGEP(..., 0, 2) to load struct elements is now codegen.Load(HashTableProxy::directory, ...).
  • Similarly done for storing elements of a struct: codegen.Store(...)
• Removed CCHashTable, added generic HashTable that will eventually be used for both joins and aggregations.

Review notes:

• Most of the heavy lifting is done in pipeline.cpp. Pipelines can be run serially or in parallel.
  • This means all operators need to pass a std::function when launching pipelines.
• Many of the changed files are in the proxies, making is simpler to create them.

TODO:

- Remove std::mutex from buffer output (This will be another PR) - Enable parallel aggregations (This will be another PR when parallel agg team is done) - Add tests for memory leak (Done)

[coveralls]

Coverage decreased (-0.2%) to 77.406% when pulling 1a70093a0eeccbe50edd2a92ad17c066c97a4a34 on pmenon:mt into 65915234edf5c33acda078f632012887291d22b7 on cmu-db:master.

[pmenon]

@tcm-marcel Want to get started on this review? It's quite big, so I don't want to leave it un-merged for too long.

[pmenon]

@tcm-marcel A small final change, then it's good to go.