- Motivation:
  - TPCH Q01 profile
   - parallelism
   - efficiency
     - more cores than DB X
     - parallelism doesn't scale infinitly
     - nonparallel bits
- Identified bottlenecks:
  - plain expression evaluation
  - aggregate transition values
  - tuple deforming
  - grouping comparisons & hashing (aggregation, set functions)
  - sorting!
- very brief intro into expression evaluation
- So, what's JIT?
  - create executable code
  - linearize expression evalution
  - remove branches for deforming
  - inline functions
- example generated code
- example assembler
- Parts:
  - expression evaluation machinery (v10)
  - known tupledescs for scan, left, right tree
  - jit infrastructure (lifetime, errors, settings, inlining)
- Remaining work:
  - planner: when to jit and how much to optimize
  - executor:
    - OOM handling
    - expression engine for hashing
    - make tupledescs known in more places
    - determine NOT NULL in upper tupledescs
    - ...
  - JIT:
    - correct inlining implementation
- Qs