Usage¶

class optimizer.Optimizer(optimization_classes=[])¶

This is the core of optimizer-model. This performs “on-line” optimization, meaning that operations are optimized as they’re added, not all at once at the end. By default it actually performs no optimizations, it just records the operations which are added, but you can easily add them:

from optimizer import Optimizer

opt = Optimizer()
opt = Optimizer([optimization, classes, here])

add_input(tp)¶

This adds input variables to a trace.

from optimizer import Types

i0 = opt.add_input(Types.INT)
i1 = opt.add_input(Types.INT)

add_operation(op, args, descr=None)¶

Adds an operation to the sequence of operations, and runs it through all of the optimizations. Returns a representation of the result.

from optimizer import Operations

i2 = opt.add_operation(Operations.INT_ADD, [i0, i1])
opt.add_operation(Operations.FINISH, [i2])

build_operations()¶

Returns a sequence of all of the operations, after optimizations:

ops = opt.build_operations()
assert len(ops) == 3

Optimizations¶

Out of the box, an Optimizer doesn’t actually run any optimizations, it just records the operations. However, optimizer-model includes many optimizations which can be plugged in.

The optimizations included with optimizer-model are:

class optimizer.optimizations.IntBounds¶: Keeps track of the possible bounds for an integer and propogates that data.

class optimizer.optimizations.ConstantFold¶: Performs constant folding on operations which do not have side-effects and which have all-constant arguments.

class optimizer.optimizations.GuardPropagation¶: Promotes values to be constant after they’ve been guarded against.

class optimizer.optimizations.Virtualize¶: Removes allocations which do not escapes the trace, and removes GETFIELD and SETFIELD operations on objects whose allocation has been removed.