Abstract InstructionForm

[cod3monk]

Have a class representation of an InstructionForm, which can be created from an assembly line or by describing a mnemonic and its parameters.

This class should be able to generate a canonical string representation and handle (or at least reference) performance relevant information.

[JanLJL]

After applying the semantics with add_semantics (and if wished the optimal throughput with assign_optimal_throughput) to the kernel, each element of the kernel list inherits the following attributes as dict:

>>> kernel[0].keys()
dict_keys(['instruction', 'operands', 'directive', 'comment', 'label', 'line', 
'line_number', 'semantic_operands', 'port_pressure', 'port_uops', 'flags', 
'throughput', 'latency', 'latency_wo_load', 'latency_cp', 'latency_lcd'])

This object includes a canonical number (line_number), the original string representation (line) and performance relevant information (basically all other attributes). Is that sufficient for you?

[cod3monk]

It's rather tedious to work with, some OO programming could be helpful here. E.g., a class InstructionForm, which captures the parsed information found in an assembly code line. Then this object can be passed a machine model, which allows it to relate it self to a mnemonic and offers performance relevant information. And so on.

Not urgent and requires some rewriting, but makes it easier to work with from a programmers point of view.