How To Use Tiramisu

Tiramisu provides few classes to enable users to represent their program:

The tiramisu::computation class: a computation is composed of an expression and an iteration space but is not associated with any memory location.
The tiramisu::function class: a function is composed of multiple computations and a vector of arguments (functions arguments).
The tiramisu::buffer: a class to represent memory buffers.

In general, in order to use Tiramisu to optimize, all what a user needs to do is the following:

Represent the program that needs to be optimized
- Instantiate a tiramisu::function,
- Instantiate a set of tiramisu::computation objects for each function,
Provide the list of optimizations (memory mapping and schedule)
- Provide the mapping of each tiramisu::computation to memory (i.e. where each computation should be stored in memory),
- Provide the schedule of each tiramisu::computation (a list of loop nest transformations and optimizations such as tiling, parallelization, vectorization, fusion, ...),
Generate code
- Generate an AST (Abstract Syntax Tree),
- Generate target code (an object file),

Provide feedback