Paper: The MLIR Transform Dialect. Your compiler is more powerful than you think

The authors adapt the MLIR framework to provide finer grained
control over the sequence of optimization steps a compiler uses.


    To take full advantage of a specific hardware target, performance
    engineers need to gain control on compilers in order to leverage their
    domain knowledge about the program and hardware. Yet, modern compilers
    are poorly controlled, usually by configuring a sequence of coarse-grained
    monolithic black-box passes, or by means of predefined compiler
    annotations/pragmas. These can be effective, but often do not let users
    precisely optimize their varying compute loads. As a consequence,
    performance engineers have to resort to implementing custom passes for a
    specific optimization heuristic, requiring compiler engineering expert
    knowledge.


    In this paper, we present a technique that provides fine-grained control
    of general-purpose compilers by introducing the Transform dialect, a
    controllable IR-based transformation system implemented in MLIR. The
    Transform dialect empowers performance engineers to optimize their
    various compute loads by composing and reusing existing - but currently hidden -
    compiler features without the need to implement new passes or even
    rebuilding the compiler.


    We demonstrate in five case studies that the Transform dialect enables
    precise, safe composition of compiler transformations and allows for
    straightforward integration with state-of-the-art search methods.


https://arxiv.org/abs/2409.03864


Regards,
John Levine, johnl@taugh.com, Taughannock Networks, Trumansburg NY
Please consider the environment before reading this e-mail. https://jl.ly

Return to the comp.compilers page.
Search the comp.compilers archives again.