PALMED: Throughput Characterization for Superscalar Architectures
Team work: Nicolas Derumigny presented "PALMED: Throughput Characterization for Superscalar Architectures" (CGO'22) at 4A312 the 26/4/2024 at 10h30.
Abstract
Short : How to characterise performances of an "unknown" CPU.
Long : In a super-scalar architecture, the scheduler dynamically assigns micro-operations (μOPs) to execution ports. The port mapping of an architecture describes how an instruction decomposes into μOPs and lists for each μOP the set of ports it can be mapped to. It is used by compilers and performance debugging tools to characterize the performance throughput of a sequence of instructions repeatedly executed as the core component of a loop.
This paper presents a technique for automated dicovery of this behavior based on an abstract representation of the CPU internals. This representation is proved to be equivalent to the usual µOPs-derived mapping on thoughput prediction, but is computationnally less complex to use for reverse engineering.
