Hi, Flexus uses Simics's checkpointing mechanism to store architectural (processor, memory, and device) state. Simics does not use live-state, it instead saves a differential image of all memory locations that have changed since the previous Simics checkpoint. This means that each flexpoint has the whole memory image available, but flexpoints are not self-contained --- each depends on all previous flexpoints to be able to build the complete memory image.
The Simics documentation describes its checkpointing features in more detail. Regards, -Thomas Wenisch On Jun 27, 2012, at 11:03 AM, [email protected] wrote: > Hi, > I need a clarification about flexus simulator. It's about what is actually > saved in a flex point. > The "Simulation Sampling with Live-Points" paper in section 5 proposes the > live state technique. > Basically the idea is to save in checkpoints (live points) only the state > (memory etc) that will be accessed during the simulation window and also all > uninitialized state accessed from the wrong path is random values (page 6 > column 1 last paragraph). > From what I understood a flex point doesn't implements live-state instead a > flex point contains the differences between the WHOLE memory state. > So did I understand correct? The flex points implement live state or not? > > Thanks - Zacharias Hadjilambrou > University of Cyprus Undergraduate student
