[PTLsim-devel] Cycle accuracy of flushes from mis-predicted branches?

Tue Mar 13 20:45:12 EDT 2007

On 3/9/07, Edward Lee <eslee3 at uiuc.edu> wrote:
> * If I'm reading this correctly, the ooocore recovers from a
> mispredict in 0.76 cycles on average?

For those curious, it seems like the reason for this is because PTLsim
was detecting self-modifying-code [1] and flushing the pipeline
anyway, so flushing again on a branch mispredict wasn't too much
worse.

Applying the patch from [1] or spacing out text and data segments
alleviates this issue of low overhead of branch mispredicts.

[1] https://ptlsim.org/pipermail/ptlsim-devel/2007-March/000013.html

-- 
Ed