December 7, 2000
Summery of efforts in Phase I :
- participation in modifying,
re-organizing, commenting, clean-up for header files inside sim_RT;
- modified other relevant files
and scripts run_bench to accommodate the new version of "RT_*.h"
- member of Profiler/Statics
group; worked on design and implementation of branch instruction profiling
and results back-annotation for X_code IR.
- Integration of profiler
module; set up environment (modified Automake files, comp_MAIN .c files
and scripts) to add profiler module "comp_PROF".
- other efforts including
updating files under comp_MAIN, comp_EMU, comp_IR to maintain and support
- added auto-generated address
information into X_code and emitted to exe.c, corrected region entry,
exit detection functions
- modified IMPACT parms
to emit debug information to X_code, added benchmark to test the results.
Summery of efforts in Phase II :
- currently working on the
design and implementation of Statistics module
- initial implementation of
STAT system (it's not our final design choice, however, parts of the work
are reusable) : initialize, record and report stats using hash table,
which allow user to specify which stats to collect and report through PARM
system; extended the interface to support variant stats type and data
structure; modified buildsim to automatically generate stat header files
and link to modules.
Besides, I'm also working on a research project that aims to
provide X_module with a static timing analysis tool. It will provide worst case
and best case scenario performance estimation, as well as other stats (e.g.
basic block execution counts) without overhead of simulation / or in case where
it's impossible to simulate all the traces. The results can also serve as a
feedback to help compiler tune the performance. Currently the interface from
X_code to our previous estimation tool (for simple embedded processor) has been
finished. An initial estimation module version is expected to be finished in