Testing the performance of instruction Scheduler

[phani narasimhan venkata sesha myreddy]

hai all,
How do i test the performance of instruction scheduler.
I have written an instruction scheduler for a database code generator.
The scheduler is simple and tuned to the graph generated, as i get
some class of graphs. It contains the global, local scheduler and the
instruciton grouping logic.

I wonder i did not got the expected performance. The gain for some
graphs
is 6 to 8 %. The worst part is the performance goes down for some
graphs.
The graph size is very huge. some times 200 blocks and 2000
instructions.
The instructions generated is very optimal.

I read the ultrasparc-II manual and in it the instructions are properly
scheduled
it is possible to bring CPI as low as 0.25. I am not able to visualise
this much percentage gain overall.

Is there any tool to calculate the number of cycles taken by the code
at run-time?

Is there any tool to find for a specific run how many data cache misses
have
occured?

Is there any tool that give the number of pieline stalls that occured ?

Is there any other way to project the performance apart from just
comparing the
execution times?

How much performance gain in terms of percentage one can expect from an
instruction scheduler?

Hoping a detailed early reply.

Thanks.
sincierly,
phani.