Patch to support big buffers

[Philippe SWARTVAGHER]

Hello,

The attached patch fxt_big_buffer_callback.patch does two things:

* it fixes buffer sizes bigger than 2 GB (size was stored in 32 bits 
variables -> changed to 64 bits variables).

* it adds a parameter to the fut_setup function to define a callback 
function to call when FUT is flushing buffer because it is full, and not 
because it's the end of the tracing. This allows to be aware that buffer 
size is too small and maybe it can impact performance of the traced 
application.

The patch starpu_fxt_big_buffer.patch is an example of integration of 
the callback in StarPU (also to test if my FUT patch works nicely !). I 
will wait for a release of the patched FUT before committing it properly 
in StarPU.

I hope the patch format is correct...


Moreover, two questions:

* is it possible to add a FUT routine to force the flush, even if the 
buffer isn't full and tracing isn't finished ? I don't know if calling 
directly fut_flush() is a proper way of doing it. (use case: flush 
buffers between each iteration of Chameleon benchmark ;) )

* I noticed the write syscall doesn't support writing more than 2 GB at 
one time (fixed in my patch). But I asked myself if there is a way to 
check if a trace is complete/not corrupted (ie it doesn't contain only 
the first 2 GB !) ? I tried some stuff with events FUT_STOP_FLUSH_CODE, 
but these events are registered only when flushing occurs during the 
recording, not at the end. Are there events which clearly define the 
beginning and the end of a trace ? Or an event counter stored in the trace ?

--
Philippe SWARTVAGHER

Doctorant
Équipe TADaaM, Inria Bordeaux Sud-Ouest

fxt_big_buffer_callback.patch
Description: Text Data

starpu_fxt_big_buffer.patch
Description: Text Data