On Sun, Apr 3, 2011 at 12:06 PM, sunil shah <[email protected]>wrote:
> Hi, > > I am trying to generate load/store address trace ( virtual address). Thus > will able to capture the program data access behavior independent of > architecture. Only want those load's /store's which are committed. > > The way I am trying to generate the load/store address trace by reading > load store queue, lsq->virtaddr in ReorderBurfferEntry::commit() function, > just before returning commit result ok. > > If you are capturing these addresses right before 'return' statement then they will be invalid as 'lsq' is reset at line 2285 in ooopipe.cpp file. If you capture these addresses before the lsq entry is reset then you should get the correct trace. - Avadh > Problem: > The generated address trace contains multiple entry for the same address > and there is no pattern in the repetition. Some time its for 3 times, at > other times its 4 times and so. This happens for load as well as store. > Running a simple test code, with array of 100,000 > int, initializing it and then accessing every 100th element. > > Is it possible by any chance that, since I am running this test code in > given Qemu image and since Qemu is a virtual machine. So there is some > internal code translation and blocking by virtual machine of Qemu. Because > of blocking different addresses are mapped to same block of virtual machine > and thus I am getting multiple load /store virtual address to be same. > > > Another issue, it has lots of load and store at virtual address 0x0, is it > for constant value zero?. If so, then why there are store's at 0x0. > > Thank you very much in anticipation. And if any one else had already > generated load/store trace can kindly share the knowledge. > > Thanks & Regards > Sunil Shah > > > _______________________________________________ > http://www.marss86.org > Marss86-Devel mailing list > [email protected] > https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel > >
_______________________________________________ http://www.marss86.org Marss86-Devel mailing list [email protected] https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
