|
From: Ali J. <a.j...@gm...> - 2008-01-31 09:11:18
|
Hi Julian
> [quick check: you are subscribed to valgrind-developers, yes?
> some messages about thread checking go only there]
Yes since few weeks i am.
> What I can do is to help to provide an evaluation. I can:
>
> * commit the 64-bit SVal patch, so you don't both have to
> include it in your own work
I have done some minor changes in your 64 bit schema as I needed in ShM2
lset, tset and TSegments.
> * fix the signed/unsigned Word problem that is now in hg_wordfm.c
Could be helpful
>
> * try out your patched versions on OOo / Firefox, and compare/
> summarise results.
>
> Is that useful?
This could be very useful to have an overview and compare the MSMs. I am
doing a review on my changes pls use the new patch that I will send you
after my revision.
I will run Konstatntin unittest with "MSMUnika" and will send you the
results.
> One comment about your MSM. It is often possible for memory
> to change from Virgin ("New" is a better name, imo) directly
> to Excl.R. Especially if the application is buggy (reading
> uninitialised memory) or due to compiler tricks, where the
> compiler loads a word from memory, part of which is uninitialised,
> and then does not use the uninitialised part. So you should add
> an edge for this transition to your Figure 1.
In the new version I changed and got a better results. thanks
Ali
|
|
From: Ali J. <a.j...@gm...> - 2008-01-31 09:23:29
|
Hi Konstantin > And a quick question: does this machine handle barriers? I should with somemore examples . I justtest with some and it is working. > My I ask you to run all tests from racecheck_unittest (take the fresh version from <http://data-race-test.googlecode.com/svn/trunk/unittest/> http://data-race-test.googlecode.com/svn/trunk/unittest/) and send us the results? Also, if you have other tests that show difference between your machine and others, please shared those with us. Will you send me pls the file " thread_wrappers_pthread.h" so that I can run the test unit. Also there are some barriers example as I see. >> It is often possible for memory to change from Virgin ("New" is a better name, imo) directly to Excl.R. Especially if the application is buggy > Another case is when a program has it's own memory management routines that first initialize memory with zeros and then call VG_USERREQ__HG_CLEAN_MEMORY. This way the memory will be "New" but a read from it is quite legal. yea i have changed and i got better results. Tahnk you Ali |