different result from the same code

[mikebell]

[howardhh]

Dear mikebell,

Really thanks for your help~
I have tried to print the arrays, but everything is right. And I think there is no other errors except the "mailbox", because I have another version of this program, which only use the local memory, is right.

I suspect that the mistake happens when I update the value of pos and vel, since the initial values are right, but the integration part is wrong. Does that means I can't change the value of variables in the mailbox once I e_write it into the shared_memory?

[sebraa]

The Epiphany cores can change all of shared memory when they feel like it; it is not related to any e_read or e_write calls from the host.

I do not see obvious mistakes either, but:
- try to use an optimization level of -Os or -O1
- declare your mailbox as volatile
- in the mailbox typedef, align it to at least 8

If your code works fine in local memory, but not in shared memory, you might see race conditions due to radically different timing. The Epiphany code might do RMW cycles to access shared memory, and if other cores do the same simultaneously, you will get half-written garbage there. Also, the compiler optimization might remove accesses completely, since you did not mark shared memory as volatile. Reducing the optimization level might help you there.

[cmcconnell]

[howardhh]

Dear all and Colin,

Thank you very much! It's the flag's mistake.
Now the problem is solved by using another kind of flag like the dotproduct.

And thank you all for giving me so much help~I have optimized the code by your suggestion.