accelerator: introduce compare_ipc_handles fnct #12352
Conversation
edgargabriel
changed the title
| int *pid_2 = (int *)&handle_2[pos]; | ||
|
|
||
| /* Note: using pid_1 != pid_2 since we need to return 0 if identical */ | ||
| return (memcmp(handle_1, handle_2, 32) && (*pid_1 != *pid_2)); |
There was a problem hiding this comment.
I'm not sure about this - should it be memcmp(handle_1, handle_2, 32) || (*pid_1 != *pid_2) instead?
There was a problem hiding this comment.
I think you might be right, I hate these brain twister. I will write down the decision matrix to check
| * and the process ID for comparison. | ||
| * We definitily need to exclude the offset component in the comparison. | ||
| */ | ||
| int pos = 32 + 2*sizeof(size_t); |
There was a problem hiding this comment.
nit static const int POS = (int) 32 + 2 * sizeof(size_t); - maybe also need a const for 32.
edgargabriel
changed the title
edgargabriel
force-pushed
the
topic/accelerator-ipc-handle-compare
branch
from
5dc1328 to
9251cd5
Compare
|
@wenduwan thank you for your review, can you please have another look? I think I incorporated both of your comments. |
| int *pid_1 = (int *)&handle_1[pos]; | ||
| int *pid_2 = (int *)&handle_2[pos]; | ||
|
|
||
| return (memcmp(handle_1, handle_2, 32) + (*pid_1 == *pid_2 ? 0 : 1)); |
There was a problem hiding this comment.
Ah this might not work because -1 + 1 = 0
There was a problem hiding this comment.
:-( you are unfortunately right again. Thanks for catching it!
There was a problem hiding this comment.
A more readable way can be
...
if (*pid_1 != *pid_2) return *pid_1 > *pid_2 ? 1 : -1;
return memcmp(handle_1, handle_2, 32);
...
There was a problem hiding this comment.
that is actually a good idea, I made just one minor change, thanks!
comparing ipc handles might not be always just a memcmp of the two handles. Introduce an abstraction for this functionality. Use the memcmp function that was used so far in the cuda and ze component, but use only certain parts of the ipc handle in rocm. Signed-off-by: Edgar Gabriel <[email protected]>
edgargabriel
force-pushed
the
topic/accelerator-ipc-handle-compare
branch
from
9251cd5 to
4002fbc
Compare
2 participants
comparing ipc handles might not be always just a memcmp of the two handles. Introduce an abstraction for this functionality. Use the memcmp function that was used so far in the cuda and ze component, but use only certain parts of the ipc handle in rocm.