C# Programming
02-18-2010, 02:31 AM
Hi everyone.
I'm translating some code from C++ to C# and I've run into a big problem due to the fact that C# is managed and I cannot get the address of the objects I'm handling. (I'm a complete C# newbie so correct me if I'm wrong; I hope I am).
Here's the thing: I'm translating a system that underlies applications and makes sure accesses to shared data are done safely (handles concurrent accesses, etc). For that, whenever a position of the shared memory is accessed there are some measures to be taken, and locks are involved. Memory is segmented so that each 4 words share a lock (too fine a granularity creates high overhead; too coarse, and it greatly reduces parallelism) -- and this is the problem for me. The address of the accessed element is used to map to the lock, but I cannot use addresses in C#, since the garbage collector may relocate objects, etc. I could pin them, but I doubt that would do any good performance-wise.
Instead of addresses I've considered using hash codes (with RuntimeHelpers.GetHashCode), but then I have no way (that I can think of) to control how many memory elements map to the same lock. When using addresses you just shift the address 4 positions to the right, and then you know that you have 4 words mapping to the same lock (assuming addressing by byte and a 32-bit processor), but if you just take a hash code and shift it 4 bits you may end up with very irregular mappings, since hash codes don't need to be consecutive, etc. You may end up with a lock that is mapped by 10 words, and another by 1 or none... This is important for both perfomance and parallelism.
Please tell me that I'm wrong. http://www.barakasoft.com/script/Forums/Images/smiley_smile.gif
Is there a way to do something similar in C#?
Thanks.
I'm translating some code from C++ to C# and I've run into a big problem due to the fact that C# is managed and I cannot get the address of the objects I'm handling. (I'm a complete C# newbie so correct me if I'm wrong; I hope I am).
Here's the thing: I'm translating a system that underlies applications and makes sure accesses to shared data are done safely (handles concurrent accesses, etc). For that, whenever a position of the shared memory is accessed there are some measures to be taken, and locks are involved. Memory is segmented so that each 4 words share a lock (too fine a granularity creates high overhead; too coarse, and it greatly reduces parallelism) -- and this is the problem for me. The address of the accessed element is used to map to the lock, but I cannot use addresses in C#, since the garbage collector may relocate objects, etc. I could pin them, but I doubt that would do any good performance-wise.
Instead of addresses I've considered using hash codes (with RuntimeHelpers.GetHashCode), but then I have no way (that I can think of) to control how many memory elements map to the same lock. When using addresses you just shift the address 4 positions to the right, and then you know that you have 4 words mapping to the same lock (assuming addressing by byte and a 32-bit processor), but if you just take a hash code and shift it 4 bits you may end up with very irregular mappings, since hash codes don't need to be consecutive, etc. You may end up with a lock that is mapped by 10 words, and another by 1 or none... This is important for both perfomance and parallelism.
Please tell me that I'm wrong. http://www.barakasoft.com/script/Forums/Images/smiley_smile.gif
Is there a way to do something similar in C#?
Thanks.