[SDL] SDL Semaphore implementation broken on Windows?

[John Bartholomew]

Sam Lantinga wrote:
>> However, assuming this is the cause of my problems, there is a very 
>> simple fix:
>> Windows provides InterlockedIncrement() and InterlockedDecrement() 
>> functions to perform
>> increments and decrements which are guaranteed to be atomic.
>> So the fix is in thread/win32/SDL_syssem.c:
>> replace occurrences of --sem->count with InterlockedDecrement(&sem->count);
>> and replace occurrences of ++sem->count with 
>> InterlockedIncrement(&sem->count);
>>     
>
> I'm pretty sure the 32-bit ++ and -- operators are atomic on x86 architectures.
> Are you sure this fixes the problem?
I wouldn't bet my life on it, but I'm 99% sure.  Certainly the problem 
hasn't recurred at all since I made that change, and although it isn't 
100% repeatable, it was happening regularly enough (say, 80 to 90% of 
the times I attempted to reproduce) that I believe it would have 
surfaced again if this hadn't fixed it.
Of course, it's always possible that the change just managed to push the 
timings off and make it far, far rarer rather than actually eliminating 
the problem.

As for ++ and -- being atomic, at the assembly code level the 'INC' and 
'DEC' instructions are only atomic operations when the 'LOCK' prefix is 
used (which is presumably how InterlockedIncrement/InterlockedDecrement 
are implemented under the hood).  I don't know whether C gives any 
guarantees about their atomicity though.  I don't think any compiler 
would emit a lock prefix for them under all circumstances, but possibly 
it does if the variable being modified is marked as volatile?

Regards,

John B