Maximum total parameter size

Accelerated Computing CUDA CUDA Programming and Performance
#1

I recall that I have read something in this forum about a maximum total parameter size when executing a kernel, but I can’t seem to find the post in question. The CUDA programming guide states (p. 23, section 4.2.3) that the specified amount of dynamically allocated shared memory can not be larger than

total available shared memory - the amount of statically allocated shared memory - the amount of shared memory used for function arguments - the amount of shared memory used for execution configuration

So, does this mean that the memory used for kernel parameters is actually shared memory? And does this in turn mean that the maximum size of parameters is only restricted by the shared memory size or are there also other limiting factors?

Thank you in advance,
Veronica

#2

Yes, kernel arguments are stored in shared memory. No, you can’t use all 16k for arguments: the limit is 256 bytes. See this recent post in the programming and development forum: http://forums.nvidia.com/index.php?showtopic=68296

#3

Thank you MisterAnderson42,
that was in fact the post I was looking for! Thanks to the post, I also found the statement in the programming guide, section 4.2.1.4. Once again, thank you!