cuMemGetInfo() and size_t limitation

1.1k views Asked by At

I would like to know my free and total memory on my GPU device thanks to the function cuMemGetInfo()

// ----- Before any variable initialization -----
size_t free;
size_t total;
CUresult result=cuMemGetInfo(&free,&total);

I'm getting the result :

Free memory : 4095 MB
Total memory : 4095 MB

I'm working on a Tesla C2070 with 6GB of memory on a 64bit Windows 7. However, my application is running in 32 bit. My code should give me something like :

Free memory : 5376 MB
Total memory : 5376 MB // values given by the deviceQuery.exe example of CUDA

I 4095*1024*1024 = 4293918720 is about 2^32 (after rounding). indeed, size_t s a pointer to a unsigned int (on 4 Bytes).

So here's my question. Is it possible to change the definition of size_t in order to point to a unsigned long for example?

thanks

1

There are 1 answers

0
user703016 On BEST ANSWER

size_t is usually 32 bits when compiling for 32 bits targets. If you want to get a 64 bits size_t, simply set your compiler to target 64 bits platforms.

If you're using the Visual Studio build chain, here is the documentation on the topic. If you're using the Qt build chain, I don't know how it's done, but it's certainly feasible somehow.

You say in the comments:

I can't run a 64bit application because of the compatibility between Qt, CUDA and Visual Studio. Is there any way to get around that?

To the best of my knowledge, it's not possible. If you build for 32 bits, then you will link against the 32 bits version of cuMemGetInfo, which works with 32 bits size_t. There's no getting around that.

However, if memory serves, I've worked in the past with 64 bits applications built with VS2012, CUDA 6 and Qt 4.8, so it should be possible to set your tool chain to work with 64 bits.