cudaMallo3D - possibly incorrectly documented in Programming Guide 2.1

[garfungiloops]

Dear All,

The reference to the “cudaMalloc3D” function is possibly incorrect in the Programming Guide 2.1.

On page 35 there is one example how to use that function.

After the lines:

The following code sample allocates a width×height×depth 3D array of

floating-point values and shows how to loop over the array elements in device code

there is an example code.

However I think that the following line in the example is incorrect:

cudaExtent extent = make_cudaExtent(64, 64, 64);

In case of floating point array it must be:

cudaExtent extent = make_cudaExtent(64*sizeof(float), 64, 64);

Otherwise, how it would be possible to allocate array of short (for ex.) values?

The issue was also pointed on the forum here:

http://forums.nvidia.com/lofiversion/index.php?t77910.html

Please notice, that in case of “cudaMalloc3DArray” function the extent would not take a number of bytes into account,

which is very confusing!

Also the example for the “cudaMemcpy3D” function is highly desired for the Programming Guide.

Just in case, my additional info:

CUDA 2.1

Nvidia Quadro FX 570M (HP compaq 8510w)

Win XP SP3

Drv. ver. 6.14.11.8585

MSVS 2005 SP2