Complete syntax of CUDA Kernels

Posted by Unknown at 08:45 | 2 comments

GPU Kernel with Streams, Run time Kernel Launch

In this article we’ll let you know the complete syntax of CUDA Kernels.

We all are love to learn and always curious about know everything in detail. I was very disappointed when I was not able to find the complete syntax of CUDA Kernels. So, I though let me give it a day to search everywhere, after the havey search, I found the syntax of CUDA Kernel and today I am presenting It you reader.

The CUDA Kernel consist in <<< >>> brackets four things.

First argument is known as “Grid Size”, followed by “Block Size”, followed by size of “Shared Memory” and end with “Stream argument”.

Here is the complete syntax;

Kernel_Name<<< GridSize,  BlockSize,  SMEMSize,   Stream  >>> (arguments,....);

Grid Size

We all know what is Grid size, in case you don’t know read further.

Grid size is defined by the number of blocks in a grid. In previous version of CUDA architecture (from Compute capability 1.x to 2.x) the grid can only be organized in two dimension (X and Y direction ). But in the current version (from Compute capability 3.x onwards) the grid can be organized in three dimension ( X , Y and Z all ).

Block Size

The blocks organized in terms of threads. Threads is the smallest unit in Parallel programming so in CUDA.

Shared Memory (SMEMSize)

This is for the size of shared memory which is to be use in CUDA Kernel for shared variable space. This is use bec. Of dynamic shared memory size in CUDA Kernels.

Streams

A stream is a sequence of operations that are performed in order on the device.

Streams allows independent concurrent in-order queues of execution. Stream tell on which device, kernel will execute.

Operations in different streams can be interleaved and overlapped, which can be used to hide data transfers between host and device.
Learn more about CUDA Streams follow this article

Got Questions? 

Feel free to ask me any question because I'd be happy to walk you through step by step! 

 
References and External links

CUDA C Programming Guide
CUDA; Nvidia
Wikipedia