CUDA Kernel Launch

 Hello All,

In my previous post related to Introduction to GPU programming using NVIDA CUDA Tool Kit(link in comment section) I have explained about how to write a simple program(Performing addition of two arrays) using CUDA.

In this post let us understand how the CUDA kernel will launch with provided block dimension and grid dimensions parameters.

A GPU will follows a single instruction multiple thread(SIMT) architecture it means that the multiple threads are issued for processing the same instruction.

These threads are organized in to blocks and blocks are organized in to grids.




Let us consider the example of CUDA Hello World Program where we launch the CUDA kernel with total number of threads in a block as 1 and there is 1 such block in a grid.

HelloWorld.cu

//Pre-processor directives
#include <stdio.h>
#include "cuda_runtime.h"
#include "device_launch_parameters.h"//Device code
__global__
void cuda_kernel()
{
    printf("Hello World!");
}//Host code
int main()
{
    cuda_kernel <<< 1, 1 >>> ();  //CUDA kernel launch
    cudaDeviceSynchronize();    
    cudaDeviceReset();  
    return 0;
}

In the above code to launch the CUDA kernel two 1's are initialized between the angular brackets. The First parameter indicates the total number of blocks in a grid and second parameter indicates the total number of threads inside a block. Thus in the above code total number of threads in a block is 1 and there is 1 such block in a grid.



Implicit variables initialized by CUDA runtime: 

threadIdx

1.     It is a dim3 variable and each dimension can be accessed by threadIdx.x, threadIdx.y, threadIdx.z.

2.     Refers to the thread ID with in a block and it starts from 0. So, if number of threads in X dim in a block is 32, then threadIdx.x ranges from 0 to 31 in each block.

blockIdx

1.     It is a dim3 variable and each dimension can be accessed by blockIdx.x, blockIdx.y, blockIdx.z.

2.     Refers to the block ID in a grid and it starts from 0.

blockDim

1.     It is a dim3 variable.

2.     Refers to the maximum number of threads in a block in all the dimension and it starts from 1.

3.     All thread blocks have the same dimension.

gridDim

1.     It is a dim3 variable.

2.     Refers to the maximum number of blocks in a grid in all the dimension and it starts from 1.

 

  

Comments

Post a Comment

Popular posts from this blog

Adding custom splash screen to yocto image

Image
 Hello All, In this article we are going to see how we can add custom splash screen to our yocto image. pre-requisite: you need to be familiar with my previous topics as it contains the necessary steps to setup the environment  and  build the image. link to my previous topic: https://learnyoctowithme.blogspot.com/2022/04/constructing-embedded-linux-image-for.html https://www.blogger.com/blog/posts/8695339245532547417?tab=rj Adding custom splash screen to yocto image:  lets check what is the default splash screen set to image. step 1: open local.conf file $ vi poky/build_x86/conf/local.conf  and append the following line to local.conf file as IMAGE_FEATURES += "splash" and build the image using below command $ bitbake core-image-minimal  after successful compilation of build , boot the image using below command $ runqemu you can observe the splash screen while booting of the image Now we will modify the custom splash screen with our image. step 2: Pspla...
Read more

Adding libraries (pciutils,usbutils and util-linux) to our image using yocto project

Image
 Hello All, In this article we are going to see how we can add libraries like pciutils,usbutils and util-linux to our image using yocto project. Before proceeding to this, you need to  be familiar with my previous articles , as it contains the necessary steps to create the build environment for qemux86-64 and steps to create our own layer. pre-requisite: https://www.blogger.com/blog/posts/8695339245532547417?tab=rj Adding libraries (pciutils,usbutils and util-linux) to our image using yocto project: step 1: switch to poky folder $ cd poky step 2: now switch to the directory build_x86/conf/ $ cd build_x86/conf step 3: open local.cong file $ vi local.conf and append the following lines as  IMAGE_INSTALL_append += "util-linux pciutils usbutils" and save it step 4: now compile the image as $ bitbake core-image-minimal  step 5: now boot the image using the command $ runqemu nographic and login as root user step 6: On terminal, enter the command as $ lsblk step 7: on ...
Read more

Adding custom python script(to findout system info) to rootfs using yocto project

Image
 Hello All, In this article, we are going to see how we can add our custom python script( python script to find out the system information) to our root filesystem using yocto project Before proceeding to this, you need to familiar with my previous articles, as it gives the overview about yocto build system and setup environment pre-requisite: https://www.blogger.com/blog/posts/8695339245532547417?tab=rj Adding custom python script (to findout system information) to rootfs using yocto: step 1: switch to meta-mylayer/recipes-kernel directory $ cd poky/meta-mylayer/recipes-kernel/ step 2 : create a folder called system_info $ mkdir system_info step 3: Inside system_info folder, create another folder called files $ mkdir files step 4: Inside files folder , create a file called sytem_info.py and edit as following $ touch system_info.py $ vi system_info.py and save and exit it step 5: now switch back to system_info folder $ cd .. step 6: create a bitbake recipe to add the system_inf...
Read more