Howto maximize blocks and grid for Jetson Xavier

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Xavier
1

How do I determine how to maximize these blocks and grid for Jetson Xavier? Thanks in advance!

Example
dimBlock=dim3(32,32);
int yBlocks = eglFrame.height/dimBlock.y+((eglFrame.height%dimBlock.y)==0?0:1);
int xBlocks = eglFrame.width/dimBlock.x+((eglFrame.width%dimBlock.x)==0?0:1);
dimGrid=dim3(xBlocks,yBlocks);

3

Hi,

You can get this information from the deviceQuery sample:

$ ./deviceQuery
./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

Detected 1 CUDA Capable device(s)

Device 0: "Xavier"
  CUDA Driver Version / Runtime Version          11.4 / 11.4
  CUDA Capability Major/Minor version number:    7.2
  ...
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
...

Thanks.

4

Number of thread block size that works for me is 32,32. What else would need to be changed when I change these values? I get this message if I use anything bigger(64,64 for example).
Cuda failure nvsample_cudaprocess.cu:945: invalid configuration argument. But I believe this is a side effect. How does this effect the grid size allowed? Here is my code snippet.

dimBlock=dim3(32,32); // this works fine using 64,64 fails
int yBlocks = eglFrame.height/dimBlock.y+((eglFrame.height%dimBlock.y)==0?0:1);
int xBlocks = eglFrame.width/dimBlock.x+((eglFrame.width%dimBlock.x)==0?0:1);
dimGrid=dim3(xBlocks,yBlocks);

5

Hi,

Based on the deviceQuery, the maximum #threads is 1024.
Block with dim(64, 64) has 64x64=4096 threads which is already over the limit.

Thanks.

6

Hi,

Got it, thanks.

Tom

8

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.