Docker headless no AVX compile but Illegal instruction #7010
Replies: 3 comments 1 reply
-
Try building with What is your HW platform / processor? Here are the SIMD instructions that we build - it should work on your system if any one is present. (SSE2 should be sufficient).
|
Beta Was this translation helpful? Give feedback.
-
I'm running this inside a docker, my local machine is Apple M3 Pro CPU (arm). Would be nice to have a runtime detection if the instruction set is present to use the right vectoriel instructions. So running the docker on x64 machine would still have the speed boost but no need for another image for non x64 native runner of a docker file. |
Beta Was this translation helpful? Give feedback.
-
I end up doing one docker for the open3d part and another docker for colmap (colmap is not arm friendly at this time). This avoid trying to run the open3d on rosetta with arm cpu under MacOS. This solve my problem with my docker images for now. But would be nice to have the avx instructions being only used when they are available and use standard (slower) instructions when not. This could be detected at module load easily with: |
Beta Was this translation helpful? Give feedback.
-
Hi,
Since I was getting illegal instruction when installing open3d from pip:
Inside the docker image if I try to use the open3d I get illegal instruction:
I realized I was not having AVX cpu in my docker env where it run. So I am trying to run open3d with visualization.rendering.OffscreenRenderer in a headless, non AVX CPU environnement. I'm trying to compile this way:
This is inefficient Docker code, just to debug ans test until I get it right
But I cannot get away from the illegale instruction when importing:
Anyway to known what is the problem? Any other things that must be turn off?
Beta Was this translation helpful? Give feedback.
All reactions