Keocc
January 30, 2018, 2:02am
1
Hello, everyone. Recently I have learned Stereo Semi Global Matching by cuda. This code is from the github https://github.com/fixstars/libSGM . It is normal when I run this code on PC (GPU is GTX1050). But there is an error on Nvidia TX2.:
CUDA ERROR #4 (using FilterMedian kernel #2): unspecified launch failure
stereo_test: /home/nvidia/newlibSGM/libSGM/src/median_filter.cu:31: void sgm::details::median_filter(const uint16_t*, uint16_t*, void*, int, int): Assertionstatus == 0' failed. Aborted (core dumped)`
Even if I delete the FilterMedian kernel, there is “unspecified launch failure ” for cudaMemcpy.
And the CUDA_NVCC_FLAGS is set to compute_62 on TX2.
set(CUDA_NVCC_FLAGS
${CUDA_NVCC_FLAGS};
-O3 -lineinfo
-gencode arch=compute_62,code=sm_62
)
What parameters should I set? What configuration parameters should I modify?
Hi,
Please modify CMakeLists to sm_62 for TX2 GPU architecture:
https://github.com/fixstars/libSGM/blob/master/CMakeLists.txt#L3
Here is an example for file CMakeLists.txt on Jetson:
cmake_minimum_required(VERSION 2.8)
project(jetson-inference)
# submodule warning
message(" ")
message("Note: this project uses git submodules in the source tree.")
message(" if you haven't already, run the following command from")
message(" the project's root directory:")
message(" ")
message(" git submodule update --init")
message("\n")
if( NOT EXISTS "${PROJECT_SOURCE_DIR}/utils/.git" )
message("Note: required git submodules have not been detected.")
message(" first, please run the following command from the")
message(" the project's root directory to clone them:")
message(" ")
message(" git submodule update --init")
message(" ")
This file has been truncated. show original
Thanks.
Keocc
January 30, 2018, 11:57am
3
Thank you for your reply. I changed another development board TX2 to test, and the code has run successfully