I installed Cuda on my arch. To verify the success of the installation I tried to compile one of the samples (deviceQuery). I get the following error: [root@archpaz paz]# cd /home/paz/cuda/ [root@archpaz cuda]# make /opt/cuda/bin/nvcc -ccbin g++ -I../../common/inc -m64 -gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=sm_35 -gencode arch=compute_37,code=sm_37 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_75,code=compute_75 -o deviceQuery.o -c deviceQuery.cpp deviceQuery.cpp:17:10: fatal error: helper_cuda.h: File o directory non esistente 17 | #include <helper_cuda.h> | ^~~~~~~~~~~~~~~ compilation terminated. make: *** [Makefile:288: deviceQuery.o] Error 1 Does anyone know what a mistake it is and how to make cuda work?
GG says executing the following lines does help: export PATH=/usr/local/cuda-10.1/bin${PATH:+:${PATH}} export LD_LIBRARY_PATH=\ /usr/local/cuda-10.1/lib64/${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}} export MANPATH=/usr/local/cuda/doc/man/${MANPATH:+:${MANPATH}} And in your case it can not find helper_cuda.h a bunch of include files are in a common area and they can not be moved. When gg tested this just now on Fedora, He first did "cd deviceQuery" before running "make". If you get further or the above does not help, email more information please. gg/Phyllis [root@archpaz paz]# cd /home/paz/cuda/
[root@archpaz cuda]# make /opt/cuda/bin/nvcc -ccbin g++ -I../../common/inc -m64 -gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=sm_35 -gencode arch=compute_37,code=sm_37 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_75,code=compute_75 -o deviceQuery.o -c deviceQuery.cpp deviceQuery.cpp:17:10: fatal error: helper_cuda.h: File o directory non esistente 17 | #include <helper_cuda.h> | ^~~~~~~~~~~~~~~ compilation terminated. make: *** [Makefile:288: deviceQuery.o] Error 1
GG says executing the following lines does help:
export PATH=/usr/local/cuda-10.1/bin${PATH:+:${PATH}} export LD_LIBRARY_PATH=\ /usr/local/cuda-10.1/lib64/${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}} export MANPATH=/usr/local/cuda/doc/man/${MANPATH:+:${MANPATH}}
He keeps giving me the exact same error. However, after the installation I had made a new login to automatically configure the PATH. - I brought the DeviceQuery folder from /opt to /home as recommended by Arch wiki. Here I gave the make. - After installation it gave as an optional "Java-runtime8" dependency that I have not installed. - Maybe my old Nvidia gtx 460M doesn't support Cuda very well (but the info would say yes; shader model 3.0 is required and the graphics card supports 5.1 and Cuda 2.1). - I do not know what other information to give (Journalctl?).
Andrea: He keeps giving me the exact same error.
However, after the installation I had made a new login to automatically configure the PATH. - I brought the DeviceQuery folder from /opt to /home as recommended by Arch wiki. Here I gave the make.
That creates the problem -- that is, you moved something. So you can either make a copy to put it where it originally was. OR if you key in the following first, you will get partial success that is usable (ignore the release error) and you will now be able to run deviceQuery. This worked for gg on a test system just now. export EXTRA_NVCCFLAGS=-I/usr/local/cuda/samples/common/inc/
You are right: Compiling in /opt works well. In any case, being able to compile "deviceQuery" is not important, only that I get the same error if I try to compile Cuda in CinGG. Do I need to put in any particular instructions? (I used --with-cuda). I have to put the string on: -I/opt/cuda/include (vedi wiki più sotto) in the compiler? How do I do it? Do I have to compile CinGG as root? With the command "printenv" I see the voice: "CUDA_PATH=/opt/cuda" I report what Arch's wiki says: "The cuda package installs all components in the directory /opt/cuda. For compiling CUDA code, add /opt/cuda/include to your include path in the compiler instructions. For example this can be accomplished by adding -I/opt/cuda/include to the compiler flags/options. To use nvcc, a gcc wrapper provided by NVIDIA, just add /opt/cuda/bin to your path. To find whether the installation was successful and if cuda is up and running, you can compile the samples installed on /opt/cuda/samples (you can simply run make inside the directory, altough is a good practice to copy the /opt/cuda/samples directory to your home directory before compiling) and running the compiled examples. A nice way to check the installation is to run one of the examples, called deviceQuery."
export EXTRA_NVCCFLAGS=-I/usr/local/cuda/samples/common/inc/
Maybe Arch installs Cuda in /opt and Fedora in /usr/local/...?
Andrea, somehow I missed this email. Sorry. Whenever you do not get a timely response on an email "Question", please email again. Although some questions I do not answer because I am totally clueless and even then I usually just say that. In any case, being able to compile "deviceQuery" is not important,
only that I get the same error if I try to compile Cuda in CinGG. Do I need to put in any particular instructions? (I used --with-cuda). I have to put the string on: -I/opt/cuda/include (vedi wiki più sotto) in the compiler? How do I do it? Do I have to compile CinGG as root?
With the command "printenv" I see the voice: "CUDA_PATH=/opt/cuda"
I report what Arch's wiki says:
"The cuda package installs all components in the directory /opt/cuda. For compiling CUDA code, add /opt/cuda/include to your include path in ... Maybe Arch installs Cuda in /opt and Fedora in /usr/local/...?
Your statement just above this is correct. GG will have to fix this for Arch. Attached is a corrected file that he will check in later (probably today) but you can just write over the one you have in Cinelerra's toplevel directory. AND as you stated you will have to use "setenv CUDA_PATH=/opt/cuda" on the command line before compiling with-cuda. We also tested on Leap the current way and that worked too. Arch is a little bit different animal!
I'm sorry, I'm very incompetent on code and environment variables. I can't compile Cuda. I have made these attempts: 1) Copied attachment.obj into ~/cinelerra5/cinelerra-5.1/ and given the command "env CUDA_PATH=/opt/cuda" (setenv does not exist in Arch). The result of "./configure ..." is: ... checking for /usr/local/cuda/include/cuda.h... no checking cuda sdk... no configure: error: required for CUDA support. (note that somewhere it still looks for /usr/local/cuda instead of /opt/cuda) 2) I deleted "plugin_cuda" and renamed "attchment.obj" to "plugin_cuda". But the resulting error is exactly the same. 3) I tried to edit the file "plugin_cuda" replacing all instances "/usr/local/cuda" with "/opt/cuda", but the result is always the same error.
Andrea, we really appreciate your continuing effort to compile Cinelerra on Arch with Cuda. GG has checked in another change to get you further along. I hope you do not get discouraged because it would be hard for us to attempt to check this on Arch here. On Wed, Jun 26, 2019 at 2:19 AM Andrea paz <[email protected]> wrote:
I'm sorry, I'm very incompetent on code and environment variables. I can't compile Cuda. I have made these attempts:
Thanks, with the latest release I have compiled CinGG with Cuda without any problems. I've loaded a three-media project: 1- h264 .mp4 2- h264 .mp4 (made with the same smartphone within seconds of each other) 3- h264 .3gp (same codec as 1 and 2 but different container, I used another smartphone) 1 and 3 give error and are not displayed in timeline. 2 don't give mistake and is displayed normally. Strange, since it is exactly equal to 1. Error is: Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. FFStream::decode: avcodec_send_packet failed. file:/home/paz/video_editing/cinelerra/cc-15d/volto.3gp err: Invalid data found when processing input FFStream::decode: failed Repeated several times (loading attempts?)
Andrea, we can duplicate the problem here. GG traced this down to "nvdec" on the decode/read-in side. He is looking into it further still. Just out of curiosity did Mandel and N_Body Cuda plugins show up in Resources Video Effects? On Wed, Jun 26, 2019 at 12:55 PM Andrea paz <[email protected]> wrote:
Thanks, with the latest release I have compiled CinGG with Cuda without any problems.
I've loaded a three-media project: 1- h264 .mp4 2- h264 .mp4 (made with the same smartphone within seconds of each other) 3- h264 .3gp (same codec as 1 and 2 but different container, I used another smartphone) 1 and 3 give error and are not displayed in timeline. 2 don't give mistake and is displayed normally. Strange, since it is exactly equal to 1. Error is:
Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. FFStream::decode: avcodec_send_packet failed. file:/home/paz/video_editing/cinelerra/cc-15d/volto.3gp err: Invalid data found when processing input FFStream::decode: failed
Repeated several times (loading attempts?) -- Cin mailing list [email protected] https://lists.cinelerra-gg.org/mailman/listinfo/cin
Andrea, Another fix was checked in for Cinelerra into GIT which was a patch to ffmpeg. We will probably contact ffmpeg to see if they can put it in the code base. gg/Phyllis On Wed, Jun 26, 2019 at 12:55 PM Andrea paz <[email protected]> wrote:
Thanks, with the latest release I have compiled CinGG with Cuda without any problems.
I've loaded a three-media project: 1- h264 .mp4 2- h264 .mp4 (made with the same smartphone within seconds of each other) 3- h264 .3gp (same codec as 1 and 2 but different container, I used another smartphone) 1 and 3 give error and are not displayed in timeline. 2 don't give mistake and is displayed normally. Strange, since it is exactly equal to 1. Error is:
Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. Failed to get HW surface format. FFStream::decode: avcodec_send_packet failed. file:/home/paz/video_editing/cinelerra/cc-15d/volto.3gp err: Invalid data found when processing input FFStream::decode: failed
Repeated several times (loading attempts?) -- Cin mailing list [email protected] https://lists.cinelerra-gg.org/mailman/listinfo/cin
The two cuda filters (Mandelbrot and N_Body) appear and can be applied to edits. But when I move the insertion point over the edit to see the result of the effect I have a crash (no dump). I attach the respective errors on the terminal for both plugins. NOTE: Loading the three assets without root I only see the edit 2 (as I said in the previous mail). Starting from root (with the .bcast remaining from previous installations, since I had not eliminated it as I do for each new compilation) I see all three edits with no errors.
NOTE: Loading the three assets without root I only see the edit 2 (as I said in the previous mail). Starting from root (with the .bcast remaining from previous installations, since I had not eliminated it as I do for each new compilation) I see all three edits with no errors.
My usual dumbness: starting CinGG as root worked without errors because I had not set "Cuda" in Preferences. I'm sorry. However, if we want to investigate these issues perhaps it is better to open a ticket in MantisBT.
Yes, please open a ticket. That way those people who do not like email because of other spam can see it more easily. Thanks, Phyllis On Thu, Jun 27, 2019 at 2:21 AM Andrea paz <[email protected]> wrote:
NOTE: Loading the three assets without root I only see the edit 2 (as I said in the previous mail). Starting from root (with the .bcast remaining from previous installations, since I had not eliminated it as I do for each new compilation) I see all three edits with no errors.
My usual dumbness: starting CinGG as root worked without errors because I had not set "Cuda" in Preferences. I'm sorry. However, if we want to investigate these issues perhaps it is better to open a ticket in MantisBT. -- Cin mailing list [email protected] https://lists.cinelerra-gg.org/mailman/listinfo/cin
Hi, I hope this is correct, I do not have enough experience with cuda to really be a good source for this kind of diagnostic. It appears that cudaErrorInsufficientDriver indicates *CUDA 10* is not compatible with driver version on your machine. There are more than one possible corrective actions. 1) upgrade the driver if your board supports newer nvidia builds. 2) downgrade the cuda development package to a version that works for your board. 3) buy a really cool new board, and then spend the next few weeks regretting the breakage of opening your machine. gg On Thu, Jun 27, 2019 at 1:29 AM Andrea paz <[email protected]> wrote:
The two cuda filters (Mandelbrot and N_Body) appear and can be applied to edits. But when I move the insertion point over the edit to see the result of the effect I have a crash (no dump). I attach the respective errors on the terminal for both plugins.
NOTE: Loading the three assets without root I only see the edit 2 (as I said in the previous mail). Starting from root (with the .bcast remaining from previous installations, since I had not eliminated it as I do for each new compilation) I see all three edits with no errors. -- Cin mailing list [email protected] https://lists.cinelerra-gg.org/mailman/listinfo/cin
3) buy a really cool new board, and then spend the next few weeks regretting the breakage of opening your machine.
Thank you, it's good news to know that it depends on my old GPU. I'm inclined towards option 3 as in a couple of months I'll be moving from notebook to desktop.
Andea GG also wrote: "1) upgrade the driver if your board supports newer nvidia builds". When I wanted to explore the Cudas and nvenc issue with my Nvidia GTX-750ti card I received the message that the Nvidia driver version (384.130) provided by my Linux Mint 18.3 was too old. Even version 19.1 of Linux Mint is not recent enough (390.116). I hope that the release of Linux Mint 19.2 (maybe next month), will bring the necessary version of the Nvidia driver (390.25 or newer). If so, I could consider replacing my Mint and reinstalling and reconfiguring all my used software. I think it is possible to update only the Nvidia driver on a version of Mint that does not offer it but I prefer to wait for the version of Mint that will include it. Perhaps it is the same under your distribution. Pierre On 19-06-28 02 h 21, Andrea paz wrote:
3) buy a really cool new board, and then spend the next few weeks regretting the breakage of opening your machine.
Thank you, it's good news to know that it depends on my old GPU. I'm inclined towards option 3 as in a couple of months I'll be moving from notebook to desktop.
I think it is possible to update only the Nvidia driver on a version of Mint that does not offer it but I prefer to wait for the version of Mint that will include it. Pierre
I think the 390.xx (legacy) drivers we both use, no longer add new features, only maintenance patches. So point 1 of GG is out of the game. The point 2 remains, that is to downgrade cuda SDK to a version (which?) that supports our drivers. Curiosity: I have compiled CinGG without "--with-cuda", which doesn't really work for me, but it has been installed the same. Do I have to use "--without-cuda" to not compile it?
When I had tried to render in h264_nvenc.mp4, each time CinGG refused to proceed And I received this message from the terminal: [h264_nvenc @ 0x7f28601e7a00] Driver does not support the required nvenc API version. Required: 9.0 Found: 8.0 [h264_nvenc @ 0x7f28601e7a00] The minimum required Nvidia driver for nvenc is 390.25 or newer 390.25 therefore seems to offer new functionalities. Pierre On 19-06-28 09 h 40, Andrea paz wrote:
I think it is possible to update only the Nvidia driver on a version of Mint that does not offer it but I prefer to wait for the version of Mint that will include it. Pierre
I think the 390.xx (legacy) drivers we both use, no longer add new features, only maintenance patches. So point 1 of GG is out of the game. The point 2 remains, that is to downgrade cuda SDK to a version (which?) that supports our drivers.
Curiosity: I have compiled CinGG without "--with-cuda", which doesn't really work for me, but it has been installed the same. Do I have to use "--without-cuda" to not compile it?
participants (4)
-
Andrea paz -
Good Guy -
Phyllis Smith -
Pierre autourduglobe