Hi Matt, The docker image worked for us. Thanks for providing this image and making it available for use. We were able to run some HIP programs.
Thank you > On Mar 21, 2020, at 5:43 PM, Anirudh <system...@gmail.com> wrote: > > Hi Matt, > > Thanks for the detailed response and link to the docker image. > > I have been looking at the post > https://www.mail-archive.com/gem5-users@gem5.org/msg15909.html > <https://www.mail-archive.com/gem5-users@gem5.org/msg15909.html>, and I used > several hints regarding rocm version, and changes to the library paths in > apu_se.py listed in the post. In an alternate post > https://www.mail-archive.com/gem5-users@gem5.org/msg16275.html > <https://www.mail-archive.com/gem5-users@gem5.org/msg16275.html>, Anthony > recommended some specific commit IDs for ROCr, ROCt, and HCC. I am yet to > install these ROC packages with these specific commit IDs. > > We will try HIP and the docker image you have provided. > > Thanks again. > > >> On Mar 21, 2020, at 4:28 PM, Matt Sinclair <mattdsincl...@gmail.com >> <mailto:mattdsincl...@gmail.com>> wrote: >> >> Hi Anirudh, >> >> This question has been asked several times before. I recommend you take a >> look, e.g., >> here:https://www.mail-archive.com/gem5-users@gem5.org/msg15909.html >> <https://www.mail-archive.com/gem5-users@gem5.org/msg15909.html> and see if >> those suggestions resolve your problems. Usually this error means you >> either have an OS that is not supported or you haven't set up the ROCm stack >> correctly. In the past, I saw this error when running on Ubuntu 14/16, but >> I thought the updated commits on the AMD staging branch had resolved this. >> So, perhaps it's a problem with how you set up ROCm then (or you are using a >> very old commit). >> >> A secondary problem is that the current GPU model does not fully support >> OpenCL, it only supports HIP: http://old.gem5.org/GPU_Models.html >> <http://old.gem5.org/GPU_Models.html> (it appears that the new website does >> appear to have that page; Jason have you already filed a JIRA ticket about >> this?). Note that the linked webpage does need to be updated to explain how >> to hipify applications. The slides from the tutorial on that webpage have >> some instructions about how to hipify that represent a good place to start >> (e.g., slide 76). >> >> In the meantime, we have created a Docker that sets up everything you need >> to run the GPU model: https://github.com/KyleRoarty/gem5_docker/ >> <https://github.com/KyleRoarty/gem5_docker/>. We are also in the process of >> merging this into gem5: >> https://gem5-review.googlesource.com/c/amd/gem5/+/26544 >> <https://gem5-review.googlesource.com/c/amd/gem5/+/26544>. I suggest you >> use one of these as your starting point, as based on the error it's probably >> an issue with how you set up ROCm (as mentioned above). >> >> In terms of running backprop, the simplest thing to do would be to take the >> CUDA version of backprop (e.g., from Rodinia), run hipify on it, remove the >> dGPU copies, compile it with hipcc, and try running that. You can see an >> example of how Kyle did the last two steps for square (similar to you vector >> addition) here: >> https://github.com/KyleRoarty/gem5_docker/blob/master/tests/run_tests.sh >> <https://github.com/KyleRoarty/gem5_docker/blob/master/tests/run_tests.sh> >> >> Thanks, >> Matt >> >> On Sat, Mar 21, 2020 at 2:42 PM Anirudh <system...@gmail.com >> <mailto:system...@gmail.com>> wrote: >> Thanks for the response. >> >> This is the command line argument: build/GCN3_X86/gem5.opt >> configs/example/apu_se.py -c backprop.out -o '-n 16’. >> >> I also tried a simple hand written vector addition program in OpenCL, and I >> was encountering the same issue. >> >>> On Mar 21, 2020, at 12:31 AM, Abhishek Singh >>> <abhishek.singh199...@gmail.com <mailto:abhishek.singh199...@gmail.com>> >>> wrote: >>> >>> Hi, >>> Can you paste your command line? >>> >>> On Fri, Mar 20, 2020 at 8:37 PM Anirudh <system...@gmail.com >>> <mailto:system...@gmail.com>> wrote: >>> Hello, >>> >>> I would like some help running the backprop benchmark on AMD gem5-gpu. The >>> backprop benchmark is written in OpenCL. I am using g++ v 4.8. I have rocm >>> 1.6 installed. >>> >>> I am seeing this error: “ sysconf(NPROCESSORS_CONF) failed: No such file or >>> directory. request to allocate mask for invalid number”. I suspect this has >>> something to do with querying some system state by OpenCL calls. >>> >>> I have been following a relevant post that encountered the same issue for >>> HCC applications >>> (https://gem5-users.gem5.narkive.com/wMel4JdA/gem5-compute-gpu-model >>> <https://gem5-users.gem5.narkive.com/wMel4JdA/gem5-compute-gpu-model>). >>> >>> Any guidance on this issue would be useful. >>> >>> Thanks. >>> _______________________________________________ >>> gem5-users mailing list >>> gem5-users@gem5.org <mailto:gem5-users@gem5.org> >>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >>> <http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users>_______________________________________________ >>> gem5-users mailing list >>> gem5-users@gem5.org <mailto:gem5-users@gem5.org> >>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >>> <http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users> >> _______________________________________________ >> gem5-users mailing list >> gem5-users@gem5.org <mailto:gem5-users@gem5.org> >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >> <http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users>_______________________________________________ >> gem5-users mailing list >> gem5-users@gem5.org <mailto:gem5-users@gem5.org> >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >
_______________________________________________ gem5-users mailing list gem5-users@gem5.org http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
