Bug#757961: Does not work with GCC 4.9, default GCC in Jessie
On 21/08/2014 23:02, Tomasz Rybak wrote: Not necessarily. I'll build (and test) PyCUDA on my machine, then build amd64 and i386 packages in pbuilder. As building does not require running any kernels, but only presence of CUDA headers, it'll work without any problems. OK, great. BTW - do you plan to upload 6.0, or 6.5 before Jessie freeze? If so, I'll need to rebuild PyCUDA with new CUDA. I have a mostly working CUDA 6.0 package in my Ubuntu PPA. I am just waiting on someone to increase my PPA size so I can upload another version, hopefully in the next day or so. I think it then be will be ready to test and for Vincent to review and possibly upload to Experimental. It will still be missing a fix for #755513, but it can go through NEW without that. I'd still like to get 5.5 fixed for Ubuntu Utopic and Trusty though. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#757961: Does not work with GCC 4.9, default GCC in Jessie
Just for reference, nvidia-cuda-toolkit 6.0.37 also prevents the use of gcc 4.8 in /usr/include/host_config.h and therefore needs the same patch. In fact, the only difference between /usr/include/host_config.h in 5.5.22 and 6.0.37 is that the version in 6.0.37 now allows ICC 13.1 instead of ICC 12.1. Tomasz: do you need this uploaded before you can upload pycuda 2014.1? -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#757961: Does not work with GCC 4.9, default GCC in Jessie
Dnia 2014-08-21, czw o godzinie 15:13 +0200, Graham Inggs pisze: Just for reference, nvidia-cuda-toolkit 6.0.37 also prevents the use of gcc 4.8 in /usr/include/host_config.h and therefore needs the same patch. In fact, the only difference between /usr/include/host_config.h in 5.5.22 and 6.0.37 is that the version in 6.0.37 now allows ICC 13.1 instead of ICC 12.1. Tomasz: do you need this uploaded before you can upload pycuda 2014.1? Not necessarily. I'll build (and test) PyCUDA on my machine, then build amd64 and i386 packages in pbuilder. As building does not require running any kernels, but only presence of CUDA headers, it'll work without any problems. BTW - do you plan to upload 6.0, or 6.5 before Jessie freeze? If so, I'll need to rebuild PyCUDA with new CUDA. Best regards. -- Tomasz Rybak GPG/PGP key ID: 2AD5 9860 Fingerprint A481 824E 7DD3 9C0E C40A 488E C654 FB33 2AD5 9860 http://member.acm.org/~tomaszrybak signature.asc Description: This is a digitally signed message part
Bug#757961: Does not work with GCC 4.9, default GCC in Jessie
tags 757961 patch thanks I plan to prepare this for NMU in the next couple of days, hopefully along with a fix for #755513. On 19/08/2014 11:27, Vincent Cheng wrote: This might be an issue fixed in a newer upstream release, but I don't think there are any active maintainers looking after nvidia-cuda-toolkit right now. If anyone who's reading this bug report would like to help out by packaging the latest upstream release (6.x or similar?), I'd be glad to consider reviewing and sponsoring your package. Thanks, I might take you up on your offer. I'll reply to bug #751522 if I make any progress. diff -Nru nvidia-cuda-toolkit-5.5.22/debian/rules nvidia-cuda-toolkit-5.5.22/debian/rules --- nvidia-cuda-toolkit-5.5.22/debian/rules 2014-05-25 01:25:00.0 +0200 +++ nvidia-cuda-toolkit-5.5.22/debian/rules 2014-08-14 14:04:23.0 +0200 @@ -96,6 +96,8 @@ chmod -x debian/tmp/usr/nvvm/include/*.h chmod -x debian/tmp/usr/nvvm/libnvvm-samples/build.bat sed -i '/^-vm$$/ d; /^..\/jre\/bin\/java$$/ d' debian/tmp/usr/libnvvp/nvvp.ini debian/tmp/usr/libnsight/nsight.ini + # CUDA 5.5 seems to work fine with GCC 4.9 + sed -i 's/__GNUC_MINOR__ 8/__GNUC_MINOR__ 9/' debian/tmp/usr/include/host_config.h # reduce 'dh_install --list-missing' noise rm -rf debian/tmp/usr/include/thrust rm -rf debian/tmp/usr/jre
Bug#757961: Does not work with GCC 4.9, default GCC in Jessie
On Tue, Aug 12, 2014 at 11:52 AM, Tomasz Rybak bogom...@post.pl wrote: Package: nvidia-cuda-toolkit Version: 5.5.22-4 Severity: important It looks like the CUDA 5.5 does not work with GCC 4.9: $ nvcc tmpTfIlUk.cu In file included from /usr/lib/nvidia-cuda-toolkit/include/cuda_runtime.h:59:0, from command-line:0: /usr/include/host_config.h:82:2: error: #error -- unsupported GNU version! gcc 4.9 and up are not supported! #error -- unsupported GNU version! gcc 4.9 and up are not supported! This might be an issue fixed in a newer upstream release, but I don't think there are any active maintainers looking after nvidia-cuda-toolkit right now. If anyone who's reading this bug report would like to help out by packaging the latest upstream release (6.x or similar?), I'd be glad to consider reviewing and sponsoring your package. Regards, Vincent -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#757961: Does not work with GCC 4.9, default GCC in Jessie
Dnia 2014-08-14, czw o godzinie 12:01 +0200, Graham Inggs pisze: Hi Thomas I tried editing /usr/include/host_config.h as below to bypass the GCC version check. After this change, I was able to build all of the CUDA 5.5 SDK samples without error. Let me know if this works for you too and I will prepare a patch. Thanks. After this change PyCUDA works like a charm - all tests pass, both for Python 2 and Python 3. Best regards. -- Tomasz Rybak GPG/PGP key ID: 2AD5 9860 Fingerprint A481 824E 7DD3 9C0E C40A 488E C654 FB33 2AD5 9860 http://member.acm.org/~tomaszrybak signature.asc Description: This is a digitally signed message part
Bug#757961: Does not work with GCC 4.9, default GCC in Jessie
Hi Thomas I tried editing /usr/include/host_config.h as below to bypass the GCC version check. After this change, I was able to build all of the CUDA 5.5 SDK samples without error. Let me know if this works for you too and I will prepare a patch. Regards Graham --- /usr/include/host_config.h.orig2014-08-14 11:44:17.760255774 +0200 +++ /usr/include/host_config.h2014-08-14 11:34:42.820275680 +0200 @@ -77,7 +77,7 @@ #if defined(__GNUC__) -#if __GNUC__ 4 || (__GNUC__ == 4 __GNUC_MINOR__ 8) +#if __GNUC__ 4 || (__GNUC__ == 4 __GNUC_MINOR__ 9) #error -- unsupported GNU version! gcc 4.9 and up are not supported! -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#757961: Does not work with GCC 4.9, default GCC in Jessie
Package: nvidia-cuda-toolkit Version: 5.5.22-4 Severity: important It looks like the CUDA 5.5 does not work with GCC 4.9: $ nvcc tmpTfIlUk.cu In file included from /usr/lib/nvidia-cuda-toolkit/include/cuda_runtime.h:59:0, from command-line:0: /usr/include/host_config.h:82:2: error: #error -- unsupported GNU version! gcc 4.9 and up are not supported! #error -- unsupported GNU version! gcc 4.9 and up are not supported! -- System Information: Debian Release: jessie/sid APT prefers unstable APT policy: (500, 'unstable'), (500, 'testing'), (1, 'experimental') Architecture: amd64 (x86_64) Foreign Architectures: i386 Kernel: Linux 3.14-2-amd64 (SMP w/3 CPU cores) Locale: LANG=pl_PL.UTF-8, LC_CTYPE=pl_PL.UTF-8 (charmap=UTF-8) Shell: /bin/sh linked to /bin/dash Versions of packages nvidia-cuda-toolkit depends on: ii g++ 4:4.9.0-4 ii gcc 4:4.9.0-4 ii libc62.19-7 ii libgcc1 1:4.9.0-10 ii libnvvm2 5.5.22-4 ii libstdc++6 4.9.0-10 ii nvidia-cuda-dev 5.5.22-4 ii nvidia-profiler 5.5.22-4 ii ocl-icd-opencl-dev [opencl-dev] 2.1.3-4 Versions of packages nvidia-cuda-toolkit recommends: ii nvidia-cuda-doc 5.5.22-4 ii nvidia-cuda-gdb 5.5.22-4 ii nvidia-visual-profiler 5.5.22-4 Versions of packages nvidia-cuda-toolkit suggests: ii libcupti-dev 5.5.22-4 -- no debconf information -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
