July 2020 - Ecp-aurora-sdk-discuss - mailman3.cels.anl.gov

New A21 SDK installed in JLSE
by Rosales-fernandez, Carlos 31 Jul '20

31 Jul '20

Hi all, We are pleased to announce that a new A21 SDK has been deployed in JLSE. This version 2020.07.30.001 includes multiple improvements and bug fixes. For additional details please refer to the latest release notes in the Wiki: https://wiki.jlse.anl.gov/display/inteldga/DGA+SDK+Release+Notes Modules will enable the newer SDK by default, but the older versions remain available for regression testing. Important notes =============== When using dynamic linking for threaded MKL calls, please include the tbb location in teh link line. An old tbb.so library is being picked up by the linker from /lib64 adn this is a workaround to avoid issues, so substitute your typicall "-ltbb" in the MKL build by (prepending location to LD_LIBRARY_PATH will not work): $ -L$TBBROOT/lib/intel64/gcc4.8 -ltbb Public Documentation ==================== * ALCF oneAPI overview - https://www.alcf.anl.gov/support-center/aurora/oneapi-overview * Intel GPGPU Software Information - https://dgpu-docs.intel.com/ * Intel oneAPI Beta Documentation - https://software.intel.com/content/www/us/en/develop/tools/oneapi/documenta… Please let us know of any issues with the release or the environment settings. Regards, Carlos

1 0

New A21 SDK coming to JLSE today
by Rosales-fernandez, Carlos 31 Jul '20

31 Jul '20

Hi all, A new A21 SDK will be installed today at 3:00 PM CST. Running jobs will not be affected. Default SDK modulefiles will be updated at that time. A notice will be sent after the install is complete. Regards, Carlos

1 0

Re: [Ecp-aurora-sdk-discuss] clang/dpcpp compilation issues; RH 4.8.5 libs
by Bertoni, Colleen 29 Jul '20

29 Jul '20

So the first errors (from using clang++ directly) are likely because clang++ by default (at least as built on JLSE) uses GCC’s STL. The default GCC on JLSE is 4.8.5 unfortunately, so this is what it picks up. You should be able to get around this by specifying the path to a more recent gcc with `clang++ --gcc-toolchain=/path/`. In fact, dpcpp is a wrapper around clang++ that, among other things, that uses `—gcc-toolchain` to pull in a more recent GCC. (You can see in CNDA location /soft/restricted/CNDA/modulefiles/dpcpp/2020.06.15.001_wrapper/dpcpp ). Since dpcpp is actually a wrapper as mentioned above, it might cause something to break inside of cmake. Can you check that dpcpp on its own will work for you (module purge; module load dpcpp and try fresh on the command line)? You might need to change stuff inside of cmake to get it to work, I don't know off the top of my head what would need to be changed in your cmake scripts, though (maybe someone else on the list has more experience). If it’s possible to share, could you send pointers to how you build? (Also: note that this email was sent to both the ECP-aurora-SDK list and intel-dga list. These two lists have different NDA levels--not everyone on ECP-aurora-SDK is on intel-dga, and we should keep the conversation at the CNDA level, or remove one of the lists.) - Colleen ________________________________ From: Ecp-aurora-sdk-discuss <ecp-aurora-sdk-discuss-bounces(a)lists.jlse.anl.gov> on behalf of Vincent R. Pascuzzi via Ecp-aurora-sdk-discuss <ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov> Sent: Wednesday, July 29, 2020 12:13 PM To: ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov <ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov>; Intel-dga-discuss <intel-dga-discuss(a)lists.jlse.anl.gov> Subject: [Ecp-aurora-sdk-discuss] clang/dpcpp compilation issues; RH 4.8.5 libs Hi, I have loaded the dpcpp and mkl modules (default/2020.06.15): $ module list Currently Loaded Modulefiles: 1) gmp/6.1.2-gcc-9.3.0-gsxwgho 12) binutils/2.34-gcc-9.3.0-rnwhrdg 2) mpfr/3.1.6-gcc-9.3.0-pcot5yh 13) isl/0.20-gcc-9.3.0-fey2k52 3) mpc/1.1.0-gcc-9.3.0-nu26nou 14) gcc/9.3.0-gcc-9.3.0-qfmcwfb 4) zlib/1.2.11-gcc-9.3.0-uubmvor 15) spack/linux-rhel7-x86_64 5) ncurses/6.2-gcc-9.3.0-xvsnay4 16) .public_intel_igc/release/1.0.3977-1 6) bzip2/1.0.8-gcc-9.3.0-vatuhza 17) public_intel_level_zero/release/20.20.16837-1 7) libiconv/1.16-gcc-9.3.0-zqp7gsc 18) dpcpp/2020.06.15.001(default) 8) tar/1.32-gcc-9.3.0-g3rvopa 19) mkl/2020.06.15.001(default) 9) xz/5.2.5-gcc-9.3.0-6rgt4w7 20) openssl/1.1.1f-gcc-9.3.0-l545wyq 10) libxml2/2.9.9-gcc-9.3.0-2yk5s47 21) cmake/3.17.0-gcc-9.3.0-5dgh2gv 11) gettext/0.20.1-gcc-9.3.0-4uprshc and end up with compilations errors when building with clang/clang++. A bunch of errors of the type: ... /soft/restricted/intel_dga/sdk/2020.6.15.1/inteloneapi/compiler/latest/linux/bin/../include/sycl/CL/sycl/intel/functional.hpp:28:17: error: too few template arguments for class template 'less' return std::less<>()(std::forward<const T>(lhs), std::forward<const U>(rhs)) ^ /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../include/c++/4.8.5/bits/stl_function.h:231:12: note: template is declared here struct less : public binary_function<_Tp, _Tp, bool> … where the distribution’s gcc 4.8.5 libraries are being referenced. If I instead try dpcpp instead of clang++: ... -- Check for working CXX compiler: /soft/restricted/intel_dga/sdk/2020.6.15.1/inteloneapi/compiler/latest/linux/bin/dpcpp - broken CMake Error at /soft/packaging/spack-builds/linux-rhel7-x86_64/gcc-9.3.0/cmake-3.17.0-5dgh2gvtfsxrvwqkw3m7jq2jhhk6yhph/share/cmake-3.17/Modules/CMakeTestCXXCompiler.cmake:53 (message): The C++ compiler "/soft/restricted/intel_dga/sdk/2020.6.15.1/inteloneapi/compiler/latest/linux/bin/dpcpp" is not able to compile a simple test program. … Linking CXX executable cmTC_44726 /soft/packaging/spack-builds/linux-rhel7-x86_64/gcc-9.3.0/cmake-3.17.0-5dgh2gvtfsxrvwqkw3m7jq2jhhk6yhph/bin/cmake -E cmake_link_script CMakeFiles/cmTC_44726.dir/link.txt --verbose=1 /soft/restricted/intel_dga/sdk/2020.6.15.1/inteloneapi/compiler/latest/linux/bin/dpcpp CMakeFiles/cmTC_44726.dir/testCXXCompiler.cxx.o -o cmTC_44726 /soft/packaging/spack-builds/linux-rhel7-x86_64/gcc-9.3.0/binutils-2.34-rnwhrdgiqluqiypg5palnxdxviv3mynt/bin/ld: /soft/restricted/CNDA/sdk/2020.6.15.1/inteloneapi/compiler/2021.1-beta07/linux/bin/../lib/libsycl.so: undefined reference to `VTT for std::__cxx11::basic_stringstream<char, std::char_traits<char>, std::allocator<char> >@GLIBCXX_3.4.21' /soft/packaging/spack-builds/linux-rhel7-x86_64/gcc-9.3.0/binutils-2.34-rnwhrdgiqluqiypg5palnxdxviv3mynt/bin/ld: /soft/restricted/CNDA/sdk/2020.6.15.1/inteloneapi/compiler/2021.1-beta07/linux/bin/../lib/libsycl.so: undefined reference to `std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::_M_assign(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)(a)GLIBCXX_3.4.xn--21-o2t … Any advice would be greatly appreciated! Vince -- Ecp-aurora-sdk-discuss mailing list Ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov https://lists.jlse.anl.gov/mailman/listinfo/ecp-aurora-sdk-discuss

1 0

clang/dpcpp compilation issues; RH 4.8.5 libs
by Vincent R. Pascuzzi 29 Jul '20

29 Jul '20

Hi, I have loaded the dpcpp and mkl modules (default/2020.06.15): $ module list Currently Loaded Modulefiles: 1) gmp/6.1.2-gcc-9.3.0-gsxwgho 12) binutils/2.34-gcc-9.3.0-rnwhrdg 2) mpfr/3.1.6-gcc-9.3.0-pcot5yh 13) isl/0.20-gcc-9.3.0-fey2k52 3) mpc/1.1.0-gcc-9.3.0-nu26nou 14) gcc/9.3.0-gcc-9.3.0-qfmcwfb 4) zlib/1.2.11-gcc-9.3.0-uubmvor 15) spack/linux-rhel7-x86_64 5) ncurses/6.2-gcc-9.3.0-xvsnay4 16) .public_intel_igc/release/1.0.3977-1 6) bzip2/1.0.8-gcc-9.3.0-vatuhza 17) public_intel_level_zero/release/20.20.16837-1 7) libiconv/1.16-gcc-9.3.0-zqp7gsc 18) dpcpp/2020.06.15.001(default) 8) tar/1.32-gcc-9.3.0-g3rvopa 19) mkl/2020.06.15.001(default) 9) xz/5.2.5-gcc-9.3.0-6rgt4w7 20) openssl/1.1.1f-gcc-9.3.0-l545wyq 10) libxml2/2.9.9-gcc-9.3.0-2yk5s47 21) cmake/3.17.0-gcc-9.3.0-5dgh2gv 11) gettext/0.20.1-gcc-9.3.0-4uprshc and end up with compilations errors when building with clang/clang++. A bunch of errors of the type: ... /soft/restricted/intel_dga/sdk/2020.6.15.1/inteloneapi/compiler/latest/linux/bin/../include/sycl/CL/sycl/intel/functional.hpp:28:17: error: too few template arguments for class template 'less' return std::less<>()(std::forward<const T>(lhs), std::forward<const U>(rhs)) ^ /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../include/c++/4.8.5/bits/stl_function.h:231:12: note: template is declared here struct less : public binary_function<_Tp, _Tp, bool> … where the distribution’s gcc 4.8.5 libraries are being referenced. If I instead try dpcpp instead of clang++: ... -- Check for working CXX compiler: /soft/restricted/intel_dga/sdk/2020.6.15.1/inteloneapi/compiler/latest/linux/bin/dpcpp - broken CMake Error at /soft/packaging/spack-builds/linux-rhel7-x86_64/gcc-9.3.0/cmake-3.17.0-5dgh2gvtfsxrvwqkw3m7jq2jhhk6yhph/share/cmake-3.17/Modules/CMakeTestCXXCompiler.cmake:53 (message): The C++ compiler "/soft/restricted/intel_dga/sdk/2020.6.15.1/inteloneapi/compiler/latest/linux/bin/dpcpp" is not able to compile a simple test program. … Linking CXX executable cmTC_44726 /soft/packaging/spack-builds/linux-rhel7-x86_64/gcc-9.3.0/cmake-3.17.0-5dgh2gvtfsxrvwqkw3m7jq2jhhk6yhph/bin/cmake -E cmake_link_script CMakeFiles/cmTC_44726.dir/link.txt --verbose=1 /soft/restricted/intel_dga/sdk/2020.6.15.1/inteloneapi/compiler/latest/linux/bin/dpcpp CMakeFiles/cmTC_44726.dir/testCXXCompiler.cxx.o -o cmTC_44726 /soft/packaging/spack-builds/linux-rhel7-x86_64/gcc-9.3.0/binutils-2.34-rnwhrdgiqluqiypg5palnxdxviv3mynt/bin/ld: /soft/restricted/CNDA/sdk/2020.6.15.1/inteloneapi/compiler/2021.1-beta07/linux/bin/../lib/libsycl.so: undefined reference to `VTT for std::__cxx11::basic_stringstream<char, std::char_traits<char>, std::allocator<char> >@GLIBCXX_3.4.21' /soft/packaging/spack-builds/linux-rhel7-x86_64/gcc-9.3.0/binutils-2.34-rnwhrdgiqluqiypg5palnxdxviv3mynt/bin/ld: /soft/restricted/CNDA/sdk/2020.6.15.1/inteloneapi/compiler/2021.1-beta07/linux/bin/../lib/libsycl.so: undefined reference to `std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::_M_assign(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)(a)GLIBCXX_3.4.xn--21-o2t … Any advice would be greatly appreciated! Vince

1 0

Re: [Ecp-aurora-sdk-discuss] Running SYCL/DPC++ Code on NVIDIA GPUs
by Bertoni, Colleen 16 Jul '20

16 Jul '20

Yes, you'll need a SYCL compiler with CUDA support. If you're interested, it looks like there are build instructions here https://intel.github.io/llvm-docs/GetStartedGuide.html#build-dpc-toolchain-… and https://github.com/codeplaysoftware/sycl-for-cuda/blob/cuda/sycl/doc/GetSta… Thanks, Colleen ________________________________ From: Ecp-aurora-sdk-discuss <ecp-aurora-sdk-discuss-bounces(a)lists.jlse.anl.gov> on behalf of David Poole via Ecp-aurora-sdk-discuss <ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov> Sent: Thursday, July 16, 2020 3:02 PM To: Hammond, Jeff R <jeff.r.hammond(a)intel.com> Cc: ecp-aurora-sdk-discuss(a)jlse.anl.gov <ecp-aurora-sdk-discuss(a)jlse.anl.gov> Subject: Re: [Ecp-aurora-sdk-discuss] Running SYCL/DPC++ Code on NVIDIA GPUs Hi all, Thank you very much for your assistance on this! It seems that the module "public_llvm_intel_sycl/20200519-cuda" is required for building DPC++ code on NVIDIA GPUs (as it recognizes the nvptx64-nvidia-cuda-sycldevice SYCL target), which I was not aware of before sending this email. Sincerely, -David Poole ----------------------------------------------- "It is the mark of an educated mind to be able to entertain a thought without accepting it." - Aristotle David Poole Iowa State University 201 Spedding Hall davpoole(a)iastate.edu "I would rather have questions that can't be answered than answers that can't be questioned." - Richard Feynman "For me, I am driven by two main philosophies: know more today about the world than I knew yesterday and lessen the suffering of others. You'd be surprised how far that gets you. " - Neil deGrasse Tyson ----------------------------------------------- On Thu, Jul 16, 2020 at 2:30 PM Hammond, Jeff R <jeff.r.hammond(a)intel.com> wrote: > > If you use SYCL 1.2.1, it should be widely portable. If you use SYCL 2020 USM, you will need to use DPC++ for now. I am optimistic that both ComputeCpp and hipSYCL will support USM in the near future, but you should contact the developers of those implementations to get their official answer. > > Intel DPC++ supports USM on NVIDIA, although there is one known bug that shouldn't affect most usages (I think it is related to events, which many codes don't need to use). > > CodePlay ComputeCpp supports SYCL 1.2.1 today and they are working towards SYCL 2020 support, but the latest release I downloaded lacked a complete implementation of USM (I tested it with simple code and I've verified the lack of function is expected with Rod Burns at CodePlay). > > If you use the DPCT, I recall there is an option to target the port to use USM or not (i.e. buffers). I do not remember what the default is, but suspect that it changed from buffers to USM at some point. > > Jeff > > On 7/16/20, 12:22 PM, "Ecp-aurora-sdk-discuss on behalf of David Poole via Ecp-aurora-sdk-discuss" <ecp-aurora-sdk-discuss-bounces(a)lists.jlse.anl.gov on behalf of ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov> wrote: > > Hi all, > > My team and I have been working on porting our codebase from CUDA to > DPC++, via construction of a mini-app. This process has been going > rather well when it comes to running on the Iris Gen9 GPU. However, I > have also had an interest in getting the ported DPC++ code to run on > NVIDIA GPUs. > > What I am wondering is, is it possible to run SYCL/DPC++ code on > NVIDIA GPUs at the moment? Based on my experiments and the contents of > https://software.intel.com/content/www/us/en/develop/articles/intel-oneapi-…, > my suspicion is that it is not yet; but I may be wrong about that. And > if it is currently possible to run SYCL/DPC++ code on NVIDIA GPUs, how > would one go about that? > > Thank you very much in advance! > > Sincerely, > -David Poole > > ----------------------------------------------- > "It is the mark of an educated mind to > be able to entertain a thought without > accepting it." - Aristotle > > David Poole > Iowa State University > 201 Spedding Hall > davpoole(a)iastate.edu > > "I would rather have questions that can't > be answered than answers that can't be > questioned." - Richard Feynman > > "For me, I am driven by two main > philosophies: know more today about > the world than I knew yesterday and > lessen the suffering of others. You'd > be surprised how far that gets > you. " - Neil deGrasse Tyson > ----------------------------------------------- > -- > Ecp-aurora-sdk-discuss mailing list > Ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov > https://lists.jlse.anl.gov/mailman/listinfo/ecp-aurora-sdk-discuss > -- Ecp-aurora-sdk-discuss mailing list Ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov https://lists.jlse.anl.gov/mailman/listinfo/ecp-aurora-sdk-discuss

1 0

Re: [Ecp-aurora-sdk-discuss] Running SYCL/DPC++ Code on NVIDIA GPUs
by David Poole 16 Jul '20

16 Jul '20

Hi all, Thank you very much for your assistance on this! It seems that the module "public_llvm_intel_sycl/20200519-cuda" is required for building DPC++ code on NVIDIA GPUs (as it recognizes the nvptx64-nvidia-cuda-sycldevice SYCL target), which I was not aware of before sending this email. Sincerely, -David Poole ----------------------------------------------- "It is the mark of an educated mind to be able to entertain a thought without accepting it." - Aristotle David Poole Iowa State University 201 Spedding Hall davpoole(a)iastate.edu "I would rather have questions that can't be answered than answers that can't be questioned." - Richard Feynman "For me, I am driven by two main philosophies: know more today about the world than I knew yesterday and lessen the suffering of others. You'd be surprised how far that gets you. " - Neil deGrasse Tyson ----------------------------------------------- On Thu, Jul 16, 2020 at 2:30 PM Hammond, Jeff R <jeff.r.hammond(a)intel.com> wrote: > > If you use SYCL 1.2.1, it should be widely portable. If you use SYCL 2020 USM, you will need to use DPC++ for now. I am optimistic that both ComputeCpp and hipSYCL will support USM in the near future, but you should contact the developers of those implementations to get their official answer. > > Intel DPC++ supports USM on NVIDIA, although there is one known bug that shouldn't affect most usages (I think it is related to events, which many codes don't need to use). > > CodePlay ComputeCpp supports SYCL 1.2.1 today and they are working towards SYCL 2020 support, but the latest release I downloaded lacked a complete implementation of USM (I tested it with simple code and I've verified the lack of function is expected with Rod Burns at CodePlay). > > If you use the DPCT, I recall there is an option to target the port to use USM or not (i.e. buffers). I do not remember what the default is, but suspect that it changed from buffers to USM at some point. > > Jeff > > On 7/16/20, 12:22 PM, "Ecp-aurora-sdk-discuss on behalf of David Poole via Ecp-aurora-sdk-discuss" <ecp-aurora-sdk-discuss-bounces(a)lists.jlse.anl.gov on behalf of ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov> wrote: > > Hi all, > > My team and I have been working on porting our codebase from CUDA to > DPC++, via construction of a mini-app. This process has been going > rather well when it comes to running on the Iris Gen9 GPU. However, I > have also had an interest in getting the ported DPC++ code to run on > NVIDIA GPUs. > > What I am wondering is, is it possible to run SYCL/DPC++ code on > NVIDIA GPUs at the moment? Based on my experiments and the contents of > https://software.intel.com/content/www/us/en/develop/articles/intel-oneapi-…, > my suspicion is that it is not yet; but I may be wrong about that. And > if it is currently possible to run SYCL/DPC++ code on NVIDIA GPUs, how > would one go about that? > > Thank you very much in advance! > > Sincerely, > -David Poole > > ----------------------------------------------- > "It is the mark of an educated mind to > be able to entertain a thought without > accepting it." - Aristotle > > David Poole > Iowa State University > 201 Spedding Hall > davpoole(a)iastate.edu > > "I would rather have questions that can't > be answered than answers that can't be > questioned." - Richard Feynman > > "For me, I am driven by two main > philosophies: know more today about > the world than I knew yesterday and > lessen the suffering of others. You'd > be surprised how far that gets > you. " - Neil deGrasse Tyson > ----------------------------------------------- > -- > Ecp-aurora-sdk-discuss mailing list > Ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov > https://lists.jlse.anl.gov/mailman/listinfo/ecp-aurora-sdk-discuss >

1 0

Re: [Ecp-aurora-sdk-discuss] Running SYCL/DPC++ Code on NVIDIA GPUs
by Hammond, Jeff R 16 Jul '20

16 Jul '20

If you use SYCL 1.2.1, it should be widely portable. If you use SYCL 2020 USM, you will need to use DPC++ for now. I am optimistic that both ComputeCpp and hipSYCL will support USM in the near future, but you should contact the developers of those implementations to get their official answer. Intel DPC++ supports USM on NVIDIA, although there is one known bug that shouldn't affect most usages (I think it is related to events, which many codes don't need to use). CodePlay ComputeCpp supports SYCL 1.2.1 today and they are working towards SYCL 2020 support, but the latest release I downloaded lacked a complete implementation of USM (I tested it with simple code and I've verified the lack of function is expected with Rod Burns at CodePlay). If you use the DPCT, I recall there is an option to target the port to use USM or not (i.e. buffers). I do not remember what the default is, but suspect that it changed from buffers to USM at some point. Jeff On 7/16/20, 12:22 PM, "Ecp-aurora-sdk-discuss on behalf of David Poole via Ecp-aurora-sdk-discuss" <ecp-aurora-sdk-discuss-bounces(a)lists.jlse.anl.gov on behalf of ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov> wrote: Hi all, My team and I have been working on porting our codebase from CUDA to DPC++, via construction of a mini-app. This process has been going rather well when it comes to running on the Iris Gen9 GPU. However, I have also had an interest in getting the ported DPC++ code to run on NVIDIA GPUs. What I am wondering is, is it possible to run SYCL/DPC++ code on NVIDIA GPUs at the moment? Based on my experiments and the contents of https://software.intel.com/content/www/us/en/develop/articles/intel-oneapi-…, my suspicion is that it is not yet; but I may be wrong about that. And if it is currently possible to run SYCL/DPC++ code on NVIDIA GPUs, how would one go about that? Thank you very much in advance! Sincerely, -David Poole ----------------------------------------------- "It is the mark of an educated mind to be able to entertain a thought without accepting it." - Aristotle David Poole Iowa State University 201 Spedding Hall davpoole(a)iastate.edu "I would rather have questions that can't be answered than answers that can't be questioned." - Richard Feynman "For me, I am driven by two main philosophies: know more today about the world than I knew yesterday and lessen the suffering of others. You'd be surprised how far that gets you. " - Neil deGrasse Tyson ----------------------------------------------- -- Ecp-aurora-sdk-discuss mailing list Ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov https://lists.jlse.anl.gov/mailman/listinfo/ecp-aurora-sdk-discuss

1 0

Re: [Ecp-aurora-sdk-discuss] Running SYCL/DPC++ Code on NVIDIA GPUs
by Bertoni, Colleen 16 Jul '20

16 Jul '20

Hi David, It should be possible on JLSE on the V100 nodes (gpu_v100_smx2). After logging into a node with a V100, could you try: ``` module load public_llvm_intel_sycl/20200519-cuda ``` and then compile with ``` clang++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda-sycldevice -Xsycl-target-backend=nvptx64-nvidia-cuda-sycldevice '--cuda-gpu-arch=sm_70' test.cpp ``` and let us know if it works ok? Thanks, Colleen ________________________________ From: Ecp-aurora-sdk-discuss <ecp-aurora-sdk-discuss-bounces(a)lists.jlse.anl.gov> on behalf of David Poole via Ecp-aurora-sdk-discuss <ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov> Sent: Thursday, July 16, 2020 2:20 PM To: ecp-aurora-sdk-discuss(a)jlse.anl.gov <ecp-aurora-sdk-discuss(a)jlse.anl.gov> Subject: [Ecp-aurora-sdk-discuss] Running SYCL/DPC++ Code on NVIDIA GPUs Hi all, My team and I have been working on porting our codebase from CUDA to DPC++, via construction of a mini-app. This process has been going rather well when it comes to running on the Iris Gen9 GPU. However, I have also had an interest in getting the ported DPC++ code to run on NVIDIA GPUs. What I am wondering is, is it possible to run SYCL/DPC++ code on NVIDIA GPUs at the moment? Based on my experiments and the contents of https://software.intel.com/content/www/us/en/develop/articles/intel-oneapi-…, my suspicion is that it is not yet; but I may be wrong about that. And if it is currently possible to run SYCL/DPC++ code on NVIDIA GPUs, how would one go about that? Thank you very much in advance! Sincerely, -David Poole ----------------------------------------------- "It is the mark of an educated mind to be able to entertain a thought without accepting it." - Aristotle David Poole Iowa State University 201 Spedding Hall davpoole(a)iastate.edu "I would rather have questions that can't be answered than answers that can't be questioned." - Richard Feynman "For me, I am driven by two main philosophies: know more today about the world than I knew yesterday and lessen the suffering of others. You'd be surprised how far that gets you. " - Neil deGrasse Tyson ----------------------------------------------- -- Ecp-aurora-sdk-discuss mailing list Ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov https://lists.jlse.anl.gov/mailman/listinfo/ecp-aurora-sdk-discuss

1 0

Re: [Ecp-aurora-sdk-discuss] Running SYCL/DPC++ Code on NVIDIA GPUs
by Vincent R. Pascuzzi 16 Jul '20

16 Jul '20

Hi, David. It’s possible to target PTX using: -fsycl-targets=nvptx64-nvidia-cuda-sycldevice and in my experience, things work rather well; it’s also a good way to measure the performance of *real* data transfers, i.e. not using Intel UMA. Note that certain things don’t yet work, e.g. experimental::printf() and mkl::rng, so you’ll need to workaround such limitations. Vince > On Jul 16, 2020, at 12:20 PM, David Poole via Ecp-aurora-sdk-discuss <ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov> wrote: > > Hi all, > > My team and I have been working on porting our codebase from CUDA to > DPC++, via construction of a mini-app. This process has been going > rather well when it comes to running on the Iris Gen9 GPU. However, I > have also had an interest in getting the ported DPC++ code to run on > NVIDIA GPUs. > > What I am wondering is, is it possible to run SYCL/DPC++ code on > NVIDIA GPUs at the moment? Based on my experiments and the contents of > https://software.intel.com/content/www/us/en/develop/articles/intel-oneapi-…, > my suspicion is that it is not yet; but I may be wrong about that. And > if it is currently possible to run SYCL/DPC++ code on NVIDIA GPUs, how > would one go about that? > > Thank you very much in advance! > > Sincerely, > -David Poole > > ----------------------------------------------- > "It is the mark of an educated mind to > be able to entertain a thought without > accepting it." - Aristotle > > David Poole > Iowa State University > 201 Spedding Hall > davpoole(a)iastate.edu > > "I would rather have questions that can't > be answered than answers that can't be > questioned." - Richard Feynman > > "For me, I am driven by two main > philosophies: know more today about > the world than I knew yesterday and > lessen the suffering of others. You'd > be surprised how far that gets > you. " - Neil deGrasse Tyson > ----------------------------------------------- > -- > Ecp-aurora-sdk-discuss mailing list > Ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov > https://lists.jlse.anl.gov/mailman/listinfo/ecp-aurora-sdk-discuss

1 0

Running SYCL/DPC++ Code on NVIDIA GPUs
by David Poole 16 Jul '20

16 Jul '20

Hi all, My team and I have been working on porting our codebase from CUDA to DPC++, via construction of a mini-app. This process has been going rather well when it comes to running on the Iris Gen9 GPU. However, I have also had an interest in getting the ported DPC++ code to run on NVIDIA GPUs. What I am wondering is, is it possible to run SYCL/DPC++ code on NVIDIA GPUs at the moment? Based on my experiments and the contents of https://software.intel.com/content/www/us/en/develop/articles/intel-oneapi-…, my suspicion is that it is not yet; but I may be wrong about that. And if it is currently possible to run SYCL/DPC++ code on NVIDIA GPUs, how would one go about that? Thank you very much in advance! Sincerely, -David Poole ----------------------------------------------- "It is the mark of an educated mind to be able to entertain a thought without accepting it." - Aristotle David Poole Iowa State University 201 Spedding Hall davpoole(a)iastate.edu "I would rather have questions that can't be answered than answers that can't be questioned." - Richard Feynman "For me, I am driven by two main philosophies: know more today about the world than I knew yesterday and lessen the suffering of others. You'd be surprised how far that gets you. " - Neil deGrasse Tyson -----------------------------------------------

1 0