December 2017 - Anl-intel-lrz-knr - mailman3.cels.anl.gov

[Anl-intel-csa] Canceled: ANL/Intel CSA Workload Characterization
by Petrini, Fabrizio 28 Dec '17

28 Dec '17

Placeholder for weekly ANL/Intel CSA Exascale workload characterization meeting ......................................................................................................................................... --> Join Skype Meeting<https://meet.intel.com/fabrizio.petrini/7K7FM32F> Trouble Joining? Try Skype Web App<https://meet.intel.com/fabrizio.petrini/7K7FM32F?sl=1> Join by phone +1(916)356-2663 (or your local bridge access #) Choose bridge 5.<tel:+1(916)356-2663%20(or%20your%20local%20bridge%20access%20#)%20Choose%20bridge%205.> (Global) English (United States) Find a local number<https://dial.intel.com> Conference ID: 791574018 Forgot your dial-in PIN?<https://dial.intel.com> |Help<https://o15.officeredir.microsoft.com/r/rlidLync15?clid=1033&p1=5&p2=2009> [!OC([1033])!] .........................................................................................................................................

1 0

[Anl-intel-csa] CSA SDK Release Notes for 2017/12/19 version
by Glossop, Kent 21 Dec '17

21 Dec '17

CSA SDK Release Notes for 2017/12/19 version The default release has been changed to 2017-12-19-release. Relevant changes since the 2017-12-04-release: Documentation - There is a brand new "Getting Started" guide, available at: $CSA_ROOT/doc/CSA-Getting-Started-Guide.pdf. This describes the flow of how to get started working with CSA directives in an application, and has some description of strategy. We would appreciate feedback on the usefulness of this document, and how to improve the initial experience of trying to modify applications to use the CSA. - A CSA-Tools.pdf document is now in the doc. area. This covers a number of topics about the simulator, including the call interface, statistics, etc. It also has information about assembler directives and attributes. - The CSA-Programming.pdf document and CSA-Tool-Chain.pdf presentation have been updated. (CSA-Architecture.pdf is still pending.) Compiler / Runtime - The SPMDization intrinsic now allows either a "cyclic" or "blocking" approach for assignment of iterations. To support this, there is an incompatible change in the __builtin_csa_spmdization directive that adds a second argument to specify which strategy to use: __builtin_csa_spmdization(integer, "blocked"|"cyclic"); The second parameter designates the approach to use: o Cyclic: the loop iterations are divided into chunks of size one, and the chunks are assigned to the workers in a round-robin fashion. (This was the only approach before.) o Blocked: the loop iterations are divided into contiguous chunks of size #iterations/#workers to be executed by each worker. In Fortran, the intrinsic was also updated to accommodate the two approaches via: call builtin_csa_spmdization(integer, "blocked|"cyclic") Also, fixed bugs related to the use of O3 with SPMDization - A new __builtin_csa_pipeline_loop(N) builtin has been introduced. It is only valid to use this builtin on a loop which has its immediate parent marked as a parallel loop. The bulitin may be useful to users who wish to experiment with ILPL. The argument "N" is an integer literal which specifies the maximum number of outer loop iterations which can share the pipeline. If set to "0", then ILPL will automatically determine a depth. (Currently this is just 32, or the value set by "-mllvm -csa-ilpl-tokens=N".) Please note that the ILPL transform is still considered to be experimental and correctness still depends on the simulator's unrealistic assumption of 128 entry LICs. A new "-mllvm -csa-ilpl-selection" option now controls when ILPL performs transformations: -csa-ilpl-selection - CSA specific: choose how pipelineable inner loops are discovered =automatic - automatically choose inner loops to pipeline =manual - pipeline according to __builtin_csa_pipellineable_loop =both - use both specified and auto-discovered pipelining candidates =disabled - disable inner loop pipelining The default is "manual". The previous automatic behavior can be restored by setting this to "both" or "automatic". This option should generally be used to control ILPL instead of "-mllvm -csa-ilpl". Finally, note that the IR pass can emit some remarks to stderr if requested: -Rpass{,-{missed,analysis}}=csa-ilpl - A new memory ordering pass has been added that should produce better performance results than the previous pass in many cases. In particular, code with non-trivial control flow that uses SPMDization should perform significantly better with this new pass. It has been enabled by default, but if you encounter issues with it you can fall back to an earlier pass using the -csa-order-memops-type flag (-mllvm -csa-order-memops-type=independent for the previous pass or -mllvm -csa-order-memops-type=wavefront for the one before that). o One very particular case that this new pass does not (yet!) handle well is individual stores inside of a loop that do not alias any other memory reference in that loop. The previous pass is able to recognize that those don't need to be explicitly ordered with respect to themselves and can break the associated edge, but that logic hasn't been implemented in the new pass yet. Until it is, this can be worked around by either switching to an older memory ordering pass or annotating the loop with __builtin_csa_parallel_{region,section}_{entry,exit} (or just __builtin_csa_parallel_loop if applicable) to explicitly indicate that the store doesn't need ordering relative to itself. - Some programs failed to compile due to a phase ordering bug that involved one of the IR optimization passes (CSAIRReductionOpt). This bug has been resolved. Simulator - The fountain operation now handles the length argument as operations normally do (consumed for each triggering). It also now verifies addresses.

1 0

Re: [Anl-intel-csa] CSA Fortran Offload
by Velesko, Paulius 20 Dec '17

20 Dec '17

Hello and update: Attached is the latest iteration of my attempt at getting some sort of offload working with fortran. To use it just replace your FC=gfortran/mpif90 with this guy. It’s set up to work on JLSE and you’ll need to source the CSA stuff prior. In order to offload functions to CSA you’ll need to put ‘!$omp declare target’ prior to the subroutine/function definition. You can then run that particular function normally or offload it off to CSA by putting ‘!$omp target’ above the function/subroutine call. Also, any omp in your code besides these two directives will be ignored. Some restrictions/use notes are included in the at the top of the file. Just vi it ( it’s a python script ) So far I’ve tried it with my toy codes, a few subroutines on XGC and ADCIRC with limited success. Main issues with ADCIRC are the inability to use imported functions/global vars. This can be solved but I really don’t want to go down that rabbit hole so a workaround is to pass everything as arguments, mark functions used inside with !$omp declare target and then inline them by hand (this script generates a bunch of files so that might help in this effort) The problem is XGC is the fact that it’s object oriented fortran that uses function polymorphism so I have to rename things and call them explicitly. This shouldn’t be too bad of a fix. It turns out that writing a compiler front end is quite laborious and there’s a lot of ways to say the same thing in fortran. I don’t have enough codes to cover all of the edge cases so if you try it and it doesn’t work I’d appreciate a tarball I could use to reproduce the error. Thanks, Paulius R Velesko From: anl-intel-csa [mailto:[email protected]] On Behalf Of Hammond, Jeff R Sent: Tuesday, December 19, 2017 10:26 AM To: anl-intel-csa(a)jlse.anl.gov; Hawes, FrederickX <frederickx.hawes(a)intel.com> Subject: Re: [Anl-intel-csa] CSA Fortran Offload Fred (cc) is working on Fortran CSA support on the Intel side. He's not able to access the Intel-hosted CSA forum so it would be great if he could be added to the email list and possibly the Argonne-hosted CSA forum. I already forwarded him the Python stuff and he has questions/comments. Thanks, Jeff From: anl-intel-csa <anl-intel-csa-bounces(a)lists.jlse.anl.gov<mailto:[email protected]>> on behalf of "Velesko, Paulius" <pvelesko(a)anl.gov<mailto:[email protected]>> Date: Tuesday, December 12, 2017 at 8:39 AM To: "Applencourt, Thomas" <tapplencourt(a)anl.gov<mailto:[email protected]>>, "Knight, Christopher J." <knightc(a)anl.gov<mailto:[email protected]>> Cc: "anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]>" <anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]>> Subject: Re: [Anl-intel-csa] CSA Fortran Offload I finished the script to allow for directive based offload using fortran. If somebody could give it a try I would appreciate any feedback. The You can open up bind_function.py to read a couple of use restrictions/instructions. Otherwise just python bind_function.py yourProgram.f. yourProgram.f should be in the same directory as the script and it takes in a name only not a path so don’t do python bind_function.py ./yourProgram.f. Thanks, Paulius From: Applencourt, Thomas Sent: Thursday, December 7, 2017 12:37 PM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>> Cc: anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]> Subject: Re: [Anl-intel-csa] CSA Fortran Offload I agree with Kris. Your binding seem correct (maybe just change `real (8)` into `C_DOUBLE` in your module to be cleaner ; anyway Fortran doesn't care about that kind of cast). You just need to pass a pointer as argument of the csa_args. Regards, Thomas Thomas Applencourt Leadership Computing Facility Argonne National Laboratory Building 240 Office - 1.D.18 9700 S Cass Av., Argonne Il, 60439 (630) 252-0036 ________________________________ From: anl-intel-csa <anl-intel-csa-bounces(a)lists.jlse.anl.gov<mailto:[email protected]>> on behalf of Christopher Knight <knightc(a)anl.gov<mailto:[email protected]>> Sent: Thursday, December 7, 2017 10:07 To: Velesko, Paulius Cc: anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]> Subject: Re: [Anl-intel-csa] CSA Fortran Offload Hi Paulius, I think the only change I needed to make was the following for the csa_args in mysubroutine.c. args[0].u.pv = (void*) x; args[1].u.pv = (void*) y; args[2].u.pv = (void*) z; [knightc@jlselogin1 2]$ ./exe Passed in 1+1=2 got back 2.0000000000000000 I copied your example here for reference. [knightc@jlselogin1 2]$ pwd /soft/restricted/intel_csa/apps/lammps/knight/work/fortran_example/2 chris On Dec 7, 2017, at 10:00 AM, Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> wrote: I meant I’ll try Real(8), intent(in), value :: x But you’re right – I should just continue passing by reference instead of dereference inside C Thank you From: Morozov, Vitali A. Sent: Thursday, December 7, 2017 9:51 AM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>>; Bertoni, Colleen <bertoni(a)anl.gov<mailto:[email protected]>>; Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Cc: anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]> Subject: Re: [Anl-intel-csa] CSA Fortran Offload Replacing "args[i].u.d" to "args[i].u.pv" is a better choice because %VAL in Fortran is non-standard and compiler dependent. On 12/07/2017 09:44 AM, Velesko, Paulius wrote: Yes, that what Vitaly said as well. For some reason I thought subroutines differ in that regard. I’ll try with the VALUE parameter. From: Bertoni, Colleen Sent: Thursday, December 7, 2017 9:41 AM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>>; Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Cc: anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]> Subject: Re: CSA Fortran Offload Assuming I understand the problem correctly, one possible problem is that the Fortran routine mysubroutine (upper left) will be expecting addresses of doubles (since Fortran is pass by reference), but in csa_call in the C function (bottom left), the csa_args are derefenced, and doubles, not pointers, are sent as arguments. (I guess it depends on what the csa graph is expecting. If it was generate from Fortran, it will be expecting addresses.) Colleen From: anl-intel-csa <anl-intel-csa-bounces(a)lists.jlse.anl.gov<mailto:[email protected]>> on behalf of Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> Sent: Thursday, December 7, 2017 8:26 AM To: Kumaran, Kalyan Cc: anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]> Subject: Re: [Anl-intel-csa] CSA Fortran Offload Hello all. I’m working on a hack to get fortran directive based offload working by wrapping functions up for AOT fortran offload through a C intermediate like the example you’ve sent me. I’m running into this issue: driver.f calls a function from a module which points to a C function. This C function does the CSA magic ultimately compiling a CSA graph from a function/subroutine written in fortran. I have fortran function calls now working on CSA but I’m having issues with subroutines. My driver executable crashes when I try to run csa_call even though my pointers for csa_assembly and csa_alloc do not seem to be failing. Here is a quick snapshot what I end up generating: Top left the original fortran function I want to offload Top right function wrapper to execute a generated C function Bottom left C function using csa calls to call the fortran csa graph Bottom right driver application If I remove the csa_call then the C function properly dereferences the pointers that were passed in from fortran so I’m certain there’s just something off about how I’m passing my csa args. Any ideas? Thank you, Paulius <image001.png> From: Velesko, Paulius Sent: Thursday, December 7, 2017 8:14 AM To: Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Cc: Jennings, Elise <ejennings(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>> Subject: RE: CSA Fortran Offload What email to do I cc this to to make it happen? From: Kumaran, Kalyan Sent: Thursday, December 7, 2017 8:05 AM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> Cc: Jennings, Elise <ejennings(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>> Subject: Re: CSA Fortran Offload I would prefer if these are sent to the ant-intel-csa list (even if Elise and Chris are the only ones who can help with them). I want them archived for future users. Could you send this? On Dec 7, 2017, at 7:57 AM, Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> wrote: Hello again. I’m working on a hack to get fortran directive based offload working by wrapping functions up for AOT fortran offload through a C intermediate like the example you’ve sent me. I’m running into this issue: driver.f calls a function from a module which points to a C function. This C function does the CSA magic ultimately compiling a CSA graph from a function/subroutine written in fortran. I have fortran function calls now working on CSA but I’m having issues with subroutines. I’m guessing it’s something to do with intent? My driver executable seg faults when I try to run csa_call even though my pointers for csa_assembly and csa_alloc do not seem to be failing. Here is a quick snapshot what I end up generating: Top left the original fortran function I want to offload Top right function wrapper to execute a generated C function Bottom left C function using csa calls to call the fortran csa graph Bottom right driver application If I remove the csa_call then the C function properly dereferences the pointers that were passed in from fortran so I’m certain there’s just something off about how I’m passing my csa args. Any ideas? Thank you, Paulius <image001.png> From: Velesko, Paulius Sent: Wednesday, December 6, 2017 5:57 PM To: Jennings, Elise <ejennings(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>>; Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Subject: RE: CSA Fortran Offload Hello again. I’m working on a hack to get fortran directive based offload working by wrapping functions up for AOT fortran offload through a C intermediate like the example you’ve sent me. I’m running into this issue: driver.f calls a function from a module which points to a C function. This C function does the CSA magic ultimately compiling a CSA graph from a function/subroutine written in fortran. I have fortran function calls now working on CSA but I’m having issues with subroutines. I’m guessing it’s something to do with intent? My driver executable crashes when I try to run csa_call even though my pointers for csa_assembly and csa_alloc do not seem to be failing. Here is a quick snapshot what I end up generating: Top left the original fortran function I want to offload Top right function wrapper to execute a generated C function Bottom left C function using csa calls to call the fortran csa graph Bottom right driver application If I remove the csa_call then the C function properly dereferences the pointers that were passed in from fortran so I’m certain there’s just something off about how I’m passing my csa args. Any ideas? Thank you, Paulius <image001.png> From: Velesko, Paulius Sent: Wednesday, November 29, 2017 5:24 PM To: Jennings, Elise <ejennings(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>>; Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Subject: RE: CSA Fortran Offload Very helpful. Thanks, Paulius From: Jennings, Elise Sent: Wednesday, November 29, 2017 3:13 PM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>>; Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Subject: Re: CSA Fortran Offload Hi Paulius, I have updated the wiki with a simple example. I hope this is useful. https://wiki.jlse.anl.gov/display/intelcsa/8.+Fortran+AOT+Example Best wishes, Elise From: Velesko, Paulius Sent: Wednesday, November 29, 2017 12:57:03 PM To: Jennings, Elise; Knight, Christopher J.; Kumaran, Kalyan Subject: RE: CSA Fortran Offload Much appreciated. Thanks, Paulius From: Jennings, Elise Sent: Wednesday, November 29, 2017 12:50 PM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>>; Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Subject: Re: CSA Fortran Offload Hi Paulius, I'm looking at adding a simple example of this to the wiki soon. It is an alternative to Jeff's code and hopefully should work. I'll let you know when I've added it (hopefully today). Best wishes, Elise From: Velesko, Paulius Sent: Wednesday, November 29, 2017 12:43:00 PM To: Knight, Christopher J.; Kumaran, Kalyan Cc: Jennings, Elise Subject: RE: CSA Fortran Offload This is not building for me out of the box. Have you built this successfully and perhaps could share the instructions? Thank you From: Knight, Christopher J. Sent: Wednesday, November 29, 2017 10:59 AM To: Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Cc: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>>; Jennings, Elise <ejennings(a)anl.gov<mailto:[email protected]>> Subject: Re: CSA Fortran Offload Hi Paulius, Attached is a small example that Jeff Hammond shared a while back. I think this is representative of the current state of things for running Fortran codes through the simulator: c bindings to csa functions and subroutine to be off-loaded written in c/c++. Elise (or others) that have worked with Fortran codes (probably not many) could say more. chris On Nov 29, 2017, at 10:52 AM, Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> wrote: Talk to Elise or chris. On Nov 29, 2017, at 10:51 AM, Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> wrote: In the C++ example you had to do some setup like csa_processor* proc = csa_alloc("autounit"); Is that also the case for fortran? Where could I find the reference to the syntax since I don’t see loop1.f example mentioned in the webpage? From: Kumaran, Kalyan Sent: Wednesday, November 29, 2017 10:34 AM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> Subject: Re: CSA Fortran Offload We do not have fortran support yet. Take a look at this: https://wiki.jlse.anl.gov/display/intelcsa/8.+Fortran+AOT+Example On Nov 29, 2017, at 10:30 AM, Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> wrote: Good morning. I’m looking to port a couple of Fortran routines to CSA. What’s the status of openmp target for fortran? Thanks _______________________________________________ anl-intel-csa mailing list anl-intel-csa(a)lists.jlse.anl.gov<mailto:[email protected]> https://lists.jlse.anl.gov/mailman/listinfo/anl-intel-csa _______________________________________________ anl-intel-csa mailing list anl-intel-csa(a)lists.jlse.anl.gov<mailto:[email protected]> https://lists.jlse.anl.gov/mailman/listinfo/anl-intel-csa _______________________________________________ anl-intel-csa mailing list anl-intel-csa(a)lists.jlse.anl.gov<mailto:[email protected]> https://lists.jlse.anl.gov/mailman/listinfo/anl-intel-csa

1 0

Re: [Anl-intel-csa] __builtin_csa_spmdization(k) with nested loops
by Jennings, Elise 19 Dec '17

19 Dec '17

Hi Daniel, Ok, thanks. The code is part of the SVM app on JLSE. The nested loop kernel offloaded to the csa is in this dir /home/elise/SVM/test Best wishes, Elise ________________________________ From: Woodworth, Daniel <daniel.woodworth(a)intel.com> Sent: Monday, December 18, 2017 12:49:35 PM To: Jennings, Elise; 'anl-intel-csa(a)jlse.anl.gov' Subject: RE: __builtin_csa_spmdization(k) with nested loops Ah, this assert is the one that triggers if the compiler runs out of lics: since we currently represent lics as physical registers, we can only have a finite number of them and this error happens if a function tries to use too many. We have someone working on improving this, but in practice code that hits the lic limits like this will usually also be too large to fit on the dataflow array anyway. You may need to decrease your worker count. Usually SPMDization should produce less code than unrolling, though, since it doesn’t need cleanup loops. What does the code that you’re working on look like? From: Jennings, Elise [mailto:[email protected]] Sent: Monday, December 18, 2017 1:09 PM To: Woodworth, Daniel <daniel.woodworth(a)intel.com>; 'anl-intel-csa(a)jlse.anl.gov' <anl-intel-csa(a)jlse.anl.gov> Subject: Re: __builtin_csa_spmdization(k) with nested loops Hi Daniel, Error message is below. Thanks, Elise icpc -g -O3 -I/soft/restricted/intel_csa/comp/fsim/2017-12-04-release -c svm.cpp icpc -g -O3 -I/soft/restricted/intel_csa/comp/fsim/2017-12-04-release -c matrix_precomputation.cpp icpc -g -O3 -I/soft/restricted/intel_csa/comp/fsim/2017-12-04-release svm-train.c svm.o -o svm-train -lm -L/soft/restricted/intel_csa/comp/fsim/2017-12-04-release -lcsasim -lrt -ldl csa-clang -S -O2 csa_kernel.cpp clang-6.0: /nfs/hd/disks/csaobj/comp/tools/2017-12-04-release/tools/src/llvm/include/llvm/MC/MCRegisterInfo.h:64: unsigned int llvm::MCRegisterClass::getRegister(unsigned int) const: Assertion `i < getNumRegs() && "Register number out of range!"' failed. #0 0x00000000031e58f8 llvm::sys::PrintStackTrace(llvm::raw_ostream&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x31e58f8) #1 0x00000000031e5989 PrintStackTraceSignalHandler(void*) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x31e5989) #2 0x00000000031e3e1c llvm::sys::RunSignalHandlers() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x31e3e1c) #3 0x00000000031e51df SignalHandler(int) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x31e51df) #4 0x00007f6963b055e0 __restore_rt (/lib64/libpthread.so.0+0xf5e0) #5 0x00007f6962c1d1f7 __GI_raise (/lib64/libc.so.6+0x351f7) #6 0x00007f6962c1e8e8 __GI_abort (/lib64/libc.so.6+0x368e8) #7 0x00007f6962c16266 __assert_fail_base (/lib64/libc.so.6+0x2e266) #8 0x00007f6962c16312 (/lib64/libc.so.6+0x2e312) #9 0x0000000001e7e097 llvm::MCRegisterClass::getRegister(unsigned int) const (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e7e097) #10 0x0000000001e7e113 llvm::TargetRegisterClass::getRegister(unsigned int) const (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e7e113) #11 0x0000000001e7e576 llvm::CSAMachineFunctionInfo::allocateLIC(llvm::TargetRegisterClass const*) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e7e576) #12 0x0000000001f035a8 llvm::CSACvtCFDFPass::assignLicForDF() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1f035a8) #13 0x0000000001efad89 llvm::CSACvtCFDFPass::runOnMachineFunction(llvm::MachineFunction&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1efad89) #14 0x00000000026bdb07 llvm::MachineFunctionPass::runOnFunction(llvm::Function&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x26bdb07) #15 0x0000000002b53260 llvm::FPPassManager::runOnFunction(llvm::Function&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b53260) #16 0x0000000002b533f3 llvm::FPPassManager::runOnModule(llvm::Module&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b533f3) #17 0x0000000002b5377f (anonymous namespace)::MPPassManager::runOnModule(llvm::Module&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b5377f) #18 0x0000000002b53e8a llvm::legacy::PassManagerImpl::run(llvm::Module&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b53e8a) #19 0x0000000002b540cb llvm::legacy::PassManager::run(llvm::Module&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b540cb) #20 0x0000000003464bbb (anonymous namespace)::EmitAssemblyHelper::EmitAssembly(clang::BackendAction, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream> >) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3464bbb) #21 0x000000000346737e clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::DataLayout const&, llvm::Module*, clang::BackendAction, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream> >) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x346737e) #22 0x0000000003f2c4df clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3f2c4df) #23 0x0000000004b827ec clang::ParseAST(clang::Sema&, bool, bool) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x4b827ec) #24 0x0000000003a5fd10 clang::ASTFrontendAction::ExecuteAction() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3a5fd10) #25 0x0000000003f2fcc6 clang::CodeGenAction::ExecuteAction() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3f2fcc6) #26 0x0000000003a5f6ff clang::FrontendAction::Execute() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3a5f6ff) #27 0x0000000003a073b6 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3a073b6) #28 0x0000000003b8f740 clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3b8f740) #29 0x0000000001e63868 cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e63868) #30 0x0000000001e59f96 ExecuteCC1Tool(llvm::ArrayRef<char const*>, llvm::StringRef) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e59f96) #31 0x0000000001e5ae92 main (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e5ae92) #32 0x00007f6962c09c05 __libc_start_main (/lib64/libc.so.6+0x21c05) #33 0x0000000001e561b1 _start /usr/src/packages/BUILD/glibc-2.11.3/csu/../sysdeps/x86_64/elf/start.S:116:0 Stack dump: 0. Program arguments: /soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0 -cc1 -triple csa -S -disable-free -main-file-name csa_kernel.cpp -mrelocation-model static -mthread-model posix -mdisable-fp-elim -no-integrated-as -mconstructor-aliases -fno-builtin-memset -fno-builtin-memcpy -include __clang_csa_math.h -dwarf-column-info -debugger-tuning=gdb -coverage-notes-file /home/elise/WFG/SVM/test/csa_kernel.gcno -resource-dir /soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/lib/clang/6.0.0 -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/ipp/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/mkl/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/pstl/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/tbb/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/tbb/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/daal/include -I/usr/intel/pkgs/ics/2013.1e/composer_xe_2013_sp1.2.144/tbb/include/ -internal-isystem /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../include/c++/4.8.5 -internal-isystem /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../include/c++/4.8.5/x86_64-redhat-linux -internal-isystem /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../include/c++/4.8.5/backward -internal-isystem /usr/local/include -internal-isystem /soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/lib/clang/6.0.0/include -internal-externc-isystem /include -internal-externc-isystem /usr/include -O2 -fdeprecated-macro -fno-dwarf-directory-asm -fdebug-compilation-dir /home/elise/WFG/SVM/test -ferror-limit 19 -fmessage-length 164 -fobjc-runtime=gcc -fcxx-exceptions -fexceptions -fdiagnostics-show-option -fcolor-diagnostics -vectorize-loops -vectorize-slp -o csa_kernel.s -x c++ csa_kernel.cpp 1. <eof> parser at end of file 2. Code generation 3. Running pass 'Function Pass Manager' on module 'csa_kernel.cpp'. 4. Running pass 'CSA Convert Control Flow to Data Flow' on function '@csa_kernel' clang-6.0: error: unable to execute command: Aborted clang-6.0: error: clang frontend command failed due to signal (use -v to see invocation) clang version 6.0.0 (ssh://git-amr-3.devtools.intel.com:29418/dcg-knp-arch-lpu-clang fe29d5a51c355eecac19d23a9cd0999976335069) (ssh://git-amr-3.devtools.intel.com:29418/dcg-knp-arch-lpu-llvm 719142b6d797956bb47096cbd9a93c2e7ab8ce51) Target: csa Thread model: posix InstalledDir: /soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/local/bin clang-6.0: note: diagnostic msg: PLEASE submit a bug report to http://llvm.org/bugs/ and include the crash backtrace, preprocessed source, and associated run script. clang-6.0: note: diagnostic msg: ******************** PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT: Preprocessed source(s) and associated run script(s) are located at: clang-6.0: note: diagnostic msg: /tmp/csa_kernel-22f338.cpp clang-6.0: note: diagnostic msg: /tmp/csa_kernel-22f338.sh clang-6.0: note: diagnostic msg: ******************** make: *** [csa_kernel.s] Error 254 [elise@jlselogin1 test]$ ________________________________ From: Woodworth, Daniel <daniel.woodworth(a)intel.com<mailto:[email protected]>> Sent: Monday, December 18, 2017 10:09:36 AM To: Jennings, Elise; 'anl-intel-csa(a)jlse.anl.gov' Subject: RE: __builtin_csa_spmdization(k) with nested loops You should be able to. What error message are you getting? From: anl-intel-csa [mailto:[email protected]] On Behalf Of Jennings, Elise Sent: Monday, December 18, 2017 10:56 AM To: 'anl-intel-csa(a)jlse.anl.gov' <anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]>> Subject: [Anl-intel-csa] __builtin_csa_spmdization(k) with nested loops Hi all, Can I use __builtin_csa_spmdization(k) on nested loops? Previously I used two __builtin_csa_parallel_region entry calls each with different unroll factors. When I replace both with SPMDization I get errors when compiling. If I replace only one __builtin_cas_parallel region is works fine. Thanks, Elise ________________________________ From: anl-intel-csa <anl-intel-csa-bounces(a)lists.jlse.anl.gov<mailto:[email protected]>> on behalf of Glossop, Kent <kent.glossop(a)intel.com<mailto:[email protected]>> Sent: Monday, December 4, 2017 9:42 PM To: 'anl-intel-csa(a)jlse.anl.gov' Subject: [Anl-intel-csa] CSA SDK Release Notes for 2017/12/04 CSA SDK Release Notes for 2017/12/04 version The default release has been changed to 2017-12-04-release. Relevant changes since the 2017-11-21-release: SDK Environment • The SDK environmental setup now provides a “csaenv” command. After invoking the csasetup script, the “latest” tools will be selected. The “csaenv” command can be used to show what is set, and to switch between versions if necessary. o Once csasetup.bash is sourced, user can use ‘csaenv’ to switch between environments o Examples: • `csaenv list` • Shows all available environments in system • `csaenv set fsim/2017-12-04-release tools/2017-11-29-release’ • Sets current environment with fsim built in 2017-12-04 and tools built in 2017-11-29 o Execute `csaenv help` for more details” Compiler / Runtime • Added a warning when the SPMDization region is not structured code; the code should be OpenMP-conforming (branches to or from the SPMDization region (continue, break, etc.) are illegal) • Fixes: o Add the zero trip count guard to each of the generated workers. Now, SPMDization should generate correct code when 0 <= #iterations < #workers Advisor • Introduced special mode for Dependencies analysis (-loop-call-count-limit=N, switched off by default) which helps to dramatically reduce Spatial Advisor Dependencies overhead o Example: advixe-cl --collect dependencies -filter-reductions -loops=”scalar,loop-height=0” -loop-call-count-limit=16 … • Improved SAPPC html report output and included new metrics (e.g. tripcounts, normalized total time) • Improved SAPPC performance by disabling Advisor “recommendations engine” (which is not yet used in SAPPC)

1 0

Re: [Anl-intel-csa] CSA Fortran Offload
by Hammond, Jeff R 19 Dec '17

19 Dec '17

Fred (cc) is working on Fortran CSA support on the Intel side. He's not able to access the Intel-hosted CSA forum so it would be great if he could be added to the email list and possibly the Argonne-hosted CSA forum. I already forwarded him the Python stuff and he has questions/comments. Thanks, Jeff From: anl-intel-csa <anl-intel-csa-bounces(a)lists.jlse.anl.gov> on behalf of "Velesko, Paulius" <pvelesko(a)anl.gov> Date: Tuesday, December 12, 2017 at 8:39 AM To: "Applencourt, Thomas" <tapplencourt(a)anl.gov>, "Knight, Christopher J." <knightc(a)anl.gov> Cc: "anl-intel-csa(a)jlse.anl.gov" <anl-intel-csa(a)jlse.anl.gov> Subject: Re: [Anl-intel-csa] CSA Fortran Offload I finished the script to allow for directive based offload using fortran. If somebody could give it a try I would appreciate any feedback. The You can open up bind_function.py to read a couple of use restrictions/instructions. Otherwise just python bind_function.py yourProgram.f. yourProgram.f should be in the same directory as the script and it takes in a name only not a path so don’t do python bind_function.py ./yourProgram.f. Thanks, Paulius From: Applencourt, Thomas Sent: Thursday, December 7, 2017 12:37 PM To: Velesko, Paulius <pvelesko(a)anl.gov>; Knight, Christopher J. <knightc(a)anl.gov> Cc: anl-intel-csa(a)jlse.anl.gov Subject: Re: [Anl-intel-csa] CSA Fortran Offload I agree with Kris. Your binding seem correct (maybe just change `real (8)` into `C_DOUBLE` in your module to be cleaner ; anyway Fortran doesn't care about that kind of cast). You just need to pass a pointer as argument of the csa_args. Regards, Thomas Thomas Applencourt Leadership Computing Facility Argonne National Laboratory Building 240 Office - 1.D.18 9700 S Cass Av., Argonne Il, 60439 (630) 252-0036 ________________________________ From: anl-intel-csa <anl-intel-csa-bounces(a)lists.jlse.anl.gov<mailto:[email protected]>> on behalf of Christopher Knight <knightc(a)anl.gov<mailto:[email protected]>> Sent: Thursday, December 7, 2017 10:07 To: Velesko, Paulius Cc: anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]> Subject: Re: [Anl-intel-csa] CSA Fortran Offload Hi Paulius, I think the only change I needed to make was the following for the csa_args in mysubroutine.c. args[0].u.pv = (void*) x; args[1].u.pv = (void*) y; args[2].u.pv = (void*) z; [knightc@jlselogin1 2]$ ./exe Passed in 1+1=2 got back 2.0000000000000000 I copied your example here for reference. [knightc@jlselogin1 2]$ pwd /soft/restricted/intel_csa/apps/lammps/knight/work/fortran_example/2 chris On Dec 7, 2017, at 10:00 AM, Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> wrote: I meant I’ll try Real(8), intent(in), value :: x But you’re right – I should just continue passing by reference instead of dereference inside C Thank you From: Morozov, Vitali A. Sent: Thursday, December 7, 2017 9:51 AM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>>; Bertoni, Colleen <bertoni(a)anl.gov<mailto:[email protected]>>; Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Cc: anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]> Subject: Re: [Anl-intel-csa] CSA Fortran Offload Replacing "args[i].u.d" to "args[i].u.pv" is a better choice because %VAL in Fortran is non-standard and compiler dependent. On 12/07/2017 09:44 AM, Velesko, Paulius wrote: Yes, that what Vitaly said as well. For some reason I thought subroutines differ in that regard. I’ll try with the VALUE parameter. From: Bertoni, Colleen Sent: Thursday, December 7, 2017 9:41 AM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>>; Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Cc: anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]> Subject: Re: CSA Fortran Offload Assuming I understand the problem correctly, one possible problem is that the Fortran routine mysubroutine (upper left) will be expecting addresses of doubles (since Fortran is pass by reference), but in csa_call in the C function (bottom left), the csa_args are derefenced, and doubles, not pointers, are sent as arguments. (I guess it depends on what the csa graph is expecting. If it was generate from Fortran, it will be expecting addresses.) Colleen From: anl-intel-csa <anl-intel-csa-bounces(a)lists.jlse.anl.gov<mailto:[email protected]>> on behalf of Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> Sent: Thursday, December 7, 2017 8:26 AM To: Kumaran, Kalyan Cc: anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]> Subject: Re: [Anl-intel-csa] CSA Fortran Offload Hello all. I’m working on a hack to get fortran directive based offload working by wrapping functions up for AOT fortran offload through a C intermediate like the example you’ve sent me. I’m running into this issue: driver.f calls a function from a module which points to a C function. This C function does the CSA magic ultimately compiling a CSA graph from a function/subroutine written in fortran. I have fortran function calls now working on CSA but I’m having issues with subroutines. My driver executable crashes when I try to run csa_call even though my pointers for csa_assembly and csa_alloc do not seem to be failing. Here is a quick snapshot what I end up generating: Top left the original fortran function I want to offload Top right function wrapper to execute a generated C function Bottom left C function using csa calls to call the fortran csa graph Bottom right driver application If I remove the csa_call then the C function properly dereferences the pointers that were passed in from fortran so I’m certain there’s just something off about how I’m passing my csa args. Any ideas? Thank you, Paulius <image001.png> From: Velesko, Paulius Sent: Thursday, December 7, 2017 8:14 AM To: Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Cc: Jennings, Elise <ejennings(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>> Subject: RE: CSA Fortran Offload What email to do I cc this to to make it happen? From: Kumaran, Kalyan Sent: Thursday, December 7, 2017 8:05 AM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> Cc: Jennings, Elise <ejennings(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>> Subject: Re: CSA Fortran Offload I would prefer if these are sent to the ant-intel-csa list (even if Elise and Chris are the only ones who can help with them). I want them archived for future users. Could you send this? On Dec 7, 2017, at 7:57 AM, Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> wrote: Hello again. I’m working on a hack to get fortran directive based offload working by wrapping functions up for AOT fortran offload through a C intermediate like the example you’ve sent me. I’m running into this issue: driver.f calls a function from a module which points to a C function. This C function does the CSA magic ultimately compiling a CSA graph from a function/subroutine written in fortran. I have fortran function calls now working on CSA but I’m having issues with subroutines. I’m guessing it’s something to do with intent? My driver executable seg faults when I try to run csa_call even though my pointers for csa_assembly and csa_alloc do not seem to be failing. Here is a quick snapshot what I end up generating: Top left the original fortran function I want to offload Top right function wrapper to execute a generated C function Bottom left C function using csa calls to call the fortran csa graph Bottom right driver application If I remove the csa_call then the C function properly dereferences the pointers that were passed in from fortran so I’m certain there’s just something off about how I’m passing my csa args. Any ideas? Thank you, Paulius <image001.png> From: Velesko, Paulius Sent: Wednesday, December 6, 2017 5:57 PM To: Jennings, Elise <ejennings(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>>; Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Subject: RE: CSA Fortran Offload Hello again. I’m working on a hack to get fortran directive based offload working by wrapping functions up for AOT fortran offload through a C intermediate like the example you’ve sent me. I’m running into this issue: driver.f calls a function from a module which points to a C function. This C function does the CSA magic ultimately compiling a CSA graph from a function/subroutine written in fortran. I have fortran function calls now working on CSA but I’m having issues with subroutines. I’m guessing it’s something to do with intent? My driver executable crashes when I try to run csa_call even though my pointers for csa_assembly and csa_alloc do not seem to be failing. Here is a quick snapshot what I end up generating: Top left the original fortran function I want to offload Top right function wrapper to execute a generated C function Bottom left C function using csa calls to call the fortran csa graph Bottom right driver application If I remove the csa_call then the C function properly dereferences the pointers that were passed in from fortran so I’m certain there’s just something off about how I’m passing my csa args. Any ideas? Thank you, Paulius <image001.png> From: Velesko, Paulius Sent: Wednesday, November 29, 2017 5:24 PM To: Jennings, Elise <ejennings(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>>; Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Subject: RE: CSA Fortran Offload Very helpful. Thanks, Paulius From: Jennings, Elise Sent: Wednesday, November 29, 2017 3:13 PM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>>; Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Subject: Re: CSA Fortran Offload Hi Paulius, I have updated the wiki with a simple example. I hope this is useful. https://wiki.jlse.anl.gov/display/intelcsa/8.+Fortran+AOT+Example Best wishes, Elise From: Velesko, Paulius Sent: Wednesday, November 29, 2017 12:57:03 PM To: Jennings, Elise; Knight, Christopher J.; Kumaran, Kalyan Subject: RE: CSA Fortran Offload Much appreciated. Thanks, Paulius From: Jennings, Elise Sent: Wednesday, November 29, 2017 12:50 PM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>>; Knight, Christopher J. <knightc(a)anl.gov<mailto:[email protected]>>; Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Subject: Re: CSA Fortran Offload Hi Paulius, I'm looking at adding a simple example of this to the wiki soon. It is an alternative to Jeff's code and hopefully should work. I'll let you know when I've added it (hopefully today). Best wishes, Elise From: Velesko, Paulius Sent: Wednesday, November 29, 2017 12:43:00 PM To: Knight, Christopher J.; Kumaran, Kalyan Cc: Jennings, Elise Subject: RE: CSA Fortran Offload This is not building for me out of the box. Have you built this successfully and perhaps could share the instructions? Thank you From: Knight, Christopher J. Sent: Wednesday, November 29, 2017 10:59 AM To: Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> Cc: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>>; Jennings, Elise <ejennings(a)anl.gov<mailto:[email protected]>> Subject: Re: CSA Fortran Offload Hi Paulius, Attached is a small example that Jeff Hammond shared a while back. I think this is representative of the current state of things for running Fortran codes through the simulator: c bindings to csa functions and subroutine to be off-loaded written in c/c++. Elise (or others) that have worked with Fortran codes (probably not many) could say more. chris On Nov 29, 2017, at 10:52 AM, Kumaran, Kalyan <kumaran(a)anl.gov<mailto:[email protected]>> wrote: Talk to Elise or chris. On Nov 29, 2017, at 10:51 AM, Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> wrote: In the C++ example you had to do some setup like csa_processor* proc = csa_alloc("autounit"); Is that also the case for fortran? Where could I find the reference to the syntax since I don’t see loop1.f example mentioned in the webpage? From: Kumaran, Kalyan Sent: Wednesday, November 29, 2017 10:34 AM To: Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> Subject: Re: CSA Fortran Offload We do not have fortran support yet. Take a look at this: https://wiki.jlse.anl.gov/display/intelcsa/8.+Fortran+AOT+Example On Nov 29, 2017, at 10:30 AM, Velesko, Paulius <pvelesko(a)anl.gov<mailto:[email protected]>> wrote: Good morning. I’m looking to port a couple of Fortran routines to CSA. What’s the status of openmp target for fortran? Thanks _______________________________________________ anl-intel-csa mailing list anl-intel-csa(a)lists.jlse.anl.gov<mailto:[email protected]> https://lists.jlse.anl.gov/mailman/listinfo/anl-intel-csa _______________________________________________ anl-intel-csa mailing list anl-intel-csa(a)lists.jlse.anl.gov<mailto:[email protected]> https://lists.jlse.anl.gov/mailman/listinfo/anl-intel-csa _______________________________________________ anl-intel-csa mailing list anl-intel-csa(a)lists.jlse.anl.gov<mailto:[email protected]> https://lists.jlse.anl.gov/mailman/listinfo/anl-intel-csa

1 0

Re: [Anl-intel-csa] __builtin_csa_spmdization(k) with nested loops
by Woodworth, Daniel 18 Dec '17

18 Dec '17

Ah, this assert is the one that triggers if the compiler runs out of lics: since we currently represent lics as physical registers, we can only have a finite number of them and this error happens if a function tries to use too many. We have someone working on improving this, but in practice code that hits the lic limits like this will usually also be too large to fit on the dataflow array anyway. You may need to decrease your worker count. Usually SPMDization should produce less code than unrolling, though, since it doesn’t need cleanup loops. What does the code that you’re working on look like? From: Jennings, Elise [mailto:[email protected]] Sent: Monday, December 18, 2017 1:09 PM To: Woodworth, Daniel <daniel.woodworth(a)intel.com>; 'anl-intel-csa(a)jlse.anl.gov' <anl-intel-csa(a)jlse.anl.gov> Subject: Re: __builtin_csa_spmdization(k) with nested loops Hi Daniel, Error message is below. Thanks, Elise icpc -g -O3 -I/soft/restricted/intel_csa/comp/fsim/2017-12-04-release -c svm.cpp icpc -g -O3 -I/soft/restricted/intel_csa/comp/fsim/2017-12-04-release -c matrix_precomputation.cpp icpc -g -O3 -I/soft/restricted/intel_csa/comp/fsim/2017-12-04-release svm-train.c svm.o -o svm-train -lm -L/soft/restricted/intel_csa/comp/fsim/2017-12-04-release -lcsasim -lrt -ldl csa-clang -S -O2 csa_kernel.cpp clang-6.0: /nfs/hd/disks/csaobj/comp/tools/2017-12-04-release/tools/src/llvm/include/llvm/MC/MCRegisterInfo.h:64: unsigned int llvm::MCRegisterClass::getRegister(unsigned int) const: Assertion `i < getNumRegs() && "Register number out of range!"' failed. #0 0x00000000031e58f8 llvm::sys::PrintStackTrace(llvm::raw_ostream&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x31e58f8) #1 0x00000000031e5989 PrintStackTraceSignalHandler(void*) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x31e5989) #2 0x00000000031e3e1c llvm::sys::RunSignalHandlers() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x31e3e1c) #3 0x00000000031e51df SignalHandler(int) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x31e51df) #4 0x00007f6963b055e0 __restore_rt (/lib64/libpthread.so.0+0xf5e0) #5 0x00007f6962c1d1f7 __GI_raise (/lib64/libc.so.6+0x351f7) #6 0x00007f6962c1e8e8 __GI_abort (/lib64/libc.so.6+0x368e8) #7 0x00007f6962c16266 __assert_fail_base (/lib64/libc.so.6+0x2e266) #8 0x00007f6962c16312 (/lib64/libc.so.6+0x2e312) #9 0x0000000001e7e097 llvm::MCRegisterClass::getRegister(unsigned int) const (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e7e097) #10 0x0000000001e7e113 llvm::TargetRegisterClass::getRegister(unsigned int) const (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e7e113) #11 0x0000000001e7e576 llvm::CSAMachineFunctionInfo::allocateLIC(llvm::TargetRegisterClass const*) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e7e576) #12 0x0000000001f035a8 llvm::CSACvtCFDFPass::assignLicForDF() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1f035a8) #13 0x0000000001efad89 llvm::CSACvtCFDFPass::runOnMachineFunction(llvm::MachineFunction&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1efad89) #14 0x00000000026bdb07 llvm::MachineFunctionPass::runOnFunction(llvm::Function&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x26bdb07) #15 0x0000000002b53260 llvm::FPPassManager::runOnFunction(llvm::Function&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b53260) #16 0x0000000002b533f3 llvm::FPPassManager::runOnModule(llvm::Module&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b533f3) #17 0x0000000002b5377f (anonymous namespace)::MPPassManager::runOnModule(llvm::Module&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b5377f) #18 0x0000000002b53e8a llvm::legacy::PassManagerImpl::run(llvm::Module&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b53e8a) #19 0x0000000002b540cb llvm::legacy::PassManager::run(llvm::Module&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b540cb) #20 0x0000000003464bbb (anonymous namespace)::EmitAssemblyHelper::EmitAssembly(clang::BackendAction, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream> >) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3464bbb) #21 0x000000000346737e clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::DataLayout const&, llvm::Module*, clang::BackendAction, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream> >) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x346737e) #22 0x0000000003f2c4df clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3f2c4df) #23 0x0000000004b827ec clang::ParseAST(clang::Sema&, bool, bool) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x4b827ec) #24 0x0000000003a5fd10 clang::ASTFrontendAction::ExecuteAction() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3a5fd10) #25 0x0000000003f2fcc6 clang::CodeGenAction::ExecuteAction() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3f2fcc6) #26 0x0000000003a5f6ff clang::FrontendAction::Execute() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3a5f6ff) #27 0x0000000003a073b6 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3a073b6) #28 0x0000000003b8f740 clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3b8f740) #29 0x0000000001e63868 cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e63868) #30 0x0000000001e59f96 ExecuteCC1Tool(llvm::ArrayRef<char const*>, llvm::StringRef) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e59f96) #31 0x0000000001e5ae92 main (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e5ae92) #32 0x00007f6962c09c05 __libc_start_main (/lib64/libc.so.6+0x21c05) #33 0x0000000001e561b1 _start /usr/src/packages/BUILD/glibc-2.11.3/csu/../sysdeps/x86_64/elf/start.S:116:0 Stack dump: 0. Program arguments: /soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0 -cc1 -triple csa -S -disable-free -main-file-name csa_kernel.cpp -mrelocation-model static -mthread-model posix -mdisable-fp-elim -no-integrated-as -mconstructor-aliases -fno-builtin-memset -fno-builtin-memcpy -include __clang_csa_math.h -dwarf-column-info -debugger-tuning=gdb -coverage-notes-file /home/elise/WFG/SVM/test/csa_kernel.gcno -resource-dir /soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/lib/clang/6.0.0 -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/ipp/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/mkl/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/pstl/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/tbb/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/tbb/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/daal/include -I/usr/intel/pkgs/ics/2013.1e/composer_xe_2013_sp1.2.144/tbb/include/ -internal-isystem /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../include/c++/4.8.5 -internal-isystem /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../include/c++/4.8.5/x86_64-redhat-linux -internal-isystem /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../include/c++/4.8.5/backward -internal-isystem /usr/local/include -internal-isystem /soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/lib/clang/6.0.0/include -internal-externc-isystem /include -internal-externc-isystem /usr/include -O2 -fdeprecated-macro -fno-dwarf-directory-asm -fdebug-compilation-dir /home/elise/WFG/SVM/test -ferror-limit 19 -fmessage-length 164 -fobjc-runtime=gcc -fcxx-exceptions -fexceptions -fdiagnostics-show-option -fcolor-diagnostics -vectorize-loops -vectorize-slp -o csa_kernel.s -x c++ csa_kernel.cpp 1. <eof> parser at end of file 2. Code generation 3. Running pass 'Function Pass Manager' on module 'csa_kernel.cpp'. 4. Running pass 'CSA Convert Control Flow to Data Flow' on function '@csa_kernel' clang-6.0: error: unable to execute command: Aborted clang-6.0: error: clang frontend command failed due to signal (use -v to see invocation) clang version 6.0.0 (ssh://git-amr-3.devtools.intel.com:29418/dcg-knp-arch-lpu-clang fe29d5a51c355eecac19d23a9cd0999976335069) (ssh://git-amr-3.devtools.intel.com:29418/dcg-knp-arch-lpu-llvm 719142b6d797956bb47096cbd9a93c2e7ab8ce51) Target: csa Thread model: posix InstalledDir: /soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/local/bin clang-6.0: note: diagnostic msg: PLEASE submit a bug report to http://llvm.org/bugs/ and include the crash backtrace, preprocessed source, and associated run script. clang-6.0: note: diagnostic msg: ******************** PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT: Preprocessed source(s) and associated run script(s) are located at: clang-6.0: note: diagnostic msg: /tmp/csa_kernel-22f338.cpp clang-6.0: note: diagnostic msg: /tmp/csa_kernel-22f338.sh clang-6.0: note: diagnostic msg: ******************** make: *** [csa_kernel.s] Error 254 [elise@jlselogin1 test]$ ________________________________ From: Woodworth, Daniel <daniel.woodworth(a)intel.com<mailto:[email protected]>> Sent: Monday, December 18, 2017 10:09:36 AM To: Jennings, Elise; 'anl-intel-csa(a)jlse.anl.gov' Subject: RE: __builtin_csa_spmdization(k) with nested loops You should be able to. What error message are you getting? From: anl-intel-csa [mailto:[email protected]] On Behalf Of Jennings, Elise Sent: Monday, December 18, 2017 10:56 AM To: 'anl-intel-csa(a)jlse.anl.gov' <anl-intel-csa(a)jlse.anl.gov<mailto:[email protected]>> Subject: [Anl-intel-csa] __builtin_csa_spmdization(k) with nested loops Hi all, Can I use __builtin_csa_spmdization(k) on nested loops? Previously I used two __builtin_csa_parallel_region entry calls each with different unroll factors. When I replace both with SPMDization I get errors when compiling. If I replace only one __builtin_cas_parallel region is works fine. Thanks, Elise ________________________________ From: anl-intel-csa <anl-intel-csa-bounces(a)lists.jlse.anl.gov<mailto:[email protected]>> on behalf of Glossop, Kent <kent.glossop(a)intel.com<mailto:[email protected]>> Sent: Monday, December 4, 2017 9:42 PM To: 'anl-intel-csa(a)jlse.anl.gov' Subject: [Anl-intel-csa] CSA SDK Release Notes for 2017/12/04 CSA SDK Release Notes for 2017/12/04 version The default release has been changed to 2017-12-04-release. Relevant changes since the 2017-11-21-release: SDK Environment • The SDK environmental setup now provides a “csaenv” command. After invoking the csasetup script, the “latest” tools will be selected. The “csaenv” command can be used to show what is set, and to switch between versions if necessary. o Once csasetup.bash is sourced, user can use ‘csaenv’ to switch between environments o Examples: • `csaenv list` • Shows all available environments in system • `csaenv set fsim/2017-12-04-release tools/2017-11-29-release’ • Sets current environment with fsim built in 2017-12-04 and tools built in 2017-11-29 o Execute `csaenv help` for more details” Compiler / Runtime • Added a warning when the SPMDization region is not structured code; the code should be OpenMP-conforming (branches to or from the SPMDization region (continue, break, etc.) are illegal) • Fixes: o Add the zero trip count guard to each of the generated workers. Now, SPMDization should generate correct code when 0 <= #iterations < #workers Advisor • Introduced special mode for Dependencies analysis (-loop-call-count-limit=N, switched off by default) which helps to dramatically reduce Spatial Advisor Dependencies overhead o Example: advixe-cl --collect dependencies -filter-reductions -loops=”scalar,loop-height=0” -loop-call-count-limit=16 … • Improved SAPPC html report output and included new metrics (e.g. tripcounts, normalized total time) • Improved SAPPC performance by disabling Advisor “recommendations engine” (which is not yet used in SAPPC)

1 0

Re: [Anl-intel-csa] __builtin_csa_spmdization(k) with nested loops
by Jennings, Elise 18 Dec '17

18 Dec '17

Hi Daniel, Error message is below. Thanks, Elise icpc -g -O3 -I/soft/restricted/intel_csa/comp/fsim/2017-12-04-release -c svm.cpp icpc -g -O3 -I/soft/restricted/intel_csa/comp/fsim/2017-12-04-release -c matrix_precomputation.cpp icpc -g -O3 -I/soft/restricted/intel_csa/comp/fsim/2017-12-04-release svm-train.c svm.o -o svm-train -lm -L/soft/restricted/intel_csa/comp/fsim/2017-12-04-release -lcsasim -lrt -ldl csa-clang -S -O2 csa_kernel.cpp clang-6.0: /nfs/hd/disks/csaobj/comp/tools/2017-12-04-release/tools/src/llvm/include/llvm/MC/MCRegisterInfo.h:64: unsigned int llvm::MCRegisterClass::getRegister(unsigned int) const: Assertion `i < getNumRegs() && "Register number out of range!"' failed. #0 0x00000000031e58f8 llvm::sys::PrintStackTrace(llvm::raw_ostream&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x31e58f8) #1 0x00000000031e5989 PrintStackTraceSignalHandler(void*) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x31e5989) #2 0x00000000031e3e1c llvm::sys::RunSignalHandlers() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x31e3e1c) #3 0x00000000031e51df SignalHandler(int) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x31e51df) #4 0x00007f6963b055e0 __restore_rt (/lib64/libpthread.so.0+0xf5e0) #5 0x00007f6962c1d1f7 __GI_raise (/lib64/libc.so.6+0x351f7) #6 0x00007f6962c1e8e8 __GI_abort (/lib64/libc.so.6+0x368e8) #7 0x00007f6962c16266 __assert_fail_base (/lib64/libc.so.6+0x2e266) #8 0x00007f6962c16312 (/lib64/libc.so.6+0x2e312) #9 0x0000000001e7e097 llvm::MCRegisterClass::getRegister(unsigned int) const (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e7e097) #10 0x0000000001e7e113 llvm::TargetRegisterClass::getRegister(unsigned int) const (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e7e113) #11 0x0000000001e7e576 llvm::CSAMachineFunctionInfo::allocateLIC(llvm::TargetRegisterClass const*) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e7e576) #12 0x0000000001f035a8 llvm::CSACvtCFDFPass::assignLicForDF() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1f035a8) #13 0x0000000001efad89 llvm::CSACvtCFDFPass::runOnMachineFunction(llvm::MachineFunction&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1efad89) #14 0x00000000026bdb07 llvm::MachineFunctionPass::runOnFunction(llvm::Function&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x26bdb07) #15 0x0000000002b53260 llvm::FPPassManager::runOnFunction(llvm::Function&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b53260) #16 0x0000000002b533f3 llvm::FPPassManager::runOnModule(llvm::Module&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b533f3) #17 0x0000000002b5377f (anonymous namespace)::MPPassManager::runOnModule(llvm::Module&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b5377f) #18 0x0000000002b53e8a llvm::legacy::PassManagerImpl::run(llvm::Module&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b53e8a) #19 0x0000000002b540cb llvm::legacy::PassManager::run(llvm::Module&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x2b540cb) #20 0x0000000003464bbb (anonymous namespace)::EmitAssemblyHelper::EmitAssembly(clang::BackendAction, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream> >) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3464bbb) #21 0x000000000346737e clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::DataLayout const&, llvm::Module*, clang::BackendAction, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream> >) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x346737e) #22 0x0000000003f2c4df clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3f2c4df) #23 0x0000000004b827ec clang::ParseAST(clang::Sema&, bool, bool) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x4b827ec) #24 0x0000000003a5fd10 clang::ASTFrontendAction::ExecuteAction() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3a5fd10) #25 0x0000000003f2fcc6 clang::CodeGenAction::ExecuteAction() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3f2fcc6) #26 0x0000000003a5f6ff clang::FrontendAction::Execute() (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3a5f6ff) #27 0x0000000003a073b6 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3a073b6) #28 0x0000000003b8f740 clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x3b8f740) #29 0x0000000001e63868 cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e63868) #30 0x0000000001e59f96 ExecuteCC1Tool(llvm::ArrayRef<char const*>, llvm::StringRef) (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e59f96) #31 0x0000000001e5ae92 main (/soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0+0x1e5ae92) #32 0x00007f6962c09c05 __libc_start_main (/lib64/libc.so.6+0x21c05) #33 0x0000000001e561b1 _start /usr/src/packages/BUILD/glibc-2.11.3/csu/../sysdeps/x86_64/elf/start.S:116:0 Stack dump: 0. Program arguments: /soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/bin/clang-6.0 -cc1 -triple csa -S -disable-free -main-file-name csa_kernel.cpp -mrelocation-model static -mthread-model posix -mdisable-fp-elim -no-integrated-as -mconstructor-aliases -fno-builtin-memset -fno-builtin-memcpy -include __clang_csa_math.h -dwarf-column-info -debugger-tuning=gdb -coverage-notes-file /home/elise/WFG/SVM/test/csa_kernel.gcno -resource-dir /soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/lib/clang/6.0.0 -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/ipp/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/mkl/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/pstl/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/tbb/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/tbb/include -I/soft/compilers/intel/compilers_and_libraries_2018.1.163/linux/daal/include -I/usr/intel/pkgs/ics/2013.1e/composer_xe_2013_sp1.2.144/tbb/include/ -internal-isystem /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../include/c++/4.8.5 -internal-isystem /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../include/c++/4.8.5/x86_64-redhat-linux -internal-isystem /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../include/c++/4.8.5/backward -internal-isystem /usr/local/include -internal-isystem /soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/install/lib/clang/6.0.0/include -internal-externc-isystem /include -internal-externc-isystem /usr/include -O2 -fdeprecated-macro -fno-dwarf-directory-asm -fdebug-compilation-dir /home/elise/WFG/SVM/test -ferror-limit 19 -fmessage-length 164 -fobjc-runtime=gcc -fcxx-exceptions -fexceptions -fdiagnostics-show-option -fcolor-diagnostics -vectorize-loops -vectorize-slp -o csa_kernel.s -x c++ csa_kernel.cpp 1. <eof> parser at end of file 2. Code generation 3. Running pass 'Function Pass Manager' on module 'csa_kernel.cpp'. 4. Running pass 'CSA Convert Control Flow to Data Flow' on function '@csa_kernel' clang-6.0: error: unable to execute command: Aborted clang-6.0: error: clang frontend command failed due to signal (use -v to see invocation) clang version 6.0.0 (ssh://git-amr-3.devtools.intel.com:29418/dcg-knp-arch-lpu-clang fe29d5a51c355eecac19d23a9cd0999976335069) (ssh://git-amr-3.devtools.intel.com:29418/dcg-knp-arch-lpu-llvm 719142b6d797956bb47096cbd9a93c2e7ab8ce51) Target: csa Thread model: posix InstalledDir: /soft/restricted/intel_csa/comp/tools/2017-12-04-release/tools/local/bin clang-6.0: note: diagnostic msg: PLEASE submit a bug report to http://llvm.org/bugs/ and include the crash backtrace, preprocessed source, and associated run script. clang-6.0: note: diagnostic msg: ******************** PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT: Preprocessed source(s) and associated run script(s) are located at: clang-6.0: note: diagnostic msg: /tmp/csa_kernel-22f338.cpp clang-6.0: note: diagnostic msg: /tmp/csa_kernel-22f338.sh clang-6.0: note: diagnostic msg: ******************** make: *** [csa_kernel.s] Error 254 [elise@jlselogin1 test]$ ________________________________ From: Woodworth, Daniel <daniel.woodworth(a)intel.com> Sent: Monday, December 18, 2017 10:09:36 AM To: Jennings, Elise; 'anl-intel-csa(a)jlse.anl.gov' Subject: RE: __builtin_csa_spmdization(k) with nested loops You should be able to. What error message are you getting? From: anl-intel-csa [mailto:[email protected]] On Behalf Of Jennings, Elise Sent: Monday, December 18, 2017 10:56 AM To: 'anl-intel-csa(a)jlse.anl.gov' <anl-intel-csa(a)jlse.anl.gov> Subject: [Anl-intel-csa] __builtin_csa_spmdization(k) with nested loops Hi all, Can I use __builtin_csa_spmdization(k) on nested loops? Previously I used two __builtin_csa_parallel_region entry calls each with different unroll factors. When I replace both with SPMDization I get errors when compiling. If I replace only one __builtin_cas_parallel region is works fine. Thanks, Elise ________________________________ From: anl-intel-csa <anl-intel-csa-bounces(a)lists.jlse.anl.gov<mailto:[email protected]>> on behalf of Glossop, Kent <kent.glossop(a)intel.com<mailto:[email protected]>> Sent: Monday, December 4, 2017 9:42 PM To: 'anl-intel-csa(a)jlse.anl.gov' Subject: [Anl-intel-csa] CSA SDK Release Notes for 2017/12/04 CSA SDK Release Notes for 2017/12/04 version The default release has been changed to 2017-12-04-release. Relevant changes since the 2017-11-21-release: SDK Environment • The SDK environmental setup now provides a “csaenv” command. After invoking the csasetup script, the “latest” tools will be selected. The “csaenv” command can be used to show what is set, and to switch between versions if necessary. o Once csasetup.bash is sourced, user can use ‘csaenv’ to switch between environments o Examples: • `csaenv list` • Shows all available environments in system • `csaenv set fsim/2017-12-04-release tools/2017-11-29-release’ • Sets current environment with fsim built in 2017-12-04 and tools built in 2017-11-29 o Execute `csaenv help` for more details” Compiler / Runtime • Added a warning when the SPMDization region is not structured code; the code should be OpenMP-conforming (branches to or from the SPMDization region (continue, break, etc.) are illegal) • Fixes: o Add the zero trip count guard to each of the generated workers. Now, SPMDization should generate correct code when 0 <= #iterations < #workers Advisor • Introduced special mode for Dependencies analysis (-loop-call-count-limit=N, switched off by default) which helps to dramatically reduce Spatial Advisor Dependencies overhead o Example: advixe-cl --collect dependencies -filter-reductions -loops=”scalar,loop-height=0” -loop-call-count-limit=16 … • Improved SAPPC html report output and included new metrics (e.g. tripcounts, normalized total time) • Improved SAPPC performance by disabling Advisor “recommendations engine” (which is not yet used in SAPPC)

1 0

Re: [Anl-intel-csa] __builtin_csa_spmdization(k) with nested loops
by Woodworth, Daniel 18 Dec '17

18 Dec '17

You should be able to. What error message are you getting? From: anl-intel-csa [mailto:[email protected]] On Behalf Of Jennings, Elise Sent: Monday, December 18, 2017 10:56 AM To: 'anl-intel-csa(a)jlse.anl.gov' <anl-intel-csa(a)jlse.anl.gov> Subject: [Anl-intel-csa] __builtin_csa_spmdization(k) with nested loops Hi all, Can I use __builtin_csa_spmdization(k) on nested loops? Previously I used two __builtin_csa_parallel_region entry calls each with different unroll factors. When I replace both with SPMDization I get errors when compiling. If I replace only one __builtin_cas_parallel region is works fine. Thanks, Elise ________________________________ From: anl-intel-csa <anl-intel-csa-bounces(a)lists.jlse.anl.gov<mailto:[email protected]>> on behalf of Glossop, Kent <kent.glossop(a)intel.com<mailto:[email protected]>> Sent: Monday, December 4, 2017 9:42 PM To: 'anl-intel-csa(a)jlse.anl.gov' Subject: [Anl-intel-csa] CSA SDK Release Notes for 2017/12/04 CSA SDK Release Notes for 2017/12/04 version The default release has been changed to 2017-12-04-release. Relevant changes since the 2017-11-21-release: SDK Environment • The SDK environmental setup now provides a “csaenv” command. After invoking the csasetup script, the “latest” tools will be selected. The “csaenv” command can be used to show what is set, and to switch between versions if necessary. o Once csasetup.bash is sourced, user can use ‘csaenv’ to switch between environments o Examples: • `csaenv list` • Shows all available environments in system • `csaenv set fsim/2017-12-04-release tools/2017-11-29-release’ • Sets current environment with fsim built in 2017-12-04 and tools built in 2017-11-29 o Execute `csaenv help` for more details” Compiler / Runtime • Added a warning when the SPMDization region is not structured code; the code should be OpenMP-conforming (branches to or from the SPMDization region (continue, break, etc.) are illegal) • Fixes: o Add the zero trip count guard to each of the generated workers. Now, SPMDization should generate correct code when 0 <= #iterations < #workers Advisor • Introduced special mode for Dependencies analysis (-loop-call-count-limit=N, switched off by default) which helps to dramatically reduce Spatial Advisor Dependencies overhead o Example: advixe-cl --collect dependencies -filter-reductions -loops=”scalar,loop-height=0” -loop-call-count-limit=16 … • Improved SAPPC html report output and included new metrics (e.g. tripcounts, normalized total time) • Improved SAPPC performance by disabling Advisor “recommendations engine” (which is not yet used in SAPPC)

1 0

[Anl-intel-csa] __builtin_csa_spmdization(k) with nested loops
by Jennings, Elise 18 Dec '17

18 Dec '17

Hi all, Can I use __builtin_csa_spmdization(k) on nested loops? Previously I used two __builtin_csa_parallel_region entry calls each with different unroll factors. When I replace both with SPMDization I get errors when compiling. If I replace only one __builtin_cas_parallel region is works fine. Thanks, Elise ________________________________ From: anl-intel-csa <anl-intel-csa-bounces(a)lists.jlse.anl.gov> on behalf of Glossop, Kent <kent.glossop(a)intel.com> Sent: Monday, December 4, 2017 9:42 PM To: 'anl-intel-csa(a)jlse.anl.gov' Subject: [Anl-intel-csa] CSA SDK Release Notes for 2017/12/04 CSA SDK Release Notes for 2017/12/04 version The default release has been changed to 2017-12-04-release. Relevant changes since the 2017-11-21-release: SDK Environment · The SDK environmental setup now provides a “csaenv” command. After invoking the csasetup script, the “latest” tools will be selected. The “csaenv” command can be used to show what is set, and to switch between versions if necessary. o Once csasetup.bash is sourced, user can use ‘csaenv’ to switch between environments o Examples: § `csaenv list` · Shows all available environments in system § `csaenv set fsim/2017-12-04-release tools/2017-11-29-release’ · Sets current environment with fsim built in 2017-12-04 and tools built in 2017-11-29 o Execute `csaenv help` for more details” Compiler / Runtime · Added a warning when the SPMDization region is not structured code; the code should be OpenMP-conforming (branches to or from the SPMDization region (continue, break, etc.) are illegal) · Fixes: o Add the zero trip count guard to each of the generated workers. Now, SPMDization should generate correct code when 0 <= #iterations < #workers Advisor · Introduced special mode for Dependencies analysis (-loop-call-count-limit=N, switched off by default) which helps to dramatically reduce Spatial Advisor Dependencies overhead o Example: advixe-cl --collect dependencies -filter-reductions -loops=”scalar,loop-height=0” -loop-call-count-limit=16 … · Improved SAPPC html report output and included new metrics (e.g. tripcounts, normalized total time) · Improved SAPPC performance by disabling Advisor “recommendations engine” (which is not yet used in SAPPC)

1 0

[Anl-intel-csa] Canceled: ANL/Intel CSA Workload Characterization
by Petrini, Fabrizio 14 Dec '17

14 Dec '17

Placeholder for weekly ANL/Intel CSA Exascale workload characterization meeting ......................................................................................................................................... --> Join Skype Meeting<https://meet.intel.com/fabrizio.petrini/7K7FM32F> Trouble Joining? Try Skype Web App<https://meet.intel.com/fabrizio.petrini/7K7FM32F?sl=1> Join by phone +1(916)356-2663 (or your local bridge access #) Choose bridge 5.<tel:+1(916)356-2663%20(or%20your%20local%20bridge%20access%20#)%20Choose%20bridge%205.> (Global) English (United States) Find a local number<https://dial.intel.com> Conference ID: 791574018 Forgot your dial-in PIN?<https://dial.intel.com> |Help<https://o15.officeredir.microsoft.com/r/rlidLync15?clid=1033&p1=5&p2=2009> [!OC([1033])!] .........................................................................................................................................

1 0