[llvm-bugs] [Bug 43505] New: PTX code generation with cuda10: shfl without .sync is not supported

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=43505

            Bug ID: 43505
           Summary: PTX code generation with cuda10: shfl without .sync is
                    not supported
           Product: clang
           Version: 9.0
          Hardware: Other
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: CUDA
          Assignee: unassignedclangbugs at nondot.org
          Reporter: smithc11 at rpi.edu
                CC: llvm-bugs at lists.llvm.org

Created attachment 22605
  --> https://bugs.llvm.org/attachment.cgi?id=22605&action=edit
tarball with source code, compilation instructions, and temporary files

Overview:

  Using Clang 9.0.0 for PTX code generation from Thrust source code results in
the following error with Cuda 10.1.243:

ptxas /var/tmp/testcase-min-af0932.s, line 4521; error   : Instruction 'shfl'
without '.sync' is not supported on .target sm_70 and higher from PTX ISA
version 6.4

Steps to Reproduce: 

   Compile the 'attached testcase-min.cu' with the following command:

   $ clang++ -O0 --cuda-gpu-arch=sm_70 testcase-min.cu

Actual Results: 

   The following error is output during the build:

ptxas /var/tmp/testcase-min-af0932.s, line 4521; error   : Instruction 'shfl'
without '.sync' is not supported on .target sm_70 and higher from PTX ISA
version 6.4
...
ptxas fatal   : Ptx assembly aborted due to errors
clang-9: error: ptxas command failed with exit code 255 (use -v to see
invocation)

Expected Results: 

    A working binary.

Build Date & Hardware: 

    System: LLNL Lassen system; 2x IBM Power9 host processors + 4x Nvidia V100
per node

$ clang++ --version
clang version 9.0.0 (/builddir/build/BUILD/ibm-llvm/tools/clang
63a7d47678dad8b206a08bdfa9380ebdb147e888) (/builddir/build/BUILD/ibm-llvm
d99a7ea8cd2b634d0dcb13c44d06c4bdd4436c4e)
Target: powerpc64le-unknown-linux-gnu
Thread model: posix
InstalledDir: /usr/tce/packages/clang/clang-upstream-2019.08.15/release/bin

$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Sun_Jul_28_19:07:52_PDT_2019
Cuda compilation tools, release 10.1, V10.1.243

$ echo $CUDA_HOME 
/usr/tce/packages/cuda/cuda-10.1.243

Additional Builds and Platforms: 

    I did not compile the problematic code with Clang on another Volta equipped
system.

Additional Information: 

    Temporary files generated with '-save-temps' are included in the attached
tarball.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20190930/6e6daf16/attachment.html>