[libomptarget] Run GPU offloading tests on both new and old runtime

[JonChesterfield]

Summary

Implemented by patching python config instead of modifying all
the tests so that -generic and XFAIL work as usual. Expectation is for
this to be reverted once the old runtime is deleted.

[JonChesterfield]

As it stands, this would double the number of offloading tests run on the cuda buildbot

[Meinersbur]

Do you know whether all current regression tests are passing with the newRTL?

openmp/libomptarget/plugins/cuda/CMakeLists.txt

68–69

[JonChesterfield]

They definitely don't on amdgpu. Cuda I assume they are, my local nvidia box is still building clang to confirm. I think they were passing on nvptx a couple of days ago.

I get a failure on sm_50:

$HOME/llvm-project/openmp/libomptarget/test/unified_shared_memory/api.c:18:22: error: Target architecture sm_50 does not support unified addressing
#pragma omp requires unified_shared_memory

which presumably means I've missed whatever should have disabled that test on my machine

edit: patch was fine, test in question doesn't work on the old devicertl either. Copied the xfail.

[JonChesterfield]

• add xfail for new cuda target

[JonChesterfield]

• review comment

openmp/libomptarget/plugins/cuda/CMakeLists.txt

68–69

better, thanks!

[Meinersbur]

LGTM

[JonChesterfield]

openmp/libomptarget/plugins/cuda/CMakeLists.txt

68–69

this ^ turned out to be a functional change. As originally written, it would only run the newRTL cases. With this form, both run

openmp/libomptarget/CMakeLists.txt

 # LLVM source tree is required at build time for libomptarget
 if (NOT LIBOMPTARGET_LLVM_INCLUDE_DIRS)
   message(FATAL_ERROR "Missing definition for LIBOMPTARGET_LLVM_INCLUDE_DIRS")
 endif()
 
 # This is a list of all the targets that are supported/tested right now.
 set (LIBOMPTARGET_ALL_TARGETS "${LIBOMPTARGET_ALL_TARGETS} aarch64-unknown-linux-gnu")
 set (LIBOMPTARGET_ALL_TARGETS "${LIBOMPTARGET_ALL_TARGETS} amdgcn-amd-amdhsa")
+set (LIBOMPTARGET_ALL_TARGETS "${LIBOMPTARGET_ALL_TARGETS} amdgcn-amd-amdhsa-newRTL")
 set (LIBOMPTARGET_ALL_TARGETS "${LIBOMPTARGET_ALL_TARGETS} powerpc64le-ibm-linux-gnu")
 set (LIBOMPTARGET_ALL_TARGETS "${LIBOMPTARGET_ALL_TARGETS} powerpc64-ibm-linux-gnu")
 set (LIBOMPTARGET_ALL_TARGETS "${LIBOMPTARGET_ALL_TARGETS} x86_64-pc-linux-gnu")
 set (LIBOMPTARGET_ALL_TARGETS "${LIBOMPTARGET_ALL_TARGETS} nvptx64-nvidia-cuda")
+set (LIBOMPTARGET_ALL_TARGETS "${LIBOMPTARGET_ALL_TARGETS} nvptx64-nvidia-cuda-newRTL")
 
 # Once the plugins for the different targets are validated, they will be added to
 # the list of supported targets in the current system.
 set (LIBOMPTARGET_SYSTEM_TARGETS "")
 set (LIBOMPTARGET_TESTED_PLUGINS "")
 
 # Check OMPT support
 set(OMPT_DEFAULT_IN_LIBOMPTARGET FALSE)

openmp/libomptarget/plugins/amdgpu/CMakeLists.txt

execute_process(COMMAND "${AMDGPU_ARCH_TOOL}" RESULT_VARIABLE amdgpu_arch_result
   OUTPUT_VARIABLE amdgpu_arch_output)
 if (${amdgpu_arch_result})
   libomptarget_say("Not generating amdgcn test targets as amdgpu-arch exited with ${amdgpu_arch_result}")
   return()
 endif()
 
 # Report to the parent scope that we are building a plugin for amdgpu
 set(LIBOMPTARGET_SYSTEM_TARGETS "${LIBOMPTARGET_SYSTEM_TARGETS} amdgcn-amd-amdhsa" PARENT_SCOPE)
-

openmp/libomptarget/plugins/cuda/CMakeLists.txt

target_link_libraries(omptarget.rtl.cuda
   "-Wl,-z,defs")
 
 # Report to the parent scope that we are building a plugin for CUDA.
 # This controls whether tests are run for the nvptx offloading target
 # Run them if libcuda is available, or if the user explicitly asked for dlopen
 # Otherwise this plugin is being built speculatively and there may be no cuda available
 if (LIBOMPTARGET_CAN_LINK_LIBCUDA OR LIBOMPTARGET_FORCE_DLOPEN_LIBCUDA)
   libomptarget_say("Enable tests using CUDA plugin")
-  set(LIBOMPTARGET_SYSTEM_TARGETS "${LIBOMPTARGET_SYSTEM_TARGETS} nvptx64-nvidia-cuda" PARENT_SCOPE)
+  set(LIBOMPTARGET_SYSTEM_TARGETS "${LIBOMPTARGET_SYSTEM_TARGETS} nvptx64-nvidia-cuda nvptx64-nvidia-cuda-newRTL" PARENT_SCOPE)
 else()
   libomptarget_say("Disabling tests using CUDA plugin as cuda may not be available")
 endif()

openmp/libomptarget/test/lit.cfg

else: # Unices
     config.test_flags += " -Wl,-rpath," + config.library_dir
     config.test_flags += " -Wl,-rpath," + config.omp_host_rtl_directory
     if config.cuda_libdir:
         config.test_flags += " -Wl,-rpath," + config.cuda_libdir
     if config.libomptarget_current_target.startswith('amdgcn'):
         config.test_flags += " --libomptarget-amdgcn-bc-path=" + config.library_dir
     if config.libomptarget_current_target.startswith('nvptx'):
         config.test_flags += " --libomptarget-nvptx-bc-path=" + config.library_dir
+    if config.libomptarget_current_target.endswith('-newRTL'):
+        config.test_flags += " -fopenmp-target-new-runtime"
 
 # substitutions
 # - for targets that exist in the system create the actual command.
 # - for valid targets that do not exist in the system, return false, so that the
 #   same test can be used for different targets.
 
 # Scan all the valid targets.
 for libomptarget_target in config.libomptarget_all_targets:

openmp/libomptarget/test/unified_shared_memory/api.c

 // RUN: %libomptarget-compile-run-and-check-generic
 // XFAIL: nvptx64-nvidia-cuda
+// XFAIL: nvptx64-nvidia-cuda-newRTL
 
 // Fails on amdgcn with error: GPU Memory Error
 // XFAIL: amdgcn-amd-amdhsa
 
 #include <stdio.h>
 #include <omp.h>
 
 // ---------------------------------------------------------------------------