[OpenMP] NVPTX: Set default/minimum compute capability to sm_35

[grokos]

Summary

The current implementation of the nvptx runtime (to be upstreamed shortly) uses the atomicMax operation on 64-bit integers. This is only supported in compute capabilities 3.5 and later. I've changed the clang default to sm_35.

[guansong]

LGTM.

CMakeLists.txt

 endif()
 
 set(CLANG_DEFAULT_OBJCOPY "objcopy" CACHE STRING
   "Default objcopy executable to use.")
 
 set(CLANG_DEFAULT_OPENMP_RUNTIME "libomp" CACHE STRING
   "Default OpenMP runtime used by -fopenmp.")
 
-# OpenMP offloading requires at least sm_30 because we use shuffle instructions
-# to generate efficient code for reductions.
-set(CLANG_OPENMP_NVPTX_DEFAULT_ARCH "sm_30" CACHE STRING
+# OpenMP offloading requires at least sm_35 because we use shuffle instructions
+# to generate efficient code for reductions and the atomicMax instruction on
+# 64-bit integers in the implementation of conditional lastprivate.
+set(CLANG_OPENMP_NVPTX_DEFAULT_ARCH "sm_35" CACHE STRING
   "Default architecture for OpenMP offloading to Nvidia GPUs.")
 string(REGEX MATCH "^sm_([0-9]+)$" MATCHED_ARCH "${CLANG_OPENMP_NVPTX_DEFAULT_ARCH}")
-if (NOT DEFINED MATCHED_ARCH OR "${CMAKE_MATCH_1}" LESS 30)
-  message(WARNING "Resetting default architecture for OpenMP offloading to Nvidia GPUs to sm_30")
-  set(CLANG_OPENMP_NVPTX_DEFAULT_ARCH "sm_30" CACHE STRING
+if (NOT DEFINED MATCHED_ARCH OR "${CMAKE_MATCH_1}" LESS 35)
+  message(WARNING "Resetting default architecture for OpenMP offloading to Nvidia GPUs to sm_35")
+  set(CLANG_OPENMP_NVPTX_DEFAULT_ARCH "sm_35" CACHE STRING
     "Default architecture for OpenMP offloading to Nvidia GPUs." FORCE)
 endif()
 
 set(CLANG_VENDOR ${PACKAGE_VENDOR} CACHE STRING
   "Vendor-specific text for showing with version information.")
 
 if( CLANG_VENDOR )
   add_definitions( -DCLANG_VENDOR="${CLANG_VENDOR} " )