[offload] - Add mainline build support for gf950 (llvm#1244)

Author: Salinas, David

offload/hostexec/CMakeLists.txt

@@ -126,6 +126,29 @@ else()
   return()
 endif()
 
+set(amdgpu_mcpus gfx700 gfx701 gfx801 gfx803 gfx900 gfx902 gfx906 gfx908 gfx90a gfx90c gfx940 gfx941 gfx942 gfx950 gfx1010 gfx1030 gfx1031 gfx1032 gfx1033 gfx1034 gfx1035 gfx1036 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx1200 gfx1201)
+if (DEFINED LIBOMPTARGET_AMDGCN_GFXLIST)
+  set(amdgpu_mcpus ${LIBOMPTARGET_AMDGCN_GFXLIST})
+endif()
+
+set(all_capabilities 35 37 50 52 53 60 61 62 70 72 75 80 86 89 90)
+set(LIBOMPTARGET_NVPTX_COMPUTE_CAPABILITIES ${all_capabilities} CACHE STRING
+  "List of CUDA Compute Capabilities to be used to compile the NVPTX DeviceRTL.")
+string(TOLOWER ${LIBOMPTARGET_NVPTX_COMPUTE_CAPABILITIES} LIBOMPTARGET_NVPTX_COMPUTE_CAPABILITIES)
+if (LIBOMPTARGET_NVPTX_COMPUTE_CAPABILITIES STREQUAL "all")
+  set(nvptx_sm_list ${all_capabilities})
+elseif(LIBOMPTARGET_NVPTX_COMPUTE_CAPABILITIES STREQUAL "auto")
+  if (NOT LIBOMPTARGET_DEP_CUDA_FOUND)
+    libomptarget_error_say("[NVPTX] Cannot auto detect compute capability as CUDA not found.")
+  endif()
+  set(nvptx_sm_list ${LIBOMPTARGET_DEP_CUDA_ARCH})
+else()
+  string(REPLACE "," ";" nvptx_sm_list "${LIBOMPTARGET_NVPTX_COMPUTE_CAPABILITIES}")
+endif()
+foreach(sm ${nvptx_sm_list})
+   list(APPEND nvptx_mcpus "sm_${sm}")
+endforeach()
+
 set(ocl_atomics_cl_filename ${CMAKE_CURRENT_SOURCE_DIR}/src/oclAtomics.cl)
 set(invoke_cpp_file_name ${CMAKE_CURRENT_SOURCE_DIR}/src/hostexec_invoke.cpp)
 set(hostexec_stubs_filename ${CMAKE_CURRENT_SOURCE_DIR}/src/hostexec_stubs.cpp)

offload/test/lit.cfg

@@ -134,7 +134,8 @@ elif config.libomptarget_current_target.startswith('amdgcn'):
     # only check the first one assuming that we will run the test on it.
     if not (config.amdgpu_test_arch.startswith("gfx90a") or
             config.amdgpu_test_arch.startswith("gfx940") or
-            config.amdgpu_test_arch.startswith("gfx942")):
+            config.amdgpu_test_arch.startswith("gfx942") or
+            config.amdgpu_test_arch.startswith("gfx950")):
        supports_unified_shared_memory = False
     # check if AMD architecture is an APU:
     if (config.amdgpu_test_arch.startswith("gfx940") or