[OpenMP] Make bc file compilation sensitive to LIBOMPTARGET_NVPTX_DEBUG flag

[guansong]

Summary

The LIBOMPTARGET_NVPTX_DEBUG flag is inconsistent between using nvcc to generate .a file and clang to generate .bc file. Sync the two setting so we can get debug messages from the bc file path as well.

[Hahnfeld]

libomptarget/deviceRTLs/nvptx/CMakeLists.txt

146

Hmm, why does Sync mean a new variable? If that's what you want, it should start with LIBOMPTARGET_NVPTX and be properly documented

[guansong]

libomptarget/deviceRTLs/nvptx/CMakeLists.txt

146

Thanks. I put a typo there, it should be

set(BC_DEBUG FALSE CACHE BOOL

As I used in the next few lines. BC_DEBUG is an internal var. The behavior to sync is the Cmake command line def LIBOMPTARGET_NVPTX_DEBUG

[guansong]

libomptarget/deviceRTLs/nvptx/CMakeLists.txt

146

Look at this again, I am thinking I don't need these few lines. The BC_DEBUG will be defined either way in the following if block.

[Hahnfeld]

libomptarget/deviceRTLs/nvptx/CMakeLists.txt

146

In that case it must not be CACHE. You can define BC_DEBUG to -DOMPTARGET_NVPTX_DEBUG=0 and overwrite it if(${LIBOMPTARGET_NVPTX_DEBUG})

[guansong]

update according to suggestion, thanks.

[grokos]

I think this is now good to go.

libomptarget/deviceRTLs/nvptx/CMakeLists.txt

if(LIBOMPTARGET_NVPTX_SELECTED_CUDA_COMPILER AND LIBOMPTARGET_NVPTX_SELECTED_BC_LINKER)
 
       # Decide which ptx version to use. Same choices as Clang.
       if(CUDA_VERSION_MAJOR GREATER 9 OR CUDA_VERSION_MAJOR EQUAL 9)
         set(CUDA_PTX_VERSION ptx60)
       else()
         set(CUDA_PTX_VERSION ptx42)
       endif()
  
+      # Activate RTL message dumps if requested by the user.
+      set(LIBOMPTARGET_BC_DEBUG FALSE CACHE BOOL
+        "Activate BC code device RTL debug messages.")
+      if(${LIBOMPTARGET_NVPTX_DEBUG})
+        set(BC_DEBUG -DOMPTARGET_NVPTX_DEBUG=-1)
+      else()
+        set(BC_DEBUG -DOMPTARGET_NVPTX_DEBUG=0)
+      endif()
+
       # Set flags for Clang cuda compilation.  Only Clang is supported because there is
       # no other compiler capable of generating bitcode from cuda sources.
       set(CUDA_FLAGS
           -emit-llvm
           -fcuda-rdc
           -O1
           -Xclang -target-feature
           -Xclang +${CUDA_PTX_VERSION}
           --cuda-device-only
-          -DOMPTARGET_NVPTX_TEST=0 -DOMPTARGET_NVPTX_DEBUG=0
+          -DOMPTARGET_NVPTX_TEST=0
+          ${BC_DEBUG}
       )
 
       # CUDA 9 header files use the nv_weak attribute which clang is not yet prepared
       # to handle. Therefore, we use 'weak' instead. We are compiling only for the
       # device, so it should be equivalent.
       if(CUDA_VERSION_MAJOR EQUAL 9)
         set(CUDA_FLAGS ${CUDA_FLAGS} -Dnv_weak=weak)
       endif()