[PATCH] D43461: [CUDA] Include single GPU binary, NFCI.

[Artem Belevich via Phabricator via cfe-commits]

tra added inline comments.


================
Comment at: lib/Driver/ToolChains/Clang.cpp:4659
   if (IsCuda) {
-    // Host-side cuda compilation receives device-side outputs as Inputs[1...].
-    // Include them with -fcuda-include-gpubinary.
+    // Host-side cuda compilation receives device-side outputs as Inputs[1].
+    // Include the binary with -fcuda-include-gpubinary.
----------------
Nit: Passing multiple things as a single input may need some more details.
E.g. `...receives all device-side outputs in a single fatbin as Inputs[1]`


================
Comment at: lib/Frontend/CompilerInvocation.cpp:1044-1045
 
-  Opts.CudaGpuBinaryFileNames =
-      Args.getAllArgValues(OPT_fcuda_include_gpubinary);
+  Opts.CudaGpuBinaryFileName =
+      Args.getLastArgValue(OPT_fcuda_include_gpubinary);
 
----------------
If more than one gpu binary is passed, all but last will be ignored.
IMO in this case we would want to either warn that some inputs were ignored or report an error that there is more than one GPU binary.


Repository:
  rC Clang

https://reviews.llvm.org/D43461