[PATCH] [cuda] Driver changes to build and stitch together host and device-side CUDA code.

[Artem Belevich]

In http://reviews.llvm.org/D9509#167188, @echristo wrote:

> Not really happy with the getTargetToolChain changes. Can you go ahead and revisit the necessity of those (and probably the naming of the StringRef argument as well)?


getToolChain does two things -- calculates a triple based on default triple, command line args and optional DarwinArch and then selects a toolchain based on the triple. I needed toolchain selection by triple, so I've extracted it into getTargetToolChain.

Alternative approach would be to extend computeTargetTriple so that it can figure out that I need to get a NVPTX triple based on DarwinArchName (renamed to ArchName?). IMO it's not as clean as using getTargetToolchain(Triple) considering that we already know the triple.

I've changed the code so it no longer uses computeTargetTriple() directly (and removed forward declaration), but kept getTargetToolchain().


http://reviews.llvm.org/D9509

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/