[llvm] [LLVM] Update CUDA ELF flags for their new ABI (PR #149534)

[Joseph Huber via llvm-commits]

================
@@ -295,7 +295,8 @@ TEST(ELFObjectFileTest, CheckOSAndTriple) {
       {ELF::EM_X86_64, ELF::ELFOSABI_AIX, "x86_64--aix"},
       {ELF::EM_X86_64, ELF::ELFOSABI_FREEBSD, "x86_64--freebsd"},
       {ELF::EM_X86_64, ELF::ELFOSABI_OPENBSD, "x86_64--openbsd"},
-      {ELF::EM_CUDA, ELF::ELFOSABI_CUDA, "nvptx64-nvidia-cuda"}};
+      {ELF::EM_CUDA, ELF::ELFOSABI_CUDA, "nvptx64-nvidia-cuda"},
+      {ELF::EM_CUDA, ELF::ELFOSABI_CUDA_V2, "nvptx64-nvidia-cuda"}};
----------------
jhuber6 wrote:

It's definitely not going to work if we use a separate triple, that'd break a lot of crap. The thing we key off of for this ABI is the processor architecture, which is in the ELF flags. I really, really want to know why NVIDIA changed it like this.

https://github.com/llvm/llvm-project/pull/149534