[clang] [CUDA] Add support for sm101 and sm120 target architectures (PR #127187)

[Artem Belevich via cfe-commits]

================
@@ -300,6 +306,10 @@ void NVPTXTargetInfo::getTargetDefines(const LangOptions &Opts,
       Builder.defineMacro("__CUDA_ARCH_FEAT_SM90_ALL", "1");
     if (GPU == OffloadArch::SM_100a)
       Builder.defineMacro("__CUDA_ARCH_FEAT_SM100_ALL", "1");
+    if (GPU == OffloadArch::SM_101a)
+      Builder.defineMacro("__CUDA_ARCH_FEAT_SM101_ALL", "1");
+    if (GPU == OffloadArch::SM_120a)
+      Builder.defineMacro("__CUDA_ARCH_FEAT_SM120_ALL", "1");
----------------
Artem-B wrote:

Right. We need to trim the last digit. A bit shorter way to do it would be `StringRef(CUDAArchCode).drop_back()` 
Actually, we can change return type of the lambda to `StringRef` which will simplify it to just `CUDAArchCode.drop_back()`



https://github.com/llvm/llvm-project/pull/127187