[llvm] [clang] [CUDA] Add support for CUDA-12.3 and sm_90a (PR #74895)

[Artem Belevich via cfe-commits]

================
@@ -80,8 +85,10 @@ class NVPTXSubtarget : public NVPTXGenSubtargetInfo {
   bool allowFP16Math() const;
   bool hasMaskOperator() const { return PTXVersion >= 71; }
   bool hasNoReturn() const { return SmVersion >= 30 && PTXVersion >= 64; }
-  unsigned int getSmVersion() const { return SmVersion; }
+  unsigned int getSmVersion() const { return FullSmVersion / 10; }
+  unsigned int getFullSmVersion() const { return FullSmVersion; }
   std::string getTargetName() const { return TargetName; }
+  bool isSm90a() const { return getFullSmVersion() == 901; }
----------------
Artem-B wrote:

According to [CUDA docs](docs.nvidia.com/cuda/parallel-thread-execution/index.html?highlight=sm_90a#ptx-module-directives-target)

> Target architectures with suffix “a”, such as sm_90a, include architecture-accelerated features that are supported on the specified architecture only, hence such targets do not follow the onion layer model. Therefore, PTX code generated for such targets cannot be run on later generation devices. Architecture-accelerated features can only be used with targets that support these features.

It's not clear where they are going with this approach.

I can make it a more generic `int hasAAFeatures() { return FullSmVersion % 10;  }` if that's what you're looking for.


https://github.com/llvm/llvm-project/pull/74895