[CUDA] update default CUDA sm ver to 75 (#20824)

This pull request updates the default NVIDIA GPU architecture used for
CUDA from `SM_50` to `SM_75`, as the newest CUDA 13 no longer supports
`SM_50`.

It also updates several tests:
* sets CUDA 10.0 as the default toolkit version used in driver-detection
tests
* bumps the SM version to 75 and the PTX version to 63 for tests that
look for specific output patterns
* adds support for the 3-operand atomic intrinsic
* adds support for the tanh.approx.f16/f16x2 intrinsic

Author: bratpiorka

clang/include/clang/Basic/OffloadArch.h

@@ -167,7 +167,7 @@ enum class OffloadArch {
   LNL_M,
   LAST,
 
-  CudaDefault = OffloadArch::SM_52,
+  CudaDefault = OffloadArch::SM_75,
   HIPDefault = OffloadArch::GFX906,
 };
 

clang/lib/Driver/Driver.cpp

@@ -5977,11 +5977,11 @@ class OffloadingActionBuilder final {
 
       // Handle defaults architectures
       for (auto &Triple : SYCLTripleList) {
-        // For NVIDIA use SM_50 as a default
+        // For NVIDIA use SM_75 as a default
         if (Triple.isNVPTX() && llvm::none_of(GpuArchList, [&](auto &P) {
               return P.first.isNVPTX();
             })) {
-          const char *DefaultArch = OffloadArchToString(OffloadArch::SM_50);
+          const char *DefaultArch = OffloadArchToString(OffloadArch::SM_75);
           GpuArchList.emplace_back(Triple, DefaultArch);
         }
 
@@ -7696,7 +7696,7 @@ Driver::getOffloadArchs(Compilation &C, const llvm::opt::DerivedArgList &Args,
       // The default arch is set for NVPTX if not provided.  For AMDGPU, emit
       // an error as the user is responsible to set the arch.
       if (TC.getTriple().isNVPTX())
-        Archs.insert(OffloadArchToString(OffloadArch::SM_50));
+        Archs.insert(OffloadArchToString(OffloadArch::SM_75));
       else if (TC.getTriple().isAMDGPU())
         C.getDriver().Diag(clang::diag::err_drv_sycl_missing_amdgpu_arch)
             << 1 << TC.getTriple().str();

clang/test/Driver/Inputs/CUDA-windows/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.0/bin/.keep

@@ -0,0 +1,2 @@
+CUDA Version 10.0.130
+