[clang] [llvm] [clang][NVPTX] Add intrinsics and builtins for CVT RS rounding mode (PR #160494)

[Srinivasa Ravi via llvm-commits]

================
@@ -1203,6 +1209,83 @@ __device__ void nvvm_cvt_sm100a_sm101a_sm120a() {
   // CHECK: ret void
 }
 
+__device__ void nvvm_cvt_sm100a_sm103a() {
+#if (PTX >= 87) && (__CUDA_ARCH_FEAT_SM100_ALL || __CUDA_ARCH_FEAT_SM103_ALL)
+  
+// CHECK_PTX87_SM100a: call <2 x half> @llvm.nvvm.ff2f16x2.rs(float 1.000000e+00, float 1.000000e+00, i32 0)
+// CHECK_PTX87_SM103a: call <2 x half> @llvm.nvvm.ff2f16x2.rs(float 1.000000e+00, float 1.000000e+00, i32 0)
+  __nvvm_ff2f16x2_rs(1.0f, 1.0f, 0);
----------------
Wolfram70 wrote:

Added assignment to these tests in the latest revision. Please take a look, thanks!

https://github.com/llvm/llvm-project/pull/160494