[llvm] [NVPTX] Add support for scaled::n1::ue8m0 in FP8, FP6 and FP4 conversions (PR #227652)

[Durgadoss R via llvm-commits]

================
@@ -3182,6 +3195,9 @@ foreach dst = ["e2m3x2", "e3m2x2"] in {
       defvar F32X2    = ToCvtIntrinsics<"ff_to_" # Suffix>;
       defvar F16X2    = ToCvtIntrinsics<"f16x2_to_" # Suffix>;
       defvar BF16X2   = ToCvtIntrinsics<"bf16x2_to_" # Suffix>;
+      defvar F32Scale = !cast<Intrinsic>("int_nvvm_ff_to_" # Suffix # "_satfinite_scale_n1_ue8m0");
+      defvar F16Scale = !cast<Intrinsic>("int_nvvm_f16x2_to_" # Suffix # "_satfinite_scale_n1_ue8m0");
+      defvar BF16Scale = !cast<Intrinsic>("int_nvvm_bf16x2_to_" # Suffix # "_satfinite_scale_n1_ue8m0");
----------------
durga4github wrote:

Using `ToCvtIntrinsics` here does not work..?
(same question for line 3264 below...)

https://github.com/llvm/llvm-project/pull/227652