[llvm] [AMDGPU] Fix canonicalization of truncated values. (PR #83054)

[Harald van Dijk via llvm-commits]

================
@@ -26258,6 +26258,24 @@ SDValue DAGCombiner::visitFP_TO_FP16(SDNode *N) {
   if (N0->getOpcode() == ISD::FP16_TO_FP)
     return N0->getOperand(0);
 
+  // fold (fp_to_fp16 (freeze (fp16_to_fp (fp_to_fp16 op))))
----------------
hvdijk wrote:

Let's just bite the bullet and do it, I guess. Some of the regressions are easily solved and I have included them in the same commit. Others are not but the degradations that remain actually seem like they are outweighed by better `fcanonicalize` optimisations elsewhere.

https://github.com/llvm/llvm-project/pull/83054