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

[Harald van Dijk via llvm-commits]

================
@@ -3590,6 +3618,17 @@ FPMinMaxPat<Instruction minmaxInst, ValueType vt, SDPatternOperator min_or_max,
               DSTCLAMP.NONE, DSTOMOD.NONE)
 >;
 
+class
+FPMinCanonMaxPat<Instruction minmaxInst, ValueType vt, SDPatternOperator min_or_max,
+                 SDPatternOperator max_or_min_oneuse> : GCNPat <
+  (min_or_max (is_canonicalized_1<fcanonicalize>
----------------
hvdijk wrote:

There may well be more, I just went for the ones that I came across from the test results. But if there are more, it doesn't result in wrong code, only suboptimal code, which I think is acceptable if the overall code still looks better?

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