[llvm] [AMDGPU] Legalize 64bit elements for BUILD_VECTOR on gfx942 (PR #145052)

[Janek van Oirschot via llvm-commits]

================
@@ -15796,6 +15814,27 @@ SITargetLowering::performExtractVectorEltCombine(SDNode *N,
     }
   }
 
+  // if PeekThoughBitcast(Vec)[MapIdx(CIdx)] == undef &&
----------------
JanekvO wrote:

Should this be split off to another PR (and into generic combiner)? AFAICT, generic combiner does not cover this

https://github.com/llvm/llvm-project/pull/145052