[llvm] [AArch64][CostModel] Increase the cost of illegal SVE int-to-fp converts (PR #130756)

[David Green via llvm-commits]

================
@@ -2811,6 +2811,19 @@ InstructionCost AArch64TTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
             BF16Tbl, ISD, DstTy.getSimpleVT(), SrcTy.getSimpleVT()))
       return AdjustCost(Entry->Cost);
 
+  // Symbolic constants for the SVE sitofp/uitofp entries in the table below
+  // The cost of unpacking twice is artificially increased for now in order
+  // to avoid regressions against NEON, which will use tbl instructions directly
+  // instead of multiple layers of [s|u]unpk[lo|hi].
+  // We use the unpacks in cases where the destination type is illegal and
+  // requires splitting of the input, even if the input type itself is legal.
+  // FIXME: Use tbl instructions for SVE as well, at least in cases where the
+  //        conversion is done in a loop.
----------------
davemgreen wrote:

I would remove this FIXME - mostly as it just doesn't feel like the right place for it and I think if we supported tbl extensions that would only apply to loops, and the cost model should probably be the worst-case of with and without tbl. (There is a chance we want to change that in the future to have something that can cost "invariant" vs "fixed" costs, similar for constants, but for the moment it is probably OK to stick with the higher).

https://github.com/llvm/llvm-project/pull/130756