[llvm] [LoopIdiomRecognize] Enable clmul optimization for CRC loops (PR #203405)

[Sean Clarke via llvm-commits]

================
@@ -1572,6 +1580,156 @@ bool LoopIdiomRecognize::avoidLIRForMultiBlockLoop(bool IsMemset,
 }
 
 bool LoopIdiomRecognize::optimizeCRCLoop(const PolynomialInfo &Info) {
+  if (ForceCRCClmul)
+    return optimizeCRCLoopUsingClmul(Info) ||
+           (!ApplyCodeSizeHeuristics && optimizeCRCLoopUsingTableLookup(Info));
+
+  // FIXME: Once intrinsic cost modeling is more reliable for clmul, that should
+  // be used to determine which optimization to use. Until then, only apply the
+  // clmul optimization when optimizing for size, since a lookup table is not
+  // viable in that case. On some platforms, TC=8 for clmul seems to be slower
+  // than even the unoptimized loop, so bail on that case as well.
+  return ApplyCodeSizeHeuristics
+             ? Info.TripCount > 8 && optimizeCRCLoopUsingClmul(Info)
+             : optimizeCRCLoopUsingTableLookup(Info);
----------------
xarkenz wrote:

The dip in the CRC-64 TC=32 case should be remediated by https://github.com/llvm/llvm-project/pull/209265. That doesn't affect the mergeability of this PR, though-- for now, this optimization only applies when the table optimization cannot be applied.

https://github.com/llvm/llvm-project/pull/203405