[llvm] [X86] Avoid shl->mul lowbit-isolate fold on Zen when unprofitable (PR #217776)

[via llvm-commits]

================
@@ -832,6 +832,12 @@ def TuningPreferMaskRegisters
                        "Prefer AVX512 mask registers over PTEST/MOVMSK",
                        [], InlineIgnore>;
 
+def TuningPreferShlOverBlsiMul
+    : SubtargetFeature<"prefer-shl-over-blsi-mul", "PreferShlOverBlsiMul", "true",
----------------
Andarwinux wrote:

Personally, I think this is unnecessary.

https://godbolt.org/z/qnMhv3TEn

Since these two sequences show identical performance on Intel, to ensure this fold is as active as possible in real-world, it should be unconditionally.

https://github.com/llvm/llvm-project/pull/217776