[llvm] [X86] Convert FP compares split by a nested logic op to FP logic (PR #226222)

[via llvm-commits]

================
@@ -52658,9 +52658,28 @@ static unsigned convertIntLogicToFPLogicOpcode(unsigned Opcode) {
   return FPOpcode;
 }
 
+/// Return true if \p V is a tree of single-use vector logic ops over vector
+/// compares of scalar FP values, as created by convertIntLogicToFPLogic. Limit
+/// the depth so that, after one more logic op is added,
+/// SimplifyDemandedVectorElts on the final extract still reaches every compare:
+/// otherwise the X86ISD::CMPM mask compares are not scalarized and are widened
+/// to 512 bits without AVX512VL.
----------------
Andarwinux wrote:

According to SDM, vcmpsh doesn't need VLX? Is this a bug?

https://intel.github.io/SDM/instruction/VCMPSH.html

https://github.com/llvm/llvm-project/pull/226222