[X86] Prefer KORTEST on Knights Landing or later for memcmp()

PTEST and especially the MOVMSK instructions are slow on Knights Landing
or later. As a bonus, this patch increases instruction parallelism by
emitting:
    KORTEST(PCMPNEQ(a, b), PCMPNEQ(c, d)) == 0
Instead of:
    KORTEST(AND(PCMPEQ(a, b), PCMPEQ(c, d))) == ~0

https://reviews.llvm.org/D69157

1 files changed, 4 insertions, 0 deletions

diff --git a/llvm/lib/Target/X86/X86Subtarget.h b/llvm/lib/Target/X86/X86Subtarget.h
index e8efe8f2afe..e77b64a743a 100644
--- a/llvm/lib/Target/X86/X86Subtarget.h
+++ b/llvm/lib/Target/X86/X86Subtarget.h
@@ -445,6 +445,9 @@ protected:
   /// Indicates target prefers 256 bit instructions.
   bool Prefer256Bit = false;
 
+  /// Indicates target prefers AVX512 mask registers.
+  bool PreferMaskRegisters = false;
+
   /// Threeway branch is profitable in this subtarget.
   bool ThreewayBranchProfitable = false;
 
@@ -706,6 +709,7 @@ public:
     return UseRetpolineIndirectBranches;
   }
   bool useRetpolineExternalThunk() const { return UseRetpolineExternalThunk; }
+  bool preferMaskRegisters() const { return PreferMaskRegisters; }
 
   unsigned getPreferVectorWidth() const { return PreferVectorWidth; }
   unsigned getRequiredVectorWidth() const { return RequiredVectorWidth; }