[llvm] r225491 - [x86] Add a flag to control the vector shuffle legality predicates that

[Chandler Carruth]

Author: chandlerc
Date: Thu Jan  8 19:24:36 2015
New Revision: 225491

URL: http://llvm.org/viewvc/llvm-project?rev=225491&view=rev
Log:
[x86] Add a flag to control the vector shuffle legality predicates that
complements the new vector shuffle lowering code path. This flag,
naturally, is *off* because we've not tested or evaluated the results of
this at all. However, the flag will make it much easier to evaluate
whether we can be this aggressive and whether there are missing vector
shuffle lowering optimizations.

Modified:
    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=225491&r1=225490&r2=225491&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Thu Jan  8 19:24:36 2015
@@ -71,6 +71,13 @@ static cl::opt<bool> ExperimentalVectorS
     cl::desc("Enable an experimental vector shuffle lowering code path."),
     cl::Hidden);
 
+static cl::opt<bool> ExperimentalVectorShuffleLegality(
+    "x86-experimental-vector-shuffle-legality", cl::init(false),
+    cl::desc("Enable experimental shuffle legality based on the experimental "
+             "shuffle lowering. Should only be used with the experimental "
+             "shuffle lowering."),
+    cl::Hidden);
+
 static cl::opt<int> ReciprocalEstimateRefinementSteps(
     "x86-recip-refinement-steps", cl::init(1),
     cl::desc("Specify the number of Newton-Raphson iterations applied to the "
@@ -20148,6 +20155,14 @@ X86TargetLowering::isShuffleMaskLegal(co
   if (VT.getSizeInBits() == 64)
     return false;
 
+  // This is an experimental legality test that is tailored to match the
+  // legality test of the experimental lowering more closely. They are gated
+  // separately to ease testing of performance differences.
+  if (ExperimentalVectorShuffleLegality)
+    // We only care that the types being shuffled are legal. The lowering can
+    // handle any possible shuffle mask that results.
+    return isTypeLegal(SVT);
+
   // If this is a single-input shuffle with no 128 bit lane crossings we can
   // lower it into pshufb.
   if ((SVT.is128BitVector() && Subtarget->hasSSSE3()) ||
@@ -20192,6 +20207,14 @@ X86TargetLowering::isVectorClearMaskLega
     return false;
 
   MVT SVT = VT.getSimpleVT();
+
+  // This is an experimental legality test that is tailored to match the
+  // legality test of the experimental lowering more closely. They are gated
+  // separately to ease testing of performance differences.
+  if (ExperimentalVectorShuffleLegality)
+    // The new vector shuffle lowering is very good at managing zero-inputs.
+    return isShuffleMaskLegal(Mask, VT);
+
   unsigned NumElts = SVT.getVectorNumElements();
   // FIXME: This collection of masks seems suspect.
   if (NumElts == 2)