[X86][SSE] Improvements to byte shift shuffle matching

Since (v)pslldq / (v)psrldq instructions resolve to a single input argument it is useful to match it much earlier than we currently do - this prevents more complicated shuffles (notably insertion into a zero vector) matching before it.

Differential Revision: http://reviews.llvm.org/D6409



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@222796 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 20 insertions, 20 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 63a3009fd82..0833079d573 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -8217,6 +8217,11 @@ static SDValue lowerV2I64VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
                     getV4X86ShuffleImm8ForMask(WidenedMask, DAG)));
   }
 
+  // Try to use byte shift instructions.
+  if (SDValue Shift = lowerVectorShuffleAsByteShift(
+          DL, MVT::v2i64, V1, V2, Mask, DAG))
+    return Shift;
+
   // If we have a single input from V2 insert that into V1 if we can do so
   // cheaply.
   if ((Mask[0] >= 2) + (Mask[1] >= 2) == 1) {
@@ -8243,11 +8248,6 @@ static SDValue lowerV2I64VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
                                                   Subtarget, DAG))
       return Blend;
 
-  // Try to use byte shift instructions.
-  if (SDValue Shift = lowerVectorShuffleAsByteShift(
-          DL, MVT::v2i64, V1, V2, Mask, DAG))
-    return Shift;
-
   // Try to use byte rotation instructions.
   // Its more profitable for pre-SSSE3 to use shuffles/unpacks.
   if (Subtarget->hasSSSE3())
@@ -8508,6 +8508,11 @@ static SDValue lowerV4I32VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
                        getV4X86ShuffleImm8ForMask(Mask, DAG));
   }
 
+  // Try to use byte shift instructions.
+  if (SDValue Shift = lowerVectorShuffleAsByteShift(
+          DL, MVT::v4i32, V1, V2, Mask, DAG))
+    return Shift;
+
   // There are special ways we can lower some single-element blends.
   if (NumV2Elements == 1)
     if (SDValue V = lowerVectorShuffleAsElementInsertion(MVT::v4i32, DL, V1, V2,
@@ -8525,11 +8530,6 @@ static SDValue lowerV4I32VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
                                                   Subtarget, DAG))
       return Blend;
 
-  // Try to use byte shift instructions.
-  if (SDValue Shift = lowerVectorShuffleAsByteShift(
-          DL, MVT::v4i32, V1, V2, Mask, DAG))
-    return Shift;
-
   // Try to use byte rotation instructions.
   // Its more profitable for pre-SSSE3 to use shuffles/unpacks.
   if (Subtarget->hasSSSE3())
@@ -8593,17 +8593,17 @@ static SDValue lowerV8I16SingleInputVectorShuffle(
                                                         Mask, Subtarget, DAG))
     return Broadcast;
 
+  // Try to use byte shift instructions.
+  if (SDValue Shift = lowerVectorShuffleAsByteShift(
+          DL, MVT::v8i16, V, V, Mask, DAG))
+    return Shift;
+
   // Use dedicated unpack instructions for masks that match their pattern.
   if (isShuffleEquivalent(Mask, 0, 0, 1, 1, 2, 2, 3, 3))
     return DAG.getNode(X86ISD::UNPCKL, DL, MVT::v8i16, V, V);
   if (isShuffleEquivalent(Mask, 4, 4, 5, 5, 6, 6, 7, 7))
     return DAG.getNode(X86ISD::UNPCKH, DL, MVT::v8i16, V, V);
 
-  // Try to use byte shift instructions.
-  if (SDValue Shift = lowerVectorShuffleAsByteShift(
-          DL, MVT::v8i16, V, V, Mask, DAG))
-    return Shift;
-
   // Try to use byte rotation instructions.
   if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
           DL, MVT::v8i16, V, V, Mask, Subtarget, DAG))
@@ -9210,6 +9210,11 @@ static SDValue lowerV8I16VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
   assert(NumV1Inputs > 0 && "All single-input shuffles should be canonicalized "
                             "to be V1-input shuffles.");
 
+  // Try to use byte shift instructions.
+  if (SDValue Shift = lowerVectorShuffleAsByteShift(
+          DL, MVT::v8i16, V1, V2, Mask, DAG))
+    return Shift;
+
   // There are special ways we can lower some single-element blends.
   if (NumV2Inputs == 1)
     if (SDValue V = lowerVectorShuffleAsElementInsertion(MVT::v8i16, DL, V1, V2,
@@ -9227,11 +9232,6 @@ static SDValue lowerV8I16VectorShuffle(SDValue Op, SDValue V1, SDValue V2,
                                                   Subtarget, DAG))
       return Blend;
 
-  // Try to use byte shift instructions.
-  if (SDValue Shift = lowerVectorShuffleAsByteShift(
-          DL, MVT::v8i16, V1, V2, Mask, DAG))
-    return Shift;
-
   // Try to use byte rotation instructions.
   if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
           DL, MVT::v8i16, V1, V2, Mask, Subtarget, DAG))