[PATCH] don't prevent a vselect of constants from becoming a single load (PR20648)

[Sanjay Patel]

Hi chandlerc, rafael, nadav,

Fix for PR20648 - http://llvm.org/bugs/show_bug.cgi?id=20648

This patch checks the operands of a vselect to see if all values are constants. If yes, bail out of any further attempts to create a blend or shuffle because SelectionDAGLegalize knows how to turn this kind of vselect into a single load. 

This already happens for machines without SSE4.1, so the added checks just send more targets down that path.

http://reviews.llvm.org/D4934

Files:
  lib/Target/X86/X86ISelLowering.cpp
  test/CodeGen/X86/sse41-blend.ll

Index: lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- lib/Target/X86/X86ISelLowering.cpp
+++ lib/Target/X86/X86ISelLowering.cpp
@@ -10220,6 +10220,13 @@
 }
 
 SDValue X86TargetLowering::LowerVSELECT(SDValue Op, SelectionDAG &DAG) const {
+  // A vselect where all conditions and data are constants can be optimized into
+  // a single vector load by SelectionDAGLegalize::ExpandBUILD_VECTOR().
+  if (ISD::isBuildVectorOfConstantSDNodes(Op.getOperand(0).getNode()) &&
+      ISD::isBuildVectorOfConstantSDNodes(Op.getOperand(1).getNode()) &&
+      ISD::isBuildVectorOfConstantSDNodes(Op.getOperand(2).getNode()))
+    return SDValue();
+  
   SDValue BlendOp = LowerVSELECTtoBlend(Op, Subtarget, DAG);
   if (BlendOp.getNode())
     return BlendOp;
@@ -20200,6 +20207,12 @@
   if (!ISD::isBuildVectorOfConstantSDNodes(Cond.getNode()))
     return SDValue();
 
+  // A vselect where all conditions and data are constants can be optimized into
+  // a single vector load by SelectionDAGLegalize::ExpandBUILD_VECTOR().
+  if (ISD::isBuildVectorOfConstantSDNodes(LHS.getNode()) &&
+      ISD::isBuildVectorOfConstantSDNodes(RHS.getNode()))
+    return SDValue();
+
   unsigned MaskValue = 0;
   if (!BUILD_VECTORtoBlendMask(cast<BuildVectorSDNode>(Cond), MaskValue))
     return SDValue();
Index: test/CodeGen/X86/sse41-blend.ll
===================================================================
--- test/CodeGen/X86/sse41-blend.ll
+++ test/CodeGen/X86/sse41-blend.ll
@@ -138,3 +138,13 @@
   %1 = shufflevector <8 x i16> %a, <8 x i16> %b, <8 x i32> <i32 0, i32 9, i32 10, i32 3, i32 4, i32 5, i32 6, i32 15>
   ret <8 x i16> %1
 }
+
+; PR20648 - a blend of constants isn't really a blend; it's just a constant pool load.
+; CHECK-LABEL: @does_not_blend
+; CHECK: movaps
+; CHECK-NEXT: ret
+define <4 x i32> @does_not_blend() {
+  %select = select <4 x i1> <i1 1, i1 0, i1 0, i1 1>, <4 x i32> <i32 1, i32 1, i32 1, i32 1>, <4 x i32> <i32 2, i32 2, i32 2, i32 2>
+  ret <4 x i32> %select
+}
+
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D4934.12565.patch
Type: text/x-patch
Size: 2049 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140815/b27bbd26/attachment.bin>