ARM: Creating a vector from a lane of another.

The VDUP instruction source register doesn't allow a non-constant lane
index, so make sure we don't construct a ARM::VDUPLANE node asking it to
do so.

rdar://13328063
http://llvm.org/bugs/show_bug.cgi?id=13963

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176413 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index 5d24e92f2357ec57f8dbdaead332e7c035bcd656..514971f01ee82fac1b2552aeb907182e15aa5493 100644 (file)
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -4452,8 +4452,11 @@ SDValue ARMTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
 
       // If we are VDUPing a value that comes directly from a vector, that will
       // cause an unnecessary move to and from a GPR, where instead we could
-      // just use VDUPLANE.
-      if (Value->getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
+      // just use VDUPLANE. We can only do this if the lane being extracted
+      // is at a constant index, as the VDUP from lane instructions only have
+      // constant-index forms.
+      if (Value->getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
+          isa<ConstantSDNode>(Value->getOperand(1))) {
         // We need to create a new undef vector to use for the VDUPLANE if the
         // size of the vector from which we get the value is different than the
         // size of the vector that we need to create. We will insert the element

diff --git a/test/CodeGen/Thumb2/2013-03-02-vduplane-nonconstant-source-index.ll b/test/CodeGen/Thumb2/2013-03-02-vduplane-nonconstant-source-index.ll
new file mode 100644 (file)
index 0000000..937ecc0
--- /dev/null
+++ b/test/CodeGen/Thumb2/2013-03-02-vduplane-nonconstant-source-index.ll
@@ -0,0 +1,14 @@
+; RUN: llc < %s -mtriple=thumbv7-apple-ios | FileCheck %s
+
+define void @bar(<4 x i32>* %p, i32 %lane, <4 x i32> %phitmp) nounwind {
+; CHECK:  vst1.64 {d{{[0-9]+}}, d{{[0-9]+}}}, [r[[SOURCE:[0-9]+]]:128]
+; CHECK:  add.w r[[ADDR:[0-9]+]], r[[SOURCE]], {{r[0-9]+}}, lsl #2
+; CHECK:  vld1.32 {[[DREG:d[0-9]+]][], [[DREG2:d[0-9]+]][]}, [r[[ADDR]]:32]
+; CHECK:  vst1.32 {[[DREG]], [[DREG2]]}, [r0]
+  %val = extractelement <4 x i32> %phitmp, i32 %lane
+  %r1 = insertelement <4 x i32> undef, i32 %val, i32 1
+  %r2 = insertelement <4 x i32> %r1, i32 %val, i32 2
+  %r3 = insertelement <4 x i32> %r2, i32 %val, i32 3
+  store <4 x i32> %r3, <4 x i32>* %p, align 4
+  ret void
+}