fix the buildvector->insertp[sd] logic to not always create a redundant

insertp[sd] $0, which is a noop.  Before:

_f32:                                   ## @f32
	pshufd	$1, %xmm1, %xmm2
	pshufd	$1, %xmm0, %xmm3
	addss	%xmm2, %xmm3
	addss	%xmm1, %xmm0
                                        ## kill: XMM0<def> XMM0<kill> XMM0<def>
	insertps	$0, %xmm0, %xmm0
	insertps	$16, %xmm3, %xmm0
	ret

after:

_f32:                                   ## @f32
	movdqa	%xmm0, %xmm2
	addss	%xmm1, %xmm2
	pshufd	$1, %xmm1, %xmm1
	pshufd	$1, %xmm0, %xmm3
	addss	%xmm1, %xmm3
	movdqa	%xmm2, %xmm0
	insertps	$16, %xmm3, %xmm0
	ret

The extra movs are due to a random (poor) scheduling decision.

llvm-svn: 112379

lib/Target/X86/X86ISelLowering.cpp

@@ -4278,14 +4278,20 @@ X86TargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const {
     if (LD.getNode())
       return LD;
     
-    // For SSE 4.1, use inserts into undef.  
+    // For SSE 4.1, use insertps to put the high elements into the low element. 
     if (getSubtarget()->hasSSE41()) {
-      V[0] = DAG.getUNDEF(VT);
+      SDValue Result;
-      for (unsigned i = 0; i < NumElems; ++i)
+      if (Op.getOperand(0).getOpcode() != ISD::UNDEF)
-        if (Op.getOperand(i).getOpcode() != ISD::UNDEF)
+        Result = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Op.getOperand(0));
-          V[0] = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, V[0],
+      else
+        Result = DAG.getUNDEF(VT);
+      
+      for (unsigned i = 1; i < NumElems; ++i) {
+        if (Op.getOperand(i).getOpcode() == ISD::UNDEF) continue;
+        Result = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, Result,
                              Op.getOperand(i), DAG.getIntPtrConstant(i));
-      return V[0];
+      }
+      return Result;
     }
     
     // Otherwise, expand into a number of unpckl*, start by extending each of

test/CodeGen/X86/2009-02-26-MachineLICMBug.ll

@@ -1,4 +1,4 @@
-; RUN: llc < %s -march=x86-64 -mattr=+sse3,+sse41 -stats |& grep {7 machine-licm}
+; RUN: llc < %s -march=x86-64 -mattr=+sse3,+sse41 -stats |& grep {6 machine-licm}
 ; RUN: llc < %s -march=x86-64 -mattr=+sse3,+sse41 | FileCheck %s
 ; rdar://6627786
 ; rdar://7792037

test/CodeGen/X86/sse41.ll

@@ -224,3 +224,28 @@ declare i32 @llvm.x86.sse41.ptestz(<4 x float>, <4 x float>) nounwind readnone
 declare i32 @llvm.x86.sse41.ptestc(<4 x float>, <4 x float>) nounwind readnone
 declare i32 @llvm.x86.sse41.ptestnzc(<4 x float>, <4 x float>) nounwind readnone
 
+; This used to compile to insertps $0  + insertps $16.  insertps $0 is always
+; pointless.
+define <2 x float> @buildvector(<2 x float> %A, <2 x float> %B) nounwind  {
+entry:
+  %tmp7 = extractelement <2 x float> %A, i32 0
+  %tmp5 = extractelement <2 x float> %A, i32 1
+  %tmp3 = extractelement <2 x float> %B, i32 0
+  %tmp1 = extractelement <2 x float> %B, i32 1
+  %add.r = fadd float %tmp7, %tmp3
+  %add.i = fadd float %tmp5, %tmp1
+  %tmp11 = insertelement <2 x float> undef, float %add.r, i32 0
+  %tmp9 = insertelement <2 x float> %tmp11, float %add.i, i32 1
+  ret <2 x float> %tmp9
+; X32: buildvector:
+; X32-NOT: insertps $0
+; X32: insertps $16
+; X32-NOT: insertps $0
+; X32: ret
+; X64: buildvector:
+; X64-NOT: insertps $0
+; X64: insertps $16
+; X64-NOT: insertps $0
+; X64: ret
+}
+

test/CodeGen/X86/vec_insert-9.ll

@@ -1,5 +1,5 @@
 ; RUN: llc < %s -march=x86 -mattr=+sse41 > %t
-; RUN: grep pinsrd %t | count 2
+; RUN: grep pinsrd %t | count 1
 
 define <4 x i32> @var_insert2(<4 x i32> %x, i32 %val, i32 %idx) nounwind  {
 entry: