When generating a vector the really slow way, via loads
and stores, handle the case where the element size is not
a valid target type correctly (PPC).



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@89521 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index 4f0a229..4cd7ca1 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -1517,6 +1517,7 @@
   // Create the stack frame object.
   EVT VT = Node->getValueType(0);
   EVT OpVT = Node->getOperand(0).getValueType();
+  EVT EltVT = VT.getVectorElementType();
   DebugLoc dl = Node->getDebugLoc();
   SDValue FIPtr = DAG.CreateStackTemporary(VT);
   int FI = cast<FrameIndexSDNode>(FIPtr.getNode())->getIndex();
@@ -1524,7 +1525,7 @@
 
   // Emit a store of each element to the stack slot.
   SmallVector<SDValue, 8> Stores;
-  unsigned TypeByteSize = OpVT.getSizeInBits() / 8;
+  unsigned TypeByteSize = EltVT.getSizeInBits() / 8;
   // Store (in the right endianness) the elements to memory.
   for (unsigned i = 0, e = Node->getNumOperands(); i != e; ++i) {
     // Ignore undef elements.
@@ -1535,8 +1536,13 @@
     SDValue Idx = DAG.getConstant(Offset, FIPtr.getValueType());
     Idx = DAG.getNode(ISD::ADD, dl, FIPtr.getValueType(), FIPtr, Idx);
 
-    Stores.push_back(DAG.getStore(DAG.getEntryNode(), dl, Node->getOperand(i),
-                                  Idx, SV, Offset));
+    // If EltVT smaller than OpVT, only store the bits necessary.
+    if (EltVT.bitsLT(OpVT))
+      Stores.push_back(DAG.getTruncStore(DAG.getEntryNode(), dl,
+                          Node->getOperand(i), Idx, SV, Offset, EltVT));
+    else
+      Stores.push_back(DAG.getStore(DAG.getEntryNode(), dl, 
+                                    Node->getOperand(i), Idx, SV, Offset));
   }
 
   SDValue StoreChain;
diff --git a/test/CodeGen/PowerPC/vec_buildvector_loadstore.ll b/test/CodeGen/PowerPC/vec_buildvector_loadstore.ll
new file mode 100644
index 0000000..015c086
--- /dev/null
+++ b/test/CodeGen/PowerPC/vec_buildvector_loadstore.ll
@@ -0,0 +1,37 @@
+; RUN: llc < %s -march=ppc32 -mtriple=powerpc-apple-darwin -mattr=+altivec  | FileCheck %s
+; Formerly this did byte loads and word stores.
+@a = external global <16 x i8>
+@b = external global <16 x i8>
+@c = external global <16 x i8>
+
+define void @foo() nounwind ssp {
+; CHECK: _foo:
+; CHECK-NOT: stw
+entry:
+    %tmp0 = load <16 x i8>* @a, align 16
+  %tmp180.i = extractelement <16 x i8> %tmp0, i32 0 ; <i8> [#uses=1]
+  %tmp181.i = insertelement <16 x i8> <i8 0, i8 0, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef>, i8 %tmp180.i, i32 2 ; <<16 x i8>> [#uses=1]
+  %tmp182.i = extractelement <16 x i8> %tmp0, i32 1 ; <i8> [#uses=1]
+  %tmp183.i = insertelement <16 x i8> %tmp181.i, i8 %tmp182.i, i32 3 ; <<16 x i8>> [#uses=1]
+  %tmp184.i = insertelement <16 x i8> %tmp183.i, i8 0, i32 4 ; <<16 x i8>> [#uses=1]
+  %tmp185.i = insertelement <16 x i8> %tmp184.i, i8 0, i32 5 ; <<16 x i8>> [#uses=1]
+  %tmp186.i = extractelement <16 x i8> %tmp0, i32 4 ; <i8> [#uses=1]
+  %tmp187.i = insertelement <16 x i8> %tmp185.i, i8 %tmp186.i, i32 6 ; <<16 x i8>> [#uses=1]
+  %tmp188.i = extractelement <16 x i8> %tmp0, i32 5 ; <i8> [#uses=1]
+  %tmp189.i = insertelement <16 x i8> %tmp187.i, i8 %tmp188.i, i32 7 ; <<16 x i8>> [#uses=1]
+  %tmp190.i = insertelement <16 x i8> %tmp189.i, i8 0, i32 8 ; <<16 x i8>> [#uses=1]
+  %tmp191.i = insertelement <16 x i8> %tmp190.i, i8 0, i32 9 ; <<16 x i8>> [#uses=1]
+  %tmp192.i = extractelement <16 x i8> %tmp0, i32 8 ; <i8> [#uses=1]
+  %tmp193.i = insertelement <16 x i8> %tmp191.i, i8 %tmp192.i, i32 10 ; <<16 x i8>> [#uses=1]
+  %tmp194.i = extractelement <16 x i8> %tmp0, i32 9 ; <i8> [#uses=1]
+  %tmp195.i = insertelement <16 x i8> %tmp193.i, i8 %tmp194.i, i32 11 ; <<16 x i8>> [#uses=1]
+  %tmp196.i = insertelement <16 x i8> %tmp195.i, i8 0, i32 12 ; <<16 x i8>> [#uses=1]
+  %tmp197.i = insertelement <16 x i8> %tmp196.i, i8 0, i32 13 ; <<16 x i8>> [#uses=1]
+%tmp201 = shufflevector <16 x i8> %tmp197.i, <16 x i8> %tmp0, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 28, i32 29>; ModuleID = 'try.c'
+    store <16 x i8> %tmp201, <16 x i8>* @c, align 16
+    br label %return
+
+return:		; preds = %bb2
+	ret void
+; CHECK: blr
+}