Improved widening loads by adding support for wider loads if
the alignment allows.  Fixed a bug where we didn't use a
vector load/store for PR5626.

llvm-svn: 94338
diff --git a/llvm/lib/Target/X86/X86InstrMMX.td b/llvm/lib/Target/X86/X86InstrMMX.td
index fc40c9a..ab169ac 100644
--- a/llvm/lib/Target/X86/X86InstrMMX.td
+++ b/llvm/lib/Target/X86/X86InstrMMX.td
@@ -501,6 +501,20 @@
                                (iPTR imm:$src3))))]>;
 }
 
+// MMX to XMM for vector types
+def MMX_X86movq2dq : SDNode<"X86ISD::MOVQ2DQ", SDTypeProfile<1, 1,
+                            [SDTCisVT<0, v2i64>, SDTCisVT<1, v1i64>]>>;
+
+def : Pat<(v2i64 (MMX_X86movq2dq VR64:$src)),
+          (v2i64 (MMX_MOVQ2DQrr VR64:$src))>;
+
+def : Pat<(v2i64 (MMX_X86movq2dq (load_mmx addr:$src))),
+          (v2i64 (MOVQI2PQIrm addr:$src))>;
+
+def : Pat<(v2i64 (MMX_X86movq2dq (v1i64 (bitconvert
+                            (v2i32 (scalar_to_vector (loadi32 addr:$src))))))),
+          (v2i64 (MOVDI2PDIrm addr:$src))>;
+
 // Mask creation
 def MMX_PMOVMSKBrr : MMXI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR64:$src),
                           "pmovmskb\t{$src, $dst|$dst, $src}",