add a note, this is actually not too bad to implement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@49466 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/README-SSE.txt b/lib/Target/X86/README-SSE.txt
index bbd93bf3ed404f3b0a86ba6a091eec5d69f98909..5cccfa8daa4b295b1efe5f9a8359c5b6ce2ebbd1 100644 (file)
--- a/lib/Target/X86/README-SSE.txt
+++ b/lib/Target/X86/README-SSE.txt
@@ -472,7 +472,13 @@ _doload64:
        addl    $12, %esp
        ret
 
-instead of movsd from the stack.
+instead of movsd from the stack.  This is actually not too bad to implement. The
+best way to do this is to implement a dag combine that turns 
+bitconvert(build_pair(load a, load b)) into one load of the right type.  The
+only trick to this is writing the predicate that determines that a/b are at the
+right offset from each other.  For the enterprising hacker, InferAlignment is a
+helpful place to start poking if interested.
+
 
 //===---------------------------------------------------------------------===//