Bitcode: Collect all MDString records into a single blob

Optimize output of MDStrings in bitcode.  This emits them in big blocks
(currently 1024) in a pair of records:
  - BULK_STRING_SIZES: the sizes of the strings in the block, and
  - BULK_STRING_DATA: a single blob, which is the concatenation of all
    the strings.

Inspired by Mehdi's similar patch, http://reviews.llvm.org/D18342, this
should (a) slightly reduce bitcode size, since there is less record
overhead, and (b) greatly improve reading speed, since blobs are super
cheap to deserialize.

I needed to add support for blobs to streaming input to get the test
suite passing.
  - StreamingMemoryObject::getPointer reads ahead and returns the
    address of the blob.
  - To avoid a possible reallocation of StreamingMemoryObject::Bytes,
    BitstreamCursor::readRecord needs to move the call to JumpToEnd
    forward so that getPointer is the last bitstream operation.

llvm-svn: 264409

llvm/lib/Bitcode/Reader/BitstreamReader.cpp

@@ -261,6 +261,10 @@ unsigned BitstreamCursor::readRecord(unsigned AbbrevID,
     }
 
     // Otherwise, inform the streamer that we need these bytes in memory.
+    // Skip over tail padding first.  We can't do it later if this is a
+    // streaming memory object, since that could reallocate the storage that
+    // the blob pointer references.
+    JumpToBit(NewEnd);
     const char *Ptr = (const char*)
       BitStream->getBitcodeBytes().getPointer(CurBitPos/8, NumElts);
 
@@ -272,8 +276,6 @@ unsigned BitstreamCursor::readRecord(unsigned AbbrevID,
       for (; NumElts; --NumElts)
         Vals.push_back((unsigned char)*Ptr++);
     }
-    // Skip over tail padding.
-    JumpToBit(NewEnd);
   }
 
   return Code;