This reliably reproduces with the existing test:

$ CONF=linux-x86-server-fastdebug make run-test TEST=java/util/Random/RandomTestMoments.java TEST_VM_OPTS="-XX:UseAVX=0 -XX:UseSSE=0"

# Internal Error (/home/shade/trunks/jdk/src/hotspot/share/opto/matcher.cpp:1681), pid=1313665, tid=1313681
# assert(false) failed: bad AD file


o1789 AddD === _ o1790 o436 [[o1785 ]]

--N: o1789 AddD === _ o1790 o436 [[o1785 ]]

   --N: o1790 CastDD === o1112 o1086 [[o1789 ]] #double unconditional dependency
   REGD 100 castDD
   LEGREGD 200 MoveD2LEG
   VLREGD 200 MoveD2VL

      --N: o1086 RoundDouble === _ o1085 [[o1189 o1792 o1790 ]]
      REGDPR 0 REGDPR
      REGDPR1 0 REGDPR1
      REGDPR2 0 REGDPR2
      REGNOTDPR1 0 REGNOTDPR1
      REGD 100 MoveVL2D
      LEGREGD 200 MoveD2LEG
      VLREGD 0 VLREGD

   --N: o436 Phi === o1762 o380 o1459 [[o1451 o730 o1458 o1789 o1791 ]] #double
   REGDPR 0 REGDPR
   REGDPR1 0 REGDPR1
   REGDPR2 0 REGDPR2
   REGNOTDPR1 0 REGNOTDPR1
   REGD 100 MoveVL2D
   LEGREGD 200 MoveD2LEG
   VLREGD 0 VLREGD

I think the newly added CastDD nodes should handle regDPR instead of just regD to avoid this mismatch error.