64-bit PowerPC code quality regression

[David Edelsohn]

>>>>> Richard Henderson writes:

Richard> This splitter matches.

Richard> Since 0xff..ff0000 is mask64_operand, and in num_insns_constant
Richard> we have 

Richard> if (mask64_operand (op, mode))
Richard> return 2;

	Yes, I know the splitter can match.  I mentioned that in my
original note.  What I don't understand is why GCC is applying the
splitter when it already has a match.

	I can tighten the final constraint on the splitter so that it does
not match, but I thought that the recognizer follows the order of the MD
file.  There is a previous pattern which should have matched before the
splitter. 

Thanks, David