gcc-64 on HP-UX 11.00

[John David Anglin]

> What is important here is whether or not an address load will ever end
> up in an FPR.  While I believe it may be possible in theory, I don't
> believe it will be possible in practice, except due to bugs elsewhere
> in the compiler (such as the "T" constraint bug).

In testing today, I got the following bootstrap error:

stage1/xgcc -Bstage1/ -B/home/dave/opt/gnu/hppa-linux/bin/ -c -DIN_GCC    -g -O2 -mpa-risc-2-0 -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -Wtraditional -pedantic -Wno-long-long  -DHAVE_CONFIG_H    -I. -I. -I../../gcc/gcc -I../../gcc/gcc/. -I../../gcc/gcc/config -I../../gcc/gcc/../include ../../gcc/gcc/global.c -o global.o
../../gcc/gcc/global.c: In function `expand_preferences':
../../gcc/gcc/global.c:882: insn does not satisfy its constraints:
(insn 806 804 162 (set (reg:SI 72 %fr24)
	(mem/f:SI (lo_sum:SI (reg/f:SI 9 %r9 [156])
		(symbol_ref:SI ("allocno_row_words"))) [3 allocno_row_words+0 S4 A32])) 68 {*pa.md:2234} (nil)
    (nil))
../../gcc/gcc/global.c:882: Internal compiler error in reload_cse_simplify_operands, at reload1.c:8367

As you can see, I was trying a bootstrap with "-mpa-risc-2-0".  Insn 806
comes from a reload for:

(insn 162 209 180 (set (reg:DI 160)
        (mult:DI (zero_extend:DI (reg:SI 157))
	    (zero_extend:DI (mem/f:SI (lo_sum:SI (reg/f:SI 156)
			(symbol_ref:SI ("allocno_row_words"))) [3 allocno_row_words+0 S4 A32])))) 170 {umulsidi3} (insn_list 159 (nil))
    (nil))

The umulsidi3 wants to arguments in FP registers.  So, it appears that
we need to handle non-pic LO_SUMs in the T constraint.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)