Re: Resend of Patches for HP-UX support on IA64

["Zack Weinberg" <zackw at stanford dot edu>]

On Tue, Mar 13, 2001 at 09:37:43AM -0800, Steve Ellcey wrote:
> 
> I sent some patches last week (March 5) for HP-UX support on IA64 and
> asked that they be checked in but I have not heard any feedback so I
> am sending them again in the hopes that someone will look them over
> and check them in.  I did tweek my config.gcc change a little to clean
> it up from the last submission.

I have some nits to pick.

> *** config/ia64/hpux.h.orig	Mon Mar  5 12:57:17 2001
> --- config/ia64/hpux.h	Mon Mar  5 10:48:23 2001
> ***************
> *** 0 ****
> --- 1,76 ----
> + /* Definitions for ia64-hpux target.  */

Missing copyright notice.

> + #include "ia64/ia64.h"
> + #include "svr4.h"
> + #include "sysv4.h"

Please use the new style, listing each relevant header explicitly in
config.gcc.

I think you mean ia64/sysv4.h.  config/ia64 is not in the -I list.

> + #undef CPP_PREDEFINES
> + #define CPP_PREDEFINES "\
> +   -D__IA64__ -D__ia64 -D__ia64__ -D__hpux -D__hpux__ -Dhpux -Dunix \
> +   -D__BIG_ENDIAN__ -D_LONGLONG -D__ELF__ \
> +   -Asystem(hpux) -Acpu(ia64) -Amachine(ia64) -D_UINT128_T"

Please use the new syntax for -A: -Asystem=hpux.  This allows cutting
and pasting -v output back into the shell without digging through to
find the parentheses and quote them.

Shouldn't you have an -Asystem=posix in there?

> + /* -D__fpreg=long double is needed to compensate for the lack of __fpreg
> +    which is a primitive type in HP C but does not exist in GNU C.  Same
> +    for __float80 and __float128.  These types appear in HP-UX header
> +    files and so must have some definition.  */
> + 
> + #undef CPP_SPEC
> + #define CPP_SPEC "\
> +   %{mcpu=itanium:-D__itanium__} %{mbig-endian:-D__BIG_ENDIAN__} \
> +   %{!milp32:-D__LP64__ -D__LONG_MAX__=9223372036854775807L} \
> +   %{!ansi: -D_HPUX_SOURCE -D__STDC_EXT__} \
> +   -D__fpreg=long\\ double \
> +   -D__float80=long\\ double \
> +   -D__float128=long\\ double"

The -mcpu, -mbig-endian, -milp32 stuff should be factored out into a
CPP_CPU_SPEC shared by all the IA64 targets.  And -milp32 should not
appear until it actually works, i.e. changes the size of long and
pointers in the compiler.

> + #undef CONST_SECTION_ASM_OP
> + #define CONST_SECTION_ASM_OP    "\t.section\t.data,\t\"a\",\t\"progbits\""

.rodata, surely?

> + /* We need this macro to output DWARF2 information correctly.  The macro
> +    is defined in dwarf2out.c, but it will not do section relative offsets
> +    which messes up our ability to debug using gdb.  */
> + 
> + #undef ASM_OUTPUT_DWARF_OFFSET
> + #define ASM_OUTPUT_DWARF_OFFSET(FILE,LABEL)				\
> +  do {									\
> + 	fprintf ((FILE), "\t%s\t", UNALIGNED_OFFSET_ASM_OP);		\
> + 	fprintf ((FILE), "@secrel(");                                   \
> + 	assemble_name (FILE, LABEL);					\
> + 	fprintf ((FILE), ")");                                          \
> +   } while (0)

Shouldn't this be made common to all IA64 targets, if the problem is
as universal as it sounds?

> + /* End of hpux.h */

Leave this out.

zw