2003-06-13 Richard Earnshaw <rearnsha@arm.com>
* tc-arm.c (FPU_DEFAULT, case TE_LINUX): Default to FPU_ARCH_FPA.
(FPU_DEFAULT, case TE_NetBSD): Default to FPU_ARCH_VFP for ELF,
FPU_ARCH_FPA for AOUT.
(md_begin): Don't try to guess the floating point architecture from
the CPU if the OS ABI (Linux, NetBSD) mandates a particular form.
...per Richard's suggestion.
pre-processor.
This makes code of the form
#define FOO(X) (BAR(X))
#define BAR(X) (X)
#define BAZ 23 /* comment, comment. */
int foo(void) { return FOO(BAZ); }
be output as expected. (The old behaviour got confused by the ','
within the comment.)
2003-04-26 Ben Elliston <bje@wasabisystems.com>
* config/arm/arm.c (arm_adjust_cost): Correct logic that tests the
return values from recog_memoized().
BFD32_BACKENDS so that the correct dependencies are registered.
Parallel builds now work as a result. Fixes PR 20602 and PR 21081.
I'll leave the bigger question of if this stuff is still needed and why it
hasn't been fed back to the binutils people to someone more qualified than
myself.
2003-03-14 Jason Thorpe <thorpej@wasabisystems.com>
* c-decl.c (flag_hosted): Move from here...
* toplev.c: ...to here.
* c-tree.h (flag_hosted): Move extern declaration from here...
* flags.h: ...to here.
* config/alpha/alpha.c (alpha_initialize_trampoline): Only
emit a call to __enable_execute_stack if flag_hosted is true.
* config/i386/i386.c (x86_initialize_trampoline): Likewise.
* config/sparc/sparc.c (sparc_initialize_trampoline)
(sparc64_initialize_trampoline): Likewise.
An updated version of this will be submitted to gcc.gnu.org.
(*mulsi3_compare0, *mulsi_compare0_scratch, *mulsi3addsi_compare0,
*mulsi3addsi_compare0_scratch): Predicate on !arm_is_xscale.
(mulhisi3, *mulhisi3addsi, *mulhidi3adddi): New patterns, predicated
on arm_is_xscale.
by the FSF binutils maintainers, who instead suggested a different way
of addressing the problem.
(Besides, the size adjustment code was broken by the binutils update,
since about 1/2 of the changes got lost.)
(and associated targets for .dvi et al), as well as man pages from .pod,
using a prefix of `# '.
This prevents a variety of build issues caused by situations such as
cvs checkouts which result in the .texinfo file being slightly newer
than the .info file, and the build process trying to unnecessarily
regenerate them, which will fail in a variety of circumstances
(build tools not available, read-only source, etc ...)
Once a better solution is found, we can switch to it.
For now, this improves things greatly and should reduce a lot of
support requests that would undoubtably appear otherwise...
