OSDN Git Service

2009-03-17 H.J. Lu <hongjiu.lu@intel.com>
[pf3gnuchains/gcc-fork.git] / gcc / ChangeLog
index 341bcc6..42240dc 100644 (file)
@@ -1,3 +1,171 @@
+2009-03-17  H.J. Lu  <hongjiu.lu@intel.com>
+
+       PR target/39477
+       * doc/extend.texi: Correct register behavior for regparm on
+       Intel 386.
+
+2009-03-17  H.J. Lu  <hongjiu.lu@intel.com>
+
+       PR target/39476
+       * config/i386/i386.c (ix86_function_regparm): Rewrite for
+       64bit.
+
+2009-03-17  H.J. Lu  <hongjiu.lu@intel.com>
+
+       PR target/39473
+       * config/i386/i386.c (ix86_expand_call): Check extra clobbers
+       for ms->sysv ABI calls only in 64bit mode.
+
+       * config/i386/i386.md (untyped_call): Support 32bit.
+
+2009-03-16  H.J. Lu  <hongjiu.lu@intel.com>
+
+       * doc/extend.texi: Replace x86_65 with x86_64.
+
+2009-03-16  Jakub Jelinek  <jakub@redhat.com>
+
+       PR tree-optimization/39455
+       * tree-ssa-loop-niter.c (number_of_iterations_lt_to_ne): Fix types
+       mismatches for POINTER_TYPE_P (type).
+       (number_of_iterations_le): Likewise.
+
+2009-03-16  Hariharan Sandanagobalane <hariharan@picochip.com>
+
+       * config/picochip/picochip.c: Removed profiling support.
+       * config/picochip/picochip.md: Removed profiling instruction.
+       * config/picochip/picochip.h: Removed profiling builtin.
+
+2009-03-16  Joseph Myers  <joseph@codesourcery.com>
+
+       * doc/install.texi (--with-host-libstdcxx): Document.
+
+2009-03-14  Anatoly Sokolov  <aesok@post.ru>
+
+       PR target/34299
+       * config/avr/avr.c (avr_handle_fndecl_attribute): Move code for
+       generate a warning if the function name does not begin with
+       "__vector" and the function has either the 'signal' or 'interrupt'
+       attribute, from here to ...
+       (avr_declare_function_name): ...here. New function.
+       * config/avr/avr.h (ASM_DECLARE_FUNCTION_NAME): Redefine.
+       * config/avr/avr-protos.h (avr_declare_function_name): Declare.
+
+2009-03-14  Jakub Jelinek  <jakub@redhat.com>
+
+       PR bootstrap/39454
+       * cse.c (fold_rtx): Don't modify original const_arg1 when
+       canonicalizing SHIFT_COUNT_TRUNCATED shift count, do it on a
+       separate variable instead.
+       * rtlanal.c (nonzero_bits1) <case ASHIFTRT>: Don't assume anything
+       from out of range shift counts.
+       (num_sign_bit_copies1) <case ASHIFTRT, case ASHIFT>: Similarly.
+
+2008-03-13  Catherine Moore  <clm@codesourcery.com>
+
+       *  gcc/config/i386/x-mingw32 (host-mingw32.o): Replace
+       diagnostic.h with $(DIAGNOSTIC_H).
+
+2008-03-12  Jakub Jelinek  <jakub@redhat.com>
+
+       PR target/39431
+       * config/i386/predicates.md (cmpxchg8b_pic_memory_operand): New
+       predicate.
+       * config/i386/sync.md (sync_compare_and_swap<mode>,
+       sync_compare_and_swap_cc<mode>): For DImode with -m32 -fpic check
+       if operands[1] is cmpxchg8b_pic_memory_operand, if not force address
+       into a register.
+       (sync_double_compare_and_swapdi_pic,
+       sync_double_compare_and_swap_ccdi_pic): Require operand 1 to be
+       cmpxchg8b_pic_memory_operand instead of just memory_operand.
+
+2009-03-12  H.J. Lu  <hongjiu.lu@intel.com>
+
+       PR target/39445
+       * config/i386/i386.c (ix86_expand_push): Don't set memory
+       alignment.
+
+2009-03-12  H.J. Lu  <hongjiu.lu@intel.com>
+
+       PR target/39327
+       * config/i386/sse.md (avx_addsubv8sf3): Correct item bits.
+       (avx_addsubv4df3): Likewise.
+       (*avx_addsubv4sf3): Likewise.
+       (sse3_addsubv4sf3): Likewise.
+
+2009-03-12  H.J. Lu  <hongjiu.lu@intel.com>
+
+       PR target/38824
+       * config/i386/i386.md: Compare REGNO on the new peephole2
+       patterns.
+
+2009-03-12  Vladimir Makarov  <vmakarov@redhat.com>
+
+       PR debug/39432
+       * ira-int.h (struct allocno): Fix comment for calls_crossed_num.
+       * ira-conflicts.c (ira_build_conflicts): Prohibit call used
+       registers for allocnos created from user-defined variables.
+
+2009-03-11  Ulrich Weigand  <Ulrich.Weigand@de.ibm.com>
+
+       PR target/39181
+       * config/spu/spu.c (spu_expand_mov): Handle invalid subregs
+       of non-integer mode as well.
+
+2009-03-11  Adam Nemet  <anemet@caviumnetworks.com>
+
+       * gimplify.c (gimplify_call_expr): Don't set CALL_CANNOT_INLINE_P
+       for functions for which the parameter types are unknown.
+
+2009-03-11  Jakub Jelinek  <jakub@redhat.com>
+
+       PR target/39137
+       * cfgexpand.c (get_decl_align_unit): Use LOCAL_DECL_ALIGNMENT
+       macro.
+       * defaults.h (LOCAL_DECL_ALIGNMENT): Define if not yet defined.
+       * config/i386/i386.h (LOCAL_DECL_ALIGNMENT): Define.
+       * config/i386/i386.c (ix86_local_alignment): For
+       -m32 -mpreferred-stack-boundary=2 use 32-bit alignment for
+       long long variables on the stack to avoid dynamic realignment.
+       Allow the first argument to be a decl rather than type.
+       * doc/tm.texi (LOCAL_DECL_ALIGNMENT): Document.
+
+2009-03-11  Nick Clifton  <nickc@redhat.com>
+
+       PR target/5362
+       * config/mcore/mcore.opt: Remove deprecated m4align and m8align
+       options.
+       Add description to mno-lsim option.
+       * config/mcore/mcore.h: Remove comment about deprecated m4align
+       option.
+       (TARGET_DEFAULT): Remove deprecated MASK_M8ALIGN.
+       * doc/invoke.texi: Add description of mno-lsim and
+       mstack-increment options.
+
+       * config/fr30/fr30.opt: Document the -mno-lsim option.
+       * doc/invoke.texi: Add descriptions of the FR30's -msmall-model
+       and -mno-lsim options.
+
+2009-03-11  Ulrich Weigand  <Ulrich.Weigand@de.ibm.com>
+
+       * fold-const.c (fold_comparison): Only call fold_inf_compare
+       if the mode supports infinities.
+
+2009-03-11  Jason Merrill  <jason@redhat.com>
+
+       PR debug/39086
+       * tree-nrv.c (tree_nrv): Don't do this optimization if the front
+       end already did.  Notice GIMPLE_CALL modifications of the result.
+       Don't copy debug information from an ignored decl or a decl from
+       another function.
+
+2009-03-10  Richard Guenther  <rguenther@suse.de>
+           Nathan Froyd  <froydnj@codesourcery.com>
+
+       PR middle-end/37850
+       * libgcc2.c (__mulMODE3): Use explicit assignments to form the
+       result.
+       (__divMODE3): Likewise.
+
 2009-03-09  Jakub Jelinek  <jakub@redhat.com>
 
        PR tree-optimization/39394