Some aspect of GCC memory consumption increased by recent patch

gcctest@suse.de gcctest@suse.de
Thu Nov 25 13:01:00 GMT 2004


Hi,
Comparing memory consumption on compilation of combine.i and generate-3.4.ii I got:


comparing combine.c compilation at -O0 level:
  Overall memory allocated via mmap and sbrk increased from 25217k to 25325k, overall 0.43%
  Amount of produced GGC garbage increased from 42515k to 42649k, overall 0.31%
    Overall memory needed: 25217k -> 25325k
    Peak memory use before GGC: 9329k
    Peak memory use after GGC: 8641k
    Maximum of released memory in single GGC run: 2926k
    Garbage: 42515k -> 42649k
    Leak: 6088k -> 6088k
    Overhead: 5657k -> 5744k
    GGC runs: 354

comparing combine.c compilation at -O1 level:
  Amount of produced GGC garbage decreased from 69782k to 67607k, overall -3.22%
    Overall memory needed: 26493k -> 25797k
    Peak memory use before GGC: 9205k -> 9204k
    Peak memory use after GGC: 8733k -> 8721k
    Maximum of released memory in single GGC run: 2066k -> 2062k
    Garbage: 69782k -> 67607k
    Leak: 6480k -> 6485k
    Overhead: 11359k -> 10611k
    GGC runs: 555 -> 544

comparing combine.c compilation at -O2 level:
  Overall memory allocated via mmap and sbrk increased from 29801k to 29917k, overall 0.39%
  Amount of produced GGC garbage decreased from 84780k to 81908k, overall -3.51%
    Overall memory needed: 29801k -> 29917k
    Peak memory use before GGC: 12705k
    Peak memory use after GGC: 12579k
    Maximum of released memory in single GGC run: 2576k
    Garbage: 84780k -> 81908k
    Leak: 6303k -> 6306k
    Overhead: 15985k -> 14823k
    GGC runs: 562 -> 547

comparing combine.c compilation at -O3 level:
  Ovarall memory allocated via mmap and sbrk decreased from 21132k to 20132k, overall -4.97%
    Overall memory needed: 21132k -> 20132k
    Peak memory use before GGC: 13093k -> 13003k
    Peak memory use after GGC: 12579k
    Maximum of released memory in single GGC run: 3413k -> 3411k
    Garbage: 113749k -> 110600k
    Leak: 6845k -> 6832k
    Overhead: 21057k -> 19812k
    GGC runs: 629 -> 612

comparing insn-attrtab.c compilation at -O0 level:
  Amount of produced GGC garbage increased from 161088k to 162163k, overall 0.67%
    Overall memory needed: 118512k -> 118468k
    Peak memory use before GGC: 79408k -> 79386k
    Peak memory use after GGC: 46137k
    Maximum of released memory in single GGC run: 43357k -> 43335k
    Garbage: 161088k -> 162163k
    Leak: 10609k -> 10610k
    Overhead: 20460k -> 21249k
    GGC runs: 295

comparing insn-attrtab.c compilation at -O1 level:
  Overall memory allocated via mmap and sbrk increased from 125048k to 129664k, overall 3.69%
  Amount of produced GGC garbage increased from 430920k to 444561k, overall 3.17%
    Overall memory needed: 125048k -> 129664k
    Peak memory use before GGC: 86894k -> 85227k
    Peak memory use after GGC: 70491k -> 70039k
    Maximum of released memory in single GGC run: 41161k -> 41127k
    Garbage: 430920k -> 444561k
    Leak: 10956k -> 10956k
    Overhead: 80011k -> 79424k
    GGC runs: 441 -> 426

comparing insn-attrtab.c compilation at -O2 level:
  Overall memory allocated via mmap and sbrk increased from 151480k to 153472k, overall 1.32%
  Amount of produced GGC garbage increased from 477316k to 491494k, overall 2.97%
    Overall memory needed: 151480k -> 153472k
    Peak memory use before GGC: 102310k -> 101239k
    Peak memory use after GGC: 85907k -> 85452k
    Maximum of released memory in single GGC run: 42321k -> 42306k
    Garbage: 477316k -> 491494k
    Leak: 10883k -> 10876k
    Overhead: 88016k -> 87675k
    GGC runs: 362 -> 358

comparing insn-attrtab.c compilation at -O3 level:
  Overall memory allocated via mmap and sbrk increased from 151488k to 153520k, overall 1.34%
  Amount of produced GGC garbage increased from 478589k to 492771k, overall 2.96%
    Overall memory needed: 151488k -> 153520k
    Peak memory use before GGC: 102312k -> 101241k
    Peak memory use after GGC: 85909k -> 85454k
    Maximum of released memory in single GGC run: 42322k -> 42306k
    Garbage: 478589k -> 492771k
    Leak: 10927k -> 10921k
    Overhead: 88206k -> 87862k
    GGC runs: 371 -> 366

comparing Gerald's testcase PR8361 compilation at -O0 level:
  Amount of produced GGC garbage increased from 250165k to 250741k, overall 0.23%
  Amount of memory still referenced at the end of compilation increased from 57105k to 57181k, overall 0.13%
    Overall memory needed: 112784k -> 112860k
    Peak memory use before GGC: 89431k -> 89476k
    Peak memory use after GGC: 88443k -> 88489k
    Maximum of released memory in single GGC run: 19812k -> 19795k
    Garbage: 250165k -> 250741k
    Leak: 57105k -> 57181k
    Overhead: 45841k -> 46082k
    GGC runs: 363 -> 364

comparing Gerald's testcase PR8361 compilation at -O1 level:
  Amount of produced GGC garbage decreased from 595564k to 553944k, overall -7.51%
  Amount of memory still referenced at the end of compilation increased from 59145k to 59218k, overall 0.12%
    Overall memory needed: 95348k -> 95388k
    Peak memory use before GGC: 88431k -> 88479k
    Peak memory use after GGC: 87373k -> 87421k
    Maximum of released memory in single GGC run: 19437k -> 19423k
    Garbage: 595564k -> 553944k
    Leak: 59145k -> 59218k
    Overhead: 133258k -> 116891k
    GGC runs: 628 -> 613

comparing Gerald's testcase PR8361 compilation at -O2 level:
  Amount of produced GGC garbage decreased from 651671k to 600725k, overall -8.48%
  Amount of memory still referenced at the end of compilation increased from 59735k to 59801k, overall 0.11%
    Overall memory needed: 95684k -> 95388k
    Peak memory use before GGC: 88431k -> 88479k
    Peak memory use after GGC: 87373k -> 87421k
    Maximum of released memory in single GGC run: 19437k -> 19423k
    Garbage: 651671k -> 600725k
    Leak: 59735k -> 59801k
    Overhead: 158238k -> 138208k
    GGC runs: 675 -> 649

comparing Gerald's testcase PR8361 compilation at -O3 level:
  Amount of produced GGC garbage decreased from 699713k to 640766k, overall -9.20%
    Overall memory needed: 103024k -> 103068k
    Peak memory use before GGC: 95610k -> 95564k
    Peak memory use after GGC: 88661k -> 88702k
    Maximum of released memory in single GGC run: 20248k -> 20227k
    Garbage: 699713k -> 640766k
    Leak: 60125k -> 60136k
    Overhead: 172044k -> 149365k
    GGC runs: 664 -> 647

Head of changelog is:

--- /usr/src/SpecTests/sandbox-britten-memory/x86_64/mem-result/ChangeLog	2004-11-04 22:09:49.000000000 +0000
+++ /usr/src/SpecTests/sandbox-britten-memory/gcc/gcc/ChangeLog	2004-11-25 11:56:20.000000000 +0000
@@ -1,3 +1,2977 @@
+2004-11-25  Andrew Haley  <aph@redhat.com>
+
+	* gcc.c (process_command): Don't supply -v to linker.
+
+2004-11-25  Joseph S. Myers  <joseph@codesourcery.com>
+
+	* config/rs6000/darwin.h (SUBTARGET_OVERRIDE_OPTIONS): Use %qs
+	instead of `%s' in diagnostic.
+
+2004-11-25  Nathan Sidwell  <nathan@codesourcery.com>
+
+	* basic-block.h (OBSTACK_ALLOC_REG_SET): Adjust.
+	(FREE_REG_SET): Adjust.
+	* bitmap.c (bitmap_obstack_free): Cope with NULL bitmap.
+	* bitmap.h (BITMAP_OBSTACK_ALLOC): Rename to ...
+	(BITMAP_ALLOC): ... here.
+	(BITMAP_OBSTACK_FREE): Rename to ...
+	(BITMAP_FREE): Don't check for NULL bitmap here.
+	* tree-ssa-pre.c (value_insert_into_set_bitmap,
+	bitmap_set_new): Use new names.
+
+	* bt-load.c (migrate_btr_defs): Remove unneeded NULL check.
+	* df.c (df_free): Likewise.
+	* ra-build.c (ra_build_free, ra_build_free_all): Likewise.
+	* tree-ssa-loop-ivopts.c (set_use_iv_cost): Likewise.
+
+	* basic-block.h (OBSTACK_ALLOC_REG_SET): Rename to ...
+	(ALLOC_REG_SET): ... here.
+	(FREE_REG_SET): Adjust.
+	(XMALLOC_REG_SET, XFREE_REG_SET): Remove.
+	* bb-reorder.c (fix_crossing_conditional_branches): Adjust.
+	* cfglayout.c (cfg_layout_duplicate_bb): Adjust.
+	* cfgrtl.c (rtl_split_block, rtl_split_edge,
+	safe_insert_insn_on_edge, cfg_layout_split_edge): Adjust.
+	* flow.c (update_life_info, calculate_global_regs_live,
+	allocate_bb_life_data, init_propagate_block_info): Adjust.
+	* global.c (build_insn_chain): Adjust.
+	* ifcvt.c (dead_or_predicable): Adjust.
+	* loop-iv.c (simplify_using_initial_values): Adjust.
+	* recog.c (peephole2_optimize): Adjust.
+	* sched-deps.c (init_deps_global): Adjust.
+
+2004-11-25  Ralf Corsepius <ralf.corsepius@rtems.org>
+
+	* config.gcc (avr-*-rtems*): Fix typo.
+
+2004-11-25  J"orn Rennecke <joern.rennecke@st.com>
+
+	* sh.c (sh_output_mi_thunk): Do not call regset_release_memory.
+
+2004-11-24  Jeff Law  <law@redhat.com>
+
+	* tree-ssa-alias.c (compute_flow_insensitive_aliasing): Reduce
+	the number of queries to random elements in the ai->written_vars
+	bitmap.
+
+2004-11-24  Roger Sayle  <roger@eyesopen.com>
+
+	* config/i386/i386.c (override_options): Disable x87 fancy math
+	intrinsics if -mfpmath= doesn't include 387 (default on x86_64).
+
+2004-11-24  Roger Sayle  <roger@eyesopen.com>
+
+	* configure.ac: Tweak test for HAVE_DECL_LDGETNAME to avoid a
+	system header conflict on AIX 5.2.
+	* configure: Regenerate.
+
+2004-11-24  Kazu Hirata  <kazu@cs.umass.edu>
+
+	* tree-outof-ssa.c (coalesce_abnormal_edges): Use e->dest_idx
+	instead of calling phi_arg_from_edge.
+
+	* tree-phinodes.c (remove_phi_args): Replace phi_arg_from_edge
+	with e->dest_idx.
+
+2004-11-24  Ben Elliston  <bje@au.ibm.com>
+
+	* config/i386/i386.h (ASM_OUTPUT_DWARF_ADDR_CONST): Remove.
+	* config/i386/i386.c (i386_dwarf_output_addr_const): Likewise.
+	* config/i386/i386-protos.h (i386_dwarf_output_addr_const):
+	Likewise.
+	
+2004-11-24  Richard Kenner  <kenner@vlsi1.ultra.nyu.edu>
+
+	* tree-sra.c (sra_walk_modify_expr): Handle RHS first, then LHS.
+
+	* tree-inline.c (copy_body_r): Explicitly copy a constant if the
+	type will be remapped.
+
+2004-11-24  Steven Bosscher  <stevenb@suse.de>
+
+	* c-opts.c (c_common_post_options): Don't clear
+	flag_inline_functions.
+	* dojump.c (clear_pending_stack_adjust): Remove check on
+	flag_inline_functions, it's always true.
+	* config/alpha/alpha.md (movdi_er_maybe_g): Remove splitter
+	that can never trigger.
+	* config/c4x/c4x.h (TARGET_CPU_CPP_BUILTINS): Don't look at
+	flag_inline_trees, now that flag_inline_functions is never
+	cleared.
+	* config/pdp11/pdp11.h (OPTIMIZATION_OPTIONS): Don't set
+	flag_inline_functions at optimization levels greater than
+	or equal to 3.  This is already done by default.
+
+2004-11-24  Kazu Hirata  <kazu@cs.umass.edu>
+
+	* tree-outof-ssa.c (eliminate_build): Use g->e->dest_idx
+	instead to find the PHI argument.  Do not take I as an
+	argument.
+	(eliminate_phi): Adjust the call to eliminate_build.  Do not
+	take I as an argument.
+	(rewrite_trees): Adjust the call to eliminate_phi.
+
+2004-11-24  Richard Sandiford  <rsandifo@redhat.com>
+
+	* optabs.h (force_expand_binop): Declare.
+	* optabs.c (force_expand_binop): Export.
+	* stmt.c (shift_return_value): Delete.
+	(expand_return): Don't call it.
+	* expr.h (shift_return_value): Declare.
+	* calls.c (shift_returned_value): Delete in favor of...
+	(shift_return_value): ...this new function.  Leave the caller to check
+	for non-BLKmode values passed in the msb of a register.  Take said mode
+	and a shift direction as argument.  Operate on the hard function value,
+	not a pseudo.
+	(expand_call): Adjust accordingly.
+	* function.c (expand_function_start): If a non-BLKmode return value
+	is padded at the last significant end of the return register, use the
+	return value's natural mode for the DECL_RESULT, not the mode of the
+	padded register.
+	(expand_function_end): Shift the same sort of return values left by
+	the appropriate amount.
+
+2004-11-24  Matt Austern  <austern@apple.com>
+
+	* recog.c (recog_memoized_1): Remove.
+	* recog.h (recog_memoized_1): Remove declaration.
+	(recog_memoized): Change from macro to inline function.
+	
+2004-11-24  Devang Patel  <dpatel@apple.com>
+
+	PR/18555
+	* gcc.h (DEFAULT_WORD_SWITCH_TAKES_ARG): Add isysroot.
+	* config/darwin-c.c (darwin_register_frameworks): Use sysroot.
+
+2004-11-24  Devang Patel  <dpatel@apple.com>
+
+	* gcc.c (process_command): Supply -v to linker.
+	
+2004-11-24  David Edelsohn  <edelsohn@gnu.org>
+	    Paolo Bonzini  <bonzini@gnu.org>
+
+	* config/rs6000/rs6000.c (rs6000_return_in_memory): Allow Altivec
+	vector modes without ALTIVEC_ABI.  Use GCC vector instead of
+	synthetic vector.
+	(rs6000_pass_by_reference): Split conditional into pieces.  Use
+	GCC vector instead of synthetic vector.
+
+2004-11-24  Aldy Hernandez  <aldyh@redhat.com>
+
+	* tree.c (recompute_tree_invarant_for_addr_expr): The address of a
+	thread-local variable is invariant.
+
+2004-11-24  Richard Kenner  <kenner@vlsi1.ultra.nyu.edu>
+
+	* tree-pretty-print.c (dump_generic_node, case POINTER_TYPE):
+	Handle TYPE_REF_CAN_ALIAS_ALL.
+	(print_declaration): Print array dimensions like dump_generic_node.
+
+	* fold-const.c (operand_equal_p): Remove kludge allowing ARG0 and
+	ARG1 to be null; instead define OP_SAME and OP_SAME_NULL and use them.
+
+2004-11-24  Nathan Sidwell  <nathan@codesourcery.com>
+
+	* bitmap.c (INLINE): Do not define.
+	(bitmap_elem_to_freelist, bitmap_element_free,
+	bitmap_element_allocate, bitmap_clear, bitmap_element_zerop,
+	bitmap_element_link, bitmap_find_bit): Use inline keyword/macro.
+
+2004-11-24  Kazu Hirata  <kazu@cs.umass.edu>
+
+	* tree-ssa-dom.c (cprop_into_successor_phis): Remove code to
+	find the index of a PHI argument.  Use e->dest_idx instead.
+	Replace hint with index.
+
+	* tree-ssa-dom.c (cprop_into_successor_phis): Replace index
+	with indx.
+
+2004-11-24  Diego Novillo  <dnovillo@redhat.com>
+
+	* tree-ssa-alias.c (merge_pointed_to_info): Fix comment
+	regarding PT_MALLOC.
+
+2004-11-24  Joseph Myers  <joseph@codesourcery.com>
+
+	* config/i386/i386.c (optimization_options): Use
+	SUBTARGET_OPTIMIZATION_OPTIONS.
+	* config/i386/sol2-10.h (SUBTARGET_OPTIMIZATION_OPTIONS): Define.
+	(TARGET_SUBTARGET_DEFAULT): Don't include
+	MASK_OMIT_LEAF_FRAME_POINTER.
+
+2004-11-24  Mark Mitchell  <mark@codesourcery.com>
+
+	* config/i386/sol2.h (X86_FILE_START_VERSION_DIRECTIVE): 
+	Define to false.
+
+2004-11-24  Joseph Myers  <joseph@codesourcery.com>
+
+	* config/i386/t-sol2-10 (MULTILIB_DIRNAMES, MULTILIB_OSDIRNAMES):
+	Use amd64 instead of 64.
+
+2004-11-24  Mark Mitchell  <mark@codesourcery.com>
+            Joseph Myers  <joseph@codesourcery.com>
+
+	* crtstuff.c (IN_LIBGCC2): Define it.
+	(EH_FRAME_SECTION_CONST): Check EH_TABLES_CAN_BE_READ_ONLY
+	instead of HAVE_LD_RO_RW_SECTION_MIXING.
+	* defaults.h (EH_TABLES_CAN_BE_READ_ONLY): New macro.
+	* dwarf2out.c (named_section_eh_frame_section): Check
+	EH_TABLES_CAN_BE_READ_ONLY.
+	* except.c (default_exception_section): Likewise.
+	* config/i386/sol2.h (EH_TABLES_CAN_BE_READ_ONLY): Define.
+	* doc/tm.texi (EH_TABLES_CAN_BE_READ_ONLY): Document.
+
+2004-11-24  Kazu Hirata  <kazu@cs.umass.edu>
+
+	* tree-ssa-pre.c (phi_translate): Use find_edge to find the
+	index of a PHI argument.
+
+	* tree-ssa-loop-manip.c (lv_adjust_loop_header_phi): Use
+	find_edge to find the index of a PHI argument.
+
+2004-11-24  Steven Bosscher  <stevenb@suse.de>
+
+	* expr.c (expand_expr_real_1): Remove cases for EXIT_BLOCK_EXPR 
+	and LABELED_BLOCK_EXPR.
+	* gimplify.c (gimplify_labeled_block_expr): Remove.
+	(gimplify_exit_block_expr): Remove.
+	(gimplify_expr): Don't call them.
+	* tree-inline.c (copy_body_r): Don't handle EXIT_BLOCK_EXPR
+	and LABELED_BLOCK_EXPR.
+	(estimate_num_insns_1): Likewise.
+	(walk_tree): Likewise.
+	* tree-pretty-print.c (dump_generic_node): Don't handle
+	EXIT_BLOCK_EXPR and LABELED_BLOCK_EXPR.
+	* tree.def (EXIT_BLOCK_EXPR): Moved to java-tree.def.
+	(LABELED_BLOCK_EXPR): Likewise.
+	* tree.h (LABELED_BLOCK_LABEL): Moved to java-tree.h.
+	(LABELED_BLOCK_BODY): Likewise.
+	(EXIT_BLOCK_LABELED_BLOCK): Likewise.
+	(EXIT_BLOCK_RETURN): Removed.
+
+2004-11-24  Steven Bosscher  <stevenb@suse.de>
+
+	* cfgrtl.c (rtl_delete_block): Fix comment.
+	* emit-rtl.c (remove_unnecessary_notes): Die if we see BLOCK_BEG
+	or BLOCK_END insn notes.
+	* jump.c (squeeze_notes): Likewise.
+	* haifa-sched.c (reemit_notes): Don't "re-emit" EH_REGION_BEG and
+	EH_REGION_END notes, we never have them to begin with.
+	* sched-deps.c (sched_analyze_insn): When updating loop notes,
+	verify that we have indeed only recorded loop notes.
+	(sched_analyze): Die if we see EH_REGION_BEG or EH_REGION_END notes.
+	Only record loop notes.
+
+	* cfgexpand.c (tree_expand_cfg): Fix comment.
+
+	* passes.c (rest_of_compilation): Don't do a second call to
+	convert_from_eh_region_ranges from here, it's already called
+	from cfgexpand.c.
+	* except.c (resolve_fixup_regions): Remove.
+	(remove_fixup_regions): Remove.
+	(convert_from_eh_region_ranges_1): Remove.
+	(convert_from_eh_region_ranges): Remove the case where EH is
+	not already lowered at the tree level.  We always lower there.
+
+2004-11-24  Paolo Bonzini  <bonzini@gnu.org>
+
+	PR c++/16882
+	* tree.c (make_vector_type): Move qualifiers to the vector type,
+	use the inner type's main variant and build a main variant for
+	the vector type if necessary.
+	(type_hash_eq): Check a vector type's TYPE_VECTOR_SUBPARTS.
+
+2004-11-24  Eric Botcazou  <ebotcazou@libertysurf.fr>
+
+	* target.h (late_rtl_prologue_epilogue): Remove.
+	* target-def.h (TARGET_LATE_RTL_PROLOGUE_EPILOGUE): Remove.
+	* system.h: Poison TARGET_LATE_RTL_PROLOGUE_EPILOGUE.
+	* passes.c (rest_of_handle_prologue_epilogue): Remove and move
+	remaining bits to...
+	(rest_of_handle_flow2): ...here.
+	(rest_of_compilation): Remove call to rest_of_handle_prologue_epilogue.
+	* doc/tm.texi (TARGET_LATE_RTL_PROLOGUE_EPILOGUE): Remove.
+
+2004-11-23  Mark Mitchell  <mark@codesourcery.com>
+
+	* hwint.h (HOST_LONG_LONG_FORMAT): New macro.  Use it throughout.
+	* config/i386/xm-mingw32.h (HOST_LONG_LONG_FORMAT): Define.
+	* doc/hostconfig.texi (HOST_LONG_LONG_FORMAT): Document.
+
+2004-11-24  Daniel Jacobowitz  <dan@codesourcery.com>
+
+	* config/i386/i386.c (override_options): Move
+	SUBTARGET_OVERRIDE_OPTIONS before defaulting
+	flag_omit_frame_pointer.
+
+2004-11-23  Richard Henderson  <rth@redhat.com>
+
+	* rtl.h (validate_subreg): Declare.
+	* emit-rtl.c (validate_subreg): New.
+	(gen_rtx_SUBREG): Use it.
+	* simplify-rtx.c (simplify_subreg): Likewise.
+	(simplify_gen_subreg): Likewise.  Remove duplicate asserts.
+	* expr.c (emit_move_insn_1): Tidy complex move code.  Use memory
+	fallback whenever gen_realpart/gen_imagpart would not be able to
+	create SUBREGs.
+
+2004-11-23  Richard Henderson  <rth@redhat.com>
+
+	* expmed.c (extract_bit_field): Use simplify_gen_subreg instead of
+	hard-coding avoiding calls to gen_rtx_SUBREG.  Split complex return
+	modes to CONCAT.
+
+2004-11-23  Diego Novillo  <dnovillo@redhat.com>
+
+	PR tree-optimization/18618
+	* tree-into-ssa.c (DEF_VEC_MALLOC_P(basic_block)): Declare.
+	(insert_phi_nodes_for): Change third argument to VEC(basic_block).
+	* tree-ssa-dom.c (avail_exprs_stack): Change type to VEC(tree_on_heap).
+	Update all users.
+	(stmts_to_rescan): Likewise.
+	(const_and_copies_stack): Likewise.
+	(nonzero_vars_stack): Likewise.
+	(vrp_variables_stack): Likewise.
+
+2004-11-23  Richard Henderson  <rth@redhat.com>
+
+	* cse.c (record_jump_cond_subreg): New.
+	(record_jump_cond): Use it instead of gen_lowpart.
+
+2004-11-23  Daniel Jacobowitz  <dan@codesourcery.com>
+	    Joseph Myers  <joseph@codesourcery.com>
+
+	* config/sol2-c.c (solaris_register_pragmas): Use
+	c_register_pragma_with_expansion.
+	* config/sol2.h (HANDLE_PRAGMA_PACK_WITH_EXPANSION): Define.
+	* c-pragma.c (c_register_pragma): Update call to
+	cpp_register_pragma.
+	(c_register_pragma_with_expansion): New function.
+	(init_pragma): Honor HANDLE_PRAGMA_PACK_WITH_EXPANSION.
+	* c-pragma.h (c_register_pragma_with_expansion): New prototype.
+	* doc/extend.texi (Solaris Pragmas): Mention macro expansion for
+	#pragma align.
+	* doc/tm.texi (c_register_pragma_with_expansion,
+	HANDLE_PRAGMA_PACK_WITH_EXPANSION): Document.
+
+2004-11-23  Richard Henderson  <rth@redhat.com>
+
+	* combine.c (gen_lowpart_for_combine): Factor out mode of x as well
+	as mode sizes into local temporaries.  Unify failure path.
+
+2004-11-23  Richard Henderson  <rth@redhat.com>
+
+	* emit-rtl.c, rtl.h (subreg_hard_regno): Remove.
+	* caller-save.c (mark_set_regs): Use subreg_regno instead.
+	* final.c (alter_subreg): Likewise.
+	* local-alloc.c (reg_is_born): Likewise.
+
+2004-11-23  Richard Henderson  <rth@redhat.com>
+
+	* simplify-rtx.c (simplify_subreg): Use subreg_regno_offset directly
+	instead of using a SUBREG temporary.
+
+2004-11-23  Kazu Hirata  <kazu@cs.umass.edu>
+
+	* tree-cfg.c (tree_forwarder_block_p): Speed up by reordering
+	two checks.
+
+2004-11-23  Zack Weinberg  <zack@codesourcery.com>
+
+	* function.h (struct function): Remove calls_longjmp.
+	(current_function_calls_longjmp): Delete.
+	* tree.h (ECF_LONGJMP): Delete.
+	(ECF_SIBCALL, ECF_PURE, ECF_SP_DEPRESSED, ECF_ALWAYS_RETURN)
+	(ECF_LIBCALL_BLOCK): Everybody slide down one.
+	(ECF_CONST, ECF_NORETURN, ECF_SIBCALL): Clarify comments.
+	* builtins.c (expand_builtin_longjmp): Don't set
+	current_function_calls_longjmp.
+	* calls.c (special_function_p): Mark longjmp and siglongjmp
+	with ECF_NORETURN, not ECF_LONGJMP.
+	(emit_call_1, expand_call, emit_library_call_value_1):
+	Don't check for ECF_LONGJMP.
+	* tree-cfg.c (make_exit_edges, is_ctrl_altering_stmt)
+	(need_fake_edge_p): Likewise.
+
+	* config/avr/avr.h, config/ip2k/ip2k.h: Don't define NON_SAVING_SETJMP.
+	* system.h: Poison NON_SAVING_SETJMP.
+	* function.c (use_register_for_decl)
+	* gcse.c (compute_hash_table_work, compute_store_table)
+	* postreload-gcse.c (record_opr_changes)
+	* reload.c (find_equiv_reg)
+	* reload1.c (reload)
+	* config/i386/i386.c (ix86_can_use_return_insn_p):
+	Remove code conditional on NON_SAVING_SETJMP.
+	* doc/tm.texi: Delete documentation of NON_SAVING_SETJMP.
+
+	* config/i386/sysv3.h: Delete file.
+	* config/i386/i386.c (ix86_svr3_asm_out_constructor): Delete.
+
+2004-11-23  Dorit Naishlos  <dorit@il.ibm.com>
+
+	* tree-vectorizer.c (vect_analyze_operations): Don't vectorize if
+	the vectorization factor is greater than the iteration count.
+	(vect_gen_niters_for_prolog_loop): No need to calculate min if the
+	iteration count is known.
+
+	(vect_build_symbol_bound): Removed. Its functionality moved to
+	vect_generate_tmps_on_preheader.
+	(vect_generate_tmps_on_preheader): Moved content of
+	vect_build_symbol_bound here instead of calling it.  Changed some
+	variables names.
+
+	(vect_build_loop_niters): We don't expect a new_bb to be generated -
+	use gcc_assert.
+	(vect_gen_niters_for_prolog_loop): Likewise.
+	(vect_gen_niters_for_prolog_loo): Likewise.
+
+	(slpeel_make_loop_iterate_ntimes): Use buildN instead of build.
+	(slpeel_add_loop_guard): Likewise.
+	(slpeel_tree_peel_loop_to_edge): Likewise.
+	(vect_do_peeling_for_alignment): Likewise.
+
+	(vect_get_first_index): Missing space.
+
+2004-11-23  Ulrich Weigand  <uweigand@de.ibm.com>
+
+	* config/s390/s390.md ("*iordi3"): Mark commutative.
+	("*iorsi3_zarch", "*iorsi3_esa"): Likewise.
+
+2004-11-23  Kazu Hirata  <kazu@cs.umass.edu>
+
+	* tree-phinode.c (resize_phi_node): Abort when LEN is strictly
+	greater than PHI_ARG_CAPACITY.
+	(reserve_phi_args_for_new_edge): Initialize the new PHI
+	argument to NULL_TREE.  Increment PHI_NUM_ARGS.
+	(add_phi_arg): Add a PHI argument to the slot given by
+	E->dest_idx.
+	(remove_phi_arg_num): Do not write to PHI_ARG_EDGE.
+	* tree-flow-inline (phi_arg_from_edge): Return E->dest_idx.
+	* tree-ssa.c (ssa_redirect_edge): Check for a missing PHI
+	argument by looking at PHI_ARG_DEF.
+	(verify_phi_args): Check for a missing PHI argument.  Remove
+	the check for duplicate PHI arguments.
+	* tree.h (PHI_ARG_EDGE): Redefine in terms of EDGE_PRED.
+	(phi_arg_d): Remove e.
+
+	* tree-cfg.c (tree_try_redirect_by_replacing_jump): Speed up
+	by restricting to the case with two outgoing edges.
+
+2004-11-23  Andreas Krebbel  <krebbel1@de.ibm.com>
+
+	* config/s390/s390.c (s390_backchain_string): Removed.
+	(s390_return_addr_rtx, s390_back_chain_rtx, s390_frame_info)
+	(s390_emit_prologue, s390_va_start, s390_gimplify_va_arg):
+	Changed users of TARGET_BACKCHAIN, TARGET_NO_BACKCHAIN and
+	TARGET_KERNEL_BACKCHAIN to reflect the new options.
+	* config/s390/s390.h (MASK_BACKCHAIN, MASK_PACKED_STACK): New macros.
+	(TARGET_KERNEL_BACKCHAIN): Removed.
+	(TARGET_BACKCHAIN): Former triple state option change to a target flag.
+	(TARGET_PACKED_STACK): New macro.
+	(TARGET_SWITCHES): New switches added.
+	(TARGET_OPTIONS): "backchain", "no-backchain" and "kernel-backchain"
+	removed.
+	(DYNAMIC_CHAIN_ADDRESS): Use TARGET_PACKED_STACK.
+	* config/s390/s390.md ("allocate_stack", "restore_stack_block")
+	("save_stack_nonlocal", "restore_stack_nonlocal"): Modified to reflect
+	the change in target switch semantics.
+	* config/s390/tpf.h (TARGET_DEFAULT_BACKCHAIN): Removed.
+	(TARGET_DEFAULT): MASK_BACKCHAIN added.
+	* doc/invoke.texi: Added documentation for the new/changed options.
+
+2004-11-23  Uros Bizjak  <uros@kss-loka.si>
+
+	PR rtl-optimization/18614
+	* simplify-rtx.c (simplify_binary_operation): Do not
+	simplify inner elements of constant arguments of
+	VEC_CONCAT insn.
+
+2004-11-23  Eric Botcazou  <ebotcazou@libertysurf.fr>
+
+	* gthr-solaris.h (__gthread_recursive_mutex_init_function): Use
+	0 instead of NULL.
+
+2004-11-23  Kazu Hirata  <kazu@cs.umass.edu>
+
+	* tree-phinode.c (make_phi_node): Use a new variable,
+	capacity, to receive the return value of ideal_phi_node_len.
+
+2004-11-23  Eric Botcazou  <ebotcazou@libertysurf.fr>
+
+	* gthr-solaris.h (__gthread_recursive_mutex_t): New type.
+	(__GTHREAD_RECURSIVE_MUTEX_INIT_FUNCTION): Define to
+	__gthread_recursive_mutex_init_function.
+	(__gthread_key_create): Properly cast -1.
+	(__gthread_recursive_mutex_init_function): New function.
+	(__gthread_recursive_mutex_lock): Rewrite.
+	(__gthread_recursive_mutex_trylock): Likewise.
+	(__gthread_recursive_mutex_unlock): Likewise.
+
+2004-11-23  Ralf Corsepius <ralf.corsepius@rtems.org>
+
+	* config/c4x/t-rtems: New.
+	* config.gcc: Reflect having added c4x/t-rtems.
+
+2004-11-23  Dorit Naishlos <dorit@il.ibm.com>
+
+	PR tree-opt/18403
+	PR tree-opt/18505
+	* tree-vectorizer.c (vect_create_data_ref_ptr): Use
+	lang_hooks.types.type_for_size instead of integer_type_node for the
+	type of ptr_update.
+
+2004-11-23  Ralf Corsepius <ralf.corsepius@rtems.org>
+
+	* config.gcc (h8300-*-rtems*): Use h8300/t-rtems.
+
+2004-11-23  Ralf Corsepius <ralf.corsepius@rtems.org>
+
+	* config/arm/t-rtems: New.
+	* config.gcc: Reflect having added arm/t-rtems.
+
+2004-11-23  Kazu Hirata  <kazu@cs.umass.edu>
+
+	* tree-cfg.c (tree_execute_on_growing_pred): New.
+	(tree_cfg_hooks): Add tree_execute_on_growing_pred.
+	* tree-flow.h: Add a prototype for
+	reserve_phi_args_for_new_edge.
+	* tree-phinodes.c (reserve_phi_args_for_new_edge): New.
+	(add_phi_arg): Don't resize a PHI array.
+
+2004-11-23  Ralf Corsepius <ralf.corsepius@rtems.org>
+
+	* config.gcc: Add avr-*-rtems*.
+	* config/avr/t-rtems: New.
+	* config/avr/rtems.h: New.
+
+2004-11-22  Daniel Berlin  <dberlin@dberlin.org>
+
+	* df.c (dataflow_set_a_op_b): Remove reslt parameter, since it's
+	always the same as op1.  Use bitmap_ior_into and bitmap_and_into.
+	Add comment to function.
+
+2004-11-22  John David Anglin  <dave.anglin@nrc-cnrc.gc.ca>
+
+	PR pch/14940
+	* config/host-linux.c (TRY_EMPTY_VM_SPACE): Define for __hppa__.
+
+	PR rtl-optimization/14838
+	* emit-rtl.c (get_first_nonnote_insn): Don't assume first insn is a
+	note.
+	(get_last_nonnote_insn): Don't assume last insn is a note.
+
+2004-11-22  Roger Sayle  <roger@eyesopen.com>
+
+	* fold-const.c (nondestructive_fold_binary_to_constant): Rename
+	to fold_binary_to_constant.
+	(nondestructive_fold_unary_to_constant): Likewise, rename to
+	fold_unary_to_constant.
+	(fold_relational_hi_lo): Update call to fold_binary_to_constant.
+	* tree.h (nondestructive_fold_binary_to_constant): Update prototype.
+	(nondestructive_fold_unary_to_constant): Likewise.
+	* tree-ssa-ccp.c (ccp_fold): Update calls to fold_unary_to_constant
+	and fold_binary_to_constant.
+	* tree-ssa-loop-niter.c (EXEC_BINARY, EXEC_UNARY): Delete macros.
+	(inverse, number_of_iterations_cond): Replace uses of EXEC_BINARY
+	and EXEC_UNARY with calls to fold_*nary_to_constant.
+	* tree-ssa-loop-ivopts.c (EXEC_BINARY, EXEC_UNARY): Delete macros.
+	(idx_find_step): Replace uses of EXEC_BINARY with calls to
+	fold_binary_to_constant.
+
+2004-11-22  Nick Clifton  <nickc@redhat.com>
+
+	* sbitmap.c (sbitmap_union_of_preds): Remove redundant
+	initialisation of 'e'.
+
+2004-11-22  Roger Sayle  <roger@eyesopen.com>
+
+	* config/i386/i386.h (TARGET_USE_FANCY_MATH_387): New macro.
+	* config/i386/i386.c (override_options):  Set MASK_NO_FANCY_MATH_387
+	automatically for targets without TARGET_80387.
+	* config/i386/i386.md (sqrtsf2, sqrtsf2_1, sqrtsf2_i387, sqrtdf2,
+	sqrtdf2_1, sqrtdf2_i387, *sqrtextendsfdf2, sqrtxf2,
+	*sqrtextenddfxf2, *sqrtextendsfxf2, fpremxf4, fmodsf3, fmoddf3,
+	fmodxf3, fprem1xf4, dremsf3, dremdf3, dremxf3, *sindf2, *sinsf2,
+	*sinextendsfdf2, *sinxf2, *cosdf2, *cossf2, *cosextendsfdf2,
+	*cosxf2, sincosdf3, sincossf3, *sincosextendsfdf3, sincosxf3,
+	*tandf3_1, tandf2, *tansf3_1, tansf2, *tanxf3_1, tanxf2,
+	atan2df3_1, atan2df3, atandf2, atan2sf3_1, atan2sf3, atansf2,
+	atan2xf3_1, atan2xf3, atanxf2, asindf2, asinsf2, asinxf2,
+	acosdf2, acossf2, acosxf2, fyl2x_xf3, logsf2, logdf2, logxf2,
+	log10sf2, log10df2, log10xf2, log2sf2, log2df2, log2xf2,
+	fyl2xp1_xf3, log1psf2, log1pdf2, log1pxf2, *fxtractxf3, logbsf2,
+	logbdf2, logbxf2, ilogbsi2, *f2xm1xf2, *fscalexf4, expsf2,
+	expdf2, expxf2, exp10sf2, exp10df2, exp10xf2, exp2sf2, exp2df2,
+	exp2xf2, expm1df2, expm1sf2, expm1xf2, frndintxf2, rintdf2,
+	rintsf2, rintxf2, frndintxf2_floor, floordf2, floorsf2,
+	floorxf2, frndintxf2_ceil, ceildf2, ceilsf2, ceilxf2,
+	frndintxf2_trunc, btruncdf2, btruncsf2, btruncxf2,
+	frndintxf2_mask_pm, nearbyintdf2, nearbyintsf2, nearbyintxf2):
+	Simplify conditionals using TARGET_USE_FANCY_MATH_387.
+
+2004-11-22  Dale Johannesen  <dalej@apple.com>
+
+	* config/darwin.h (JUMP_TABLES_IN_TEXT_SECTION):  Conditionalize
+	properly and comment.
+
+2004-11-22  James A. Morrison  <phython@gcc.gnu.org
+
+	* config/sparc/sparc.c: Include insn-codes.h and langhooks.h.
+	(sparc_init_builtins): New function.
+	(sparc_init_vis_builtins): Create builtin functions for VIS
+	instructions.
+	(sparc_expand_builtin): Expand builtin functions for VIS instructions.
+	(TARGET_INIT_BUILTINS): Define to sparc_init_builtins.
+	(TARGET_EXPAND_BUILTIN): Define to sparc_expand_builtin.
+	(def_builtin): New macro for creating builtin functions.
+	(P): New mode macro for pointer types.
+	(UNSPEC_FPACK16, UNSPEC_FPACK32, UNSPEC_FPACKFIX, UNSPEC_FEXPAND,
+	UNSPEC_FPMERGE, UNSPEC_MUL16AL, UNSPEC_MUL8UL, UNSPEC_MULDUL,
+	UNSPEC_ALIGNDATA, UNSPEC_ALIGNADDR, UNSPEC_PDIST): New constants.
+	(fpack16_vis, fpackfix_vis, fpack32_vis, fexpand_vis, fpmerge_vis,
+	fmul8x16_vis, fmul8x16au_vis, fmul8x16al_vis, fmul8sux16_vis,
+	fmul8ulx16_vis, fmuld8sux16_vis, fmuld8ulx16_vis, pdist_vis,
+	faligndata<V64:mode>_vis, alignaddr<P:mode>_vis): New patterns.
+
+2004-11-22  Devang Patel  <dpatel@apple.com>
+
+	* tree-if-conv.c (clean_predicate_lists): Clean all basic blocks.
+
+2004-11-22  Devang Patel  <dpatel@apple.com>
+
+	* tree-if-conv.c (if_convertable_phi_p): Rename to ...
+	(if_convertible_phi_p): ... new name.
+	(if_convertable_modify_expr_p): Rename to ...
+	(if_convertible_modify_expr_p): ... new name.
+	(if_convertable_stmt_p): Rename to ...
+	(if_convertible_stmt_p): ... new name.
+	(if_convertable_bb_p): Rename to ...
+	(if_convertible_bb_p): ... new name.
+	(if_convertable_loop_p): Rename to ...
+	(if_convertible_loop_p): ... new name.
+
+2004-11-22  Eric Botcazou  <ebotcazou@libertysurf.fr>
+
+	PR target/18444
+	* configure.ac (threading): Accept 'posix95'.
+	* configure: Regenerate.
+	* config.gcc (i[34567]86-*-solaris2*): On Solaris up to 2.6,
+	include sol26.h and default to posix95 threads if have_pthread_h.
+	(sparc-*-solaris2*): Likewise.
+	* gthr.h: Include gthr-posix95.h if _PTHREADS95 is defined.
+	* gthr-posix95.h: New file.
+	* config/sol26.h: New file.
+	* doc/install.texi (--enable-threads): Document 'posix95'.
+
+2004-11-23  Ben Elliston  <bje@au.ibm.com>
+
+	* pointer-set.h (pointer_set_contains): Remove.
+	* pointer-set.c (pointer_set_contains): Likewise.
+
+2004-11-22  Dale Johannesen  <dalej@apple.com>
+
+	* calls.c (expand_call): Do not call preserve_temp_slots.
+
+2004-11-22  Kazu Hirata  <kazu@cs.umass.edu>
+
+	* tree-flow.h: Remove the prototype for remove_phi_arg.
+	Add a prototype for remove_phi_args.
+	* tree-phinodes.c (remove_phi_arg): Remove.
+	(remove_phi_args): New.
+	* tree-ssa.c (ssa_remove_edge): Call remove_phi_args instead
+	of remove_phi_arg.
+
+	* cfg.c (unchecked_make_edge): Call execute_on_growing_pred
+	after making an edge.
+	(remove_edge): Call execute_on_shrinking_pred before removing
+	an edge.
+	(redirect_edge_succ): Call execute_on_growing_pred and
+	execute_on_shrinking_pred.
+	* cfghooks.c (execute_on_growing_pred): New.
+	(execute_on_shrinking_pred): Likewise.
+	* cfghooks.h (cfg_hooks): Add execute_on_growing_pred and
+	execute_on_shrinking_pred.
+	Add prototypes for execute_on_growing_pred and
+	execute_on_shrinking_pred.
+	* cfgrtl.c (rtl_cfg_hooks): Add NULL hooks to
+	execute_on_growing_pred and execute_on_shrinking_pred.
+	(cfg_layout_rtl_cfg_hook): Likewise.
+	* tree-cfg.c (tree_cfg_hooks): Likewise.
+
+	* tree-cfg.c (tree_execute_on_shrinking_pred): New.
+	(tree_cfg_hooks): Use it.
+	* tree-ssa.c (ssa_remove_edge): Don't call remove_phi_args.
+	(ssa_redirect_edge): Don't call remove_phi_arg_num.
+
+	* tree-phinodes.c (remove_phi_arg_num): Make it static.
+	* tree-flow.h: Remove the corresponding prototype.
+
+	* tree-ssa.c (ssa_remove_edge): Remove.
+	* tree-flow.h: Remove the corresponding prototype.
+	* tree-cfg.c: Replace ssa_remove_edge with remove_edge.
+	* basic-block.h: Likewise.
+	* tree-if-conv.c: Likewise.
+	* tree-ssa-threadupdate.c: Likewise.
+
+2004-11-23  Ben Elliston  <bje@au.ibm.com>
+
+	* doc/cfg.texi (Maintaining the CFG): Use @ftable instead of
+	@table to list the BSI methods.  Use individual @findex directives
+	instead of one with a comma-separated list of index entries.
+
+2004-11-22  Zdenek Dvorak  <dvorakz@suse.cz>
+
+	PR tree-optimization/18529
+	* fold-const.c (fold_to_nonsharp_ineq_using_bound): New function.
+	(simple_operand_p): Use STRIP_NOPS.  Consider SSA names simple.
+	(fold): Call fold_to_nonsharp_ineq_using_bound.
+	* tree-ssa-loop-niter.c (simplify_replace_tree): New function.
+	(number_of_iterations_cond): Fold the expressions before futher
+	processing.
+	(tree_simplify_using_condition): Handle case when cond or expr is
+	an EQ_EXPR specially.
+
+2004-11-22 Daniel Berlin  <dberlin@dberlin.org>
+
+	* tree-ssa.c (verify_ssa): SSA_OP_ALL_USES should be
+	SSA_OP_ALL_USES | SSA_OP_ALL_KILLS for verification purposes.
+
+2004-11-22  David Edelsohn  <edelsohn@gnu.org>
+	    Fariborz Jahanian  <fjahanian@apple.com>
+
+	* config/rs6000/rs6000.c (rs6000_return_in_memory): Test
+	TARGET_ALTIVEC_ABI.
+	(function_arg_boundary): Align 8 to 16 byte vectors like SPE.
+	Align 16 byte and larger vectors like Altivec.
+	(function_arg_advance): Pass any 16 byte vector like Altivec vector.
+	(function_arg): Same.
+	(rs6000_pass_by_references): Test TARGET_ALTIVEC_ABI.
+
+2004-11-22  Andrew Pinski  <pinskia@physics.uc.edu>
+
+	PR tree-opt/18572
+	* tree-sra.c (scalarize_init): Unshare the rhs before gimplifying
+	it.
+
+2004-11-22  Richard Henderson  <rth@redhat.com>
+
+	* function.c (assign_parm_setup_block): Revert 11-13 change to
+	forcably re-mode a (possibly BLKmode) hard register.
+
+2004-11-22  Eric Botcazou  <ebotcazou@libertysurf.fr>
+
+	* config/sparc/sparc.c (mem_min_alignment): Check MEM_ALIGN.
+
+2004-11-22  David Edelsohn  <edelsohn@gnu.org>
+
+	* config/rs6000/power4.md: Remove delay between dispatch and issue
+	associated with dispatch group slots 3 and 4.
+	* config/rs6000/power5.md: Same.
+
+2004-11-21  Jeff Law  <law@redhat.com>
+
+	* cfg.c (cached_make_edge): Use find_edge rather than an inlined
+	variant.
+	* cfgbuild.c (make_edges): Likewise.
+	* cfghooks.c (can_duplicate_block_p): Likewise.
+	* cfgloop.c (loop_latch_edge): Likewise.
+	* cfgloopmanip.c (force_single_succ_latches): Likewise.
+	* cfgrtl.c (rtl_flow_call_edges_add): Likewise.
+	* predict.c (predict_loops, propagate_freq): Likewise.
+	* tracer.c (tail_duplicate): Likewise.
+	* tree-cfg.c (disband_implicit_edges): Likewise.
+	(tree_forwarder_block_p, tree_flow_call_edges_add): Likewise.
+
+2004-11-22  Nick Clifton  <nickc@redhat.com>
+
+	* sbitmap.c (sbitmap_union_of_preds): Remove redundant
+	initialisation of 'e'.
+
+2004-11-22  Kazu Hirata  <kazu@cs.umass.edu>
+
+	PR rtl-optimization/18599
+	* regrename.c (copyprop_hardreg_forward): Speed up by putting
+	BB_VISITED flags on basic blocks as we process them.
+
+2004-11-22  Nathan Sidwell  <nathan@codesourcery.com>
+
+	* config/rs6000/altivec.md (VI_char): New mode attribute.
+	(addv16qi3, addv8hi3, addv4ai3): Replace with ...
+	(add<mode>3): ... this.
+	(subv16qi3, subv8hi3, subv4ai3): Replace with ...
+	(sub<mode>3): ... this.
+	(smaxv16qi3, smaxv8hi3, smaxv4ai3): Replace with ...
+	(smax<mode>3): ... this.
+	(sminv16qi3, sminv8hi3, sminv4ai3): Replace with ...
+	(smin<mode>3): ... this.
+	(umaxv16qi3, umaxv8hi3, umaxv4ai3): Replace with ...
+	(umax<mode>3): ... this.
+	(uminv16qi3, uminv8hi3, uminv4ai3): Replace with ...
+	(umin<mode>3): ... this.
+	(andv16qi3, andv8hi3, andv4ai3): Replace with ...
+	(and<mode>3): ... this.
+	(iorv16qi3, iorv8hi3, iorv4ai3): Replace with ...
+	(ior<mode>3): ... this.
+	(xorv16qi3, xorv8hi3, xorv4ai3): Replace with ...
+	(xor<mode>3): ... this.
+	(andv16qi3, andv8hi3, andv4ai3): Replace with ...
+	(and<mode>3): ... this.
+	(iorv16qi3, iorv8hi3, iorv4ai3): Replace with ...
+	(ior<mode>3): ... this.
+	(altivec_vnor): Replace with ...
+	(altivec_nor<mode>): ... this.
+	(one_cmplv16qi2, one_cmplv8hi2, one_complv4ai2): Replace with ...
+	(one_cmpl<mode>2): ... this.
+	(altivec_vandc): New expander.
+	(*andc<mode>3): New insn.
+	* config/rs6000/rs6000.c (bdesc_2arg): Adjust for new insn names.
+
+2004-11-22  Dorit Naishlos  <dorit@il.ibm.com>
+
+	PR tree-opt/18536
+	* tree-vectorizer.c (make_loop_iterate_ntimes): Use type of niters when
+	creating iv instead of integer_type.
+	(vect_build_loop_niters): Remove redundant code.
+	(vect_transform_loop_bound): Removed (duplicates functionality of
+	make_loop_iterate_ntimes.
+	(vect_gen_niters_for_prolog_loop): Add documentation. Call
+	lang_hooks.types.type_for_size to create a type of the required size.
+	Use that type instead of integer_type. Remove redundant code.
+	(vect_update_niters_after_peeling): Use type of niters instead of
+	integer_type.
+	(vect_transform_loop): Unify handling of known and unknown loop bound
+	cases. Call make_loop_iterate_ntimes instead of
+	vect_transform_loop_bound.
+
+2004-11-22  Andrew Pinski  <pinskia@physics.uc.edu>
+
+	PR target/18217
+	* config/rs6000/t-darwin (LIB2FUNCS_EXTRA): Move darwin-fallback.c to
+	(LIB2ADDEH): Here.
+
+2004-11-22  Dorit Naishlos  <dorit@il.ibm.com>
+
+	PR tree-opt/18544
+	* tree-vectorizer.c (vect_analyze_data_refs): Fail if memtag not found.
+
+2004-11-22 Kazuhiro Inaoka <inaoka.kazuhiro@renesas.com>
+
+	* config.gcc: Use t-slibgcc-elf to build libgcc_s.so on
+	m32r*linux.
+	* config/m32r/t-linux (SHLIB_MAPFILES): Override to use
+	m32r/libgcc-glibc.ver.
+	* config/m32r/libgcc-glibc.ver: Add New file.
+
+2004-11-22  Nathan Sidwell  <nathan@codesourcery.com>
+
+	* bitmap.h (struct bitmap_obstack): New obstack type.
+	(struct bitmap_head_def): Replace using_obstack with obstack
+	pointer.
+	(bitmap_default_obstack): New.
+	(bitmap_initialize): Make inline, does not do allocation.
+	(bitmap_release_memory): Remove.
+	(bitmap_obstack_initialize, bitmap_obstack_release): Declare.
+	(bitmap_obstack_alloc, bitmap_malloc_alloc, bitmap_gc_alloc,
+	bitmap_obstack_free, bitmap_malloc_free): Declare.
+	(BITMAP_OBSTACK_ALLOC, BITMAP_GGC_ALLOC, BITMAP_XMALLOC): Adjust.
+	(BITMAP_FREE): Replace with ...
+	(BITMAP_OBSTACK_FREE): ... this.
+	(BITMAP_XFREE): Adjust.
+	(BITMAP_INIT_ONCE): Remove.
+	* bitmap.c (bitmap_obstack, bitmap_obstack_init, bitmap_free: Remove.
+	(bitmap_default_obstack): New.
+	(bitmap_elem_to_freelist): Adjust.
+	(bitmap_element_allocate): Adjust. Break initialization into ...
+	(bitmap_obstack_initialize): ... here.
+	(bitmap_release_memory): Replace with ...
+	(bitmap_obstack_release): ... this.
+	(bitmap_obstack_alloc, bitmap_malloc_alloc, bitmap_gc_alloc,
+	bitmap_obstack_free, bitmap_malloc_free): New.
+	(bitmap_ior_and_compl, bitmap_ior_and_compl_into): Use
+	bitmap_initialize.
+	(bitmap_initialize): Move to bitmap.h.
+	* gengtype.c (open_base_files): Add obstack.h to ifiles.
+	* Makefile.in (BASIC_BLOCK_H, REGS_H): Add obstack.h.
+	* basic-block.h (INIT_REG_SET): Allocate from reg_obstack.
+	(INITIALIZE_REG_SET): Remove.
+	(FREE_REG_SET): Use BITMAP_OBSTACK_FREE.
+	(INIT_ONCE_REG_SET, MAX_REGNO_REG_SET): Remove.
+	(flow_obstack): Do not declare.
+	(reg_obstack): Declare.
+	* regs.h: Include obstack.h.
+	* tree-optimize.c (tree_rest_of_compilation): Initialize and
+	release bitmap obstack here.
+	* bb-reorder.c: #include regs, not basic-block.
+	(fix_crossing_conditional_branches): Allocate regsets from
+	reg_obstack.
+	* bt-load.c: Do not inlude bitmap.h, sbitmap.h, basic-block.h or
+	obstack.h.
+	* caller-save.c: Include regs.h earlier.
+	* cfg.c: Do not include basic-block.h or obstack.h.
+	(reg_obstack): Define.
+	* cfganal.c: Include obstack.h
+	* cfgcleanyp.c: Do not include basic-block.h. Include regs.h
+	earlier.
+	* cfglayout.c: Do not include obstack.h.
+	(flow_obstack): Remove declaration.
+	(cfg_layout_duplicate_bb): Use reg_obstack.
+	* cfgloop.c, cfgloopanal.c, cfgloopmanip.c: Include obstack.h.
+	* cfgrtl.c (rtl_split_block): Use reg_obstack.
+	(force_nonfallthru_and_redirect, rtl_split_edge): Likewise.
+	(safe_insert_insn_on_edge): Use OBSTACK_ALLOC_REG_SET, adjust.
+	(cfg_layout_split_edge): Use reg_obstack.
+	* cse.c: Include regs.h earlier.
+	* ddg.c: Do not include basic-block.h.
+	* dominance.c: Inlude obstack.h.
+	* flow.c (update_life_info): Use OBSTACK_ALLOC_REG_SET, adjust.
+	(calculate_global_regs_live): Likewise.
+	(allocate_bb_life_data): Use reg_obstack.
+	(init_propagate_block_info): Use OBSTACK_ALLOC_REGSET.
+	* global.c: Do not include basic-block.h.
+	(build_insn_chain): Use OBSTACK_ALLOC_REG_SET, adjust.
+	* graph.c: Include obstack.h.
+	* haifa-sched.c: Do not include basic-block.h.
+	* ifcvt.c: Use OBSTACK_ALLOC_REG_SET, adjust.
+	* local-alloc.c: Do not include basic-block.h.
+	* loop-init.c, loop-invariant.c: Include obstack.h.
+	* loop-iv.c: Likewise.
+	(simplify_using_initial_values): Use OBSTACK_ALLOC_REG_SET,
+	adjust.
+	* loop-unroll.c, loop-unswitch.c: Inlude obstack.h.
+	* modulo-sched.c: Do not include basic-block.h.
+	* passes.c (rest_of_handle_final): Do not call
+	regset_release_memory.
+	* ra-debug.c: Include regs.h earlier. Do not include
+	basic-block.h.
+	* recog.c (peephole2_optimize): Use OBSTACK_ALLOC_REG_SET, adjust.
+	* regclass.c (init_reg_sets): Do not call INIT_ONCE_REG_SET.
+	(allocate_reg_info): Do not call MAX_REGNO_REG_SET.
+	(regset_release_memory): Remove.
+	* resource.c: Do not include basic-block.h.
+	* rtlanal.c: Do not include basic-block.h.
+	* sbitmap.c: Include obstack.h.
+	* sched-deps.c: Do not include basic-block.h.
+	(reg_pending_sets_head, reg_pending_clobbers_head,
+	reg_pending_uses_head): Remove.
+	(init_deps_global): Use OBSTACK_ALLOC_REG_SET.
+	* sched-ebb.c: Do not include basic-block.h.
+	* sched-rgn.c: Likewise.
+	* tree-if-conv.c (get_loop_body_in_if_conv_order): Use
+	BITMAP_XFREE.
+	* tree-outof-ssa.c (perform_edge_inserts): Use BITMAP_XFREE.
+	* tree-sra.c (decide_instantiations): Adjust bitmap
+	initialization.
+	* tree-ssa-dce.c: Include obstack.h.
+	* tree-ssa-pre.c (grand_bitmap_obstack): Make a bitmap_obstack.
+	(value_insert_into_set_bitmap): Remove useless bitmap_clear.
+	(bitmap_set_new): Likewise.
+	(init_pre): Initialize bitmap obstack.
+	(fini_pre): Release bitmap obstack.
+	* tree-ssanames.c (ssa_names_to_rewrite): Make static.
+	(marked_for_rewrite_p): ssa_names_to_rewrite is never NULL.
+	(mark_for_rewrite, unmark_for_rewrite): Likewise.
+	(marked_ssa_names): Likewise.
+	(init_ssanames): Use BITMAP_XMALLOC.
+	(fini_ssanames): Use BITMAP_XFREE.
+	* web.c: Include obstack.h
+
+2004-11-22  Nick Clifton  <nickc@redhat.com>
+
+	* sbitmap.c (sbitmap_union_of_preds): Set 'e' to the next edge
+	predecessor in the first for-loop.
+
+2004-11-21  Stan Shebs  <shebs@apple.com>
+
+	* config/rs6000/rs6000.c: (rs6000_darwin64_function_arg): Add
+	UNION_TYPE case.
+	(function_arg): Move darwin test up, sniff mode first for
+	efficiency, test for union type.
+
+2004-11-21  Dale Johannesen  <dalej@apple.com>
+
+	* config/darwin.h (JUMP_TABLES_IN_TEXT_SECTION):  Define.
+
+2004-11-21  Jeff Law  <law@redhat.com>
+
+	* tree-ssa.c (verify_use): Fix comment.

I am friendly script caring about memory consumption in GCC.  Please contact
jh@suse.cz if something is going wrong.

The results can be reproduced by building compiler with
--enable-gather-detailed-mem-stats targetting x86-64 and compiling preprocessed
combine.c or testcase from PR8632 with:

-fmem-report --param=ggc-min-heapsize=1024 --param=ggc-min-expand=1 -Ox -Q

The memory consumption summary appears in the dump after detailed listing of
the places they are allocated in.  Peak memory consumption is actually computed
by looking for maximal value in {GC XXXX -> YYYY} report.

Yours testing script.



More information about the Gcc-regression mailing list