A recent patch increased GCC's memory consumption!

[gcctest@suse.de]

Hi,

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

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


comparing combine.c compilation at -O0 level:
  Amount of memory still referenced at the end of compilation increased from 6706k to 6732k, overall 0.38%
    Overall memory needed: 24828k -> 25288k
    Peak memory use before GGC: 9653k -> 9657k
    Peak memory use after GGC: 8965k -> 8970k
    Maximum of released memory in single GGC run: 2790k
    Garbage: 42280k -> 42259k
    Leak: 6706k -> 6732k
    Overhead: 5890k -> 5889k
    GGC runs: 329 -> 328

comparing combine.c compilation at -O1 level:
  Amount of memory still referenced at the end of compilation increased from 7088k to 7097k, overall 0.14%
    Overall memory needed: 27616k -> 27624k
    Peak memory use before GGC: 9161k -> 9165k
    Peak memory use after GGC: 8720k -> 8724k
    Maximum of released memory in single GGC run: 2204k
    Garbage: 63153k -> 63135k
    Leak: 7088k -> 7097k
    Overhead: 7793k -> 7791k
    GGC runs: 521

comparing combine.c compilation at -O2 level:
  Amount of memory still referenced at the end of compilation increased from 7076k to 7087k, overall 0.15%
    Overall memory needed: 24816k -> 24824k
    Peak memory use before GGC: 18274k -> 18282k
    Peak memory use after GGC: 18093k -> 18101k
    Maximum of released memory in single GGC run: 2524k
    Garbage: 86280k -> 86238k
    Leak: 7076k -> 7087k
    Overhead: 10936k -> 10928k
    GGC runs: 479 -> 478

comparing combine.c compilation at -O3 level:
  Amount of memory still referenced at the end of compilation increased from 7147k to 7174k, overall 0.37%
    Overall memory needed: 25060k -> 25056k
    Peak memory use before GGC: 18276k -> 18284k
    Peak memory use after GGC: 18093k -> 18101k
    Maximum of released memory in single GGC run: 3099k
    Garbage: 114883k -> 114882k
    Leak: 7147k -> 7174k
    Overhead: 14474k -> 14471k
    GGC runs: 527 -> 528

comparing insn-attrtab.c compilation at -O0 level:
  Amount of memory still referenced at the end of compilation increased from 11303k to 11542k, overall 2.11%
    Overall memory needed: 85596k -> 85580k
    Peak memory use before GGC: 73395k -> 73379k
    Peak memory use after GGC: 45365k
    Maximum of released memory in single GGC run: 37613k -> 37597k
    Garbage: 153738k -> 153195k
    Leak: 11303k -> 11542k
    Overhead: 19826k -> 19639k
    GGC runs: 268

comparing insn-attrtab.c compilation at -O1 level:
    Overall memory needed: 101156k
    Peak memory use before GGC: 76439k
    Peak memory use after GGC: 65594k
    Maximum of released memory in single GGC run: 37093k -> 37076k
    Garbage: 305059k -> 304775k
    Leak: 11605k -> 11603k
    Overhead: 38740k -> 38553k
    GGC runs: 381

comparing insn-attrtab.c compilation at -O2 level:
    Overall memory needed: 154668k -> 154652k
    Peak memory use before GGC: 121110k
    Peak memory use after GGC: 92200k
    Maximum of released memory in single GGC run: 32950k
    Garbage: 403219k -> 402929k
    Leak: 11445k -> 11445k
    Overhead: 51578k -> 51392k
    GGC runs: 303

comparing insn-attrtab.c compilation at -O3 level:
    Overall memory needed: 154684k
    Peak memory use before GGC: 121110k
    Peak memory use after GGC: 92202k
    Maximum of released memory in single GGC run: 32950k
    Garbage: 403986k -> 403699k
    Leak: 11467k -> 11466k
    Overhead: 51700k -> 51513k
    GGC runs: 309

comparing Gerald's testcase PR8361 compilation at -O0 level:
    Overall memory needed: 127328k -> 127332k
    Peak memory use before GGC: 103080k -> 103083k
    Peak memory use after GGC: 102057k -> 102060k
    Maximum of released memory in single GGC run: 21522k -> 21524k
    Garbage: 247550k -> 247510k
    Leak: 53798k -> 53777k
    Overhead: 43016k -> 42959k
    GGC runs: 347 -> 346

comparing Gerald's testcase PR8361 compilation at -O1 level:
    Overall memory needed: 121432k -> 121444k
    Peak memory use before GGC: 112522k -> 112525k
    Peak memory use after GGC: 100700k -> 100703k
    Maximum of released memory in single GGC run: 20025k -> 20026k
    Garbage: 662821k -> 662526k
    Leak: 58927k -> 58909k
    Overhead: 86717k -> 86378k
    GGC runs: 517

comparing Gerald's testcase PR8361 compilation at -O2 level:
    Overall memory needed: 121424k -> 121428k
    Peak memory use before GGC: 112523k -> 112526k
    Peak memory use after GGC: 100700k -> 100703k
    Maximum of released memory in single GGC run: 20025k -> 20027k
    Garbage: 758463k -> 758068k
    Leak: 59614k -> 59604k
    Overhead: 104856k -> 104509k
    GGC runs: 600

comparing Gerald's testcase PR8361 compilation at -O3 level:
    Overall memory needed: 124364k -> 124368k
    Peak memory use before GGC: 115265k -> 115268k
    Peak memory use after GGC: 102526k -> 102529k
    Maximum of released memory in single GGC run: 21385k -> 21386k
    Garbage: 814674k -> 814285k
    Leak: 60958k -> 60972k
    Overhead: 111421k -> 111049k
    GGC runs: 603 -> 602

Head of the ChangeLog is:

--- /usr/src/SpecTests/sandbox-britten-memory/x86_64/mem-result/ChangeLog	2005-06-02 17:30:02.000000000 +0000
+++ /usr/src/SpecTests/sandbox-britten-memory/gcc/gcc/ChangeLog	2005-06-02 20:26:56.000000000 +0000
@@ -1,3 +1,79 @@
+2005-06-02  Kazu Hirata  <kazu@codesourcery.com>
+
+	* gimplify.c (gimplify_ctx): Change the type of case_labels to
+	VEC from VARRAY.
+	(gimplify_switch_expr, gimplify_case_label_expr): Adjust uses
+	of case_labels.
+
+2005-06-02  Richard Guenther  <rguenth@gcc.gnu.org>
+
+	* c-typeck.c (build_indirect_ref): Build INDIRECT_REF
+	with correct type.
+
+2005-06-02  Ulrich Weigand  <uweigand@de.ibm.com>
+
+	* config/s390/s390.c (s390_add_execute): Do not handle out-of-pool
+	execute templates.
+	(s390_dump_pool): Likewise.
+	(s390_mainpool_start, s390_mainpool_finish): Likewise.
+	(s390_chunkify_start): Likewise.
+	(s390_dump_execute): Remove.
+	(s390_reorg): Handle out-of-pool execute templates.
+
+2005-06-02  Adrian Straetling  <straetling@de.ibm.com>
+
+	* optabs.c: (expand_bool_compare_and_swap): Emit barrier after
+	  unconditional jump.
+
+2005-06-02  Jan Hubicka  <jh@suse.cz>
+
+	* cgraph.c (cgraph_node): Maintain master clones.
+	(cgraph_remove_node): Likewise.
+	(availability_names): New static variable.
+	(dump_cgraph_node): Dump availability.
+	(dump_cgraph_varpool_node): Likewise.
+	(cgraph_is_master_clone, cgraph_master_clone,
+	cgraph_function_body_availability,
+	cgraph_variable_initializer_availability): New functions.
+	* cgraph.h (availability): New enum.
+	(struct cgraph_node): Add master_clone.
+	(cgraph_is_master_clone, cgraph_master_clone,
+	cgraph_function_body_availability,
+	cgraph_variable_initializer_availability): Declare.
+	* cgraphunit.c (cgraph_expand_function): Setcgraph_function_flags_ready.
+	(cgraph_remove_unreachable_nodes): Remove unreachable nodes.
+	* ipa-inline.c (cgraph_decide_inlining): Do not call
+	cgraph_remove_unreachable_nodes.
+
+	* cgraphunit.c (cgraph_function_and_variable_visibility): Fix typo in
+	previous patch.
+
+2005-06-02  Diego Novillo  <dnovillo@redhat.com>
+
+	PR 21582
+	* tree-vrp.c (nonnull_arg_p): New.
+	(get_value_range): Call it.
+
+2005-06-02  Eric Christopher  <echristo@redhat.com>
+
+	* config/mips/mips.h (processor_type): Remove PROCESSOR_DEFAULT,
+	add PROCESSOR_MAX.
+	(mips_rtx_cost_data): New datatype.
+	(MEMORY_MOVE_COST): Use data from structure.
+	(BRANCH_COST): Ditto.
+	(LOGICAL_OP_NON_SHORT_CIRCUIT): Define to zero.
+	* config/mips/mips.md (cpu): Rework for processor_type changes.
+	* config/mips/mips.c (mips_cost): New variable.
+	(DEFAULT_COSTS): Define.
+	(mips_rtx_cost_data): New.
+	(mips_rtx_costs): Use. Minor formatting changes. Use COSTS_N_INSNS
+	for NEG cost. Add support for FLOAT, UNSIGNED_FLOAT, FIX,
+	FLOAT_EXTEND, FLOAT_TRUNCATE, and SQRT.
+	(override_options): Set cost data.
+	(mips_register_move_cost): Formatting changes.
+	(bdesc_arrays): Use PROCESSOR_MAX.
+	(mips_init_builtins): Ditto.
+
 2005-06-02  Diego Novillo  <dnovillo@redhat.com>
 
 	PR 21765
--- /usr/src/SpecTests/sandbox-britten-memory/x86_64/mem-result/ChangeLog.cp	2005-06-02 09:59:38.000000000 +0000
+++ /usr/src/SpecTests/sandbox-britten-memory/gcc/gcc/cp/ChangeLog	2005-06-02 20:26:57.000000000 +0000
@@ -1,5 +1,20 @@
 2005-06-02  Nathan Sidwell  <nathan@codesourcery.com>
 
+	* method.c (synthesize_method): Add addtional arg to warning call.
+
+	PR c++/21280
+	* Make-lang.in (method.o): Add diagnostic.h
+	* decl.c (start_preparsed_function): Use decl's location for file
+	info.
+	* decl2.c (cp_finish_file): Set input_location before synthesizing
+	a function.
+	(mark_used): When deferring a synthesized function, save current
+	location.  Do not set function's location when actually
+	synthesizing it.
+	* method.c: #include diagnostic.h.
+	(synthesize_method): Set the functions source location.  Show
+	needed location if errors are emitted.
+
 	* decl.c (start_decl): Simplify specialization handling. Remove
 	unneeded CLASSTYPE_TEMPLATE_INSTANTIATION check.
 	* mangle.c (discriminator_for_local_entity): Use VEC_index.


The results can be reproduced by building a 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.

Your testing script.