GCC memory consumption increased by recent patch!

[gcctest at suse dot de]

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


comparing combine.c compilation at -O0 level:
Peak memory use before GGC: 12120k -> 12120k
Peak memory use after GGC: 11641k -> 11641k
Maximum of released memory in single GGC run: 2368k -> 2368k
Garbage: 42976k -> 42976k
Leak: 10910k -> 10910k
Overhead: 5810k -> 5810k
GGC runs: 447 -> 447

comparing combine.c compilation at -O1 level:
Peak amount of GGC memory allocated before garbage collecting increased from 11101k to 11131k, overall 0.27%
Peak amount of GGC memory still allocated after garbage collectin increased from 10424k to 10454k, overall 0.29%
Amount of produced GGC garbage increased from 79689k to 86850k, overall 8.99%
Amount of memory still referenced at the end of compilation increased from 9880k to 10005k, overall 1.26%
Peak memory use before GGC: 11101k -> 11131k
Peak memory use after GGC: 10424k -> 10454k
Maximum of released memory in single GGC run: 1726k -> 1767k
Garbage: 79689k -> 86850k
Leak: 10005k -> 9880k
Overhead: 13224k -> 14989k
GGC runs: 655 -> 668

comparing combine.c compilation at -O2 level:
Peak amount of GGC memory allocated before garbage collecting increased from 13545k to 13613k, overall 0.50%
Amount of produced GGC garbage increased from 95694k to 102943k, overall 7.58%
Amount of memory still referenced at the end of compilation increased from 9922k to 10025k, overall 1.04%
Peak memory use before GGC: 13545k -> 13613k
Peak memory use after GGC: 13160k -> 13160k
Maximum of released memory in single GGC run: 2103k -> 2104k
Garbage: 95694k -> 102943k
Leak: 10025k -> 9922k
Overhead: 18928k -> 19886k
GGC runs: 665 -> 673

comparing combine.c compilation at -O3 level:
Peak amount of GGC memory allocated before garbage collecting increased from 17431k to 17554k, overall 0.71%
Peak amount of GGC memory still allocated after garbage collectin increased from 15406k to 15529k, overall 0.80%
Amount of produced GGC garbage increased from 124342k to 135722k, overall 9.15%
Amount of memory still referenced at the end of compilation increased from 12005k to 12211k, overall 1.72%
Peak memory use before GGC: 17431k -> 17554k
Peak memory use after GGC: 15406k -> 15529k
Maximum of released memory in single GGC run: 3010k -> 3047k
Garbage: 124342k -> 135722k
Leak: 12211k -> 12005k
Overhead: 26002k -> 26200k
GGC runs: 721 -> 727

comparing Gerald's testcase PR8361 compilation at -O0 level:
Peak memory use before GGC: 93183k -> 93183k
Peak memory use after GGC: 91694k -> 91694k
Maximum of released memory in single GGC run: 20289k -> 20289k
Garbage: 255124k -> 255124k
Leak: 72853k -> 72853k
Overhead: 34609k -> 34609k
GGC runs: 538 -> 538

comparing Gerald's testcase PR8361 compilation at -O1 level:
Amount of produced GGC garbage increased from 624678k to 675502k, overall 8.14%
Amount of memory still referenced at the end of compilation increased from 88861k to 90181k, overall 1.49%
Peak memory use before GGC: 97174k -> 97174k
Peak memory use after GGC: 90561k -> 90561k
Maximum of released memory in single GGC run: 19488k -> 19488k
Garbage: 624678k -> 675502k
Leak: 90181k -> 88861k
Overhead: 142560k -> 147559k
GGC runs: 792 -> 820

comparing Gerald's testcase PR8361 compilation at -O2 level:
Amount of produced GGC garbage increased from 690390k to 742184k, overall 7.50%
Amount of memory still referenced at the end of compilation increased from 87885k to 89222k, overall 1.52%
Peak memory use before GGC: 97175k -> 97175k
Peak memory use after GGC: 90562k -> 90562k
Maximum of released memory in single GGC run: 19489k -> 19489k
Garbage: 690390k -> 742184k
Leak: 89222k -> 87885k
Overhead: 165700k -> 170323k
GGC runs: 840 -> 867

comparing Gerald's testcase PR8361 compilation at -O3 level:
Amount of produced GGC garbage increased from 729959k to 793516k, overall 8.71%
Amount of memory still referenced at the end of compilation increased from 89010k to 90657k, overall 1.85%
Peak memory use before GGC: 92864k -> 92864k
Peak memory use after GGC: 91396k -> 91396k
Maximum of released memory in single GGC run: 20194k -> 20194k
Garbage: 729959k -> 793516k
Leak: 90657k -> 89010k
Overhead: 184792k -> 185145k
GGC runs: 846 -> 866

Head of changelog is:

--- /usr/src/SpecTests/sandbox-britten-memory/x86_64/mem-result/ChangeLog	2004-09-05 03:25:24.000000000 +0000
+++ /usr/src/SpecTests/sandbox-britten-memory/gcc/gcc/ChangeLog	2004-09-05 12:55:23.000000000 +0000
@@ -1,3 +1,63 @@
+2004-09-05  Diego Novillo  <dnovillo@redhat.com>
+
+	* tree-if-conv.c (gate_tree_if_conversion): Enable only if the
+	vectorizer is enabled.
+
+2004-09-05  Andreas Jaeger  <aj@suse.de>
+
+	* tree-if-conv.c: Spell check comments and clean up whitespace.
+
+2004-09-05  Zdenek Dvorak  <rakdver@atrey.karlin.mff.cuni.cz>
+
+	* tree-ssa-loop-ivopts.c: New file.
+	* Makefile.in (tree-ssa-loop-ivopts.c): Add.
+	* cfgloop.h (target_avail_regs, target_res_regs, target_small_cost,
+	target_pres_cost, target_spill_cost): Declare.
+	* cfgloopanal.c (avail_regs, res_regs, small_cost, pres_cost,
+	spill_cost): Renamed to ...
+	(target_avail_regs, target_res_regs, target_small_cost,
+	target_pres_cost, target_spill_cost): ... and exported.
+	(init_set_costs, global_cost_for_size): Work with renamed variables.
+	* common.opt (flag_ivopts): New flag.
+	* expr.c (expand_expr_real_1): Handle SSA_NAME case.  Handle
+	REF_ORIGINAL.
+	* gimplify.c (struct gimplify_ctx): Add into_ssa field.
+	(internal_get_tmp_var, gimplify_modify_expr, gimplify_expr): Support
+	generating SSA form.
+	(force_gimple_operand): New function.
+	* timevar.def (TV_TREE_LOOP_IVOPTS): New timevar.
+	* tree-cfg.c (stmt_bsi): New function.
+	* params.def (PARAM_IV_CONSIDER_ALL_CANDIDATES_BOUND,
+	PARAM_IV_MAX_CONSIDERED_USES): New.
+	* tree-flow.h (stmt_bsi, tree_ssa_iv_optimize, split_loop_exit_edge,
+	bsi_insert_on_edge_immediate_loop. standard_iv_increment_position,
+	ip_end_pos, ip_normal_pos, force_gimple_operand): Declare.
+	* tree-gimple.c (is_gimple_formal_tmp_var): Accept ssa names.
+	* tree-nested.c (build_addr): Export.
+	* tree-optimize.c (init_tree_optimization_passes): Add
+	pass_iv_optimize.
+	* tree-pass.h (pass_iv_optimize): Declare.
+	* tree-ssa-loop-im.c (for_each_index): Handle REALPART_EXPR and
+	IMAGPART_EXPR.
+	* tree-ssa-loop-manip.c (create_iv): Force the base to be acceptable
+	as a phi node argument.
+	(split_loop_exit_edge, bsi_insert_on_edge_immediate_loop,
+	ip_end_pos, ip_normal_pos, standard_iv_increment_position): New
+	functions.
+	* tree-ssa-loop-niter.c (zero_p, unsigned_type_for): Export.
+	* tree-ssa-loop.c (tree_ssa_loop_ivopts, gate_tree_ssa_loop_ivopts,
+	pass_iv_optimize): New pass.
+	* tree-ssa-operands.c (get_indirect_ref_operands): Handle REF_ORIGINAL.
+	* tree-ssanames.c (release_ssa_name): Allow calling with var = NULL.
+	* tree.c (build_int_cst_type, cst_and_fits_in_hwi): New functions.
+	* tree.h (REF_ORIGINAL): New macro.
+	(build_int_cst_type, unsigned_type_for, zero_p,
+	cst_and_fits_in_hwi, build_addr): Declare.
+	* doc/invoke.texi (-fivopts): Document.
+	(PARAM_IV_CONSIDER_ALL_CANDIDATES_BOUND,
+	PARAM_IV_MAX_CONSIDERED_USES): Document.
+	* doc/passes.texi: Document induction variable optimizations pass.
+
 2004-09-04  Kaveh R. Ghazi  <ghazi@caip.rutgers.edu>
 
 	* builtin-attrs.def (ATTR_NOTHROW_SENTINEL_1): New.
@@ -102,7 +162,7 @@
 	Handle new rounding modes.
 
 	* config/i386/i386.h (enum fp_cw_mode): Delete.
-	(MODE_NEEDED): Handle new rounding modes. 
+	(MODE_NEEDED): Handle new rounding modes.
 	(EMIT_MODE_SET): Change condition to handle new rounding modes.
 
 	* config/i386/i386.md (UNSPEC_FRNDINT_FLOOR, UNSPEC_FRNDINT_CEIL,
@@ -167,7 +227,7 @@
 	* tree-ssa-operands.c (get_expr_operands): Handle COND_EXPR.
 	* tree-if-conv.c: New file.
 	* doc/passes.texi: Document tree if-conversion pass.
-	* doc/tree-ssa.texi: Same.	
+	* doc/tree-ssa.texi: Same.
 
 2004-09-03  H.J. Lu  <hongjiu.lu@intel.com>
 
@@ -389,7 +449,7 @@
 2004-09-03  Devang Patel  <dpatel@apple.com>
 
 	* dwarf2out.c (gen_field_die). Equate decl number to die.
-	
+
 2004-09-03  Vladimir Makarov  <vmakarov@redhat.com>
 
 	PR target/15832
@@ -417,7 +477,7 @@
 	* tsystem.h: Declare memcpy and memset when inhibit_libc.
 
 2004-09-02  Roman Zippel  <zippel@linux-m68k.org>
- 
+
 	* combine.c (try_combine): Delay modifying of insns which cannot
 	be easily undone.
 
@@ -449,7 +509,7 @@
 	TARGET_BPABI_CPP_BUILTINS.
 
 2004-09-02  Roman Zippel  <zippel@linux-m68k.org>
- 
+
 	* combine.c (distribute_notes): Don't add REG_LABEL to jump insn.
 
 2004-09-02  Eric Christopher  <echristo@redhat.com>
@@ -7091,7 +7151,7 @@
 	(mover2side, moverside, movemside, movei, op3, andu): Replace
 	live define_peephole:s with define_peephole2 near-equivalents.
 	Delete the rest.
- 	("*mov_sidesisf_biap"): Rename from "*mov_sidesi_biap".  Match all
+	("*mov_sidesisf_biap"): Rename from "*mov_sidesi_biap".  Match all
 	word-size modes.
 	("*mov_sidesisf", "*mov_sidesisf_biap_mem", "*mov_sidesisf_mem"):
 	Similar for "*mov_sidesi", "*mov_sidesi_biap_mem" and

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.