Some aspect of GCC memory consumption increased by recent patch
gcctest@suse.de
gcctest@suse.de
Sat Oct 30 01:36:00 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 needed: 25241k -> 25245k
Peak memory use before GGC: 9329k
Peak memory use after GGC: 8641k
Maximum of released memory in single GGC run: 2926k
Garbage: 42520k
Leak: 6088k
Overhead: 5657k
GGC runs: 354
comparing combine.c compilation at -O1 level:
Overall memory needed: 26577k -> 26573k
Peak memory use before GGC: 9291k -> 9256k
Peak memory use after GGC: 8730k -> 8733k
Maximum of released memory in single GGC run: 2064k -> 2067k
Garbage: 71893k -> 70549k
Leak: 6484k -> 6484k
Overhead: 11597k -> 11476k
GGC runs: 569 -> 563
comparing combine.c compilation at -O2 level:
Overall memory allocated via mmap and sbrk increased from 29805k to 29857k, overall 0.17%
Overall memory needed: 29805k -> 29857k
Peak memory use before GGC: 12705k
Peak memory use after GGC: 12579k
Maximum of released memory in single GGC run: 2578k -> 2579k
Garbage: 87424k -> 85991k
Leak: 6305k -> 6305k
Overhead: 16312k -> 16184k
GGC runs: 570 -> 567
comparing combine.c compilation at -O3 level:
Amount of memory still referenced at the end of compilation increased from 6813k to 6845k, overall 0.47%
Overall memory needed: 21400k -> 21356k
Peak memory use before GGC: 13132k -> 13091k
Peak memory use after GGC: 12579k
Maximum of released memory in single GGC run: 3418k -> 3420k
Garbage: 117386k -> 115186k
Leak: 6813k -> 6845k
Overhead: 21474k -> 21275k
GGC runs: 636 -> 631
comparing insn-attrtab.c compilation at -O0 level:
Overall memory needed: 118512k
Peak memory use before GGC: 79408k
Peak memory use after GGC: 46137k
Maximum of released memory in single GGC run: 43357k
Garbage: 161090k
Leak: 10609k
Overhead: 20460k
GGC runs: 295
comparing insn-attrtab.c compilation at -O1 level:
Peak amount of GGC memory allocated before garbage collecting run decreased from 89840k to 86893k, overall -3.39%
Amount of produced GGC garbage decreased from 462072k to 445811k, overall -3.65%
Overall memory needed: 130240k
Peak memory use before GGC: 89840k -> 86893k
Peak memory use after GGC: 70491k
Maximum of released memory in single GGC run: 40959k
Garbage: 462072k -> 445811k
Leak: 10954k
Overhead: 77023k -> 75545k
GGC runs: 453 -> 451
comparing insn-attrtab.c compilation at -O2 level:
Amount of produced GGC garbage decreased from 511440k to 495175k, overall -3.28%
Overall memory needed: 205748k -> 205752k
Peak memory use before GGC: 105256k -> 102310k
Peak memory use after GGC: 85907k
Maximum of released memory in single GGC run: 35657k
Garbage: 511440k -> 495175k
Leak: 10882k
Overhead: 85123k -> 83644k
GGC runs: 375 -> 373
comparing insn-attrtab.c compilation at -O3 level:
Amount of produced GGC garbage decreased from 512715k to 496448k, overall -3.28%
Overall memory needed: 205756k -> 205752k
Peak memory use before GGC: 105258k -> 102312k
Peak memory use after GGC: 85909k
Maximum of released memory in single GGC run: 35657k
Garbage: 512715k -> 496448k
Leak: 10926k
Overhead: 85307k -> 83828k
GGC runs: 384 -> 382
comparing Gerald's testcase PR8361 compilation at -O0 level:
Overall memory needed: 111960k -> 111956k
Peak memory use before GGC: 88676k
Peak memory use after GGC: 87634k
Maximum of released memory in single GGC run: 19564k
Garbage: 249098k
Leak: 57105k
Overhead: 45784k
GGC runs: 364
comparing Gerald's testcase PR8361 compilation at -O1 level:
Overall memory needed: 94576k
Peak memory use before GGC: 87850k
Peak memory use after GGC: 86403k
Maximum of released memory in single GGC run: 19363k
Garbage: 593544k -> 592152k
Leak: 59030k -> 59030k
Overhead: 132538k -> 132417k
GGC runs: 628 -> 627
comparing Gerald's testcase PR8361 compilation at -O2 level:
Overall memory needed: 95168k -> 95216k
Peak memory use before GGC: 87850k
Peak memory use after GGC: 86403k
Maximum of released memory in single GGC run: 19363k
Garbage: 651202k -> 649821k
Leak: 59617k -> 59617k
Overhead: 158863k -> 158740k
GGC runs: 680 -> 677
comparing Gerald's testcase PR8361 compilation at -O3 level:
Overall memory needed: 102916k -> 102808k
Peak memory use before GGC: 94792k
Peak memory use after GGC: 87880k
Maximum of released memory in single GGC run: 19973k
Garbage: 695233k -> 693804k
Leak: 59953k -> 59953k
Overhead: 171043k -> 170913k
GGC runs: 660 -> 659
Head of changelog is:
--- /usr/src/SpecTests/sandbox-britten-memory/x86_64/mem-result/ChangeLog 2004-10-29 15:11:45.000000000 +0000
+++ /usr/src/SpecTests/sandbox-britten-memory/gcc/gcc/ChangeLog 2004-10-30 00:23:34.000000000 +0000
@@ -1,3 +1,54 @@
+2004-10-30 Kazu Hirata <kazu@cs.umass.edu>
+
+ * tree-phinodes.c (create_phi_node): Don't zero PHI_REWRITTEN.
+
+2004-10-29 Jeff Law <law@redhat.com>
+
+ * tree-ssa-dom.c (struct edge_info): New structure holding
+ edge equivalences and edge redirection information.
+ (get_eq_expr_value, record_dominating_conditions): Kill.
+ (propagate_to_outgoing_edges): Renamed from cprop_into_phis.
+ Call record_edge_info.
+ (allocate_edge_info, free_edge_info): New.
+ (tree_ssa_dominator_optimize): Use propagate_to_outgoing_edges
+ rather than cprop_into_phis. Free all edge infos before threading
+ jumps.
+ (thread_across_edge): Allocate new edge info structures as needed
+ and store the redirection target into the edge info structure
+ instead of the edge's AUX field.
+ (dom_opt_initialize_block): Mark unused argument with ATTRIBUTE_UNUSED.
+ (record_equivalence_from_incoming_edge): Lose unnecessary argument.
+ Revamp code which finds and records equivalences associated with
+ edges to use saved data in the edge_info structure.
+ (record_equivalencs_from_phis): Similarly.
+ (dom_opt_finalize_block): Revamp code which finds and records
+ equivalences associated with edges to use saved data in the
+ edge_info structure.
+ (build_and_record_new_cond): New function.
+ (record_conditions): Use build_and_record_new_cond to record
+ dominating conditions.
+ (record_edge_info): New function.
+ (record_range): Tighten test for conditions which create
+ useful range records.
+
+2004-10-29 Geoffrey Keating <geoffk@apple.com>
+
+ * config/rs6000/darwin-fallback.c: Include <mach/thread_status.h>.
+
+2004-10-29 Stan Shebs <shebs@apple.com>
+
+ * config/rs6000/darwin-tramp.asm: Make more 64-bit-friendly.
+
+2004-10-29 Kazu Hirata <kazu@cs.umass.edu>
+
+ * tree-phinodes.c (allocate_phi_node): New.
+ (make_phi_node, resize_phi_node): Use it.
+
+2004-10-29 David Edelsohn <edelsohn@gnu.org>
+
+ * config/rs6000/sysv4.h (TARGET_POWER): Define as 0.
+ * config/rs6000/darwin.h (TARGET_POWER): Define as 0.
+
2004-10-29 Kazu Hirata <kazu@cs.umass.edu>
* tree-phinodes.c (make_phi_node, resize_phi_node): Don't zero
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