A recent patch increased GCC's memory consumption in some cases!
gcctest@suse.de
gcctest@suse.de
Thu Dec 21 06:35:00 GMT 2006
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 empty function compilation at -O0 level:
Overall memory needed: 18281k -> 18293k
Peak memory use before GGC: 2235k
Peak memory use after GGC: 1942k
Maximum of released memory in single GGC run: 293k
Garbage: 423k -> 423k
Leak: 2273k
Overhead: 446k
GGC runs: 3
comparing empty function compilation at -O0 -g level:
Overall memory needed: 18297k -> 18309k
Peak memory use before GGC: 2263k
Peak memory use after GGC: 1970k
Maximum of released memory in single GGC run: 293k
Garbage: 425k -> 426k
Leak: 2305k
Overhead: 450k
GGC runs: 3
comparing empty function compilation at -O1 level:
Overall memory needed: 18381k -> 18397k
Peak memory use before GGC: 2235k
Peak memory use after GGC: 1942k
Maximum of released memory in single GGC run: 293k
Garbage: 427k -> 427k
Leak: 2275k
Overhead: 446k
GGC runs: 4
comparing empty function compilation at -O2 level:
Overall memory needed: 18393k -> 18409k
Peak memory use before GGC: 2236k
Peak memory use after GGC: 1942k
Maximum of released memory in single GGC run: 294k
Garbage: 430k -> 430k
Leak: 2275k
Overhead: 447k
GGC runs: 4
comparing empty function compilation at -O3 level:
Overall memory needed: 18393k -> 18409k
Peak memory use before GGC: 2236k
Peak memory use after GGC: 1942k
Maximum of released memory in single GGC run: 294k
Garbage: 430k -> 430k
Leak: 2275k
Overhead: 447k
GGC runs: 4
comparing combine.c compilation at -O0 level:
Overall memory needed: 28441k -> 28461k
Peak memory use before GGC: 9288k
Peak memory use after GGC: 8804k -> 8803k
Maximum of released memory in single GGC run: 2642k
Garbage: 37525k -> 37557k
Leak: 6454k
Overhead: 4872k -> 4872k
GGC runs: 280 -> 281
comparing combine.c compilation at -O0 -g level:
Overall memory needed: 30525k -> 30537k
Peak memory use before GGC: 10835k -> 10834k
Peak memory use after GGC: 10464k -> 10463k
Maximum of released memory in single GGC run: 2320k
Garbage: 38104k -> 38132k
Leak: 9330k
Overhead: 5574k -> 5574k
GGC runs: 272
comparing combine.c compilation at -O1 level:
Amount of memory still referenced at the end of compilation increased from 6481k to 6489k, overall 0.12%
Overall memory needed: 29474k
Peak memory use before GGC: 16962k -> 16963k
Peak memory use after GGC: 16792k
Maximum of released memory in single GGC run: 2254k -> 2252k
Garbage: 55702k -> 55738k
Leak: 6481k -> 6489k
Overhead: 9959k -> 9959k
GGC runs: 358
comparing combine.c compilation at -O2 level:
Overall memory needed: 29474k
Peak memory use before GGC: 16967k
Peak memory use after GGC: 16792k
Maximum of released memory in single GGC run: 2363k -> 2371k
Garbage: 71797k -> 71830k
Leak: 6611k -> 6602k
Overhead: 11849k -> 11849k
GGC runs: 412
comparing combine.c compilation at -O3 level:
Overall memory needed: 29602k
Peak memory use before GGC: 18066k -> 18067k
Peak memory use after GGC: 17599k -> 17600k
Maximum of released memory in single GGC run: 3676k -> 3678k
Garbage: 106280k -> 106302k
Leak: 6684k
Overhead: 16905k -> 16904k
GGC runs: 460
comparing insn-attrtab.c compilation at -O0 level:
Amount of produced GGC garbage increased from 132119k to 132390k, overall 0.21%
Overall memory needed: 89654k -> 89650k
Peak memory use before GGC: 71193k
Peak memory use after GGC: 44700k -> 44699k
Maximum of released memory in single GGC run: 37868k
Garbage: 132119k -> 132390k
Leak: 9518k -> 9278k
Overhead: 16954k -> 16954k
GGC runs: 211 -> 212
comparing insn-attrtab.c compilation at -O0 -g level:
Amount of memory still referenced at the end of compilation increased from 10981k to 11221k, overall 2.19%
Overall memory needed: 90826k
Peak memory use before GGC: 72355k -> 72354k
Peak memory use after GGC: 45967k -> 45966k
Maximum of released memory in single GGC run: 37869k -> 37868k
Garbage: 133525k -> 133284k
Leak: 10981k -> 11221k
Overhead: 17349k -> 17349k
GGC runs: 209
comparing insn-attrtab.c compilation at -O1 level:
Overall memory needed: 93726k -> 93730k
Peak memory use before GGC: 71858k
Peak memory use after GGC: 67990k
Maximum of released memory in single GGC run: 31671k -> 31668k
Garbage: 229875k -> 229900k
Leak: 9343k
Overhead: 29520k -> 29520k
GGC runs: 220
comparing insn-attrtab.c compilation at -O2 level:
Overall memory needed: 123986k -> 124006k
Peak memory use before GGC: 79550k
Peak memory use after GGC: 73714k -> 73715k
Maximum of released memory in single GGC run: 30216k -> 30212k
Garbage: 282685k -> 282733k
Leak: 9345k
Overhead: 35777k -> 35781k
GGC runs: 243 -> 244
comparing insn-attrtab.c compilation at -O3 level:
Ovarall memory allocated via mmap and sbrk decreased from 128814k to 123766k, overall -4.08%
Overall memory needed: 128814k -> 123766k
Peak memory use before GGC: 79575k -> 79576k
Peak memory use after GGC: 73740k
Maximum of released memory in single GGC run: 30410k -> 30407k
Garbage: 283525k -> 283552k
Leak: 9350k
Overhead: 36009k -> 36009k
GGC runs: 247
comparing Gerald's testcase PR8361 compilation at -O0 level:
Amount of produced GGC garbage increased from 211915k to 212865k, overall 0.45%
Overall memory needed: 119066k
Peak memory use before GGC: 92332k -> 92331k
Peak memory use after GGC: 91417k
Maximum of released memory in single GGC run: 19250k -> 19248k
Garbage: 211915k -> 212865k
Leak: 48116k
Overhead: 21222k -> 21222k
GGC runs: 416 -> 417
comparing Gerald's testcase PR8361 compilation at -O0 -g level:
Amount of produced GGC garbage increased from 218494k to 219448k, overall 0.44%
Overall memory needed: 131642k
Peak memory use before GGC: 104689k -> 104692k
Peak memory use after GGC: 103652k
Maximum of released memory in single GGC run: 18932k -> 18936k
Garbage: 218494k -> 219448k
Leak: 71548k
Overhead: 27127k -> 27126k
GGC runs: 389 -> 390
comparing Gerald's testcase PR8361 compilation at -O1 level:
Overall memory needed: 119502k
Peak memory use before GGC: 96665k
Peak memory use after GGC: 94462k
Maximum of released memory in single GGC run: 17940k
Garbage: 442643k -> 442789k
Leak: 50182k -> 50182k
Overhead: 103830k -> 103827k
GGC runs: 562
comparing Gerald's testcase PR8361 compilation at -O2 level:
Overall memory needed: 119554k
Peak memory use before GGC: 96692k
Peak memory use after GGC: 94489k
Maximum of released memory in single GGC run: 18081k
Garbage: 497476k -> 497651k
Leak: 51150k
Overhead: 58753k -> 58753k
GGC runs: 615 -> 616
comparing Gerald's testcase PR8361 compilation at -O3 level:
Overall memory needed: 121138k
Peak memory use before GGC: 97648k
Peak memory use after GGC: 96144k
Maximum of released memory in single GGC run: 18476k
Garbage: 517975k -> 518089k
Leak: 51123k
Overhead: 58758k -> 58756k
GGC runs: 621
comparing PR rtl-optimization/28071 testcase compilation at -O0 level:
Overall memory needed: 137642k -> 137634k
Peak memory use before GGC: 81588k -> 81587k
Peak memory use after GGC: 58467k
Maximum of released memory in single GGC run: 45167k -> 45166k
Garbage: 148519k -> 148526k
Leak: 7542k
Overhead: 25329k
GGC runs: 82
comparing PR rtl-optimization/28071 testcase compilation at -O0 -g level:
Overall memory needed: 138014k
Peak memory use before GGC: 82234k -> 82233k
Peak memory use after GGC: 59113k
Maximum of released memory in single GGC run: 45232k
Garbage: 148730k -> 148736k
Leak: 9309k
Overhead: 25824k
GGC runs: 88
comparing PR rtl-optimization/28071 testcase compilation at -O1 level:
Overall memory needed: 418466k -> 418794k
Peak memory use before GGC: 199519k
Peak memory use after GGC: 192216k -> 192217k
Maximum of released memory in single GGC run: 94104k -> 109875k
Garbage: 284119k -> 284125k
Leak: 29778k
Overhead: 31548k -> 31548k
GGC runs: 98
comparing PR rtl-optimization/28071 testcase compilation at -O2 level:
Overall memory needed: 343286k -> 343378k
Peak memory use before GGC: 199512k
Peak memory use after GGC: 192209k -> 192210k
Maximum of released memory in single GGC run: 96120k -> 111890k
Garbage: 364357k -> 364364k
Leak: 30361k
Overhead: 47297k -> 47297k
GGC runs: 104
comparing PR rtl-optimization/28071 testcase compilation at -O3 -fno-tree-pre -fno-tree-fre level:
Ovarall memory allocated via mmap and sbrk decreased from 771946k to 748734k, overall -3.10%
Overall memory needed: 771946k -> 748734k
Peak memory use before GGC: 317621k -> 317622k
Peak memory use after GGC: 296096k
Maximum of released memory in single GGC run: 168283k -> 186389k
Garbage: 504436k -> 504442k
Leak: 45414k
Overhead: 60274k -> 60274k
GGC runs: 98 -> 99
Head of the ChangeLog is:
--- /usr/src/SpecTests/sandbox-britten-memory/x86_64/mem-result/ChangeLog 2006-12-20 13:14:45.000000000 +0000
+++ /usr/src/SpecTests/sandbox-britten-memory/gcc/gcc/ChangeLog 2006-12-21 04:54:51.000000000 +0000
@@ -1,3 +1,111 @@
+2006-12-20 Roger Sayle <roger@eyesopen.com>
+
+ * simplify-rtx.c (simplify_subreg): Use the correct mode when
+ determining whether a SUBREG of a CONCAT refers to the first or
+ second component.
+
+2006-12-21 Ben Elliston <bje@au.ibm.com>
+
+ * config/spu/spu.c (spu_builtin_mul_widen_even): Remove unused
+ local variable `d'.
+
+2006-12-20 Jan Hubicka <jh@suse.cz>
+
+ * tree-dfa.c (add_referenced_var): Walk initializers of
+ non-constant/readonly static vars.
+
+2006-12-20 Jan Hubicka <jh@suse.cz>
+
+ * tree-flow-inline.h (gimple_var_anns): New function.
+ (var_ann): Use hashtable for static functions.
+ * tree-dfa.c (create_var_ann): Likewise.
+ * tree-ssa.c (var_ann_eq, var_ann_hash): New functions.
+ (init_tree_ssa): Initialize var anns.
+ (delete_tree_ssa): Delete var anns; also clear out gimple_df.
+ * tree-flow.h (struct static_var_ann_d): New structure.
+ (gimple_df): Add var_anns.
+
+2006-12-20 Carlos O'Donell <carlos@codesourcery.com>
+
+ PR bootstrap/30242
+ * gcc/c-incpath.c (add_standard_paths): Only relocate paths that
+ begin with the configured prefix.
+
+2006-12-20 Jan Hubicka <jh@suse.cz>
+
+ PR target/30213
+ * i386.c (expand_setmem_epilogue): Fix formating.
+ (dsmalest_pow2_greater_than): New function.
+ (ix86_expand_movmem): Improve comments; avoid re-computing of
+ epilogue size.
+ (promote_duplicated_reg_to_size): Break out from ...
+ (expand_setmem): ... this one; reorganize promotion code;
+ improve comments; avoid recomputation of epilogue size.
+
+2006-12-20 Andrew Pinski <pinskia@gmail.com>
+
+ PR middle-end/30143
+ * omp-low.c (init_tmp_var): New function.
+ (save_tmp_var): New function.
+ (lower_omp_1): Use them for VAR_DECL.
+
+2006-12-20 Andrew Pinski <pinskia@gmail.com>
+
+ * tree-gimple.c (is_gimple_min_invariant): Treat constant vector
+ CONSTRUCTORs as min invariants.
+
+2006-12-20 Joseph Myers <joseph@codesourcery.com>
+
+ * rtlanal.c (struct subreg_info, subreg_get_info, subreg_nregs):
+ New.
+ (subreg_regno_offset, subreg_offset_representable_p): Change to
+ wrappers about subreg_get_info.
+ (refers_to_regno_p, reg_overlap_mentioned_p): Use subreg_nregs.
+ * rtl.h (subreg_nregs): Declare.
+ * doc/tm.texi (HARD_REGNO_NREGS_HAS_PADDING): Update to refer to
+ subreg_get_info.
+ * caller-save.c (mark_set_regs, add_stored_regs): Use
+ subreg_nregs.
+ * df-scan.c (df_ref_record): Use subreg_nregs.
+ * flow.c (mark_set_1): Use subreg_nregs.
+ * postreload.c (move2add_note_store): Use subreg_nregs.
+ * reload.c (decompose, refers_to_regno_for_reload_p,
+ reg_overlap_mentioned_for_reload_p): Use subreg_nregs.
+ * resource.c (update_live_status, mark_referenced_resources,
+ mark_set_resources): Use subreg_nregs.
+
+2006-12-20 Zdenek Dvorak <dvorakz@suse.cz>
+
+ * loop-unswitch.c (unswitch_loop): Update arguments of
+ duplicate_loop_to_header_edge call.
+ * tree-ssa-loop-ivcanon.c (try_unroll_loop_completely): Ditto.
+ * loop-unroll.c (peel_loop_completely, unroll_loop_constant_iterations,
+ unroll_loop_runtime_iterations, peel_loop_simple, unroll_loop_stupid):
+ Ditto.
+ * cfgloopmanip.c (loop_version): Ditto.
+ (duplicate_loop_to_header_edge): Change
+ type of to_remove to VEC(edge), remove n_to_remove argument.
+ * tree-ssa-loop-manip.c (tree_duplicate_loop_to_header_edge):
+ Change type of to_remove to VEC(edge), remove n_to_remove argument.
+ (tree_unroll_loop): Update arguments of
+ tree_duplicate_loop_to_header_edge call.
+ * cfghooks.c (cfg_hook_duplicate_loop_to_header_edge):
+ Change type of to_remove to VEC(edge), remove n_to_remove argument.
+ * cfghooks.h (struct cfg_hooks): Type of
+ cfg_hook_duplicate_loop_to_header_edge changed.
+ (cfg_hook_duplicate_loop_to_header_edge): Declaration changed.
+ * cfgloop.h (duplicate_loop_to_header_edge): Ditto.
+ * tree-flow.h (tree_duplicate_loop_to_header_edge): Ditto.
+
+2006-12-20 Dorit Nuzman <dorit@il.ibm.com>
+
+ * config/spu/spu.md (vec_widen_umult_hi_v8hi): New.
+ (vec_widen_umult_lo_v8hi, vec_widen_smult_hi_v8hi): New.
+ (vec_widen_smult_lo_v8hi): New.
+ * config/spu/spu.c (TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN): Defined.
+ (TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD): Defined.
+ (spu_builtin_mul_widen_even, spu_builtin_mul_widen_odd): New.
+
2006-12-20 Jan Hubicka <jh@suse.cz>
* cgraph.c: Update overall comment; fix vertical spacing.
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.
More information about the Gcc-regression
mailing list