Re: A recent patch increased GCC's memory consumption in some cases!

[Jan Hubicka <jh at suse dot cz>]

This is for switching compiler back to mainline.
combine.c -O1 and -O2 improved noticeably relative to last mainline run.
There is regression on PR27071 -O3 but not too serve.

Honza
> 
> 
> comparing empty function compilation at -O0 level:
>   Peak amount of GGC memory allocated before garbage collecting increased from 2263k to 2324k, overall 2.70%
>   Peak amount of GGC memory still allocated after garbage collecting increased from 1952k to 1997k, overall 2.31%
>   Amount of produced GGC garbage increased from 447k to 479k, overall 7.27%
>   Amount of memory still referenced at the end of compilation increased from 2285k to 2374k, overall 3.87%
>     Overall memory needed: 7491k -> 7412k
>     Peak memory use before GGC: 2263k -> 2324k
>     Peak memory use after GGC: 1952k -> 1997k
>     Maximum of released memory in single GGC run: 311k -> 327k
>     Garbage: 447k -> 479k
>     Leak: 2285k -> 2374k
>     Overhead: 456k -> 516k
>     GGC runs: 3
> 
> comparing empty function compilation at -O0 -g level:
>   Peak amount of GGC memory allocated before garbage collecting increased from 2290k to 2352k, overall 2.71%
>   Peak amount of GGC memory still allocated after garbage collecting increased from 1979k to 2025k, overall 2.32%
>   Amount of produced GGC garbage increased from 449k to 482k, overall 7.18%
>   Amount of memory still referenced at the end of compilation increased from 2318k to 2407k, overall 3.82%
>     Overall memory needed: 7507k -> 7428k
>     Peak memory use before GGC: 2290k -> 2352k
>     Peak memory use after GGC: 1979k -> 2025k
>     Maximum of released memory in single GGC run: 311k -> 327k
>     Garbage: 449k -> 482k
>     Leak: 2318k -> 2407k
>     Overhead: 461k -> 521k
>     GGC runs: 3
> 
> comparing empty function compilation at -O1 level:
>   Peak amount of GGC memory allocated before garbage collecting increased from 2263k to 2324k, overall 2.70%
>   Peak amount of GGC memory still allocated after garbage collecting increased from 1952k to 1997k, overall 2.31%
>   Amount of produced GGC garbage increased from 452k to 485k, overall 7.19%
>   Amount of memory still referenced at the end of compilation increased from 2288k to 2376k, overall 3.87%
>     Overall memory needed: 7527k -> 7516k
>     Peak memory use before GGC: 2263k -> 2324k
>     Peak memory use after GGC: 1952k -> 1997k
>     Maximum of released memory in single GGC run: 311k -> 327k
>     Garbage: 452k -> 485k
>     Leak: 2288k -> 2376k
>     Overhead: 457k -> 517k
>     GGC runs: 4 -> 3
> 
> comparing empty function compilation at -O2 level:
>   Peak amount of GGC memory allocated before garbage collecting increased from 2263k to 2325k, overall 2.74%
>   Peak amount of GGC memory still allocated after garbage collecting increased from 1952k to 1997k, overall 2.31%
>   Amount of produced GGC garbage increased from 455k to 488k, overall 7.17%
>   Amount of memory still referenced at the end of compilation increased from 2288k to 2376k, overall 3.87%
>     Overall memory needed: 7555k -> 7524k
>     Peak memory use before GGC: 2263k -> 2325k
>     Peak memory use after GGC: 1952k -> 1997k
>     Maximum of released memory in single GGC run: 311k -> 328k
>     Garbage: 455k -> 488k
>     Leak: 2288k -> 2376k
>     Overhead: 457k -> 517k
>     GGC runs: 4
> 
> comparing empty function compilation at -O3 level:
>   Peak amount of GGC memory allocated before garbage collecting increased from 2263k to 2325k, overall 2.74%
>   Peak amount of GGC memory still allocated after garbage collecting increased from 1952k to 1997k, overall 2.31%
>   Amount of produced GGC garbage increased from 455k to 488k, overall 7.17%
>   Amount of memory still referenced at the end of compilation increased from 2288k to 2376k, overall 3.87%
>     Overall memory needed: 7555k -> 7524k
>     Peak memory use before GGC: 2263k -> 2325k
>     Peak memory use after GGC: 1952k -> 1997k
>     Maximum of released memory in single GGC run: 311k -> 328k
>     Garbage: 455k -> 488k
>     Leak: 2288k -> 2376k
>     Overhead: 457k -> 517k
>     GGC runs: 4
> 
> comparing combine.c compilation at -O0 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 22651k to 17676k, overall -28.15%
>   Peak amount of GGC memory allocated before garbage collecting run decreased from 9308k to 9015k, overall -3.25%
>   Peak amount of GGC memory still allocated after garbage collecting decreased from 8827k to 8255k, overall -6.93%
>   Amount of produced GGC garbage increased from 37622k to 37687k, overall 0.17%
>   Amount of memory still referenced at the end of compilation increased from 6580k to 7000k, overall 6.37%
>     Overall memory needed: 22651k -> 17676k
>     Peak memory use before GGC: 9308k -> 9015k
>     Peak memory use after GGC: 8827k -> 8255k
>     Maximum of released memory in single GGC run: 2622k -> 1874k
>     Garbage: 37622k -> 37687k
>     Leak: 6580k -> 7000k
>     Overhead: 5250k -> 4753k
>     GGC runs: 282 -> 278
> 
> comparing combine.c compilation at -O0 -g level:
>   Ovarall memory allocated via mmap and sbrk decreased from 24619k to 19572k, overall -25.79%
>   Peak amount of GGC memory allocated before garbage collecting run decreased from 11116k to 10751k, overall -3.40%
>   Peak amount of GGC memory still allocated after garbage collecting decreased from 10579k to 9975k, overall -6.06%
>   Amount of memory still referenced at the end of compilation increased from 9473k to 9892k, overall 4.43%
>     Overall memory needed: 24619k -> 19572k
>     Peak memory use before GGC: 11116k -> 10751k
>     Peak memory use after GGC: 10579k -> 9975k
>     Maximum of released memory in single GGC run: 2595k -> 1558k
>     Garbage: 38213k -> 38022k
>     Leak: 9473k -> 9892k
>     Overhead: 5955k -> 5459k
>     GGC runs: 271 -> 270
> 
> comparing combine.c compilation at -O1 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 40555k to 29988k, overall -35.24%
>   Peak amount of GGC memory allocated before garbage collecting run decreased from 19322k to 17844k, overall -8.28%
>   Peak amount of GGC memory still allocated after garbage collecting decreased from 19127k to 17651k, overall -8.36%
>   Amount of produced GGC garbage decreased from 57382k to 52532k, overall -9.23%
>   Amount of memory still referenced at the end of compilation increased from 6609k to 7051k, overall 6.69%
>     Overall memory needed: 40555k -> 29988k
>     Peak memory use before GGC: 19322k -> 17844k
>     Peak memory use after GGC: 19127k -> 17651k
>     Maximum of released memory in single GGC run: 2077k -> 1450k
>     Garbage: 57382k -> 52532k
>     Leak: 6609k -> 7051k
>     Overhead: 6547k -> 5910k
>     GGC runs: 352 -> 357
> 
> comparing combine.c compilation at -O2 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 41059k to 34344k, overall -19.55%
>   Peak amount of GGC memory allocated before garbage collecting run decreased from 19372k to 17874k, overall -8.38%
>   Peak amount of GGC memory still allocated after garbage collecting decreased from 19179k to 17667k, overall -8.56%
>   Amount of memory still referenced at the end of compilation increased from 6732k to 7167k, overall 6.45%
>     Overall memory needed: 41059k -> 34344k
>     Peak memory use before GGC: 19372k -> 17874k
>     Peak memory use after GGC: 19179k -> 17667k
>     Maximum of released memory in single GGC run: 2159k -> 1392k
>     Garbage: 69376k -> 69072k
>     Leak: 6732k -> 7167k
>     Overhead: 8392k -> 7992k
>     GGC runs: 406 -> 414
> 
> comparing combine.c compilation at -O3 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 45611k to 40692k, overall -12.09%
>   Peak amount of GGC memory allocated before garbage collecting run decreased from 19484k to 18146k, overall -7.37%
>   Peak amount of GGC memory still allocated after garbage collecting decreased from 19266k to 17822k, overall -8.10%
>   Amount of produced GGC garbage increased from 90771k to 94381k, overall 3.98%
>   Amount of memory still referenced at the end of compilation increased from 6843k to 7279k, overall 6.37%
>     Overall memory needed: 45611k -> 40692k
>     Peak memory use before GGC: 19484k -> 18146k
>     Peak memory use after GGC: 19266k -> 17822k
>     Maximum of released memory in single GGC run: 3635k -> 3636k
>     Garbage: 90771k -> 94381k
>     Leak: 6843k -> 7279k
>     Overhead: 11439k -> 11263k
>     GGC runs: 428 -> 443
> 
> comparing insn-attrtab.c compilation at -O0 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 151663k to 92884k, overall -63.28%
>   Peak amount of GGC memory allocated before garbage collecting run decreased from 69577k to 58836k, overall -18.26%
>   Peak amount of GGC memory still allocated after garbage collecting decreased from 44191k to 33331k, overall -32.58%
>   Amount of produced GGC garbage decreased from 133641k to 129388k, overall -3.29%
>   Amount of memory still referenced at the end of compilation increased from 9583k to 9612k, overall 0.30%
>     Overall memory needed: 151663k -> 92884k
>     Peak memory use before GGC: 69577k -> 58836k
>     Peak memory use after GGC: 44191k -> 33331k
>     Maximum of released memory in single GGC run: 37379k -> 33675k
>     Garbage: 133641k -> 129388k
>     Leak: 9583k -> 9612k
>     Overhead: 17418k -> 13888k
>     GGC runs: 216
> 
> comparing insn-attrtab.c compilation at -O0 -g level:
>   Ovarall memory allocated via mmap and sbrk decreased from 153171k to 94124k, overall -62.73%
>   Peak amount of GGC memory allocated before garbage collecting run decreased from 70739k to 59998k, overall -17.90%
>   Peak amount of GGC memory still allocated after garbage collecting decreased from 45459k to 34492k, overall -31.80%
>   Amount of produced GGC garbage decreased from 135123k to 129349k, overall -4.46%
>   Amount of memory still referenced at the end of compilation increased from 11051k to 11544k, overall 4.46%
>     Overall memory needed: 153171k -> 94124k
>     Peak memory use before GGC: 70739k -> 59998k
>     Peak memory use after GGC: 45459k -> 34492k
>     Maximum of released memory in single GGC run: 37380k -> 33676k
>     Garbage: 135123k -> 129349k
>     Leak: 11051k -> 11544k
>     Overhead: 17815k -> 14285k
>     GGC runs: 215 -> 213
> 
> comparing insn-attrtab.c compilation at -O1 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 183279k to 110116k, overall -66.44%
>   Peak amount of GGC memory allocated before garbage collecting run decreased from 85948k to 63440k, overall -35.48%
>   Peak amount of GGC memory still allocated after garbage collecting decreased from 80070k to 60815k, overall -31.66%
>   Amount of produced GGC garbage decreased from 268762k to 233043k, overall -15.33%
>   Amount of memory still referenced at the end of compilation increased from 9403k to 9732k, overall 3.50%
>     Overall memory needed: 183279k -> 110116k
>     Peak memory use before GGC: 85948k -> 63440k
>     Peak memory use after GGC: 80070k -> 60815k
>     Maximum of released memory in single GGC run: 32250k -> 25048k
>     Garbage: 268762k -> 233043k
>     Leak: 9403k -> 9732k
>     Overhead: 28313k -> 26091k
>     GGC runs: 228 -> 244
> 
> comparing insn-attrtab.c compilation at -O2 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 200087k to 169404k, overall -18.11%
>   Peak amount of GGC memory allocated before garbage collecting run decreased from 87235k to 63584k, overall -37.20%
>   Peak amount of GGC memory still allocated after garbage collecting decreased from 80140k to 61114k, overall -31.13%
>   Amount of produced GGC garbage decreased from 302019k to 269075k, overall -12.24%
>   Amount of memory still referenced at the end of compilation increased from 9401k to 9725k, overall 3.45%
>     Overall memory needed: 200087k -> 169404k
>     Peak memory use before GGC: 87235k -> 63584k
>     Peak memory use after GGC: 80140k -> 61114k
>     Maximum of released memory in single GGC run: 30040k -> 22146k
>     Garbage: 302019k -> 269075k
>     Leak: 9401k -> 9725k
>     Overhead: 33722k -> 31669k
>     GGC runs: 248 -> 266
> 
> comparing insn-attrtab.c compilation at -O3 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 200079k to 182268k, overall -9.77%
>   Peak amount of GGC memory allocated before garbage collecting run decreased from 87249k to 75615k, overall -15.39%
>   Peak amount of GGC memory still allocated after garbage collecting decreased from 80154k to 71528k, overall -12.06%
>   Amount of memory still referenced at the end of compilation increased from 9405k to 9729k, overall 3.44%
>     Overall memory needed: 200079k -> 182268k
>     Peak memory use before GGC: 87249k -> 75615k
>     Peak memory use after GGC: 80154k -> 71528k
>     Maximum of released memory in single GGC run: 30102k -> 24450k
>     Garbage: 302705k -> 300339k
>     Leak: 9405k -> 9729k
>     Overhead: 33928k -> 32926k
>     GGC runs: 254 -> 267
> 
> comparing Gerald's testcase PR8361 compilation at -O0 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 152987k to 145871k, overall -4.88%
>   Amount of memory still referenced at the end of compilation increased from 49217k to 51177k, overall 3.98%
>     Overall memory needed: 152987k -> 145871k
>     Peak memory use before GGC: 90246k -> 89182k
>     Peak memory use after GGC: 89353k -> 88299k
>     Maximum of released memory in single GGC run: 17774k -> 18129k
>     Garbage: 209978k -> 207129k
>     Leak: 49217k -> 51177k
>     Overhead: 24709k -> 23513k
>     GGC runs: 412 -> 408
> 
> comparing Gerald's testcase PR8361 compilation at -O0 -g level:
>   Ovarall memory allocated via mmap and sbrk decreased from 174771k to 163539k, overall -6.87%
>   Amount of memory still referenced at the end of compilation increased from 72532k to 74492k, overall 2.70%
>     Overall memory needed: 174771k -> 163539k
>     Peak memory use before GGC: 102853k -> 101739k
>     Peak memory use after GGC: 101834k -> 100731k
>     Maximum of released memory in single GGC run: 18129k -> 18434k
>     Garbage: 216510k -> 212808k
>     Leak: 72532k -> 74492k
>     Overhead: 30605k -> 29409k
>     GGC runs: 385 -> 381
> 
> comparing Gerald's testcase PR8361 compilation at -O1 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 150728k to 141579k, overall -6.46%
>   Amount of memory still referenced at the end of compilation increased from 50296k to 51765k, overall 2.92%
>     Overall memory needed: 150728k -> 141579k
>     Peak memory use before GGC: 101899k -> 100523k
>     Peak memory use after GGC: 100884k -> 99511k
>     Maximum of released memory in single GGC run: 17236k -> 17479k
>     Garbage: 347581k -> 342084k
>     Leak: 50296k -> 51765k
>     Overhead: 31201k -> 30479k
>     GGC runs: 531 -> 527
> 
> comparing Gerald's testcase PR8361 compilation at -O2 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 155252k to 147867k, overall -4.99%
>   Amount of produced GGC garbage increased from 378710k to 388895k, overall 2.69%
>   Amount of memory still referenced at the end of compilation increased from 51032k to 52844k, overall 3.55%
>     Overall memory needed: 155252k -> 147867k
>     Peak memory use before GGC: 102512k -> 101308k
>     Peak memory use after GGC: 101495k -> 100305k
>     Maximum of released memory in single GGC run: 17233k -> 17476k
>     Garbage: 378710k -> 388895k
>     Leak: 51032k -> 52844k
>     Overhead: 35520k -> 35877k
>     GGC runs: 569 -> 577
> 
> comparing Gerald's testcase PR8361 compilation at -O3 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 157080k to 149127k, overall -5.33%
>   Amount of produced GGC garbage increased from 397752k to 421739k, overall 6.03%
>   Amount of memory still referenced at the end of compilation increased from 51318k to 53142k, overall 3.55%
>     Overall memory needed: 157080k -> 149127k
>     Peak memory use before GGC: 104309k -> 102907k
>     Peak memory use after GGC: 103276k -> 101887k
>     Maximum of released memory in single GGC run: 17610k -> 17909k
>     Garbage: 397752k -> 421739k
>     Leak: 51318k -> 53142k
>     Overhead: 37071k -> 38435k
>     GGC runs: 575 -> 599
> 
> comparing PR rtl-optimization/28071 testcase compilation at -O0 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 344531k to 244568k, overall -40.87%
>   Peak amount of GGC memory allocated before garbage collecting increased from 80256k to 81018k, overall 0.95%
>   Peak amount of GGC memory still allocated after garbage collecting increased from 57995k to 58757k, overall 1.31%
>   Amount of produced GGC garbage increased from 142770k to 144428k, overall 1.16%
>   Amount of memory still referenced at the end of compilation increased from 7421k to 7724k, overall 4.08%
>     Overall memory needed: 344531k -> 244568k
>     Peak memory use before GGC: 80256k -> 81018k
>     Peak memory use after GGC: 57995k -> 58757k
>     Maximum of released memory in single GGC run: 43619k -> 44133k
>     Garbage: 142770k -> 144428k
>     Leak: 7421k -> 7724k
>     Overhead: 25956k -> 23300k
>     GGC runs: 74 -> 79
> 
> comparing PR rtl-optimization/28071 testcase compilation at -O0 -g level:
>   Ovarall memory allocated via mmap and sbrk decreased from 345431k to 245388k, overall -40.77%
>   Peak amount of GGC memory allocated before garbage collecting increased from 80902k to 81665k, overall 0.94%
>   Peak amount of GGC memory still allocated after garbage collecting increased from 58641k to 59403k, overall 1.30%
>   Amount of produced GGC garbage increased from 143036k to 144537k, overall 1.05%
>   Amount of memory still referenced at the end of compilation increased from 9188k to 9492k, overall 3.30%
>     Overall memory needed: 345431k -> 245388k
>     Peak memory use before GGC: 80902k -> 81665k
>     Peak memory use after GGC: 58641k -> 59403k
>     Maximum of released memory in single GGC run: 43608k -> 44122k
>     Garbage: 143036k -> 144537k
>     Leak: 9188k -> 9492k
>     Overhead: 26452k -> 23796k
>     GGC runs: 84 -> 87
> 
> comparing PR rtl-optimization/28071 testcase compilation at -O1 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 303191k to 243816k, overall -24.35%
>   Amount of produced GGC garbage decreased from 231144k to 223287k, overall -3.52%
>   Amount of memory still referenced at the end of compilation increased from 20856k to 20967k, overall 0.53%
>     Overall memory needed: 303191k -> 243816k
>     Peak memory use before GGC: 84146k -> 83516k
>     Peak memory use after GGC: 74847k -> 74900k
>     Maximum of released memory in single GGC run: 34777k -> 39417k
>     Garbage: 231144k -> 223287k
>     Leak: 20856k -> 20967k
>     Overhead: 33134k -> 29202k
>     GGC runs: 81
> 
> comparing PR rtl-optimization/28071 testcase compilation at -O2 level:
>   Ovarall memory allocated via mmap and sbrk decreased from 559707k to 264060k, overall -111.96%
>   Amount of produced GGC garbage decreased from 239608k to 229973k, overall -4.19%
>   Amount of memory still referenced at the end of compilation increased from 20946k to 21057k, overall 0.53%
>     Overall memory needed: 559707k -> 264060k
>     Peak memory use before GGC: 79833k -> 79886k
>     Peak memory use after GGC: 74847k -> 74900k
>     Maximum of released memory in single GGC run: 33336k -> 33022k
>     Garbage: 239608k -> 229973k
>     Leak: 20946k -> 21057k
>     Overhead: 35480k -> 31220k
>     GGC runs: 92 -> 91
> 
> comparing PR rtl-optimization/28071 testcase compilation at -O3 -fno-tree-pre -fno-tree-fre level:
>   Overall memory allocated via mmap and sbrk increased from 1046099k to 1297752k, overall 24.06%
>   Peak amount of GGC memory allocated before garbage collecting run decreased from 201851k to 190659k, overall -5.87%
>   Peak amount of GGC memory still allocated after garbage collecting decreased from 190309k to 178175k, overall -6.81%
>   Amount of produced GGC garbage decreased from 378489k to 363316k, overall -4.18%
>   Amount of memory still referenced at the end of compilation increased from 46312k to 46424k, overall 0.24%
>     Overall memory needed: 1046099k -> 1297752k
>     Peak memory use before GGC: 201851k -> 190659k
>     Peak memory use after GGC: 190309k -> 178175k
>     Maximum of released memory in single GGC run: 80846k -> 80667k
>     Garbage: 378489k -> 363316k
>     Leak: 46312k -> 46424k
>     Overhead: 51551k -> 43892k
>     GGC runs: 72
> 
> Head of the ChangeLog is:
> 
> --- /usr/src/SpecTests/sandbox-britten-memory/x86_64/mem-result/ChangeLog	2007-04-29 17:39:02.000000000 +0000
> +++ /usr/src/SpecTests/sandbox-britten-memory/gcc/gcc/ChangeLog	2007-05-02 14:07:52.000000000 +0000
> @@ -1,3 +1,230 @@
> +2007-05-02  Richard Guenther  <rguenther@suse.de>
> +
> +	PR tree-optimization/31146
> +	* tree-ssa-forwprop.c (forward_propagate_addr_expr_1): New
> +	argument, single_use_p.  If we have a single use that is
> +	a conversion to the definition rhs type, propagate that rhs.
> +	(forward_propagate_addr_expr): Pass single_use_p argument
> +	to forward_propagate_addr_expr_1.
> +