# Tests that call 'compiler_stats_num_field' are skipped when debugging is on. # See testsuite/driver/testlib.py. def no_lint(name, opts): opts.compiler_always_flags = \ [opt for opt in opts.compiler_always_flags if opt != '-dcore-lint' and opt != '-dcmm-lint'] setTestOpts(no_lint) # Note [residency] # # Residency (peak_megabytes_allocated and max_bytes_used) is sensitive # to when the major GC runs, which makes it inherently inaccurate. # Sometime an innocuous change somewhere can shift things around such # that the samples occur at a different time, and the residency # appears to change (up or down) when the underlying profile hasn't # really changed. # # However, please don't just ignore changes in residency. If you see # a change in one of these figures, please check whether it is real or # not as follows: # # * Run the test with old and new compilers, adding +RTS -h -i0.01 # (you don't need to compile anything for profiling or enable profiling # libraries to get a heap profile). # * view the heap profiles, read off the maximum residency. If it has # really changed, then you know there's an issue. test('T1969', [# expect_broken(12437), compiler_stats_num_field('peak_megabytes_allocated', # Note [residency] [(wordsize(32), 30, 15), # 2010-05-17 14 (x86/Windows) # 15 (x86/OS X) # 19 (x86/OS X) # 2013-02-10 13 (x86/Windows) # 2013-02-10 14 (x86/OSX) # 2013-11-13 17 (x86/Windows, 64bit machine) # 2015-07-11 21 (x86/Linux, 64bit machine) use +RTS -G1 # 2016-04-06 30 (x86/Linux, 64bit machine) (wordsize(64), 83, 20)]), # 28 (amd64/Linux) # 34 (amd64/Linux) # 2012-09-20 23 (amd64/Linux) # 2012-10-03 25 (amd64/Linux if .hi exists) # 2013-02-13 23, but unstable so increased to 10% range # 2013-02-13 27, very unstable! # 2014-09-10 29 (amd64/Linux) post-AMP-cleanup # 2013-09-11 30, 10 (amd64/Linux) # 2013-09-11 30, 15 (adapt to Phab CI) # 2015-06-03 41, (amd64/Linux) use +RTS -G1 # 2015-10-28 55, (amd64/Linux) emit Typeable at definition site # 2016-10-20 68, (amd64/Linux) allow top-level string literals # See the comment 16 on #8472. # 2017-02-17 83 (amd64/Linux) Type-indexed Typeable compiler_stats_num_field('max_bytes_used', [(platform('i386-unknown-mingw32'), 5719436, 20), # 2010-05-17 5717704 (x86/Windows) # 2013-02-10 5159748 (x86/Windows) # 2013-02-10 5030080 (x86/Windows) # 2013-11-13 7295012 (x86/Windows, 64bit machine) # 2014-04-24 5719436 (x86/Windows, 64bit machine) (wordsize(32), 9093608, 1), # 6707308 (x86/OS X) # 2009-12-31 6149572 (x86/Linux) # 2014-01-22 6429864 (x86/Linux) # 2014-06-29 5949188 (x86/Linux) # 2015-07-11 6241108 (x86/Linux, 64bit machine) use +RTS -G1 # 2016-04-06 9093608 (x86/Linux, 64bit machine) (wordsize(64), 19924328, 15)]), # 2014-09-10 10463640, 10 # post-AMP-update (somewhat stabelish) # looks like the peak is around ~10M, but we're # unlikely to GC exactly on the peak. # varies quite a lot with CLEANUP and BINDIST, # hence 10% range. # See Note [residency] to get an accurate view. # 2014-09-14 9684256, 10 # try to lower it a bit more to match Phab's CI # 2014-11-03 10584344, # ghcspeed reports higher numbers consistently # 2015-07-11 11670120 (amd64/Linux) # 2015-10-28 15017528 (amd64/Linux) emit typeable at definition site # 2016-10-12 17285216 (amd64/Linux) it's not entirely clear why # 2017-02-01 19924328 (amd64/Linux) Join points (#12988) # 2017-02-14 16393848 Early inline patch compiler_stats_num_field('bytes allocated', [(platform('i386-unknown-mingw32'), 301784492, 5), # 215582916 (x86/Windows) # 2012-10-29 298921816 (x86/Windows) # 2013-02-10 310633884 (x86/Windows) # 2013-11-13 317975916 (x86/Windows, 64bit machine) # 2014-04-04 301784492 (x86/Windows, 64bit machine) (wordsize(32), 344730660, 1), # 221667908 (x86/OS X) # 274932264 (x86/Linux) # 2012-10-08 303930948 (x86/Linux, new codegen) # 2013-02-10 322937684 (x86/OSX) # 2014-01-22 316103268 (x86/Linux) # 2014-06-29 303300692 (x86/Linux) # 2015-07-11 288699104 (x86/Linux, 64-bit machine) use +RTS -G1 # 2016-04-06 344730660 (x86/Linux, 64-bit machine) (wordsize(64), 695354904, 5)]), # 2009-11-17 434845560 (amd64/Linux) # 2009-12-08 459776680 (amd64/Linux) # 2010-05-17 519377728 (amd64/Linux) # 2011-08-05 561382568 (amd64/OS X) # 2012-07-16 589168872 (amd64/Linux) # 2012-07-20 595936240 (amd64/Linux) # 2012-08-23 606230880 (amd64/Linux) # 2012-08-29 633334184 (amd64/Linux) new codegen # 2012-09-18 641959976 (amd64/Linux) # 2012-10-19 661832592 (amd64/Linux) -fPIC turned on # 2012-10-23 642594312 (amd64/Linux) -fPIC turned off again # 2012-11-12 658786936 (amd64/Linux) UNKNOWN REASON # 2013-91-17 667160192 (x86_64/Linux) new demand analyser # 2013-10-18 698612512 (x86_64/Linux) fix for #8456 # 2014-02-10 660922376 (x86_64/Linux) call arity analysis # 2014-07-17 651626680 (x86_64/Linux) roundabout update # 2014-09-10 630299456 (x86_64/Linux) post-AMP-cleanup # 2015-06-03 581460896 (x86_64/Linux) use +RTS -G1 # 2015-10-28 695430728 (x86_64/Linux) emit Typeable at definition site # 2015-10-28 756138176 (x86_64/Linux) inst-decl defaults go via typechecker (#12220) # 2017-02-17 831733376 (x86_64/Linux) Type-indexed Typeable # 2017-02-25 695354904 (x86_64/Linux) Early inlining patch only_ways(['normal']), extra_hc_opts('-dcore-lint -static'), # Leave -dcore-lint on for this one test, so that we have something # that will catch a regression in -dcore-lint performance. # Use `+RTS -G1` for more stable residency measurements. Note [residency]. extra_hc_opts('+RTS -G1 -RTS') ], compile, ['']) # This one tests for the space leak in the native code generator, # where it holds onto the entire asm output until the end. The space # leak reappears from time to time, so it's a good idea to have a test # for it. The residency of this test will jump by 10MB to 40MB or so # on x86-64 if the space leak appears. # Only run this one if we have an NCG: if 'optasm' in config.compile_ways: conf_3294 = only_ways(['normal']) else: conf_3294 = skip test('T3294', [ compiler_stats_num_field('max_bytes_used', # Note [residency] [(wordsize(32), 28686588, 15), # 17725476 (x86/OS X) # 14593500 (Windows) # 2013-02-10 20651576 (x86/Windows) # 2013-02-10 20772984 (x86/OSX) # 2013-11-13 24009436 (x86/Windows, 64bit machine) # 2014-04-24 19882188 (x86/Windows, 64bit machine) # 2014-12-22 26525384 (x86/Windows) Increase due to silent superclasses? # 2015-07-11 43196344 (x86/Linux, 64-bit machine) use +RTS -G1 # 2016-04-06 28686588 (x86/Linux, 64-bit machine) (wordsize(64), 63131248, 20)]), # prev: 25753192 (amd64/Linux) # 29/08/2012: 37724352 (amd64/Linux) # (increase due to new codegen, see #7198) # 13/13/2012: 44894544 (amd64/Linux) # (reason for increase unknown) # 15/5/2013: 36904752 (amd64/Linux) # (reason for decrease unknown) # 29/5/2013: 43224080 (amd64/Linux) # (reason for increase back to earlier value unknown) # 2014-07-14: 36670800 (amd64/Linux) # (reason unknown, setting expected value somewhere in between) # 2015-01-22: 45000000 (amd64/Linux) # varies between 40959592 and 52914488... increasing to +-20% # 2015-10-28: 50367248 (amd64/Linux) # D757: emit Typeable instances at site of type definition # 2016-07-11: 54609256 (Windows) before fix for #12227 # 2016-07-11: 52992688 (Windows) after fix for #12227 # 2017-02-17: 63131248 (amd64/Linux) Type indexed Typeable compiler_stats_num_field('bytes allocated', [(wordsize(32), 1377050640, 5), # previous: 815479800 (x86/Linux) # (^ increase due to new codegen, see #7198) # 2012-10-08: 1373514844 (x86/Linux) # 2013-11-13: 1478325844 (x86/Windows, 64bit machine) # 2014-01-12: 1565185140 (x86/Linux) # 2013-04-04: 1377050640 (x86/Windows, 64bit machine) (wordsize(64), 2758641264, 5)]), # old: 1357587088 (amd64/Linux) # 29/08/2012: 2961778696 (amd64/Linux) # (^ increase due to new codegen, see #7198) # 18/09/2012: 2717327208 (amd64/Linux) # 08/06/2013: 2901451552 (amd64/Linux) (reason unknown) # 12/12/2013: 3083825616 (amd64/Linux) (reason unknown) # 18/02/2014: 2897630040 (amd64/Linux) (call arity improvements) # 12/03/2014: 2705289664 (amd64/Linux) (more call arity improvements) # 2014-17-07: 2671595512 (amd64/Linux) (round-about update) # 2014-09-10: 2709595808 (amd64/Linux) post-AMP cleanup # 2016-07-11: 2664479936 (Windows) before fix for #12227 # 2016-07-11: 2739731144 (Windows) after fix for #12227 (ignoring) # 2016-02-17: 2758641264 (amd64/Linux) (Type indexed Typeable) conf_3294, # Use `+RTS -G1` for more stable residency measurements. Note [residency]. extra_hc_opts('+RTS -G1 -RTS') ], compile, ['']) test('T4801', [ # expect_broken(5224), # temporarily unbroken (#5227) ################################### # deactivated for now, as this metric became too volatile recently # compiler_stats_num_field('peak_megabytes_allocated',# Note [residency] # [(platform('x86_64-apple-darwin'), 70, 1), # # expected value: 58 (amd64/OS X) # # 13/01/2014 - 70 # (wordsize(32), 30, 20), # (wordsize(64), 48, 20)]), # # prev: 50 (amd64/Linux) # # 19/10/2012: 64 (amd64/Linux) # # (^ REASON UNKNOWN!) # # 12/11/2012: 49 (amd64/Linux) # # (^ REASON UNKNOWN!) # # 28/8/13: 60 (amd64/Linux) # # (^ REASON UNKNOWN!) # # 2014-09-10: 55 post-AMP-cleanup # # 2014-10-08: 62 (jumps between 55 and 71 observed -- GC tipping point?) # # 2014-10-13: 48 stricter seqDmdType compiler_stats_num_field('bytes allocated', [(platform('x86_64-apple-darwin'), 465653312, 10), # prev: 510938976 (amd64/OS X): # 2015-12-11: 465653312 (amd64/OS X) Update, bump tolerance to +/-10% (wordsize(32), 239556572, 10), # prev: 185669232 (x86/OSX) # 2014-01-22: 211198056 (x86/Linux) # 2014-09-03: 185242032 (Windows laptop) # 2014-12-01: 203962148 (Windows laptop) # 2016-04-06: 239556572 (x86/Linux) (wordsize(64), 388898280, 10)]), # prev: 360243576 (amd64/Linux) # 19/10/2012: 447190832 (amd64/Linux) (-fPIC turned on) # 19/10/2012: 392409984 (amd64/Linux) (-fPIC turned off) # 2014-04-08: 362939272 (amd64/Linux) cumulation of various smaller improvements over recent commits # 2014-10-08: 382056344 (amd64/Linux) stricter foldr2 488e95b # 2015-10-28: 434278248 (amd64/Linux) emit Typeable at definition site # 2016-10-19: 388898280 (amd64/Linux) Refactor traceRn interface (#12617) ################################### # deactivated for now, as this metric became too volatile recently # # compiler_stats_num_field('max_bytes_used', # [(platform('x86_64-apple-darwin'), 25145320, 5), # (wordsize(32), 11829000, 15), # # 9651948 (x86/OSX) # # 10290952 (windows) # # 2013-02-10 11071060 (x86/Windows) # # 2013-02-10: 11207828 (x86/OSX) # # (some date): 11139444 # # 2013-11-13: 11829000 (x86/Windows, 64bit machine) # (wordsize(64), 19296544, 15)]), # # prev: 20486256 (amd64/OS X) # # 30/08/2012: 17305600--20391920 (varies a lot) # # 19/10/2012: 26882576 (-fPIC turned on) # # 19/10/2012: 18619912 (-fPIC turned off) # # 24/12/2012: 21657520 (perhaps gc sampling time wibbles?) # # 10/01/2014: 25166280 # # 13/01/2014: 22646000 (mostly due to #8647) # # 18/02/2014: 25002136 (call arity analysis changes) # # 12/05/2014: 25002136 (specialisation and inlining changes) # # 10/09/2014: 19296544, 10 (post-AMP-cleanup) # # 14/09/2014: 19585456, 15 (adapt to Phab CI env) only_ways(['normal']), extra_hc_opts('-static'), # Use `+RTS -G1` for more stable residency measurements. Note [residency]. extra_hc_opts('+RTS -G1 -RTS') ], compile, ['']) test('T3064', [compiler_stats_num_field('peak_megabytes_allocated',# Note [residency] [(wordsize(32), 28, 20), # expected value: 14 (x86/Linux 28-06-2012): # 2013-11-13: 18 (x86/Windows, 64bit machine) # 2014-01-22: 23 (x86/Linux) # 2014-12-22: 23 (x86/Linux) death to silent superclasses # 2015-07-11 28 (x86/Linux, 64-bit machine) use +RTS -G1 (wordsize(64), 66, 20)]), # (amd64/Linux): 18 # (amd64/Linux) 2012-02-07: 26 # (amd64/Linux) 2013-02-12: 23; increased range to 10% # (amd64/Linux) 2013-04-03: 26 # (amd64/Linux) 2013-09-11: 30; result of AMP patch # Increased range to 20%. peak-usage varies from 22 to 26, # depending on whether the old .hi file exists # (amd64/Linux) 2013-09-11: 37; better arity analysis (weird) # (amd64/Linux) (09/09/2014): 42, AMP changes (larger interfaces, more loading) # (amd64/Linux) 2014-10-13: 38: Stricter seqDmdType # (amd64/Linux) 2014-12-22: 27: death to silent superclasses # (amd64/Linux) 2015-01-22: 32: Varies from 30 to 34, at least here. # (amd64/Linux) 2015-06-03: 54: use +RTS -G1 # (amd64/Linux) 2016-10-25: 66: Presumably creep compiler_stats_num_field('bytes allocated', [(wordsize(32), 153261024, 10), # 2011-06-28: 56380288 (x86/Linux) # 2012-10-30: 111189536 (x86/Windows) # 2013-11-13: 146626504 (x86/Windows, 64bit machine) # 2014-01-22: 162457940 (x86/Linux) # 2014-12-01: 162457940 (Windows) # 2014-12-22: 122836340 (Windows) Death to silent superclasses # 2016-04-06: 153261024 (x86/Linux) probably wildcard refactor (wordsize(64), 259815560, 5)]), # (amd64/Linux) (2011-06-28): 73259544 # (amd64/Linux) (2013-02-07): 224798696 # (amd64/Linux) (2013-08-02): 236404384, increase from roles # (amd64/Linux) (2013-09-11): 290165632, increase from AMP warnings # (amd64/Linux) (2013-11-22): 308300448, GND via Coercible and counters for constraints solving # (amd64/Linux) (2013-12-02): 329795912, Coercible refactor # (amd64/Linux) (2014-02-11): 308422280, optimize Coercions in simpleOptExpr # (amd64/Linux) (2014-05-23): 324022680, unknown cause # (amd64/Linux) (2014-07-17): 332702112, general round of updates # (amd64/Linux) (2014-08-29): 313638592, w/w for INLINABLE things # (amd64/Linux) (2014-09-09): 407416464, AMP changes (larger interfaces, more loading) # (amd64/Linux) (2014-09-14): 385145080, BPP changes (more NoImplicitPrelude in base) # (amd64/Linux) (2014-12-10): 363103840, improvements in constraint solver # (Mac) (2014-12-18): 350418600, improvements to flattener # (amd64/Linux) (2014-12-22): 243670824, Ha! Death to superclass constraints, makes # much less code for Monad instances # (amd64/Linux) (2015-12-01): 264952256, Regression due to Simon's wildcard refactor # Tracked as #11151. # (amd64/Linux) (2015-12-11): 304344936, Regression due to TypeInType # Tracked as #11196 # (amd64/Linux) (2016-04-15): 287460128 Improvement due to using coercionKind instead # of zonkTcType (Trac #11882) # (amd64/Darwin) (2017-01-23): 306222424 Presumably creep from recent changes (Typeable?) # (amd64/Linux) (2017-02-14): 259815560 Early inline patch: 9% improvement ################################### # deactivated for now, as this metric became too volatile recently # # compiler_stats_num_field('max_bytes_used', # [(wordsize(32), 11202304, 20), # # 2011-06-28: 2247016 (x86/Linux) (28/6/2011): # #(some date): 5511604 # # 2013-11-13: 7218200 (x86/Windows, 64bit machine) # # 2014-04-04: 11202304 (x86/Windows, 64bit machine) # (wordsize(64), 13251728, 20)]), # # (amd64/Linux, intree) (28/06/2011): 4032024 # # (amd64/Linux, intree) (07/02/2013): 9819288 # # (amd64/Linux) (14/02/2013): 8687360 # # (amd64/Linux) (18/02/2013): 9397488 # # (amd64/Linux) (02/08/2013): 10742536, increase from roles # # (amd64/Linux) (19/08/2013): 9211816, decrease apparently from better eta reduction # # (amd64/Linux) (11/09/2013): 12000480, increase from AMP warnings # # 933cdf15a2d85229d3df04b437da31fdfbf4961f # # (amd64/Linux) (22/11/2013): 16266992, GND via Coercible and counters for constraints solving # # (amd64/Linux) (12/12/2013): 19821544, better One shot analysis # # (amd64/Linux) (09/09/2014): 24357392, AMP changes (larger interfaces, more loading) # # (amd64/Linux) (14/09/2014): 16053888, BPP changes (more NoImplicitPrelude in base) # # (amd64/Linux) (19/09/2014): 18744992, unknown # # (amd64/Linux) 2014-10-13: 13251728, Stricter seqDmdType only_ways(['normal']), # Use `+RTS -G1` for more stable residency measurements. Note [residency]. extra_hc_opts('+RTS -G1 -RTS') ], compile, ['']) test('T4007', normal, run_command, ['$MAKE -s --no-print-directory T4007']) test('T5030', [compiler_stats_num_field('bytes allocated', [(wordsize(32), 345668088, 10), # previous: 196457520 # 2012-10-08: 259547660 (x86/Linux, new codegen) # 2013-11-21: 198573456 (x86 Windows, 64 bit machine) # 2014-12-10: 227205560 constraint solver got worse again; more aggressive solving # of family-applications leads to less sharing, I think # 2015-07-11: 201882912 reason unknown # 2016-04-06: 345668088 likely TypeInType (wordsize(64), 794426536, 10)]), # Previously 530000000 (+/- 10%) # 17/1/13: 602993184 (x86_64/Linux) # (new demand analyser) # 2013-06-08 538467496 (x86_64/Linux) # ^ reason unknown # 2013-08-02 454498592 (amd64/Linux) # decrease from more aggressive coercion optimisations from roles # 2013-11-12 397672152 (amd64/Linux) # big decrease following better CSE and arity # 2014-07-17 409314320 (amd64/Linux) # general round of updates # 2014-09-10 385152728 post-AMP-cleanup # 2014-12-08 340969128 constraint solver perf improvements (esp kick-out) # 2014-12-10 449042120 constraint solver got worse again; more aggressive solving # of family-applications leads to less sharing, I think # 2015-03-17 403932600 tweak to solver algorithm # 2015-12-11 653710960 TypeInType (see #11196) # 2016-10-17 794426536 20% big increase following # 31621b12 * A collection of type-inference refactorings. # See ticket for more info only_ways(['normal']) ], compile, ['-freduction-depth=300']) test('T5631', [compiler_stats_num_field('bytes allocated', [(wordsize(32), 570137436, 10), # expected value: 392904228 (x86/Linux) # 2014-04-04: 346389856 (x86 Windows, 64 bit machine) # 2014-12-01: 390199244 (Windows laptop) # 2016-04-06: 570137436 (amd64/Linux) many reasons (wordsize(64), 1065147968, 5)]), # expected value: 774595008 (amd64/Linux): # expected value: 735486328 (amd64/Linux) 2012/12/12: # expected value: 690742040 (amd64/Linux) Call Arity improvements # 2014-09-09: 739704712 (amd64/Linux) AMP changes # 2014-11-04: 776121120 (amd64/Linux) new-flatten-skolems # 2015-06-01: 812288344 (amd64/Linux) unknown cause # 2015-12-11: 1128828928 (amd64/Linux) TypeInType (see #11196) # 2015-12-21: 1198327544 (Mac) TypeApplications (will fix with #11196) # 2015-03-18: 1124068664 (Mac) optimize Unify & zonking # 2016-10-19: 1024926024 (amd64/Linux) Refactor traceRn interface (#12617) # 2016-11-10: 1077429456 (amd64/Linux) Stop -dno-debug-output suppressing -ddump-tc-trace # 2017-02-17: 1517484488 (amd64/Linux) Type-indexed Typeable # 2017-03-03: 1065147968 (amd64/Linux) Share Typeable KindReps only_ways(['normal']) ], compile, ['']) test('parsing001', [compiler_stats_num_field('bytes allocated', [(wordsize(32), 274000576, 10), (wordsize(64), 463931280, 5)]), # expected value: 587079016 (amd64/Linux) # 2016-09-01: 581551384 (amd64/Linux) Restore w/w limit (#11565) # 2016-12-19: 493730288 (amd64/Linux) Join points (#12988) # 2017-02-14: 463931280 Early inlining patch; acutal improvement 7% only_ways(['normal']), ], compile_fail, ['']) test('T783', [ only_ways(['normal']), # no optimisation for this one # expected value: 175,569,928 (x86/Linux) compiler_stats_num_field('bytes allocated', [(wordsize(32), 249332816, 5), # 2012-10-08: 226907420 (x86/Linux) # 2013-02-10: 329202116 (x86/Windows) # 2013-02-10: 338465200 (x86/OSX) # 2014-04-04: 319179104 (x86 Windows, 64 bit machine) # 2014-09-03: 223377364 (Windows) better specialisation, raft of core-to-core optimisations # 2014-12-22: 235002220 (Windows) not sure why # 2016-04-06: 249332816 (x86/Linux, 64-bit machine) (wordsize(64), 436978192, 10)]), # prev: 349263216 (amd64/Linux) # 07/08/2012: 384479856 (amd64/Linux) # 29/08/2012: 436927840 (amd64/Linux) # 12/11/2012: 640324528 (amd64/Linux) # (OldCmm removed: not sure why this got worse, the # other perf tests remained about the same) # 18/10/2013: 734038080 (amd64/Linux) # (fix for #8456) # 24/10/2013: 654804144 (amd64/Linux) # (fix previous fix for #8456) # 2014-07-17: 640031840 (amd64/Linux) # (general round of updates) # 2014-08-29: 441932632 (amd64/Linux) # (better specialisation, raft of core-to-core optimisations) # 2014-08-29: 719814352 (amd64/Linux) # (changed order of cmm block causes analyses to allocate much more, # but the changed order is slighly better in terms of runtime, and # this test seems to be an extreme outlier.) # 2015-05-16: 548288760 (amd64/Linux) # (improved sequenceBlocks in nativeCodeGen, #10422) # 2015-08-07: 470738808 (amd64/Linux) # (simplifying the switch plan code path for simple checks, #10677) # 2015-08-28: 526230456 (amd64/Linux) # (D757: Emit Typeable instances at site of type definition) # 2015-12-04: 1134085384 (amd64/Linux) # (D1535: Major overhaul of pattern match checker, #11162) # 2016-02-03: 488592288 (amd64/Linux) # (D1795: Another overhaul of pattern match checker, #11374) # 2017-02-14 436978192 Early inlining: 5% improvement extra_hc_opts('-static') ], compile,['']) test('T5321Fun', [ only_ways(['normal']), # no optimisation for this one compiler_stats_num_field('bytes allocated', [(wordsize(32), 279922360, 10), # prev: 300000000 # 2012-10-08: 344416344 x86/Linux # (increase due to new codegen) # 2014-09-03: 299656164 (specialisation and inlining) # 2014-12-10: 206406188 # Improvements in constraint solver # 2016-04-06: 279922360 x86/Linux (wordsize(64), 488295304, 5)]) # prev: 585521080 # 2012-08-29: 713385808 # (increase due to new codegen) # 2013-05-15: 628341952 # (reason for decrease unknown) # 2013-06-24: 694019152 # (reason for re-increase unknown) # 2014-05-12: 614409344 # (specialisation and inlining changes) # 2014-09-10: 601629032 # post-AMP-cleanup # 2014-11-06: 541287000 # Simon's flat-skol changes to the constraint solver # 2014-12-10: 408110888 # Improvements in constraint solver # 2014-12-16: 429921312 # Flattener parameterized over roles # 2015-08-10: 509921312 # (undefined now takes an implicit parameter and GHC -O0 does # not recognize that the application is bottom) # 2015-12-11: 565883176 # TypeInType (see #11196) # 2017-01-06: 497356688 # Small coercion optimisations # The actual decrease was only 2%; earlier # commits had drifted down # 2017-01-22: 525895608 # Allow top-level string literals in Core. I'm not # convinced that this patch is # responsible for all of this # change, however. Namely I am # quite skeptical of the downward # "drift" reported above # 2017-01-31: 498135752 # Join points (#12988) # 2017-02-23: 524706256 # Type-indexed Typeable? (on Darwin) # 2017-02-25: 488295304 # Early inlining patch ], compile,['']) test('T5321FD', [ only_ways(['normal']), # no optimisation for this one compiler_stats_num_field('bytes allocated', [(wordsize(32), 250757460, 10), # prev: 213380256 # 2012-10-08: 240302920 (x86/Linux) # (increase due to new codegen) # 2014-07-31: 211699816 (Windows) (-11%) # (due to better optCoercion, 5e7406d9, #9233) # 2016-04-06: 250757460 (x86/Linux) (wordsize(64), 477840432, 10)]) # prev: 418306336 # 29/08/2012: 492905640 # (increase due to new codegen) # 15/05/2013: 406039584 # (reason for decrease unknown) # 08/06/2013: 476497048 # (reason for increase unknown) # before 2014-07-17: 441997096 # (with -8%, still in range, hence cause not known) # 2014-07-17: 426960992 (-11% of previous value) # (due to better optCoercion, 5e7406d9, #9233) # 2014-10-08 410895536 # (various changes; biggest improvements due to 949ad67 and FastString package ids) # 2015-08-10: 470895536 # (undefined now takes an implicit parameter and GHC -O0 does # not recognize that the application is bottom) # 2015-10-28: 532365376 # D757: emit Typeable instances at site of type definition # 2016-07-16: 477840432 # Optimize handling of built-in OccNames ], compile,['']) test('T5642', [ only_ways(['normal']), normal, compiler_stats_num_field('bytes allocated', [(wordsize(32), 462677300, 10), # sample from x86/Linux # prev: 650000000 # 2014-09-03: 753045568 # 2014-12-10: 641085256 Improvements in constraints solver # 2016-04-06: 462677300 (wordsize(64), 838316496, 10)]) # prev: 1300000000 # 2014-07-17: 1358833928 (general round of updates) # 2014-08-07: 1402242360 (caused by 1fc60ea) # Watch out for: # 23/05/2014: 1452688392 (More aggressive specialisation means we get # specialised copies of imported functions that # are ultimately discarded by trimAutoRules # It's a bizarre program with LOTS of data types) # 2014-09-10: 1536924976 post-AMP-cleanup # 2014-12-10: 1282916024 Improvements in constraints solver # 2015-10-28: 1412808976 Emit Typeable at definition site # 2015-11-22: 1071915072 Use TypeLits in the metadata encoding # 2016-02-08: 950004816 Pattern match checker re-rework # 2016-05-12: 1300685592 Make Generic1 poly-kinded # 2016-06-05: 916484672 Refactor derived Generic instances to reduce allocations # 2016-09-03: 838316496 Derive the Generic instance in perf/compiler/T5642 ], compile,['-O']) test('T5837', [ only_ways(['normal']), compiler_stats_num_field('bytes allocated', [(wordsize(32), 24199320, 10), # 40000000 (x86/Linux) # 2013-11-13: 45520936 (x86/Windows, 64bit machine) # 2014-09-03: 37096484 (Windows laptop, w/w for INLINABLE things # 2014-12-01: 135914136 (Windows laptop, regression see below) # 2014-12-08: 115905208 Constraint solver perf improvements (esp kick-out) # 2016-04-06: 24199320 (x86/Linux, 64-bit machine) TypeInType (platform('x86_64-unknown-mingw32'), 59161648, 7), # 2017-02-19 59161648 (x64/Windows) - Unknown (wordsize(64), 52625920, 7)]) # sample: 3926235424 (amd64/Linux, 15/2/2012) # 2012-10-02 81879216 # 2012-09-20 87254264 amd64/Linux # 2013-09-18 90587232 amd64/Linux # 2013-11-21 86795752 amd64/Linux, GND via Coercible and counters # for constraints solving # 2014-08-29 73639840 amd64/Linux, w/w for INLINABLE things # 2014-10-08 73639840 amd64/Linux, Burning Bridges and other small changes # 2014-11-06 271028976 Linux, Accept big regression; # See Note [An alternative story for the inert substitution] in TcFlatten # 2014-12-08 234790312 Constraint solver perf improvements (esp kick-out) # 2014-12-16 231155640 Mac Flattener parameterized over roles; # some optimization # 2015-03-17 53424304 Mac Better depth checking; fails earlier # 2015-06-09 38834096 Better "improvement"; I'm not sure whey it improves things # 2015-12-11 43877520 amd64/Linux, TypeInType (see #11196) # 2016-03-18 48507272 Mac, accept small regression in exchange # for other optimisations # 2016-09-15 42445672 Linux; fixing #12422 # 2016-09-25 41832056 amd64/Linux, Rework handling of names (D2469) # 2016-10-25 52597024 amd64/Linux, the test now passes (hooray), and so # allocates more because it goes right down the # compilation pipeline # 2017-01-24 57861352 amd64/Linux, very likely due to the top-level strings # in Core patch. # 2017-02-07 50253880 Another improvement in SetLevels. I don't think # all the gain here is from this patch, but I think it # just pushed it over the edge, so I'm re-centreing, and # changing to 5% tolerance # 2017-02-07 53592736 amd64/Linux Simon's earlier decrease appears # to be environmentally-dependent. # Also bumped acceptance threshold to 7%. # 2017-02-20 58648600 amd64/Linux Type-indexed Typeable # 2017-02-28 54151864 amd64/Linux Likely drift due to recent simplifier improvements # 2017-02-25 52625920 amd64/Linux Early inlining patch ], compile, ['-freduction-depth=50']) test('T6048', [ only_ways(['optasm']), compiler_stats_num_field('bytes allocated', [(wordsize(32), 55701280, 10), # prev: 38000000 (x86/Linux) # 2012-10-08: 48887164 (x86/Linux) # 2014-04-04: 62618072 (x86 Windows, 64 bit machine) # 2014-09-03: 56315812 (x86 Windows, w/w for INLINABLE) # 2014-12-01: 49987836 (x86 Windows) # 2016-04-06: 55701280 (x86/Linux, 64-bit machine) (wordsize(64), 115714216, 10)]) # 2012-09-18 97247032 amd64/Linux # 2014-01-16 108578664 amd64/Linux (unknown, likely foldl-via-foldr) # 2014-01-18 95960720 amd64/Linux Call Arity improvements # 2014-02-28 105556793 amd64/Linux (unknown, tweak in base/4d9e7c9e3 resulted in change) # 2014-03-05 110646312 amd64/Linux Call Arity became more elaborate # 2014-07-14 125431448 amd64/Linux unknown reason. Even worse in GHC-7.8.3. *shurg* # 2014-08-29 108354472 amd64/Linux w/w for INLINABLE things # 2014-09-14 88186056 amd64/Linux BPP part1 change (more NoImplicitPreludes in base) # 2014-01-08 95946688 amd64/Linux Mostly 4c834fd. Occasional spikes to 103822120! # 2016-03-11 108225624 amd64/Linux unknown reason sadly; likely gradual creep. # 2016-11-25 94327392 amd64/Linux Back down again hooray; still not sure why # 2017-02-17 115715592 amd64/Linux Type-indexed Typeable ], compile,['']) test('T9020', [ only_ways(['optasm']), compiler_stats_num_field('bytes allocated', [(wordsize(32), 343005716, 10), # Original: 381360728 # 2014-07-31: 343005716 (Windows) (general round of updates) (wordsize(64), 500707080, 10)]) # prev: 795469104 # 2014-07-17: 728263536 (general round of updates) # 2014-09-10: 785871680 post-AMP-cleanup # 2014-11-03: 680162056 Further Applicative and Monad adjustments # 2015-10-21: 786189008 Make stronglyConnCompFromEdgedVertices deterministic # 2016-01-26: 698401736 improvement from using ExpTypes instead of ReturnTvs # 2016-04-06: 852298336 Refactoring of CSE #11781 # 2016-04-06: 698401736 Use thenIO in Applicative IO # 2017-02-03: 764866144 Join points # 2017-02-14: 500707080 Early inline patch; 35% decrease! # Program size collapses in first simplification ], compile,['']) test('T9675', [ only_ways(['optasm']), compiler_stats_num_field('max_bytes_used', # Note [residency] [(wordsize(64), 38776008, 15), # 2014-10-13 29596552 # 2014-10-13 26570896 seq the DmdEnv in seqDmdType as well # 2014-10-13 18582472 different machines giving different results.. # 2014-10-13 22220552 use the mean # 2015-06-21 28056344 switch to `+RTS -G1`, tighten bound to 15% # 2015-10-28 23776640 emit Typeable at definition site # 2015-12-11 30837312 TypeInType (see #11196) # 2016-04-14 38776008 Final demand analyzer run (wordsize(32), 18043224, 15) # 2015-07-11 15341228 (x86/Linux, 64-bit machine) use +RTS -G1 # 2016-04-06 18043224 (x86/Linux, 64-bit machine) ]), compiler_stats_num_field('peak_megabytes_allocated', # Note [residency] [(wordsize(64), 121, 15), # 2014-10-13 66 # 2014-10-13 58 seq the DmdEnv in seqDmdType as well # 2014-10-13 49 different machines giving different results... # 2014-10-13 53 use the mean # 2015-06-15 44 reduced for some reason # 2015-06-21 105 switch to `+RTS -G1` # 2015-12-04 88 new pattern checker (D1535) # 2015-12-11 113 TypeInType (see #11196) # 2016-04-14 144 Final demand analyzer run # 2016-07-26 121 Unboxed sums? (wordsize(32), 56, 15) # 2015-07-11 56 (x86/Linux, 64-bit machine) use +RTS -G1 ]), compiler_stats_num_field('bytes allocated', [(wordsize(64), 731171072, 10) # 2014-10-13 544489040 # 2015-10-28 608284152 emit Typeable at definition site # 2017-02-17 731171072 Type-indexed Typeable ,(wordsize(32), 279480696, 10) # 2015-07-11 279480696 (x86/Linux, 64-bit machine) use +RTS -G1 ]), # Use `+RTS -G1` for more stable residency measurements. Note [residency]. extra_hc_opts('+RTS -G1 -RTS') ], compile, ['']) test('T9872a', [ only_ways(['normal']), compiler_stats_num_field('bytes allocated', [(wordsize(64), 3005891848, 5), # 2014-12-10 5521332656 Initally created # 2014-12-16 5848657456 Flattener parameterized over roles # 2014-12-18 2680733672 Reduce type families even more eagerly # 2015-12-11 3581500440 TypeInType (see #11196) # 2016-04-07 3352882080 CSE improvements # 2016-10-19 3134866040 Refactor traceRn interface (#12617) # 2017-02-17 3298422648 Type-indexed Typeable # 2017-02-25 3005891848 Early inlining patch (wordsize(32), 1740903516, 5) # was 1325592896 # 2016-04-06 1740903516 x86/Linux ]), ], compile_fail, ['']) test('T9872b', [ only_ways(['normal']), compiler_stats_num_field('bytes allocated', [(wordsize(64), 3730686224, 5), # 2014-12-10 6483306280 Initally created # 2014-12-16 6892251912 Flattener parameterized over roles # 2014-12-18 3480212048 Reduce type families even more eagerly # 2015-12-11 5199926080 TypeInType (see #11196) # 2016-02-08 4918990352 Improved a bit by tyConRolesRepresentational # 2016-04-06: 4600233488 Refactoring of CSE #11781 # 2016-09-15: 4069522928 Fix #12422 # 2017-02-14 3730686224 Early inlining: 5% improvement (wordsize(32), 2422750696, 5) # was 1700000000 # 2016-04-06 2422750696 x86/Linux ]), ], compile_fail, ['']) test('T9872c', [ only_ways(['normal']), compiler_stats_num_field('bytes allocated', [(wordsize(64), 3404346032, 5), # 2014-12-10 5495850096 Initally created # 2014-12-16 5842024784 Flattener parameterized over roles # 2014-12-18 2963554096 Reduce type families even more eagerly # 2015-12-11 4723613784 TypeInType (see #11196) # 2016-02-08 4454071184 Improved a bit by tyConRolesRepresentational # 2016-04-06: 4306667256 Refactoring of CSE #11781 # 2016-09-15: 3702580928 Fixing #12422 # 2017-02-14 3404346032 Early inlining: 5% improvement (wordsize(32), 2257242896, 5) # was 1500000000 # 2016-04-06 2257242896 ]), ], compile_fail, ['']) test('T9872d', [ only_ways(['normal']), compiler_stats_num_field('bytes allocated', [(wordsize(64), 462817352, 5), # 2014-12-18 796071864 Initally created # 2014-12-18 739189056 Reduce type families even more eagerly # 2015-01-07 687562440 TrieMap leaf compression # 2015-03-17 726679784 tweak to solver; probably flattens more # 2015-12-11 566134504 TypeInType; see #11196 # 2016-02-08 534693648 Improved a bit by tyConRolesRepresentational # 2016-03-18 506691240 optimize Unify & zonking # 2016-12-05 478169352 using tyConIsTyFamFree, I think, but only # a 1% improvement 482 -> 478 # 2017-02-17 535565128 Type-indexed Typeable # 2017-02-25 498855104 Early inlining # 2017-03-03 462817352 Share Typeable KindReps (wordsize(32), 264566040, 5) # some date 328810212 # 2015-07-11 350369584 # 2016-04-06 264566040 x86/Linux ]), ], compile, ['']) test('T9961', [ only_ways(['normal']), compiler_stats_num_field('bytes allocated', [(wordsize(64), 498326216, 5), # 2015-01-12 807117816 Initally created # 2015-spring 772510192 Got better # 2015-05-22 663978160 Fix for #10370 improves it more # 2015-10-28 708680480 x86_64/Linux Emit Typeable at definition site # 2015-12-17 745044392 x86_64/Darwin Creep upwards # 2016-03-20 519436672 x64_64/Linux Don't use build desugaring for large lists (#11707) # 2016-03-24 568526784 x64_64/Linux Add eqInt* variants (#11688) # 2016-09-01 537297968 x64_64/Linux Restore w/w limit (#11565) # 2016-12-19 571246936 x64_64/Linux Join points (#12988) # 2017-02-14 498326216 Early inline patch; 13% improvement (wordsize(32), 275264188, 5) # was 375647160 # 2016-04-06 275264188 x86/Linux ]), ], compile, ['-O']) test('T9233', [ only_ways(['normal']), compiler_stats_num_field('bytes allocated', [(wordsize(64), 974530192, 5), # 2015-08-04 999826288 initial value # 2016-04-14 1066246248 Final demand analyzer run # 2016-06-18 984268712 shuffling around of Data.Functor.Identity # 2017-01-20 920101608 Improvement to SetLevels apparently saved 4.2% in # compiler allocation. Program size seems virtually # unchanged; maybe the compiler itself is a little faster # 2017-01-23 861862608 worker/wrapper evald-ness flags; another 5% improvement! # 2017-02-01 894486272 Join points # 2017-02-07 884436192 Another improvement to SetLevels # 2017-02-17 974530192 Type-indexed Typeable (wordsize(32), 515672240, 5) # Put in your value here if you hit this # 2016-04-06 515672240 (x86/Linux) initial value ]), extra_clean(['T9233a.hi', 'T9233a.o']) ], multimod_compile, ['T9233', '-v0 -O2 -fno-spec-constr']) test('T10370', [ only_ways(['optasm']), compiler_stats_num_field('max_bytes_used', # Note [residency] [(wordsize(64), 41291976, 15), # 2015-10-22 19548720 # 2016-02-24 22823976 Changing Levity to RuntimeRep; not sure why this regresses though, even after some analysis # 2016-04-14 28256896 final demand analyzer run # 2016-08-08 33049304 # This change happened because we changed the behavior # of inlining across hs-boot files, so that we don't # inline if something comes from a boot file. This # affected stats on bootstrapped GHC. However, # when I set -i0.01 with profiling, the heap profiles # were identical, so I think it's just GC noise. # 2016-10-20 38221184 Allow top-level string literals. # See the comment 16 on #8472. # 2017-02-17 51126304 Type-indexed Typeable # 2017-02-27 43455848 Likely drift from recent simplifier improvements # 2017-02-25 41291976 Early inline patch (wordsize(32), 11371496, 15), # 2015-10-22 11371496 ]), compiler_stats_num_field('peak_megabytes_allocated', # Note [residency] [(wordsize(64), 154, 15), # 2015-10-22 76 # 2016-04-14 101 final demand analyzer run # 2016-08-08 121 see above # 2017-01-18 146 Allow top-level string literals in Core # 2017-02-17 187 Type-indexed Typeable # 2017-02-25 154 Early inline patch (wordsize(32), 39, 15), # 2015-10-22 39 ]), # Use `+RTS -G1` for more stable residency measurements. Note [residency]. extra_hc_opts('+RTS -G1 -RTS') ], compile, ['']) test('T10547', [ compiler_stats_num_field('bytes allocated', [(platform('x86_64-unknown-mingw32'), 37485128, 20), # 2017-02-19 37485128 (x64/Windows) - Unknown (wordsize(64), 37681360, 20), # initial: 39165544 # 2016-11-25: 31041520 Linux Around the time of refactoring the constraint solver; # but I think that only pushed it over the edge # 2017-02-20: 38681216 Linux Type-indexed Typeable ]), ], compile_fail, ['-fprint-expanded-synonyms']) test('T12227', [ only_ways(['normal']), compiler_stats_num_field('bytes allocated', [(wordsize(64), 1060158624, 5), # 2016-07-11 5650186880 (Windows) before fix for #12227 # 2016-07-11 1822822016 (Windows) after fix for #12227 # 2016-12-20 1715827784 after d250d493 (INLINE in Traversable dms) # (or thereabouts in the commit history) # 2017-02-14 1060158624 Early inlining: 35% improvement ]), ], compile, # Use `-M1G` to prevent memory thrashing with ghc-8.0.1. ['-O2 -ddump-hi -ddump-to-file +RTS -M1G']) test('T12425', [ only_ways(['optasm']), compiler_stats_num_field('bytes allocated', [(wordsize(64), 142256192, 5), # initial: 125831400 # 2017-01-18: 133380960 Allow top-level string literals in Core # 2017-02-17: 153611448 Type-indexed Typeable # 2017-03-03: 142256192 Share Typeable KindReps ]), ], compile, ['']) test('T12234', [ only_ways(['optasm']), compiler_stats_num_field('bytes allocated', [(platform('x86_64-unknown-mingw32'), 89180624, 5), # initial: 83032768 # 2017-02-19 89180624 (x64/Windows) - Unknown (wordsize(64), 80245640, 5), # initial: 72958288 # 2016-01-17: 76848856 (x86-64, Linux. drift?) # 2017-02-01: 80882208 (Use superclass instances when solving) # 2017-02-05: 74374440 (Probably OccAnal fixes) # 2017-02-17: 86525344 (Type-indexed Typeable) # 2017-02-25: 83032768 (Early inline patch) ]), ], compile, ['']) test('T13035', [ only_ways(['normal']), compiler_stats_num_field('bytes allocated', [(wordsize(64), 98390488, 5), # 2017-01-05 90595208 initial # 2017-01-19 95269000 Allow top-level string literals in Core # 2017-02-05 88806416 Probably OccAnal fixes # 2017-02-17 103890200 Type-indexed Typeable # 2017-02-25 98390488 Early inline patch ]), ], compile, [''] ) test('T13056', [ only_ways(['optasm']), compiler_stats_num_field('bytes allocated', [(wordsize(64), 440548592, 5), # 2017-01-06 520166912 initial # 2017-01-31 546800240 Join points (#12988) # 2017-02-07 524611224 new SetLevels # 2017-02-14 440548592 Early inline patch: 16% improvement ]), ], compile, ['-O1']) test('T12707', [ compiler_stats_num_field('bytes allocated', [(wordsize(64), 1231809592, 5), # initial: 1271577192 # 2017-01-22: 1348865648 Allow top-level strings in Core # 2017-01-31: 1280336112 Join points (#12988) # 2017-02-11: 1310037632 Check local family instances vs imports # 2017-02-23: 1386110512 Type-indexed Typeable? (on Darwin) # 2017-03-02: 1231809592 Drift from recent simplifier improvements ]), ], compile, [''])