Commit Graph

3085 Commits

Author SHA1 Message Date
zhxchen17
4b76c684bb Add "prof.dump_prefix" to override filename prefixes for dumps. 2019-09-12 22:26:03 -07:00
zhxchen17
242af439b8 Rename "prof_dump_seq_mtx" to "prof_dump_filename_mtx". 2019-09-12 22:26:03 -07:00
Giridhar Prasath R
e06658cb24 check GNU make exists in path
Signed-off-by: Giridhar Prasath R <cristianoprasath@gmail.com>
2019-09-11 16:36:19 -07:00
Qi Wang
22bc75ee3e Workaround the stringop-overflow check false positives. 2019-09-09 11:35:04 -07:00
Yinan Zhang
93d6151800 Pass tsd down to prof_backtrace() 2019-09-05 10:57:43 -07:00
Yinan Zhang
671f120e26 Fix prof_backtrace() reentrancy level 2019-09-05 10:57:43 -07:00
Qi Wang
785b84e603 Make cache_bin_sz_t unsigned.
The bin size type was made signed only because the low_water could go -1, which
was already removed.
2019-09-04 13:37:07 -07:00
Qi Wang
23dc7a7fba Fix index type for cache_bin_alloc_easy. 2019-09-04 13:37:07 -07:00
Qi Wang
2abb02ecd7 Fix MSVC 2015 build, as proposed by @christianaguilera-foundry. 2019-08-28 23:37:24 -07:00
Qi Wang
719583f14a Fix large.nflushes in the merged stats. 2019-08-28 23:37:00 -07:00
Yinan Zhang
adce29c885 Optimize for prof_active off
Move the handling of `prof_active` off case completely to slow path,
so as to reduce register pressure on malloc fast path.
2019-08-27 14:48:56 -07:00
Yinan Zhang
49e6fbce78 Always adjust thread_(de)allocated 2019-08-26 11:56:41 -07:00
Yinan Zhang
57b81c078e Pull thread_(de)allocated out of config_stats 2019-08-26 11:56:41 -07:00
Yinan Zhang
9e031c1d11 Bug fix for prof_active switch
The bug is subtle but critical: if application performs the following
three actions in sequence: (a) turn `prof_active` off, (b) make at
least one allocation that triggers the malloc slow path via the
`if (unlikely(bytes_until_sample < 0))` path, and (c) turn
`prof_active` back on, then the application would never get another
sample (until a very very long time later).

The fix is to properly reset `bytes_until_sample` rather than
throwing it all the way to `SSIZE_MAX`.

A side minor change is to call `prof_active_get_unlocked()` rather
than directly grabbing the `prof_active` variable - it is the very
reason why we defined the `prof_active_get_unlocked()` function.
2019-08-22 13:00:10 -07:00
Qi Wang
0043e68d4c Track low_water == -1 case explicitly.
The -1 value of low_water indicates if the cache has been depleted and
refilled.  Track the status explicitly in the tcache struct.

This allows the fast path to check if (cur_ptr > low_water), instead of >=,
which avoids reaching slow path when the last item is allocated.
2019-08-21 16:00:38 -07:00
Qi Wang
937ca1db9f Store ncached_max * ptr_size in tcache_bin_info.
With the cache bin metadata switched to pointers, ncached_max is usually
accessed and timed by sizeof(ptr). Store the results in tcache_bin_info for
direct access, and add a helper function for the ncached_max value.
2019-08-19 12:23:24 -07:00
Qi Wang
7599c82d48 Redesign the cache bin metadata for fast path.
Implement the pointer-based metadata for tcache bins --
- 3 pointers are maintained to represent each bin;
- 2 of the pointers are compressed on 64-bit;
- is_full / is_empty done through pointer comparison;

Comparing to the previous counter based design --
- fast-path speed up ~15% in benchmarks
- direct pointer comparison and de-reference
- no need to access tcache_bin_info in common case
2019-08-19 12:21:44 -07:00
Qi Wang
d2dddfb82a Add hint in the bogus version string. 2019-08-16 16:08:18 -07:00
Qi Wang
d6b7995c16 Update INSTALL.md about the default doc build. 2019-08-16 10:03:34 -07:00
Qi Wang
e2c7584361 Simplify / refactor tcache_dalloc_large. 2019-08-14 13:08:23 -07:00
Qi Wang
9c5c2a2c86 Unify the signature of tcache_flush small and large. 2019-08-14 13:08:23 -07:00
Yinan Zhang
28ed9b9a51 Buffer stats printing
Without buffering `malloc_stats_print` would invoke the write back
call (which could mean an expensive `malloc_write_fd` call) for every
single `printf` (including printing each line break and each leading
tab/space for indentation).
2019-08-13 09:40:11 -07:00
Yinan Zhang
eb70fef8ca Make compact json format as default
Saves 20-50% of the output size.
2019-08-12 13:59:50 -07:00
Yinan Zhang
a219cfcda3 Clear tcache prof_accumbytes in tcache_flush_cache
`tcache->prof_accumbytes` should always be cleared after being
transferred to arena; otherwise the allocations would be double
counted, leading to excessive prof dumps.
2019-08-12 09:08:09 -07:00
Yinan Zhang
ad3f7dbfa0 Buffer prof_log_stop
Make use of the new buffered writer for the output of `prof_log_stop`.
2019-08-12 09:06:01 -07:00
Qi Wang
5934846612 Fix large bin index accessed through cache bin descriptor. 2019-08-11 16:31:12 -07:00
Qi Wang
22746d3c9f Properly dalloc prof nodes with idalloctm.
The prof_alloc_node is allocated through ialloc as internal.  Switch to
idalloctm with tcache and is_internal properly set.
2019-08-09 10:29:49 -07:00
Yinan Zhang
8c8466fa6e Add compact json option for emitter
JSON format is largely meant for machine-machine communication, so
adding the option to the emitter.  According to local testing, the
savings in terms of bytes outputted is around 50% for stats printing
and around 25% for prof log printing.
2019-08-09 09:53:41 -07:00
Yinan Zhang
7fc6b1b259 Add buffered writer
The buffered writer adopts a signature identical to `write_cb`,
so that it can be plugged into anywhere `write_cb` appears.
2019-08-09 09:44:29 -07:00
Yinan Zhang
39343555d6 Report stats for tdatas_mtx and prof_dump_mtx 2019-08-09 09:24:16 -07:00
Qi Wang
87e2400cbb Fix tcaches mutex pre- / post-fork handling. 2019-08-08 10:55:32 -07:00
Yinan Zhang
07ce2434bf Refactor profiling
Refactored core profiling codebase into two logical parts:

(a) `prof_data.c`: core internal data structure managing & dumping;
(b) `prof.c`: mutexes & outward-facing APIs.

Some internal functions had to be exposed out, but there are not
that many of them if the modularization is (hopefully) clean enough.
2019-08-07 19:48:28 -07:00
Yinan Zhang
56126d0d2d Refactor prof log
Prof logging is conceptually seperate from core profiling, so
split it out as a module of its own.  There are a few internal
functions that had to be exposed but I think it is a fair trade-off.
2019-08-07 13:53:45 -07:00
Yinan Zhang
56c8ecffc1 Correct tsd layout graph
Augmented the tsd layout graph so that the two recently added fields,
`offset_state` and `bytes_until_sample`, are properly reflected.
As is shown, the cache footprint is 16 bytes larger than before.
2019-08-05 15:30:20 -07:00
Qi Wang
ea6b3e973b Merge branch 'dev' 2019-08-05 12:59:21 -07:00
Qi Wang
0cfa36a58a Update Changelog for 5.2.1. 2019-08-05 12:52:43 -07:00
Qi Wang
8a94ac25d5 Sanity check on prof dump buffer size. 2019-08-01 17:55:45 -07:00
Yinan Zhang
82b8aaaeb6 Quick fix for prof log printing
The emitter APIs used were incorrect, a side effect of which was
extra lines being printed.
2019-07-30 19:31:28 -07:00
Yinan Zhang
9344d25488 Workaround to address g++ unused variable warnings
g++ 5.5.0+ complained `parameter ‘expected’ set but not used
[-Werror=unused-but-set-parameter]` (despite that `expected` is in
fact used).
2019-07-30 11:37:56 -07:00
Qi Wang
c9cdc1b27f Limit to exact fit on Windows with retain off.
W/o retain, split and merge are disallowed on Windows.  Avoid doing first-fit
which needs splitting almost always.  Instead, try exact fit only and bail out
early.
2019-07-29 16:19:36 -07:00
Qi Wang
5742473cc8 Revert "Refactor prof log"
This reverts commit 7618b0b8e4.
2019-07-29 14:10:15 -07:00
Qi Wang
1a0503367b Revert "Refactor profiling"
This reverts commit 0b462407ae.
2019-07-29 14:10:15 -07:00
Yinan Zhang
0b462407ae Refactor profiling
Refactored core profiling codebase into two logical parts:

(a) `prof_data.c`: core internal data structure managing & dumping;
(b) `prof.c`: mutexes & outward-facing APIs.

Some internal functions had to be exposed out, but there are not
that many of them if the modularization is (hopefully) clean enough.
2019-07-29 13:55:00 -07:00
Yinan Zhang
7618b0b8e4 Refactor prof log
`prof.c` is growing too long, so trying to modularize it.  There are
a few internal functions that had to be exposed but I think it is a
fair trade-off.
2019-07-29 13:55:00 -07:00
Qi Wang
85f0cb2d0c Add indent to individual options for confirm_conf. 2019-07-25 17:00:31 -07:00
Qi Wang
9f6a9f4c1f Update manual for opt.retain (new default on Windows). 2019-07-25 15:25:58 -07:00
Qi Wang
10fcff6c38 Lower nthreads in test/unit/retained on 32-bit to avoid OOM. 2019-07-25 13:10:03 -07:00
Qi Wang
a3fa597921 Refactor arena_dalloc() / _sdalloc(). 2019-07-24 18:30:54 -07:00
Qi Wang
bc0998a905 Invoke arena_dalloc_promoted() properly w/o tcache.
When tcache was disabled, the dalloc promoted case was missing.
2019-07-24 18:30:54 -07:00
Qi Wang
1d148f353a Optimize max_active_fit in first_fit.
Stop scanning once reached the first max_active_fit size.
2019-07-24 11:28:45 -07:00