mirror/tracy
1
0
Fork 0
mirror of https://github.com/wolfpld/tracy.git synced 2024-09-20 13:52:18 +00:00
Code Issues Packages Projects Releases Wiki Activity
4,063 Commits 2 Branches 28 Tags 40 MiB
c0f49c648b
Commit Graph

534 Commits

Author SHA1 Message Date
Bartosz Taudul
50b96c757e Context switch usage reconstruction skeleton. 2019-11-05 01:28:44 +01:00
Bartosz Taudul
d9c3238462 Save 2 bytes per PlotItem. 
Memory savings:

android     2614 MB -> 2487 MB (95%)
chicken     1932 MB -> 1852 MB (95%)
mem         6067 MB -> 5747 MB (94%)
q3bsp-mt    5059 MB -> 5017 MB (99%)
q3bsp-st    1211 MB -> 1171 MB (96%)
 2019-11-03 16:29:45 +01:00
Bartosz Taudul
4bc1588a5e Clear proper vector. 2019-11-02 16:57:18 +01:00
Bartosz Taudul
0df29d1e0b Use short ptr for source location payload data. 2019-11-02 16:54:12 +01:00
Bartosz Taudul
72efbe28ed Use short ptr for message data. 2019-11-02 16:54:12 +01:00
Bartosz Taudul
1e4022e05b Use proper comparison. 2019-11-02 16:54:12 +01:00
Bartosz Taudul
a40bbacb17 Use short ptr for CPU zone data. 2019-11-02 16:54:12 +01:00
Bartosz Taudul
cb20bf01f9 Use short ptr for GPU zone data. 2019-11-02 16:54:11 +01:00
Bartosz Taudul
c7664b0a98 Use short ptr in LockEventPtr. 2019-11-02 16:17:45 +01:00
Bartosz Taudul
45ff14d678 Fix saving source location payload data. 2019-11-02 14:28:59 +01:00
Bartosz Taudul
16bc862904 Save sizes of children vectors to prevent reallocation. 2019-11-02 12:38:32 +01:00
Bartosz Taudul
13b656fe61 Make srcloc dynamic color depend on function name. 2019-11-01 20:17:25 +01:00
Bartosz Taudul
39988ad636 Check for shutdown in background processing thread. 2019-10-31 21:41:21 +01:00
Bartosz Taudul
456deefdbc Keep child idx on stack. 2019-10-30 23:55:21 +01:00
Bartosz Taudul
25b610a36f Pack child into GPU start/end in GpuEvent (saves 4 bytes). 
long    5152 MB -> 5061 MB
 2019-10-30 23:50:37 +01:00
Bartosz Taudul
e8286600d1 Use -1 as invalid GPU start time. 2019-10-30 23:12:43 +01:00
Bartosz Taudul
7ce8c772ad Disallow negative GPU times. 
Shouldn't happen, but GPU timestamps are a shitshow, so better be safe
than sorry.
 2019-10-30 22:37:07 +01:00
Bartosz Taudul
ae4794ab4c Save 2 bytes in ContextSwitchData and ContextSwitchCpu. 2019-10-30 22:25:46 +01:00
Bartosz Taudul
99d198d0bf Pack csAlloc in MemEvent (saves 3 bytes). 
Memory usage change on selected traces:

android     2699 MB -> 2613 MB
chicken     2019 MB -> 2007 MB
mem         6308 MB -> 6068 MB
q3bsp-mt    5283 MB -> 5252 MB
q3bsp-st    1241 MB -> 1211 MB
 2019-10-30 22:01:13 +01:00
Bartosz Taudul
6f0dc2885f Fix connection abort. 2019-10-28 23:32:51 +01:00
Bartosz Taudul
8050622b0f Read and decompress network data on a separate thread. 2019-10-28 23:22:50 +01:00
Bartosz Taudul
e0356ae01e Cosmetics. 2019-10-28 22:53:06 +01:00
Bartosz Taudul
99b7e8ad92 Close socket when shutting down. 2019-10-28 22:52:52 +01:00
Bartosz Taudul
788ca2e5df Spawn no-op network thread. 2019-10-28 22:45:10 +01:00
Bartosz Taudul
7f07f5beb4 Free child time stack. 2019-10-26 23:32:16 +02:00
Bartosz Taudul
01985f50ef Cache source location zones counter search. 2019-10-26 16:33:40 +02:00
Bartosz Taudul
1d0084aa28 Add cache for last accessed source location zones. 2019-10-25 21:29:55 +02:00
Bartosz Taudul
b5419944aa Only write to memory if value has changed. 2019-10-25 21:28:55 +02:00
Bartosz Taudul
779063a18b Cache last shrinked source location. 2019-10-25 21:07:28 +02:00
Bartosz Taudul
294793367f Cache last CheckSourceLocation query. 
Just knowing that the query was performed is enough here -- this
function adds a new source location entry, if there already isn't one.
 2019-10-25 21:01:33 +02:00
Bartosz Taudul
0f2503d334 Send time deltas in GPU time events. 2019-10-25 19:52:01 +02:00
Bartosz Taudul
8fa5188176 Send delta times for context switches. 2019-10-25 19:13:11 +02:00
Bartosz Taudul
29c42cc8d7 Fix assert. 2019-10-25 01:00:32 +02:00
Bartosz Taudul
17a51c898e No need to check if vector is empty. 2019-10-25 00:54:46 +02:00
Bartosz Taudul
b5e759bc5a Don't calculate child index twice. 2019-10-25 00:54:46 +02:00
Bartosz Taudul
70f1074490 Don't iterate over children to calculate zone self time. 2019-10-25 00:33:44 +02:00
Bartosz Taudul
1fe76be955 Don't reconstruct lock event time on insert. 2019-10-24 23:25:04 +02:00
Bartosz Taudul
b83d0f46d9 Improve updating last time. 
Avoid LHS, don't write if don't need to.
 2019-10-24 23:23:52 +02:00
Bartosz Taudul
721f3c8925 Callstack is already zero-initialized. 2019-10-24 23:05:39 +02:00
Bartosz Taudul
c9da5f1474 Use cached thread retriever. 2019-10-24 22:34:18 +02:00
Bartosz Taudul
5873561b54 Add cached thread retriever. 2019-10-24 22:33:48 +02:00
Bartosz Taudul
06bc802107 Avoid load-hit-store. 2019-10-24 22:24:00 +02:00
Bartosz Taudul
1cfb5adc44 Count transferred data size. 2019-10-24 00:47:16 +02:00
Bartosz Taudul
ba61a9ed84 Transfer time deltas, not absolute times. 
This change significantly reduces network bandwidth requirements.

Implemented for:
- CPU zones,
- GPU zones,
- locks,
- plots,
- memory events.
 2019-10-24 00:06:41 +02:00
Bartosz Taudul
5c92eae3ed Add early exit for invalid times. 2019-10-20 18:47:50 +02:00
Bartosz Taudul
4d761def61 Microoptimize comparison. 2019-10-16 20:26:39 +02:00
Bartosz Taudul
3ae5c125f6 Implement counting CPU usage (ctx switch) at a given time. 2019-10-15 16:54:43 +02:00
Bartosz Taudul
3ce6b1205f Don't iterate over 256 CPUs. 2019-10-15 16:13:53 +02:00
Bartosz Taudul
eccb0b1e4a Track max CPU present in context switch data. 2019-10-15 16:13:53 +02:00
Bartosz Taudul
bdb8516d04 Make sure context switch end time wasn't set already. 2019-10-15 14:54:28 +02:00
Bartosz Taudul
215dc8a804 More compact GpuEvent struct (save 4 bytes). 
Memory usage reduction of various traces:

big         9011 -> 9007
frameimages 561  -> 552
fi-big      4144 -> 4139
long        5253 -> 5125
 2019-10-13 14:42:52 +02:00
Bartosz Taudul
5e1894dd79 Count GPU zones. 2019-10-13 14:13:04 +02:00
Bartosz Taudul
65ea33a60f Store memory callstack data as 24-bit ints. 
This reduces MemEvent size from 40 to 38 bytes.

Memory usage reduction:

chicken     2027 -> 2019
mem         6468 -> 6308
q3bsp-mt    5304 -> 5283
 2019-10-01 22:38:17 +02:00
Bartosz Taudul
f0b957ec56 Store callstacks on 24 bits. 
ZoneEvent is now 27 bytes.

Memory usage reduction on selected traces (sizes in MB):

big             9224 -> 9011  (97%)
chicken         2044 -> 2027  (99%)
drl-l-b         1443 -> 1383  (95%)
long            5327 -> 5253  (98%)
q3bsp-mt        5400 -> 5304  (98%)
selfprofile     1403 -> 1382  (98%)
 2019-10-01 22:38:17 +02:00
Bartosz Taudul
717a212563 Save another 2 bytes per ZoneEvent. 
ZoneEvent is not 28 bytes.

Memory usage reduction on selected traces (sizes in MB):

big             9527 -> 9224  (96%)
chicken         2107 -> 2044  (97%)
drl-l-b         1479 -> 1443  (97%)
long            5412 -> 5327  (98%)
q3bsp-mt        5592 -> 5400  (96%)
selfprofile     1443 -> 1403  (97%)
 2019-10-01 01:05:37 +02:00
Bartosz Taudul
2470936050 Don't perform background tasks during trace upgrade. 2019-09-29 20:52:25 +02:00
Bartosz Taudul
d228bcb622 Pack StringIdx in 24 bits. 
This reduces ZoneEvent size from 32 to 30 bytes.

Memory usage reduction on selected traces (sizes in MB):

big             9902 -> 9527  (96%)
chicken         2172 -> 2107  (97%)
ctx-big          311 ->  309  (99%)
drl-l-b         1570 -> 1479  (94%)
long            5496 -> 5412  (98%)
mem             6468 -> 6468  (100%)
q3bsp-mt        5784 -> 5592  (96%)
selfprofile     1486 -> 1443  (97%)
 2019-09-29 20:32:42 +02:00
Aleksei Skriabin
c0c2f4536a strstr_nocase() typo fix. 2019-09-28 14:20:29 +05:00
Bartosz Taudul
a5ba74ed13 Handle multiple Vulkan threads. 2019-09-23 17:27:49 +02:00
Bartosz Taudul
82cd667b30 Allow specifying network port in server. 2019-09-21 15:43:01 +02:00
Bartosz Taudul
d8e0853cd8 Multithreaded frame image compression. 2019-09-20 23:03:12 +02:00
Bartosz Taudul
e1e5d6bd47 Add const version of PackFrameImage(). 
Temporary buffer needs to be handled outside of the function.
 2019-09-20 22:55:55 +02:00
Bartosz Taudul
8fe9b56b6f Calculate frame statistics. 2019-09-16 22:02:47 +02:00
Bartosz Taudul
7673028dba Fix skipping memory data. 2019-09-16 15:42:25 +02:00
Bartosz Taudul
cdc4575dba Setup tid -> thread data mapping when loading trace. 2019-09-08 14:15:40 +02:00
Bartosz Taudul
ea6a0a58a7 Thread data accessor. 2019-09-08 14:07:16 +02:00
Bartosz Taudul
aac0a36a2d Don't use source location zones before they are ready. 2019-09-07 17:23:11 +02:00
Bartosz Taudul
86cb477811 Pack ZoneThreadData. 
This reduces struct size from 10 to 8 bytes. Assumes 48-bit pointers
(4-level paging)!

Memory savings (MB):

android     2766    ->  2757    (99%)
big         10.29 G ->  9902    (96%)
chicken     2244    ->  2172    (96%)
ctx-android 228     ->  224     (98%)
drl-l-b     1635    ->  1570    (96%)
gn-vulkan   244     ->  240     (98%)
long        5656    ->  5496    (97%)
q3bsp-mt    6043    ->  5784    (95%)
selfprofile 1554    ->  1486    (95%)
 2019-08-31 00:55:51 +02:00
Bartosz Taudul
3ec534cdf3 Prevent "ntdll.dll" from appearing as a thread name. 2019-08-30 23:09:07 +02:00
Bartosz Taudul
1c0c6311ec Fix skipping data when loading traces. 2019-08-30 01:16:42 +02:00
Bartosz Taudul
a2f968d843 Compress thread id in MessageData. 2019-08-28 21:03:01 +02:00
Bartosz Taudul
fd5014be6f GetThreadString() is no longer used. 2019-08-28 20:08:16 +02:00
Bartosz Taudul
3c092b4bec Add thread name getter combining local and external thread names. 2019-08-27 23:00:13 +02:00
Bartosz Taudul
f76f38777e Signed minus unsigned is unsigned... 2019-08-26 19:09:12 +02:00
Bartosz Taudul
1712431dfd Compress external threads. Saves 4 bytes per ctx switch. 
Dropped support for loading context switch data in previous versions of
traces.
 2019-08-19 23:09:58 +02:00
Bartosz Taudul
21e7a4bb16 Extract thread compression into a separate class. 2019-08-19 22:56:58 +02:00
Bartosz Taudul
94382f54ca Move FileVersion() to TracyFileHeader.hpp. 2019-08-19 22:56:58 +02:00
Bartosz Taudul
19857473e3 Also collect information on local threads. 2019-08-18 14:56:17 +02:00
Bartosz Taudul
3b8518f7b6 Save/load CPU thread data. 2019-08-18 01:53:38 +02:00
Bartosz Taudul
62dbe522c5 Add accessors. 2019-08-18 01:51:02 +02:00
Bartosz Taudul
103645c2fa Calculate cpu thread data statistics. 2019-08-18 01:50:49 +02:00
Bartosz Taudul
1498417a8d Save/load tid to pid mapping. 2019-08-17 22:36:21 +02:00
Bartosz Taudul
20e8a5ecc8 Create tid to pid mapping. 2019-08-17 22:32:41 +02:00
Bartosz Taudul
678e942e9f Transfer PID of profiled program. 2019-08-17 22:19:04 +02:00
Bartosz Taudul
414f903cc5 Collect thread wakeup data. 2019-08-17 17:05:29 +02:00
Bartosz Taudul
26be78530f Use signed number to calculate frame offset. 2019-08-17 15:22:54 +02:00
Bartosz Taudul
6c53cac15e Fix uninitialized variable. 2019-08-16 21:20:04 +02:00
Bartosz Taudul
e975c4d7bf Also retrieve external thread names. 2019-08-16 19:49:16 +02:00
Bartosz Taudul
ccaf92afc4 Save/load external process names. 2019-08-16 19:24:38 +02:00
Bartosz Taudul
fe7f56b022 Implement retrieval of external process names. 2019-08-16 19:22:23 +02:00
Bartosz Taudul
c212661714 Allow determining whether thread is local to profiled program. 2019-08-16 17:59:25 +02:00
Bartosz Taudul
cef7e4b8d0 Save/load per-cpu context switches. 2019-08-16 16:51:18 +02:00
Bartosz Taudul
8bc4258e29 Display count of per-cpu context switch data. 2019-08-16 16:51:18 +02:00
Bartosz Taudul
69527d2f71 Collect per-cpu context switch data. 2019-08-16 16:51:18 +02:00
Bartosz Taudul
42c71d7e46 Fix loading old traces. 2019-08-16 00:24:29 +02:00
Bartosz Taudul
889eddd646 Pack ContextSwitchData. Saves 3 bytes per context switch region. 2019-08-15 23:53:47 +02:00
Bartosz Taudul
c22c259a13 Pack time and thread in MemEvent. 
This saves 4 bytes per logged memory allocation. Memory savings for
selected traces:

android     2945 MB -> 2766 MB
chicken     2261 MB -> 2245 MB
q3bsp-mt    6085 MB -> 6043 MB
mem         6788 MB -> 6468 MB
 2019-08-15 23:02:43 +02:00
Bartosz Taudul
9618ee3581 Fix skipping locks. 2019-08-15 22:24:27 +02:00
Bartosz Taudul
8b73dece98 Preserve magic time values when loading old traces. 2019-08-15 21:30:37 +02:00
Bartosz Taudul
3db3952135 Hackfix for broken lock terminate times. 2019-08-15 20:45:00 +02:00
Bartosz Taudul
5e20b3f28a Pack time and source location in LockEvent. 2019-08-15 20:39:16 +02:00
Bartosz Taudul
bf3ad57456 Pack start time and srcloc together in ZoneEvent. 
This reduces ZoneEvent struct size by 2 bytes. Memory savings on various
captures:

10.62 GB -> 10.29 GB
 2342 MB ->  2276 MB
 1706 MB ->  1635 MB
 6277 MB ->  6085 MB
 2019-08-15 20:17:36 +02:00
Bartosz Taudul
042e6c9e11 Set initial time of old traces to 0. 2019-08-15 20:17:36 +02:00
Bartosz Taudul
b322d20c19 Store received timestamps offset to 0. 2019-08-15 20:17:36 +02:00
Bartosz Taudul
659907c972 Store srcloc identifiers using 16 bit. 
This reduces various structure sizes by 2 bytes. Memory usage reduction
on various traces:

big               11 GB -> 10.62 GB
chicken         2436 MB ->  2342 MB
drl-light-big   1761 MB ->  1706 MB
q3bsp-mt        6469 MB ->  6277 MB
 2019-08-15 20:15:48 +02:00
Bartosz Taudul
416113fdcb Drop support for ETC1 frame images. 2019-08-15 16:29:50 +02:00
Bartosz Taudul
9a364fe5fe Cache context switch data queries. 2019-08-14 20:16:11 +02:00
Bartosz Taudul
3e01ca3269 Calculate how long thread was in running time. 2019-08-14 17:12:48 +02:00
Bartosz Taudul
0bb0c10e3c Revert "Save one byte on ContextSwitchData." 
Counting bits is hard, let's go shopping.
 2019-08-14 13:55:05 +02:00
Bartosz Taudul
3996516fce One more SetThreadName() to change. 2019-08-14 02:27:01 +02:00
Bartosz Taudul
f285e0f5cc Save one byte on ContextSwitchData. 2019-08-13 15:16:46 +02:00
Bartosz Taudul
9417ad994d Save/load context switch data. 2019-08-13 13:10:37 +02:00
Bartosz Taudul
1c937ad9bb Implement skipping frame image data. 2019-08-13 02:35:32 +02:00
Bartosz Taudul
8c494eabbf Display number of context switch regions. 2019-08-13 02:35:32 +02:00
Bartosz Taudul
0b03fed61c Add context switch accessor. 2019-08-13 02:35:32 +02:00
Bartosz Taudul
419f74280d Store context switches. 2019-08-13 02:35:32 +02:00
Bartosz Taudul
8aa0be39d5 Drop support for CPU id queries. 2019-08-12 23:05:34 +02:00
Bartosz Taudul
d6f32a0839 Serialize lock processing. 
This makes is much easier to process on the server and opens new
optimization possibilities. It also fixes theoretical problems, which
may be caused by invalid ordering of events with the same timestamp.
 2019-08-12 13:51:01 +02:00
Bartosz Taudul
6398ecb344 Drop support for pre-0.4 traces. 2019-08-12 12:36:37 +02:00
Bartosz Taudul
a76622d17a Cache last searched ThreadData. 2019-08-03 14:35:01 +02:00
Bartosz Taudul
12969ee497 Track thread context. 
This change exploits the fact that events are processed in batches
originating from a single thread. A single message changing thread
context is enough to handle multiple messages, as opposed to inclusion
of thread identifier in each message.
 2019-08-02 20:18:08 +02:00
Bartosz Taudul
a4e7a341c0 Proper handling of disconnect request. 2019-08-01 23:14:09 +02:00
Bartosz Taudul
dc49f2f76a Move DXT1 index conversion to server. 2019-07-19 21:46:58 +02:00
Bartosz Taudul
2e774f4626 Save/load application info. 2019-07-12 18:45:35 +02:00
Bartosz Taudul
d64ab7db5a Store app info messages. 2019-07-12 18:34:46 +02:00
Bartosz Taudul
10bcc8c770 Switch to DXT1 textures in profiler utility. 2019-06-27 19:14:51 +02:00
Bartosz Taudul
7dc7ece2bd Add staging area for frame images. 
Compressing frame images on a separate thread may cause frame image
arrival before frames are sent. Fix this issue by creating a staging
area in which frame images will wait for frames to arrive.

This probably breaks playback functionality, as non-existent frames may
be queried, but this problem seems to be very hard to find, so let's
ignore it for now.
 2019-06-27 13:24:35 +02:00
Bartosz Taudul
bb35f9a897 Compress frame images in a separate thread. 2019-06-27 13:24:35 +02:00
Bartosz Taudul
1c41229766 Use proper type for buffer size comparison. 2019-06-22 14:07:53 +02:00
Bartosz Taudul
6a82f666a7 Cosmetics. 2019-06-22 14:05:18 +02:00
Bartosz Taudul
eb4c7ca9ea Ignore useless warnings. 2019-06-22 13:40:00 +02:00
Bartosz Taudul
850815534e Insert frame mark at beginning of on-demand connection. 2019-06-21 19:39:41 +02:00
Bartosz Taudul
31a4a45b14 Ignore memory free faults if running on apple. 
There's a case in MoltenVK initialization where overloading operator new
and operator delete works for std::string destruction, but not
construction.
 2019-06-13 14:15:17 +02:00
Bartosz Taudul
37d1457b44 Frame image may need flipping. 2019-06-12 15:28:32 +02:00
Bartosz Taudul
eb6ac5e6e1 Store frame reference in frame images. 2019-06-12 00:55:02 +02:00
Bartosz Taudul
5f8eadfb16 Release zone id stack. 2019-06-09 17:56:41 +02:00
Bartosz Taudul
b1f8d9fba1 Send server termination query on server disconnect. 2019-06-09 16:10:49 +02:00
Bartosz Taudul
2c780f1af4 Allow sending immediate termination query from server. 2019-06-09 16:10:49 +02:00
Bartosz Taudul
d6d7b82529 Ignore invalid frame images in on-demand mode. 2019-06-09 15:37:49 +02:00
Bartosz Taudul
50cda7720f Handle frame image instrumentation failures. 2019-06-09 13:44:53 +02:00
Bartosz Taudul
bef1988800 Compress frame images using LZ4. 2019-06-08 12:17:18 +02:00
Bartosz Taudul
fc5a8f7e3a Assign frame image to the correct frame (including offset). 2019-06-07 20:13:08 +02:00
Bartosz Taudul
42a30bffe1 Frame images are now ETC1 compressed. 2019-06-07 00:31:51 +02:00
Bartosz Taudul
646e7327b8 Show loading progress of frame images. 2019-06-06 23:40:37 +02:00
Bartosz Taudul
6b2741ccdb Save/load frame images. 2019-06-06 23:08:19 +02:00
Bartosz Taudul
cd2f572a2f Use proper index. 2019-06-06 22:22:57 +02:00
Bartosz Taudul
af56f41e32 Add frame image accessor. 2019-06-06 22:14:51 +02:00
Bartosz Taudul
34b84bb284 Add frame image index to frame data. 2019-06-06 21:44:48 +02:00
Bartosz Taudul
e5bb6011c5 Frame image transfer prototype. 2019-06-06 21:39:54 +02:00
Bartosz Taudul
5681096486 Track status of worker background tasks. 2019-06-02 15:00:38 +02:00