Bartosz Taudul
51043ebc47
Callstack payload transfer.
2018-06-19 19:31:16 +02:00
Bartosz Taudul
55e6a4a484
No return status is needed here.
2018-06-19 19:00:57 +02:00
Bartosz Taudul
4a01eb7fc4
Windows callstack inspection plumbing.
2018-06-19 01:17:19 +02:00
Bartosz Taudul
7a23f677dd
Vulkan and OpenGL must share idx pool.
2018-06-18 01:10:43 +02:00
Bartosz Taudul
9c11e0fc5b
Vulkan tracing.
2018-06-17 18:14:37 +02:00
Bartosz Taudul
5b6d9769af
Properly separate HW timer from MSVC rdtscp optimization.
2018-04-27 19:40:47 +02:00
Bartosz Taudul
237aee30a8
Test if HW timer can be used on arm.
2018-04-27 16:58:45 +02:00
Bartosz Taudul
8cc9464082
Use GetTime() in CalibrateTimer().
2018-04-26 15:29:09 +02:00
Bartosz Taudul
48665cc09b
s/TRACY_RDTSCP_SUPPORTED/TRACY_HW_TIMER/
2018-04-26 15:25:54 +02:00
Bartosz Taudul
459890ef0e
Don't hold lock on serial queue during dequeue.
2018-04-14 15:46:11 +02:00
Bartosz Taudul
7c4075c9ce
Fix MemRead() call.
2018-04-03 17:57:12 +02:00
Bartosz Taudul
3ea5600900
Fix UB, lose type safety.
2018-04-03 17:51:53 +02:00
Bartosz Taudul
794f199bdc
Serial queue dequeuing.
2018-04-01 20:04:35 +02:00
Bartosz Taudul
faeecdd773
Add serial queue to profiler.
2018-04-01 19:53:05 +02:00
Bartosz Taudul
3b03e849f0
Harden client code against unaligned memory access.
...
There shouldn't be any changes in generated code on modern
architectures, as the memcpy will be reduced to a store/load operation
identical to the one generated with plain struct member access.
GetTime( cpu ) needs special handling, as the MSVC intrinsic for rdtscp
can't store cpu identifier in a register. Using intermediate variable
would cause store to stack, read from stack, store to the destination
address. Since rdtscp is only available on x86, which handles unaligned
stores without any problems, we can have one place with direct struct
member access.
2018-03-31 14:15:04 +02:00
Bartosz Taudul
7300c2e46e
Fix TRACY_NO_EXIT behavior.
...
Terminate event could be the first event that was sent. In such case
server immediately closed the connection, as there was no outstanding
data to receive. Fix by sending all data in the queue before sending
terminate event.
2018-01-11 13:45:13 +01:00
Bartosz Taudul
f67465e784
Reduce timer calibration delay to 200 ms.
2017-11-25 13:34:26 +01:00
Bartosz Taudul
48da593ab2
Increase calibration time to half a second.
2017-11-24 01:43:35 +01:00
Bartosz Taudul
c431747f06
Favor transfer of zones without predicted payload.
2017-11-22 02:28:12 +01:00
Bartosz Taudul
630db7112a
Leaner iteration in Profiler::Dequeue().
2017-11-22 02:07:23 +01:00
Bartosz Taudul
2f669aea41
Workaround gcc issues.
2017-11-15 10:56:27 +01:00
Bartosz Taudul
c43eb29ce0
Don't send source location pointer in query reply.
...
Since reply order is the same as the query order, the server already
knows what source location it receives. This observation allows placing
zone name into the source location struct.
2017-11-14 23:06:45 +01:00
Bartosz Taudul
5c872b2137
Simplify GPU context handling.
2017-11-14 00:48:26 +01:00
Bartosz Taudul
3c00ce0958
GPU context registration.
2017-11-11 19:44:09 +01:00
Bartosz Taudul
81735aea2f
Support for setting zone names in lua.
2017-11-11 17:56:41 +01:00
Bartosz Taudul
59ec40c045
Preemptive transfer of source location payload.
2017-11-11 15:59:30 +01:00
Bartosz Taudul
7f3b8f4647
Preemptive message text delivery.
2017-11-11 15:41:21 +01:00
Bartosz Taudul
76e11174dc
Preemptive sending of custom strings.
2017-11-11 15:22:55 +01:00
Bartosz Taudul
c2797a4cc7
Data packets can't cross data buffer boundary.
2017-11-11 15:08:03 +01:00
Bartosz Taudul
49bce256bc
Fix type mismatch.
2017-11-11 14:35:46 +01:00
Bartosz Taudul
0d15d45c3a
Don't send source location through the queue.
2017-11-11 14:24:22 +01:00
Bartosz Taudul
065964b216
Send data before sleeping during shutdown.
2017-11-11 14:23:55 +01:00
Bartosz Taudul
fc4e31bb8f
Wait for send buffer to be full before sending data.
2017-11-11 14:16:37 +01:00
Bartosz Taudul
fb46b3cf24
Restore process name retrieval on cygwin.
2017-11-10 19:46:31 +01:00
Bartosz Taudul
f6af913fd3
Remove ability to disable LZ4 compression.
2017-11-10 17:34:11 +01:00
Bartosz Taudul
bc77aa8d26
Source location payload transfer. No storage yet.
2017-11-05 16:46:00 +01:00
Bartosz Taudul
3659afac2a
Move another big buffer out of stack.
2017-11-02 17:37:10 +01:00
Bartosz Taudul
cea7cdd716
Don't create big buffers on stack.
2017-11-02 12:58:38 +01:00
Bartosz Taudul
cf3d8eb810
Manual collection of thread names on android.
2017-10-30 20:45:00 +01:00
Bartosz Taudul
8f280b147b
Getting program name on android.
2017-10-30 14:59:05 +01:00
Bartosz Taudul
a220043114
Add no-cpu GetTime() variant.
...
In this version the address of cpu output variable is const, so there's
no stack address calculation involved.
2017-10-29 16:12:16 +01:00
Bartosz Taudul
03289175ab
Lock counter also must be initialized early.
2017-10-24 22:02:49 +02:00
Bartosz Taudul
ad338a7cfd
Fix message literals.
2017-10-21 12:39:26 +02:00
Bartosz Taudul
1e645665fe
Initialize rpmalloc in profiler worker thread.
...
Thread local variables on gcc are apparently not initialized on thread
startup, but on first access to thread local variables block. Previously
it was working, because s_token was accessed before any rpmalloc
allocation could be performed. Now the first rpmalloc allocation is the
Socket class, and rpmalloc is not initialized there, as there was no
thread local access yet.
2017-10-18 23:30:54 +02:00
Bartosz Taudul
9c4316879c
Add TRACY_NO_EXIT macro.
2017-10-18 20:01:12 +02:00
Bartosz Taudul
51013dc0e6
Manual allocation of socket memory.
2017-10-18 19:50:28 +02:00
Bartosz Taudul
fc94378e0c
Move TracyAlloc.hpp to common. Use rpmalloc only if TRACY_ENABLE.
2017-10-18 19:50:28 +02:00
Bartosz Taudul
c5ea9c744c
Do not disable lz4 in debug builds.
2017-10-18 19:50:28 +02:00
Bartosz Taudul
6a2cbe2842
Rename DISABLE_LZ4 to TRACY_DISABLE_LZ4.
2017-10-18 19:50:22 +02:00
Bartosz Taudul
7c47edc64f
Terminate connection handshake.
2017-10-18 18:48:51 +02:00
Bartosz Taudul
d942b7edf1
Don't exit until all data is sent.
2017-10-17 22:02:47 +02:00
Bartosz Taudul
652dccd163
Also no need to construct more than one welcome message.
2017-10-17 21:55:40 +02:00
Bartosz Taudul
5421164f33
No need to get process name more than once.
2017-10-17 21:53:09 +02:00
Bartosz Taudul
1e3476cf36
Transfer profiler initialization time.
2017-10-17 01:10:38 +02:00
Bartosz Taudul
51f5ae4796
More precise profiler init end time measurement.
2017-10-17 01:07:54 +02:00
Bartosz Taudul
0ed789825a
Measure time of initialization start.
2017-10-17 01:07:34 +02:00
Bartosz Taudul
866081bf29
Initialize tracy before anything else.
2017-10-17 00:36:15 +02:00
Bartosz Taudul
9d01b508ed
One more type cast.
2017-10-17 00:25:32 +02:00
Bartosz Taudul
8a6e4d2971
Change TRACY_DISABLE to TRACY_ENABLE.
...
By default tracy is now disabled.
2017-10-16 21:34:39 +02:00
Bartosz Taudul
e04bd05606
Always use ShouldExit() to determine if worker should exit.
2017-10-16 21:21:42 +02:00
Bartosz Taudul
9f28205548
Use custom threading wrapper instead of std::thread.
...
std::thread may perform memory allocation when a thread is created (it
does so on MSVC). Tracy heap is managed by its own allocator and this
changes prevents accessing application heap.
2017-10-16 21:17:58 +02:00
Bartosz Taudul
31fc2335dd
Silence some type mismatch warnings.
2017-10-16 21:17:58 +02:00
Bartosz Taudul
5b9fcddfb3
String literal message transfer.
2017-10-15 13:06:49 +02:00
Bartosz Taudul
9a60c3fb6e
Remove unused variable.
2017-10-14 20:03:55 +02:00
Bartosz Taudul
472b5a521f
Preallocation size is in number of elements, not bytes.
2017-10-14 17:33:05 +02:00
Bartosz Taudul
c497966c7f
Use rpmalloc to allocate tracy client memory.
2017-10-14 17:15:18 +02:00
Bartosz Taudul
e8968efea7
Initialize rpmalloc.
2017-10-14 17:00:25 +02:00
Bartosz Taudul
fa8030009f
Store messages.
2017-10-14 14:28:04 +02:00
Bartosz Taudul
ec789d60e8
Store source location color in 24 bits.
2017-10-13 19:59:18 +02:00
Bartosz Taudul
fe0366c792
Receive plot data.
2017-10-13 03:36:59 +02:00
Bartosz Taudul
e23da05a65
Workaround gcc stupidity.
2017-10-11 01:44:35 +02:00
Bartosz Taudul
77dfefb5d0
Remove one stack address load.
2017-10-11 01:27:22 +02:00
Bartosz Taudul
af3773dc9a
Remove one level of indirection.
2017-10-11 01:04:21 +02:00
Bartosz Taudul
cc8b357f09
Avoid excessive stack operations for cpu query.
2017-10-10 23:21:30 +02:00
Bartosz Taudul
75457c1465
Remove +x flag from files.
2017-10-10 21:56:15 +02:00
Bartosz Taudul
2c252226fc
Force proper initialization order on gcc.
2017-10-09 00:39:12 +02:00
Bartosz Taudul
f8e7f7ed83
Cygwin can't determine process name using winapi.
2017-10-04 01:22:22 +02:00
Bartosz Taudul
3f0bd793fd
Send program start time, not connection time.
2017-10-04 00:34:05 +02:00
Bartosz Taudul
b2252de9c8
Send and display program execution date.
2017-10-03 23:26:41 +02:00
Bartosz Taudul
cf07383db8
Send program name in welcome message.
2017-10-03 23:17:58 +02:00
Bartosz Taudul
6485457518
Process name getter.
2017-10-03 23:17:16 +02:00
Bartosz Taudul
d1edd30ca6
Zone ids are unnecessary.
2017-10-03 16:41:32 +02:00
Bartosz Taudul
2fb4c47491
Remember to calibrate timer.
2017-10-03 15:35:43 +02:00
Bartosz Taudul
ba037e5798
Do not store tail index in memory.
2017-10-03 14:50:55 +02:00
Bartosz Taudul
353fda95a3
Expose profiler internals to make it easier for inlining.
...
concurrentqueue.h doesn't bring any poisonous includes, only STL.
2017-10-03 14:22:49 +02:00
Bartosz Taudul
a1abf1f015
Record CPU id.
2017-10-01 19:17:08 +02:00
Bartosz Taudul
f46781808c
Construct queue items directly in queue memory.
2017-10-01 17:49:45 +02:00
Bartosz Taudul
99b8c4c77e
Prevent fake loop from optimizing out.
2017-10-01 17:42:22 +02:00
Bartosz Taudul
efda50acb1
Send timer resolution to server.
2017-09-29 18:32:07 +02:00
Bartosz Taudul
6a2cb2c14e
Calculate timer resolution.
2017-09-29 18:29:39 +02:00
Bartosz Taudul
b9aa10913a
Rename internal enum to avoid #define conflicts.
2017-09-28 21:20:33 +02:00
Bartosz Taudul
8c1c395cec
Allow sending custom zone names.
2017-09-28 19:28:24 +02:00
Bartosz Taudul
d1bbb731fc
Zone text (custom string) transfer.
2017-09-27 02:18:17 +02:00
Bartosz Taudul
3c0ce01954
Simplify access to queue producer token.
...
Note that calibration loop needs separate token, as the thread_local
instance is created after the profiler (and its calibration loop).
2017-09-27 01:03:29 +02:00
Bartosz Taudul
842721a754
Make profiler instance static.
2017-09-27 01:03:01 +02:00
Bartosz Taudul
3cc7cc596e
Remove GetNewId() from Profiler interface.
2017-09-27 01:02:04 +02:00
Bartosz Taudul
f584bf76e8
Profiler ID can be static (one less instruction).
2017-09-27 00:30:02 +02:00
Bartosz Taudul
e076d1d475
Send source location answer in stream, not as separate packet.
2017-09-26 19:00:25 +02:00
Bartosz Taudul
e90a86e06e
Store zone color in source location struct.
2017-09-26 18:54:48 +02:00
Bartosz Taudul
7424077d70
Store source location in a single object.
...
Source file, function name and line number are now stored in a const
static container object. This has the following benefits:
- Slightly lighter profiling workload (3 instructions less).
- Profiling queue event size is significantly reduced, by 12 bytes. This
has an effect on all queue event types.
- Source location grouping has now no cost, as it's performed at the
compilation stage. This allows simplification of server code.
The downside is that the full source location resolution is now
performed in two steps, as the server has to query both source location
container and strings contained within. This has almost no real impact
on profiler operation.
2017-09-26 02:39:08 +02:00