Profiling part of a program's execution - c++

I have a complex application that executes in a number of phases. I would like to profile only one of the phases.
The C++ application runs on Linux, x86-64.
This program takes several minutes to run. If I use perf to profile the whole thing, the resulting data set is too large for perf report to process. However, at this point I am interested only in profiling the execution of one phase of the program that takes maybe 1/3 of the total time. Perhaps this data set will be easier for perf to report on.
Ideally, I'd like something along the lines of "send yourself SIGUSR1 to start profiling, and SIGUSR2 to stop it". At that point I can easily delineate the execution phase that I want profile information for.
I can always write my own (albeit basic) profiler using SIGPROF, but is there a way I can do this with existing tools such as perf?

A possible way to do this is to attach perf to an existing process.
So, start your program, check out its pid. Then start profiling with the -p <pid> option, when it's appropriate. And use CTRL-C or SIGINT to stop the profiling. But this trick works only if you don't need to start/stop profiling a lot of times, as the data append functionality has been removed from perf long time ago.
Or maybe you can just decrease sampling frequency with -F, so the resulting data becomes more tractable.

Related

How do I measure GPU time on Metal?

I want to see programmatically how much GPU time a part of my application consumes on macOS and iOS. On OpenGL and D3D I can use GPU timer query objects. I searched and couldn't find anything similar for Metal. How do I measure GPU time on Metal without using Instruments etc. I'm using Objective-C.
There are a couple of problems with this method:
1) You really want to know what is the GPU side latency within a command buffer most of the time, not round trip to CPU. This is better measured as the time difference between running 20 instances of the shader and 10 instances of the shader. However, that approach can add noise since the error is the sum of the errors associated with the two measurements.
2) Waiting for completion causes the GPU to clock down when it stops executing. When it starts back up again, the clock is in a low power state and may take quite a while to come up again, skewing your results. This can be a serious problem and may understate your performance in benchmark vs. actual by a factor of two or more.
3) if you start the clock on scheduled and stop on completed, but the GPU is busy running other work, then your elapsed time includes time spent on the other workload. If the GPU is not busy, then you get the clock down problems described in (2).
This problem is considerably harder to do right than most benchmarking cases I've worked with, and I have done a lot of performance measurement.
The best way to measure these things is to use on device performance monitor counters, as it is a direct measure of what is going on, using the machine's own notion of time. I favor ones that report cycles over wall clock time because that tends to weed out clock slewing, but there is not universal agreement about that. (Not all parts of the hardware run at the same frequency, etc.) I would look to the developer tools for methods to measure based on PMCs and if you don't find them, ask for them.
You can add scheduled and completed handler blocks to a command buffer. You can take timestamps in each and compare. There's some latency, since the blocks are executed on the CPU, but it should get you close.
With Metal 2.1, Metal now provides "events", which are more like fences in other APIs. (The name MTLFence was already used for synchronizing shared heap stuff.) In particular, with MTLSharedEvent, you can encode commands to modify the event's value at particular points in the command buffer(s). Then, you can either way for the event to have that value or ask for a block to be executed asynchronously when the event reaches a target value.
That still has problems with latency, etc. (as Ian Ollmann described), but is more fine grained than command buffer scheduling and completion. In particular, as Klaas mentions in a comment, a command buffer being scheduled does not indicate that it has started executing. You could put commands to set an event's value at the beginning and (with a different value) at the end of a sequence of commands, and those would only notify at actual execution time.
Finally, on iOS 10.3+ but not macOS, MTLCommandBuffer has two properties, GPUStartTime and GPUEndTime, with which you can determine how much time a command buffer took to execute on the GPU. This should not be subject to latency in the same way as the other techniques.
As an addition to Ken's comment above, GPUStartTime and GPUEndTime is now available on macOS too (10.15+):
https://developer.apple.com/documentation/metal/mtlcommandbuffer/1639926-gpuendtime?language=objc

How to use .. QNX Momentics Application Profiler?

I'd like to profile my (multi-threaded) application in terms of timing. Certain threads are supposed to be re-activated frequently, i.e. a thread executes its main job once every fixed time interval. In other words, there's a fixed time slice in which all the threads a getting re-activated.
More precisely, I expect certain threads to get activated every 2ms (since this is the cycle period). I made some simplified measurements which confirmed the 2ms to be indeed effective.
For the purpose of profiling my app more accurately it seemed suitable to use Momentics' tool "Application Profiler".
However when I do so, I fail to interpret the timing figures that I selected. I would be interested in the average as well in the min and max time it takes before a certain thread is re-activated. So far it seems, the idea is to be only able to monitor the times certain functions occupy. However, even that does not really seem to be the case. E.g. I've got 2 lines of code that are put literally next to each other:
if (var1 && var2 && var3) var5=1; takes 1ms (avg)
if (var4) var5=0; takes 5ms (avg)
What is that supposed to tell me?
Another thing confuses me - the parent thread "takes" up 33ms on avg, 2ms on max and 1ms on min. Aside the fact that the avg shouldn't be bigger than max (i.e. even more I expect avg to be not bigger than 2ms - since this is the cycle time), it's actually increasing the longer I run the the profiling tool. So, if I would run the tool for half an hour the 33ms would actually be something like 120s. So, it seems that avg is actually the total amount of time the thread occupies the CPU.
If that is the case, I would assume to be able to offset against the total time using the count figure which doesn't work either. Mostly due to the figure being almost never available - i.e. there is only as a separate list entry (for every parent thread) called which does not represent a specific process scope.
So, I read QNX community wiki about the "Application Profiler", incl. the manual about "New IDE Application Profiler Enhancements", as well as the official manual articles about how to use the profiler tool.. but I couldn't figure out how I would use the tool to serve my interest.
Bottom line: I'm pretty sure I'm misinterpreting and misusing the tool for what it was intended to be used. Thus my question - how would I interpret the numbers or use the tool's feedback properly to get my 2ms cycle time confirmed?
Additional information
CPU: single core
QNX SDP 6.5 / Momentics 4.7.0
Profiling Method: Sampling and Call Count Instrumentation
Profiling Scope: Single Application
I enabled "Build for Profiling (Sampling and Call Count Instrumentation)" in the Build Options1
The System Profiler should give you what you are looking for. It hooks into the micro kernel and lets you see the state of all threads on the system. I used it in a similar setup to find out what our system was getting unexpected time-outs. (The cause turned out to be Page Waits on critical threads.)

C++: How to set a timeout (not reading input, not threaded)?

Got a large C++ function in Linux that calls a whole lot of other functions, making up an algorithm. At various points given certain bad inputs, the algorithm can get "stuck" and go on forever. Adding a timeout seems appropriate as all potential "stuck" points cannot be predicted. But despite scouring the Internet for timeout examples I've only found how to apply timeouts when either the thing your timing is a separate thread or it's reading inputs. My code is a single thread and does not modify file descriptors, so not coming up with any luck. Do I basically have no choice but to thread it?
I am not sure about the situation, actually server applications or embedded applications often run for years in background without stopping. I think one option is to let your program run in background and log to a file(or screen) timely, and, if you really want to stop the program after certain time, you can use timeout command or a script to kill your program after that time, say, timeout 15s your-prog.

How to diagnose / profile momentary performance hit, C++

Solved: For when simple profiling isn't effective enough, I have written a tool to show me where performance hits occur. Basic information about how the tool works is in the accepted answer below. The source can be found here: http://pastebin.com/ETiW8hE8 (be sure to turn debugging symbols on in the program you're testing)
I've built a game engine in C++ and I have noticed in one particular area of a level that there is a brief performance hit. The game will stop completely for about half a second, and then continue on merrily. I've tried to profile this, but it's difficult isolate the condition since I also have to load the map and perform the in-game task which causes the performance hit. I can make a map load automatically and skip showing menus, etc, and comparing those profile results against a set of similar control data (all the same steps but without actually initiating the performance hit), but it doesn't show anything obvious.
I'm using gmon to profile.
This is a large application with many, many classes and functions. The performance hit only happens once, so there's no way to just trigger the problem many times during one execution to saturate my profiling results in order to make the offending functions more obvious in the profiling results.
What else can I do?
What I would do is try to grab a stack sample in that half second when it's frozen.
This would require an alarm clock timer set to go off some small time in the future, like 100ms.
Then in some loop, like the frame display loop, that normally takes less than 100ms to repeat, keep resetting the timer.
That way, it will act as a watchdog that barks if you don't keep petting it.
Then, stick a breakpoint in the timer interrupt handler.
When it gets there, you know you're in the bad slice of time.
Then just display the call stack, and it should show you what the problem is.
You might have to repeat the process a few times.
You are not saying anything about whether your application is threaded, but I will assume that it is not.
As per suggestion from mike, get insights by getting a stack trace at and see where it is freezing, you can do that with a bit of luck using pstack, so
while usleep 100000; do
pstack processid
done >/tmp/stack.log
Should give you some output to go on -- my guess is that you are calling a blocking IO operation, like reading some assets from disk.

C++ executing a bash script which terminates and restarts the current process

So here is the situation, we have a C++ datafeed client program which we run ~30 instances of with different parameters, and there are 3 scripts written to run/stop them: start.sh stop.sh and restart.sh (which runs stop.sh and then start.sh).
When there is a high volume of data the client "falls behind" real time. We test this by comparing the system time to the most recent data entry times listed. If any of the clients falls behind more than 10 minutes or so, I want to call the restart script to start all the binaries fresh so our data is as close to real time as possible.
Normally I call a script using System(script.sh), however the restart script looks up and kills the process using kill, BUT calling System() also makes the current program execution ignore SIGQUIT and SIGINT until system() returns.
On top of this if there are two concurrent executions with the same arguments they will conflict and the program will hang (this stems from establishing database connections), so I can not start the new instance until the old one is killed and I can not kill the current one if it ignores SIGQUIT.
Is there any way around this? The current state of the binary and missing some data does not matter at all if it has reached the threshold, I also can not just have the program restart itself, since if one of the instances falls behind, we want to restart all 30 of the instances (so gaps in the data are at uniform times). Is there a clean way to call a script from within C++ which hands over control and allows the script to restart the program from scratch?
FYI we are running on CentOS 6.3
Use exec() instead of system(). It will replace your process with the new one. Note there is a significant different in how exec() is called and how it behaves: system() passes its string argument to the system shell to run. exec() actually executes an executable file, and you need to supply the arguments to the process one at a time, instead of letting the shell parse them apart for you.
Here's my two cents.
Temporary solution: Use SIGKILL.
Long-term solution: Optimize your code or the general logic of your service tree, using other system calls like exec or by rewritting it to use threads.
If you want better answers maybe you should post some code and or degeneralize the issue.