Add Python wrap to `CallProfiler` #315

q40603 · 2024-04-20T17:09:49Z

Following the issue : #302
This PR includes :

Wrap CallProfiler to python
Use decorator to wrap the python function with call_profiler_probe to get the profiling result
Convert RadixTree to python dict so user can better access the profiling result.

yungyuc · 2024-04-21T00:31:53Z

Thanks, @q40603 . Before the PR should be reviewed, please make all CI checks passing. After that, please annotate (using PR in-line comments) and make a global comment stating the readiness of review.

If there are issues resolving the CI failures, we may discuss here or on discord.

q40603 · 2024-04-21T02:35:33Z

cpp/modmesh/toggle/RadixTree.hpp

@@ -51,8 +51,8 @@ class RadixTreeNode
    static_assert(std::is_integral_v<key_type> && std::is_signed_v<key_type>, "signed integral required");

    RadixTreeNode(std::string const & name, key_type key)
-        : m_name(name)
-        , m_key(key)
+        : m_key(key)


fix Wreorder-ctor error

q40603

@yungyuc THE PR is ready for review.

q40603 · 2024-04-21T02:35:49Z

cpp/modmesh/toggle/RadixTree.hpp

@@ -225,17 +225,16 @@ class CallProfiler
    ~CallProfiler() = default;

    // Called when a function starts
-    void start_caller(const std::string & caller_name, std::function<void()> cancel_callback)


remove unused arg

Please don't mark conversation as resolved. If you want to say it is resolved, leave a comment saying so.

Marking conversation as resolved makes it hard to keep track of discussion history.

q40603 · 2024-04-21T02:36:26Z

cpp/modmesh/toggle/RadixTree.hpp

@@ -265,14 +264,10 @@ class CallProfilerProbe
 {
 public:
    CallProfilerProbe(CallProfiler & profiler, const char * caller_name)
-        : m_profiler(profiler)
-        , m_caller_name(caller_name)
+        : m_caller_name(caller_name)


fix Wreorder-ctor error

q40603 · 2024-04-21T02:39:33Z

cpp/modmesh/toggle/RadixTree.hpp

        m_radix_tree.entry(caller_name);
        CallerProfile & callProfile = m_radix_tree.get_current_node()->data();
        callProfile.caller_name = caller_name;
        callProfile.start_stopwatch();
    }

    // Called when a function ends
-    void end_caller(const std::string & caller_name)


remove unused arg

q40603 · 2024-04-21T02:40:58Z

cpp/modmesh/toggle/pymod/wrap_profile.cpp

@@ -163,12 +163,43 @@ class MODMESH_PYTHON_WRAPPER_VISIBILITY WrapTimeRegistry

 }; /* end class WrapTimeRegistry */

+class WrapCallProfiler : public WrapBase<WrapCallProfiler, CallProfiler>


Expose CallProfiler to python so profiling util can be shared.
RadixTree is internal data structure, hence no need to expose this to python

q40603 · 2024-04-21T02:41:49Z

tests/test_callprofiler.py

+        foo1()
+        result = modmesh.call_profiler.print_profiling_result()
+        pattern = r"([-+]?\d*\.\d+|\d+) ms"
+        match = re.search(pattern, result)


Not sure if this is a good practice to extract value from the string and assert the value

No it's not. Please add relevant data accessors in CallProfiler for the profiling data.

By writing tests, we will discover many things for improving the API (system) omitted when during the initial design. This "design-for-testing" is a powerful techniques for software development too, although it is not exactly the same as how it is used in IC design.

q40603 · 2024-04-21T02:43:16Z

tests/test_callprofiler.py

+        self.assertNotIn("foo1", result)
+
+    @unittest.skipUnless("nt" != os.name,
+                         "timing code on windows has resoltion issue")


Notice there's time resolution issue in windows, causing the CI to fail as the measured time is always less than what it should be.
Wonder if there's a better way to handle this on windows.

Much much less it is in windows than in other systems?

Preferably this can be fixed (or worked around) in this PR. We use the skipUnless trick if can't do it soon enough.

can we have a workaround for Windows? e.g. increasing the time or enlarging the error tolerance?

yungyuc

Remove functool comment in the test code.
Add comment for the future plan of profile_function.
Use a little bit of time to resolve the windows timing issue for testing.
Implement basic data access in CallProfiler for testing.

Again, please add in-line annotation to mark the points that you address or would like to discuss, and when everything is ready for review, leave a global comment. Combining the in-line comments and the global using the code review section/feature/whatever may be a good idea. Let's use this protocol from now on.

yungyuc · 2024-04-21T08:15:09Z

tests/test_callprofiler.py

+import modmesh
+
+
+# https://docs.python.org/3/library/functools.html#functools.wraps


Please remove the comment pointing to Python document. This is a simple helper in Python standard library that every Python developer should be familiar with, so I expect all contributors to modmesh to be familiar with it. We should not add comment for people who do not know how to learn easy things themselves.

On the other hand, if there are tricky interaction among popular API, or obscure bug, we usually add comments pointing to bug trackers.

yungyuc · 2024-04-21T08:19:19Z

tests/test_callprofiler.py

+
+
+# https://docs.python.org/3/library/functools.html#functools.wraps
+def profile_function(func):


This will be a helpful utility function for developing modmesh code, e.g., GUI (FYI @j8xixo12). Eventually I think a more tailored implementation should be made available somewhere under the modmesh Python package.

At the time being, could you please add a comment for the above-mentioned argument?

When it comes to the to-be-useful-and-mature helper function in the future, I don't like the function name (profile_function). But it is OK for a local helper in the test module.

I am not seeing the descriptions in comment. Please point it out.

This will be a helpful utility function for developing modmesh code, e.g., GUI (FYI @j8xixo12). Eventually I think a more tailored implementation should be made available somewhere under the modmesh Python package.

At the time being, could you please add a comment for the above-mentioned argument?

By "add a comment for ...", I meant to add the comment in the code, so that a future maintainer will read it and perform the refactoring.

yungyuc · 2024-04-21T08:23:47Z

tests/test_callprofiler.py

+        self.assertNotIn("foo1", result)
+
+    @unittest.skipUnless("nt" != os.name,
+                         "timing code on windows has resoltion issue")


Much much less it is in windows than in other systems?

Preferably this can be fixed (or worked around) in this PR. We use the skipUnless trick if can't do it soon enough.

yungyuc · 2024-04-21T08:25:03Z

tests/test_callprofiler.py

+        foo1()
+        result = modmesh.call_profiler.print_profiling_result()
+        pattern = r"([-+]?\d*\.\d+|\d+) ms"
+        match = re.search(pattern, result)


No it's not. Please add relevant data accessors in CallProfiler for the profiling data.

By writing tests, we will discover many things for improving the API (system) omitted when during the initial design. This "design-for-testing" is a powerful techniques for software development too, although it is not exactly the same as how it is used in IC design.

yungyuc · 2024-04-21T08:28:07Z

tests/test_callprofiler.py

+
+    @unittest.skipUnless("nt" != os.name,
+                         "timing code on windows has resoltion issue")
+    def test_single_fun_profile(self):


I suggest to use func as a shorthand for function, but it's just my personal preference. I kind of feel fun is for fun.

You don't have to change if you don't feel it the same way.

func makes more sense. If possible, I prefer spelling out the full name.

tigercosmos · 2024-04-21T11:26:40Z

cpp/modmesh/toggle/RadixTree.hpp

-        {
-            cancel();
-        };
-        m_profiler.start_caller(m_caller_name, cancel_callback);


why do you remove the cancel_ballback?

@tigercosmos
Because the compiler raised unused variable error.
Also, I am not sure about the true behavior of this callback.
Could you please elaborate more or maybe kindly add some tests in gtests/test_nopython_callprofiler.cpp?

@q40603 Originally, cancel_callback is for canceling the profiling. For example, if we restart the profiler, we need to cancel the previous profiling, otherwise the profiler will have unexpected behavior. Surprisingly, I didn't fully support the cancel mechanism before. Let me file another PR for this.

please check the PR:
#318

tigercosmos · 2024-04-21T11:29:34Z

cpp/modmesh/toggle/pymod/wrap_profile.cpp

+                { return wrapped_type::instance(); })
+            .def("start_caller", &wrapped_type::start_caller, py::arg("caller_name"))
+            .def("end_caller", &wrapped_type::end_caller)
+            .def("print_profiling_result", [](const CallProfiler & profiler)


maybe consider taking a stream object as the argument.

tigercosmos · 2024-04-21T11:29:59Z

cpp/modmesh/toggle/toggle.hpp

@@ -31,6 +31,7 @@
 #include <modmesh/base.hpp>
 #include <modmesh/buffer/buffer.hpp>
 #include <modmesh/toggle/profile.hpp>
+#include <modmesh/toggle/RadixTree.hpp>


do we need to put the include here?

tigercosmos · 2024-04-21T11:34:57Z

tests/test_callprofiler.py

+        match = re.search(foo3_pattern, result)
+        self.assertIsNotNone(match, result)
+        foo3_total_time = float(match.group(1))
+        self.assertGreater(foo3_total_time, 3000)


the number can be fixed. please refer to test_nopython_radixtree.cpp

sorry I am wrong here, should be test_nopython_callprofiler.cpp

q40603 · 2024-04-27T08:31:55Z

I will try to finalize this PR in 2 days.

@yungyuc @tigercosmos
As cancel callback feature has been implemented by @tigercosmos in PR 318. I found it hard to wrap the call_profiler using pybind11 because of the cancell_callback argument.

Hence, I tried to wrap the CallProfilerProbe instead, but I ran into a issue.

Consider the following 2 code snippets:
with the difference modmesh.CallProfilerProbe is assigning to a variable or not

def CallProfileProbeA(func):
    @wraps(func)
    def wrapper(*args, **kwargs):
        modmesh.CallProfilerProbe(func.__name__)
        result = func(*args, **kwargs)
        return result
    return wrapper

def CallProfileProbeB(func):
    @wraps(func)
    def wrapper(*args, **kwargs):
        tmp = modmesh.CallProfilerProbe(func.__name__)
        result = func(*args, **kwargs)
        return result
    return wrapper

and a simple python code that drives the Profiler

@CallProfileProbeA
def foo1():
    time.sleep(2)

@CallProfileProbeA
def foo2():
    time.sleep(3)
    foo1()


if __name__ == '__main__':
    foo2()
    print(modmesh.call_profiler.print_profiling_result())

The two different implementations gave different result

CallProfileProbeA - unexpected

Profiling Result
  foo2 - Total Time: 0.002378 ms, Call Count: 1
  foo1 - Total Time: 0.018734 ms, Call Count: 1

CallProfileProbeB - expected

Profiling Result
  foo2 - Total Time: 5002.27 ms, Call Count: 1
    foo1 - Total Time: 2001.02 ms, Call Count: 1

For the CallProfileProbeA, the CallProfilerProbe object is deleted almost immediately because of the python garbage collection, reference counting of CallProfilerProbe is zero. (Spent a few hours debugging why CallProfileProbeA doesn't work 😂, also tried to implement CallProfileProbe as context manager, but give up.)

I just wonder if there's a more elegant way for the CallProfilerProbe object to be persisted before the wrapper function return?

yungyuc · 2024-04-28T03:43:48Z

@tigercosmos why did a helper in a profiler need to use callback? Callback is generally hard to understand and maintain. It does not sound like necessary.

Is it possible to implement the similar capability without a callback? Or is there a strong reason to use a callback?

tigercosmos · 2024-04-28T11:34:49Z

@q40603 @yungyuc
I utilized a callback in this context to invoke the cancellation function of the probe, without exposing its entirety. However, if the callback presents an issue, we can modify the logic to pass the probe pointer to CallProfiler, so it can call probe->cancel later.

yungyuc · 2024-04-28T13:20:48Z

Thanks, @q40603 @tigercosmos . Now there seem to be two things:

Avoid using callback in the profiler.
Wrap the profiler to Python while there is the cancellation callback.

Can they be handled independently? If so, let's do it this way.

If not, what is the dependency?

tigercosmos · 2024-04-28T13:30:26Z

If they are dependent, in my opinion, we can refactor the callback to another design, for now maybe we can skip supporting cancel() in Python if it has an issue.

yungyuc · 2024-04-28T13:32:44Z

If they are dependent, in my opinion, we can refactor the callback to another design, for now maybe we can skip supporting cancel() in Python if it has an issue.

Thanks, @tigercosmos . Does that mean that @q40603 can continue the profiler wrapper work by skipping the wrapper of cancel()?

tigercosmos · 2024-04-28T13:53:30Z

@yungyuc yes, I think the cancel function is not the priority.

yungyuc · 2024-04-30T10:27:42Z

@yungyuc yes, I think the cancel function is not the priority.

Thanks for the confirmation. I think I know what you meant now.

@q40603 do you have sufficient information to proceed?

q40603 · 2024-04-30T16:57:47Z

@yungyuc @tigercosmos
Thanks for the comments.
I will skip cancel callback at this moment and implement the CallProfiler data access ASAP, which will need to expose the RadixTree to python as well.

tigercosmos · 2024-05-01T07:31:25Z

btw, can we rename the file to CallProfiler.h rather than RadixTree.h? RadixTree is the internal data structure, and the CallProfiler is the public API. @q40603 @yungyuc

yungyuc · 2024-05-01T08:39:23Z

btw, can we rename the file to CallProfiler.h rather than RadixTree.h? RadixTree is the internal data structure, and the CallProfiler is the public API. @q40603 @yungyuc

@tigercosmos feel free to create a refactoring PR for discussing the proposal. It may conflict with what @q40603 is doing, but discussing with code is still better than only words.

yungyuc · 2024-06-02T08:10:56Z

@q40603 Is there progress on this PR?

q40603 · 2024-06-04T04:05:35Z

I am revising the data accessor of profiling result by making it returned in nested python dictionary format.
The dict can then be used with better assertion.
Some fields are also required to add to distinguish where is profiling probe called, like in python or in c++.

yungyuc · 2024-06-05T12:12:57Z

It could be an issue to use Python dict for storing the data in the long run. It is slow. But we can do it in the short term like in this PR. Performance issues need to be addressed by profiling.

remove unwanted checkin callprofiler driver code clang-format clang-format clang-format RadixTree flake8 floating number skip windows timeing test fix typo fix conflict wrap CallProfilerProbe wrap CallProfilerProbe profliing result in python dict profliing result in python dict add jsonschema assertGreaterEqual add MODMESH_PYTHON_WRAPPER_VISIBILITY busy loop fix flake8 use perf_counter flake8 jsonschema missing jsonschema

q40603

Remove functool comment in the test code.

Add comment for the future plan of profile_function.

Use a little bit of time to resolve the windows timing issue for testing.

Implement basic data access in CallProfiler for testing.

@yungyuc, @tigercosmos The issues are addressed and the code is ready for review.

q40603 · 2024-06-18T00:39:12Z

tests/test_callprofiler.py

+
+def busy_loop(duration):
+    end_time = time.perf_counter() + duration
+    while time.perf_counter() < end_time:


use time.perf_counter() to get higher time resolution instead of time.time()

q40603 · 2024-06-18T00:41:09Z

tests/test_callprofiler.py

+        result = modmesh.call_profiler.result()
+        self.assertEqual(result, {})
+
+    def test_profiler_result_schema(self):


add schema validation using jsoncshema library.

q40603 · 2024-06-18T00:46:40Z

tests/test_callprofiler.py

+        self.assertEqual(second_bar_result["count"], 1)
+        self.assertGreaterEqual(second_bar_result["total_time"], 500)
+
+    def test_get_result_during_profiling(self):


As the previous unit test all assert the result after profiling a series of function calls.
But user should be able to get the profiling result within the function too, so add this test.

Sounds good.

q40603 · 2024-06-18T00:47:45Z

tests/data/profiler_python_schema.json

@@ -0,0 +1,41 @@
+{


Simple json schema that regulates the output python dict format.

not sure if the schema should be put in this path.

we should also mention this schema in the cpp implementation.

q40603 · 2024-06-18T00:50:24Z

cpp/modmesh/toggle/pymod/wrap_profile.cpp

+            .def("result", [](CallProfiler & profiler)
+                 {
+            const RadixTreeNode<CallerProfile> * root = profiler.radix_tree().get_root();
+            if (root->no_children()) {


To avoid root only dict is returned when the radix tree is empty.

yungyuc

We may use some discussions, which can take place here or on the discord.

Review and correct coding style
Missing comments for the future plan of profile_function().

yungyuc · 2024-06-18T12:48:25Z

cpp/modmesh/toggle/RadixTree.hpp

@@ -72,6 +73,11 @@ class RadixTreeNode
    const T & data() const { return m_data; }
    const child_list_type & children() const { return m_children; }

+    bool no_children() const


There are two options for this getter: no_children() or has_children().

no_children() acts like the implementation detail m_children.empty(). If we choose this, what is the reason that we do not name the function as empty_children()? empty is more consistent to the style of STL.

Please consider the options and we should discuss.

I agree that no_chlidren should be renamed to empty_children, no_children is too verbal.
But I am curios, in practical, how to choose between negative check and positive check?
like empty_children or has_children in this case

Time will tell.

yungyuc · 2024-06-18T12:49:55Z

tests/test_callprofiler.py

+# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+# POSSIBILITY OF SUCH DAMAGE.


Need two blank lines separating comment block and code.

cc @ThreeMonth03

yungyuc · 2024-06-18T12:50:54Z

tests/test_callprofiler.py

+
+
+# https://docs.python.org/3/library/functools.html#functools.wraps
+def profile_function(func):


I am not seeing the descriptions in comment. Please point it out.

yungyuc · 2024-06-18T12:52:44Z

tests/test_callprofiler.py

+        self.assertEqual(second_bar_result["count"], 1)
+        self.assertGreaterEqual(second_bar_result["total_time"], 500)
+
+    def test_get_result_during_profiling(self):


Sounds good.

tigercosmos · 2024-06-18T15:10:25Z

tests/test_callprofiler.py

+def busy_loop(duration):
+    end_time = time.perf_counter() + duration
+    while time.perf_counter() < end_time:
+        time.sleep(0.00001)


sleep is usually not precise in a very short time. why not just use pass for the busy loop?

Yes, I think pass would be more straightforward.
Originally, the sleep is just for reducing the CPU usage.

tigercosmos · 2024-06-18T15:11:14Z

tests/test_callprofiler.py

+    return wrapper
+
+
+def busy_loop(duration):


let's have a comment to explain the time unit.

fix coding style

q40603 · 2024-06-19T03:25:22Z

We may use some discussions, which can take place here or on the discord.

Review and correct coding style

Missing comments for the future plan of profile_function().

Issues addressed.
Regarding profile_function(), here is the future plan:

Implement profile_funcion somewhere else under python modmesh lib
Add a filed in each node in the profiler to distinguish where is the profiler run, like from: C++, from: Python
Try profile a real app that has both python and C++ codes, and add a unit test for this scenario.

empty_children

yungyuc

Still miss comments for the future plan of profile_function().

yungyuc · 2024-06-19T09:43:47Z

tests/test_callprofiler.py

+
+
+# https://docs.python.org/3/library/functools.html#functools.wraps
+def profile_function(func):


This will be a helpful utility function for developing modmesh code, e.g., GUI (FYI @j8xixo12). Eventually I think a more tailored implementation should be made available somewhere under the modmesh Python package.

At the time being, could you please add a comment for the above-mentioned argument?

By "add a comment for ...", I meant to add the comment in the code, so that a future maintainer will read it and perform the refactoring.

yungyuc · 2024-06-19T09:46:02Z

@q40603 Could you please find @terrychan999 for handling the failure of creating artifacts in the windows CI run? I think we should use some discussions to deal with it. Please bring it to the discord channel.

q40603 · 2024-06-19T10:06:45Z

Still miss comments for the future plan of profile_function().

Do you mean leaving comments in the code?

yungyuc · 2024-06-19T10:30:04Z

Correct

q40603 · 2024-06-20T08:03:04Z

Comment added.

q40603

@yungyuc The code is ready for next review.

q40603 · 2024-06-21T01:03:51Z

tests/test_callprofiler.py

+import modmesh
+
+
+# TODO


Comment added.

LGTM. Thanks.

yungyuc

Almost there (while I might find more to change later):

Import the module and use . to access the functions in it.

yungyuc · 2024-06-21T09:37:12Z

tests/test_callprofiler.py

+import modmesh
+
+
+# TODO


LGTM. Thanks.

yungyuc · 2024-06-21T09:42:58Z

tests/test_callprofiler.py

+import unittest
+import time
+import json
+from jsonschema import validate, ValidationError


Please import the module and use . to access the functions in it, like:

import jsonschema ... jsonschema.validate(...)

Please review other Python files in the test and code directories and follow the style. @ThreeMonth03 this may be captured in the style guide.

q40603

Almost there (while I might find more to change later):

Import the module and use . to access the functions in it.

Issues addressed, the code is ready for next review.

q40603 · 2024-06-21T14:58:46Z

tests/test_callprofiler.py

+            schema = json.load(schema_file)
+
+        try:
+            jsonschema.validate(instance=result, schema=schema)


Fix module import and use dot operator.

q40603 · 2024-06-21T14:59:11Z

tests/test_callprofiler.py

+# 3. Try profile a real app that has both python and C++ codes
+#    and add a unit test for this scenario.
+def profile_function(func):
+    @functools.wraps(func)


Fix module import and use dot operator.

yungyuc

Thanks a lot. Merging.

q40603 commented Apr 21, 2024

View reviewed changes

yungyuc added the performance Profiling, runtime, and memory consumption label Apr 21, 2024

q40603 commented Apr 21, 2024

View reviewed changes

yungyuc requested changes Apr 21, 2024

View reviewed changes

yungyuc assigned q40603 Apr 21, 2024

tigercosmos reviewed Apr 21, 2024

View reviewed changes

tigercosmos mentioned this pull request Apr 23, 2024

add missing cancel() for CallProfiler #318

Merged

q40603 force-pushed the quentin/wrap_call_profiler branch 2 times, most recently from 40333e1 to 085d34a Compare April 27, 2024 07:20

yungyuc mentioned this pull request Jun 2, 2024

Wrap the C++ profile object (CallProfile) to Python #344

Open

yungyuc changed the title ~~Quentin/wrap call profiler~~ PR: Add Python wrap to CallProfiler Jun 2, 2024

yungyuc changed the title ~~PR: Add Python wrap to CallProfiler~~ Add Python wrap to CallProfiler Jun 2, 2024

q40603 force-pushed the quentin/wrap_call_profiler branch 3 times, most recently from d28d200 to 567dea3 Compare June 17, 2024 17:43

q40603 force-pushed the quentin/wrap_call_profiler branch from 567dea3 to 1f07576 Compare June 17, 2024 17:49

q40603 commented Jun 18, 2024

View reviewed changes

q40603 requested a review from yungyuc June 18, 2024 07:07

yungyuc requested changes Jun 18, 2024

View reviewed changes

tigercosmos reviewed Jun 18, 2024

View reviewed changes

fix coding style

0522d83

fix coding style

q40603 force-pushed the quentin/wrap_call_profiler branch from d7af8db to 0522d83 Compare June 18, 2024 15:45

q40603 added 2 commits June 19, 2024 11:27

Update RadixTree.hpp

f57e85d

empty_children

Update wrap_profile.cpp

9e147ef

yungyuc requested changes Jun 19, 2024

View reviewed changes

add future plan of profile_function

c9fa78d

q40603 added 2 commits June 20, 2024 17:52

update json schema

a7c3d70

update json schema: current_node

5c067fc

q40603 requested a review from yungyuc June 20, 2024 15:40

q40603 commented Jun 21, 2024

View reviewed changes

yungyuc requested changes Jun 21, 2024

View reviewed changes

modules import fix

20d245e

q40603 commented Jun 21, 2024

View reviewed changes

yungyuc approved these changes Jun 22, 2024

View reviewed changes

yungyuc merged commit e2dd7cc into solvcon:master Jun 22, 2024
13 checks passed

		@@ -163,12 +163,43 @@ class MODMESH_PYTHON_WRAPPER_VISIBILITY WrapTimeRegistry

		}; /* end class WrapTimeRegistry */

		class WrapCallProfiler : public WrapBase<WrapCallProfiler, CallProfiler>

		import modmesh


		# https://docs.python.org/3/library/functools.html#functools.wraps



		# https://docs.python.org/3/library/functools.html#functools.wraps
		def profile_function(func):

Add Python wrap to CallProfiler #315

Add Python wrap to CallProfiler #315

Conversation

q40603 commented Apr 20, 2024 • edited Loading

yungyuc commented Apr 21, 2024

Choose a reason for hiding this comment

q40603 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

yungyuc left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

q40603 commented Apr 27, 2024 • edited Loading

yungyuc commented Apr 28, 2024

tigercosmos commented Apr 28, 2024

yungyuc commented Apr 28, 2024

tigercosmos commented Apr 28, 2024

yungyuc commented Apr 28, 2024

tigercosmos commented Apr 28, 2024

yungyuc commented Apr 30, 2024

q40603 commented Apr 30, 2024

tigercosmos commented May 1, 2024

yungyuc commented May 1, 2024

yungyuc commented Jun 2, 2024

q40603 commented Jun 4, 2024

yungyuc commented Jun 5, 2024

q40603 left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

yungyuc left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

q40603 commented Jun 19, 2024

yungyuc left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

yungyuc commented Jun 19, 2024 • edited Loading

q40603 commented Jun 19, 2024 • edited Loading

yungyuc commented Jun 19, 2024

q40603 commented Jun 20, 2024

Add Python wrap to `CallProfiler` #315

Add Python wrap to `CallProfiler` #315

q40603 commented Apr 20, 2024 •

edited

Loading

yungyuc left a comment •

edited

Loading

q40603 commented Apr 27, 2024 •

edited

Loading

q40603 left a comment •

edited

Loading

yungyuc left a comment •

edited

Loading

yungyuc commented Jun 19, 2024 •

edited

Loading

q40603 commented Jun 19, 2024 •

edited

Loading

yungyuc left a comment •

edited

Loading