mirror-github-bspeice/speice.io
1
0
Fork 0
mirror of https://github.com/bspeice/speice.io synced 2024-11-14 22:18:10 -05:00
Code Issues Releases Wiki Activity

Proofreading

Browse Source
This commit is contained in:
Bradlee Speice 2020-06-30 17:53:02 -04:00
parent fb958ac92a
commit 8027538bb0
1 changed files with 10 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
_posts/2020-06-29-release-the-gil-pt.-2.md
View File

@ -92,8 +92,8 @@ PYBIND11_MODULE(speiceio_pybind11, m) {
} }
``` ```
After the code is installed into a `virtualenv` or similar setup, we can use the functions to After building the C++ module, those functions can be used to demonstrate the effect of unlocking
demonstrate GIL unlocking: the GIL.
```python ```python
# The billionth Fibonacci number overflows `std::uint64_t`, but that's OK; # The billionth Fibonacci number overflows `std::uint64_t`, but that's OK;
@ -103,7 +103,7 @@ N = 1_000_000_000;
from speiceio_pybind11 import fibonacci_gil, fibonacci_nogil from speiceio_pybind11 import fibonacci_gil, fibonacci_nogil
``` ```
Even when using two threads, the code is effectively serial: Even though two threads are used, the GIL prevents those threads from running in parallel:
```python ```python
%%time %%time
@ -123,9 +123,11 @@ t1.join(); t2.join()
> Wall time: 705 ms > Wall time: 705 ms
> </pre> > </pre>
The elapsed ("wall") time is effectively the same as the time spent executing on the CPU ("user"). Because the elapsed ("wall") time is effectively the same as the time spent executing on the CPU
("user"), there was no benefit to using multiple threads.
However, if one thread unlocks the GIL first, then the threads will execute in parallel: However, if one thread unlocks the GIL first, the Python interpreter is allowed to execute the
second thread in parallel:
```python ```python
%%time %%time
@ -141,9 +143,10 @@ t1.join(); t2.join()
> Wall time: 372 ms > Wall time: 372 ms
> </pre> > </pre>
The CPU time ("user") hasn't changed, but the elapsed time ("wall") is effectively cut in half. The CPU time ("user") hasn't changed much, but the elapsed time ("wall") is effectively cut in half.
TODO: Note about double-unlocking: Caution is advised though; attempting to unlock the GIL when it isn't locked will terminate the
current process:
```c++ ```c++
void recurse_unlock() { void recurse_unlock() {