The reason I want to use a JIT compiler for running embedded Django applications is this:
"More recently, developers began pioneering a new field in dynamic compilation called trace compiling, or tracing. During compilation into the intermediate language, the compiler can mark the branches within the code, for example, a label in front of a return statement, a jmp operation at a break statement, or call or return. The reasoning is simple, if you strip away all of the jumps and function calls, no matter how long the program is, as long as its written by a human, then from an engineer's perspective, the program will always run in negligible time. Where programs begins to eat up memory however, are the various loops, recursive calls, and whatnot. Essentially, the introduction of branches makes the complexity of the program unpredictable, hence by optimizing the frequently used loops and function calls, we can effectively balance the efficiency of the program with the resources required to compile the code (which is still not trivial by any means). It also solves another major problem. Whenever the program counter (or instruction pointer) hits one of the branch markers, its entire execution flow is traced and stored (and will eventually be compiled if that route is taken frequently enough). Because the program flow of a string concatenation differs from those of an integer addition or even an overloaded function, each of these flow generates unique traces with their unique signatures (the input). For this very reason, if we encounter a string concatenation before an integer addition, the program will not only be able to identify the integer case as a completely different route, but will also be able to break down the complexity of the environment so that further optimizations during the compilation of the integer route can safely assume that its inputs are strictly integers. Neat huh?" [1]
https://gist.github.com/leegao/1073233
--
Hi,
So what is your motivation to even try to make Python(and thus Django to convert it to C code (or run it on top of something like LLVM)?
Is there some part which is too slow in your application? Are you even sure that doing things that Django does would even be faster when compiling it to something like C-code?
In my projects I have few really complex calculation systems and I have resolved them by writing the slow part in C(++) and using C-types to bind C code to my Python (Django) app.
If you're just seeking generic speedup, there exists PyPy project [1] which can run Django and relevant database backends at least successfully. PyPy is JIT for python, works quite well without any modifications to Python code.
PyPy can run Django library just fine, I have seen reports about 15 times faster execution in most operations. Never used it myself though since I haven't really found out Python be too slow in my apps.
[1] http://pypy.org/
On Sun, Jan 28, 2018 at 10:53 AM, Etienne Robillard <tkadm30@yandex.com> wrote:
Hi Jani,
I don't want to optimize my code manually except with machine-compiled instructions.
Cheers,
Etienne
Le 2018-01-28 à 03:12, Jani Tiainen a écrit :
To view this discussion on the web visit https://groups.google.com/d/Hi
Why you even want to add such a complexity?
Have you measured what part in your apps are really slow?
Have you tried to optimize your python code before trying such an extreme actions?--
27.1.2018 9.04 ip. "Etienne Robillard" <tkadm30@yandex.com> kirjoitti:
Hi,
I would like to know if it's possible to compile Django code with a JIT compiler to a C file using LLVM?
What do you think?
Etienne
Etienne Robillard
tkadm30@yandex.com
https://www.isotopesoftware.ca/
--
You received this message because you are subscribed to the Google Groups "Django users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to django-users+unsubscribe@googlegroups.com .
To post to this group, send email to django-users@googlegroups.com.
Visit this group at https://groups.google.com/group/django-users .
To view this discussion on the web visit https://groups.google.com/d/msgid/django-users/ec8bf736-c316 .-8e9a-01ec-16383c78e746%40yand ex.com
For more options, visit https://groups.google.com/d/optout .
You received this message because you are subscribed to the Google Groups "Django users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to django-users+unsubscribe@googlegroups.com .
To post to this group, send email to django-users@googlegroups.com.
Visit this group at https://groups.google.com/group/django-users .
msgid/django-users/ .CAHn91ofbbm2uNyWmJ% 3DH9H8eWMmFEpOPNnN3JkJy1UWmcw6 kACQ%40mail.gmail.com
For more options, visit https://groups.google.com/d/optout .
-- Etienne Robillard tkadm30@yandex.com https://www.isotopesoftware.ca/
--
Jani Tiainen
- Well planned is half done, and a half done has been sufficient before...
You received this message because you are subscribed to the Google Groups "Django users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to django-users+unsubscribe@googlegroups.com.
To post to this group, send email to django-users@googlegroups.com.
Visit this group at https://groups.google.com/group/django-users.
To view this discussion on the web visit https://groups.google.com/d/msgid/django-users/CAHn91odnLhi9JQFzGD%2BJqLP26s0Eh1J4iJAh3mJYtiCvSDgQoQ%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.
-- Etienne Robillard tkadm30@yandex.com https://www.isotopesoftware.ca/
No comments:
Post a Comment