Author: Antonio Cuni <[email protected]>
Branch: extradoc
Changeset: r5815:f172adc02a04
Date: 2017-07-12 10:59 +0200
http://bitbucket.org/pypy/extradoc/changeset/f172adc02a04/
Log: add slides about v5
diff --git a/talk/ep2017/the-joy-of-pypy-jit/talk.rst
b/talk/ep2017/the-joy-of-pypy-jit/talk.rst
--- a/talk/ep2017/the-joy-of-pypy-jit/talk.rst
+++ b/talk/ep2017/the-joy-of-pypy-jit/talk.rst
@@ -394,6 +394,74 @@
:scale: 30%
:align: center
+
+Version 5
+----------
+
+|scriptsize|
+
+.. sourcecode:: python
+
+ from pypytools.codegen import Code
+
+ def Kernel(matrix):
+ height = len(matrix)
+ width = len(matrix[0])
+ code = Code()
+ with code.block('def apply(img, p):'):
+ code.w('value = 0.0')
+ for j, row in enumerate(matrix, -(height/2)):
+ for i, k in enumerate(row, -(width/2)):
+ if k == 0:
+ continue
+ code.w('value += img[p+{delta}] * {k}',
+ delta=(i, j), k=k)
+ code.w('return value')
+
+ code.compile()
+ return code['apply']
+
+|end_scriptsize|
+
+Version 5
+----------
+
+|scriptsize|
+
+.. sourcecode:: python
+
+ # GENERATED CODE
+
+ def apply(img, p):
+ value = 0.0
+ value += img[p+(-1, -1)] * -1.0
+ value += img[p+(1, -1)] * 1.0
+ value += img[p+(-1, 0)] * -2.0
+ value += img[p+(1, 0)] * 2.0
+ value += img[p+(-1, 1)] * -1.0
+ value += img[p+(1, 1)] * 1.0
+ return value
+
+|end_scriptsize|
+
+Version 5
+----------
+
+|column1|
+
+.. image:: sobel/CPython-v5.png
+ :scale: 30%
+
+|column2|
+
+.. image:: sobel/PyPy-v5.png
+ :scale: 30%
+
+|end_columns|
+
+* PyPy ~428x faster again
+
+
The cost of abstraction
------------------------
@@ -407,7 +475,7 @@
- abstractions (almost) for free
- - v5 is ~20% slower than v0, v1, v2
+ - v5 is ~18% slower than v0, v1, v2
PyPy JIT 101
@@ -417,7 +485,15 @@
* Which code is optimized away?
+* Very rough explanation
+
+* For a deeper view:
+ - http://speakerdeck.com/u/antocuni/p/pypy-jit-under-the-hood
+
+ - http://www.youtube.com/watch?v=cMtBUvORCfU
+
+
Loops and guards
-----------------
@@ -677,3 +753,4 @@
* Or, just talk to us :)
+* @pypyproject, @antocuni
_______________________________________________
pypy-commit mailing list
[email protected]
https://mail.python.org/mailman/listinfo/pypy-commit
