Author: Ronan Lamy <ronan.l...@gmail.com>
Branch: extradoc
Changeset: r5404:d32cb87fb33e
Date: 2014-09-18 05:04 +0100
http://bitbucket.org/pypy/extradoc/changeset/d32cb87fb33e/
Log: progress
diff --git a/talk/pycon-uk-2014/Makefile b/talk/pycon-uk-2014/Makefile
--- a/talk/pycon-uk-2014/Makefile
+++ b/talk/pycon-uk-2014/Makefile
@@ -1,6 +1,7 @@
talk.pdf: talk.rst author.latex stylesheet.latex
- rst2beamer --stylesheet=stylesheet.latex --documentoptions=14pt
--overlaybullets=False talk.rst talk.latex || exit
+ rst2beamer --stylesheet=stylesheet.latex --overlaybullets=False
--codeblocks-use-pygments --documentoptions=14pt talk.rst talk.latex || exit
sed 's/\\date{}/\\input{author.latex}/' -i talk.latex || exit
+ sed 's/XXXFAKETITLE//' -i talk.latex || exit
#sed 's/\\maketitle/\\input{title.latex}/' -i talk.latex || exit
pdflatex talk.latex || exit
diff --git a/talk/pycon-uk-2014/talk.rst b/talk/pycon-uk-2014/talk.rst
--- a/talk/pycon-uk-2014/talk.rst
+++ b/talk/pycon-uk-2014/talk.rst
@@ -1,7 +1,7 @@
.. include:: beamerdefs.txt
===========================
-PyPy and its ecosystem
+XXXFAKETITLE
===========================
About me
@@ -43,19 +43,21 @@
- **PyPy**
+|pause|
+
- HippyVM: implementing PHP
* ~7x faster than standard PHP
* http://hippyvm.com/
+|pause|
+
- Topaz: implementing Ruby
- * most of the language implemented
+ * https://github.com/topazproject/topaz
- * "definitely faster than MRI"
-
- * https://github.com/topazproject/topaz
+|pause|
- Pyrolog (Prolog)
@@ -122,6 +124,8 @@
* Trace one code path through the loop
* Compile (magic!)
+
+ - with guards to ensure correctness
* Profit!
@@ -131,7 +135,7 @@
|scriptsize|
-.. code:: python
+.. sourcecode:: python
@wrap(['space', str, W_Root, Optional(int)])
def strpos(space, haystack, w_needle, offset=0):
@@ -156,6 +160,62 @@
|end_scriptsize|
+
+JITting strpos()
+----------------
+
+|example<| strpos_jit.php |>|
+
+|scriptsize|
+
+.. sourcecode:: php
+
+ <?php
+ function compute($n) {
+ $sum = 0;
+ for ($i=0; $i < 100000; $i++) {
+ $sum += strpos($i, '1');
+ }
+ return $sum;
+ }
+
+ echo compute(100000) . "\n";
+ ?>
+
+|end_scriptsize|
+
+|end_example|
+
+JITting strpos()
+----------------
+
+|scriptsize|
+
+.. parsed-literal::
+
+ label(p0, p1, p3, p7, i48, i52, p33, p26, descr=TargetToken(140099127632144))
+ i57 = int_eq(i52, 100000)
+ guard_false(i57, descr=<Guard0x7f6b5ebd5a60>) [p1, p0, i52, p3, p7, i48]
+ i58 = int_lt(i52, 100000)
+ guard_true(i58, descr=<Guard0x7f6b5ebd59f0>) [p1, p0, p3, p7, i52, i48]
+ guard_not_invalidated(descr=<Guard0x7f6b5ebd5980>) [p1, p0, p3, p7, p33,
i52, i48]
+ **p59 = call(ConstClass(ll_str__IntegerR_SignedConst_Signed), i52,
descr=<Callr 8 i EF=3>)**
+ guard_no_exception(descr=<Guard0x7f6b5ebd5910>) [p1, p0, p59, p3, p7, p33,
i52, i48]
+ i60 = strlen(p59)
+ **i61 = call(ConstClass(ll_find_char__rpy_stringPtr_Char_Signed_Signed),
p59, 49, 0, i60, descr=<Calli 8 riii EF=0>)**
+ i62 = int_eq(i61, -1)
+ guard_false(i62, descr=<Guard0x7f6b5ebd58a0>) [p1, p0, i61, p3, p7, p33,
i52, i48]
+ i63 = int_add_ovf(i48, i61)
+ guard_no_overflow(descr=<Guard0x7f6b5ebd5830>) [p1, p0, i63, i61, i48, p3,
p7, p33, i52, None]
+ i64 = int_add(i52, 1)
+ i65 = getfield_raw(24210432, descr=<FieldS pypysig_long_struct.c_value 0>)
+ i66 = int_lt(i65, 0)
+ guard_false(i66, descr=<Guard0x7f6b5ebd57c0>) [p1, p0, p3, p7, i63, i64,
None, None, None]
+ i67 = arraylen_gc(p26, descr=<ArrayU 1>)
+ jump(p0, p1, p3, p7, i63, i64, p33, p26, descr=TargetToken(140099127632144))
+
+|end_scriptsize|
+
PyPy: past two years (1)
-----------------------------
@@ -209,21 +269,13 @@
- Python code: "it just works"
-- C code: better than ever!
+- 2.7: very stable
- * cpyext: more complete, but still slow
+- 3.2: stable
- * CFFI: the future
+ * Some missing optimizations
- * Native PyPy C API for embedding
-
- * cppyy for C++
-
-- Lots of CFFI modules around:
-
- * pygame_cffi, psycopg2_cffi, lxml
-
-- numpy: in-progress (more later)
+- 3.3: in-progress
Speed: 6.3x faster than CPython
@@ -247,16 +299,22 @@
- Distributed as part of Raspbian OS
-py3k
-----
+Compiled extensions
+-------------------
-- 3.2: stable
+- cpyext: incomplete, slow
-- 3.3: branch started, in-progress
+- ctypes: inconvenient, slow
-- Some missing optimizations
+- "Don't write your own C, PyPy is fast enough"
- * getting better
+|pause|
+
+- Native PyPy C API for embedding
+
+- CFFI: the future
+
+- numpy: in-progress (more later)
CFFI
@@ -264,148 +322,109 @@
- Python <-> C interfacing done right
+ * Inspired by LuaJIT's FFI
+
* existing shared libraries
* custom C code
-- Inspired by LuaJIT's FFI
-
- Alternative to C-API, ctypes, Cython, etc.
- Fast on CPython, super-fast on PyPy
+- Lots of CFFI modules around:
+
+ * pygame_cffi, psycopg2_cffi, lxml, cryptography, pyzmq, ...
+
+
+CFFI
+--------------------
+
+|example<| Simple example |>|
+|scriptsize|
+
+.. code:: python
+
+ >>> from cffi import FFI
+ >>> ffi = FFI()
+ >>> ffi.cdef("""
+ ... int printf(const char *format, ...); // from man page
+ ... """)
+ >>> C = ffi.dlopen(None) # loads whole C namespace
+ >>> arg = ffi.new("char[]", "world") # char arg[] = "world";
+ >>> C.printf("hi there, %s!\n", arg) # call printf()
+ hi there, world!
+
+|end_scriptsize|
+|end_example|
+
+- Docs: http://cffi.readthedocs.org
+
numpy
-----
-- As usual, in-progress
+- In-progress
- ~80% of numpy implemented
- * 2336 passing tests out of 3265
-
- * http://buildbot.pypy.org/numpy-status/latest.html
+ * 2415 passing tests out of 3265
- Just try it
- No scipy :-/
-cppyy
-------
-
-- Interface to C++
-
-- Based on reflection, no need to write wrappers
-
-- PyPy-only, similar to PyCintex for CPython
-
-- Main use case: ROOT
-
- * http://root.cern.ch
-
- * "a set of OO frameworks with all the functionality needed to handle and
- analyze large amounts of data in a very efficient way"
-
-- 3x faster than CPython
-
-
The future: STM
----------------
- Software Transactional Memory
-- Strategy to solve race conditions
-
-- "Finger crossed", rollback in case of conflicts
+- "Easier parallelism for everybody"
- On-going research project
* by Armin Rigo and Remi Meier
-STM semantics
--------------
+Transactional Memory
+--------------------
-- N threads
+* like GIL, but instead of blocking, each thread runs optimistically
-- Each thread split into atomic blocks
+* "easy" to implement:
-- Sequential execution in some arbitrary order
+ - GIL acquire -> transaction start
-- In practice:
+ - GIL release -> transaction commit
-- Parallel execution, conflicts solved by STM
+* overhead: cross-checking conflicting memory reads and writes,
+ and if necessary, cancel and restart transactions
-Unit of execution (1)
----------------------
-
-- Atomic blocks == 1 Python bytecode
-
-- Threads are executed in arbitrary order, but bytecodes are atomic
-
-- ==> Same semantics as GIL
-
-- "and this will solve the GIL problem" (A. Rigo, EuroPython 2011 lighting
talk)
-
-Unit of execution (2)
-----------------------
+Longer transactions
+-------------------
- Larger atomic blocks
-- ``with atomic:``
+- threads and application-level locks still needed...
-- Much easier to use than explicit locks
+- but *can be very coarse:*
+
+ - two transactions can optimistically run in parallel
+
+ - even if they both *acquire and release the same lock*
- Can be hidden by libraries to provide even higher level paradigms
* e.g.: Twisted apps made parallel out of the box
-Race conditions
----------------
-- They don't magically disappear
-
-- With explicit locks
-
- * ==> BOOM
-
- * you fix bugs by preventing race conditions
-
-- With atomic blocks
-
- * ==> Rollback
-
- * Performance penalty
-
- * You optimize by preventing race conditions
-
-- Fast&broken vs. Slower&correct
-
-
-Implementation
----------------
-
-- Conflicts detection, commit and rollback is costly
-
-- Original goal (2011): 2x-5x slower than PyPy without STM
-
- * But parallelizable!
-
-|pause|
-
-- Current goal (2014): 25% slower than PyPy without STM
-
-- Yes, that's 10x less overhead than original goal
-
-- mmap black magic
-
-Current status
----------------
+STM status
+----------
- Preliminary versions of pypy-jit-stm available
-- The JIT overhead is still a bit too high
+- Best case 25-40% overhead (much better than originally planned)
- Lots of polishing needed
@@ -417,9 +436,11 @@
- numpy: 48,412 $ of 60,000 $ (80.7%)
-- STM, 1st call: 25,000 $
+- STM (2nd call): 13,939 $ of 80,000 $ (17.4%)
-- STM, 2nd call: 13,939 $ of 80,000 $ (17.4%)
+- General PyPy progress
+
+- **September only**: PSF matches donations
- Thanks to all donors!
@@ -429,8 +450,8 @@
- http://pypy.org
-- http://morepypy.blogspot.com/
+- Commercial support: http://baroquesoftware.com
-- IRC: #p...@freenode.net
+- IRC: ``#pypy`` on freenode.net
- Any question?
_______________________________________________
pypy-commit mailing list
pypy-commit@python.org
https://mail.python.org/mailman/listinfo/pypy-commit
