Update README.rst

Fix ImportError for C extensions on Python 3.4b2

.gitattributes

@@ -0,0 +1 @@
+llvm/_version.py export-subst

.gitignore

@@ -2,4 +2,7 @@
 build
 _build
 *.pyc
+*.so
 /llvm/_intrinsic_ids.py
+llvm_
+newbinding/api/*

.travis.yml

@@ -0,0 +1,24 @@
+language: python
+
+python:
+  - "2.6"
+  - "2.7"
+  - "3.2"
+  - "3.3"
+
+branches:
+  only:
+    - master
+
+install:
+  - wget "http://repo.continuum.io/pkgs/free/linux-64/llvm-3.2-0.tar.bz2"
+  - tar -xjf llvm-3.2-0.tar.bz2
+  - PATH+=":`pwd`/bin"
+  - export LD_LIBRARY_PATH="`pwd`/lib"
+  - export LLVM_CONFIG_PATH="`pwd`/bin/llvm-config"
+  - $LLVM_CONFIG_PATH --cflags # test llvm-config
+  - export LLVMPY_DYNLINK=1
+  - python setup.py install -q
+
+script: cd ~; python -c "import sys;import llvm;sys.exit(0 if llvm.test() == 0 else 1)"
+

CHANGELOG

@@ -1,14 +1,88 @@
+2014-04-28    0.12.5:
+---------------------
+  * Fixes memory leaks (#92)
+  * Fixes tarball (#99)
 
-0.7, in progress:
+2014-03-20    0.12.4:
+---------------------
+  * Add dylib_import_library and friends
+  * Fix BasicBlock downcast
+  * Module hashing
+  * Fix test script
 
+2014-02-18    0.12.3:
+---------------------
+  * Fix deprecation message for py2.6
+  * Fix llvm_cbuilder for using deprecated_alloca
+  * Merged PR #88 by cantora
+  * Merged PR #94 by cgohlke
+
+2014-02-04    0.12.2:
+---------------------
+  * enhance wrapper efficiency by moving some capsule code into C++
+  * fix unclosed file handler in avx_support
+  * multiple-dimension insert_value, extract_value
+  * various minor fixes
+
+2013-11-11    0.12.1:
+---------------------
+  * various bug fixes
+
+2013-08-28    0.12.0:
+---------------------
+  * update to LLVM 3.3 and maintain compatibility with LLVM 3.2
+  * add LLRT for minimal support for 64-bit divmod on 32-bit platform
+  * start to adopt MCJIT (not quite usable on win32)
+  * various bug fixes
+
+2013-03-05   0.11.1:
+--------------------
+  * fix test when cc is not available
+  * fix Python 3 division (Hernan Grecco) (Issue #59)
+  * add relocation enums and add reloc argument for TargetMachine
+
+
+2013-03-01   0.11.0:
+--------------------
+  * fix Python 3 support on Windows
+  * New llvm binding
+
+
+2013-02-01   0.10.2:
+--------------------
+  * change default to link dynamically to LLVM use:
+      $ export LLVMPY_DYNLINK=0   # link statically
+      $ export LLVMPY_DYNLINK=1   # link dynamically
+      $ unset LLVMPY_DYNLINK  # tries to link dynamically if LLVM shared
+                              # objects are found and statically otherwise
+  * fix llpython for Python 2.6 support
+
+
+2013-01-25   0.10.1:
+--------------------
+  * fix support for Python 2.6
+
+
+2013-01-18   0.10.0:
+--------------------
+  * Add LLVM 3.2 support.
+  * New TargetData class.
+  * Fixed windows issue (Issue #42).
+  * Add ExecutionEngine.add_global_mapping and .get_pointer_to_global.
+  * Improved TargetMachine class and added code-model constants (CM_*).
+  * Added llvm.passes.build_pass_managers as a simpler way to build PassManagers.
+
+
+in progress, 0.7:
+-----------------
   * Add llvm.core.Argument.alignment property.
   * Migrate to LLVM 2.8.
   * Fix ffi link issue on darwin (Albert Mietus) (Issue #29).
   * LLVM tutorial ported (Max Shawabkeh) (Issue #33).
 
 
-0.6, 31-Aug-2010:
-
+2010-08-31   0.6:
+-----------------
   * Add and remove function attributes (Krzysztof Goj) (Issue #21).
   * Wrap fadd,fsub,fmul (Aaron S Lav) (Issue #31).
   * Migrate to LLVM 2.7.
@@ -27,16 +101,16 @@
   * Migrate to LLVM 2.5.
 
 
-0.5, 22-Nov-2008:
-
+2008-11-22   0.5:
+-----------------
   * Added vicmp, vfcmp instructions and constant expressions.
   * Builds on FreeBSD.
   * Updated documentation.
   * Migrate to LLVM 2.4.
 
 
-0.4, 21-Nov-2008:
-
+2008-11-21   0.4:
+-----------------
   * Code cleanup, added license headers.
   * Added llvm.core.load_library_permanently() (Issue #12).
   * Fix comparison using != (Issue #11).
@@ -45,8 +119,8 @@
   * Added viewCFG methods to Function (Paulo Silva).
 
 
-0.3, 8-Sep-2008:
-
+2008-09-08   0.3:
+-----------------
   * Passes added.
   * Assembly support: create modules from .ll files.
   * Various bug fixes.
@@ -58,13 +132,13 @@
   * Updated documentation.
 
 
-0.2.1, 18-Jun-2008:
-
+2008-06-28   0.2.1:
+-------------------
   * Build cleanly with LLVM 2.3 and 2.3svn.
 
 
-0.2, 15-Jun-2008:
-
+2008-06-15   0.2:
+-----------------
   * Independent package, need not be unpacked into llvm/bindings.
   * Fixed ownership issues with Module/ModuleProvider.
   * Modules, values and types can be stringified, to get their LLVM
@@ -79,7 +153,6 @@
   * Lots of cleanup.
 
 
-0.1, 10-May-2008:
-
-  * Initial release.
-
+2008-05-10   0.1:
+-----------------
+  * Initial release

MANIFEST.in

@@ -1,4 +1,6 @@
-include CHANGELOG LICENSE README setup.py MANIFEST.in
+include CHANGELOG LICENSE README.rst setup.py MANIFEST.in versioneer.py
 recursive-include llvm *
+recursive-include llvmpy *
 recursive-include www *
 recursive-include test *
+recursive-include tools *

README.rst

@@ -10,27 +10,55 @@ http://www.llvmpy.org
 Versions
 --------
 
-This package has only been tested with LLVM 3.1, and Python 2.7, (not Python 3.x).
+This package has been tested with LLVM 3.2, Python 2.6, 2.7 and 3.3.
+Other Python versions may work.
 
 Quickstart
 ----------
 
-1. Get 3.1 version of LLVM, build it.  Make sure ``--enable-pic`` is passed to
-   LLVM's ``configure``.
+1. Get and extract LLVM 3.2 source tarball from
+   `llvm.org <http://llvm.org/releases/download.html#3.2>`_.  Then, ``cd`` into
+   the extracted directory.
 
-2. Get llvm-py and install it::
+2. Run ``./configure --enable-optimized --prefix=LLVM_INSTALL_PATH``.
+
+    **Note**: Without the ``--enable-optimized`` flag, debug build will be
+    selected.  Unless you are developing LLVM or llvmpy, it is recommended
+    that the flag is used to reduce build time and binary size.
+    
+    **Note**: Use prefix to select the installation path.  It is recommended
+    to separate your custom build from the default system package.  Please
+    replace ``LLVM_INSTALL_PATH`` with your own path.
+
+3. Run ``REQUIRES_RTTI=1 make install`` to build and install.
+
+    **Note**: With LLVM 3.2, the default build configuration has C++ RTTI 
+    disabled.  However, llvmpy requires RTTI.
+    
+    **Note**: Use ``make -j2 install`` to enable concurrent build.  
+    Replace ``2`` with the actual number of processor you have. 
+
+4. Get llvm-py and install it::
 
    $ git clone git@github.com:llvmpy/llvmpy.git
    $ cd llvmpy
-   $ python setup.py install
+   $ LLVM_CONFIG_PATH=LLVM_INSTALL_PATH/bin/llvm-config python setup.py install
 
    Run the tests::
 
    $ python -c "import llvm; llvm.test()"
 
-3. See documentation at 'http://www.llvmpy.org/pages.html' and examples
+5. See documentation at 'http://www.llvmpy.org' and examples
    under 'test'.
    
+Common Build Problems
+---------------------
+
+1. If llvmpy cannot be imported due to "undefined symbol:
+   _ZTIN4llvm24PassRegistrationListenerE", it is because RTTI is not enabled
+   when building LLVM.  "_ZTIN4llvm24PassRegistrationListenerE" is the typeinfo
+   of PassRegistrationListener class.
+
 LICENSE
 -------
 

README_LLVM_CBUILDER.md

@@ -0,0 +1,144 @@
+LLVM CBuilder
+=============
+
+A few short examples:
+(TODO: add more later)
+
+```python
+from llvm.core import *
+from llvm_cbuilder import *
+import llvm_cbuilder.shortnames as C
+```
+
+```python
+class Square(CDefinition):
+    _name_ = 'square'
+    _retty_ = C.double
+    _argtys_ = [ ('x', C.double) ]
+
+    def body(self, x):
+        y = x * x
+        self.ret(y)
+```
+
+```python
+m = Module.new('my_module')
+llvm_square = Square()(m)
+print(m)
+```
+
+```
+; ModuleID = 'my_module'
+
+define double @square(double %x) {
+decl:
+  %0 = fmul double %x, %x
+  ret double %0
+}
+```
+
+
+```python
+class IsPrime(CDefinition):
+    _name_ = 'isprime'
+    _retty_ = C.int
+    _argtys_ = [('x', C.int)]
+
+    def body(self, x):
+        false = zero = self.constant(C.int, 0)
+        true = one = self.constant(C.int, 1)
+
+        two = self.constant(C.int, 2)
+
+        with self.ifelse( x <= two ) as ifelse:
+            with ifelse.then():
+                self.ret(true)
+
+        with self.ifelse( (x % two) == zero ) as ifelse:
+            with ifelse.then():
+                self.ret(false)
+
+        idx = self.var(C.int, 3, name='idx')
+
+        with self.loop() as loop:
+            with loop.condition() as setcond:
+                setcond( idx < x )
+
+            with loop.body():
+                with self.ifelse( (x % idx ) == zero ) as ifelse:
+                    with ifelse.then():
+                        self.ret(false)
+                idx += two
+
+        self.ret(true)
+```
+
+
+
+```
+define i32 @isprime(i32 %x) {
+decl:
+  %0 = icmp sle i32 %x, 2
+  br i1 %0, label %if.then, label %if.end
+
+if.then:                                          ; preds = %loop.body, %loop.cond, %if.end, %decl
+  %merge = phi i32 [ 1, %decl ], [ 0, %if.end ], [ 1, %loop.cond ], [ 0, %loop.body ]
+  ret i32 %merge
+
+if.end:                                           ; preds = %decl
+  %1 = srem i32 %x, 2
+  %2 = icmp eq i32 %1, 0
+  br i1 %2, label %if.then, label %if.end4
+
+if.end4:                                          ; preds = %if.end
+  br label %loop.cond
+
+loop.cond:                                        ; preds = %if.end7, %if.end4
+  %idx.0 = phi i32 [ 3, %if.end4 ], [ %6, %if.end7 ]
+  %3 = icmp slt i32 %idx.0, %x
+  br i1 %3, label %loop.body, label %if.then
+
+loop.body:                                        ; preds = %loop.cond
+  %4 = srem i32 %x, %idx.0
+  %5 = icmp eq i32 %4, 0
+  br i1 %5, label %if.then, label %if.end7
+
+if.end7:                                          ; preds = %loop.body
+  %6 = add i32 %idx.0, 2
+  br label %loop.cond
+}
+
+; ModuleID = 'my_module'
+
+define i32 @isprime(i32 %x) {
+decl:
+  %0 = icmp sle i32 %x, 2
+  br i1 %0, label %if.then, label %if.end
+
+if.then:                                          ; preds = %loop.body, %loop.cond, %if.end, %decl
+  %merge = phi i32 [ 1, %decl ], [ 0, %if.end ], [ 1, %loop.cond ], [ 0, %loop.body ]
+  ret i32 %merge
+
+if.end:                                           ; preds = %decl
+  %1 = srem i32 %x, 2
+  %2 = icmp eq i32 %1, 0
+  br i1 %2, label %if.then, label %if.end4
+
+if.end4:                                          ; preds = %if.end
+  br label %loop.cond
+
+loop.cond:                                        ; preds = %if.end7, %if.end4
+  %idx.0 = phi i32 [ 3, %if.end4 ], [ %6, %if.end7 ]
+  %3 = icmp slt i32 %idx.0, %x
+  br i1 %3, label %loop.body, label %if.then
+
+loop.body:                                        ; preds = %loop.cond
+  %4 = srem i32 %x, %idx.0
+  %5 = icmp eq i32 %4, 0
+  br i1 %5, label %if.then, label %if.end7
+
+if.end7:                                          ; preds = %loop.body
+  %6 = add i32 %idx.0, 2
+  br label %loop.cond
+}
+```

buildscripts/condarecipe/bld.bat

@@ -0,0 +1,6 @@
+set LLVMPY_DYNLINK=0
+set INCLUDE=%LIBRARY_INC%
+set LIBPATH=%LIBRARY_LIB%
+set LIB=%LIBRARY_LIB%
+%PYTHON% setup.py install
+if errorlevel 1 exit 1

buildscripts/condarecipe/build.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+
+if [[ (`uname` == Linux) && (`uname -m` != armv6l) ]]
+then
+    export CC=gcc
+	#gcc44
+    export CXX=g++
+	#g++44
+fi
+
+export LLVMPY_DYNLINK=$DISTRO_BUILD
+
+$PYTHON setup.py install

buildscripts/condarecipe/meta.yaml

@@ -0,0 +1,28 @@
+package:
+  name: llvmpy
+  version: 99.9.9
+  
+source:
+  git_url: git@github.com:llvmpy/llvmpy.git
+#  git_tag: 0.12.0
+
+requirements:
+  build:
+    - llvm
+    - python
+    #- chrpath        [linux]
+  run:
+    - llvm           [unix]
+    - python
+
+test:
+  imports:
+    - llvm
+    - llvmpy
+    - llvmpy._api
+    - llvmpy._capsule
+    - llpython
+    - llvm_array
+    - llvm_cbuilder
+
+

buildscripts/condarecipe/run_test.py

@@ -0,0 +1,22 @@
+import sys
+import platform
+import llvm
+
+from llvm.core import Module
+from llvm.ee import EngineBuilder
+from llvm.utils import check_intrinsics
+
+m = Module.new('fjoidajfa')
+eb = EngineBuilder.new(m)
+target = eb.select_target()
+
+print('target.triple=%r' % target.triple)
+if sys.platform == 'darwin':
+    s = {'64bit': 'x86_64', '32bit': 'x86'}[platform.architecture()[0]]
+    assert target.triple.startswith(s + '-apple-darwin')
+
+assert llvm.test(verbosity=2, run_isolated=False) == 0
+#check_intrinsics.main()
+
+print('llvm.__version__: %s' % llvm.__version__)
+#assert llvm.__version__ == '0.12.0'

docs/gh-pages.py

@@ -30,7 +30,7 @@ from subprocess import Popen, PIPE, CalledProcessError, check_call
 pages_dir = 'gh-pages'
 html_dir = '_build/html'
 pdf_dir = '_build/latex'
-pages_repo = 'git@github.com:llvmpy/llvmpy-doc.git'
+pages_repo = 'https://github.com/llvmpy/llvmpy-doc.git'
 
 #-----------------------------------------------------------------------------
 # Functions

docs/source/conf.py

@@ -11,12 +11,23 @@
 # All configuration values have a default; values that are commented out
 # serve to show the default.
 
-import sys, os
+import sys, os, glob
 
 # If extensions (or modules to document with autodoc) are in another directory,
 # add these directories to sys.path here. If the directory is relative to the
 # documentation root, use os.path.abspath to make it absolute, like shown here.
-#sys.path.insert(0, os.path.abspath('.'))
+#sys.path.insert(0, os.path.abspath('../..'))
+
+# Support sphinx.ext.autodoc to extract docstrings from modules without installing
+# complete package.
+# The python modules depend on _core, so we must build entire package first though.
+built_lib = glob.glob('../../build/lib.*-%d.%d/' % sys.version_info[:2])
+if not built_lib:
+    sys.stderr.write("WARNING: To build complete documentation you must build "
+                     "package first\n")
+else:
+    # lib dir has platform suffix
+    sys.path.insert(0, os.path.abspath(built_lib[0]))
 
 # -- General configuration -----------------------------------------------------
 
@@ -25,7 +36,7 @@ import sys, os
 
 # Add any Sphinx extension module names here, as strings. They can be extensions
 # coming with Sphinx (named 'sphinx.ext.*') or your custom ones.
-extensions = ['sphinx.ext.mathjax']
+extensions = ['sphinx.ext.mathjax', 'sphinx.ext.autodoc']
 
 # Add any paths that contain templates here, relative to this directory.
 templates_path = ['_templates']
@@ -41,16 +52,22 @@ master_doc = 'index'
 
 # General information about the project.
 project = u'llvmpy'
-copyright = u'2012, Mahadevan R (2008-2010), Continuum Analytics (2012)'
+copyright = u'2013, Mahadevan R (2008-2010), Continuum Analytics (2012-2013)'
 
 # The version info for the project you're documenting, acts as replacement for
 # |version| and |release|, also used in various other places throughout the
 # built documents.
 #
+try:
+    import llvm
+    version_strs = llvm.__version__.split('.')
     # The short X.Y version.
-version = '0.8'
+    version = '.'.join(version_strs[:2])
     # The full version, including alpha/beta/rc tags.
-release = '0.8.2'
+    release = '%s.%s' % (version, '-'.join(version_strs[2].split('-')[:2]))
+except ImportError:
+    version = 'unknown-version'
+    release = 'unknown-release'
 
 # The language for content autogenerated by Sphinx. Refer to documentation
 # for a list of supported languages.

docs/source/doc/core.rst

@@ -0,0 +1,9 @@
+********************************
+llvm.core
+********************************
+
+.. toctree::
+   :titlesonly:
+   :glob:
+
+   llvm.core.*

docs/source/doc/examples/JITTutorial2.rst

@@ -1,5 +1,5 @@
 A More Complicated Function
-====================
+===========================
 
 .. code-block:: python
 

docs/source/doc/getting_started.rst

@@ -26,7 +26,7 @@ Both LLVM and llvmpy are distributed under (different) permissive open
 source licenses. llvmpy uses the `new BSD
 license <http://opensource.org/licenses/bsd-license.php>`_. More
 information is available
-`here <https://github.com/numba/llvmpy/blob/master/LICENSE>`_.
+`here <https://github.com/llvmpy/llvmpy/blob/master/LICENSE>`_.
 
 Platforms
 ---------
@@ -38,12 +38,11 @@ supported, for a variety of reasons.
 Versions
 --------
 
-llvmpy 0.8.2 requires version 3.1 of LLVM. It may not work with
+llvmpy 0.11.2 uses LLVM 3.2 (or at least 3.1). It may not work with
 previous versions.
 
-llvmpy has been built and tested with Python 2.7. It should work with
-earlier versions. It has not been tried with Python 3.x (patches
-welcome).
+llvmpy has been built and tested with Python 2.7 and 3.2. It should work with
+earlier versions.
 
 
 Installation
@@ -66,22 +65,6 @@ It does not matter which compiler LLVM itself was built with (``g++``,
 ``llvm-g++`` or any other); llvmpy can be built with any compiler. It
 has been tried only with gcc/g++ though.
 
-
-LLVM and ``--enable-pic``
--------------------------
-
-The result of an LLVM build is a set of static libraries and object
-files. The llvmpy contains an extension package that is built into a
-shared object (\_core.so) which links to these static libraries and
-object files. It is therefore required that the LLVM libraries and
-object files be built with the ``-fPIC`` option (generate position
-independent code). Be sure to use the ``--enable-pic`` option while
-configuring LLVM (default is no PIC), like this:
-
-.. code-block:: bash
-  
-   $ ~/llvm ./configure --enable-pic --enable-optimized
-
 llvm-config
 -----------
 
@@ -99,13 +82,34 @@ is different from that of 'root', so even if ``llvm-config`` is in your
 Steps
 -----
 
-Get 3.1 version of LLVM, build it. Make sure '--enable-pic' is passed to
-LLVM's 'configure'.
+1. Get and extract LLVM 3.2 source tarball from
+   `llvm.org <http://llvm.org/releases/download.html#3.2>`_.  Then, ``cd`` into
+   the extracted directory.
 
-Get llvmpy and install it:
+2. Run ``./configure --enable-optimized --prefix=LLVM_INSTALL_PATH``.
 
+    **Note**: Without the ``--enable-optimized`` flag, debug build will be
+    selected.  Unless you are developing LLVM or llvmpy, it is recommended
+    that the flag is used to reduce build time and binary size.
     
-.. code-block:: bash
+    **Note**: Use prefix to select the installation path.  It is recommended
+    to separate your custom build from the default system package.  Please
+    replace ``LLVM_INSTALL_PATH`` with your own path.
 
-   $ git clone git@github.com:numba/llvmpy.git
-   $ cd llvmpy $ python setup.py install
+3. Run ``REQUIRES_RTTI=1 make`` to build.
+
+    **Note**: With LLVM 3.2, the default build configuration has C++ RTTI 
+    disabled.  However, llvmpy requires RTTI.
+
+4. Get llvm-py and install it::
+
+   $ git clone git@github.com:llvmpy/llvmpy.git
+   $ cd llvmpy
+   $ LLVM_CONFIG_PATH=LLVM_INSTALL_PATH/bin/llvm-config python setup.py install
+
+   Run the tests::
+
+   $ python -c "import llvm; llvm.test()"
+
+5. See documentation at 'http://www.llvmpy.org' and examples
+   under 'test'.

docs/source/doc/kaleidoscope/PythonLangImpl1.rst

@@ -6,7 +6,7 @@ Written by `Chris Lattner <mailto:sabre@nondot.org>`_ and `Max
 Shawabkeh <http://max99x.com>`_
 
 Introduction
-=========
+============
 
 Welcome to the "Implementing a language with LLVM" tutorial. This
 tutorial runs through the implementation of a simple language, showing
@@ -115,13 +115,13 @@ following simple example computes `Fibonacci
 numbers <http://en.wikipedia.org/wiki/Fibonacci_number>`_:
 
 
-.. code-block:: python
+.. code-block:: text
 
    # Compute the x'th fibonacci number. 
-   def fib(x):
-      if x < 3:
+   def fib(x)
+      if x < 3
          return 1
-      else:
+      else
          return fib(x-1)+fib(x-2)
    
    # This expression will compute the 40th number.
@@ -226,7 +226,7 @@ the line.
    
    # Regular expressions that tokens and comments of our language.
    REGEX_NUMBER = re.compile('[0-9]+(?:.[0-9]+)?')
-   REGEX_IDENTIFIER = re.compile('[a-zA-Z][a-zA-Z0-9]\ *')
+   REGEX_IDENTIFIER = re.compile('[a-zA-Z][a-zA-Z0-9]*')
    REGEX_COMMENT = re.compile('#.*')
    
 

docs/source/doc/kaleidoscope/PythonLangImpl5.rst

@@ -217,7 +217,8 @@ Kaleidoscope looks something like this:
       br i1 %ifcond, label %then, label %else 
 
    then:       ; preds = %entry 
-      %calltmp1 = call double @bar() 
+      %calltmp = call double @foo() 
+      br label %ifcont 
 
    else:       ; preds = %entry 
       %calltmp1 = call double @bar() 

docs/source/doc/kaleidoscope/PythonLangImpl7.rst

@@ -133,11 +133,11 @@ instruction <http://www.llvm.org/docs/LangRef.html#i_alloca>`_:
 
    define i32 @example() { 
    entry: 
-         %X = alloca i32          ; type of %X is i32 *. 
+         %X = alloca i32                ; type of %X is i32* 
          ... 
-         %tmp = load i32* %X      ; load the stack value %X from the stack. 
-         %tmp2 = add i32 %tmp, 1  ; increment it store i32 %tmp2,
-         i32* %X                  ; store it back 
+         %tmp = load i32* %X            ; load the stack value %X from the stack 
+         %tmp2 = add i32 %tmp, 1        ; increment it 
+         store i32 %tmp2, i32* %X       ; store it back 
          ...
 
 
@@ -295,7 +295,7 @@ variables is a useful thing regardless of whether you will be mutating
 them. Here's a motivating example that shows how we could use these:
 
 
-.. code-block:: 
+.. code-block:: none
 
    # Define ':' for sequencing: as a low-precedence operator that ignores operands 
    # and just returns the RHS. 
@@ -352,7 +352,7 @@ function that ensures that the allocas are created in the entry block of
 the function:
 
 
-.. code-block:: 
+.. code-block:: python
 
    # Creates an alloca instruction in the entry block of the function. This is used 
    # for mutable variables. 
@@ -475,7 +475,7 @@ It is interesting to see what the code looks like before and after the
 mem2reg optimization runs. For example, this is the before/after code
 for our recursive fib function. Before the optimization:
 
-.. code-block:: 
+.. code-block:: llvm
    
    define double @fib(double %x) { 
    entry: 
@@ -515,7 +515,7 @@ still just make the PHI.
 
 Here is the code after the mem2reg pass runs:
 
-.. code-block:: 
+.. code-block:: llvm
    
    define double @fib(double %x) { 
    entry: 
@@ -651,7 +651,7 @@ Now that we have an assignment operator, we can mutate loop variables
 and arguments. For example, we can now run code like this:
 
 
-.. code-block::
+.. code-block:: none
 
    # Function to print a double. 
    extern printd(x)

docs/source/doc/kaleidoscope/PythonLangImpl8.rst

@@ -1,6 +1,6 @@
-*************************************************
+***************************************************
 Chapter 8: Conclusion and other useful LLVM tidbits
-*************************************************
+***************************************************
 
 Written by Chris Lattner
 
@@ -106,7 +106,7 @@ generating LLVM IR. These are some of the more subtle things that may not be obv
 but are very useful if you want to take advantage of LLVM's capabilities.
 
 Properties of the LLVM IR
-========================
+=========================
 
 We have a couple common questions about code in the LLVM IR form - let's 
 just get these out of the way right now, shall we?
@@ -164,7 +164,7 @@ This can make sense for specialized domains such as an in-kernel language.
 --------------
 
 Safety Guarantees
-----------------
+-----------------
 
 Many of the languages above are also "safe" languages: it is 
 impossible for a program written in Java to corrupt its address space and 
@@ -236,7 +236,7 @@ you desire in your front-end, on the language-specific AST.
 --------------
 
 Tips and Tricks
-==============
+===============
 
 There is a variety of useful tips and tricks that you come to 
 know after working on/with LLVM that aren't obvious at first glance. 

docs/source/doc/llpython/articles.rst

@@ -0,0 +1,10 @@
+LLPython Articles
+=================
+
+.. toctree::
+   :titlesonly:
+   :maxdepth: 1
+
+   intro_llpython.rst
+   nobitey_dev.rst
+   ctmp_in_llpython.rst

docs/source/doc/llpython/ctmp_in_llpython.rst

@@ -0,0 +1,5 @@
+Compile-time Metaprogramming in LLPython
+========================================
+
+In this article, we discuss how LLPython supports compile-time
+metaprogramming.

docs/source/doc/llpython/index.rst

@@ -0,0 +1,9 @@
+LLPython Documentation
+======================
+
+.. toctree::
+   :titlesonly:
+   :maxdepth: 2
+
+   articles.rst
+   reference.rst

docs/source/doc/llpython/intro_llpython.rst

@@ -0,0 +1,46 @@
+====================
+Introducing LLPython
+====================
+
+In this article, we introduce the llpython package.  The primary goal
+of the llpython package is to provide a Python dialect/subset that
+maps directly to LLVM code.  LLPython differs from its originating
+LLVM translator, Numba, in the following aspects:
+
+  * LLPython code is not intended to work in Python if not translated
+    and wrapped.
+  * The LLPython translator only uses LLVM types.
+  * LLPython is explicitly typed, and does not support type inference.
+    LLPython does not support implicit casts, all casts must be explicit.
+  * LLPython supports code that directly calls the C API, the Python C
+    API, and the llvm.core.Builder methods.
+
+Additionally, we designed the sub-package to have the following
+engineering properties:
+
+  * Usable from Python 2.7, and 3.X.  At the time of writing, we plan
+    to support Python 2.6.
+  * Clean from Numba dependencies (other than llvmpy), and can be used
+    as a standalone code generator without a full Numba installation.
+  * Provides a series of Python bytecode passes that can be easily
+    used by other projects.
+
+
+LLPython Origins
+================
+
+We developed LLPython with the initial goal of simplifying writing
+LLVM code.
+
+
+LLPython Internals
+==================
+
+In this section, we describe the various passes performed by the
+LLPython translator.
+
+
+Conclusions
+===========
+
+LLPython is neat.

docs/source/doc/llpython/llpython.byte_control.rst

@@ -0,0 +1,6 @@
+=====================
+llpython.byte_control
+=====================
+
+.. automodule:: llpython.byte_control
+   :members:

docs/source/doc/llpython/llpython.byte_flow.rst

@@ -0,0 +1,6 @@
+==================
+llpython.byte_flow
+==================
+
+.. automodule:: llpython.byte_flow
+   :members:

docs/source/doc/llpython/llpython.byte_translator.rst

@@ -0,0 +1,6 @@
+========================
+llpython.byte_translator
+========================
+
+.. automodule:: llpython.byte_translator
+   :members:

docs/source/doc/llpython/llpython.bytecode_visitor.rst

@@ -0,0 +1,6 @@
+=========================
+llpython.bytecode_visitor
+=========================
+
+.. automodule:: llpython.bytecode_visitor
+   :members:

docs/source/doc/llpython/llpython.bytetype.rst

@@ -0,0 +1,6 @@
+=================
+llpython.bytetype
+=================
+
+.. automodule:: llpython.bytetype
+   :members:

docs/source/doc/llpython/llpython.control_flow.rst

@@ -0,0 +1,6 @@
+=====================
+llpython.control_flow
+=====================
+
+.. automodule:: llpython.control_flow
+   :members:

docs/source/doc/llpython/llpython.gen_bytecode_visitor.rst

@@ -0,0 +1,6 @@
+=============================
+llpython.gen_bytecode_visitor
+=============================
+
+.. automodule:: llpython.gen_bytecode_visitor
+   :members:

docs/source/doc/llpython/llpython.nobitey.rst

@@ -0,0 +1,6 @@
+================
+llpython.nobitey
+================
+
+.. automodule:: llpython.nobitey
+   :members:

docs/source/doc/llpython/llpython.opcode_util.rst

@@ -0,0 +1,6 @@
+====================
+llpython.opcode_util
+====================
+
+.. automodule:: llpython.opcode_util
+   :members:

docs/source/doc/llpython/llpython.phi_injector.rst

@@ -0,0 +1,6 @@
+=====================
+llpython.phi_injector
+=====================
+
+.. automodule:: llpython.phi_injector
+   :members:

docs/source/doc/llpython/llpython.pyaddfunc.rst

@@ -0,0 +1,6 @@
+==================
+llpython.pyaddfunc
+==================
+
+.. automodule:: llpython.pyaddfunc
+   :members:

docs/source/doc/llpython/llpython.rst

@@ -0,0 +1,6 @@
+========
+llpython
+========
+
+.. automodule:: llpython
+   :members:

docs/source/doc/llpython/nobitey_dev.rst

@@ -0,0 +1,5 @@
+nobitey: Using ctypes and llvmpy to Bypass ctypes
+=================================================
+
+In this article, we show how nobitey uses llvmpy to eliminate the
+ctypes call overhead.

docs/source/doc/llpython/reference.rst

@@ -0,0 +1,21 @@
+LLPython Module Reference
+=========================
+
+Contents:
+
+.. toctree::
+   :titlesonly:
+   :maxdepth: 2
+
+   llpython.rst
+   llpython.bytecode_visitor.rst
+   llpython.byte_control.rst
+   llpython.byte_flow.rst
+   llpython.byte_translator.rst
+   llpython.bytetype.rst
+   llpython.control_flow.rst
+   llpython.gen_bytecode_visitor.rst
+   llpython.nobitey.rst
+   llpython.opcode_util.rst
+   llpython.phi_injector.rst
+   llpython.pyaddfunc.rst

docs/source/doc/llvm.core.Argument.rst

@@ -56,3 +56,10 @@ Add an attribute ``attr`` to the argument, from the set listed above.
 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 Remove the attribute ``attr`` of the argument.
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.Argument
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.ArrayType.rst

@@ -28,3 +28,10 @@ A ``Type`` object representing the type of the element of the array.
 [read-only]
 
 The number of elements in the array.
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.ArrayType
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.BasicBlock.rst

@@ -42,3 +42,10 @@ The parent function of this basicblock.
 ~~~~~~~~~~~~~~~~
 
 A list of instructions in this basicblock.
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.BasicBlock
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.Builder.rst

@@ -409,3 +409,10 @@ positioned.
 ~~~~~~~~~
 
 Deprecated. Same as ``basic_block``
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.Builder
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.Constant.rst

@@ -4,6 +4,9 @@
 | title: Constant (llvm.core)   |
 +-------------------------------+
 
+llvm.core.Constant
+==================
+
 ``Constant``-s represents constants that appear within the code. The
 values of such objects are known at creation time. Constants can be
 created from Python constants. A constant expression is also a constant
@@ -23,14 +26,15 @@ some examples:
    
    tr = Type.float()
    
-   r1 = Constant.real(tr, "3.141592") # create from a string r2 =
-   Constant.real(tr, 1.61803399) # create from a Python float {%
-   endhighlight %}
+   r1 = Constant.real(tr, "3.141592") # create from a string 
+   r2 = Constant.real(tr, 1.61803399) # create from a Python float
+
 
 # llvm.core.Constant
 -  This will become a table of contents (this text will be scraped).
 {:toc}
 
+
 Static factory methods
 ----------------------
 
@@ -351,3 +355,9 @@ some examples:
 
 assert isinstance(k1, ConstantInt) assert isinstance(k2, ConstantArray)
 
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.Constant
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.Function.rst

@@ -149,3 +149,10 @@ function bodies.
 
 Verifies the function. See `LLVM
 docs <http://llvm.org/docs/Passes.html#verify>`_.
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.Function
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.FunctionType.rst

@@ -46,3 +46,10 @@ the function. Used like this:
    Type.int(), Type.int() ] ) for arg in func_type.args: assert arg.kind
    == TYPE_INTEGER assert arg == Type.int() assert func_type.arg_count
    == len(func_type.args)
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.FunctionType
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.GlobalValue.rst

@@ -95,3 +95,9 @@ A power-of-2 integer indicating the boundary to align to.
 
     The module object to which this global belongs to.
 
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.GlobalValue
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.GlobalVariable.rst

@@ -4,6 +4,9 @@
 | title: GlobalVariable (llvm.core)   |
 +-------------------------------------+
 
+llvm.core.GlobalVariable
+========================
+
 Global variables (``llvm.core.GlobalVariable``) are subclasses of
 `llvm.core.GlobalValue <llvm.core.GlobalValue.html>`_ and represent
 module-level variables. These can have optional initializers and can be
@@ -37,3 +40,10 @@ class, or by using the static method ``GlobalVariable.new``.
    # list all global variables in a module
    for gv in module_obj.global_variables: print gv.name, "of type",
    gv.type
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.GlobalVariable
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.Instruction.rst

@@ -240,3 +240,10 @@ Properties
 
 The predicate of the compare instruction, one of the ``ICMP_*`` or
 ``FCMP_*`` constants.
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.Instruction
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.IntegerType.rst

@@ -21,3 +21,10 @@ Properties
 [read-only]
 
 The width of the integer type, in number of bits.
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.IntegerType
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.Module.rst

@@ -4,16 +4,26 @@
 | title: Module (llvm.core)   |
 +-----------------------------+
 
+llvm.core.Module
+================
+
 Modules are top-level container objects. You need to create a module
 object first, before you can add global variables, aliases or functions.
 Modules are created using the static method ``Module.new``:
 
-
 .. code-block:: python
 
    #!/usr/bin/env python
    
-   from llvm import \* from llvm.core import \*
+   from llvm import *
+   from llvm.core import *
    
    # create a module
    my_module = Module.new('my_module')
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.Module
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.PointerType.rst

@@ -29,3 +29,10 @@ The address space of the pointer.
 
 A `Type <llvm.core.Type.html>`_ object representing the type of the
 value pointed to.
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.PointerType
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.StructType.rst

@@ -0,0 +1,8 @@
+llvm.core.StructType
+====================
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.StructType
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.Type.rst

@@ -126,3 +126,10 @@ Example:
 
    assert Type.int().kind == TYPE_INTEGER assert
    Type.void().kind == TYPE_VOID
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.Type
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.User.rst

@@ -37,3 +37,10 @@ The list of operands (values, of type
 
 The number of operands that this value referes to. Same as
 ``len(uses.operands)`` but faster if you just want the count.
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.User
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.Value.rst

@@ -63,3 +63,10 @@ representation.
 ``Value`` objects can be compared for equality. Internally, this
 converts both arguments into their LLVM assembly representations and
 compares the resultant strings.
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.Value
+   :members:
+   :undoc-members:

docs/source/doc/llvm.core.VectorType.rst

@@ -29,3 +29,10 @@ element of the vector.
 [read-only]
 
 The number of elements in the vector.
+
+
+Automatically Generated Documentation
+-------------------------------------
+.. autoclass:: llvm.core.VectorType
+   :members:
+   :undoc-members:

docs/source/doc/llvm_cbuilder.rst

@@ -0,0 +1,12 @@
+llvm_cbuilder
+=============
+
+llvm_cbuilder is a set of Python-contexts you can use to write C-like
+constructs in Python which generates llvmpy code directly. 
+
+Similar to llpython it allows you to build llvm IR without using the
+llvmpy interface directly. 
+
+Look in the llvm_cbuilder tests directory for examples of use.
+
+

docs/source/doc/llvm_concepts.rst

@@ -28,6 +28,7 @@ follow these steps:
 -  add a *basic block* to the function
 -  using a helper object called an *instruction builder*, add two
    instructions into the basic block: 
+
     - an instruction to add the two
       arguments and store the result into a temporary variable 
     - a return

docs/source/doc/llvmcore.rst

@@ -0,0 +1,9 @@
+********************************
+llvm.core
+********************************
+
+.. toctree::
+   :titlesonly:
+   :glob:
+
+   llvm.core.*

docs/source/doc/llvmee.rst

@@ -0,0 +1,9 @@
+********************************
+llvm.ee
+********************************
+
+.. toctree::
+   :titlesonly:
+   :glob:
+
+   llvm.ee.*

docs/source/doc/llvmpasses.rst

@@ -0,0 +1,9 @@
+********************************
+llvm.passes
+********************************
+
+.. toctree::
+   :titlesonly:
+   :glob:
+
+   llvm.passes.*

docs/source/index.rst

@@ -3,8 +3,8 @@
    You can adapt this file completely to your liking, but it should at least
    contain the root `toctree` directive.
 
-Documentation for llvmpy
-=================
+llvmpy
+======
 
 Contents:
 
@@ -13,11 +13,16 @@ Contents:
    :maxdepth: 2
 
    doc/userguide.rst
+   doc/llvm_cbuilder.rst
+   doc/llpython/index.rst
    doc/examples.rst
+   doc/llvmcore.rst 
+   doc/llvmee.rst
+   doc/llvmpasses.rst
 
 
 Indices and tables
-==================
+------------------
 
 * :ref:`genindex`
 * :ref:`modindex`

example/vector_instr.py

@@ -0,0 +1,152 @@
+'''
+This example shows:
+1) how to use vector instructions
+2) how to take advantage of LLVM loop vectorization to transform scalar     
+   operations to vector operations
+'''
+
+from __future__ import print_function
+import llvm.core as lc
+import llvm.ee as le
+import llvm.passes as lp
+from ctypes import CFUNCTYPE, POINTER, c_int, c_float
+
+def build_manual_vector():
+    mod = lc.Module.new('manual.vector')
+    intty = lc.Type.int(32)
+    vecty = lc.Type.vector(lc.Type.float(), 4)
+    aryty = lc.Type.pointer(lc.Type.float())
+    fnty = lc.Type.function(lc.Type.void(), [aryty, aryty, aryty, intty])
+    fn = mod.add_function(fnty, name='vector_add')
+    bbentry = fn.append_basic_block('entry')
+    bbloopcond = fn.append_basic_block('loop.cond')
+    bbloopbody = fn.append_basic_block('loop.body')
+    bbexit = fn.append_basic_block('exit')
+    builder = lc.Builder.new(bbentry)
+
+    # populate function body
+    in1, in2, out, size = fn.args
+    ZERO = lc.Constant.null(intty)
+    loopi_ptr = builder.alloca(intty)
+    builder.store(ZERO, loopi_ptr)
+
+    builder.branch(bbloopcond)
+    builder.position_at_end(bbloopcond)
+
+    loopi = builder.load(loopi_ptr)
+    loopcond = builder.icmp(lc.ICMP_ULT, loopi, size)
+
+    builder.cbranch(loopcond, bbloopbody, bbexit)
+    builder.position_at_end(bbloopbody)
+
+    vecaryty = lc.Type.pointer(vecty)
+    in1asvec = builder.bitcast(builder.gep(in1, [loopi]), vecaryty)
+    in2asvec = builder.bitcast(builder.gep(in2, [loopi]), vecaryty)
+    outasvec = builder.bitcast(builder.gep(out, [loopi]), vecaryty)
+
+    vec1 = builder.load(in1asvec)
+    vec2 = builder.load(in2asvec)
+
+    vecout = builder.fadd(vec1, vec2)
+
+    builder.store(vecout, outasvec)
+
+    next = builder.add(loopi, lc.Constant.int(intty, 4))
+    builder.store(next, loopi_ptr)
+
+    builder.branch(bbloopcond)
+    builder.position_at_end(bbexit)
+
+    builder.ret_void()
+
+    return mod, fn
+
+
+def build_auto_vector():
+    mod = lc.Module.new('auto.vector')
+    # Loop vectorize is sensitive to the size of the index size(!?)
+    intty = lc.Type.int(tuple.__itemsize__ * 8)
+    aryty = lc.Type.pointer(lc.Type.float())
+    fnty = lc.Type.function(lc.Type.void(), [aryty, aryty, aryty, intty])
+    fn = mod.add_function(fnty, name='vector_add')
+    bbentry = fn.append_basic_block('entry')
+    bbloopcond = fn.append_basic_block('loop.cond')
+    bbloopbody = fn.append_basic_block('loop.body')
+    bbexit = fn.append_basic_block('exit')
+    builder = lc.Builder.new(bbentry)
+
+    # populate function body
+    in1, in2, out, size = fn.args
+    in1.add_attribute(lc.ATTR_NO_ALIAS)
+    in2.add_attribute(lc.ATTR_NO_ALIAS)
+    out.add_attribute(lc.ATTR_NO_ALIAS)
+    ZERO = lc.Constant.null(intty)
+    loopi_ptr = builder.alloca(intty)
+    builder.store(ZERO, loopi_ptr)
+
+    builder.branch(bbloopcond)
+    builder.position_at_end(bbloopcond)
+
+    loopi = builder.load(loopi_ptr)
+    loopcond = builder.icmp(lc.ICMP_ULT, loopi, size)
+
+    builder.cbranch(loopcond, bbloopbody, bbexit)
+    builder.position_at_end(bbloopbody)
+
+    in1elem = builder.load(builder.gep(in1, [loopi]))
+    in2elem = builder.load(builder.gep(in2, [loopi]))
+
+    outelem = builder.fadd(in1elem, in2elem)
+
+    builder.store(outelem, builder.gep(out, [loopi]))
+
+    next = builder.add(loopi, lc.Constant.int(intty, 1))
+    builder.store(next, loopi_ptr)
+
+    builder.branch(bbloopcond)
+    builder.position_at_end(bbexit)
+
+    builder.ret_void()
+
+    return mod, fn
+
+def example(title, module_builder, opt):
+    print(title.center(80, '='))
+    mod, fn = module_builder()
+
+    eb = le.EngineBuilder.new(mod).opt(3)
+    if opt:
+        print('opt')
+        tm = eb.select_target()
+        pms = lp.build_pass_managers(mod=mod, tm=tm, opt=3, loop_vectorize=True,
+                                     fpm=False)
+        pms.pm.run(mod)
+
+    print(mod)
+    print(mod.to_native_assembly())
+
+    engine = eb.create()
+    ptr = engine.get_pointer_to_function(fn)
+
+    callable = CFUNCTYPE(None, POINTER(c_float), POINTER(c_float),
+                         POINTER(c_float), c_int)(ptr)
+
+    N = 20
+    in1 = (c_float * N)(*range(N))
+    in2 = (c_float * N)(*range(N))
+    out = (c_float * N)()
+
+    print('in1: ', list(in1))
+    print('in1: ', list(in2))
+
+    callable(in1, in2, out, N)
+
+    print('out', list(out))
+
+
+def main():
+    example('manual vector function', build_manual_vector, False)
+    example('auto vector function', build_auto_vector, True)
+
+if __name__ == '__main__':
+    main()

llpython/byte_control.py

@@ -0,0 +1,125 @@
+# ______________________________________________________________________
+from __future__ import absolute_import
+import opcode
+from . import opcode_util
+import pprint
+
+from .bytecode_visitor import BasicBlockVisitor, BenignBytecodeVisitorMixin
+from .control_flow import ControlFlowGraph
+
+# ______________________________________________________________________
+
+class ControlFlowBuilder (BenignBytecodeVisitorMixin, BasicBlockVisitor):
+    '''Visitor responsible for traversing a bytecode basic block map and
+    building a control flow graph (CFG).
+
+    The primary purpose of this transformation is to create a CFG,
+    which is used by later transformers for dataflow analysis.
+    '''
+    def visit (self, flow, nargs = 0, *args, **kws):
+        '''Given a bytecode flow, and an optional number of arguments,
+        return a :py:class:`llpython.control_flow.ControlFlowGraph`
+        instance describing the full control flow of the bytecode
+        flow.'''
+        self.nargs = nargs
+        ret_val = super(ControlFlowBuilder, self).visit(flow, *args, **kws)
+        del self.nargs
+        return ret_val
+
+    def enter_blocks (self, blocks):
+        super(ControlFlowBuilder, self).enter_blocks(blocks)
+        self.blocks = blocks
+        self.block_list = list(blocks.keys())
+        self.block_list.sort()
+        self.cfg = ControlFlowGraph()
+        self.loop_stack = []
+        for block in self.block_list:
+            self.cfg.add_block(block, blocks[block])
+
+    def exit_blocks (self, blocks):
+        super(ControlFlowBuilder, self).exit_blocks(blocks)
+        assert self.blocks == blocks
+        self.cfg.compute_dataflow()
+        self.cfg.update_for_ssa()
+        ret_val = self.cfg
+        del self.loop_stack
+        del self.cfg
+        del self.block_list
+        del self.blocks
+        return ret_val
+
+    def enter_block (self, block):
+        self.block = block
+        assert block in self.cfg.blocks
+        if block == 0:
+            for local_index in range(self.nargs):
+                self.op_STORE_FAST(0, opcode.opmap['STORE_FAST'], local_index)
+        return True
+
+    def _get_next_block (self, block):
+        return self.block_list[self.block_list.index(block) + 1]
+
+    def exit_block (self, block):
+        assert block == self.block
+        del self.block
+        i, op, arg = self.blocks[block][-1]
+        opname = opcode.opname[op]
+        if op in opcode.hasjabs:
+            self.cfg.add_edge(block, arg)
+        elif op in opcode.hasjrel:
+            self.cfg.add_edge(block, i + arg + 3)
+        elif opname == 'BREAK_LOOP':
+            loop_i, _, loop_arg = self.loop_stack[-1]
+            self.cfg.add_edge(block, loop_i + loop_arg + 3)
+        elif opname != 'RETURN_VALUE':
+            self.cfg.add_edge(block, self._get_next_block(block))
+        if op in opcode_util.hascbranch:
+            self.cfg.add_edge(block, self._get_next_block(block))
+
+    def op_LOAD_FAST (self, i, op, arg, *args, **kws):
+        self.cfg.blocks_reads[self.block].add(arg)
+        return super(ControlFlowBuilder, self).op_LOAD_FAST(i, op, arg, *args,
+                                                            **kws)
+
+    def op_STORE_FAST (self, i, op, arg, *args, **kws):
+        self.cfg.writes_local(self.block, i, arg)
+        return super(ControlFlowBuilder, self).op_STORE_FAST(i, op, arg, *args,
+                                                             **kws)
+
+    def op_SETUP_LOOP (self, i, op, arg, *args, **kws):
+        self.loop_stack.append((i, op, arg))
+        return super(ControlFlowBuilder, self).op_SETUP_LOOP(i, op, arg, *args,
+                                                             **kws)
+
+    def op_POP_BLOCK (self, i, op, arg, *args, **kws):
+        self.loop_stack.pop()
+        return super(ControlFlowBuilder, self).op_POP_BLOCK(i, op, arg, *args,
+                                                            **kws)
+
+# ______________________________________________________________________
+
+def build_cfg (func):
+    '''Given a Python function, create a bytecode flow, visit the flow
+    object, and return a control flow graph.'''
+    co_obj = opcode_util.get_code_object(func)
+    return ControlFlowBuilder().visit(opcode_util.build_basic_blocks(co_obj),
+                                      co_obj.co_argcount)
+
+# ______________________________________________________________________
+# Main (self-test) routine
+
+def main (*args, **kws):
+    from tests import llfuncs
+    if not args:
+        args = ('doslice',)
+    for arg in args:
+        build_cfg(getattr(llfuncs, arg)).pprint()
+
+# ______________________________________________________________________
+
+if __name__ == "__main__":
+    import sys
+    main(*sys.argv[1:])
+
+# ______________________________________________________________________
+# End of byte_control.py

llpython/byte_flow.py

@@ -0,0 +1,266 @@
+# ______________________________________________________________________
+from __future__ import absolute_import
+import dis
+import opcode
+
+from .bytecode_visitor import BasicBlockVisitor
+from . import opcode_util
+
+# ______________________________________________________________________
+
+class BytecodeFlowBuilder (BasicBlockVisitor):
+    '''Transforms a CFG into a bytecode "flow tree".
+
+    The flow tree is a Python dictionary, described loosely by the
+    following set of productions:
+
+      * `flow_tree` ``:=`` ``{`` `blocks` ``*`` ``}``
+      * `blocks` ``:=`` `block_index` ``:`` ``[`` `bytecode_tree` ``*`` ``]``
+      * `bytecode_tree` ``:=`` ``(`` `opcode_index` ``,`` `opcode` ``,``
+          `opname` ``,`` `arg` ``,`` ``[`` `bytecode_tree` ``*`` ``]`` ``)``
+
+    The primary purpose of this transformation is to simulate the
+    value stack, removing it and any stack-specific opcodes.'''
+
+    def __init__ (self, *args, **kws):
+        super(BytecodeFlowBuilder, self).__init__(*args, **kws)
+        om_items = opcode_util.OPCODE_MAP.items()
+        self.opmap = dict((opcode.opmap[opname], (opname, pops, pushes, stmt))
+                          for opname, (pops, pushes, stmt) in om_items
+                          if opname in opcode.opmap)
+
+    def _visit_op (self, i, op, arg, opname, pops, pushes, appends):
+        assert pops is not None, ('%s not well defined in opcode_util.'
+                                  'OPCODE_MAP' % opname)
+        if pops:
+            if pops < 0:
+                pops = arg - pops - 1
+            assert pops <= len(self.stack), ("Stack underflow at instruction "
+                                             "%d (%s)!" % (i, opname))
+            stk_args = self.stack[-pops:]
+            del self.stack[-pops:]
+        else:
+            stk_args = []
+        ret_val = (i, op, opname, arg, stk_args)
+        if pushes:
+            self.stack.append(ret_val)
+        if appends:
+            self.block.append(ret_val)
+        return ret_val
+
+    def _op (self, i, op, arg):
+        opname, pops, pushes, appends = self.opmap[op]
+        return self._visit_op(i, op, arg, opname, pops, pushes, appends)
+
+    def visit_cfg (self, cfg):
+        self.cfg = cfg
+        ret_val = self.visit(cfg.blocks)
+        del self.cfg
+        return ret_val
+
+    def enter_blocks (self, blocks):
+        labels = list(blocks.keys())
+        labels.sort()
+        self.blocks = dict((index, [])
+                           for index in labels)
+        self.loop_stack = []
+        self.stacks = {}
+
+    def exit_blocks (self, blocks):
+        ret_val = self.blocks
+        del self.stacks
+        del self.loop_stack
+        del self.blocks
+        return ret_val
+
+    def enter_block (self, block):
+        self.block_no = block
+        self.block = self.blocks[block]
+        in_blocks = self.cfg.blocks_in[block]
+        if len(in_blocks) == 0:
+            self.stack = []
+        else:
+            pred_stack = None
+            for pred in in_blocks:
+                if pred in self.stacks:
+                    pred_stack = self.stacks[pred]
+                    break
+            if pred_stack is not None:
+                self.stack = pred_stack[:]
+            else:
+                raise NotImplementedError()
+
+    def exit_block (self, block):
+        assert self.block_no == block
+        self.stacks[block] = self.stack
+        del self.stack
+        del self.block
+        del self.block_no
+
+    op_BINARY_ADD = _op
+    op_BINARY_AND = _op
+    op_BINARY_DIVIDE = _op
+    op_BINARY_FLOOR_DIVIDE = _op
+    op_BINARY_LSHIFT = _op
+    op_BINARY_MODULO = _op
+    op_BINARY_MULTIPLY = _op
+    op_BINARY_OR = _op
+    op_BINARY_POWER = _op
+    op_BINARY_RSHIFT = _op
+    op_BINARY_SUBSCR = _op
+    op_BINARY_SUBTRACT = _op
+    op_BINARY_TRUE_DIVIDE = _op
+    op_BINARY_XOR = _op
+
+    def op_BREAK_LOOP (self, i, op, arg):
+        loop_i, _, loop_arg = self.loop_stack[-1]
+        assert arg is None
+        return self._op(i, op, loop_i + loop_arg + 3)
+
+    #op_BUILD_CLASS = _op
+    op_BUILD_LIST = _op
+    op_BUILD_MAP = _op
+    op_BUILD_SLICE = _op
+    op_BUILD_TUPLE = _op
+    op_CALL_FUNCTION = _op
+    op_CALL_FUNCTION_KW = _op
+    op_CALL_FUNCTION_VAR = _op
+    op_CALL_FUNCTION_VAR_KW = _op
+    op_COMPARE_OP = _op
+    #op_CONTINUE_LOOP = _op
+    op_DELETE_ATTR = _op
+    op_DELETE_FAST = _op
+    op_DELETE_GLOBAL = _op
+    op_DELETE_NAME = _op
+    op_DELETE_SLICE = _op
+    op_DELETE_SUBSCR = _op
+
+    def op_DUP_TOP (self, i, op, arg):
+        self.stack.append(self.stack[-1])
+
+    def op_DUP_TOPX (self, i, op, arg):
+        self.stack += self.stack[-arg:]
+
+    #op_END_FINALLY = _op
+    op_EXEC_STMT = _op
+    #op_EXTENDED_ARG = _op
+    op_FOR_ITER = _op
+    op_GET_ITER = _op
+    op_IMPORT_FROM = _op
+    op_IMPORT_NAME = _op
+    op_IMPORT_STAR = _op
+    op_INPLACE_ADD = _op
+    op_INPLACE_AND = _op
+    op_INPLACE_DIVIDE = _op
+    op_INPLACE_FLOOR_DIVIDE = _op
+    op_INPLACE_LSHIFT = _op
+    op_INPLACE_MODULO = _op
+    op_INPLACE_MULTIPLY = _op
+    op_INPLACE_OR = _op
+    op_INPLACE_POWER = _op
+    op_INPLACE_RSHIFT = _op
+    op_INPLACE_SUBTRACT = _op
+    op_INPLACE_TRUE_DIVIDE = _op
+    op_INPLACE_XOR = _op
+    op_JUMP_ABSOLUTE = _op
+    op_JUMP_FORWARD = _op
+
+    def op_JUMP_IF_FALSE (self, i, op, arg):
+        opname, _, _, _ = self.opmap[op]
+        ret_val = (i, op, opname, arg, [self.stack[-1]])
+        self.block.append(ret_val)
+        return ret_val
+
+    op_JUMP_IF_TRUE = op_JUMP_IF_FALSE
+    op_LIST_APPEND = _op
+    op_LOAD_ATTR = _op
+    op_LOAD_CLOSURE = _op
+    op_LOAD_CONST = _op
+    op_LOAD_DEREF = _op
+    op_LOAD_FAST = _op
+    op_LOAD_GLOBAL = _op
+    op_LOAD_LOCALS = _op
+    op_LOAD_NAME = _op
+    op_MAKE_CLOSURE = _op
+    op_MAKE_FUNCTION = _op
+    op_NOP = _op
+
+    def op_POP_BLOCK (self, i, op, arg):
+        self.loop_stack.pop()
+        return self._op(i, op, arg)
+
+    op_POP_JUMP_IF_FALSE = _op
+    op_POP_JUMP_IF_TRUE = _op
+    op_POP_TOP = _op
+    op_PRINT_EXPR = _op
+    op_PRINT_ITEM = _op
+    op_PRINT_ITEM_TO = _op
+    op_PRINT_NEWLINE = _op
+    op_PRINT_NEWLINE_TO = _op
+    op_RAISE_VARARGS = _op
+    op_RETURN_VALUE = _op
+
+    def op_ROT_FOUR (self, i, op, arg):
+        self.stack[-4:] = (self.stack[-1], self.stack[-4], self.stack[-3],
+                           self.stack[-2])
+
+    def op_ROT_THREE (self, i, op, arg):
+        self.stack[-3:] = (self.stack[-1], self.stack[-3], self.stack[-2])
+
+    def op_ROT_TWO (self, i, op, arg):
+        self.stack[-2:] = (self.stack[-1], self.stack[-2])
+
+    #op_SETUP_EXCEPT = _op
+    #op_SETUP_FINALLY = _op
+
+    def op_SETUP_LOOP (self, i, op, arg):
+        self.loop_stack.append((i, op, arg))
+        self.block.append((i, op, self.opnames[op], arg, []))
+
+    op_SLICE = _op
+    #op_STOP_CODE = _op
+    op_STORE_ATTR = _op
+    op_STORE_DEREF = _op
+    op_STORE_FAST = _op
+    op_STORE_GLOBAL = _op
+    op_STORE_MAP = _op
+    op_STORE_NAME = _op
+    op_STORE_SLICE = _op
+    op_STORE_SUBSCR = _op
+    op_UNARY_CONVERT = _op
+    op_UNARY_INVERT = _op
+    op_UNARY_NEGATIVE = _op
+    op_UNARY_NOT = _op
+    op_UNARY_POSITIVE = _op
+    op_UNPACK_SEQUENCE = _op
+    #op_WITH_CLEANUP = _op
+    op_YIELD_VALUE = _op
+
+# ______________________________________________________________________
+
+def build_flow (func):
+    '''Given a Python function, return a bytecode flow tree for that
+    function.'''
+    import byte_control
+    cfg = byte_control.build_cfg(func)
+    return BytecodeFlowBuilder().visit_cfg(cfg)
+
+# ______________________________________________________________________
+# Main (self-test) routine
+
+def main (*args):
+    import pprint
+    from tests import llfuncs
+    if not args:
+        args = ('doslice',)
+    for arg in args:
+        pprint.pprint(build_flow(getattr(llfuncs, arg)))
+
+# ______________________________________________________________________
+
+if __name__ == '__main__':
+    import sys
+    main(*sys.argv[1:])
+
+# ______________________________________________________________________
+# End of byte_flow.py

llpython/byte_translator.py

@@ -0,0 +1,612 @@
+# ______________________________________________________________________
+'''Defines a bytecode based LLVM translator for llpython code.
+'''
+# ______________________________________________________________________
+# Module imports
+from __future__ import absolute_import
+import opcode
+import types
+import logging
+
+import llvm.core as lc
+
+from . import opcode_util
+from . import bytetype
+from .bytecode_visitor import BytecodeFlowVisitor
+from .byte_flow import BytecodeFlowBuilder
+from .byte_control import ControlFlowBuilder
+from .phi_injector import PhiInjector, synthetic_opname
+
+# ______________________________________________________________________
+# Module data
+
+logger = logging.getLogger(__name__)
+
+# XXX Stolen from numba.translate:
+
+_compare_mapping_float = {'>':lc.FCMP_OGT,
+                           '<':lc.FCMP_OLT,
+                           '==':lc.FCMP_OEQ,
+                           '>=':lc.FCMP_OGE,
+                           '<=':lc.FCMP_OLE,
+                           '!=':lc.FCMP_ONE}
+
+_compare_mapping_sint = {'>':lc.ICMP_SGT,
+                          '<':lc.ICMP_SLT,
+                          '==':lc.ICMP_EQ,
+                          '>=':lc.ICMP_SGE,
+                          '<=':lc.ICMP_SLE,
+                          '!=':lc.ICMP_NE}
+
+# XXX Stolen from numba.llvm_types:
+
+class LLVMCaster (object):
+    def build_pointer_cast(_, builder, lval1, lty2):
+        return builder.bitcast(lval1, lty2)
+
+    def build_int_cast(_, builder, lval1, lty2, unsigned = False):
+        width1 = lval1.type.width
+        width2 = lty2.width
+        ret_val = lval1
+        if width2 > width1:
+            if unsigned:
+                ret_val = builder.zext(lval1, lty2)
+            else:
+                ret_val = builder.sext(lval1, lty2)
+        elif width2 < width1:
+            ret_val = builder.trunc(lval1, lty2)
+        return ret_val
+
+    def build_float_ext(_, builder, lval1, lty2):
+        return builder.fpext(lval1, lty2)
+
+    def build_float_trunc(_, builder, lval1, lty2):
+        return builder.fptrunc(lval1, lty2)
+
+    def build_int_to_float_cast(_, builder, lval1, lty2, unsigned = False):
+        ret_val = None
+        if unsigned:
+            ret_val = builder.uitofp(lval1, lty2)
+        else:
+            ret_val = builder.sitofp(lval1, lty2)
+        return ret_val
+
+    def build_int_to_ptr_cast(_, builder, lval1, lty2):
+        return builder.inttoptr(lval1, lty2)
+
+    def build_float_to_int_cast(_, builder, lval1, lty2, unsigned = False):
+        ret_val = None
+        if unsigned:
+            ret_val = builder.fptoui(lval1, lty2)
+        else:
+            ret_val = builder.fptosi(lval1, lty2)
+        return ret_val
+
+    CAST_MAP = {
+        lc.TYPE_POINTER : build_pointer_cast,
+        lc.TYPE_INTEGER: build_int_cast,
+        (lc.TYPE_FLOAT, lc.TYPE_DOUBLE) : build_float_ext,
+        (lc.TYPE_DOUBLE, lc.TYPE_FLOAT) : build_float_trunc,
+        (lc.TYPE_INTEGER, lc.TYPE_FLOAT) : build_int_to_float_cast,
+        (lc.TYPE_INTEGER, lc.TYPE_DOUBLE) : build_int_to_float_cast,
+        (lc.TYPE_INTEGER, lc.TYPE_POINTER) : build_int_to_ptr_cast,
+        (lc.TYPE_FLOAT, lc.TYPE_INTEGER) : build_float_to_int_cast,
+        (lc.TYPE_DOUBLE, lc.TYPE_INTEGER) : build_float_to_int_cast,
+
+    }
+
+    @classmethod
+    def build_cast(cls, builder, lval1, lty2, *args, **kws):
+        ret_val = lval1
+        lty1 = lval1.type
+        lkind1 = lty1.kind
+        lkind2 = lty2.kind
+
+        if lkind1 == lkind2:
+
+            if lkind1 in cls.CAST_MAP:
+                ret_val = cls.CAST_MAP[lkind1](cls, builder, lval1, lty2,
+                                               *args, **kws)
+            else:
+                raise NotImplementedError(lkind1)
+        else:
+            map_index = (lkind1, lkind2)
+            if map_index in cls.CAST_MAP:
+                ret_val = cls.CAST_MAP[map_index](cls, builder, lval1, lty2,
+                                                  *args, **kws)
+            else:
+                raise NotImplementedError(lkind1, lkind2)
+        return ret_val
+
+# ______________________________________________________________________
+# Class definitions
+
+class LLVMTranslator (BytecodeFlowVisitor):
+    '''Transformer responsible for visiting a set of bytecode flow
+    trees, emitting LLVM code.
+
+    Unlike other translators in :py:mod:`llpython`, this
+    incorporates the full transformation chain, starting with
+    :py:class:`llpython.byte_flow.BytecodeFlowBuilder`, then
+    :py:class:`llpython.byte_control.ControlFlowBuilder`, and
+    then :py:class:`llpython.phi_injector.PhiInjector`.'''
+
+    def __init__ (self, llvm_module = None, *args, **kws):
+        '''Constructor for LLVMTranslator.'''
+        super(LLVMTranslator, self).__init__(*args, **kws)
+        if llvm_module is None:
+            llvm_module = lc.Module.new('Translated_Module_%d' % (id(self),))
+        self.llvm_module = llvm_module
+        self.bytecode_flow_builder = BytecodeFlowBuilder()
+        self.control_flow_builder = ControlFlowBuilder()
+        self.phi_injector = PhiInjector()
+
+    def translate (self, function, llvm_type = None, llvm_function = None,
+                   env = None):
+        '''Translate a function to the given LLVM function type.
+
+        If no type is given, then assume the function is of LLVM type
+        "void ()".
+
+        The optional env parameter allows extension of the global
+        environment.'''
+        if llvm_type is None:
+            if llvm_function is None:
+                llvm_type = lc.Type.function(bytetype.lvoid, ())
+            else:
+                llvm_type = llvm_function.type.pointee
+        if env is None:
+            env = {}
+        else:
+            env = env.copy()
+        env.update((name, method)
+                   for name, method in lc.Builder.__dict__.items()
+                   if not name.startswith('_'))
+        env.update((name, value)
+                   for name, value in bytetype.__dict__.items()
+                   if not name.startswith('_'))
+        self.loop_stack = []
+        self.llvm_type = llvm_type
+        self.target_function_name = env.get('target_function_name',
+                                            function.__name__)
+        self.function = function
+        self.code_obj = opcode_util.get_code_object(function)
+        func_globals = getattr(function, 'func_globals',
+                               getattr(function, '__globals__', {})).copy()
+        func_globals.update(env)
+        self.globals = func_globals
+        nargs = self.code_obj.co_argcount
+        self.cfg = self.control_flow_builder.visit(
+            opcode_util.build_basic_blocks(self.code_obj), nargs)
+        self.cfg.blocks = self.bytecode_flow_builder.visit_cfg(self.cfg)
+        self.llvm_function = llvm_function
+        flow = self.phi_injector.visit_cfg(self.cfg, nargs)
+        ret_val = self.visit(flow)
+        del self.cfg
+        del self.globals
+        del self.code_obj
+        del self.target_function_name
+        del self.function
+        del self.llvm_type
+        del self.loop_stack
+        return ret_val
+
+    def enter_flow_object (self, flow):
+        super(LLVMTranslator, self).enter_flow_object(flow)
+        if self.llvm_function is None:
+            self.llvm_function = self.llvm_module.add_function(
+                self.llvm_type, self.target_function_name)
+        self.llvm_blocks = {}
+        self.llvm_definitions = {}
+        self.pending_phis = {}
+        for block in self.block_list:
+            if 0 in self.cfg.blocks_reaching[block]:
+                bb = self.llvm_function.append_basic_block(
+                    'BLOCK_%d' % (block,))
+                self.llvm_blocks[block] = bb
+
+    def exit_flow_object (self, flow):
+        super(LLVMTranslator, self).exit_flow_object(flow)
+        ret_val = self.llvm_function
+        del self.pending_phis
+        del self.llvm_definitions
+        del self.llvm_blocks
+        if __debug__ and logger.getEffectiveLevel() < logging.DEBUG:
+            logger.debug(str(ret_val))
+        return ret_val
+
+    def enter_block (self, block):
+        ret_val = False
+        if block in self.llvm_blocks:
+            self.llvm_block = self.llvm_blocks[block]
+            self.builder = lc.Builder.new(self.llvm_block)
+            ret_val = True
+        return ret_val
+
+    def exit_block (self, block):
+        bb_instrs = self.llvm_block.instructions
+        if ((len(bb_instrs) == 0) or
+                (not bb_instrs[-1].is_terminator)):
+            out_blocks = list(self.cfg.blocks_out[block])
+            assert len(out_blocks) == 1
+            self.builder.branch(self.llvm_blocks[out_blocks[0]])
+        del self.llvm_block
+        del self.builder
+
+    def visit_synthetic_op (self, i, op, arg, *args, **kws):
+        method = getattr(self, 'op_%s' % (synthetic_opname[op],))
+        return method(i, op, arg, *args, **kws)
+
+    def op_REF_ARG (self, i, op, arg, *args, **kws):
+        return [self.llvm_function.args[arg]]
+
+    def op_BUILD_PHI (self, i, op, arg, *args, **kws):
+        phi_type = None
+        incoming = []
+        pending = []
+        for child_arg in arg:
+            child_block, _, child_opname, child_arg, _ = child_arg
+            assert child_opname == 'REF_DEF'
+            if child_arg in self.llvm_definitions:
+                child_def = self.llvm_definitions[child_arg]
+                if phi_type is None:
+                    phi_type = child_def.type
+                incoming.append((child_block, child_def))
+            else:
+                pending.append((child_arg, child_block))
+        phi = self.builder.phi(phi_type)
+        for block_index, defn in incoming:
+            phi.add_incoming(defn, self.llvm_blocks[block_index])
+        for defn_index, block_index in pending:
+            if defn_index not in self.pending_phis:
+                self.pending_phis[defn_index] = []
+            self.pending_phis[defn_index].append((phi, block_index))
+        return [phi]
+
+    def op_DEFINITION (self, i, op, def_index, *args, **kws):
+        assert len(args) == 1
+        arg = args[0]
+        if def_index in self.pending_phis:
+            for phi, block_index in self.pending_phis[def_index]:
+                phi.add_incoming(arg, self.llvm_blocks[block_index])
+        self.llvm_definitions[def_index] = arg
+        return args
+
+    def op_REF_DEF (self, i, op, arg, *args, **kws):
+        return [self.llvm_definitions[arg]]
+
+    def op_BINARY_ADD (self, i, op, arg, *args, **kws):
+        arg1, arg2 = args
+        if arg1.type.kind == lc.TYPE_INTEGER:
+            ret_val = [self.builder.add(arg1, arg2)]
+        elif arg1.type.kind in (lc.TYPE_FLOAT, lc.TYPE_DOUBLE):
+            ret_val = [self.builder.fadd(arg1, arg2)]
+        elif arg1.type.kind == lc.TYPE_POINTER:
+            ret_val = [self.builder.gep(arg1, [arg2])]
+        else:
+            raise NotImplementedError("LLVMTranslator.op_BINARY_ADD for %r" %
+                                      (args,))
+        return ret_val
+
+    def op_BINARY_AND (self, i, op, arg, *args, **kws):
+        return [self.builder.and_(args[0], args[1])]
+
+    def op_BINARY_DIVIDE (self, i, op, arg, *args, **kws):
+        arg1, arg2 = args
+        if arg1.type.kind == lc.TYPE_INTEGER:
+            ret_val = [self.builder.sdiv(arg1, arg2)]
+        elif arg1.type.kind in (lc.TYPE_FLOAT, lc.TYPE_DOUBLE):
+            ret_val = [self.builder.fdiv(arg1, arg2)]
+        else:
+            raise NotImplementedError("LLVMTranslator.op_BINARY_DIVIDE for %r"
+                                      % (args,))
+        return ret_val
+
+    def op_BINARY_FLOOR_DIVIDE (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_BINARY_FLOOR_DIVIDE")
+
+    def op_BINARY_LSHIFT (self, i, op, arg, *args, **kws):
+        return [self.builder.shl(args[0], args[1])]
+
+    def op_BINARY_MODULO (self, i, op, arg, *args, **kws):
+        arg1, arg2 = args
+        if arg1.type.kind == lc.TYPE_INTEGER:
+            ret_val = [self.builder.srem(arg1, arg2)]
+        elif arg1.type.kind in (lc.TYPE_FLOAT, lc.TYPE_DOUBLE):
+            ret_val = [self.builder.frem(arg1, arg2)]
+        else:
+            raise NotImplementedError("LLVMTranslator.op_BINARY_MODULO for %r"
+                                      % (args,))
+        return ret_val
+
+    def op_BINARY_MULTIPLY (self, i, op, arg, *args, **kws):
+        arg1, arg2 = args
+        if arg1.type.kind == lc.TYPE_INTEGER:
+            ret_val = [self.builder.mul(arg1, arg2)]
+        elif arg1.type.kind in (lc.TYPE_FLOAT, lc.TYPE_DOUBLE):
+            ret_val = [self.builder.fmul(arg1, arg2)]
+        else:
+            raise NotImplementedError("LLVMTranslator.op_BINARY_MULTIPLY for "
+                                      "%r" % (args,))
+        return ret_val
+
+    def op_BINARY_OR (self, i, op, arg, *args, **kws):
+        return [self.builder.or_(args[0], args[1])]
+
+    def op_BINARY_POWER (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_BINARY_POWER")
+
+    def op_BINARY_RSHIFT (self, i, op, arg, *args, **kws):
+        return [self.builder.lshr(args[0], args[1])]
+
+    def op_BINARY_SUBSCR (self, i, op, arg, *args, **kws):
+        arr_val = args[0]
+        index_vals = args[1:]
+        ret_val = gep_result = self.builder.gep(arr_val, index_vals)
+        if (gep_result.type.kind == lc.TYPE_POINTER and
+            gep_result.type.pointee.kind != lc.TYPE_POINTER):
+            ret_val = self.builder.load(gep_result)
+        return [ret_val]
+
+    def op_BINARY_SUBTRACT (self, i, op, arg, *args, **kws):
+        arg1, arg2 = args
+        if arg1.type.kind == lc.TYPE_INTEGER:
+            ret_val = [self.builder.sub(arg1, arg2)]
+        elif arg1.type.kind in (lc.TYPE_FLOAT, lc.TYPE_DOUBLE):
+            ret_val = [self.builder.fsub(arg1, arg2)]
+        else:
+            raise NotImplementedError("LLVMTranslator.op_BINARY_SUBTRACT for "
+                                      "%r" % (args,))
+        return ret_val
+
+    op_BINARY_TRUE_DIVIDE = op_BINARY_DIVIDE
+
+    def op_BINARY_XOR (self, i, op, arg, *args, **kws):
+        return [self.builder.xor(args[0], args[1])]
+
+    def op_BREAK_LOOP (self, i, op, arg, *args, **kws):
+        return [self.builder.branch(self.llvm_blocks[arg])]
+
+    def op_BUILD_SLICE (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_BUILD_SLICE")
+
+    def op_BUILD_TUPLE (self, i, op, arg, *args, **kws):
+        return args
+
+    def op_CALL_FUNCTION (self, i, op, arg, *args, **kws):
+        fn = args[0]
+        args = args[1:]
+        fn_name = getattr(fn, '__name__', None)
+        if isinstance(fn, (types.FunctionType, types.MethodType)):
+            ret_val = [fn(self.builder, *args)]
+        elif isinstance(fn, lc.Value):
+            ret_val = [self.builder.call(fn, args)]
+        elif isinstance(fn, lc.Type):
+            if isinstance(fn, lc.FunctionType):
+                ret_val = [self.builder.call(
+                        self.llvm_module.get_or_insert_function(fn, fn_name),
+                        args)]
+            else:
+                assert len(args) == 1
+                ret_val = [LLVMCaster.build_cast(self.builder, args[0], fn)]
+        else:
+            raise NotImplementedError("Don't know how to call %s() (%r @ %d)!"
+                                      % (fn_name, fn, i))
+        return ret_val
+
+    def op_CALL_FUNCTION_KW (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_CALL_FUNCTION_KW")
+
+    def op_CALL_FUNCTION_VAR (self, i, op, arg, *args, **kws):
+        args = list(args)
+        var_args = list(args.pop())
+        args.extend(var_args)
+        return self.op_CALL_FUNCTION(i, op, arg, *args, **kws)
+
+    def op_CALL_FUNCTION_VAR_KW (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_CALL_FUNCTION_VAR_KW")
+
+    def op_COMPARE_OP (self, i, op, arg, *args, **kws):
+        arg1, arg2 = args
+        cmp_kind = opcode.cmp_op[arg]
+        if isinstance(arg1.type, lc.IntegerType):
+            ret_val = [self.builder.icmp(_compare_mapping_sint[cmp_kind],
+                                         arg1, arg2)]
+        elif arg1.type.kind in (lc.TYPE_FLOAT, lc.TYPE_DOUBLE):
+            ret_val = [self.builder.fcmp(_compare_mapping_float[cmp_kind],
+                                         arg1, arg2)]
+        else:
+            raise NotImplementedError('Comparison of type %r' % (arg1.type,))
+        return ret_val
+
+    def op_CONTINUE_LOOP (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_CONTINUE_LOOP")
+
+    def op_DELETE_ATTR (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_DELETE_ATTR")
+
+    def op_DELETE_SLICE (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_DELETE_SLICE")
+
+    def op_FOR_ITER (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_FOR_ITER")
+
+    def op_GET_ITER (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_GET_ITER")
+
+    op_INPLACE_ADD = op_BINARY_ADD
+    op_INPLACE_AND = op_BINARY_AND
+    op_INPLACE_DIVIDE = op_BINARY_DIVIDE
+    op_INPLACE_FLOOR_DIVIDE = op_BINARY_FLOOR_DIVIDE
+    op_INPLACE_LSHIFT = op_BINARY_LSHIFT
+    op_INPLACE_MODULO = op_BINARY_MODULO
+    op_INPLACE_MULTIPLY = op_BINARY_MULTIPLY
+    op_INPLACE_OR = op_BINARY_OR
+    op_INPLACE_POWER = op_BINARY_POWER
+    op_INPLACE_RSHIFT = op_BINARY_RSHIFT
+    op_INPLACE_SUBTRACT = op_BINARY_SUBTRACT
+    op_INPLACE_TRUE_DIVIDE = op_BINARY_TRUE_DIVIDE
+    op_INPLACE_XOR = op_BINARY_XOR
+
+    def op_JUMP_ABSOLUTE (self, i, op, arg, *args, **kws):
+        return [self.builder.branch(self.llvm_blocks[arg])]
+
+    def op_JUMP_FORWARD (self, i, op, arg, *args, **kws):
+        return [self.builder.branch(self.llvm_blocks[i + arg + 3])]
+
+    def op_JUMP_IF_FALSE (self, i, op, arg, *args, **kws):
+        cond = args[0]
+        block_false = self.llvm_blocks[i + 3 + arg]
+        block_true = self.llvm_blocks[i + 3]
+        return [self.builder.cbranch(cond, block_true, block_false)]
+        # raise NotImplementedError("LLVMTranslator.op_JUMP_IF_FALSE")
+
+    def op_JUMP_IF_FALSE_OR_POP (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_JUMP_IF_FALSE_OR_POP")
+
+    def op_JUMP_IF_TRUE (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_JUMP_IF_TRUE")
+
+    def op_JUMP_IF_TRUE_OR_POP (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_JUMP_IF_TRUE_OR_POP")
+
+    def op_LOAD_ATTR (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_LOAD_ATTR")
+
+    def op_LOAD_CONST (self, i, op, arg, *args, **kws):
+        py_val = self.code_obj.co_consts[arg]
+        if isinstance(py_val, int):
+            ret_val = [lc.Constant.int(bytetype.lc_int, py_val)]
+        elif isinstance(py_val, float):
+            ret_val = [lc.Constant.double(py_val)]
+        elif py_val == None:
+            ret_val = [None]
+        else:
+            raise NotImplementedError('Constant converstion for %r' %
+                                      (py_val,))
+        return ret_val
+
+    def op_LOAD_DEREF (self, i, op, arg, *args, **kws):
+        name = self.code_obj.co_freevars[arg]
+        ret_val = self.globals[name]
+        if isinstance(ret_val, lc.Type) and not hasattr(ret_val, '__name__'):
+            ret_val.__name__ = name
+        return [ret_val]
+
+    def op_LOAD_GLOBAL (self, i, op, arg, *args, **kws):
+        name = self.code_obj.co_names[arg]
+        ret_val = self.globals[name]
+        if isinstance(ret_val, lc.Type) and not hasattr(ret_val, '__name__'):
+            ret_val.__name__ = name
+        return [ret_val]
+
+    def op_POP_BLOCK (self, i, op, arg, *args, **kws):
+        self.loop_stack.pop()
+        return [self.builder.branch(self.llvm_blocks[i + 1])]
+
+    def op_POP_JUMP_IF_FALSE (self, i, op, arg, *args, **kws):
+        return [self.builder.cbranch(args[0], self.llvm_blocks[i + 3],
+                                     self.llvm_blocks[arg])]
+
+    def op_POP_JUMP_IF_TRUE (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_POP_JUMP_IF_TRUE")
+
+    def op_POP_TOP (self, i, op, arg, *args, **kws):
+        return args
+
+    def op_RETURN_VALUE (self, i, op, arg, *args, **kws):
+        if args[0] is None:
+            ret_val = [self.builder.ret_void()]
+        else:
+            ret_val = [self.builder.ret(args[0])]
+        return ret_val
+
+    def op_SETUP_LOOP (self, i, op, arg, *args, **kws):
+        self.loop_stack.append((i, arg))
+        return [self.builder.branch(self.llvm_blocks[i + 3])]
+
+    def op_SLICE (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_SLICE")
+
+    def op_STORE_ATTR (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_STORE_ATTR")
+
+    def op_STORE_SLICE (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_STORE_SLICE")
+
+    def op_STORE_SUBSCR (self, i, op, arg, *args, **kws):
+        store_val, arr_val, index_val = args
+        dest_addr = self.builder.gep(arr_val, [index_val])
+        return [self.builder.store(store_val, dest_addr)]
+
+    def op_UNARY_CONVERT (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_UNARY_CONVERT")
+
+    def op_UNARY_INVERT (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_UNARY_INVERT")
+
+    def op_UNARY_NEGATIVE (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_UNARY_NEGATIVE")
+
+    def op_UNARY_NOT (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_UNARY_NOT")
+
+    def op_UNARY_POSITIVE (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("LLVMTranslator.op_UNARY_POSITIVE")
+
+# ______________________________________________________________________
+
+def translate_function (func, lltype, llvm_module = None, **kws):
+    '''Given a function and an LLVM function type, emit LLVM code for
+    that function using a new LLVMTranslator instance.'''
+    translator = LLVMTranslator(llvm_module)
+    ret_val = translator.translate(func, lltype, env = kws)
+    return ret_val
+
+# ______________________________________________________________________
+
+def translate_into_function (py_function, llvm_function, **kws):
+    translator = LLVMTranslator(llvm_function.module)
+    ret_val = translator.translate(py_function, llvm_function = llvm_function,
+                                   env = kws)
+    return ret_val
+
+# ______________________________________________________________________
+
+def llpython (lltype, llvm_module = None, **kws):
+    '''Decorator version of translate_function().'''
+    def _llpython (func):
+        return translate_function(func, lltype, llvm_module, **kws)
+    return _llpython
+
+# ______________________________________________________________________
+
+def llpython_into (llvm_function, **kws):
+    def _llpython_into (func):
+        return translate_into_function(llvm_function, func, **kws)
+    return _llpython_into
+
+# ______________________________________________________________________
+# Main (self-test) routine
+
+def main (*args):
+    from tests import llfuncs, llfunctys
+    if not args:
+        args = ('doslice',)
+    elif 'all' in args:
+        args = [llfunc
+                for llfunc in dir(llfuncs) if not llfunc.startswith('_')]
+    llvm_module = lc.Module.new('test_module')
+    for arg in args:
+        translate_function(getattr(llfuncs, arg), getattr(llfunctys, arg),
+                           llvm_module)
+    print(llvm_module)
+
+# ______________________________________________________________________
+
+if __name__ == '__main__':
+    import sys
+    main(*sys.argv[1:])
+
+# ______________________________________________________________________
+# End of byte_translator.py

llpython/bytecode_visitor.py

@@ -0,0 +1,360 @@
+# ______________________________________________________________________
+from __future__ import absolute_import
+import itertools
+
+import opcode
+from .opcode_util import itercode
+
+# ______________________________________________________________________
+
+class BytecodeVisitor (object):
+    opnames = [name.split('+')[0] for name in opcode.opname]
+
+    def visit_op (self, i, op, arg, *args, **kws):
+        if op < 0:
+            ret_val = self.visit_synthetic_op(i, op, arg, *args, **kws)
+        else:
+            method = getattr(self, 'op_' + self.opnames[op])
+            ret_val = method(i, op, arg, *args, **kws)
+        return ret_val
+
+    def visit_synthetic_op (self, i, op, arg, *args, **kws):
+        raise NotImplementedError(
+            'BytecodeVisitor.visit_synthetic_op() must be overloaded if using '
+            'synthetic opcodes.')
+
+    def _not_implemented (self, i, op, arg, *args, **kws):
+        raise NotImplementedError("BytecodeVisitor.op_%s (@ bytecode index %d)"
+                                  % (self.opnames[op], i))
+
+    op_BINARY_ADD = _not_implemented
+    op_BINARY_AND = _not_implemented
+    op_BINARY_DIVIDE = _not_implemented
+    op_BINARY_FLOOR_DIVIDE = _not_implemented
+    op_BINARY_LSHIFT = _not_implemented
+    op_BINARY_MODULO = _not_implemented
+    op_BINARY_MULTIPLY = _not_implemented
+    op_BINARY_OR = _not_implemented
+    op_BINARY_POWER = _not_implemented
+    op_BINARY_RSHIFT = _not_implemented
+    op_BINARY_SUBSCR = _not_implemented
+    op_BINARY_SUBTRACT = _not_implemented
+    op_BINARY_TRUE_DIVIDE = _not_implemented
+    op_BINARY_XOR = _not_implemented
+    op_BREAK_LOOP = _not_implemented
+    op_BUILD_CLASS = _not_implemented
+    op_BUILD_LIST = _not_implemented
+    op_BUILD_MAP = _not_implemented
+    op_BUILD_SET = _not_implemented
+    op_BUILD_SLICE = _not_implemented
+    op_BUILD_TUPLE = _not_implemented
+    op_CALL_FUNCTION = _not_implemented
+    op_CALL_FUNCTION_KW = _not_implemented
+    op_CALL_FUNCTION_VAR = _not_implemented
+    op_CALL_FUNCTION_VAR_KW = _not_implemented
+    op_COMPARE_OP = _not_implemented
+    op_CONTINUE_LOOP = _not_implemented
+    op_DELETE_ATTR = _not_implemented
+    op_DELETE_DEREF = _not_implemented
+    op_DELETE_FAST = _not_implemented
+    op_DELETE_GLOBAL = _not_implemented
+    op_DELETE_NAME = _not_implemented
+    op_DELETE_SLICE = _not_implemented
+    op_DELETE_SUBSCR = _not_implemented
+    op_DUP_TOP = _not_implemented
+    op_DUP_TOPX = _not_implemented
+    op_DUP_TOP_TWO = _not_implemented
+    op_END_FINALLY = _not_implemented
+    op_EXEC_STMT = _not_implemented
+    op_EXTENDED_ARG = _not_implemented
+    op_FOR_ITER = _not_implemented
+    op_GET_ITER = _not_implemented
+    op_IMPORT_FROM = _not_implemented
+    op_IMPORT_NAME = _not_implemented
+    op_IMPORT_STAR = _not_implemented
+    op_INPLACE_ADD = _not_implemented
+    op_INPLACE_AND = _not_implemented
+    op_INPLACE_DIVIDE = _not_implemented
+    op_INPLACE_FLOOR_DIVIDE = _not_implemented
+    op_INPLACE_LSHIFT = _not_implemented
+    op_INPLACE_MODULO = _not_implemented
+    op_INPLACE_MULTIPLY = _not_implemented
+    op_INPLACE_OR = _not_implemented
+    op_INPLACE_POWER = _not_implemented
+    op_INPLACE_RSHIFT = _not_implemented
+    op_INPLACE_SUBTRACT = _not_implemented
+    op_INPLACE_TRUE_DIVIDE = _not_implemented
+    op_INPLACE_XOR = _not_implemented
+    op_JUMP_ABSOLUTE = _not_implemented
+    op_JUMP_FORWARD = _not_implemented
+    op_JUMP_IF_FALSE = _not_implemented
+    op_JUMP_IF_FALSE_OR_POP = _not_implemented
+    op_JUMP_IF_TRUE = _not_implemented
+    op_JUMP_IF_TRUE_OR_POP = _not_implemented
+    op_LIST_APPEND = _not_implemented
+    op_LOAD_ATTR = _not_implemented
+    op_LOAD_BUILD_CLASS = _not_implemented
+    op_LOAD_CLOSURE = _not_implemented
+    op_LOAD_CONST = _not_implemented
+    op_LOAD_DEREF = _not_implemented
+    op_LOAD_FAST = _not_implemented
+    op_LOAD_GLOBAL = _not_implemented
+    op_LOAD_LOCALS = _not_implemented
+    op_LOAD_NAME = _not_implemented
+    op_MAKE_CLOSURE = _not_implemented
+    op_MAKE_FUNCTION = _not_implemented
+    op_MAP_ADD = _not_implemented
+    op_NOP = _not_implemented
+    op_POP_BLOCK = _not_implemented
+    op_POP_EXCEPT = _not_implemented
+    op_POP_JUMP_IF_FALSE = _not_implemented
+    op_POP_JUMP_IF_TRUE = _not_implemented
+    op_POP_TOP = _not_implemented
+    op_PRINT_EXPR = _not_implemented
+    op_PRINT_ITEM = _not_implemented
+    op_PRINT_ITEM_TO = _not_implemented
+    op_PRINT_NEWLINE = _not_implemented
+    op_PRINT_NEWLINE_TO = _not_implemented
+    op_RAISE_VARARGS = _not_implemented
+    op_RETURN_VALUE = _not_implemented
+    op_ROT_FOUR = _not_implemented
+    op_ROT_THREE = _not_implemented
+    op_ROT_TWO = _not_implemented
+    op_SETUP_EXCEPT = _not_implemented
+    op_SETUP_FINALLY = _not_implemented
+    op_SETUP_LOOP = _not_implemented
+    op_SETUP_WITH = _not_implemented
+    op_SET_ADD = _not_implemented
+    op_SLICE = _not_implemented
+    op_STOP_CODE = _not_implemented
+    op_STORE_ATTR = _not_implemented
+    op_STORE_DEREF = _not_implemented
+    op_STORE_FAST = _not_implemented
+    op_STORE_GLOBAL = _not_implemented
+    op_STORE_LOCALS = _not_implemented
+    op_STORE_MAP = _not_implemented
+    op_STORE_NAME = _not_implemented
+    op_STORE_SLICE = _not_implemented
+    op_STORE_SUBSCR = _not_implemented
+    op_UNARY_CONVERT = _not_implemented
+    op_UNARY_INVERT = _not_implemented
+    op_UNARY_NEGATIVE = _not_implemented
+    op_UNARY_NOT = _not_implemented
+    op_UNARY_POSITIVE = _not_implemented
+    op_UNPACK_EX = _not_implemented
+    op_UNPACK_SEQUENCE = _not_implemented
+    op_WITH_CLEANUP = _not_implemented
+    op_YIELD_VALUE = _not_implemented
+
+# ______________________________________________________________________
+
+class BytecodeIterVisitor (BytecodeVisitor):
+    def visit (self, co_obj):
+        self.enter_code_object(co_obj)
+        for i, op, arg in itercode(co_obj.co_code):
+            self.visit_op(i, op, arg)
+        return self.exit_code_object(co_obj)
+
+    def enter_code_object (self, co_obj):
+        pass
+
+    def exit_code_object (self, co_obj):
+        pass
+
+# ______________________________________________________________________
+
+class BasicBlockVisitor (BytecodeVisitor):
+    def visit (self, blocks):
+        self.enter_blocks(blocks)
+        block_indices = list(blocks.keys())
+        block_indices.sort()
+        for block_index in block_indices:
+            self.enter_block(block_index)
+            for i, op, arg in blocks[block_index]:
+                self.visit_op(i, op, arg)
+            self.exit_block(block_index)
+        return self.exit_blocks(blocks)
+
+    def enter_blocks (self, blocks):
+        pass
+
+    def exit_blocks (self, blocks):
+        pass
+
+    def enter_block (self, block_index):
+        pass
+
+    def exit_block (self, block_index):
+        pass
+
+# ______________________________________________________________________
+
+class BytecodeFlowVisitor (BytecodeVisitor):
+    def visit (self, flow):
+        self.block_list = list(flow.keys())
+        self.block_list.sort()
+        self.enter_flow_object(flow)
+        for block in self.block_list:
+            prelude = self.enter_block(block)
+            prelude_isa_list = isinstance(prelude, list)
+            if prelude or prelude_isa_list:
+                if not prelude_isa_list:
+                    prelude = []
+                new_stmts = list(self.visit_op(i, op, arg, *args)
+                                 for i, op, _, arg, args in flow[block])
+                self.new_flow[block] = list(itertools.chain(
+                    prelude, *new_stmts))
+            self.exit_block(block)
+        del self.block_list
+        return self.exit_flow_object(flow)
+
+    def visit_op (self, i, op, arg, *args, **kws):
+        new_args = []
+        for child_i, child_op, _, child_arg, child_args in args:
+            new_args.extend(self.visit_op(child_i, child_op, child_arg,
+                                          *child_args))
+        ret_val = super(BytecodeFlowVisitor, self).visit_op(i, op, arg,
+                                                            *new_args)
+        return ret_val
+
+    def enter_flow_object (self, flow):
+        self.new_flow = {}
+
+    def exit_flow_object (self, flow):
+        ret_val = self.new_flow
+        del self.new_flow
+        return ret_val
+
+    def enter_block (self, block):
+        pass
+
+    def exit_block (self, block):
+        pass
+
+# ______________________________________________________________________
+
+class BenignBytecodeVisitorMixin (object):
+    def _do_nothing (self, i, op, arg, *args, **kws):
+        return [(i, op, self.opnames[op], arg, args)]
+
+    op_BINARY_ADD = _do_nothing
+    op_BINARY_AND = _do_nothing
+    op_BINARY_DIVIDE = _do_nothing
+    op_BINARY_FLOOR_DIVIDE = _do_nothing
+    op_BINARY_LSHIFT = _do_nothing
+    op_BINARY_MODULO = _do_nothing
+    op_BINARY_MULTIPLY = _do_nothing
+    op_BINARY_OR = _do_nothing
+    op_BINARY_POWER = _do_nothing
+    op_BINARY_RSHIFT = _do_nothing
+    op_BINARY_SUBSCR = _do_nothing
+    op_BINARY_SUBTRACT = _do_nothing
+    op_BINARY_TRUE_DIVIDE = _do_nothing
+    op_BINARY_XOR = _do_nothing
+    op_BREAK_LOOP = _do_nothing
+    op_BUILD_CLASS = _do_nothing
+    op_BUILD_LIST = _do_nothing
+    op_BUILD_MAP = _do_nothing
+    op_BUILD_SET = _do_nothing
+    op_BUILD_SLICE = _do_nothing
+    op_BUILD_TUPLE = _do_nothing
+    op_CALL_FUNCTION = _do_nothing
+    op_CALL_FUNCTION_KW = _do_nothing
+    op_CALL_FUNCTION_VAR = _do_nothing
+    op_CALL_FUNCTION_VAR_KW = _do_nothing
+    op_COMPARE_OP = _do_nothing
+    op_CONTINUE_LOOP = _do_nothing
+    op_DELETE_ATTR = _do_nothing
+    op_DELETE_DEREF = _do_nothing
+    op_DELETE_FAST = _do_nothing
+    op_DELETE_GLOBAL = _do_nothing
+    op_DELETE_NAME = _do_nothing
+    op_DELETE_SLICE = _do_nothing
+    op_DELETE_SUBSCR = _do_nothing
+    op_DUP_TOP = _do_nothing
+    op_DUP_TOPX = _do_nothing
+    op_DUP_TOP_TWO = _do_nothing
+    op_END_FINALLY = _do_nothing
+    op_EXEC_STMT = _do_nothing
+    op_EXTENDED_ARG = _do_nothing
+    op_FOR_ITER = _do_nothing
+    op_GET_ITER = _do_nothing
+    op_IMPORT_FROM = _do_nothing
+    op_IMPORT_NAME = _do_nothing
+    op_IMPORT_STAR = _do_nothing
+    op_INPLACE_ADD = _do_nothing
+    op_INPLACE_AND = _do_nothing
+    op_INPLACE_DIVIDE = _do_nothing
+    op_INPLACE_FLOOR_DIVIDE = _do_nothing
+    op_INPLACE_LSHIFT = _do_nothing
+    op_INPLACE_MODULO = _do_nothing
+    op_INPLACE_MULTIPLY = _do_nothing
+    op_INPLACE_OR = _do_nothing
+    op_INPLACE_POWER = _do_nothing
+    op_INPLACE_RSHIFT = _do_nothing
+    op_INPLACE_SUBTRACT = _do_nothing
+    op_INPLACE_TRUE_DIVIDE = _do_nothing
+    op_INPLACE_XOR = _do_nothing
+    op_JUMP_ABSOLUTE = _do_nothing
+    op_JUMP_FORWARD = _do_nothing
+    op_JUMP_IF_FALSE = _do_nothing
+    op_JUMP_IF_FALSE_OR_POP = _do_nothing
+    op_JUMP_IF_TRUE = _do_nothing
+    op_JUMP_IF_TRUE_OR_POP = _do_nothing
+    op_LIST_APPEND = _do_nothing
+    op_LOAD_ATTR = _do_nothing
+    op_LOAD_BUILD_CLASS = _do_nothing
+    op_LOAD_CLOSURE = _do_nothing
+    op_LOAD_CONST = _do_nothing
+    op_LOAD_DEREF = _do_nothing
+    op_LOAD_FAST = _do_nothing
+    op_LOAD_GLOBAL = _do_nothing
+    op_LOAD_LOCALS = _do_nothing
+    op_LOAD_NAME = _do_nothing
+    op_MAKE_CLOSURE = _do_nothing
+    op_MAKE_FUNCTION = _do_nothing
+    op_MAP_ADD = _do_nothing
+    op_NOP = _do_nothing
+    op_POP_BLOCK = _do_nothing
+    op_POP_EXCEPT = _do_nothing
+    op_POP_JUMP_IF_FALSE = _do_nothing
+    op_POP_JUMP_IF_TRUE = _do_nothing
+    op_POP_TOP = _do_nothing
+    op_PRINT_EXPR = _do_nothing
+    op_PRINT_ITEM = _do_nothing
+    op_PRINT_ITEM_TO = _do_nothing
+    op_PRINT_NEWLINE = _do_nothing
+    op_PRINT_NEWLINE_TO = _do_nothing
+    op_RAISE_VARARGS = _do_nothing
+    op_RETURN_VALUE = _do_nothing
+    op_ROT_FOUR = _do_nothing
+    op_ROT_THREE = _do_nothing
+    op_ROT_TWO = _do_nothing
+    op_SETUP_EXCEPT = _do_nothing
+    op_SETUP_FINALLY = _do_nothing
+    op_SETUP_LOOP = _do_nothing
+    op_SETUP_WITH = _do_nothing
+    op_SET_ADD = _do_nothing
+    op_SLICE = _do_nothing
+    op_STOP_CODE = _do_nothing
+    op_STORE_ATTR = _do_nothing
+    op_STORE_DEREF = _do_nothing
+    op_STORE_FAST = _do_nothing
+    op_STORE_GLOBAL = _do_nothing
+    op_STORE_LOCALS = _do_nothing
+    op_STORE_MAP = _do_nothing
+    op_STORE_NAME = _do_nothing
+    op_STORE_SLICE = _do_nothing
+    op_STORE_SUBSCR = _do_nothing
+    op_UNARY_CONVERT = _do_nothing
+    op_UNARY_INVERT = _do_nothing
+    op_UNARY_NEGATIVE = _do_nothing
+    op_UNARY_NOT = _do_nothing
+    op_UNARY_POSITIVE = _do_nothing
+    op_UNPACK_EX = _do_nothing
+    op_UNPACK_SEQUENCE = _do_nothing
+    op_WITH_CLEANUP = _do_nothing
+    op_YIELD_VALUE = _do_nothing
+
+# ______________________________________________________________________
+# End of bytecode_visitor.py

llpython/bytetype.py

@@ -0,0 +1,42 @@
+# ______________________________________________________________________
+
+import ctypes
+
+import llvm.core as lc
+
+# ______________________________________________________________________
+
+lvoid = lc.Type.void()
+li1 = lc.Type.int(1)
+li8 = lc.Type.int(8)
+li16 = lc.Type.int(16)
+li32 = lc.Type.int(32)
+li64 = lc.Type.int(64)
+liptr = lc.Type.int(ctypes.sizeof(ctypes.c_void_p) * 8)
+lc_size_t = lc.Type.int(ctypes.sizeof(
+        getattr(ctypes, 'c_ssize_t', getattr(ctypes, 'c_size_t'))) * 8)
+lfloat = lc.Type.float()
+ldouble = lc.Type.double()
+li8_ptr = lc.Type.pointer(li8)
+
+lc_int = lc.Type.int(ctypes.sizeof(ctypes.c_int) * 8)
+lc_long = lc.Type.int(ctypes.sizeof(ctypes.c_long) * 8)
+
+l_pyobject_head = [lc_size_t, lc.Type.pointer(li32)]
+l_pyobject_head_struct = lc.Type.struct(l_pyobject_head)
+l_pyobj_p = l_pyobject_head_struct_p = lc.Type.pointer(l_pyobject_head_struct)
+l_pyfunc = lc.Type.function(l_pyobj_p, (l_pyobj_p, l_pyobj_p))
+
+strlen = lc.Type.function(lc_size_t, (li8_ptr,))
+strncpy = lc.Type.function(li8_ptr, (li8_ptr, li8_ptr, lc_size_t))
+strndup = lc.Type.function(li8_ptr, (li8_ptr, lc_size_t))
+malloc = lc.Type.function(li8_ptr, (lc_size_t,))
+free = lc.Type.function(lvoid, (li8_ptr,))
+
+Py_BuildValue = lc.Type.function(l_pyobj_p, [li8_ptr], True)
+PyArg_ParseTuple = lc.Type.function(lc_int, [l_pyobj_p, li8_ptr], True)
+PyEval_SaveThread = lc.Type.function(li8_ptr, [])
+PyEval_RestoreThread = lc.Type.function(lc.Type.void(), [li8_ptr])
+
+# ______________________________________________________________________
+# End of bytetype.py

llpython/control_flow.py

@@ -0,0 +1,217 @@
+# ______________________________________________________________________
+
+import pprint
+
+# ______________________________________________________________________
+
+class ControlFlowGraph (object):
+    def __init__ (self):
+        self.blocks = {}
+        self.blocks_in = {}
+        self.blocks_out = {}
+        self.blocks_reads = {}
+        self.blocks_writes = {}
+        self.blocks_writer = {}
+        self.blocks_dom = {}
+        self.blocks_reaching = {}
+
+    def add_block (self, key, value = None):
+        self.blocks[key] = value
+        if key not in self.blocks_in:
+            self.blocks_in[key] = set()
+            self.blocks_out[key] = set()
+            self.blocks_reads[key] = set()
+            self.blocks_writes[key] = set()
+            self.blocks_writer[key] = {}
+
+    def add_edge (self, from_block, to_block):
+        self.blocks_out[from_block].add(to_block)
+        self.blocks_in[to_block].add(from_block)
+
+    def unlink_unreachables (self):
+        changed = True
+        next_blocks = self.blocks.keys()
+        next_blocks.remove(0)
+        while changed:
+            changed = False
+            blocks = next_blocks
+            next_blocks = blocks[:]
+            for block in blocks:
+                if len(self.blocks_in[block]) == 0:
+                    blocks_out = self.blocks_out[block]
+                    for out_edge in blocks_out:
+                        self.blocks_in[out_edge].discard(block)
+                    blocks_out.clear()
+                    next_blocks.remove(block)
+                    changed = True
+
+    def compute_dataflow (self):
+        '''Compute the dominator and reaching dataflow relationships
+        in the CFG.'''
+        blocks = set(self.blocks.keys())
+        nonentry_blocks = blocks.copy()
+        for block in blocks:
+            self.blocks_dom[block] = blocks
+            self.blocks_reaching[block] = set((block,))
+            if len(self.blocks_in[block]) == 0:
+                self.blocks_dom[block] = set((block,))
+                nonentry_blocks.remove(block)
+        changed = True
+        while changed:
+            changed = False
+            for block in nonentry_blocks:
+                olddom = self.blocks_dom[block]
+                newdom = set.intersection(*[self.blocks_dom[pred]
+                                            for pred in self.blocks_in[block]])
+                newdom.add(block)
+                if newdom != olddom:
+                    changed = True
+                    self.blocks_dom[block] = newdom
+                oldreaching = self.blocks_reaching[block]
+                newreaching = set.union(
+                    *[self.blocks_reaching[pred]
+                      for pred in self.blocks_in[block]])
+                newreaching.add(block)
+                if newreaching != oldreaching:
+                    changed = True
+                    self.blocks_reaching[block] = newreaching
+        return self.blocks_dom, self.blocks_reaching
+
+    def update_for_ssa (self):
+        '''Modify the blocks_writes map to reflect phi nodes inserted
+        for static single assignment representations.'''
+        joins = [block for block in self.blocks.keys()
+                 if len(self.blocks_in[block]) > 1]
+        changed = True
+        while changed:
+            changed = False
+            for block in joins:
+                phis_needed = self.phi_needed(block)
+                for affected_local in phis_needed:
+                    if affected_local not in self.blocks_writes[block]:
+                        changed = True
+                        # NOTE: For this to work, we assume that basic
+                        # blocks are indexed by their instruction
+                        # index in the VM bytecode.
+                        self.writes_local(block, block, affected_local)
+            if changed:
+                # Any modifications have invalidated the reaching
+                # definitions, so delete any memoized results.
+                if hasattr(self, 'reaching_definitions'):
+                    del self.reaching_definitions
+
+    def idom (self, block):
+        '''Compute the immediate dominator (idom) of the given block
+        key.  Returns None if the block has no in edges.
+
+        Note that in the case where there are multiple immediate
+        dominators (a join after a non-loop branch), this returns one
+        of the predecessors, but is not guaranteed to reliably select
+        one over the others (depends on the ordering of the set type
+        iterator).'''
+        preds = self.blocks_in[block]
+        npreds = len(preds)
+        if npreds == 0:
+            ret_val = None
+        elif npreds == 1:
+            ret_val = tuple(preds)[0]
+        else:
+            ret_val = [pred for pred in preds
+                       if block not in self.blocks_dom[pred]][0]
+        return ret_val
+
+    def block_writes_to_writer_map (self, block):
+        ret_val = {}
+        for local in self.blocks_writes[block]:
+            ret_val[local] = block
+        return ret_val
+
+    def get_reaching_definitions (self, block):
+        '''Return a nested map for the given block
+        s.t. ret_val[pred][local] equals the block key for the
+        definition of local that reaches the argument block via that
+        predecessor.
+
+        Useful for actually populating phi nodes, once you know you
+        need them.'''
+        has_memoized = hasattr(self, 'reaching_definitions')
+        if has_memoized and block in self.reaching_definitions:
+            ret_val = self.reaching_definitions[block]
+        else:
+            preds = self.blocks_in[block]
+            ret_val = {}
+            for pred in preds:
+                ret_val[pred] = self.block_writes_to_writer_map(pred)
+                crnt = self.idom(pred)
+                while crnt != None:
+                    crnt_writer_map = self.block_writes_to_writer_map(crnt)
+                    # This order of update favors the first definitions
+                    # encountered in the traversal since the traversal
+                    # visits blocks in reverse execution order.
+                    crnt_writer_map.update(ret_val[pred])
+                    ret_val[pred] = crnt_writer_map
+                    crnt = self.idom(crnt)
+            if not has_memoized:
+                self.reaching_definitions = {}
+            self.reaching_definitions[block] = ret_val
+        return ret_val
+
+    def nreaches (self, block):
+        '''For each local, find the number of unique reaching
+        definitions the current block has.'''
+        reaching_definitions = self.get_reaching_definitions(block)
+        definition_map = {}
+        for pred in self.blocks_in[block]:
+            reaching_from_pred = reaching_definitions[pred]
+            for local in reaching_from_pred.keys():
+                if local not in definition_map:
+                    definition_map[local] = set()
+                definition_map[local].add(reaching_from_pred[local])
+        ret_val = {}
+        for local in definition_map.keys():
+            ret_val[local] = len(definition_map[local])
+        return ret_val
+
+    def writes_local (self, block, write_instr_index, local_index):
+        self.blocks_writes[block].add(local_index)
+        block_writers = self.blocks_writer[block]
+        old_index = block_writers.get(local_index, -1)
+        # This checks for a corner case that would impact
+        # numba.translate.Translate.build_phi_nodes().
+        assert old_index != write_instr_index, (
+            "Found corner case for STORE_FAST at a CFG join!")
+        block_writers[local_index] = max(write_instr_index, old_index)
+
+    def phi_needed (self, join):
+        '''Return the set of locals that will require a phi node to be
+        generated at the given join.'''
+        nreaches = self.nreaches(join)
+        return set([local for local in nreaches.keys()
+                    if nreaches[local] > 1])
+
+    def pprint (self, *args, **kws):
+        pprint.pprint(self.__dict__, *args, **kws)
+
+    def pformat (self, *args, **kws):
+        return pprint.pformat(self.__dict__, *args, **kws)
+
+    def to_dot (self, graph_name = None):
+        '''Return a dot (digraph visualizer in Graphviz) graph
+        description as a string.'''
+        if graph_name is None:
+            graph_name = 'CFG_%d' % id(self)
+        lines_out = []
+        for block_index in self.blocks:
+            lines_out.append(
+                'BLOCK_%r [shape=box, label="BLOCK_%r\\nr: %r, w: %r"];' %
+                (block_index, block_index,
+                 tuple(self.blocks_reads[block_index]),
+                 tuple(self.blocks_writes[block_index])))
+        for block_index in self.blocks:
+            for out_edge in self.blocks_out[block_index]:
+                lines_out.append('BLOCK_%r -> BLOCK_%r;' %
+                                 (block_index, out_edge))
+        return 'digraph %s {\n%s\n}\n' % (graph_name, '\n'.join(lines_out))
+
+# ______________________________________________________________________
+# End of control_flow.py

llpython/gen_bytecode_visitor.py

@@ -0,0 +1,26 @@
+# ______________________________________________________________________
+from __future__ import absolute_import
+from . import opcode_util
+
+# ______________________________________________________________________
+
+def generate_bytecode_visitor (classname = 'BytecodeVisitor',
+                               baseclass = 'object'):
+    opnames = list(set((opname.split('+')[0]
+                        for opname in opcode_util.OPCODE_MAP.keys())))
+    opnames.sort()
+    return 'class %s (%s):\n%s\n' % (
+        classname, baseclass,
+        '\n\n'.join(('    def op_%s (self, i, op, arg):\n'
+                     '        raise NotImplementedError("%s.op_%s")' %
+                     (opname, classname, opname)
+                     for opname in opnames)))
+
+# ______________________________________________________________________
+
+if __name__ == "__main__":
+    import sys
+    print(generate_bytecode_visitor(*sys.argv[1:]))
+
+# ______________________________________________________________________
+# End of gen_bytecode_visitor.py

llpython/nobitey.py

@@ -0,0 +1,367 @@
+# ______________________________________________________________________
+from __future__ import absolute_import
+import sys
+import os.path
+import imp
+import io
+import types
+
+import llvm.core as lc
+import llvm.ee as le
+
+from . import bytetype, byte_translator
+from .pyaddfunc import pyaddfunc
+
+LLVM_TO_INT_PARSE_STR_MAP = {
+    8 : 'b',
+    16 : 'h',
+    32 : 'i', # Note that on 32-bit systems sizeof(int) == sizeof(long)
+    64 : 'L', # Seeing sizeof(long long) == 8 on both 32 and 64-bit platforms
+}
+
+LLVM_TO_PARSE_STR_MAP = {
+    lc.TYPE_FLOAT : 'f',
+    lc.TYPE_DOUBLE : 'd',
+}
+
+# ______________________________________________________________________
+
+# XXX Stolen from numba.translate
+
+def get_string_constant (module, const_str):
+    const_name = "__STR_%x" % (hash(const_str),)
+    try:
+        ret_val = module.get_global_variable_named(const_name)
+    except:
+        lconst_str = lc.Constant.stringz(const_str)
+        ret_val = module.add_global_variable(lconst_str.type, const_name)
+        ret_val.initializer = lconst_str
+        ret_val.linkage = lc.LINKAGE_INTERNAL
+    return ret_val
+
+# ______________________________________________________________________
+
+class NoBitey (object):
+    def __init__ (self, target_module = None, type_annotations = None):
+        if target_module is None:
+            target_module = lc.Module.new('NoBitey_%d' % id(self))
+        if type_annotations is None:
+            type_annotations = {}
+        self.target_module = target_module
+        self.type_aliases = type_annotations # Reserved for future use.
+
+    def _build_parse_string (self, llvm_type):
+        kind = llvm_type.kind
+        if kind == lc.TYPE_INTEGER:
+            ret_val = LLVM_TO_INT_PARSE_STR_MAP[llvm_type.width]
+        elif kind in LLVM_TO_PARSE_STR_MAP:
+            ret_val = LLVM_TO_PARSE_STR_MAP[kind]
+        else:
+            raise TypeError('Unsupported LLVM type: %s' % str(llvm_type))
+        return ret_val
+
+    def build_parse_string (self, llvm_tys):
+        """Given a set of LLVM types, return a string for parsing
+        them via PyArg_ParseTuple."""
+        return ''.join((self._build_parse_string(ty)
+                        for ty in llvm_tys))
+
+    def handle_abi_casts (self, builder, result):
+        if result.type.kind == lc.TYPE_FLOAT:
+            # NOTE: The C ABI apparently casts floats to doubles when
+            # an argument must be pushed on the stack, as is the case
+            # when calling a variable argument function.
+            # XXX Is there documentation on this where I can find all
+            # coercion rules?  Do we still need some libffi
+            # integration?
+            result = builder.fpext(result, bytetype.ldouble)
+        return result
+
+    def build_wrapper_function (self, llvm_function, engine = None):
+        arg_types = llvm_function.type.pointee.args
+        return_type = llvm_function.type.pointee.return_type
+        li32_0 = lc.Constant.int(bytetype.li32, 0)
+        def get_llvm_function (builder):
+            if self.target_module != llvm_function.module:
+                llvm_function_ptr = self.target_module.add_global_variable(
+                    llvm_function.type, llvm_function.name)
+                llvm_function_ptr.initializer = lc.Constant.inttoptr(
+                    lc.Constant.int(
+                        bytetype.liptr,
+                        engine.get_pointer_to_function(llvm_function)),
+                    llvm_function.type)
+                llvm_function_ptr.linkage = lc.LINKAGE_INTERNAL
+                ret_val = builder.load(llvm_function_ptr)
+            else:
+                ret_val = llvm_function
+            return ret_val
+        def build_parse_args (builder):
+            return [builder.alloca(arg_type) for arg_type in arg_types]
+        def build_parse_string (builder):
+            parse_str = get_string_constant(
+                self.target_module, self.build_parse_string(arg_types))
+            return builder.gep(parse_str, (li32_0, li32_0))
+        def load_target_args (builder, args):
+            return [builder.load(arg) for arg in args]
+        def build_build_string (builder):
+            build_str = get_string_constant(
+                self.target_module, self._build_parse_string(return_type))
+            return builder.gep(build_str, (li32_0, li32_0))
+        handle_abi_casts = self.handle_abi_casts
+        target_function_name = llvm_function.name + "_wrapper"
+        # __________________________________________________
+        @byte_translator.llpython(bytetype.l_pyfunc, self.target_module,
+                                   **locals())
+        def _wrapper (self, args):
+            ret_val = l_pyobj_p(0)
+            parse_args = build_parse_args()
+            parse_result = PyArg_ParseTuple(args, build_parse_string(),
+                                            *parse_args)
+            if parse_result != li32(0):
+                thread_state = PyEval_SaveThread()
+                target_args = load_target_args(parse_args)
+                llresult = handle_abi_casts(get_llvm_function()(*target_args))
+                PyEval_RestoreThread(thread_state)
+                ret_val = Py_BuildValue(build_build_string(), llresult)
+            return ret_val
+        # __________________________________________________
+        return _wrapper
+
+    def wrap_llvm_module (self, llvm_module, engine = None, py_module = None):
+        '''
+        Shamefully adapted from bitey.bind.wrap_llvm_module().
+        '''
+        functions = [func for func in llvm_module.functions
+                     if not func.name.startswith("_")
+                     and not func.is_declaration
+                     and func.linkage == lc.LINKAGE_EXTERNAL]
+        if engine is None:
+            engine = le.ExecutionEngine.new(llvm_module)
+        wrappers = [self.build_wrapper_function(func, engine)
+                    for func in functions]
+        if __debug__: print(self.target_module)
+        if self.target_module != llvm_module:
+            engine.add_module(self.target_module)
+        py_wrappers = [pyaddfunc(wrapper.name,
+                                 engine.get_pointer_to_function(wrapper))
+                       for wrapper in wrappers]
+        if py_module:
+            for py_wrapper in py_wrappers:
+                setattr(py_module, py_wrapper.__name__[:-8], py_wrapper)
+            setattr(py_module, '_llvm_module', llvm_module)
+            setattr(py_module, '_llvm_engine', engine)
+            if self.target_module != llvm_module:
+                setattr(py_module, '_llvm_wrappers', self.target_module)
+        return engine, py_wrappers
+
+    def wrap_llvm_module_in_python (self, llvm_module, py_module = None):
+        '''
+        Mildly reworked and abstracted bitey.bind.wrap_llvm_bitcode().
+        Abstracted to accept any existing LLVM Module object, and
+        return a Python wrapper module (even if one wasn't originally
+        specified).
+        '''
+        if py_module is None:
+            py_module = types.ModuleType(str(llvm_module.id))
+        engine = le.ExecutionEngine.new(llvm_module)
+        self.wrap_llvm_module(llvm_module, engine, py_module)
+        return py_module
+
+    def wrap_llvm_bitcode (self, bitcode, py_module = None):
+        '''
+        Intended to be drop-in replacement of
+        bitey.bind.wrap_llvm_bitcode().
+        '''
+        return self.wrap_llvm_module_in_python(
+            lc.Module.from_bitcode(io.BytesIO(bitcode)), py_module)
+
+    def wrap_llvm_assembly (self, llvm_asm, py_module = None):
+        return self.wrap_llvm_module_in_python(
+            lc.Module.from_assembly(io.BytesIO(llvm_asm)), py_module)
+
+# ______________________________________________________________________
+
+class NoBiteyLoader(object):
+    """
+    Load LLVM compiled bitcode and autogenerate a ctypes binding.
+
+    Initially copied and adapted from bitey.loader module.
+    """
+    def __init__(self, pkg, name, source, preload, postload):
+        self.package = pkg
+        self.name = name
+        self.fullname = '.'.join((pkg,name))
+        self.source = source
+        self.preload = preload
+        self.postload = postload
+
+    @classmethod
+    def _check_magic(cls, filename):
+        if os.path.exists(filename):
+            magic = open(filename,"rb").read(4)
+            if magic == b'\xde\xc0\x17\x0b':
+                return True
+            elif magic[:2] == b'\x42\x43':
+                return True
+            else:
+                return False
+        else:
+            return False
+
+    @classmethod
+    def build_module(cls, fullname, source_path, source_data, preload=None,
+                     postload=None):
+        name = fullname.split(".")[-1]
+        mod = imp.new_module(name)
+        if preload:
+            exec(preload, mod.__dict__, mod.__dict__)
+        type_annotations = getattr(mod, '_type_annotations', None)
+        nb = NoBitey(type_annotations = type_annotations)
+        if source_path.endswith(('.o', '.bc')):
+            nb.wrap_llvm_bitcode(source_data, mod)
+        elif source_path.endswith('.s'):
+            nb.wrap_llvm_assembly(source_data, mod)
+        if postload:
+            exec(postload, mod.__dict__, mod.__dict__)
+        return mod
+
+    @classmethod
+    def find_module(cls, fullname, paths = None):
+        if paths is None:
+            paths = sys.path
+        names = fullname.split('.')
+        modname = names[-1]
+        source_paths = None
+        for f in paths:
+            path = os.path.join(os.path.realpath(f), modname)
+            source = path + '.o'
+            if cls._check_magic(source):
+                source_paths = path, source
+                break
+            source = path + '.bc'
+            if os.path.exists(source):
+                source_paths = path, source
+                break
+            source = path + '.s'
+            if os.path.exists(source):
+                source_paths = path, source
+                break
+        if source_paths:
+            path, source = source_paths
+            return cls('.'.join(names[:-1]), modname, source,
+                       path + ".pre.py", path + ".post.py")
+
+    def get_code(self, module):
+        pass
+
+    def get_data(self, module):
+        pass
+
+    def get_filename(self, name):
+        return self.source
+
+    def get_source(self, name):
+        with open(self.source, 'rb') as f:
+             return f.read()
+
+    def is_package(self, *args, **kw):
+        return False
+
+    def load_module(self, fullname):
+        if fullname in sys.modules:
+            return sys.modules[fullname]
+
+        preload = None
+        postload = None
+
+        # Get the preload file (if any)
+        if os.path.exists(self.preload):
+            with open(self.preload) as f:
+                preload = f.read()
+
+        # Get the source
+        with open(self.source, 'rb') as f:
+            source_data = f.read()
+
+        # Get the postload file (if any)
+        if os.path.exists(self.postload):
+            with open(self.postload) as f:
+                postload = f.read()
+
+        mod = self.build_module(fullname, self.get_filename(None), source_data,
+                                preload, postload)
+        sys.modules[fullname] = mod
+        mod.__loader__ = self
+        mod.__file__ = self.source
+        return mod
+
+    @classmethod
+    def install(cls):
+        if cls not in sys.meta_path:
+            sys.meta_path.append(cls)
+
+    @classmethod
+    def remove(cls):
+        sys.meta_path.remove(cls)
+
+# ______________________________________________________________________
+
+def _mk_add_42 (llvm_module, at_type = bytetype.lc_long):
+    f = llvm_module.add_function(
+        lc.Type.function(at_type, [at_type]), 'add_42_%s' % str(at_type))
+    block = f.append_basic_block('entry')
+    builder = lc.Builder.new(block)
+    if at_type.kind == lc.TYPE_INTEGER:
+        const_42 = lc.Constant.int(at_type, 42)
+        add = builder.add
+    elif at_type.kind in (lc.TYPE_FLOAT, lc.TYPE_DOUBLE):
+        const_42 = lc.Constant.real(at_type, 42.)
+        add = builder.fadd
+    else:
+        raise TypeError('Unsupported type: %s' % str(at_type))
+    builder.ret(add(f.args[0], const_42))
+    return f
+
+# ______________________________________________________________________
+
+def build_test_module ():
+    llvm_module = lc.Module.new('nobitey_test')
+    for ty in (bytetype.li32, bytetype.li64, bytetype.lfloat,
+               bytetype.ldouble):
+        fn = _mk_add_42(llvm_module, ty)
+    return llvm_module
+
+# ______________________________________________________________________
+
+def test_wrap_module (arg = None):
+    # Build up a module.
+    m = build_test_module()
+    if arg and arg.lower() == 'separated':
+        wrap_module = NoBitey().wrap_llvm_module_in_python(m)
+    else:
+        wrap_module = NoBitey(m).wrap_llvm_module_in_python(m)
+    # Now try running the generated wrappers.
+    for py_wf_name in ('add_42_i32', 'add_42_i64', 'add_42_float',
+                       'add_42_double'):
+        py_wf = getattr(wrap_module, py_wf_name)
+        for i in range(42):
+            result = py_wf(i)
+            expected = i + 42
+            assert result == expected, "%r != %r in %r" % (
+                result, expected, py_wf)
+    return wrap_module
+
+# ______________________________________________________________________
+
+def main (*args):
+    if args:
+        for arg in args:
+            test_wrap_module(arg)
+    else:
+        test_wrap_module()
+
+if __name__ == "__main__":
+    main(*sys.argv[1:])
+
+# ______________________________________________________________________
+# End of nobitey.py

llpython/opcode_util.py

@@ -0,0 +1,225 @@
+# ______________________________________________________________________
+
+import dis
+import opcode
+
+# ______________________________________________________________________
+# Module data
+
+hasjump = opcode.hasjrel + opcode.hasjabs
+hascbranch = [op for op in hasjump
+              if 'IF' in opcode.opname[op]
+              or opcode.opname[op] in ('FOR_ITER', 'SETUP_LOOP')]
+
+# Since the actual opcode value may change, manage opcode abstraction
+# data by opcode name.
+
+OPCODE_MAP = {
+    'BINARY_ADD': (2, 1, None),
+    'BINARY_AND': (2, 1, None),
+    'BINARY_DIVIDE': (2, 1, None),
+    'BINARY_FLOOR_DIVIDE': (2, 1, None),
+    'BINARY_LSHIFT': (2, 1, None),
+    'BINARY_MODULO': (2, 1, None),
+    'BINARY_MULTIPLY': (2, 1, None),
+    'BINARY_OR': (2, 1, None),
+    'BINARY_POWER': (2, 1, None),
+    'BINARY_RSHIFT': (2, 1, None),
+    'BINARY_SUBSCR': (2, 1, None),
+    'BINARY_SUBTRACT': (2, 1, None),
+    'BINARY_TRUE_DIVIDE': (2, 1, None),
+    'BINARY_XOR': (2, 1, None),
+    'BREAK_LOOP': (0, None, 1),
+    'BUILD_CLASS': (None, None, None),
+    'BUILD_LIST': (-1, 1, None),
+    'BUILD_MAP': (None, None, None),
+    'BUILD_SET': (None, None, None),
+    'BUILD_SLICE': (None, None, None),
+    'BUILD_TUPLE': (-1, 1, None),
+    'CALL_FUNCTION': (-2, 1, None),
+    'CALL_FUNCTION_KW': (-3, 1, None),
+    'CALL_FUNCTION_VAR': (-3, 1, None),
+    'CALL_FUNCTION_VAR_KW': (-4, 1, None),
+    'COMPARE_OP': (2, 1, None),
+    'CONTINUE_LOOP': (None, None, None),
+    'DELETE_ATTR': (1, None, 1),
+    'DELETE_DEREF': (None, None, None),
+    'DELETE_FAST': (0, None, 1),
+    'DELETE_GLOBAL': (0, None, 1),
+    'DELETE_NAME': (0, None, 1),
+    'DELETE_SLICE+0': (1, None, 1),
+    'DELETE_SLICE+1': (2, None, 1),
+    'DELETE_SLICE+2': (2, None, 1),
+    'DELETE_SLICE+3': (3, None, 1),
+    'DELETE_SUBSCR': (2, None, 1),
+    'DUP_TOP': (None, None, None),
+    'DUP_TOPX': (None, None, None),
+    'DUP_TOP_TWO': (None, None, None),
+    'END_FINALLY': (None, None, None),
+    'EXEC_STMT': (None, None, None),
+    'EXTENDED_ARG': (None, None, None),
+    'FOR_ITER': (1, 1, 1),
+    'GET_ITER': (1, 1, None),
+    'IMPORT_FROM': (None, None, None),
+    'IMPORT_NAME': (None, None, None),
+    'IMPORT_STAR': (1, None, 1),
+    'INPLACE_ADD': (2, 1, None),
+    'INPLACE_AND': (2, 1, None),
+    'INPLACE_DIVIDE': (2, 1, None),
+    'INPLACE_FLOOR_DIVIDE': (2, 1, None),
+    'INPLACE_LSHIFT': (2, 1, None),
+    'INPLACE_MODULO': (2, 1, None),
+    'INPLACE_MULTIPLY': (2, 1, None),
+    'INPLACE_OR': (2, 1, None),
+    'INPLACE_POWER': (2, 1, None),
+    'INPLACE_RSHIFT': (2, 1, None),
+    'INPLACE_SUBTRACT': (2, 1, None),
+    'INPLACE_TRUE_DIVIDE': (2, 1, None),
+    'INPLACE_XOR': (2, 1, None),
+    'JUMP_ABSOLUTE': (0, None, 1),
+    'JUMP_FORWARD': (0, None, 1),
+    'JUMP_IF_FALSE': (1, 1, 1),
+    'JUMP_IF_FALSE_OR_POP': (None, None, None),
+    'JUMP_IF_TRUE': (1, 1, 1),
+    'JUMP_IF_TRUE_OR_POP': (None, None, None),
+    'LIST_APPEND': (2, 0, 1),
+    'LOAD_ATTR': (1, 1, None),
+    'LOAD_BUILD_CLASS': (None, None, None),
+    'LOAD_CLOSURE': (None, None, None),
+    'LOAD_CONST': (0, 1, None),
+    'LOAD_DEREF': (0, 1, None),
+    'LOAD_FAST': (0, 1, None),
+    'LOAD_GLOBAL': (0, 1, None),
+    'LOAD_LOCALS': (None, None, None),
+    'LOAD_NAME': (0, 1, None),
+    'MAKE_CLOSURE': (None, None, None),
+    'MAKE_FUNCTION': (-2, 1, None),
+    'MAP_ADD': (None, None, None),
+    'NOP': (0, None, None),
+    'POP_BLOCK': (0, None, 1),
+    'POP_EXCEPT': (None, None, None),
+    'POP_JUMP_IF_FALSE': (1, None, 1),
+    'POP_JUMP_IF_TRUE': (1, None, 1),
+    'POP_TOP': (1, None, 1),
+    'PRINT_EXPR': (1, None, 1),
+    'PRINT_ITEM': (1, None, 1),
+    'PRINT_ITEM_TO': (2, None, 1),
+    'PRINT_NEWLINE': (0, None, 1),
+    'PRINT_NEWLINE_TO': (1, None, 1),
+    'RAISE_VARARGS': (None, None, None),
+    'RETURN_VALUE': (1, None, 1),
+    'ROT_FOUR': (None, None, None),
+    'ROT_THREE': (None, None, None),
+    'ROT_TWO': (None, None, None),
+    'SETUP_EXCEPT': (None, None, None),
+    'SETUP_FINALLY': (None, None, None),
+    'SETUP_LOOP': (None, None, None),
+    'SETUP_WITH': (None, None, None),
+    'SET_ADD': (None, None, None),
+    'SLICE+0': (1, 1, None),
+    'SLICE+1': (2, 1, None),
+    'SLICE+2': (2, 1, None),
+    'SLICE+3': (3, 1, None),
+    'STOP_CODE': (None, None, None),
+    'STORE_ATTR': (2, None, 1),
+    'STORE_DEREF': (1, 0, 1),
+    'STORE_FAST': (1, None, 1),
+    'STORE_GLOBAL': (1, None, 1),
+    'STORE_LOCALS': (None, None, None),
+    'STORE_MAP': (1, None, 1),
+    'STORE_NAME': (1, None, 1),
+    'STORE_SLICE+0': (1, None, 1),
+    'STORE_SLICE+1': (2, None, 1),
+    'STORE_SLICE+2': (2, None, 1),
+    'STORE_SLICE+3': (3, None, 1),
+    'STORE_SUBSCR': (3, None, 1),
+    'UNARY_CONVERT': (1, 1, None),
+    'UNARY_INVERT': (1, 1, None),
+    'UNARY_NEGATIVE': (1, 1, None),
+    'UNARY_NOT': (1, 1, None),
+    'UNARY_POSITIVE': (1, 1, None),
+    'UNPACK_EX': (None, None, None),
+    'UNPACK_SEQUENCE': (None, None, None),
+    'WITH_CLEANUP': (None, None, None),
+    'YIELD_VALUE': (1, None, 1),
+}
+
+# ______________________________________________________________________
+# Module functions
+
+def itercode(code, start = 0):
+    """Return a generator of byte-offset, opcode, and argument
+    from a byte-code-string
+    """
+    i = 0
+    extended_arg = 0
+    if isinstance(code[0], str):
+        code = [ord(c) for c in code]
+    n = len(code)
+    while i < n:
+        op = code[i]
+        num = i + start
+        i = i + 1
+        oparg = None
+        if op >= opcode.HAVE_ARGUMENT:
+            oparg = code[i] + (code[i + 1] * 256) + extended_arg
+            extended_arg = 0
+            i = i + 2
+            if op == opcode.EXTENDED_ARG:
+                extended_arg = oparg * 65536
+
+        delta = yield num, op, oparg
+        if delta is not None:
+            abs_rel, dst = delta
+            assert abs_rel == 'abs' or abs_rel == 'rel'
+            i = dst if abs_rel == 'abs' else i + dst
+
+# ______________________________________________________________________
+
+def extendlabels(code, labels = None):
+    """Extend the set of jump target labels to account for the
+    passthrough targets of conditional branches.
+
+    This allows us to create a control flow graph where there is at
+    most one branch per basic block.
+    """
+    if labels is None:
+        labels = []
+    if isinstance(code[0], str):
+        code = [ord(c) for c in code]
+    n = len(code)
+    i = 0
+    while i < n:
+        op = code[i]
+        i += 1
+        if op >= dis.HAVE_ARGUMENT:
+            i += 2
+            label = -1
+            if op in hasjump:
+                label = i
+            if label >= 0:
+                if label not in labels:
+                    labels.append(label)
+        elif op == opcode.opmap['BREAK_LOOP']:
+            if i not in labels:
+                labels.append(i)
+    return labels
+
+# ______________________________________________________________________
+
+def get_code_object (func):
+    return getattr(func, '__code__', getattr(func, 'func_code', None))
+
+# ______________________________________________________________________
+
+def build_basic_blocks (co_obj):
+    co_code = co_obj.co_code
+    labels = extendlabels(co_code, dis.findlabels(co_code))
+    labels.sort()
+    blocks = dict((index, list(itercode(co_code[index:next_index], index)))
+                  for index, next_index in zip([0] + labels,
+                                               labels + [len(co_code)]))
+    return blocks
+
+# ______________________________________________________________________
+# End of opcode_util.py

llpython/phi_injector.py

@@ -0,0 +1,153 @@
+# ______________________________________________________________________
+
+from .bytecode_visitor import BytecodeFlowVisitor, BenignBytecodeVisitorMixin
+
+# ______________________________________________________________________
+
+synthetic_opname = []
+synthetic_opmap = {}
+
+def def_synth_op (opname):
+    global synthetic_opname, synthetic_opmap
+    ret_val = -(len(synthetic_opname) + 1)
+    synthetic_opname.insert(0, opname)
+    synthetic_opmap[opname] = ret_val
+    return ret_val
+
+REF_ARG = def_synth_op('REF_ARG')
+BUILD_PHI = def_synth_op('BUILD_PHI')
+DEFINITION = def_synth_op('DEFINITION')
+REF_DEF = def_synth_op('REF_DEF')
+
+# ______________________________________________________________________
+
+class PhiInjector (BenignBytecodeVisitorMixin, BytecodeFlowVisitor):
+    '''Transformer responsible for modifying a bytecode flow, removing
+    LOAD_FAST and STORE_FAST opcodes, and replacing them with a static
+    single assignment (SSA) representation.
+
+    In order to support SSA, PhiInjector adds the following synthetic
+    opcodes to transformed flows:
+
+      * REF_ARG: Specifically reference an incomming argument value.
+
+      * BUILD_PHI: Build a phi node to disambiguate between several
+        possible definitions at a control flow join.
+
+      * DEFINITION: Unique value definition indexed by the "arg" field
+        in the tuple.
+
+      * REF_DEF: Reference a specific value definition.'''
+
+    def visit_cfg (self, cfg, nargs = 0, *args, **kws):
+        self.cfg = cfg
+        ret_val = self.visit(cfg.blocks, nargs)
+        del self.cfg
+        return ret_val
+
+    def visit (self, flow, nargs = 0, *args, **kws):
+        self.nargs = nargs
+        self.definitions = []
+        self.phis = []
+        self.prev_blocks = []
+        self.blocks_locals = dict((block, {})
+                                  for block in self.cfg.blocks.keys())
+        ret_val = super(PhiInjector, self).visit(flow, *args, **kws)
+        for block, _, _, args, _ in self.phis:
+            local = args.pop()
+            reaching_definitions = self.cfg.reaching_definitions[block]
+            for prev in reaching_definitions.keys():
+                if 0 in self.cfg.blocks_reaching[prev]:
+                    args.append((prev, REF_DEF, 'REF_DEF',
+                                 self.blocks_locals[prev][local], ()))
+            args.sort()
+        del self.blocks_locals
+        del self.prev_blocks
+        del self.phis
+        del self.definitions
+        del self.nargs
+        return ret_val
+
+    def add_definition (self, index, local, arg):
+        definition_index = len(self.definitions)
+        definition = (index, DEFINITION, 'DEFINITION', definition_index,
+                      (arg,))
+        self.definitions.append(definition)
+        self.blocks_locals[self.block][local] = definition_index
+        return definition
+
+    def add_phi (self, index, local):
+        ret_val = (index, BUILD_PHI, 'BUILD_PHI', [local], ())
+        self.phis.append(ret_val)
+        return ret_val
+
+    def enter_block (self, block):
+        ret_val = False
+        self.block = block
+        if block == 0:
+            if self.nargs > 0:
+                ret_val = [self.add_definition(-1, arg,
+                                                (-1, REF_ARG, 'REF_ARG', arg,
+                                                  ()))
+                           for arg in range(self.nargs)]
+            else:
+                ret_val = True
+        elif 0 in self.cfg.blocks_reaching[block]:
+            ret_val = True
+            prev_block_locals = None
+            for pred_block in self.cfg.blocks_in[block]:
+                if pred_block in self.prev_blocks:
+                    prev_block_locals = self.blocks_locals[pred_block]
+                    break
+            assert prev_block_locals is not None, "Internal translation error"
+            self.blocks_locals[block] = prev_block_locals.copy()
+            phis_needed = self.cfg.phi_needed(block)
+            if phis_needed:
+                ret_val = [self.add_definition(block, local,
+                                               self.add_phi(block, local))
+                           for local in phis_needed]
+        return ret_val
+
+    def exit_block (self, block):
+        if 0 in self.cfg.blocks_reaching[block]:
+            self.prev_blocks.append(block)
+        del self.block
+
+    def op_STORE_FAST (self, i, op, arg, *args, **kws):
+        assert len(args) == 1
+        return [self.add_definition(i, arg, args[0])]
+
+    def op_LOAD_FAST (self, i, op, arg, *args, **kws):
+        return [(i, REF_DEF, 'REF_DEF', self.blocks_locals[self.block][arg],
+                 args)]
+
+# ______________________________________________________________________
+
+def inject_phis (func):
+    '''Given a Python function, return a bytecode flow object that has
+    been transformed by a fresh PhiInjector instance.'''
+    import byte_control, byte_flow
+    argcount = byte_control.opcode_util.get_code_object(func).co_argcount
+    cfg = byte_control.build_cfg(func)
+    cfg.blocks = byte_flow.BytecodeFlowBuilder().visit_cfg(cfg)
+    return PhiInjector().visit_cfg(cfg, argcount)
+
+# ______________________________________________________________________
+# Main (self-test) routine
+
+def main (*args):
+    import pprint
+    from tests import llfuncs
+    if not args:
+        args = ('doslice',)
+    for arg in args:
+        pprint.pprint(inject_phis(getattr(llfuncs, arg)))
+
+# ______________________________________________________________________
+
+if __name__ == "__main__":
+    import sys
+    main(*sys.argv[1:])
+
+# ______________________________________________________________________
+# End of phi_injector.py

llpython/pyaddfunc.py

@@ -0,0 +1,40 @@
+# ______________________________________________________________________
+
+import ctypes
+
+# ______________________________________________________________________
+class PyMethodDef (ctypes.Structure):
+    _fields_ = [
+        ('ml_name', ctypes.c_char_p),
+        ('ml_meth', ctypes.c_void_p),
+        ('ml_flags', ctypes.c_int),
+        ('ml_doc', ctypes.c_char_p),
+        ]
+
+PyCFunction_NewEx = ctypes.pythonapi.PyCFunction_NewEx
+PyCFunction_NewEx.argtypes = (ctypes.POINTER(PyMethodDef),
+                              ctypes.c_void_p,
+                              ctypes.c_void_p)
+PyCFunction_NewEx.restype = ctypes.py_object
+
+cache = {} # Unsure if this is necessary to keep the PyMethodDef
+           # structures from being garbage collected.  Assuming so...
+
+def pyaddfunc (func_name, func_ptr, func_doc = None):
+    global cache
+    if bytes != str:
+        func_name = bytes(ord(ch) for ch in func_name)
+    key = (func_name, func_ptr)
+    if key in cache:
+        _, ret_val = cache[key]
+    else:
+        mdef = PyMethodDef(bytes(func_name),
+                           func_ptr,
+                           1, # == METH_VARARGS (hopefully remains so...)
+                           func_doc)
+        ret_val = PyCFunction_NewEx(ctypes.byref(mdef), 0, 0)
+        cache[key] = (mdef, ret_val)
+    return ret_val
+
+# ______________________________________________________________________
+# End of pyaddfunc.py

llpython/tests/llfuncs.py

@@ -0,0 +1,41 @@
+# ______________________________________________________________________
+
+def doslice (in_string, lower, upper):
+    l = strlen(in_string)
+    if lower < lc_size_t(0):
+        lower += l
+    if upper < lc_size_t(0):
+        upper += l
+    temp_len = upper - lower
+    if temp_len < lc_size_t(0):
+        temp_len = lc_size_t(0)
+    ret_val = alloca_array(li8, temp_len + lc_size_t(1))
+    strncpy(ret_val, in_string + lower, temp_len)
+    ret_val[temp_len] = li8(0)
+    return ret_val
+
+def ipow (val, exp):
+    ret_val = 1
+    temp = val
+    w = exp
+    while w > 0:
+        if (w & 1) != 0:
+            ret_val *= temp
+            # TODO: Overflow check on ret_val
+        w >>= 1
+        if w == 0: break
+        temp *= temp
+        # TODO: Overflow check on temp
+    return ret_val
+
+def pymod (arg1, arg2):
+    ret_val = arg1 % arg2
+    if ret_val < 0:
+        if arg2 > 0:
+            ret_val += arg2
+    elif arg2 < 0:
+        ret_val += arg2
+    return ret_val
+
+# ______________________________________________________________________
+# End of llfuncs.py

llpython/tests/llfunctys.py

@@ -0,0 +1,19 @@
+# ______________________________________________________________________
+
+import llvm.core as lc
+
+from llpython import bytetype
+
+# ______________________________________________________________________
+
+doslice = lc.Type.function(bytetype.li8_ptr, (
+        bytetype.li8_ptr, bytetype.lc_size_t, bytetype.lc_size_t))
+
+ipow = lc.Type.function(bytetype.li32, (bytetype.li32,
+                                        bytetype.li32))
+
+pymod = lc.Type.function(bytetype.li32, (bytetype.li32,
+                                         bytetype.li32))
+
+# ______________________________________________________________________
+# End of llfunctys.py

llrtc/Makefile

@@ -0,0 +1,22 @@
+all:
+	make -C lib
+
+ir:
+	make -C lib ir
+
+test:
+	make -C lib test
+
+clean-test:
+	make -C lib clean-test
+
+clean-temp:
+	make -C lib clean-temp
+
+clean:
+	make -C lib clean
+
+install: ir
+	cp llrt_*.ll ../llvm/llrt
+	make -C lib clean-temp
+	

llrtc/README.md

@@ -0,0 +1,25 @@
+# LLRT: Low Level Runtime
+
+## Why?
+
+The same reason for LLVM compiler-rt.  LLVM generates libgcc symbols, such as
+__divdi3 for 64-bit division on 32-bit platform.  They are not also available.
+We need to ship compiler-rt but it is not Windows ready.
+This subproject aims to provide a small portable subset of compiler-rt.
+Start small and add only the things we really needed.
+Performance is not crucial but should not be terrible.
+Functionality and usefullness should be more important than performance.
+
+## Developer Instructions
+
+LLRT implements some functionalities in compiler-rt in ANSI C.
+The C files are compiled using clang to produce LLVM IR which are shipped.
+The IR files are committed in the repository.
+So, remember to build the IR files commit them after modifying the C files.
+
+## Build Requirement
+
+- Make
+- Clang
+- Python
+

llrtc/lib/.gitignore

@@ -0,0 +1,4 @@
+*.o
+*.run
+*.out
+*.ll

llrtc/lib/Makefile

@@ -0,0 +1,66 @@
+OUTPUT = llrt
+SOURCES = udivmod64.c sdivmod64.c div64.c mod64.c
+TESTS = test_udivmod64.c test_sdivmod64.c
+
+CLANG = clang
+LLVM_LINK = llvm-link
+CF = -Wall -ansi
+CF_TEST = $(CF) -ftrapv
+CF_BUILD = $(CF) -O0 -emit-llvm
+OUTDIR = ..
+STRIPPER = ../tools/striptriple.py
+
+all: ir
+
+ir: $(OUTDIR)/$(OUTPUT)_x86.ll $(OUTDIR)/$(OUTPUT)_x86_64.ll
+
+$(OUTDIR)/$(OUTPUT)_x86.ll: $(SOURCES:.c=_x86.bc)
+	$(LLVM_LINK) -S $+ -o $@
+	python $(STRIPPER) $@
+
+$(OUTDIR)/$(OUTPUT)_x86_64.ll: $(SOURCES:.c=_x86_64.bc)
+	$(LLVM_LINK) -S $+ -o $@
+	python $(STRIPPER) $@
+
+build-test: $(SOURCES:.c=.o) $(TESTS:.c=.run)
+
+lib$(OUTPUT).a: $(SOURCES:.c=.o)
+	$(CLANG) -static $+ -o $@
+
+test: $(TESTS:.c=.run)
+	for src in $+; do \
+		echo "testing $${src}";				 \
+		python $${src%.*}.py > $${src%.*}.out;	 \
+	done;
+
+clean-test:
+	rm -f *.out
+	rm -f *.o
+	rm -f *.run
+
+clean-dist: clean-temp
+	rm -f *.ll
+
+clean-temp:
+	rm -f *.bc
+	rm -f *.o
+	rm -f *.out
+
+clean: clean-test clean-dist
+
+%.c: llrt.h
+
+%_x86.bc: %.c
+	$(CLANG) -m32 $(CF_BUILD) -c $< -o $@
+
+%_x86_64.bc: %.c
+	$(CLANG) -m64 $(CF_BUILD) -c $< -o $@
+
+%.o: %.c
+	$(CLANG) $(CF_TEST) -c $<
+
+%.run: %.c
+	$(CLANG) $(CF_TEST) -o $@ $+
+
+test_udivmod64.run: udivmod64.o
+test_sdivmod64.run: udivmod64.o sdivmod64.o

llrtc/lib/div64.c

@@ -0,0 +1,11 @@
+#include "llrt.h"
+
+uint64_t udiv64(uint64_t dividend, uint64_t divisor)
+{
+    return udivmod64(dividend, divisor, NULL);
+}
+
+int64_t sdiv64(int64_t dividend, int64_t divisor)
+{
+    return sdivmod64(dividend, divisor, NULL);
+}

llrtc/lib/llrt.h

@@ -0,0 +1,19 @@
+#ifndef LLRT_H_
+#define LLRT_H_
+
+#include <stdint.h>
+
+#define NULL 0
+#define BITS_PER_BYTE 8
+
+uint64_t udivmod64(uint64_t dividend, uint64_t divisor, uint64_t *remainder);
+int64_t sdivmod64(int64_t dividend, int64_t divisor, int64_t *remainder);
+
+uint64_t udiv64(uint64_t dividend, uint64_t divisor);
+int64_t sdiv64(int64_t dividend, int64_t divisor);
+
+uint64_t umod64(uint64_t dividend, uint64_t divisor);
+int64_t smod64(int64_t dividend, int64_t divisor);
+
+#endif /* LLRT_H_ */
+

llrtc/lib/mod64.c

@@ -0,0 +1,15 @@
+#include "llrt.h"
+
+uint64_t umod64(uint64_t dividend, uint64_t divisor)
+{
+    uint64_t rem;
+    udivmod64(dividend, divisor, &rem);
+    return rem;
+}
+
+int64_t smod64(int64_t dividend, int64_t divisor)
+{
+    int64_t rem;
+    sdivmod64(dividend, divisor, &rem);
+    return rem;
+}

llrtc/lib/sdivmod64.c

@@ -0,0 +1,40 @@
+#include "llrt.h"
+#include <stdio.h>
+
+/*
+Calls to udivmod64 internally.
+Note: remainder uses sign of divisor.
+*/
+int64_t sdivmod64(int64_t dividend, int64_t divisor, int64_t *remainder)
+{
+    int signbitidx = BITS_PER_BYTE * sizeof(dividend) - 1;
+    int signed_dividend = dividend < 0;
+    int signed_divisor = divisor < 0;
+    int signed_result = signed_divisor ^ signed_dividend;
+
+    int64_t quotient;
+    uint64_t udvd, udvr, uquotient, uremainder;
+
+    udvd = signed_dividend ? -dividend : dividend;
+    udvr = signed_divisor ? -divisor : divisor;
+    uquotient = udivmod64(udvd, udvr, &uremainder);
+
+    if (signed_result){
+        if (uremainder) {
+            quotient = -(int64_t)uquotient - 1;
+        } else {
+            quotient = -(int64_t)uquotient;
+        }
+        if (remainder) {
+            /* if signed, there could be unsigned overflow
+               causing undefined behavior */
+            *remainder = (uint64_t)dividend - (uint64_t)quotient * (uint64_t)divisor;
+        }
+    } else {
+        quotient = (int64_t)uquotient;
+        if (remainder) {
+            *remainder = signed_divisor ? -uremainder : uremainder;
+        }
+    }
+    return quotient;
+}

llrtc/lib/test_sdivmod64.c

@@ -0,0 +1,21 @@
+#include <stdio.h>
+#include <stdint.h>
+#include "llrt.h"
+
+int main(int argc, char * argv[]){
+    int64_t n, d, q, r;
+
+    if (argc != 3) {
+        printf("invalid argument: %s dividend divisor", argv[0]);
+        return 1;
+    }
+    sscanf(argv[1], "%lld", &n);
+    sscanf(argv[2], "%lld", &d);
+
+    q = sdivmod64(n, d, &r);
+
+    printf("%lld\n", q);
+    printf("%lld\n", r);
+
+    return 0;
+}

llrtc/lib/test_sdivmod64.py

@@ -0,0 +1,56 @@
+import math
+import os
+import subprocess
+udt = os.path.join('.', 'test_sdivmod64.run')
+
+def testcase(dividend, divisor):
+    print 'divmod64(%d, %d)' % (dividend, divisor)
+
+    procargs = ('%s %s %s' % (udt, dividend, divisor)).split()
+    result = subprocess.check_output(procargs)
+    gotQ, gotR = map(int, result.splitlines())
+
+    expectQ = dividend // divisor
+    expectR = dividend % divisor
+
+    print 'Q = %d, R = %d' % (gotQ, gotR)
+
+    if expectQ != gotQ:
+        raise ValueError("invalid quotient: got=%d but expect=%d" %
+                         (gotQ, expectQ))
+    if expectR != gotR:
+        raise ValueError("invalid remainder: got=%d but expect=%d" %
+                         (gotR, expectR))
+    print 'OK'
+
+def testsequence():
+    subjects = [
+        (0, 1),
+        (0, 0xffffffff),
+        (1, 2),
+        (1, 983219),
+        (2, 2),
+        (3, 2),
+        (1024, 2),
+        (2048, 512),
+        (21321, 512),
+        (9329189, 1031),
+        (0xffffffff, 2),
+        (0xffffffff, 0xffff),
+        (0x1ffffffff, 2),
+        (0x1ffffffff, 0xffff),
+        (0xffff, 0xffffffff),
+        (0x0fffffffffffffff, 0xffff),
+        (0x7fffffffffffffff, 0x7fffffffffffffff),
+        (0x7fffffffffffffff, 0x7ffffffffffffff0),
+        (0x7fffffffffffffff, 87655678587161901),
+    ]
+
+    for dvd, dvr in subjects:
+        testcase(dvd, dvr)
+        testcase(dvd, -dvr)
+        testcase(-dvd, dvr)
+        testcase(-dvd, -dvr)
+
+if __name__ == '__main__':
+    testsequence()

llrtc/lib/test_udivmod64.c

@@ -0,0 +1,20 @@
+#include <stdio.h>
+#include <stdint.h>
+#include "llrt.h"
+
+int main(int argc, char * argv[]){
+    uint64_t n, d, q, r;
+    if (argc != 3) {
+        printf("invalid argument: %s dividend divisor", argv[0]);
+        return 1;
+    }
+    sscanf(argv[1], "%llu", &n);
+    sscanf(argv[2], "%llu", &d);
+
+    q = udivmod64(n, d, &r);
+
+    printf("%llu\n", q);
+    printf("%llu\n", r);
+
+    return 0;
+}

llrtc/lib/test_udivmod64.py

@@ -0,0 +1,53 @@
+import math
+import os
+import subprocess
+udt = os.path.join('.', 'test_udivmod64.run')
+
+def testcase(dividend, divisor):
+    print 'divmod64(%d, %d)' % (dividend, divisor)
+
+    procargs = ('%s %s %s' % (udt, dividend, divisor)).split()
+    result = subprocess.check_output(procargs)
+    gotQ, gotR = map(int, result.splitlines())
+
+    expectQ = dividend // divisor
+    expectR = dividend % divisor
+
+    print 'Q = %d, R = %d' % (gotQ, gotR)
+
+    if expectQ != gotQ:
+        raise ValueError("invalid quotient: got=%d but expect=%d" %
+                         (gotQ, expectQ))
+    if expectR != gotR:
+        raise ValueError("invalid remainder: got=%d but expect=%d" %
+                         (gotR, expectR))
+    print 'OK'
+
+def testsequence():
+    subjects = [
+        (0, 1),
+        (0, 0xffffffffffffffff),
+        (1, 2),
+        (1, 983219),
+        (2, 2),
+        (3, 2),
+        (1024, 2),
+        (2048, 512),
+        (21321, 512),
+        (9329189, 1031),
+        (0xffffffff, 2),
+        (0xffffffff, 0xffff),
+        (0x1ffffffff, 2),
+        (0x1ffffffff, 0xffff),
+        (0xffff, 0xffffffff),
+        (0xffffffffffffffff, 0xffff),
+        (0xffffffffffffffff, 0x7fffffffffffffff),
+        (0xffffffffffffffff, 0xfffffffffffffff0),
+        (0xffffffffffffffff, 87655678587161901),
+    ]
+
+    for dvd, dvr in subjects:
+        testcase(dvd, dvr)
+
+if __name__ == '__main__':
+    testsequence()

llrtc/lib/udivmod64.c

@@ -0,0 +1,84 @@
+/*
+Implements unsigned divmod using for platform missing 64-bit division and/or
+modulo functions.
+*/
+#include "llrt.h"
+
+/*
+count left zero for 64-bit words
+
+*/
+static
+int clz64(uint64_t x)
+{
+    const int total_bits = sizeof(x) * BITS_PER_BYTE;
+    int zc = 0;
+
+    while (zc < total_bits && ((x >> (total_bits - zc - 1)) & 1) == 0) {
+        ++zc;
+    }
+    return zc;
+}
+
+typedef struct div_state_
+{
+    uint64_t tmp, dvd;
+} div_state;
+
+/*
+Left shift div_state by 1 bit
+*/
+static
+void div_state_lshift(div_state *state)
+{
+    state->tmp = (state->tmp << 1) | (state->dvd >> 63);
+    state->dvd = state->dvd << 1;
+}
+
+/*
+Division of unsigned 64-bit word using 64-bit addition and subtration following
+the shift-restore division algorithm.
+For those interested in 32-bit implementation,
+mapping of 64-bit addition and subtraction to 32-bit should be trivial.
+
+Reference: 
+    - IBM. The PowerPC Compiler Writer's Guide
+    - LLVM compiler-rt
+
+Assumptions:
+    - all operands and results are positive
+    - unsigned wrapped around
+*/
+uint64_t udivmod64(uint64_t dividend, uint64_t divisor, uint64_t *remainder)
+{
+    div_state state = {0, dividend};
+    uint64_t quotient = 0;
+    int i;
+    int skipahead;
+
+    if (divisor == 0) {
+        return 1 / 0;               /* intentionally div by zero */
+    }
+
+    /*
+    skipahead to reduce iteration
+    */
+    skipahead = clz64(dividend);
+
+    for (i = 0; i < skipahead; ++i) {
+        div_state_lshift(&state);
+    }
+
+    /*
+    division loop
+    */
+    for (i = skipahead; i < 64; ++i) {
+        div_state_lshift(&state);
+        if (state.tmp >= divisor) {
+            state.tmp = state.tmp - divisor;
+            quotient |= 1ull << (63 - i);
+        }
+    }
+    if (remainder) *remainder = state.tmp;
+    return quotient;
+}

llrtc/tools/striptriple.py

@@ -0,0 +1,15 @@
+import sys
+import re
+
+buf = []
+with open(sys.argv[1], 'r') as fin:
+    tripleline = re.compile('^target\s+triple\s+=\s+')
+    for line in fin.readlines():
+        if not tripleline.match(line):
+            buf.append(line)
+
+with open(sys.argv[1], 'w') as fout:
+    for line in buf:
+        fout.write(line)
+
+

llvm-config-win32.py

@@ -1,76 +1,88 @@
-import sys, os
+import re
+import sys
+from distutils.spawn import find_executable
+from os.path import abspath, dirname, isfile, join
+from os import listdir
+from subprocess import Popen, PIPE
 
-def find_path_of(filename, envvar='PATH'):
-    """Finds the path from $PATH where the file exists, returns None if not found."""
-    pathlist = os.getenv(envvar).split(os.pathsep)
-    for path in pathlist:
-        if os.path.exists(os.path.join(path, filename)):
-            return os.path.abspath(path)
-    return None
 
-if sys.argv[1] == '--version':
-    cmd = 'llvm-tblgen --version'
-    # Hardcoded extraction, only tested on llvm 3.1
-    result = os.popen(cmd).read().split('\n')[1].strip().split(' ')[2]
-    print result
-elif sys.argv[1] == '--libs':
+def find_llvm_tblgen():
+    path = find_executable('llvm-tblgen')
+    if path is None:
+        sys.exit('Error: could not locate llvm-tblgen')
+    return path
+
+
+def find_llvm_prefix():
+    return abspath(dirname(dirname(find_llvm_tblgen())))
+
+
+def ensure_file(path):
+    if not isfile(path):
+        sys.exit('Error: no file: %r' % path)
+
+
+def get_llvm_version():
+    args = [find_llvm_tblgen(), '--version']
+    p = Popen(args, stdout=PIPE, stderr=PIPE)
+    stdout, stderr = p.communicate()
+    if stderr:
+        sys.exit("Error: %r stderr is:\n%s" % (args, stderr.decode()))
+    out = stdout.decode().strip()
+    pat = re.compile(r'llvm\s+version\s+(\d+\.\d+\S*)', re.I)
+    m = pat.search(out)
+    if m is None:
+        sys.exit('Error: could not parse version in:' + out)
+    return m.group(1)
+
+
+def libs_options():
     # NOTE: instead of actually looking at the components requested,
-    #       we just spit out a bunch of libs
+    #       we just print out a bunch of libs
     for lib in """
-LLVMAnalysis
-LLVMAsmParser
-LLVMAsmPrinter
-LLVMBitReader
-LLVMBitWriter
-LLVMCodeGen
-LLVMCore
-LLVMExecutionEngine
-LLVMInstCombine
-LLVMInstrumentation
-LLVMInterpreter
-LLVMipa
-LLVMipo
-LLVMJIT
-LLVMLinker
-LLVMMC
-LLVMMCParser
-LLVMScalarOpts
-LLVMSelectionDAG
-LLVMSupport
-LLVMTarget
-LLVMTransformUtils
-LLVMVectorize
-LLVMX86AsmParser
-LLVMX86AsmPrinter
-LLVMX86CodeGen
-LLVMX86Desc
-LLVMX86Info
-LLVMX86Utils
 Advapi32
 Shell32
 """.split():
         print('-l%s' % lib)
-        # Look for the PTX .lib in %LIBPATH%
-        if find_path_of('LLVMPTXCodeGen.lib', 'LIBPATH') != None:
-            print('-lLLVMPTXAsmPrinter')
-            print('-lLLVMPTXCodeGen')
-            print('-lLLVMPTXDesc')
-            print('-lLLVMPTXInfo')
-elif sys.argv[1] == '--includedir': 
-    llvmbin = find_path_of('llvm-tblgen.exe')
-    if llvmbin is None:
-        raise RuntimeError('Could not find LLVM')
-    incdir = os.path.abspath(os.path.join(llvmbin, '../include'))
-    if not os.path.exists(os.path.join(incdir, 'llvm/BasicBlock.h')):
-        raise RuntimeError('Could not find LLVM include dir')
-    print incdir
-elif sys.argv[1] == '--libdir': 
-    llvmbin = find_path_of('llvm-tblgen.exe')
-    if llvmbin is None:
-        raise RuntimeError('Could not find LLVM')
-    libdir = os.path.abspath(os.path.join(llvmbin, '../lib'))
-    if not os.path.exists(os.path.join(libdir, 'LLVMCore.lib')):
-        raise RuntimeError('Could not find LLVM lib dir')
-    print libdir
+
+    bpath = join(find_llvm_prefix(), 'lib')
+    for filename in listdir(bpath):
+        filepath = join(bpath, filename)
+        if isfile(filepath) and filename.endswith('.lib') and filename.startswith('LLVM'):
+            name = filename.split('.', 1)[0]
+            print('-l%s' % name)
+
+def main():
+    try:
+        option = sys.argv[1]
+    except IndexError:
+        sys.exit('Error: option missing')
+
+    if option == '--version':
+        print(get_llvm_version())
+
+    elif option == '--targets-built':
+        print('X86')  # just do X86
+
+    elif option == '--libs':
+        libs_options()
+
+    elif option == '--includedir':
+        incdir = join(find_llvm_prefix(), 'include')
+        ensure_file(join(incdir, 'llvm' , 'Linker.h'))
+        print(incdir)
+
+    elif option == '--libdir':
+        libdir = join(find_llvm_prefix(), 'lib')
+        ensure_file(join(libdir, 'LLVMCore.lib'))
+        print(libdir)
+
+    elif option in ('--ldflags', '--components'):
+        pass
+
     else:
-    raise RuntimeError('Unrecognized llvm-config command %s' % sys.argv[1])
+        sys.exit('Error: Unrecognized llvm-config option %r' % option)
+
+
+if __name__ == '__main__':
+    main()

llvm/__init__.py

@@ -1,140 +1,47 @@
-"""
-Common classes related to LLVM.
-"""
-
-__version__ = '0.8.3'
+from ._version import get_versions
+__version__ = get_versions()['version']
+del get_versions
 
 
-from weakref import WeakValueDictionary
+from llvmpy import extra
+
+version = extra.get_llvm_version()
+del extra
+
+class Wrapper(object):
+    __slots__ = '__ptr'
+
+    def __init__(self, ptr):
+        assert ptr
+        self.__ptr = ptr
+
+    @property
+    def _ptr(self):
+        try:
+            return self.__ptr
+        except AttributeError:
+            raise AttributeError("_ptr resource has been removed")
+
+    @_ptr.deleter
+    def _ptr(self):
+        del self.__ptr
 
 
-#===----------------------------------------------------------------------===
-# Exceptions
-#===----------------------------------------------------------------------===
+def _extract_ptrs(objs):
+    return [(x._ptr if x is not None else None)
+            for x in objs]
 
 class LLVMException(Exception):
-    """Generic LLVM exception."""
-
-    def __init__(self, msg=""):
-        Exception.__init__(self, msg)
-
-
-#===----------------------------------------------------------------------===
-# Ownables
-#===----------------------------------------------------------------------===
-
-class Ownable(object):
-    """Objects that can be owned.
-
-    Modules and Module Providers can be owned, i.e., the responsibility of
-    destruction of ownable objects can be handed over to other objects. The
-    llvm.Ownable class represents objects that can be so owned. This class
-    is NOT intended for public use.
-    """
-
-    def __init__(self, ptr, del_fn):
-        self.ptr = ptr
-        self.owner = None
-        self.del_fn = del_fn
-
-    def _own(self, owner):
-        if self.owner:
-            raise LLVMException("object already owned")
-        self.owner = owner
-
-    def _disown(self):
-        if not self.owner:
-            raise LLVMException("not owned")
-        self.owner = None
-
-    def __del__(self):
-        if not self.owner:
-            self.del_fn(self.ptr)
-
-
-#===----------------------------------------------------------------------===
-# Dummy owner, will not delete ownee. Be careful.
-#===----------------------------------------------------------------------===
-
-class DummyOwner(object):
     pass
 
-
-#===----------------------------------------------------------------------===
-# A metaclass to prevent aliasing.  It stores a (weak) reference to objects
-# constructed based on a PyCObject.  If an object is constructed based on a
-# PyCObject with the same underlying pointer as a previous object, a reference
-# to the previous object is returned rather than a new one.
-#===----------------------------------------------------------------------===
-
-class _ObjectCache(type):
-    """A metaclass to prevent aliasing.
-
-    Classes using 'ObjectCache' as a metaclass must have constructors
-    that take a PyCObject as their first argument.  When the class is
-    called (to create a new instance of the class), the value of the
-    pointer wrapped by the PyCObj is checked:
-
-        If no previous object has been created based on the same
-        underlying pointer (note that different PyCObject objects can
-        wrap the same pointer), the object will be initialized as
-        usual and returned.
-
-        If a previous has been created based on the same pointer,
-        then a reference to that object will be returned, and no
-        object initialization is performed.
-    """
-
-    __instances = WeakValueDictionary()
-
-    def __call__(cls, ptr, *args, **kwargs):
-        objid = _core.PyCObjectVoidPtrToPyLong(ptr)
-        key = "%s:%d" % (cls.__name__, objid)
-        obj = _ObjectCache.__instances.get(key)
-        if obj is None:
-            obj = super(_ObjectCache, cls).__call__(ptr, *args, **kwargs)
-            _ObjectCache.__instances[key] = obj
-        return obj
-
-    @staticmethod
-    def forget(obj):
-        objid = _core.PyCObjectVoidPtrToPyLong(obj.ptr)
-        key = "%s:%d" % (type(obj).__name__, objid)
-        if key in _ObjectCache.__instances:
-            del _ObjectCache.__instances[key]
-
-
-#===----------------------------------------------------------------------===
-# Cacheables
-#===----------------------------------------------------------------------===
-
-# version 2/3 compatibility help
-#   version 2 metaclass
-#       class Cacheable(object):
-#           __metaclass__ = _ObjectCache  # Doing nothing for version 3
-#
-#   version 3 metaclass
-#       class Cacheable(metaclass=_ObjectCache):
-#
-# Reference: http://mikewatkins.ca/2008/11/29/python-2-and-3-metaclasses/#using-the-metaclass-in-python-3-x
-ObjectCache = _ObjectCache('ObjectCache', (object, ), {})
-
-class Cacheable(ObjectCache):
-    """Objects that can be cached.
-
-    Objects that wrap a PyCObject are cached to avoid "aliasing", i.e.,
-    two Python objects each containing a PyCObject which internally points
-    to the same C pointer."""
-
-    def forget(self):
-        ObjectCache.forget(self)
-
-
-def test(verbosity=1):
+def test(verbosity=3, run_isolated=True):
     """test(verbosity=1) -> TextTestResult
 
-    Run self-test, and return unittest.runner.TextTestResult object.
+        Run self-test, and return the number of failures + errors
         """
-    from llvm.test_llvmpy import run
+    from llvm.tests import run
 
-    return run(verbosity=verbosity)
+    result = run(verbosity=verbosity, run_isolated=run_isolated)
+    errct = len(result.failures) + len(result.errors)
+
+    return errct

llvm/_util.py

@@ -1,103 +0,0 @@
-#
-# Copyright (c) 2008-10, Mahadevan R All rights reserved.
-#
-# Redistribution and use in source and binary forms, with or without
-# modification, are permitted provided that the following conditions are met:
-#
-#  * Redistributions of source code must retain the above copyright notice,
-#    this list of conditions and the following disclaimer.
-#
-#  * Redistributions in binary form must reproduce the above copyright notice,
-#    this list of conditions and the following disclaimer in the documentation
-#    and/or other materials provided with the distribution.
-#
-#  * Neither the name of this software, nor the names of its
-#    contributors may be used to endorse or promote products derived from
-#    this software without specific prior written permission.
-#
-# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-# DATA, OR PROFITS; OR BUSINESS 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.
-#
-
-"""Utility functions and classes.
-
-Used only in other modules, not for public use."""
-
-import llvm
-import llvm._core as _core  # for PyCObjectVoidPtrToPyLong
-
-
-#===----------------------------------------------------------------------===
-# A set of helpers to check various things. Raises exceptions on
-# failures.
-#===----------------------------------------------------------------------===
-
-def check_gen(obj, typ):
-    if not isinstance(obj, typ):
-        typ_str = typ.__name__
-        msg = "argument not an instance of llvm.core.%s" % typ_str
-        raise TypeError(msg)
-
-def check_is_unowned(ownable):
-    if ownable.owner:
-        raise llvm.LLVMException("object is already owned")
-
-
-#===----------------------------------------------------------------------===
-# A set of helpers to unpack a list of Python wrapper objects
-# into a list of PyCObject wrapped objects, checking types along
-# the way.
-#===----------------------------------------------------------------------===
-
-def unpack_gen(objlist, check_fn):
-    for obj in objlist: check_fn(obj)
-    return [ obj.ptr for obj in objlist ]
-
-def unpack_gen_allow_none(objlist, check_fn):
-    for obj in objlist:
-        if obj is not None:
-            check_fn(obj)
-    return [ (obj.ptr if obj is not None else None) for obj in objlist ]
-
-#===----------------------------------------------------------------------===
-# Helper to wrap over iterables (LLVMFirstXXX, LLVMNextXXX). This used
-# to be a generator, but that loses subscriptability of the result, so
-# we now return a list.
-#===----------------------------------------------------------------------===
-
-def wrapiter(first, next, container, wrapper):
-    ret = []
-    ptr = first(container)
-    while ptr:
-        ret.append(wrapper(ptr))
-        ptr = next(ptr)
-    return ret
-
-
-#===----------------------------------------------------------------------===
-# Py2/3 compatibility string check
-#===----------------------------------------------------------------------===
-def _isstring_py2(x):
-    return isinstance(x, basestring)
-
-def _isstring_py3(x):
-    return isinstance(x, str)
-
-def _isstring_choose():
-    try:
-        basestring
-        return _isstring_py2
-    except:
-        return _isstring_py3
-
-isstring = _isstring_choose()
-

llvm/_version.py

@@ -0,0 +1,193 @@
+
+IN_LONG_VERSION_PY = True
+# This file helps to compute a version number in source trees obtained from
+# git-archive tarball (such as those provided by github's download-from-tag
+# feature). Distribution tarballs (build by setup.py sdist) and build
+# directories (produced by setup.py build) will contain a much shorter file
+# that just contains the computed version number.
+
+# This file is released into the public domain. Generated by
+# versioneer-0.7+ (https://github.com/warner/python-versioneer)
+
+# these strings will be replaced by git during git-archive
+git_refnames = "$Format:%d$"
+git_full = "$Format:%H$"
+GIT = "git"
+
+import subprocess
+import sys
+
+def run_command(args, cwd=None, verbose=False):
+    try:
+        # remember shell=False, so use git.cmd on windows, not just git
+        p = subprocess.Popen(args, stdout=subprocess.PIPE, cwd=cwd)
+    except EnvironmentError:
+        e = sys.exc_info()[1]
+        if verbose:
+            print("unable to run %s" % args[0])
+            print(e)
+        return None
+    stdout = p.communicate()[0].strip()
+    if sys.version >= '3':
+        stdout = stdout.decode()
+    if p.returncode != 0:
+        if verbose:
+            print("unable to run %s (error)" % args[0])
+        return None
+    return stdout
+
+
+import sys
+import re
+import os.path
+
+def get_expanded_variables(versionfile_source):
+    # the code embedded in _version.py can just fetch the value of these
+    # variables. When used from setup.py, we don't want to import
+    # _version.py, so we do it with a regexp instead. This function is not
+    # used from _version.py.
+    variables = {}
+    try:
+        for line in open(versionfile_source,"r").readlines():
+            if line.strip().startswith("git_refnames ="):
+                mo = re.search(r'=\s*"(.*)"', line)
+                if mo:
+                    variables["refnames"] = mo.group(1)
+            if line.strip().startswith("git_full ="):
+                mo = re.search(r'=\s*"(.*)"', line)
+                if mo:
+                    variables["full"] = mo.group(1)
+    except EnvironmentError:
+        pass
+    return variables
+
+def versions_from_expanded_variables(variables, tag_prefix, verbose=False):
+    refnames = variables["refnames"].strip()
+    if refnames.startswith("$Format"):
+        if verbose:
+            print("variables are unexpanded, not using")
+        return {} # unexpanded, so not in an unpacked git-archive tarball
+    refs = set([r.strip() for r in refnames.strip("()").split(",")])
+    for ref in list(refs):
+        if not re.search(r'\d', ref):
+            if verbose:
+                print("discarding '%s', no digits" % ref)
+            refs.discard(ref)
+            # Assume all version tags have a digit. git's %d expansion
+            # behaves like git log --decorate=short and strips out the
+            # refs/heads/ and refs/tags/ prefixes that would let us
+            # distinguish between branches and tags. By ignoring refnames
+            # without digits, we filter out many common branch names like
+            # "release" and "stabilization", as well as "HEAD" and "master".
+    if verbose:
+        print("remaining refs: %s" % ",".join(sorted(refs)))
+    for ref in sorted(refs):
+        # sorting will prefer e.g. "2.0" over "2.0rc1"
+        if ref.startswith(tag_prefix):
+            r = ref[len(tag_prefix):]
+            if verbose:
+                print("picking %s" % r)
+            return { "version": r,
+                     "full": variables["full"].strip() }
+    # no suitable tags, so we use the full revision id
+    if verbose:
+        print("no suitable tags, using full revision id")
+    return { "version": variables["full"].strip(),
+             "full": variables["full"].strip() }
+
+def versions_from_vcs(tag_prefix, versionfile_source, verbose=False):
+    # this runs 'git' from the root of the source tree. That either means
+    # someone ran a setup.py command (and this code is in versioneer.py, so
+    # IN_LONG_VERSION_PY=False, thus the containing directory is the root of
+    # the source tree), or someone ran a project-specific entry point (and
+    # this code is in _version.py, so IN_LONG_VERSION_PY=True, thus the
+    # containing directory is somewhere deeper in the source tree). This only
+    # gets called if the git-archive 'subst' variables were *not* expanded,
+    # and _version.py hasn't already been rewritten with a short version
+    # string, meaning we're inside a checked out source tree.
+
+    try:
+        here = os.path.abspath(__file__)
+    except NameError:
+        # some py2exe/bbfreeze/non-CPython implementations don't do __file__
+        return {} # not always correct
+
+    # versionfile_source is the relative path from the top of the source tree
+    # (where the .git directory might live) to this file. Invert this to find
+    # the root from __file__.
+    root = here
+    if IN_LONG_VERSION_PY:
+        for i in range(len(versionfile_source.split("/"))):
+            root = os.path.dirname(root)
+    else:
+        root = os.path.dirname(here)
+    if not os.path.exists(os.path.join(root, ".git")):
+        if verbose:
+            print("no .git in %s" % root)
+        return {}
+
+    stdout = run_command([GIT, "describe", "--tags", "--dirty", "--always"],
+                         cwd=root)
+    if stdout is None:
+        return {}
+    if not stdout.startswith(tag_prefix):
+        if verbose:
+            print("tag '%s' doesn't start with prefix '%s'" % (stdout, tag_prefix))
+        return {}
+    tag = stdout[len(tag_prefix):]
+    stdout = run_command([GIT, "rev-parse", "HEAD"], cwd=root)
+    if stdout is None:
+        return {}
+    full = stdout.strip()
+    if tag.endswith("-dirty"):
+        full += "-dirty"
+    return {"version": tag, "full": full}
+
+
+def versions_from_parentdir(parentdir_prefix, versionfile_source, verbose=False):
+    if IN_LONG_VERSION_PY:
+        # We're running from _version.py. If it's from a source tree
+        # (execute-in-place), we can work upwards to find the root of the
+        # tree, and then check the parent directory for a version string. If
+        # it's in an installed application, there's no hope.
+        try:
+            here = os.path.abspath(__file__)
+        except NameError:
+            # py2exe/bbfreeze/non-CPython don't have __file__
+            return {} # without __file__, we have no hope
+        # versionfile_source is the relative path from the top of the source
+        # tree to _version.py. Invert this to find the root from __file__.
+        root = here
+        for i in range(len(versionfile_source.split("/"))):
+            root = os.path.dirname(root)
+    else:
+        # we're running from versioneer.py, which means we're running from
+        # the setup.py in a source tree. sys.argv[0] is setup.py in the root.
+        here = os.path.abspath(sys.argv[0])
+        root = os.path.dirname(here)
+
+    # Source tarballs conventionally unpack into a directory that includes
+    # both the project name and a version string.
+    dirname = os.path.basename(root)
+    if not dirname.startswith(parentdir_prefix):
+        if verbose:
+            print("guessing rootdir is '%s', but '%s' doesn't start with prefix '%s'" %
+                  (root, dirname, parentdir_prefix))
+        return None
+    return {"version": dirname[len(parentdir_prefix):], "full": ""}
+
+tag_prefix = ""
+parentdir_prefix = "llvmpy-"
+versionfile_source = "llvm/_version.py"
+
+def get_versions(default={"version": "unknown", "full": ""}, verbose=False):
+    variables = { "refnames": git_refnames, "full": git_full }
+    ver = versions_from_expanded_variables(variables, tag_prefix, verbose)
+    if not ver:
+        ver = versions_from_vcs(tag_prefix, versionfile_source, verbose)
+    if not ver:
+        ver = versions_from_parentdir(parentdir_prefix, versionfile_source,
+                                      verbose)
+    if not ver:
+        ver = default
+    return ver

llvm/capsulethunk.h

@@ -1,134 +0,0 @@
-#ifndef __CAPSULETHUNK_H
-#define __CAPSULETHUNK_H
-
-#if (    (PY_VERSION_HEX <  0x02070000) \
-     || ((PY_VERSION_HEX >= 0x03000000) \
-      && (PY_VERSION_HEX <  0x03010000)) )
-
-#define __PyCapsule_GetField(capsule, field, default_value) \
-    ( PyCapsule_CheckExact(capsule) \
-        ? (((PyCObject *)capsule)->field) \
-        : (default_value) \
-    ) \
-
-#define __PyCapsule_SetField(capsule, field, value) \
-    ( PyCapsule_CheckExact(capsule) \
-        ? (((PyCObject *)capsule)->field = value), 1 \
-        : 0 \
-    ) \
-
-
-#define PyCapsule_Type PyCObject_Type
-
-#define PyCapsule_CheckExact(capsule) (PyCObject_Check(capsule))
-#define PyCapsule_IsValid(capsule, name) (PyCObject_Check(capsule))
-
-
-#define PyCapsule_New(pointer, name, destructor) \
-    (PyCObject_FromVoidPtr(pointer, destructor))
-
-
-#define PyCapsule_GetPointer(capsule, name) \
-    (PyCObject_AsVoidPtr(capsule))
-
-/* Don't call PyCObject_SetPointer here, it fails if there's a destructor */
-#define PyCapsule_SetPointer(capsule, pointer) \
-    __PyCapsule_SetField(capsule, cobject, pointer)
-
-
-#define PyCapsule_GetDestructor(capsule) \
-    __PyCapsule_GetField(capsule, destructor)
-
-#define PyCapsule_SetDestructor(capsule, dtor) \
-    __PyCapsule_SetField(capsule, destructor, dtor)
-
-
-/*
- * Sorry, there's simply no place
- * to store a Capsule "name" in a CObject.
- */
-#define PyCapsule_GetName(capsule) NULL
-
-static int
-PyCapsule_SetName(PyObject *capsule, const char *unused)
-{
-    unused = unused;
-    PyErr_SetString(PyExc_NotImplementedError,
-        "can't use PyCapsule_SetName with CObjects");
-    return 1;
-}
-
-
-
-#define PyCapsule_GetContext(capsule) \
-    __PyCapsule_GetField(capsule, descr)
-
-#define PyCapsule_SetContext(capsule, context) \
-    __PyCapsule_SetField(capsule, descr, context)
-
-
-static void *
-PyCapsule_Import(const char *name, int no_block)
-{
-    PyObject *object = NULL;
-    void *return_value = NULL;
-    char *trace;
-    size_t name_length = (strlen(name) + 1) * sizeof(char);
-    char *name_dup = (char *)PyMem_MALLOC(name_length);
-
-    if (!name_dup) {
-        return NULL;
-    }
-
-    memcpy(name_dup, name, name_length);
-
-    trace = name_dup;
-    while (trace) {
-        char *dot = strchr(trace, '.');
-        if (dot) {
-            *dot++ = '\0';
-        }
-
-        if (object == NULL) {
-            if (no_block) {
-                object = PyImport_ImportModuleNoBlock(trace);
-            } else {
-                object = PyImport_ImportModule(trace);
-                if (!object) {
-                    PyErr_Format(PyExc_ImportError,
-                        "PyCapsule_Import could not "
-                        "import module \"%s\"", trace);
-                }
-            }
-        } else {
-            PyObject *object2 = PyObject_GetAttrString(object, trace);
-            Py_DECREF(object);
-            object = object2;
-        }
-        if (!object) {
-            goto EXIT;
-        }
-
-        trace = dot;
-    }
-
-    if (PyCObject_Check(object)) {
-        PyCObject *cobject = (PyCObject *)object;
-        return_value = cobject->cobject;
-    } else {
-        PyErr_Format(PyExc_AttributeError,
-            "PyCapsule_Import \"%s\" is not valid",
-            name);
-    }
-
-EXIT:
-    Py_XDECREF(object);
-    if (name_dup) {
-        PyMem_FREE(name_dup);
-    }
-    return return_value;
-}
-
-#endif /* #if PY_VERSION_HEX < 0x02070000 */
-
-#endif /* __CAPSULETHUNK_H */