Update README.rst

Fix ImportError for C extensions on Python 3.4b2

CHANGELOG

@@ -1,3 +1,40 @@
+2014-04-28    0.12.5:
+---------------------
+  * Fixes memory leaks (#92)
+  * Fixes tarball (#99)
+
+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

MANIFEST.in

@@ -1,5 +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

@@ -30,10 +30,13 @@ Quickstart
     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`` to build.
+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::
 

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/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:
 
@@ -22,7 +22,7 @@ Contents:
 
 
 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

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 from __future__ import absolute_import
 import opcode

llpython/byte_flow.py

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 from __future__ import absolute_import
 import dis

llpython/byte_translator.py

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 '''Defines a bytecode based LLVM translator for llpython code.
 '''

llpython/bytecode_visitor.py

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 from __future__ import absolute_import
 import itertools

llpython/bytetype.py

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 
 import ctypes

llpython/control_flow.py

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 
 import pprint

llpython/gen_bytecode_visitor.py

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 from __future__ import absolute_import
 from . import opcode_util

llpython/nobitey.py

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 from __future__ import absolute_import
 import sys

llpython/opcode_util.py

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 
 import dis

llpython/phi_injector.py

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 
 from .bytecode_visitor import BytecodeFlowVisitor, BenignBytecodeVisitorMixin

llpython/pyaddfunc.py

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 
 import ctypes

llpython/tests/llfuncs.py

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 
 def doslice (in_string, lower, upper):

llpython/tests/llfunctys.py

@@ -1,4 +1,3 @@
-#! /usr/bin/env python
 # ______________________________________________________________________
 
 import llvm.core as lc

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

@@ -2,6 +2,7 @@ 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
 
 
@@ -39,55 +40,17 @@ def libs_options():
     # NOTE: instead of actually looking at the components requested,
     #       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
-LLVMObject
-LLVMRuntimeDyld
-LLVMScalarOpts
-LLVMSelectionDAG
-LLVMSupport
-LLVMTarget
-LLVMTransformUtils
-LLVMVectorize
-LLVMX86AsmParser
-LLVMX86AsmPrinter
-LLVMX86CodeGen
-LLVMX86Desc
-LLVMX86Disassembler
-LLVMX86Info
-LLVMX86Utils
 Advapi32
 Shell32
 """.split():
         print('-l%s' % lib)
 
-    if isfile(join(find_llvm_prefix(), 'lib', 'LLVMPTXCodeGen.lib')):
+    bpath = join(find_llvm_prefix(), 'lib')
-        print('-lLLVMPTXAsmPrinter')
+    for filename in listdir(bpath):
-        print('-lLLVMPTXCodeGen')
+        filepath = join(bpath, filename)
-        print('-lLLVMPTXDesc')
+        if isfile(filepath) and filename.endswith('.lib') and filename.startswith('LLVM'):
-        print('-lLLVMPTXInfo')
+            name = filename.split('.', 1)[0]
-
+            print('-l%s' % name)
-    elif isfile(join(find_llvm_prefix(), 'lib', 'LLVMNVPTXCodeGen.lib')):
-        print('-lLLVMNVPTXAsmPrinter')
-        print('-lLLVMNVPTXCodeGen')
-        print('-lLLVMNVPTXDesc')
-        print('-lLLVMNVPTXInfo')
-
 
 def main():
     try:
@@ -98,12 +61,15 @@ def main():
     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' , 'BasicBlock.h'))
+        ensure_file(join(incdir, 'llvm' , 'Linker.h'))
         print(incdir)
 
     elif option == '--libdir':

llvm/__init__.py

@@ -4,10 +4,13 @@ del get_versions
 
 
 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
@@ -31,14 +34,14 @@ def _extract_ptrs(objs):
 class LLVMException(Exception):
     pass
 
-def test(verbosity=1):
+def test(verbosity=3, run_isolated=True):
     """test(verbosity=1) -> TextTestResult
 
         Run self-test, and return the number of failures + errors
         """
-    from llvm.test_llvmpy import run
+    from llvm.tests import run
 
-    result = run(verbosity=verbosity)
+    result = run(verbosity=verbosity, run_isolated=run_isolated)
-
+    errct = len(result.failures) + len(result.errors)
-    return len(result.failures) + len(result.errors)
 
+    return errct

llvm/deprecated.py

@@ -0,0 +1,25 @@
+"""
+Shameless borrowed from Smart_deprecation_warnings
+https://wiki.python.org/moin/PythonDecoratorLibrary
+"""
+
+import warnings
+import functools
+
+
+def deprecated(func):
+    """This is a decorator which can be used to mark functions
+    as deprecated. It will result in a warning being emitted
+    when the function is used."""
+
+    @functools.wraps(func)
+    def new_func(*args, **kwargs):
+        warnings.warn_explicit(
+            "Call to deprecated function %s." % (func.__name__,),
+            category=DeprecationWarning,
+            filename=func.func_code.co_filename,
+            lineno=func.func_code.co_firstlineno + 1
+        )
+        return func(*args, **kwargs)
+
+    return new_func

llvm/ee.py

@@ -30,33 +30,23 @@
 
 "Execution Engine and related classes."
 
-from io import BytesIO
+import sys
-import contextlib
 
 import llvm
 from llvm import core
 from llvm.passes import TargetData, TargetTransformInfo
 from llvmpy import api, extra
+
 #===----------------------------------------------------------------------===
-# Enumerations
+# import items which were moved to target module
 #===----------------------------------------------------------------------===
+from llvm.target import (initialize_all, initialize_target,
+    print_registered_targets, get_host_cpu_name, get_default_triple,
+    TargetMachine,
+    BO_BIG_ENDIAN, BO_LITTLE_ENDIAN,
+    CM_DEFAULT, CM_JITDEFAULT, CM_SMALL, CM_KERNEL, CM_MEDIUM, CM_LARGE,
+    RELOC_DEFAULT, RELOC_STATIC, RELOC_PIC, RELOC_DYNAMIC_NO_PIC)
 
-BO_BIG_ENDIAN       = 0
-BO_LITTLE_ENDIAN    = 1
-
-# CodeModel
-CM_DEFAULT      = api.llvm.CodeModel.Model.Default
-CM_JITDEFAULT   = api.llvm.CodeModel.Model.JITDefault
-CM_SMALL        = api.llvm.CodeModel.Model.Small
-CM_KERNEL       = api.llvm.CodeModel.Model.Kernel
-CM_MEDIUM       = api.llvm.CodeModel.Model.Medium
-CM_LARGE        = api.llvm.CodeModel.Model.Large
-
-# Reloc
-RELOC_DEFAULT        = api.llvm.Reloc.Model.Default
-RELOC_STATIC         = api.llvm.Reloc.Model.Static
-RELOC_PIC            = api.llvm.Reloc.Model.PIC_
-RELOC_DYNAMIC_NO_PIC = api.llvm.Reloc.Model.DynamicNoPIC
 
 #===----------------------------------------------------------------------===
 # Generic value
@@ -150,9 +140,17 @@ class EngineBuilder(llvm.Wrapper):
         '''
         if tm is not None:
             engine = self._ptr.create(tm._ptr)
+        elif (sys.platform.startswith('win32') and
+                    getattr(self, '_use_mcjit', False)):
+            # force ELF generation on MCJIT on win32
+            triple = get_default_triple()
+            tm = TargetMachine.new('%s-elf' % triple)
+            engine = self._ptr.create(tm._ptr)
         else:
             engine = self._ptr.create()
-        return ExecutionEngine(engine)
+        ee = ExecutionEngine(engine)
+        ee.finalize_object()                # no effect for legacy JIT
+        return ee
 
     def select_target(self, *args):
         '''get the corresponding target machine
@@ -167,6 +165,12 @@ class EngineBuilder(llvm.Wrapper):
             ptr = self._ptr.selectTarget()
         return TargetMachine(ptr)
 
+    def mcjit(self, enable):
+        '''Enable/disable MCJIT
+        '''
+        self._ptr.setUseMCJIT(enable)
+        self._use_mcjit = True
+        return self
 
 #===----------------------------------------------------------------------===
 # Execution engine
@@ -188,6 +192,9 @@ class ExecutionEngine(llvm.Wrapper):
         ptr = self._ptr.runFunction(fn._ptr, list(map(lambda x: x._ptr, args)))
         return GenericValue(ptr)
 
+    def get_pointer_to_named_function(self, name, abort=True):
+        return self._ptr.getPointerToNamedFunction(name, abort)
+
     def get_pointer_to_function(self, fn):
         return self._ptr.getPointerToFunction(fn._ptr)
 
@@ -213,151 +220,45 @@ class ExecutionEngine(llvm.Wrapper):
     def remove_module(self, module):
         return self._ptr.removeModule(module._ptr)
 
+    def finalize_object(self):
+        return self._ptr.finalizeObject()
+
     @property
     def target_data(self):
         ptr = self._ptr.getDataLayout()
         return TargetData(ptr)
 
+
 #===----------------------------------------------------------------------===
-# Target machine
+# Dynamic Library
 #===----------------------------------------------------------------------===
 
-def initialize_target(target, noraise=False):
+def dylib_add_symbol(name, ptr):
-    """Initialize target by name.
+    api.llvm.sys.DynamicLibrary.AddSymbol(name, ptr)
-    It is safe to initialize the same target multiple times.
+
+def dylib_address_of_symbol(name):
+    return api.llvm.sys.DynamicLibrary.SearchForAddressOfSymbol(name)
+
+def dylib_import_library(filename):
+    """Permanently import a dynamic library.
+
+    Returns a DynamicLibrary object
+
+    Raises RuntimeError
     """
-    prefix = 'LLVMInitialize'
+    return DynamicLibrary(filename)
-    postfixes = ['Target', 'TargetInfo', 'TargetMC', 'AsmPrinter']
-    try:
-        for postfix in postfixes:
-            getattr(api, '%s%s%s' % (prefix, target, postfix))()
-    except AttributeError:
-        if noraise:
-            return False
-        else:
-            raise
-    else:
-        return True
 
 
-def print_registered_targets():
+class DynamicLibrary(object):
-    '''
+    def __init__(self, filename):
-    Note: print directly to stdout
+        """
-    '''
+        Raises RuntimeError
-    api.llvm.TargetRegistry.printRegisteredTargetsForVersion()
+        """
-
+        self._ptr = api.llvm.sys.DynamicLibrary.getPermanentLibrary(
-def get_host_cpu_name():
+            filename)
-    '''return the string name of the host CPU
-    '''
-    return api.llvm.sys.getHostCPUName()
-
-def get_default_triple():
-    '''return the target triple of the host in str-rep
-    '''
-    return api.llvm.sys.getDefaultTargetTriple()
-
-
-class TargetMachine(llvm.Wrapper):
-
-    @staticmethod
-    def new(triple='', cpu='', features='', opt=2, cm=CM_DEFAULT,
-            reloc=RELOC_DEFAULT):
-        if not triple:
-            triple = get_default_triple()
-        if not cpu:
-            cpu = get_host_cpu_name()
-        with contextlib.closing(BytesIO()) as error:
-            target = api.llvm.TargetRegistry.lookupTarget(triple, error)
-            if not target:
-                raise llvm.LLVMException(error)
-            if not target.hasTargetMachine():
-                raise llvm.LLVMException(target, "No target machine.")
-            target_options = api.llvm.TargetOptions.new()
-            tm = target.createTargetMachine(triple, cpu, features,
-                                            target_options,
-                                            reloc, cm, opt)
-            if not tm:
-                raise llvm.LLVMException("Cannot create target machine")
-            return TargetMachine(tm)
-
-    @staticmethod
-    def lookup(arch, cpu='', features='', opt=2, cm=CM_DEFAULT,
-               reloc=RELOC_DEFAULT):
-        '''create a targetmachine given an architecture name
-
-            For a list of architectures,
-            use: `llc -help`
-
-            For a list of available CPUs,
-            use: `llvm-as < /dev/null | llc -march=xyz -mcpu=help`
-
-            For a list of available attributes (features),
-            use: `llvm-as < /dev/null | llc -march=xyz -mattr=help`
-            '''
-        triple = api.llvm.Triple.new()
-        with contextlib.closing(BytesIO()) as error:
-            target = api.llvm.TargetRegistry.lookupTarget(arch, triple, error)
-            if not target:
-                raise llvm.LLVMException(error.getvalue())
-            if not target.hasTargetMachine():
-                raise llvm.LLVMException(target, "No target machine.")
-            target_options = api.llvm.TargetOptions.new()
-            tm = target.createTargetMachine(str(triple), cpu, features,
-                                            target_options,
-                                            reloc, cm, opt)
-            if not tm:
-                raise llvm.LLVMException("Cannot create target machine")
-            return TargetMachine(tm)
-
-    def _emit_file(self, module, cgft):
-        pm = api.llvm.PassManager.new()
-        os = extra.make_raw_ostream_for_printing()
-        pm.add(api.llvm.DataLayout.new(str(self.target_data)))
-        failed = self._ptr.addPassesToEmitFile(pm, os, cgft)
-        pm.run(module)
-
-
-        CGFT = api.llvm.TargetMachine.CodeGenFileType
-        if cgft == CGFT.CGFT_ObjectFile:
-            return os.bytes()
-        else:
-            return os.str()
-
-    def emit_assembly(self, module):
-        '''returns byte string of the module as assembly code of the target machine
-        '''
-        CGFT = api.llvm.TargetMachine.CodeGenFileType
-        return self._emit_file(module._ptr, CGFT.CGFT_AssemblyFile)
-
-    def emit_object(self, module):
-        '''returns byte string of the module as native code of the target machine
-        '''
-        CGFT = api.llvm.TargetMachine.CodeGenFileType
-        return self._emit_file(module._ptr, CGFT.CGFT_ObjectFile)
-
-    @property
-    def target_data(self):
-        '''get target data of this machine
-        '''
-        return TargetData(self._ptr.getDataLayout())
-
-    @property
-    def target_name(self):
-        return self._ptr.getTarget().getName()
-
-    @property
-    def target_short_description(self):
-        return self._ptr.getTarget().getShortDescription()
-
-    @property
-    def triple(self):
-        return self._ptr.getTargetTriple()
-
-    @property
-    def cpu(self):
-        return self._ptr.getTargetCPU()
-
-    @property
-    def feature_string(self):
-        return self._ptr.getTargetFeatureString()
 
+    def get_address_of_symbol(self, symbol):
+        """
+        Get the address of `symbol` (str) as integer
+        """
+        return self._ptr.getAddressOfSymbol(symbol)

llvm/llrt.py

@@ -0,0 +1,79 @@
+import os
+import llvm.core as lc
+import llvm.passes as lp
+import llvm.ee as le
+
+def replace_divmod64(lfunc):
+    '''Replaces all 64-bit integer division (sdiv, udiv) and modulo (srem, urem)
+    '''
+    int64 = lc.Type.int(64)
+    int64ptr = lc.Type.pointer(lc.Type.int(64))
+
+    functy = lc.Type.function(int64, [int64, int64])
+    udiv64 = lfunc.module.get_or_insert_function(functy, '__llrt_udiv64')
+    sdiv64 = lfunc.module.get_or_insert_function(functy, '__llrt_sdiv64')
+    umod64 = lfunc.module.get_or_insert_function(functy, '__llrt_umod64')
+    smod64 = lfunc.module.get_or_insert_function(functy, '__llrt_smod64')
+
+    builder = lc.Builder.new(lfunc.entry_basic_block)
+    for bb in lfunc.basic_blocks:
+        for inst in bb.instructions:
+            if inst.opcode_name == 'sdiv' and inst.type == int64:
+                _replace_with(builder, inst, sdiv64)
+            elif inst.opcode_name == 'udiv' and inst.type == int64:
+                _replace_with(builder, inst, udiv64)
+            elif inst.opcode_name == 'srem' and inst.type == int64:
+                _replace_with(builder, inst, smod64)
+            elif inst.opcode_name == 'urem' and inst.type == int64:
+                _replace_with(builder, inst, umod64)
+
+def _replace_with(builder, inst, func):
+    '''Replace instruction with a call to the function with the same operands
+    as arguments.
+    '''
+    builder.position_before(inst)
+    replacement = builder.call(func, inst.operands)
+    inst.replace_all_uses_with(replacement._ptr)
+    inst.erase_from_parent()
+
+def load(arch):
+    '''Load the LLRT module corresponding to the given architecture
+    Creates a new module and optimizes it using the information from
+    the host machine.
+    '''
+    if arch != 'x86_64':
+        arch = 'x86'
+    path = os.path.join(os.path.dirname(__file__), 'llrt', 'llrt_%s.ll' % arch)
+    with open(path) as fin:
+        lib = lc.Module.from_assembly(fin)
+
+    # run passes to optimize
+    tm = le.TargetMachine.new()
+    pms = lp.build_pass_managers(tm, opt=3, fpm=False)
+    pms.pm.run(lib)
+    return lib
+
+class LLRT(object):
+    def __init__(self):
+        arch = le.get_default_triple().split('-', 1)[0]
+        self.module = load(arch)
+        self.engine = le.EngineBuilder.new(self.module).opt(3).create()
+        self.installed_symbols = set()
+
+    def install_symbols(self):
+        '''Bind all the external symbols to the global symbol map.
+        Any future reference to these symbols will be automatically resolved
+        by LLVM.
+        '''
+        for lfunc in self.module.functions:
+            if lfunc.linkage == lc.LINKAGE_EXTERNAL:
+                mangled = '__llrt_' + lfunc.name
+                self.installed_symbols.add(mangled)
+                ptr = self.engine.get_pointer_to_function(lfunc)
+                le.dylib_add_symbol(mangled, ptr)
+
+    def uninstall_symbols(self):
+        for sym in self.installed_symbols:
+            le.dylib_add_symbol(sym, 0)
+
+

llvm/llrt/llrt_x86.ll

@@ -0,0 +1,371 @@
+; ModuleID = 'udivmod64_x86.bc'
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32-S128"
+
+%struct.div_state_ = type { i64, i64 }
+
+define i64 @udivmod64(i64 %dividend, i64 %divisor, i64* %remainder) nounwind ssp {
+  %1 = alloca i64, align 4
+  %2 = alloca i64, align 8
+  %3 = alloca i64, align 8
+  %4 = alloca i64*, align 4
+  %state = alloca %struct.div_state_, align 4
+  %quotient = alloca i64, align 8
+  %i = alloca i32, align 4
+  %skipahead = alloca i32, align 4
+  store i64 %dividend, i64* %2, align 8
+  store i64 %divisor, i64* %3, align 8
+  store i64* %remainder, i64** %4, align 4
+  %5 = getelementptr inbounds %struct.div_state_* %state, i32 0, i32 0
+  store i64 0, i64* %5, align 4
+  %6 = getelementptr inbounds %struct.div_state_* %state, i32 0, i32 1
+  %7 = load i64* %2, align 8
+  store i64 %7, i64* %6, align 4
+  store i64 0, i64* %quotient, align 8
+  %8 = load i64* %3, align 8
+  %9 = icmp eq i64 %8, 0
+  br i1 %9, label %10, label %11
+
+; <label>:10                                      ; preds = %0
+  store i64 0, i64* %1
+  br label %57
+
+; <label>:11                                      ; preds = %0
+  %12 = load i64* %2, align 8
+  %13 = call i32 @clz64(i64 %12)
+  store i32 %13, i32* %skipahead, align 4
+  store i32 0, i32* %i, align 4
+  br label %14
+
+; <label>:14                                      ; preds = %19, %11
+  %15 = load i32* %i, align 4
+  %16 = load i32* %skipahead, align 4
+  %17 = icmp slt i32 %15, %16
+  br i1 %17, label %18, label %22
+
+; <label>:18                                      ; preds = %14
+  call void @div_state_lshift(%struct.div_state_* %state)
+  br label %19
+
+; <label>:19                                      ; preds = %18
+  %20 = load i32* %i, align 4
+  %21 = add nsw i32 %20, 1
+  store i32 %21, i32* %i, align 4
+  br label %14
+
+; <label>:22                                      ; preds = %14
+  %23 = load i32* %skipahead, align 4
+  store i32 %23, i32* %i, align 4
+  br label %24
+
+; <label>:24                                      ; preds = %45, %22
+  %25 = load i32* %i, align 4
+  %26 = icmp slt i32 %25, 64
+  br i1 %26, label %27, label %48
+
+; <label>:27                                      ; preds = %24
+  call void @div_state_lshift(%struct.div_state_* %state)
+  %28 = getelementptr inbounds %struct.div_state_* %state, i32 0, i32 0
+  %29 = load i64* %28, align 4
+  %30 = load i64* %3, align 8
+  %31 = icmp uge i64 %29, %30
+  br i1 %31, label %32, label %44
+
+; <label>:32                                      ; preds = %27
+  %33 = getelementptr inbounds %struct.div_state_* %state, i32 0, i32 0
+  %34 = load i64* %33, align 4
+  %35 = load i64* %3, align 8
+  %36 = sub i64 %34, %35
+  %37 = getelementptr inbounds %struct.div_state_* %state, i32 0, i32 0
+  store i64 %36, i64* %37, align 4
+  %38 = load i32* %i, align 4
+  %39 = sub nsw i32 63, %38
+  %40 = zext i32 %39 to i64
+  %41 = shl i64 1, %40
+  %42 = load i64* %quotient, align 8
+  %43 = or i64 %42, %41
+  store i64 %43, i64* %quotient, align 8
+  br label %44
+
+; <label>:44                                      ; preds = %32, %27
+  br label %45
+
+; <label>:45                                      ; preds = %44
+  %46 = load i32* %i, align 4
+  %47 = add nsw i32 %46, 1
+  store i32 %47, i32* %i, align 4
+  br label %24
+
+; <label>:48                                      ; preds = %24
+  %49 = load i64** %4, align 4
+  %50 = icmp ne i64* %49, null
+  br i1 %50, label %51, label %55
+
+; <label>:51                                      ; preds = %48
+  %52 = getelementptr inbounds %struct.div_state_* %state, i32 0, i32 0
+  %53 = load i64* %52, align 4
+  %54 = load i64** %4, align 4
+  store i64 %53, i64* %54, align 4
+  br label %55
+
+; <label>:55                                      ; preds = %51, %48
+  %56 = load i64* %quotient, align 8
+  store i64 %56, i64* %1
+  br label %57
+
+; <label>:57                                      ; preds = %55, %10
+  %58 = load i64* %1
+  ret i64 %58
+}
+
+define internal i32 @clz64(i64 %x) nounwind ssp {
+  %1 = alloca i64, align 8
+  %total_bits = alloca i32, align 4
+  %zc = alloca i32, align 4
+  store i64 %x, i64* %1, align 8
+  store i32 64, i32* %total_bits, align 4
+  store i32 0, i32* %zc, align 4
+  br label %2
+
+; <label>:2                                       ; preds = %16, %0
+  %3 = load i32* %zc, align 4
+  %4 = icmp slt i32 %3, 64
+  br i1 %4, label %5, label %14
+
+; <label>:5                                       ; preds = %2
+  %6 = load i64* %1, align 8
+  %7 = load i32* %zc, align 4
+  %8 = sub nsw i32 64, %7
+  %9 = sub nsw i32 %8, 1
+  %10 = zext i32 %9 to i64
+  %11 = lshr i64 %6, %10
+  %12 = and i64 %11, 1
+  %13 = icmp eq i64 %12, 0
+  br label %14
+
+; <label>:14                                      ; preds = %5, %2
+  %15 = phi i1 [ false, %2 ], [ %13, %5 ]
+  br i1 %15, label %16, label %19
+
+; <label>:16                                      ; preds = %14
+  %17 = load i32* %zc, align 4
+  %18 = add nsw i32 %17, 1
+  store i32 %18, i32* %zc, align 4
+  br label %2
+
+; <label>:19                                      ; preds = %14
+  %20 = load i32* %zc, align 4
+  ret i32 %20
+}
+
+define internal void @div_state_lshift(%struct.div_state_* %state) nounwind ssp {
+  %1 = alloca %struct.div_state_*, align 4
+  store %struct.div_state_* %state, %struct.div_state_** %1, align 4
+  %2 = load %struct.div_state_** %1, align 4
+  %3 = getelementptr inbounds %struct.div_state_* %2, i32 0, i32 0
+  %4 = load i64* %3, align 4
+  %5 = shl i64 %4, 1
+  %6 = load %struct.div_state_** %1, align 4
+  %7 = getelementptr inbounds %struct.div_state_* %6, i32 0, i32 1
+  %8 = load i64* %7, align 4
+  %9 = lshr i64 %8, 63
+  %10 = or i64 %5, %9
+  %11 = load %struct.div_state_** %1, align 4
+  %12 = getelementptr inbounds %struct.div_state_* %11, i32 0, i32 0
+  store i64 %10, i64* %12, align 4
+  %13 = load %struct.div_state_** %1, align 4
+  %14 = getelementptr inbounds %struct.div_state_* %13, i32 0, i32 1
+  %15 = load i64* %14, align 4
+  %16 = shl i64 %15, 1
+  %17 = load %struct.div_state_** %1, align 4
+  %18 = getelementptr inbounds %struct.div_state_* %17, i32 0, i32 1
+  store i64 %16, i64* %18, align 4
+  ret void
+}
+
+define i64 @sdivmod64(i64 %dividend, i64 %divisor, i64* %remainder) nounwind ssp {
+  %1 = alloca i64, align 8
+  %2 = alloca i64, align 8
+  %3 = alloca i64*, align 4
+  %signbitidx = alloca i32, align 4
+  %signed_dividend = alloca i32, align 4
+  %signed_divisor = alloca i32, align 4
+  %signed_result = alloca i32, align 4
+  %quotient = alloca i64, align 8
+  %udvd = alloca i64, align 8
+  %udvr = alloca i64, align 8
+  %uquotient = alloca i64, align 8
+  %uremainder = alloca i64, align 8
+  store i64 %dividend, i64* %1, align 8
+  store i64 %divisor, i64* %2, align 8
+  store i64* %remainder, i64** %3, align 4
+  store i32 63, i32* %signbitidx, align 4
+  %4 = load i64* %1, align 8
+  %5 = icmp slt i64 %4, 0
+  %6 = zext i1 %5 to i32
+  store i32 %6, i32* %signed_dividend, align 4
+  %7 = load i64* %2, align 8
+  %8 = icmp slt i64 %7, 0
+  %9 = zext i1 %8 to i32
+  store i32 %9, i32* %signed_divisor, align 4
+  %10 = load i32* %signed_divisor, align 4
+  %11 = load i32* %signed_dividend, align 4
+  %12 = xor i32 %10, %11
+  store i32 %12, i32* %signed_result, align 4
+  %13 = load i32* %signed_dividend, align 4
+  %14 = icmp ne i32 %13, 0
+  br i1 %14, label %15, label %18
+
+; <label>:15                                      ; preds = %0
+  %16 = load i64* %1, align 8
+  %17 = sub nsw i64 0, %16
+  br label %20
+
+; <label>:18                                      ; preds = %0
+  %19 = load i64* %1, align 8
+  br label %20
+
+; <label>:20                                      ; preds = %18, %15
+  %21 = phi i64 [ %17, %15 ], [ %19, %18 ]
+  store i64 %21, i64* %udvd, align 8
+  %22 = load i32* %signed_divisor, align 4
+  %23 = icmp ne i32 %22, 0
+  br i1 %23, label %24, label %27
+
+; <label>:24                                      ; preds = %20
+  %25 = load i64* %2, align 8
+  %26 = sub nsw i64 0, %25
+  br label %29
+
+; <label>:27                                      ; preds = %20
+  %28 = load i64* %2, align 8
+  br label %29
+
+; <label>:29                                      ; preds = %27, %24
+  %30 = phi i64 [ %26, %24 ], [ %28, %27 ]
+  store i64 %30, i64* %udvr, align 8
+  %31 = load i64* %udvd, align 8
+  %32 = load i64* %udvr, align 8
+  %33 = call i64 @udivmod64(i64 %31, i64 %32, i64* %uremainder)
+  store i64 %33, i64* %uquotient, align 8
+  %34 = load i32* %signed_result, align 4
+  %35 = icmp ne i32 %34, 0
+  br i1 %35, label %36, label %57
+
+; <label>:36                                      ; preds = %29
+  %37 = load i64* %uremainder, align 8
+  %38 = icmp ne i64 %37, 0
+  br i1 %38, label %39, label %43
+
+; <label>:39                                      ; preds = %36
+  %40 = load i64* %uquotient, align 8
+  %41 = sub nsw i64 0, %40
+  %42 = sub nsw i64 %41, 1
+  store i64 %42, i64* %quotient, align 8
+  br label %46
+
+; <label>:43                                      ; preds = %36
+  %44 = load i64* %uquotient, align 8
+  %45 = sub nsw i64 0, %44
+  store i64 %45, i64* %quotient, align 8
+  br label %46
+
+; <label>:46                                      ; preds = %43, %39
+  %47 = load i64** %3, align 4
+  %48 = icmp ne i64* %47, null
+  br i1 %48, label %49, label %56
+
+; <label>:49                                      ; preds = %46
+  %50 = load i64* %1, align 8
+  %51 = load i64* %quotient, align 8
+  %52 = load i64* %2, align 8
+  %53 = mul i64 %51, %52
+  %54 = sub i64 %50, %53
+  %55 = load i64** %3, align 4
+  store i64 %54, i64* %55, align 4
+  br label %56
+
+; <label>:56                                      ; preds = %49, %46
+  br label %73
+
+; <label>:57                                      ; preds = %29
+  %58 = load i64* %uquotient, align 8
+  store i64 %58, i64* %quotient, align 8
+  %59 = load i64** %3, align 4
+  %60 = icmp ne i64* %59, null
+  br i1 %60, label %61, label %72
+
+; <label>:61                                      ; preds = %57
+  %62 = load i32* %signed_divisor, align 4
+  %63 = icmp ne i32 %62, 0
+  br i1 %63, label %64, label %67
+
+; <label>:64                                      ; preds = %61
+  %65 = load i64* %uremainder, align 8
+  %66 = sub i64 0, %65
+  br label %69
+
+; <label>:67                                      ; preds = %61
+  %68 = load i64* %uremainder, align 8
+  br label %69
+
+; <label>:69                                      ; preds = %67, %64
+  %70 = phi i64 [ %66, %64 ], [ %68, %67 ]
+  %71 = load i64** %3, align 4
+  store i64 %70, i64* %71, align 4
+  br label %72
+
+; <label>:72                                      ; preds = %69, %57
+  br label %73
+
+; <label>:73                                      ; preds = %72, %56
+  %74 = load i64* %quotient, align 8
+  ret i64 %74
+}
+
+define i64 @udiv64(i64 %dividend, i64 %divisor) nounwind ssp {
+  %1 = alloca i64, align 8
+  %2 = alloca i64, align 8
+  store i64 %dividend, i64* %1, align 8
+  store i64 %divisor, i64* %2, align 8
+  %3 = load i64* %1, align 8
+  %4 = load i64* %2, align 8
+  %5 = call i64 @udivmod64(i64 %3, i64 %4, i64* null)
+  ret i64 %5
+}
+
+define i64 @sdiv64(i64 %dividend, i64 %divisor) nounwind ssp {
+  %1 = alloca i64, align 8
+  %2 = alloca i64, align 8
+  store i64 %dividend, i64* %1, align 8
+  store i64 %divisor, i64* %2, align 8
+  %3 = load i64* %1, align 8
+  %4 = load i64* %2, align 8
+  %5 = call i64 @sdivmod64(i64 %3, i64 %4, i64* null)
+  ret i64 %5
+}
+
+define i64 @umod64(i64 %dividend, i64 %divisor) nounwind ssp {
+  %1 = alloca i64, align 8
+  %2 = alloca i64, align 8
+  %rem = alloca i64, align 8
+  store i64 %dividend, i64* %1, align 8
+  store i64 %divisor, i64* %2, align 8
+  %3 = load i64* %1, align 8
+  %4 = load i64* %2, align 8
+  %5 = call i64 @udivmod64(i64 %3, i64 %4, i64* %rem)
+  %6 = load i64* %rem, align 8
+  ret i64 %6
+}
+
+define i64 @smod64(i64 %dividend, i64 %divisor) nounwind ssp {
+  %1 = alloca i64, align 8
+  %2 = alloca i64, align 8
+  %rem = alloca i64, align 8
+  store i64 %dividend, i64* %1, align 8
+  store i64 %divisor, i64* %2, align 8
+  %3 = load i64* %1, align 8
+  %4 = load i64* %2, align 8
+  %5 = call i64 @sdivmod64(i64 %3, i64 %4, i64* %rem)
+  %6 = load i64* %rem, align 8
+  ret i64 %6
+}

llvm/llrt/llrt_x86_64.ll

@@ -0,0 +1,371 @@
+; ModuleID = 'udivmod64_x86_64.bc'
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+
+%struct.div_state_ = type { i64, i64 }
+
+define i64 @udivmod64(i64 %dividend, i64 %divisor, i64* %remainder) nounwind uwtable ssp {
+  %1 = alloca i64, align 8
+  %2 = alloca i64, align 8
+  %3 = alloca i64, align 8
+  %4 = alloca i64*, align 8
+  %state = alloca %struct.div_state_, align 8
+  %quotient = alloca i64, align 8
+  %i = alloca i32, align 4
+  %skipahead = alloca i32, align 4
+  store i64 %dividend, i64* %2, align 8
+  store i64 %divisor, i64* %3, align 8
+  store i64* %remainder, i64** %4, align 8
+  %5 = getelementptr inbounds %struct.div_state_* %state, i32 0, i32 0
+  store i64 0, i64* %5, align 8
+  %6 = getelementptr inbounds %struct.div_state_* %state, i32 0, i32 1
+  %7 = load i64* %2, align 8
+  store i64 %7, i64* %6, align 8
+  store i64 0, i64* %quotient, align 8
+  %8 = load i64* %3, align 8
+  %9 = icmp eq i64 %8, 0
+  br i1 %9, label %10, label %11
+
+; <label>:10                                      ; preds = %0
+  store i64 0, i64* %1
+  br label %57
+
+; <label>:11                                      ; preds = %0
+  %12 = load i64* %2, align 8
+  %13 = call i32 @clz64(i64 %12)
+  store i32 %13, i32* %skipahead, align 4
+  store i32 0, i32* %i, align 4
+  br label %14
+
+; <label>:14                                      ; preds = %19, %11
+  %15 = load i32* %i, align 4
+  %16 = load i32* %skipahead, align 4
+  %17 = icmp slt i32 %15, %16
+  br i1 %17, label %18, label %22
+
+; <label>:18                                      ; preds = %14
+  call void @div_state_lshift(%struct.div_state_* %state)
+  br label %19
+
+; <label>:19                                      ; preds = %18
+  %20 = load i32* %i, align 4
+  %21 = add nsw i32 %20, 1
+  store i32 %21, i32* %i, align 4
+  br label %14
+
+; <label>:22                                      ; preds = %14
+  %23 = load i32* %skipahead, align 4
+  store i32 %23, i32* %i, align 4
+  br label %24
+
+; <label>:24                                      ; preds = %45, %22
+  %25 = load i32* %i, align 4
+  %26 = icmp slt i32 %25, 64
+  br i1 %26, label %27, label %48
+
+; <label>:27                                      ; preds = %24
+  call void @div_state_lshift(%struct.div_state_* %state)
+  %28 = getelementptr inbounds %struct.div_state_* %state, i32 0, i32 0
+  %29 = load i64* %28, align 8
+  %30 = load i64* %3, align 8
+  %31 = icmp uge i64 %29, %30
+  br i1 %31, label %32, label %44
+
+; <label>:32                                      ; preds = %27
+  %33 = getelementptr inbounds %struct.div_state_* %state, i32 0, i32 0
+  %34 = load i64* %33, align 8
+  %35 = load i64* %3, align 8
+  %36 = sub i64 %34, %35
+  %37 = getelementptr inbounds %struct.div_state_* %state, i32 0, i32 0
+  store i64 %36, i64* %37, align 8
+  %38 = load i32* %i, align 4
+  %39 = sub nsw i32 63, %38
+  %40 = zext i32 %39 to i64
+  %41 = shl i64 1, %40
+  %42 = load i64* %quotient, align 8
+  %43 = or i64 %42, %41
+  store i64 %43, i64* %quotient, align 8
+  br label %44
+
+; <label>:44                                      ; preds = %32, %27
+  br label %45
+
+; <label>:45                                      ; preds = %44
+  %46 = load i32* %i, align 4
+  %47 = add nsw i32 %46, 1
+  store i32 %47, i32* %i, align 4
+  br label %24
+
+; <label>:48                                      ; preds = %24
+  %49 = load i64** %4, align 8
+  %50 = icmp ne i64* %49, null
+  br i1 %50, label %51, label %55
+
+; <label>:51                                      ; preds = %48
+  %52 = getelementptr inbounds %struct.div_state_* %state, i32 0, i32 0
+  %53 = load i64* %52, align 8
+  %54 = load i64** %4, align 8
+  store i64 %53, i64* %54, align 8
+  br label %55
+
+; <label>:55                                      ; preds = %51, %48
+  %56 = load i64* %quotient, align 8
+  store i64 %56, i64* %1
+  br label %57
+
+; <label>:57                                      ; preds = %55, %10
+  %58 = load i64* %1
+  ret i64 %58
+}
+
+define internal i32 @clz64(i64 %x) nounwind uwtable ssp {
+  %1 = alloca i64, align 8
+  %total_bits = alloca i32, align 4
+  %zc = alloca i32, align 4
+  store i64 %x, i64* %1, align 8
+  store i32 64, i32* %total_bits, align 4
+  store i32 0, i32* %zc, align 4
+  br label %2
+
+; <label>:2                                       ; preds = %16, %0
+  %3 = load i32* %zc, align 4
+  %4 = icmp slt i32 %3, 64
+  br i1 %4, label %5, label %14
+
+; <label>:5                                       ; preds = %2
+  %6 = load i64* %1, align 8
+  %7 = load i32* %zc, align 4
+  %8 = sub nsw i32 64, %7
+  %9 = sub nsw i32 %8, 1
+  %10 = zext i32 %9 to i64
+  %11 = lshr i64 %6, %10
+  %12 = and i64 %11, 1
+  %13 = icmp eq i64 %12, 0
+  br label %14
+
+; <label>:14                                      ; preds = %5, %2
+  %15 = phi i1 [ false, %2 ], [ %13, %5 ]
+  br i1 %15, label %16, label %19
+
+; <label>:16                                      ; preds = %14
+  %17 = load i32* %zc, align 4
+  %18 = add nsw i32 %17, 1
+  store i32 %18, i32* %zc, align 4
+  br label %2
+
+; <label>:19                                      ; preds = %14
+  %20 = load i32* %zc, align 4
+  ret i32 %20
+}
+
+define internal void @div_state_lshift(%struct.div_state_* %state) nounwind uwtable ssp {
+  %1 = alloca %struct.div_state_*, align 8
+  store %struct.div_state_* %state, %struct.div_state_** %1, align 8
+  %2 = load %struct.div_state_** %1, align 8
+  %3 = getelementptr inbounds %struct.div_state_* %2, i32 0, i32 0
+  %4 = load i64* %3, align 8
+  %5 = shl i64 %4, 1
+  %6 = load %struct.div_state_** %1, align 8
+  %7 = getelementptr inbounds %struct.div_state_* %6, i32 0, i32 1
+  %8 = load i64* %7, align 8
+  %9 = lshr i64 %8, 63
+  %10 = or i64 %5, %9
+  %11 = load %struct.div_state_** %1, align 8
+  %12 = getelementptr inbounds %struct.div_state_* %11, i32 0, i32 0
+  store i64 %10, i64* %12, align 8
+  %13 = load %struct.div_state_** %1, align 8
+  %14 = getelementptr inbounds %struct.div_state_* %13, i32 0, i32 1
+  %15 = load i64* %14, align 8
+  %16 = shl i64 %15, 1
+  %17 = load %struct.div_state_** %1, align 8
+  %18 = getelementptr inbounds %struct.div_state_* %17, i32 0, i32 1
+  store i64 %16, i64* %18, align 8
+  ret void
+}
+
+define i64 @sdivmod64(i64 %dividend, i64 %divisor, i64* %remainder) nounwind uwtable ssp {
+  %1 = alloca i64, align 8
+  %2 = alloca i64, align 8
+  %3 = alloca i64*, align 8
+  %signbitidx = alloca i32, align 4
+  %signed_dividend = alloca i32, align 4
+  %signed_divisor = alloca i32, align 4
+  %signed_result = alloca i32, align 4
+  %quotient = alloca i64, align 8
+  %udvd = alloca i64, align 8
+  %udvr = alloca i64, align 8
+  %uquotient = alloca i64, align 8
+  %uremainder = alloca i64, align 8
+  store i64 %dividend, i64* %1, align 8
+  store i64 %divisor, i64* %2, align 8
+  store i64* %remainder, i64** %3, align 8
+  store i32 63, i32* %signbitidx, align 4
+  %4 = load i64* %1, align 8
+  %5 = icmp slt i64 %4, 0
+  %6 = zext i1 %5 to i32
+  store i32 %6, i32* %signed_dividend, align 4
+  %7 = load i64* %2, align 8
+  %8 = icmp slt i64 %7, 0
+  %9 = zext i1 %8 to i32
+  store i32 %9, i32* %signed_divisor, align 4
+  %10 = load i32* %signed_divisor, align 4
+  %11 = load i32* %signed_dividend, align 4
+  %12 = xor i32 %10, %11
+  store i32 %12, i32* %signed_result, align 4
+  %13 = load i32* %signed_dividend, align 4
+  %14 = icmp ne i32 %13, 0
+  br i1 %14, label %15, label %18
+
+; <label>:15                                      ; preds = %0
+  %16 = load i64* %1, align 8
+  %17 = sub nsw i64 0, %16
+  br label %20
+
+; <label>:18                                      ; preds = %0
+  %19 = load i64* %1, align 8
+  br label %20
+
+; <label>:20                                      ; preds = %18, %15
+  %21 = phi i64 [ %17, %15 ], [ %19, %18 ]
+  store i64 %21, i64* %udvd, align 8
+  %22 = load i32* %signed_divisor, align 4
+  %23 = icmp ne i32 %22, 0
+  br i1 %23, label %24, label %27
+
+; <label>:24                                      ; preds = %20
+  %25 = load i64* %2, align 8
+  %26 = sub nsw i64 0, %25
+  br label %29
+
+; <label>:27                                      ; preds = %20
+  %28 = load i64* %2, align 8
+  br label %29
+
+; <label>:29                                      ; preds = %27, %24
+  %30 = phi i64 [ %26, %24 ], [ %28, %27 ]
+  store i64 %30, i64* %udvr, align 8
+  %31 = load i64* %udvd, align 8
+  %32 = load i64* %udvr, align 8
+  %33 = call i64 @udivmod64(i64 %31, i64 %32, i64* %uremainder)
+  store i64 %33, i64* %uquotient, align 8
+  %34 = load i32* %signed_result, align 4
+  %35 = icmp ne i32 %34, 0
+  br i1 %35, label %36, label %57
+
+; <label>:36                                      ; preds = %29
+  %37 = load i64* %uremainder, align 8
+  %38 = icmp ne i64 %37, 0
+  br i1 %38, label %39, label %43
+
+; <label>:39                                      ; preds = %36
+  %40 = load i64* %uquotient, align 8
+  %41 = sub nsw i64 0, %40
+  %42 = sub nsw i64 %41, 1
+  store i64 %42, i64* %quotient, align 8
+  br label %46
+
+; <label>:43                                      ; preds = %36
+  %44 = load i64* %uquotient, align 8
+  %45 = sub nsw i64 0, %44
+  store i64 %45, i64* %quotient, align 8
+  br label %46
+
+; <label>:46                                      ; preds = %43, %39
+  %47 = load i64** %3, align 8
+  %48 = icmp ne i64* %47, null
+  br i1 %48, label %49, label %56
+
+; <label>:49                                      ; preds = %46
+  %50 = load i64* %1, align 8
+  %51 = load i64* %quotient, align 8
+  %52 = load i64* %2, align 8
+  %53 = mul i64 %51, %52
+  %54 = sub i64 %50, %53
+  %55 = load i64** %3, align 8
+  store i64 %54, i64* %55, align 8
+  br label %56
+
+; <label>:56                                      ; preds = %49, %46
+  br label %73
+
+; <label>:57                                      ; preds = %29
+  %58 = load i64* %uquotient, align 8
+  store i64 %58, i64* %quotient, align 8
+  %59 = load i64** %3, align 8
+  %60 = icmp ne i64* %59, null
+  br i1 %60, label %61, label %72
+
+; <label>:61                                      ; preds = %57
+  %62 = load i32* %signed_divisor, align 4
+  %63 = icmp ne i32 %62, 0
+  br i1 %63, label %64, label %67
+
+; <label>:64                                      ; preds = %61
+  %65 = load i64* %uremainder, align 8
+  %66 = sub i64 0, %65
+  br label %69
+
+; <label>:67                                      ; preds = %61
+  %68 = load i64* %uremainder, align 8
+  br label %69
+
+; <label>:69                                      ; preds = %67, %64
+  %70 = phi i64 [ %66, %64 ], [ %68, %67 ]
+  %71 = load i64** %3, align 8
+  store i64 %70, i64* %71, align 8
+  br label %72
+
+; <label>:72                                      ; preds = %69, %57
+  br label %73
+
+; <label>:73                                      ; preds = %72, %56
+  %74 = load i64* %quotient, align 8
+  ret i64 %74
+}
+
+define i64 @udiv64(i64 %dividend, i64 %divisor) nounwind uwtable ssp {
+  %1 = alloca i64, align 8
+  %2 = alloca i64, align 8
+  store i64 %dividend, i64* %1, align 8
+  store i64 %divisor, i64* %2, align 8
+  %3 = load i64* %1, align 8
+  %4 = load i64* %2, align 8
+  %5 = call i64 @udivmod64(i64 %3, i64 %4, i64* null)
+  ret i64 %5
+}
+
+define i64 @sdiv64(i64 %dividend, i64 %divisor) nounwind uwtable ssp {
+  %1 = alloca i64, align 8
+  %2 = alloca i64, align 8
+  store i64 %dividend, i64* %1, align 8
+  store i64 %divisor, i64* %2, align 8
+  %3 = load i64* %1, align 8
+  %4 = load i64* %2, align 8
+  %5 = call i64 @sdivmod64(i64 %3, i64 %4, i64* null)
+  ret i64 %5
+}
+
+define i64 @umod64(i64 %dividend, i64 %divisor) nounwind uwtable ssp {
+  %1 = alloca i64, align 8
+  %2 = alloca i64, align 8
+  %rem = alloca i64, align 8
+  store i64 %dividend, i64* %1, align 8
+  store i64 %divisor, i64* %2, align 8
+  %3 = load i64* %1, align 8
+  %4 = load i64* %2, align 8
+  %5 = call i64 @udivmod64(i64 %3, i64 %4, i64* %rem)
+  %6 = load i64* %rem, align 8
+  ret i64 %6
+}
+
+define i64 @smod64(i64 %dividend, i64 %divisor) nounwind uwtable ssp {
+  %1 = alloca i64, align 8
+  %2 = alloca i64, align 8
+  %rem = alloca i64, align 8
+  store i64 %dividend, i64* %1, align 8
+  store i64 %divisor, i64* %2, align 8
+  %3 = load i64* %1, align 8
+  %4 = load i64* %2, align 8
+  %5 = call i64 @sdivmod64(i64 %3, i64 %4, i64* %rem)
+  %6 = load i64* %rem, align 8
+  ret i64 %6
+}

llvm/mc/__init__.py

@@ -0,0 +1,242 @@
+import sys
+
+import llvm
+if llvm.version < (3, 4):
+    raise Exception("mc is not supported for llvm version less than 3.4")
+
+from io import BytesIO
+import contextlib
+
+from llvmpy import api, extra
+from llvmpy.api.llvm import MCDisassembler
+
+class Operand(object):
+
+    def __init__(self, mcoperand, target_machine):
+        '''
+        @mcoperand: an MCOperand object
+        @target_machine: an llvm.target.TargetMachine object
+        '''
+
+        self.op = mcoperand
+        if not self.op:
+            raise llvm.LLVMException("null MCOperand argument")
+
+        self.tm = target_machine
+
+    def __str__(self):
+        s = "invalid"
+        if self.is_reg():
+            s = "reg(%s)" % (self.reg_name())
+        elif self.is_imm():
+            s = "imm(0x%02x)" % (self.op.getImm())
+        elif self.is_fp_imm():
+            s = "imm(%r)" % (self.op.getFPImm())
+        elif self.is_expr():
+            s = "expr(%r)" % (self.op.getExpr().getKind())
+        elif self.is_inst():
+            s = repr(Instr(self.op.getInst()))
+
+        return s
+
+    def __repr__(self):
+        return str(self)
+
+    def reg_name(self):
+        if self.is_reg():
+            s = self.tm.reg_info.getName(self.op.getReg())
+            if s.strip() == "":
+                return "?"
+            else:
+                return s
+        else:
+            return ""
+
+    def is_reg(self):
+        return self.op.isReg()
+
+    def is_imm(self):
+        return self.op.isImm()
+
+    def is_fp_imm(self):
+        return self.op.isFPImm()
+
+    def is_expr(self):
+        return self.op.isExpr()
+
+    def is_inst(self):
+        return self.op.isInst()
+
+    def get_imm(self):
+        if self.is_imm():
+            return self.op.getImm()
+        else:
+            return None
+
+    def get_fp_imm(self):
+        if self.is_fp_imm():
+            return self.op.getFPImm()
+        else:
+            return None
+
+    def get_inst(self):
+        if self.is_inst():
+            return Instr(self.op.getInst())
+        else:
+            return None
+
+class Instr(object):
+
+    def __init__(self, mcinst, target_machine):
+        '''
+        @mcinst: an MCInst object
+        @target_machine: an llvm.target.TargetMachine object
+        '''
+
+        self.mcinst = mcinst
+        if not self.mcinst:
+            raise llvm.LLVMException("null MCInst argument")
+
+        self.tm = target_machine
+
+    def __str__(self):
+        os = extra.make_raw_ostream_for_printing()
+        self.tm.inst_printer.printInst(self.mcinst, os, "")
+        return str(os.str())
+
+    def __repr__(self):
+        return str(self)
+
+    def __len__(self):
+        ''' the number of operands '''
+        return int(self.mcinst.size())
+
+    def operands(self):
+        amt = self.mcinst.getNumOperands()
+        if amt < 1:
+            return []
+
+        l = []
+        for i in range(0, amt):
+            l.append(Operand(self.mcinst.getOperand(i), self.tm))
+
+        return l
+
+    @property
+    def instr_desc(self):
+        return self.tm.instr_info.get(self.opcode)
+
+    @property
+    def flags(self):
+        return self.instr_desc.getFlags()
+
+    @property
+    def ts_flags(self):
+        return self.instr_desc.TSFlags
+
+    @property
+    def opcode(self):
+        return self.mcinst.getOpcode()
+
+    def is_branch(self):
+        return self.instr_desc.isBranch()
+
+    def is_cond_branch(self):
+        return self.instr_desc.isConditionalBranch()
+
+    def is_uncond_branch(self):
+        return self.instr_desc.isUnconditionalBranch()
+
+    def is_indirect_branch(self):
+        return self.instr_desc.isIndirectBranch()
+
+    def is_call(self):
+        return self.instr_desc.isCall()
+
+    def is_return(self):
+        return self.instr_desc.isReturn()
+
+    def is_terminator(self):
+        return self.instr_desc.isTerminator()
+
+    def is_barrier(self):
+        return self.instr_desc.isBarrier()
+
+class BadInstr(Instr):
+    pass
+
+class Disassembler(object):
+
+    def __init__(self, target_machine):
+        self.tm = target_machine
+
+    @property
+    def mdasm(self):
+        return self.tm.disassembler
+
+    @property
+    def mai(self):
+        return self.tm.asm_info
+
+    def instr(self, mcinst):
+        return Instr(mcinst, self.tm)
+
+    def bad_instr(self, mcinst):
+        return BadInstr(mcinst, self.tm)
+
+    def decode(self, bs, base_addr, align=None):
+        '''
+        decodes the bytes in @bs into instructions and yields
+        each instruction as it is decoded. @base_addr is the base address
+        where the instruction bytes are from (not an offset into
+        @bs). yields instructions in the form of (addr, data, inst) where
+        addr is an integer, data is a tuple of integers and inst is an instance of
+        llvm.mc.Instr. @align specifies the byte alignment of instructions and
+        is only used if an un-decodable instruction is encountered, in which
+        case the disassembler will skip the following bytes until the next
+        aligned address. if @align is unspecified, the default alignment
+        for the architecture will be used, however this may not be ideal
+        for disassembly. for example, the default alignment for ARM is 1, but you
+        probably want it to be 4 for the purposes of disassembling ARM
+        instructions.
+        '''
+
+        if isinstance(bs, str) and sys.version_info.major >= 3:
+            bs = bytes(map(lambda c: ord(c), bs))
+        elif not isinstance(bs, bytes):
+            raise TypeError("expected bs to be either 'str' or 'bytes' but got %s" % type(bs))
+
+        code = api.llvm.StringRefMemoryObject.new(bs, base_addr)
+        idx = 0
+        if not isinstance(align, int) or align < 1:
+            align = self.mai.getMinInstAlignment()
+
+        while(idx < code.getExtent()):
+            inst = api.llvm.MCInst.new()
+            addr = code.getBase() + idx
+            status, size = self.mdasm.getInstruction(inst, code, addr)
+
+            if size < 1:
+                size = (align - (idx % align))
+
+            amt_left = code.getExtent() - idx
+            if amt_left >= size:
+                data = code.readBytes(addr, size)
+            elif amt_left < 1:
+                break
+            else:
+                data = code.readBytes(addr, amt_left)
+
+            if sys.version_info.major < 3:
+                data = tuple(map(lambda b: ord(b), data))
+            else:
+                data = tuple(data)
+
+            if status == MCDisassembler.DecodeStatus.Fail:
+                yield (addr, data, None)
+            elif status == MCDisassembler.DecodeStatus.SoftFail:
+                yield (addr, data, self.bad_instr(inst))
+            else:
+                yield (addr, data, self.instr(inst))
+
+            idx += size

llvm/passes.py

@@ -70,13 +70,33 @@ class PassManagerBuilder(llvm.Wrapper):
     def size_level(self, sizelevel):
         self._ptr.SizeLevel = sizelevel
 
-    @property
+    if llvm.version >= (3, 3):
-    def vectorize(self):
+        @property
-        return self._ptr.Vectorize
+        def bbvectorize(self):
+            return self._ptr.BBVectorize
 
-    @vectorize.setter
+        @bbvectorize.setter
-    def vectorize(self, enable):
+        def bbvectorize(self, enable):
-        self._ptr.Vectorize = enable
+            self._ptr.BBVectorize = enable
+
+        vectorize = bbvectorize
+
+        @property
+        def slpvectorize(self):
+            return self._ptr.SLPVectorize
+
+        @slpvectorize.setter
+        def slpvectorize(self, enable):
+            self._ptr.SLPVectorize = enable
+
+    else:
+        @property
+        def vectorize(self):
+            return self._ptr.Vectorize
+
+        @vectorize.setter
+        def vectorize(self, enable):
+            self._ptr.Vectorize = enable
 
 
     @property
@@ -109,13 +129,14 @@ class PassManagerBuilder(llvm.Wrapper):
     def disable_unroll_loops(self, disable):
         self._ptr.DisableUnrollLoops = disable
 
-    @property
+    if llvm.version <= (3, 3):
-    def disable_simplify_lib_calls(self):
+        @property
-        return self._ptr.DisableSimplifyLibCalls
+        def disable_simplify_lib_calls(self):
+            return self._ptr.DisableSimplifyLibCalls
 
-    @disable_simplify_lib_calls.setter
+        @disable_simplify_lib_calls.setter
-    def disable_simplify_lib_calls(self, disable):
+        def disable_simplify_lib_calls(self, disable):
-        self._ptr.DisableSimplifyLibCalls = disable
+            self._ptr.DisableSimplifyLibCalls = disable
 
     def use_inliner_with_threshold(self, threshold):
         self._ptr.Inliner = api.llvm.createFunctionInliningPass(threshold)
@@ -299,13 +320,15 @@ class TargetTransformInfo(Pass):
 # Helpers
 #===----------------------------------------------------------------------===
 
-def build_pass_managers(tm, opt=2, loop_vectorize=False, vectorize=False,
+def build_pass_managers(tm, opt=2, size=0, loop_vectorize=False,
-                        inline_threshold=2000, pm=True, fpm=True, mod=None):
+                        slp_vectorize=False, vectorize=False,
+                        inline_threshold=None, pm=True, fpm=True, mod=None):
     '''
         tm --- The TargetMachine for which the passes are optimizing for.
         The TargetMachine must stay alive until the pass managers
         are removed.
         opt --- [0-3] Optimization level. Default to 2.
+        size --- [0-2] Optimize for size. Default to 0.
         loop_vectorize --- [boolean] Whether to use loop-vectorizer.
         vectorize --- [boolean] Whether to use basic-block vectorizer.
         inline_threshold --- [int] Threshold for the inliner.
@@ -314,6 +337,18 @@ def build_pass_managers(tm, opt=2, loop_vectorize=False, vectorize=False,
         fpm --- [boolean] Whether to build a function-level pass-manager.
         mod --- [Module] The module object for the FunctionPassManager.
         '''
+    if inline_threshold is None:
+        if 0 < opt < 3:
+            inline_threshold = 225
+
+        if size == 1:
+            inline_threshold = 75
+        elif size == 2:
+            inline_threshold = 25
+
+        if opt >= 3:
+            inline_threshold = 275
+
     if pm:
         pm = PassManager.new()
     if fpm:
@@ -326,22 +361,27 @@ def build_pass_managers(tm, opt=2, loop_vectorize=False, vectorize=False,
     pmb.opt_level = opt
     pmb.vectorize = vectorize
     pmb.loop_vectorize = loop_vectorize
+    if llvm.version >= (3, 3):
+        pmb.slp_vectorize = slp_vectorize
     if inline_threshold:
         pmb.use_inliner_with_threshold(inline_threshold)
     if pm:
         pm.add(tm.target_data.clone())
         pm.add(TargetLibraryInfo.new(tm.triple))
-        if llvm.version >= (3, 2):
+        if llvm.version <= (3, 2):
             pm.add(TargetTransformInfo.new(tm))
+        else:
+            tm.add_analysis_passes(pm)
         pmb.populate(pm)
 
     if fpm:
         fpm.add(tm.target_data.clone())
         fpm.add(TargetLibraryInfo.new(tm.triple))
-        if llvm.version >= (3, 2):
+        if llvm.version <= (3, 2):
             fpm.add(TargetTransformInfo.new(tm))
+        else:
+            tm.add_analysis_passes(fpm)
         pmb.populate(fpm)
-        fpm.initialize()
 
     from collections import namedtuple
     return namedtuple('passmanagers', ['pm', 'fpm'])(pm=pm, fpm=fpm)

llvm/target.py

@@ -0,0 +1,266 @@
+import llvm
+from llvmpy import api, extra
+from io import BytesIO
+import contextlib
+from llvm.passes import TargetData
+
+#===----------------------------------------------------------------------===
+# Enumerations
+#===----------------------------------------------------------------------===
+
+BO_BIG_ENDIAN       = 0
+BO_LITTLE_ENDIAN    = 1
+
+# CodeModel
+CM_DEFAULT      = api.llvm.CodeModel.Model.Default
+CM_JITDEFAULT   = api.llvm.CodeModel.Model.JITDefault
+CM_SMALL        = api.llvm.CodeModel.Model.Small
+CM_KERNEL       = api.llvm.CodeModel.Model.Kernel
+CM_MEDIUM       = api.llvm.CodeModel.Model.Medium
+CM_LARGE        = api.llvm.CodeModel.Model.Large
+
+# Reloc
+RELOC_DEFAULT        = api.llvm.Reloc.Model.Default
+RELOC_STATIC         = api.llvm.Reloc.Model.Static
+RELOC_PIC            = api.llvm.Reloc.Model.PIC_
+RELOC_DYNAMIC_NO_PIC = api.llvm.Reloc.Model.DynamicNoPIC
+
+def initialize_all():
+    api.llvm.InitializeAllTargets()
+    api.llvm.InitializeAllTargetInfos()
+    api.llvm.InitializeAllTargetMCs()
+    api.llvm.InitializeAllAsmPrinters()
+    api.llvm.InitializeAllDisassemblers()
+    api.llvm.InitializeAllAsmParsers()
+
+def initialize_target(target, noraise=False):
+    """Initialize target by name.
+    It is safe to initialize the same target multiple times.
+    """
+    prefix = 'LLVMInitialize'
+    postfixes = ['Target', 'TargetInfo', 'TargetMC', 'AsmPrinter', 'AsmParser']
+    try:
+        for postfix in postfixes:
+            getattr(api, '%s%s%s' % (prefix, target, postfix))()
+    except AttributeError:
+        if noraise:
+            return False
+        else:
+            raise
+    else:
+        return True
+
+
+def print_registered_targets():
+    '''
+    Note: print directly to stdout
+    '''
+    api.llvm.TargetRegistry.printRegisteredTargetsForVersion()
+
+def get_host_cpu_name():
+    '''return the string name of the host CPU
+    '''
+    return api.llvm.sys.getHostCPUName()
+
+def get_default_triple():
+    '''return the target triple of the host in str-rep
+    '''
+    return api.llvm.sys.getDefaultTargetTriple()
+
+class TargetMachine(llvm.Wrapper):
+
+    @staticmethod
+    def new(triple='', cpu='', features='', opt=2, cm=CM_DEFAULT,
+            reloc=RELOC_DEFAULT):
+        if not triple:
+            triple = get_default_triple()
+        if not cpu:
+            cpu = get_host_cpu_name()
+        with contextlib.closing(BytesIO()) as error:
+            target = api.llvm.TargetRegistry.lookupTarget(triple, error)
+            if not target:
+                raise llvm.LLVMException(error.getvalue())
+            if not target.hasTargetMachine():
+                raise llvm.LLVMException(target, "No target machine.")
+            target_options = api.llvm.TargetOptions.new()
+            tm = target.createTargetMachine(triple, cpu, features,
+                                            target_options,
+                                            reloc, cm, opt)
+            if not tm:
+                raise llvm.LLVMException("Cannot create target machine")
+            return TargetMachine(tm)
+
+    @staticmethod
+    def lookup(arch, cpu='', features='', opt=2, cm=CM_DEFAULT,
+               reloc=RELOC_DEFAULT):
+        '''create a targetmachine given an architecture name
+
+            For a list of architectures,
+            use: `llc -help`
+
+            For a list of available CPUs,
+            use: `llvm-as < /dev/null | llc -march=xyz -mcpu=help`
+
+            For a list of available attributes (features),
+            use: `llvm-as < /dev/null | llc -march=xyz -mattr=help`
+            '''
+        triple = api.llvm.Triple.new()
+        with contextlib.closing(BytesIO()) as error:
+            target = api.llvm.TargetRegistry.lookupTarget(arch, triple, error)
+            if not target:
+                raise llvm.LLVMException(error.getvalue())
+            if not target.hasTargetMachine():
+                raise llvm.LLVMException(target, "No target machine.")
+            target_options = api.llvm.TargetOptions.new()
+            tm = target.createTargetMachine(str(triple), cpu, features,
+                                            target_options,
+                                            reloc, cm, opt)
+            if not tm:
+                raise llvm.LLVMException("Cannot create target machine")
+            return TargetMachine(tm)
+
+    @staticmethod
+    def x86():
+        return TargetMachine.lookup('x86')
+
+    @staticmethod
+    def x86_64():
+        return TargetMachine.lookup('x86-64')
+
+    @staticmethod
+    def arm():
+        return TargetMachine.lookup('arm')
+
+    @staticmethod
+    def thumb():
+        return TargetMachine.lookup('thumb')
+
+    def _emit_file(self, module, cgft):
+        pm = api.llvm.PassManager.new()
+        os = extra.make_raw_ostream_for_printing()
+        pm.add(api.llvm.DataLayout.new(str(self.target_data)))
+        failed = self._ptr.addPassesToEmitFile(pm, os, cgft)
+        pm.run(module)
+
+
+        CGFT = api.llvm.TargetMachine.CodeGenFileType
+        if cgft == CGFT.CGFT_ObjectFile:
+            return os.bytes()
+        else:
+            return os.str()
+
+    def emit_assembly(self, module):
+        '''returns byte string of the module as assembly code of the target machine
+        '''
+        CGFT = api.llvm.TargetMachine.CodeGenFileType
+        return self._emit_file(module._ptr, CGFT.CGFT_AssemblyFile)
+
+    def emit_object(self, module):
+        '''returns byte string of the module as native code of the target machine
+        '''
+        CGFT = api.llvm.TargetMachine.CodeGenFileType
+        return self._emit_file(module._ptr, CGFT.CGFT_ObjectFile)
+
+    @property
+    def target_data(self):
+        '''get target data of this machine
+        '''
+        return TargetData(self._ptr.getDataLayout())
+
+    @property
+    def target_name(self):
+        return self._ptr.getTarget().getName()
+
+    @property
+    def target_short_description(self):
+        return self._ptr.getTarget().getShortDescription()
+
+    @property
+    def triple(self):
+        return self._ptr.getTargetTriple()
+
+    @property
+    def cpu(self):
+        return self._ptr.getTargetCPU()
+
+    @property
+    def feature_string(self):
+        return self._ptr.getTargetFeatureString()
+
+    @property
+    def target(self):
+        return self._ptr.getTarget()
+
+    if llvm.version >= (3, 3):
+        def add_analysis_passes(self, pm):
+            self._ptr.addAnalysisPasses(pm._ptr)
+
+    if llvm.version >= (3, 4):
+        @property
+        def reg_info(self):
+            mri = self._ptr.getRegisterInfo()
+            if not mri:
+                raise llvm.LLVMException("no reg info for this machine")
+
+            return mri
+
+        @property
+        def subtarget_info(self):
+            sti = self._ptr.getSubtargetImpl()
+            if not sti:
+                raise llvm.LLVMException("no subtarget info for this machine")
+
+            return sti
+
+        @property
+        def asm_info(self):
+            ai = self._ptr.getMCAsmInfo()
+            if not ai:
+                raise llvm.LLVMException("no asm info for this machine")
+
+            return ai
+
+        @property
+        def instr_info(self):
+            ii = self._ptr.getInstrInfo()
+            if not ii:
+                raise llvm.LLVMException("no instr info for this machine")
+
+            return ii
+
+        @property
+        def instr_analysis(self):
+            if not getattr(self, '_mia', False):
+                self._mia = self.target.createMCInstrAnalysis(self.instr_info)
+            if not self._mia:
+                raise llvm.LLVMException("no instr analysis for this machine")
+
+            return self._mia
+
+        @property
+        def disassembler(self):
+            if not getattr(self, '_dasm', False):
+                self._dasm = self.target.createMCDisassembler(self.subtarget_info)
+            if not self._dasm:
+                raise llvm.LLVMException("no disassembler for this machine")
+
+            return self._dasm
+
+        @property
+        def inst_printer(self):
+            if not getattr(self, '_mip', False):
+                self._mip = self.target.createMCInstPrinter(
+                                     self.asm_info.getAssemblerDialect(),
+                                     self.asm_info,
+                                     self.instr_info,
+                                     self.reg_info,
+                                     self.subtarget_info
+                                     )
+            if not self._mip:
+                raise llvm.LLVMException("no instr printer for this machine")
+
+            return self._mip
+
+        def is_little_endian(self):
+            return self.asm_info.isLittleEndian()
+

llvm/tests/__init__.py

@@ -0,0 +1,72 @@
+from __future__ import print_function
+import sys
+import os
+import unittest
+import subprocess
+import llvm
+
+tests = []  # stores unittest.TestCase objects
+
+# Isolated tests
+# Tests that affect process-wide settings
+isolated_tests = []  # stores modue name
+
+
+def run(verbosity=1, run_isolated=True):
+    print('llvmpy is installed in: ' + os.path.dirname(__file__))
+    print('llvmpy version: ' + llvm.__version__)
+    print(sys.version)
+
+    files = filter(lambda s: s.startswith('test_') and s.endswith('.py'),
+                   os.listdir(os.path.dirname(__file__)))
+
+    for f in files:
+        fname = f.split('.', 1)[0]
+        __import__('.'.join([__name__, fname]))
+
+    suite = unittest.TestSuite()
+    for cls in tests:
+        if cls:
+            suite.addTest(unittest.makeSuite(cls))
+
+    # The default stream fails in IPython qtconsole on Windows,
+    # so just using sys.stdout
+
+    kwargs = dict(verbosity=verbosity, stream=sys.stdout)
+
+    if sys.version_info[:2] > (2, 6):
+        kwargs['buffer'] = True
+    runner = unittest.TextTestRunner(**kwargs)
+
+    try:
+        from guppy import hpy
+    except ImportError:
+        testresult = runner.run(suite)
+    else:
+        hp = hpy()
+        hp.setref()
+        testresult = runner.run(suite)
+        print(hp.heap())
+
+    if testresult and run_isolated:
+        # Run isolated tests
+        print("run isolated tests".center(80, '-'))
+
+        for test in isolated_tests:
+            print(('testing %s' % test).center(80))
+
+            cmd = [sys.executable, '-m', test]
+            p = subprocess.Popen(cmd, stdout=subprocess.PIPE,
+                                 stderr=subprocess.STDOUT)
+            for line in p.stdout:
+                print(line.decode('utf8'), end='')
+            p.wait()
+            if p.returncode:
+                raise Exception("%s returned: %d" % (test, p.returncode))
+
+    return testresult
+
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/support.py

@@ -0,0 +1,78 @@
+from __future__ import print_function, division
+import sys
+import platform
+import unittest
+import contextlib
+import types
+from llvm.tests import tests, isolated_tests # re-expose symbol
+
+IS_PY3K = sys.version_info[0] >= 3
+BITS = tuple.__itemsize__ * 8
+OS = sys.platform
+MACHINE = platform.machine()
+INTEL_CPUS = 'i386', 'x86_64'
+
+if sys.version_info[:2] <= (2, 6):
+    # create custom TestCase
+    class _TestCase(unittest.TestCase):
+        def assertIn(self, item, container):
+            self.assertTrue(item in container)
+
+        def assertNotIn(self, item, container):
+            self.assertFalse(item in container)
+
+        def assertLess(self, a, b):
+            self.assertTrue(a < b)
+
+        def assertIs(self, a, b):
+            self.assertTrue(a is b)
+
+        @contextlib.contextmanager
+        def assertRaises(self, exc):
+            try:
+                yield
+            except exc:
+                pass
+            else:
+                raise self.failureException("Did not raise %s" % exc)
+
+else:
+    _TestCase = unittest.TestCase
+
+class TestCase(_TestCase):
+    def assertClose(self, got, expect):
+        rel = abs(got - expect) / expect
+        self.assertTrue(rel < 1e-6, 'relative error = %f' % rel)
+
+#-------------------------------------------------------------------------------
+# Tests decorators
+
+def _skipped(name, msg):
+    def _test(self):
+        if hasattr(unittest, 'SkipTest'):
+            raise unittest.SkipTest(msg)
+        else:
+            print('skipped %s' % name, msg)
+    return _test
+
+def skip_if(cond, msg=''):
+    def skipper(test):
+        if not isinstance(test, types.FunctionType):
+            repl = None
+        else:
+            repl = _skipped(test, msg)
+        return repl if cond else test
+    return skipper
+
+skip_if_not_64bits = skip_if(BITS != 64, msg='skipped not 64-bit')
+
+skip_if_not_32bits = skip_if(BITS != 32, msg='skipped not 32-bits')
+
+skip_if_win32 = skip_if(OS.startswith('win32'), msg='skipped win32')
+
+skip_if_not_win32 = skip_if(not OS.startswith('win32'),
+                            msg='skipped not win32')
+skip_if_not_intel_cpu = skip_if(MACHINE not in INTEL_CPUS,
+                                msg='skipped not Intel CPU')
+
+

llvm/tests/test_alloca.py

@@ -0,0 +1,24 @@
+import unittest
+from llvm.core import Type, Module, Builder, Constant
+from .support import TestCase, tests
+
+class TestAlloca(TestCase):
+    def test_alloca_alignment(self):
+        m = Module.new('')
+        f = m.add_function(Type.function(Type.void(), []), "foo")
+        b = Builder.new(f.append_basic_block(''))
+        inst = b.alloca(Type.int(32))
+        inst.alignment = 4
+        b.ret_void()
+        m.verify()
+
+        self.assertTrue(inst.is_static)
+        self.assertFalse(inst.is_array)
+        self.assertEqual(inst.alignment, 4)
+        self.assertEqual(str(inst.array_size), 'i32 1')
+
+tests.append(TestAlloca)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_arg_attr.py

@@ -0,0 +1,24 @@
+import unittest
+from llvm.core import Module, Type
+import llvm.core as lc
+from .support import TestCase, tests
+
+class TestArgAttr(TestCase):
+    def test_arg_attr(self):
+        m = Module.new('oifjda')
+        vptr = Type.pointer(Type.float())
+        fnty = Type.function(Type.void(), [vptr] * 5)
+        func = m.add_function(fnty, 'foo')
+        attrs = [lc.ATTR_STRUCT_RET, lc.ATTR_BY_VAL, lc.ATTR_NEST,
+                 lc.ATTR_NO_ALIAS, lc.ATTR_NO_CAPTURE]
+        for i, attr in enumerate(attrs):
+            arg = func.args[i]
+            self.assertEqual(i, arg.arg_no)
+            arg.add_attribute(attr)
+            self.assertTrue(attr in func.args[i])
+
+tests.append(TestArgAttr)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_arith.py

@@ -0,0 +1,73 @@
+import unittest
+import llvm
+from llvm.core import (Module, Type, Builder)
+from llvm.ee import EngineBuilder
+from .support import TestCase, tests, skip_if, skip_if_not_64bits
+
+@skip_if(llvm.version < (3, 3))
+class TestArith(TestCase):
+    '''
+    Test basic arithmetic support with LLVM MCJIT
+    '''
+    def func_template(self, ty, op):
+        m = Module.new('dofjaa')
+        fnty = Type.function(ty, [ty, ty])
+        fn = m.add_function(fnty, 'foo')
+        bldr = Builder.new(fn.append_basic_block(''))
+        bldr.ret(getattr(bldr, op)(*fn.args))
+
+        engine = EngineBuilder.new(m).mcjit(True).create()
+        ptr = engine.get_pointer_to_function(fn)
+
+        from ctypes import c_uint32, c_uint64, c_float, c_double, CFUNCTYPE
+
+        maptypes = {
+            Type.int(32): c_uint32,
+            Type.int(64): c_uint64,
+            Type.float(): c_float,
+            Type.double(): c_double,
+        }
+        cty = maptypes[ty]
+        prototype = CFUNCTYPE(*[cty] * 3)
+        callee = prototype(ptr)
+        callee(12, 23)
+
+    def template(self, iop, fop):
+        inttys = [Type.int(32), Type.int(64)]
+        flttys = [Type.float(), Type.double()]
+
+        if iop:
+            for ty in inttys:
+                self.func_template(ty, iop)
+        if fop:
+            for ty in flttys:
+                self.func_template(ty, fop)
+
+    def test_add(self):
+        self.template('add', 'fadd')
+
+    def test_sub(self):
+        self.template('sub', 'fsub')
+
+    def test_mul(self):
+        self.template('mul', 'fmul')
+
+    @skip_if_not_64bits
+    def test_div(self):
+        '''
+        known failure due to unresolved external symbol __udivdi3
+        '''
+        self.template('udiv', None) # 'fdiv')
+
+    @skip_if_not_64bits
+    def test_rem(self):
+        '''
+        known failure due to unresolved external symbol __umoddi3
+        '''
+        self.template('urem', None) # 'frem')
+
+tests.append(TestArith)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_asm.py

@@ -0,0 +1,58 @@
+import os
+import unittest
+import tempfile
+import shutil
+from llvm.core import Module, Type
+from .support import TestCase, tests
+
+class TestAsm(TestCase):
+    def setUp(self):
+        self.tmpdir = tempfile.mkdtemp()
+
+    def tearDown(self):
+        shutil.rmtree(self.tmpdir)
+
+    def test_asm(self):
+        # create a module
+        m = Module.new('module1')
+        m.add_global_variable(Type.int(), 'i')
+
+        # write it's assembly representation to a file
+        asm = str(m)
+
+        testasm_ll = os.path.join(self.tmpdir, 'testasm.ll')
+        with open(testasm_ll, "w") as fout:
+            fout.write(asm)
+
+        # read it back into a module
+        with open(testasm_ll) as fin:
+            m2 = Module.from_assembly(fin)
+            # The default `m.id` is '<string>'.
+            m2.id = m.id # Copy the name from `m`
+
+        self.assertEqual(str(m2).strip(), asm.strip())
+
+    def test_bitcode(self):
+        # create a module
+        m = Module.new('module1')
+        m.add_global_variable(Type.int(), 'i')
+
+        # write it's assembly representation to a file
+        asm = str(m)
+
+        testasm_bc = os.path.join(self.tmpdir, 'testasm.bc')
+        with open(testasm_bc, "wb") as fout:
+            m.to_bitcode(fout)
+
+        # read it back into a module
+        with open(testasm_bc, "rb") as fin:
+            m2 = Module.from_bitcode(fin)
+            # The default `m.id` is '<string>'.
+            m2.id = m.id # Copy the name from `m`
+
+        self.assertEqual(str(m2).strip(), asm.strip())
+
+tests.append(TestAsm)
+
+if __name__ == '__main__':
+    unittest.main()

llvm/tests/test_atomic.py

@@ -0,0 +1,85 @@
+import unittest
+from llvm.core import (Module, Type, Builder, Constant)
+from .support import TestCase, tests
+
+class TestAtomic(TestCase):
+    orderings = ['unordered', 'monotonic', 'acquire',
+                 'release', 'acq_rel', 'seq_cst']
+
+    atomic_op = ['xchg', 'add', 'sub', 'and', 'nand', 'or', 'xor',
+                 'max', 'min', 'umax', 'umin']
+
+    def test_atomic_cmpxchg(self):
+        mod = Module.new('mod')
+        functype = Type.function(Type.void(), [])
+        func = mod.add_function(functype, name='foo')
+        bb = func.append_basic_block('entry')
+        bldr = Builder.new(bb)
+        ptr = bldr.alloca(Type.int())
+
+        old = bldr.load(ptr)
+        new = Constant.int(Type.int(), 1234)
+
+        for ordering in self.orderings:
+            inst = bldr.atomic_cmpxchg(ptr, old, new, ordering)
+            self.assertEqual(ordering, str(inst).strip().split(' ')[-1])
+
+        inst = bldr.atomic_cmpxchg(ptr, old, new, ordering, crossthread=False)
+        self.assertEqual('singlethread', str(inst).strip().split(' ')[-2])
+
+    def test_atomic_rmw(self):
+        mod = Module.new('mod')
+        functype = Type.function(Type.void(), [])
+        func = mod.add_function(functype, name='foo')
+        bb = func.append_basic_block('entry')
+        bldr = Builder.new(bb)
+        ptr = bldr.alloca(Type.int())
+
+        val = Constant.int(Type.int(), 1234)
+
+        for ordering in self.orderings:
+            inst = bldr.atomic_rmw('xchg', ptr, val, ordering)
+            self.assertEqual(ordering, str(inst).split(' ')[-1])
+
+        for op in self.atomic_op:
+            inst = bldr.atomic_rmw(op, ptr, val, ordering)
+            self.assertEqual(op, str(inst).strip().split(' ')[3])
+
+        inst = bldr.atomic_rmw('xchg', ptr, val, ordering, crossthread=False)
+        self.assertEqual('singlethread', str(inst).strip().split(' ')[-2])
+
+        for op in self.atomic_op:
+            atomic_op = getattr(bldr, 'atomic_%s' % op)
+            inst = atomic_op(ptr, val, ordering)
+            self.assertEqual(op, str(inst).strip().split(' ')[3])
+
+    def test_atomic_ldst(self):
+        mod = Module.new('mod')
+        functype = Type.function(Type.void(), [])
+        func = mod.add_function(functype, name='foo')
+        bb = func.append_basic_block('entry')
+        bldr = Builder.new(bb)
+        ptr = bldr.alloca(Type.int())
+
+        for ordering in self.orderings:
+            loaded = bldr.atomic_load(ptr, ordering)
+            self.assert_('load atomic' in str(loaded))
+            self.assertEqual(ordering,
+                             str(loaded).strip().split(' ')[-3].rstrip(','))
+            self.assert_('align 1' in str(loaded))
+
+            stored = bldr.atomic_store(loaded, ptr, ordering)
+            self.assert_('store atomic' in str(stored))
+            self.assertEqual(ordering,
+                             str(stored).strip().split(' ')[-3].rstrip(','))
+            self.assert_('align 1' in str(stored))
+
+            fenced = bldr.fence(ordering)
+            self.assertEqual(['fence', ordering],
+                             str(fenced).strip().split(' '))
+
+tests.append(TestAtomic)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_attr.py

@@ -0,0 +1,33 @@
+import unittest
+try:
+    from StringIO import StringIO
+except ImportError:
+    from io import StringIO
+
+from llvm.core import Module
+from .support import TestCase, tests
+
+class TestAttr(TestCase):
+    def make_module(self):
+        test_module = """
+            define void @sum(i32*, i32*) {
+            entry:
+                ret void
+            }
+        """
+        buf = StringIO(test_module)
+        return Module.from_assembly(buf)
+
+    def test_align(self):
+        m = self.make_module()
+        f = m.get_function_named('sum')
+        f.args[0].alignment = 16
+        self.assert_("align 16" in str(f))
+        self.assertEqual(f.args[0].alignment, 16)
+
+tests.append(TestAttr)
+
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_cmp.py

@@ -0,0 +1,26 @@
+import unittest
+from llvm.core import (Module, Type, Builder, Constant)
+import llvm.core as lc
+from .support import TestCase, tests
+
+class TestCmp(TestCase):
+    def test_arg_attr(self):
+        m = Module.new('oifjda')
+        fnty = Type.function(Type.void(), [Type.int()])
+        func = m.add_function(fnty, 'foo')
+        bb = func.append_basic_block('')
+        bldr = Builder.new(bb)
+
+        cmpinst = bldr.icmp(lc.ICMP_ULE, func.args[0],
+                            Constant.int(Type.int(), 123))
+        self.assertTrue(repr(cmpinst.predicate).startswith('ICMP_ULE'))
+        self.assertEqual(cmpinst.predicate, lc.ICMP_ULE)
+        bldr.ret_void()
+
+        func.verify()
+
+tests.append(TestCmp)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_const_expr.py

@@ -0,0 +1,23 @@
+import unittest
+from llvm.core import (Module, Type, Builder, Constant)
+import llvm.core as lc
+
+from .support import TestCase, tests
+
+class TestConstExpr(TestCase):
+
+    def test_constexpr_opcode(self):
+        mod = Module.new('test_constexpr_opcode')
+        func = mod.add_function(Type.function(Type.void(), []), name="foo")
+        builder = Builder.new(func.append_basic_block('entry'))
+        a = builder.inttoptr(Constant.int(Type.int(), 123),
+                             Type.pointer(Type.int()))
+        self.assertTrue(isinstance(a, lc.ConstantExpr))
+        self.assertEqual(a.opcode, lc.OPCODE_INTTOPTR)
+        self.assertEqual(a.opcode_name, "inttoptr")
+
+tests.append(TestConstExpr)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_cpu_support.py

@@ -0,0 +1,89 @@
+import unittest
+import math
+from llvm.core import (Module, Type, Function, Builder,
+                       Constant)
+from llvm.ee import EngineBuilder
+import llvm.core as lc
+import llvm.ee as le
+from llvm.workaround.avx_support import detect_avx_support
+from .support import TestCase, tests, skip_if_not_intel_cpu, skip_if
+
+@skip_if_not_intel_cpu
+class TestCPUSupport(TestCase):
+
+    def _build_test_module(self):
+        mod     = Module.new('test')
+
+        float   = Type.double()
+        mysinty = Type.function( float, [float] )
+        mysin   = mod.add_function(mysinty, "mysin")
+        block   = mysin.append_basic_block("entry")
+        b       = Builder.new(block)
+
+        sqrt = Function.intrinsic(mod, lc.INTR_SQRT, [float])
+        pow  = Function.intrinsic(mod, lc.INTR_POWI, [float])
+        cos  = Function.intrinsic(mod, lc.INTR_COS,  [float])
+
+        mysin.args[0].name = "x"
+        x    = mysin.args[0]
+        one  = Constant.real(float, "1")
+        cosx = b.call(cos, [x], "cosx")
+        cos2 = b.call(pow, [cosx, Constant.int(Type.int(), 2)], "cos2")
+        onemc2 = b.fsub(one, cos2, "onemc2") # Should use fsub
+        sin  = b.call(sqrt, [onemc2], "sin")
+        b.ret(sin)
+        return mod, mysin
+
+    def _template(self, mattrs):
+        mod, func = self._build_test_module()
+        ee = self._build_engine(mod, mattrs=mattrs)
+        arg = le.GenericValue.real(Type.double(), 1.234)
+        retval = ee.run_function(func, [arg])
+
+        golden = math.sin(1.234)
+        answer = retval.as_real(Type.double())
+        self.assertTrue(abs(answer-golden)/golden < 1e-5)
+
+
+    def _build_engine(self, mod, mattrs):
+        if mattrs:
+            return EngineBuilder.new(mod).mattrs(mattrs).create()
+        else:
+            return EngineBuilder.new(mod).create()
+
+    def test_cpu_support2(self):
+        features = 'sse3', 'sse41', 'sse42', 'avx'
+        mattrs = ','.join(map(lambda s: '-%s' % s, features))
+        print('disable mattrs', mattrs)
+        self._template(mattrs)
+
+    def test_cpu_support3(self):
+        features = 'sse41', 'sse42', 'avx'
+        mattrs = ','.join(map(lambda s: '-%s' % s, features))
+        print('disable mattrs', mattrs)
+        self._template(mattrs)
+
+    def test_cpu_support4(self):
+        features = 'sse42', 'avx'
+        mattrs = ','.join(map(lambda s: '-%s' % s, features))
+        print('disable mattrs', mattrs)
+        self._template(mattrs)
+
+    def test_cpu_support5(self):
+        features = 'avx',
+        mattrs = ','.join(map(lambda s: '-%s' % s, features))
+        print('disable mattrs', mattrs)
+        self._template(mattrs)
+
+    @skip_if(not detect_avx_support(), msg="no AVX support")
+    def test_cpu_support6(self):
+        features = []
+        mattrs = ','.join(map(lambda s: '-%s' % s, features))
+        print('disable mattrs', mattrs)
+        self._template(mattrs)
+
+tests.append(TestCPUSupport)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_engine_builder.py

@@ -0,0 +1,65 @@
+import unittest
+from llvm.core import Module, Type, Builder, Constant
+from llvm.ee import EngineBuilder
+import llvm.ee as le
+import llvmpy
+
+from .support import TestCase, tests
+
+class TestEngineBuilder(TestCase):
+
+    def make_test_module(self):
+        module = Module.new("testmodule")
+        fnty = Type.function(Type.int(), [])
+        function = module.add_function(fnty, 'foo')
+        bb_entry = function.append_basic_block('entry')
+        builder = Builder.new(bb_entry)
+        builder.ret(Constant.int(Type.int(), 0xcafe))
+        module.verify()
+        return module
+
+    def run_foo(self, ee, module):
+        function = module.get_function_named('foo')
+        retval = ee.run_function(function, [])
+        self.assertEqual(retval.as_int(), 0xcafe)
+
+
+    def test_enginebuilder_basic(self):
+        module = self.make_test_module()
+        self.assertTrue(llvmpy.capsule.has_ownership(module._ptr._ptr))
+        ee = EngineBuilder.new(module).create()
+        self.assertFalse(llvmpy.capsule.has_ownership(module._ptr._ptr))
+        self.run_foo(ee, module)
+
+
+    def test_enginebuilder_with_tm(self):
+        tm = le.TargetMachine.new()
+        module = self.make_test_module()
+        self.assertTrue(llvmpy.capsule.has_ownership(module._ptr._ptr))
+        ee = EngineBuilder.new(module).create(tm)
+        self.assertFalse(llvmpy.capsule.has_ownership(module._ptr._ptr))
+        self.run_foo(ee, module)
+
+    def test_enginebuilder_force_jit(self):
+        module = self.make_test_module()
+        ee = EngineBuilder.new(module).force_jit().create()
+
+        self.run_foo(ee, module)
+#
+#    def test_enginebuilder_force_interpreter(self):
+#        module = self.make_test_module()
+#        ee = EngineBuilder.new(module).force_interpreter().create()
+#
+#        self.run_foo(ee, module)
+
+    def test_enginebuilder_opt(self):
+        module = self.make_test_module()
+        ee = EngineBuilder.new(module).opt(3).create()
+
+        self.run_foo(ee, module)
+
+tests.append(TestEngineBuilder)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_exact.py

@@ -0,0 +1,37 @@
+import unittest
+from llvm.core import (Module, Type, Builder)
+from .support import TestCase, tests
+
+class TestExact(TestCase):
+    def make_module(self):
+        mod = Module.new('asdfa')
+        fnty = Type.function(Type.void(), [Type.int()] * 2)
+        func = mod.add_function(fnty, 'foo')
+        bldr = Builder.new(func.append_basic_block(''))
+        return mod, func, bldr
+
+    def has_exact(self, inst, op):
+        self.assertTrue(('%s exact' % op) in str(inst), "exact flag does not work")
+
+    def _test_template(self, opf, opname):
+        mod, func, bldr = self.make_module()
+        a, b = func.args
+        self.has_exact(opf(bldr, a, b, exact=True), opname)
+
+    def test_udiv_exact(self):
+        self._test_template(Builder.udiv, 'udiv')
+
+    def test_sdiv_exact(self):
+        self._test_template(Builder.sdiv, 'sdiv')
+
+    def test_lshr_exact(self):
+        self._test_template(Builder.lshr, 'lshr')
+
+    def test_ashr_exact(self):
+        self._test_template(Builder.ashr, 'ashr')
+
+tests.append(TestExact)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_execution_engine.py

@@ -0,0 +1,47 @@
+import unittest
+from llvm.core import Type
+import llvm.core as lc
+import llvm.ee as le
+
+from .support import TestCase, tests
+
+class TestExecutionEngine(TestCase):
+    def test_get_pointer_to_global(self):
+        module = lc.Module.new(str(self))
+        gvar = module.add_global_variable(Type.int(), 'hello')
+        X = 1234
+        gvar.initializer = lc.Constant.int(Type.int(), X)
+
+        ee = le.ExecutionEngine.new(module)
+        ptr = ee.get_pointer_to_global(gvar)
+        from ctypes import c_void_p, cast, c_int, POINTER
+        casted = cast(c_void_p(ptr), POINTER(c_int))
+        self.assertEqual(X, casted[0])
+
+    def test_add_global_mapping(self):
+        module = lc.Module.new(str(self))
+        gvar = module.add_global_variable(Type.int(), 'hello')
+
+        fnty = lc.Type.function(Type.int(), [])
+        foo = module.add_function(fnty, name='foo')
+        bldr = lc.Builder.new(foo.append_basic_block('entry'))
+        bldr.ret(bldr.load(gvar))
+
+        ee = le.ExecutionEngine.new(module)
+        from ctypes import c_int, addressof, CFUNCTYPE
+        value = 0xABCD
+        value_ctype = c_int(value)
+        value_pointer = addressof(value_ctype)
+
+        ee.add_global_mapping(gvar, value_pointer)
+
+        foo_addr = ee.get_pointer_to_function(foo)
+        prototype = CFUNCTYPE(c_int)
+        foo_callable = prototype(foo_addr)
+        self.assertEqual(foo_callable(), value)
+
+tests.append(TestExecutionEngine)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_inlining.py

@@ -0,0 +1,34 @@
+import unittest
+from llvm.core import (Module, Type, Builder, Constant, inline_function)
+from .support import TestCase, tests
+
+class TestInlining(TestCase):
+    def test_inline_call(self):
+        mod = Module.new(__name__)
+        callee = mod.add_function(Type.function(Type.int(), [Type.int()]),
+                                  name='bar')
+
+        builder = Builder.new(callee.append_basic_block('entry'))
+        builder.ret(builder.add(callee.args[0], callee.args[0]))
+
+        caller = mod.add_function(Type.function(Type.int(), []),
+                                  name='foo')
+
+        builder = Builder.new(caller.append_basic_block('entry'))
+        callinst = builder.call(callee, [Constant.int(Type.int(), 1234)])
+        builder.ret(callinst)
+
+        pre_inlining = str(caller)
+        self.assertIn('call', pre_inlining)
+
+        self.assertTrue(inline_function(callinst))
+
+        post_inlining = str(caller)
+        self.assertNotIn('call', post_inlining)
+        self.assertIn('2468', post_inlining)
+
+tests.append(TestInlining)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_intel_native_asm.py

@@ -0,0 +1,31 @@
+import sys
+import os
+import unittest
+from llvm.core import Builder, Module, Type
+import llvm.core as lc
+from .support import TestCase, skip_if_not_intel_cpu, isolated_tests
+
+@skip_if_not_intel_cpu
+class TestNativeAsm(TestCase):
+
+    def test_asm(self):
+        m = Module.new('module1')
+
+        foo = m.add_function(Type.function(Type.int(),
+                                           [Type.int(), Type.int()]),
+                             name="foo")
+        bldr = Builder.new(foo.append_basic_block('entry'))
+        x = bldr.add(foo.args[0], foo.args[1])
+        bldr.ret(x)
+
+        att_syntax = m.to_native_assembly()
+        os.environ["LLVMPY_OPTIONS"] = "-x86-asm-syntax=intel"
+        lc.parse_environment_options(sys.argv[0], "LLVMPY_OPTIONS")
+        intel_syntax = m.to_native_assembly()
+
+        self.assertNotEqual(att_syntax, intel_syntax)
+
+isolated_tests.append(__name__)
+
+if __name__ == '__main__':
+    unittest.main()

llvm/tests/test_intrinsic.py

@@ -0,0 +1,113 @@
+import unittest
+import sys
+import math
+from llvm.core import (Module, Type, Function, Builder, Constant)
+import llvm.core as lc
+import llvm.ee as le
+from .support import TestCase, tests, BITS
+
+class TestIntrinsic(TestCase):
+    def test_bswap(self):
+        # setup a function and a builder
+        mod    = Module.new('test')
+        functy = Type.function(Type.int(), [])
+        func   = mod.add_function(functy, "showme")
+        block  = func.append_basic_block("entry")
+        b      = Builder.new(block)
+
+        # let's do bswap on a 32-bit integer using llvm.bswap
+        val   = Constant.int(Type.int(), 0x42)
+        bswap = Function.intrinsic(mod, lc.INTR_BSWAP, [Type.int()])
+
+        bswap_res = b.call(bswap, [val])
+        b.ret(bswap_res)
+
+        # logging.debug(mod)
+
+        # the output is:
+        #
+        #    ; ModuleID = 'test'
+        #
+        #    define void @showme() {
+        #    entry:
+        #      %0 = call i32 @llvm.bswap.i32(i32 42)
+        #      ret i32 %0
+        #    }
+
+        # let's run the function
+        ee = le.ExecutionEngine.new(mod)
+        retval = ee.run_function(func, [])
+        self.assertEqual(retval.as_int(), 0x42000000)
+
+    def test_mysin(self):
+        if sys.platform == 'win32' and BITS == 32:
+            # float32 support is known to fail on 32-bit Windows
+            return
+
+        # mysin(x) = sqrt(1.0 - pow(cos(x), 2))
+        mod     = Module.new('test')
+
+        float   = Type.float()
+        mysinty = Type.function( float, [float] )
+        mysin   = mod.add_function(mysinty, "mysin")
+        block   = mysin.append_basic_block("entry")
+        b       = Builder.new(block)
+
+        sqrt = Function.intrinsic(mod, lc.INTR_SQRT, [float])
+        pow  = Function.intrinsic(mod, lc.INTR_POWI, [float])
+        cos  = Function.intrinsic(mod, lc.INTR_COS,  [float])
+
+        mysin.args[0].name = "x"
+        x    = mysin.args[0]
+        one  = Constant.real(float, "1")
+        cosx = b.call(cos, [x], "cosx")
+        cos2 = b.call(pow, [cosx, Constant.int(Type.int(), 2)], "cos2")
+        onemc2 = b.fsub(one, cos2, "onemc2") # Should use fsub
+        sin  = b.call(sqrt, [onemc2], "sin")
+        b.ret(sin)
+        #logging.debug(mod)
+
+#   ; ModuleID = 'test'
+#
+#   define void @showme() {
+#   entry:
+#       call i32 @llvm.bswap.i32( i32 42 )              ; <i32>:0 [#uses
+#   }
+#
+#   declare i32 @llvm.bswap.i32(i32) nounwind readnone
+#
+#   define float @mysin(float %x) {
+#   entry:
+#       %cosx = call float @llvm.cos.f32( float %x )            ; <float
+#       %cos2 = call float @llvm.powi.f32( float %cosx, i32 2 )
+#       %onemc2 = sub float 1.000000e+00, %cos2         ; <float> [#uses
+#       %sin = call float @llvm.sqrt.f32( float %onemc2 )
+#       ret float %sin
+#   }
+#
+#   declare float @llvm.sqrt.f32(float) nounwind readnone
+#
+#   declare float @llvm.powi.f32(float, i32) nounwind readnone
+#
+#   declare float @llvm.cos.f32(float) nounwind readnone
+
+        # let's run the function
+
+        from llvm.workaround.avx_support import detect_avx_support
+        if not detect_avx_support():
+            ee = le.EngineBuilder.new(mod).mattrs("-avx").create()
+        else:
+            ee = le.EngineBuilder.new(mod).create()
+
+        arg = le.GenericValue.real(Type.float(), 1.234)
+        retval = ee.run_function(mysin, [arg])
+
+        golden = math.sin(1.234)
+        answer = retval.as_real(Type.float())
+        self.assertTrue(abs(answer-golden)/golden < 1e-5)
+
+tests.append(TestIntrinsic)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_intrinsic_basic.py

@@ -0,0 +1,99 @@
+import unittest
+import sys
+import math
+from llvm.core import (Module, Type, Function, Builder)
+import llvm.core as lc
+import llvm.ee as le
+from .support import TestCase, BITS, tests
+
+class TestIntrinsicBasic(TestCase):
+
+    def _build_module(self, float):
+        mod     = Module.new('test')
+        functy  = Type.function(float, [float])
+        func    = mod.add_function(functy, "mytest%s" % float)
+        block   = func.append_basic_block("entry")
+        b       = Builder.new(block)
+        return mod, func, b
+
+    def _template(self, mod, func, pyfunc):
+        float = func.type.pointee.return_type
+
+        from llvm.workaround.avx_support import detect_avx_support
+        if not detect_avx_support():
+            ee = le.EngineBuilder.new(mod).mattrs("-avx").create()
+        else:
+            ee = le.EngineBuilder.new(mod).create()
+        arg = le.GenericValue.real(float, 1.234)
+        retval = ee.run_function(func, [arg])
+        golden = pyfunc(1.234)
+        answer = retval.as_real(float)
+        self.assertTrue(abs(answer - golden) / golden < 1e-7)
+
+    def test_sqrt_f32(self):
+        float = Type.float()
+        mod, func, b = self._build_module(float)
+        intr = Function.intrinsic(mod, lc.INTR_SQRT, [float])
+        b.ret(b.call(intr, func.args))
+        self._template(mod, func, math.sqrt)
+
+    def test_sqrt_f64(self):
+        float = Type.double()
+        mod, func, b = self._build_module(float)
+        intr = Function.intrinsic(mod, lc.INTR_SQRT, [float])
+        b.ret(b.call(intr, func.args))
+        self._template(mod, func, math.sqrt)
+
+    def test_cos_f32(self):
+        if sys.platform == 'win32' and BITS == 32:
+            # float32 support is known to fail on 32-bit Windows
+            return
+        float = Type.float()
+        mod, func, b = self._build_module(float)
+        intr = Function.intrinsic(mod, lc.INTR_COS, [float])
+        b.ret(b.call(intr, func.args))
+        self._template(mod, func, math.cos)
+
+    def test_cos_f64(self):
+        float = Type.double()
+        mod, func, b = self._build_module(float)
+        intr = Function.intrinsic(mod, lc.INTR_COS, [float])
+        b.ret(b.call(intr, func.args))
+        self._template(mod, func, math.cos)
+
+    def test_sin_f32(self):
+        if sys.platform == 'win32' and BITS == 32:
+            # float32 support is known to fail on 32-bit Windows
+            return
+        float = Type.float()
+        mod, func, b = self._build_module(float)
+        intr = Function.intrinsic(mod, lc.INTR_SIN, [float])
+        b.ret(b.call(intr, func.args))
+        self._template(mod, func, math.sin)
+
+    def test_sin_f64(self):
+        float = Type.double()
+        mod, func, b = self._build_module(float)
+        intr = Function.intrinsic(mod, lc.INTR_SIN, [float])
+        b.ret(b.call(intr, func.args))
+        self._template(mod, func, math.sin)
+
+    def test_powi_f32(self):
+        float = Type.float()
+        mod, func, b = self._build_module(float)
+        intr = Function.intrinsic(mod, lc.INTR_POWI, [float])
+        b.ret(b.call(intr, [func.args[0], lc.Constant.int(Type.int(), 2)]))
+        self._template(mod, func, lambda x: x**2)
+
+    def test_powi_f64(self):
+        float = Type.double()
+        mod, func, b = self._build_module(float)
+        intr = Function.intrinsic(mod, lc.INTR_POWI, [float])
+        b.ret(b.call(intr, [func.args[0], lc.Constant.int(Type.int(), 2)]))
+        self._template(mod, func, lambda x: x**2)
+
+tests.append(TestIntrinsicBasic)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_issue_10.py

@@ -0,0 +1,26 @@
+import unittest
+from llvm.core import (Module, Type, Builder)
+from .support import TestCase, tests
+
+class TestIssue10(TestCase):
+    def test_issue10(self):
+        m = Module.new('a')
+        ti = Type.int()
+        tf = Type.function(ti, [ti, ti])
+
+        f = m.add_function(tf, "func1")
+
+        bb = f.append_basic_block('entry')
+
+        b = Builder.new(bb)
+
+        # There are no instructions in bb. Positioning of the
+        # builder at beginning (or end) should succeed (trivially).
+        b.position_at_end(bb)
+        b.position_at_beginning(bb)
+
+tests.append(TestIssue10)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_llrt.py

@@ -0,0 +1,34 @@
+import unittest
+import llvm.core as lc
+import llvm.ee as le
+from .support import TestCase, tests
+
+class TestLLRT(TestCase):
+    def test_llrt_divmod(self):
+        from llvm import llrt
+        m = lc.Module.new('testllrt')
+        longlong = lc.Type.int(64)
+        lfunc = m.add_function(lc.Type.function(longlong, [longlong, longlong]), 'foo')
+        bldr = lc.Builder.new(lfunc.append_basic_block(''))
+        bldr.ret(bldr.udiv(*lfunc.args))
+
+        llrt.replace_divmod64(lfunc)
+
+        rt = llrt.LLRT()
+        rt.install_symbols()
+
+        engine = le.EngineBuilder.new(m).create()
+        pointer = engine.get_pointer_to_function(lfunc)
+
+        from ctypes import CFUNCTYPE, c_uint64
+        func = CFUNCTYPE(c_uint64, c_uint64, c_uint64)(pointer)
+        a, b = 98342, 2231
+        self.assertEqual(func(98342, 2231), 98342 // 2231)
+
+        rt.uninstall_symbols()
+
+tests.append(TestLLRT)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_mcjit.py

@@ -0,0 +1,73 @@
+import unittest
+import sys
+import llvm
+from llvm.core import (Module, Type, Builder)
+from llvm.ee import EngineBuilder
+
+import llvm.ee as le
+from .support import TestCase, tests, BITS
+
+class TestMCJIT(TestCase):
+    def test_mcjit(self):
+        m = Module.new('oidfjs')
+        fnty = Type.function(Type.int(), [Type.int(), Type.int()])
+        func = m.add_function(fnty, 'foo')
+        bb = func.append_basic_block('')
+        bldr = Builder.new(bb)
+        bldr.ret(bldr.add(*func.args))
+
+        func.verify()
+
+        engine = EngineBuilder.new(m).mcjit(True).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        from ctypes import c_int, CFUNCTYPE
+        callee = CFUNCTYPE(c_int, c_int, c_int)(ptr)
+        self.assertEqual(321 + 123, callee(321, 123))
+
+    def test_multi_module_linking(self):
+        # generate external library module
+        m = Module.new('external-library-module')
+        fnty = Type.function(Type.int(), [Type.int(), Type.int()])
+        libfname = 'myadd'
+        func = m.add_function(fnty, libfname)
+        bb = func.append_basic_block('')
+        bldr = Builder.new(bb)
+        bldr.ret(bldr.add(*func.args))
+        func.verify()
+
+        # JIT the lib module and bind dynamic symbol
+        libengine = EngineBuilder.new(m).mcjit(True).create()
+        myadd_ptr = libengine.get_pointer_to_function(func)
+        le.dylib_add_symbol(libfname, myadd_ptr)
+
+        # reference external library
+        m = Module.new('user')
+        fnty = Type.function(Type.int(), [Type.int(), Type.int()])
+        func = m.add_function(fnty, 'foo')
+        bb = func.append_basic_block('')
+        bldr = Builder.new(bb)
+        extadd = m.get_or_insert_function(fnty, name=libfname)
+        bldr.ret(bldr.call(extadd, func.args))
+        func.verify()
+
+        # JIT the user module
+        engine = EngineBuilder.new(m).mcjit(True).create()
+        ptr = engine.get_pointer_to_function(func)
+        self.assertEqual(myadd_ptr,
+                         engine.get_pointer_to_named_function(libfname))
+
+        from ctypes import c_int, CFUNCTYPE
+        callee = CFUNCTYPE(c_int, c_int, c_int)(ptr)
+        self.assertEqual(321 + 123, callee(321, 123))
+
+
+if (llvm.version >= (3, 3) and
+    not (sys.platform.startswith('win32') and BITS == 64)):
+    # MCJIT broken in 3.2, the test will segfault in OSX?
+    # Compatbility problem on windows 7 64-bit?
+    tests.append(TestMCJIT)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_metadata.py

@@ -0,0 +1,25 @@
+import unittest
+from llvm.core import (Module, Type, Constant, MetaData, MetaDataString)
+from .support import TestCase, tests
+
+class TestMetaData(TestCase):
+    # test module metadata
+    def test_metadata(self):
+        m = Module.new('a')
+        t = Type.int()
+        metadata = MetaData.get(m, [Constant.int(t, 100),
+                                    MetaDataString.get(m, 'abcdef'),
+                                    None])
+        MetaData.add_named_operand(m, 'foo', metadata)
+        self.assertEqual(MetaData.get_named_operands(m, 'foo'), [metadata])
+        self.assertEqual(MetaData.get_named_operands(m, 'bar'), [])
+        self.assertEqual(len(metadata.operands), 3)
+        self.assertEqual(metadata.operands[0].z_ext_value, 100)
+        self.assertEqual(metadata.operands[1].string, 'abcdef')
+        self.assertTrue(metadata.operands[2] is None)
+
+tests.append(TestMetaData)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_named_metadata.py

@@ -0,0 +1,20 @@
+import unittest
+from llvm.core import (Module, Type, Constant, MetaData)
+from .support import TestCase, tests
+
+
+class TestNamedMetaData(TestCase):
+    def test_named_md(self):
+        m = Module.new('test_named_md')
+        nmd = m.get_or_insert_named_metadata('something')
+        md = MetaData.get(m, [Constant.int(Type.int(), 0xbeef)])
+        nmd.add(md)
+        self.assertTrue(str(nmd).startswith('!something'))
+        ir = str(m)
+        self.assertTrue('!something' in ir)
+
+tests.append(TestNamedMetaData)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_native.py

@@ -0,0 +1,81 @@
+import unittest
+import os
+import sys
+import shutil
+import subprocess
+import tempfile
+from distutils.spawn import find_executable
+from llvm.core import (Module, Type, Builder, Constant)
+from .support import TestCase, IS_PY3K, tests, skip_if
+
+@skip_if(sys.platform in ('win32', 'darwin'))
+class TestNative(TestCase):
+
+    def setUp(self):
+        self.tmpdir = tempfile.mkdtemp()
+
+    def tearDown(self):
+        shutil.rmtree(self.tmpdir)
+
+
+    def _make_module(self):
+        m = Module.new('module1')
+        m.add_global_variable(Type.int(), 'i')
+
+        fty = Type.function(Type.int(), [])
+        f = m.add_function(fty, name='main')
+
+        bldr = Builder.new(f.append_basic_block('entry'))
+        bldr.ret(Constant.int(Type.int(), 0xab))
+
+        return m
+
+    def _compile(self, src):
+        cc = find_executable('cc')
+        if not cc:
+            return
+
+        dst = os.path.join(self.tmpdir, 'llvmobj.out')
+        s = subprocess.call([cc, '-o', dst, src])
+        if s != 0:
+            raise Exception("Cannot compile")
+
+        s = subprocess.call([dst])
+        self.assertEqual(s, 0xab)
+
+    def test_assembly(self):
+        #        if sys.platform == 'darwin':
+        #            # skip this test on MacOSX for now
+        #            return
+
+        m = self._make_module()
+        output = m.to_native_assembly()
+
+        src = os.path.join(self.tmpdir, 'llvmasm.s')
+        with open(src, 'wb') as fout:
+            if IS_PY3K:
+                fout.write(output.encode('utf-8'))
+            else:
+                fout.write(output)
+
+        self._compile(src)
+
+    def test_object(self):
+        '''
+        Note: Older Darwin with GCC will report missing _main symbol when
+              compile the object file to an executable.
+        '''
+        m = self._make_module()
+        output = m.to_native_object()
+
+        src = os.path.join(self.tmpdir, 'llvmobj.o')
+        with open(src, 'wb') as fout:
+            fout.write(output)
+
+        self._compile(src)
+
+tests.append(TestNative)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_nuw_nsw.py

@@ -0,0 +1,48 @@
+import unittest
+from llvm.core import Module, Type, Builder
+from .support import TestCase, tests
+
+class TestNUWNSW(TestCase):
+    def make_module(self):
+        mod = Module.new('asdfa')
+        fnty = Type.function(Type.void(), [Type.int()] * 2)
+        func = mod.add_function(fnty, 'foo')
+        bldr = Builder.new(func.append_basic_block(''))
+        return mod, func, bldr
+
+    def has_nsw(self, inst, op):
+        self.assertTrue(('%s nsw' % op) in str(inst), "NSW flag does not work")
+
+    def has_nuw(self, inst, op):
+        self.assertTrue(('%s nuw' % op) in str(inst), "NUW flag does not work")
+
+    def _test_template(self, opf, opname):
+        mod, func, bldr = self.make_module()
+        a, b = func.args
+        self.has_nsw(opf(bldr, a, b, nsw=True), opname)
+        self.has_nuw(opf(bldr, a, b, nuw=True), opname)
+
+    def test_add_nuw_nsw(self):
+        self._test_template(Builder.add, 'add')
+
+    def test_sub_nuw_nsw(self):
+        self._test_template(Builder.sub, 'sub')
+
+    def test_mul_nuw_nsw(self):
+        self._test_template(Builder.mul, 'mul')
+
+    def test_shl_nuw_nsw(self):
+        self._test_template(Builder.shl, 'shl')
+
+    def test_neg_nuw_nsw(self):
+        mod, func, bldr = self.make_module()
+        a, b = func.args
+        self.has_nsw(bldr.neg(a, nsw=True), 'sub')
+        self.has_nuw(bldr.neg(a, nuw=True), 'sub')
+
+
+tests.append(TestNUWNSW)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_obj_cache.py

@@ -0,0 +1,107 @@
+import unittest
+from llvm.core import Module, Type, GlobalVariable, Function, Builder
+from .support import TestCase, tests
+
+class TestObjCache(TestCase):
+
+    def test_objcache(self):
+        # Testing module aliasing
+        m1 = Module.new('a')
+        t = Type.int()
+        ft = Type.function(t, [t])
+        f1 = m1.add_function(ft, "func")
+        m2 = f1.module
+        self.assert_(m1 is m2)
+
+        # Testing global vairable aliasing 1
+        gv1 = GlobalVariable.new(m1, t, "gv")
+        gv2 = GlobalVariable.get(m1, "gv")
+        self.assert_(gv1 is gv2)
+
+        # Testing global vairable aliasing 2
+        gv3 = m1.global_variables[0]
+        self.assert_(gv1 is gv3)
+
+        # Testing global vairable aliasing 3
+        gv2 = None
+        gv3 = None
+
+        gv1.delete()
+
+        gv4 = GlobalVariable.new(m1, t, "gv")
+
+        self.assert_(gv1 is not gv4)
+
+        # Testing function aliasing 1
+        b1 = f1.append_basic_block('entry')
+        f2 = b1.function
+        self.assert_(f1 is f2)
+
+        # Testing function aliasing 2
+        f3 = m1.get_function_named("func")
+        self.assert_(f1 is f3)
+
+        # Testing function aliasing 3
+        f4 = Function.get_or_insert(m1, ft, "func")
+        self.assert_(f1 is f4)
+
+        # Testing function aliasing 4
+        f5 = Function.get(m1, "func")
+        self.assert_(f1 is f5)
+
+        # Testing function aliasing 5
+        f6 = m1.get_or_insert_function(ft, "func")
+        self.assert_(f1 is f6)
+
+        # Testing function aliasing 6
+        f7 = m1.functions[0]
+        self.assert_(f1 is f7)
+
+        # Testing argument aliasing
+        a1 = f1.args[0]
+        a2 = f1.args[0]
+        self.assert_(a1 is a2)
+
+        # Testing basic block aliasing 1
+        b2 = f1.basic_blocks[0]
+        self.assert_(b1 is b2)
+
+        # Testing basic block aliasing 2
+        b3 = f1.entry_basic_block
+        self.assert_(b1 is b3)
+
+        # Testing basic block aliasing 3
+        b31 = f1.entry_basic_block
+        self.assert_(b1 is b31)
+
+        # Testing basic block aliasing 4
+        bldr = Builder.new(b1)
+        b4 = bldr.basic_block
+        self.assert_(b1 is b4)
+
+        # Testing basic block aliasing 5
+        i1 = bldr.ret_void()
+        b5 = i1.basic_block
+        self.assert_(b1 is b5)
+
+        # Testing instruction aliasing 1
+        i2 = b5.instructions[0]
+        self.assert_(i1 is i2)
+
+        # phi node
+        phi = bldr.phi(t)
+        phi.add_incoming(f1.args[0], b1)
+        v2 = phi.get_incoming_value(0)
+        b6 = phi.get_incoming_block(0)
+
+        # Testing PHI / basic block aliasing 5
+        self.assert_(b1 is b6)
+
+        # Testing PHI / value aliasing
+        self.assert_(f1.args[0] is v2)
+
+tests.append(TestObjCache)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_opaque.py

@@ -0,0 +1,38 @@
+import unittest
+import llvm
+from llvm.core import Type
+from .support import TestCase, tests
+
+class TestOpaque(TestCase):
+
+    def test_opaque(self):
+        # Create an opaque type
+        ts = Type.opaque('mystruct')
+        self.assertTrue('type opaque' in str(ts))
+        self.assertTrue(ts.is_opaque)
+        self.assertTrue(ts.is_identified)
+        self.assertFalse(ts.is_literal)
+        #print(ts)
+
+        # Create a recursive type
+        ts.set_body([Type.int(), Type.pointer(ts)])
+
+        self.assertEqual(ts.elements[0], Type.int())
+        self.assertEqual(ts.elements[1], Type.pointer(ts))
+        self.assertEqual(ts.elements[1].pointee, ts)
+        self.assertFalse(ts.is_opaque) # is not longer a opaque type
+        #print(ts)
+
+        with self.assertRaises(llvm.LLVMException):
+            # Cannot redefine
+            ts.set_body([])
+
+    def test_opaque_with_no_name(self):
+        with self.assertRaises(llvm.LLVMException):
+            Type.opaque('')
+
+tests.append(TestOpaque)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_operands.py

@@ -0,0 +1,66 @@
+import unittest
+try:
+    from StringIO import StringIO
+except ImportError:
+    from io import StringIO
+
+from llvm.core import Module
+from .support import TestCase, tests
+
+class TestOperands(TestCase):
+    # implement a test function
+    test_module = """
+define i32 @prod(i32, i32) {
+entry:
+        %2 = mul i32 %0, %1
+        ret i32 %2
+}
+
+define i32 @test_func(i32, i32, i32) {
+entry:
+        %tmp1 = call i32 @prod(i32 %0, i32 %1)
+        %tmp2 = add i32 %tmp1, %2
+        %tmp3 = add i32 %tmp2, 1
+        %tmp4 = add i32 %tmp3, -1
+        %tmp5 = add i64 -81985529216486895, 12297829382473034410
+        ret i32 %tmp4
+}
+"""
+    def test_operands(self):
+        m = Module.from_assembly(StringIO(self.test_module))
+
+        test_func = m.get_function_named("test_func")
+        prod = m.get_function_named("prod")
+
+        # test operands
+        i1 = test_func.basic_blocks[0].instructions[0]
+        i2 = test_func.basic_blocks[0].instructions[1]
+        i3 = test_func.basic_blocks[0].instructions[2]
+        i4 = test_func.basic_blocks[0].instructions[3]
+        i5 = test_func.basic_blocks[0].instructions[4]
+
+        self.assertEqual(i1.operand_count, 3)
+        self.assertEqual(i2.operand_count, 2)
+
+        self.assertEqual(i3.operands[1].z_ext_value, 1)
+        self.assertEqual(i3.operands[1].s_ext_value, 1)
+        self.assertEqual(i4.operands[1].z_ext_value, 0xffffffff)
+        self.assertEqual(i4.operands[1].s_ext_value, -1)
+        self.assertEqual(i5.operands[0].s_ext_value, -81985529216486895)
+        self.assertEqual(i5.operands[1].z_ext_value, 12297829382473034410)
+
+        self.assert_(i1.operands[-1] is prod)
+        self.assert_(i1.operands[0] is test_func.args[0])
+        self.assert_(i1.operands[1] is test_func.args[1])
+        self.assert_(i2.operands[0] is i1)
+        self.assert_(i2.operands[1] is test_func.args[2])
+        self.assertEqual(len(i1.operands), 3)
+        self.assertEqual(len(i2.operands), 2)
+
+        self.assert_(i1.called_function is prod)
+
+tests.append(TestOperands)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_passes.py

@@ -0,0 +1,143 @@
+import unittest
+try:
+    from StringIO import StringIO
+except ImportError:
+    from io import StringIO
+import llvm
+from llvm.core import Module
+import llvm.passes as lp
+import llvm.ee as le
+from .support import TestCase, tests
+
+
+class TestPasses(TestCase):
+    # Create a module.
+    asm = """
+
+define i32 @test() nounwind {
+    ret i32 42
+}
+
+define i32 @test1() nounwind  {
+entry:
+        %tmp = alloca i32
+        store i32 42, i32* %tmp, align 4
+        %tmp1 = load i32* %tmp, align 4
+        %tmp2 = call i32 @test()
+        %tmp3 = load i32* %tmp, align 4
+        %tmp4 = load i32* %tmp, align 4
+    ret i32 %tmp1
+}
+
+define i32 @test2() nounwind  {
+entry:
+        %tmp = call i32 @test()
+    ret i32 %tmp
+}
+"""
+    def test_passes(self):
+        m = Module.from_assembly(StringIO(self.asm))
+
+        fn_test1 = m.get_function_named('test1')
+        fn_test2 = m.get_function_named('test2')
+
+        original_test1 = str(fn_test1)
+        original_test2 = str(fn_test2)
+
+        # Let's run a module-level inlining pass. First, create a pass manager.
+        pm = lp.PassManager.new()
+
+        # Add the target data as the first "pass". This is mandatory.
+        pm.add(le.TargetData.new(''))
+
+        # Add the inlining pass.
+        pm.add(lp.PASS_INLINE)
+
+        # Run it!
+        pm.run(m)
+
+        # Done with the pass manager.
+        del pm
+
+        # Make sure test2 is inlined
+        self.assertNotEqual(str(fn_test2).strip(), original_test2.strip())
+
+        bb_entry = fn_test2.basic_blocks[0]
+
+        self.assertEqual(len(bb_entry.instructions), 1)
+        self.assertEqual(bb_entry.instructions[0].opcode_name, 'ret')
+
+        # Let's run a DCE pass on the the function 'test1' now. First create a
+        # function pass manager.
+        fpm = lp.FunctionPassManager.new(m)
+
+        # Add the target data as first "pass". This is mandatory.
+        fpm.add(le.TargetData.new(''))
+
+        # Add a DCE pass
+        fpm.add(lp.PASS_ADCE)
+
+        # Run the pass on the function 'test1'
+        fpm.run(m.get_function_named('test1'))
+
+        # Make sure test1 is modified
+        self.assertNotEqual(str(fn_test1).strip(), original_test1.strip())
+
+    def test_passes_with_pmb(self):
+        m = Module.from_assembly(StringIO(self.asm))
+
+        fn_test1 = m.get_function_named('test1')
+        fn_test2 = m.get_function_named('test2')
+
+        original_test1 = str(fn_test1)
+        original_test2 = str(fn_test2)
+
+        # Try out the PassManagerBuilder
+
+        pmb = lp.PassManagerBuilder.new()
+
+        self.assertEqual(pmb.opt_level, 2)  # ensure default is level 2
+        pmb.opt_level = 3
+        self.assertEqual(pmb.opt_level, 3) # make sure it works
+
+        self.assertEqual(pmb.size_level, 0) # ensure default is level 0
+        pmb.size_level = 2
+        self.assertEqual(pmb.size_level, 2) # make sure it works
+
+        self.assertFalse(pmb.vectorize) # ensure default is False
+        pmb.vectorize = True
+        self.assertTrue(pmb.vectorize) # make sure it works
+
+        # make sure the default is False
+        self.assertFalse(pmb.disable_unit_at_a_time)
+        self.assertFalse(pmb.disable_unroll_loops)
+        if llvm.version <= (3, 3):
+            self.assertFalse(pmb.disable_simplify_lib_calls)
+
+        pmb.disable_unit_at_a_time = True
+        self.assertTrue(pmb.disable_unit_at_a_time)
+
+        # Do function pass
+        fpm = lp.FunctionPassManager.new(m)
+        pmb.populate(fpm)
+        fpm.run(fn_test1)
+
+        # Make sure test1 has changed
+        self.assertNotEqual(str(fn_test1).strip(), original_test1.strip())
+
+        # Do module pass
+        pm = lp.PassManager.new()
+        pmb.populate(pm)
+        pm.run(m)
+
+        # Make sure test2 has changed
+        self.assertNotEqual(str(fn_test2).strip(), original_test2.strip())
+
+    def test_dump_passes(self):
+        self.assertTrue(len(lp.PASSES)>0, msg="Cannot have no passes")
+
+tests.append(TestPasses)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_struct.py

@@ -0,0 +1,28 @@
+import unittest
+from llvm.core import Type, Module, Builder, Constant
+from .support import TestCase, tests
+
+class TestStruct(TestCase):
+    def test_struct_identical(self):
+        ta = Type.struct([Type.int(32), Type.float()], name='ta')
+        tb = Type.struct([Type.int(32), Type.float()])
+        self.assertTrue(ta.is_layout_identical(tb))
+
+    def test_struct_extract_value_2d(self):
+        ta = Type.struct([Type.int(32), Type.float()])
+        tb = Type.struct([ta, Type.float()])
+        m = Module.new('')
+        f = m.add_function(Type.function(Type.void(), []), "foo")
+        b = Builder.new(f.append_basic_block(''))
+        v = Constant.undef(tb)
+        ins = b.insert_value(v, Constant.real(Type.float(), 1.234), [0, 1])
+        ext = b.extract_value(ins, [0, 1])
+        b.ret_void()
+        m.verify()
+        self.assertEqual(str(ext), 'float 0x3FF3BE76C0000000')
+
+tests.append(TestStruct)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_struct_args.py

@@ -0,0 +1,685 @@
+from __future__ import print_function
+from . import tests
+import sys
+import unittest
+from ctypes import Structure, c_float, c_double, c_uint8, CFUNCTYPE
+from llvm import core as lc
+from llvm import ee as le
+
+from .support import (skip_if_win32, skip_if_not_win32, skip_if_not_32bits,
+                      skip_if_not_64bits, skip_if_not_intel_cpu, TestCase)
+
+class TwoDoubleOneByte(Structure):
+    _fields_ = ('x', c_double), ('y', c_double), ('z', c_uint8)
+
+    def __repr__(self):
+        return '<x=%f y=%f z=%d>' % (self.x, self.y, self.z)
+
+class TwoDouble(Structure):
+    _fields_ = ('x', c_double), ('y', c_double)
+
+    def __repr__(self):
+        return '<x=%f y=%f>' % (self.x, self.y)
+
+class TwoFloat(Structure):
+    _fields_ = ('x', c_float), ('y', c_float)
+
+    def __repr__(self):
+        return '<x=%f y=%f>' % (self.x, self.y)
+
+class OneByte(Structure):
+    _fields_ = [('x', c_uint8)]
+
+    def __repr__(self):
+        return '<x=%d>' % (self.x,)
+
+@skip_if_not_intel_cpu
+@skip_if_win32
+class TestStructSystemVABI(TestCase):
+    '''
+    Non microsoft convention
+    '''
+
+    #----------------------------------------------------------------------
+    # 64 bits
+
+    @skip_if_not_64bits
+    def test_bigger_than_two_words_64(self):
+        m = lc.Module.new('test_struct_arg')
+
+        double_type = lc.Type.double()
+        uint8_type = lc.Type.int(8)
+        struct_type = lc.Type.struct([double_type, double_type, uint8_type])
+        struct_ptr_type = lc.Type.pointer(struct_type)
+        func_type = lc.Type.function(lc.Type.void(),
+                                     [struct_ptr_type, struct_ptr_type])
+        func = m.add_function(func_type, name='foo')
+
+        # return value pointer
+        func.args[0].add_attribute(lc.ATTR_STRUCT_RET)
+
+        # pass structure by value
+        func.args[1].add_attribute(lc.ATTR_BY_VAL)
+
+        # define function body
+        builder = lc.Builder.new(func.append_basic_block(''))
+
+        arg = builder.load(func.args[1])
+        e1, e2, e3 = [builder.extract_value(arg, i) for i in range(3)]
+        se1 = builder.fmul(e1, e2)
+        se2 = builder.fdiv(e1, e2)
+        ret = builder.insert_value(lc.Constant.undef(struct_type), se1, 0)
+        ret = builder.insert_value(ret, se2, 1)
+        ret = builder.insert_value(ret, e3, 2)
+        builder.store(ret, func.args[0])
+        builder.ret_void()
+
+        del builder
+
+        # verify
+        m.verify()
+
+        print(m)
+        # use with ctypes
+        engine = le.EngineBuilder.new(m).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        cfunctype = CFUNCTYPE(TwoDoubleOneByte, TwoDoubleOneByte)
+        cfunc = cfunctype(ptr)
+
+        arg = TwoDoubleOneByte(x=1.321321, y=6.54352, z=128)
+        ret = cfunc(arg)
+        print(arg)
+        print(ret)
+
+        self.assertClose(arg.x * arg.y, ret.x)
+        self.assertClose(arg.x / arg.y, ret.y)
+        self.assertEqual(arg.z, ret.z)
+
+    @skip_if_not_64bits
+    def test_just_two_words_64(self):
+        m = lc.Module.new('test_struct_arg')
+
+        double_type = lc.Type.double()
+        struct_type = lc.Type.struct([double_type, double_type])
+        func_type = lc.Type.function(struct_type, [struct_type])
+        func = m.add_function(func_type, name='foo')
+
+        # define function body
+        builder = lc.Builder.new(func.append_basic_block(''))
+
+        arg = func.args[0]
+        e1, e2 = [builder.extract_value(arg, i) for i in range(2)]
+        se1 = builder.fmul(e1, e2)
+        se2 = builder.fdiv(e1, e2)
+        ret = builder.insert_value(lc.Constant.undef(struct_type), se1, 0)
+        ret = builder.insert_value(ret, se2, 1)
+        builder.ret(ret)
+
+        del builder
+
+        # verify
+        m.verify()
+
+        print(m)
+        # use with ctypes
+        engine = le.EngineBuilder.new(m).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        cfunctype = CFUNCTYPE(TwoDouble, TwoDouble)
+        cfunc = cfunctype(ptr)
+
+        arg = TwoDouble(x=1.321321, y=6.54352)
+        ret = cfunc(arg)
+        print(arg)
+        print(ret)
+
+        self.assertClose(arg.x * arg.y, ret.x)
+        self.assertClose(arg.x / arg.y, ret.y)
+
+    @skip_if_not_64bits
+    def test_two_halfwords(self):
+        '''Arguments smaller or equal to a word is packed into a word.
+
+        Passing as struct { float, float } occupies two XMM registers instead
+        of one.
+        The output must be in XMM.
+        '''
+        m = lc.Module.new('test_struct_arg')
+
+        float_type = lc.Type.float()
+        struct_type = lc.Type.vector(float_type, 2)
+        func_type = lc.Type.function(struct_type, [struct_type])
+        func = m.add_function(func_type, name='foo')
+
+        # define function body
+        builder = lc.Builder.new(func.append_basic_block(''))
+
+        arg = func.args[0]
+        constint = lambda x: lc.Constant.int(lc.Type.int(), x)
+        e1, e2 = [builder.extract_element(arg, constint(i))
+                  for i in range(2)]
+        se1 = builder.fmul(e1, e2)
+        se2 = builder.fdiv(e1, e2)
+        ret = builder.insert_element(lc.Constant.undef(struct_type), se1,
+                                     constint(0))
+        ret = builder.insert_element(ret, se2, constint(1))
+        builder.ret(ret)
+
+        del builder
+
+        # verify
+        m.verify()
+
+        print(m)
+        # use with ctypes
+        engine = le.EngineBuilder.new(m).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        cfunctype = CFUNCTYPE(TwoFloat, TwoFloat)
+        cfunc = cfunctype(ptr)
+
+        arg = TwoFloat(x=1.321321, y=6.54352)
+        ret = cfunc(arg)
+        print(arg)
+        print(ret)
+
+        self.assertClose(arg.x * arg.y, ret.x)
+        self.assertClose(arg.x / arg.y, ret.y)
+
+    #----------------------------------------------------------------------
+    # 32 bits
+
+    @skip_if_not_32bits
+    def test_structure_abi_32_1(self):
+        '''x86 is simple.  Always pass structure as memory.
+        '''
+        m = lc.Module.new('test_struct_arg')
+
+        double_type = lc.Type.double()
+        uint8_type = lc.Type.int(8)
+        struct_type = lc.Type.struct([double_type, double_type, uint8_type])
+        struct_ptr_type = lc.Type.pointer(struct_type)
+        func_type = lc.Type.function(lc.Type.void(),
+                                     [struct_ptr_type, struct_ptr_type])
+        func = m.add_function(func_type, name='foo')
+
+        # return value pointer
+        func.args[0].add_attribute(lc.ATTR_STRUCT_RET)
+
+        # pass structure by value
+        func.args[1].add_attribute(lc.ATTR_BY_VAL)
+
+        # define function body
+        builder = lc.Builder.new(func.append_basic_block(''))
+
+        arg = builder.load(func.args[1])
+        e1, e2, e3 = [builder.extract_value(arg, i) for i in range(3)]
+        se1 = builder.fmul(e1, e2)
+        se2 = builder.fdiv(e1, e2)
+        ret = builder.insert_value(lc.Constant.undef(struct_type), se1, 0)
+        ret = builder.insert_value(ret, se2, 1)
+        ret = builder.insert_value(ret, e3, 2)
+        builder.store(ret, func.args[0])
+        builder.ret_void()
+
+        del builder
+
+        # verify
+        m.verify()
+
+        print(m)
+        # use with ctypes
+        engine = le.EngineBuilder.new(m).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        cfunctype = CFUNCTYPE(TwoDoubleOneByte, TwoDoubleOneByte)
+        cfunc = cfunctype(ptr)
+
+        arg = TwoDoubleOneByte(x=1.321321, y=6.54352, z=128)
+        ret = cfunc(arg)
+        print(arg)
+        print(ret)
+
+        self.assertClose(arg.x * arg.y, ret.x)
+        self.assertClose(arg.x / arg.y, ret.y)
+        self.assertEqual(arg.z, ret.z)
+
+
+    @skip_if_not_32bits
+    def test_structure_abi_32_2(self):
+        '''x86 is simple.  Always pass structure as memory.
+        '''
+        m = lc.Module.new('test_struct_arg')
+
+        float_type = lc.Type.float()
+        struct_type = lc.Type.struct([float_type, float_type])
+        struct_ptr_type = lc.Type.pointer(struct_type)
+        func_type = lc.Type.function(lc.Type.void(),
+                                     [struct_ptr_type, struct_ptr_type])
+        func = m.add_function(func_type, name='foo')
+
+        # return value pointer
+        func.args[0].add_attribute(lc.ATTR_STRUCT_RET)
+
+        # pass structure by value
+        func.args[1].add_attribute(lc.ATTR_BY_VAL)
+
+        # define function body
+        builder = lc.Builder.new(func.append_basic_block(''))
+
+        arg = builder.load(func.args[1])
+        e1, e2 = [builder.extract_value(arg, i) for i in range(2)]
+        se1 = builder.fmul(e1, e2)
+        se2 = builder.fdiv(e1, e2)
+        ret = builder.insert_value(lc.Constant.undef(struct_type), se1, 0)
+        ret = builder.insert_value(ret, se2, 1)
+        builder.store(ret, func.args[0])
+        builder.ret_void()
+
+        del builder
+
+        # verify
+        m.verify()
+
+        print(m)
+        # use with ctypes
+        engine = le.EngineBuilder.new(m).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        cfunctype = CFUNCTYPE(TwoFloat, TwoFloat)
+        cfunc = cfunctype(ptr)
+
+        arg = TwoFloat(x=1.321321, y=6.54352)
+        ret = cfunc(arg)
+        print(arg)
+        print(ret)
+
+        self.assertClose(arg.x * arg.y, ret.x)
+        self.assertClose(arg.x / arg.y, ret.y)
+
+
+    @skip_if_not_32bits
+    def test_structure_abi_32_3(self):
+        '''x86 is simple.  Always pass structure as memory.
+        '''
+        m = lc.Module.new('test_struct_arg')
+
+        uint8_type = lc.Type.int(8)
+        struct_type = lc.Type.struct([uint8_type])
+        struct_ptr_type = lc.Type.pointer(struct_type)
+        func_type = lc.Type.function(lc.Type.void(),
+                                     [struct_ptr_type, struct_ptr_type])
+        func = m.add_function(func_type, name='foo')
+
+        # return value pointer
+        func.args[0].add_attribute(lc.ATTR_STRUCT_RET)
+
+        # pass structure by value
+        func.args[1].add_attribute(lc.ATTR_BY_VAL)
+
+        # define function body
+        builder = lc.Builder.new(func.append_basic_block(''))
+
+        arg = builder.load(func.args[1])
+        e1 = builder.extract_value(arg, 0)
+        se1 = builder.mul(e1, e1)
+        ret = builder.insert_value(lc.Constant.undef(struct_type), se1, 0)
+        builder.store(ret, func.args[0])
+        builder.ret_void()
+
+        del builder
+
+        # verify
+        m.verify()
+
+        print(m)
+        # use with ctypes
+        engine = le.EngineBuilder.new(m).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        cfunctype = CFUNCTYPE(OneByte, OneByte)
+        cfunc = cfunctype(ptr)
+
+        arg = OneByte(x=8)
+        ret = cfunc(arg)
+        print(arg)
+        print(ret)
+
+        self.assertEqual(arg.x * arg.x, ret.x)
+
+tests.append(TestStructSystemVABI)
+
+@skip_if_not_intel_cpu
+@skip_if_not_win32
+class TestStructMicrosoftABI(TestCase):
+    '''
+    Microsoft convention
+    '''
+
+    #----------------------------------------------------------------------
+    # 64 bits
+
+    @skip_if_not_64bits
+    def test_bigger_than_two_words_64(self):
+        m = lc.Module.new('test_struct_arg')
+
+        double_type = lc.Type.double()
+        uint8_type = lc.Type.int(8)
+        struct_type = lc.Type.struct([double_type, double_type, uint8_type])
+        struct_ptr_type = lc.Type.pointer(struct_type)
+        func_type = lc.Type.function(lc.Type.void(),
+                                     [struct_ptr_type, struct_ptr_type])
+        func = m.add_function(func_type, name='foo')
+
+        # return value pointer
+        func.args[0].add_attribute(lc.ATTR_STRUCT_RET)
+
+        # pass structure by value
+        func.args[1].add_attribute(lc.ATTR_BY_VAL)
+
+        # define function body
+        builder = lc.Builder.new(func.append_basic_block(''))
+
+        arg = builder.load(func.args[1])
+        e1, e2, e3 = [builder.extract_value(arg, i) for i in range(3)]
+        se1 = builder.fmul(e1, e2)
+        se2 = builder.fdiv(e1, e2)
+        ret = builder.insert_value(lc.Constant.undef(struct_type), se1, 0)
+        ret = builder.insert_value(ret, se2, 1)
+        ret = builder.insert_value(ret, e3, 2)
+        builder.store(ret, func.args[0])
+        builder.ret_void()
+
+        del builder
+
+        # verify
+        m.verify()
+
+        print(m)
+        # use with ctypes
+        engine = le.EngineBuilder.new(m).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        cfunctype = CFUNCTYPE(TwoDoubleOneByte, TwoDoubleOneByte)
+        cfunc = cfunctype(ptr)
+
+        arg = TwoDoubleOneByte(x=1.321321, y=6.54352, z=128)
+        ret = cfunc(arg)
+        print(arg)
+        print(ret)
+
+        self.assertClose(arg.x * arg.y, ret.x)
+        self.assertClose(arg.x / arg.y, ret.y)
+        self.assertEqual(arg.z, ret.z)
+
+    @skip_if_not_64bits
+    def test_just_two_words_64(self):
+        m = lc.Module.new('test_struct_arg')
+
+        double_type = lc.Type.double()
+        struct_type = lc.Type.struct([double_type, double_type])
+        struct_ptr_type = lc.Type.pointer(struct_type)
+        func_type = lc.Type.function(lc.Type.void(),
+                                     [struct_ptr_type, struct_ptr_type])
+        func = m.add_function(func_type, name='foo')
+
+        # return value pointer
+        func.args[0].add_attribute(lc.ATTR_STRUCT_RET)
+
+        # pass structure by value
+        func.args[1].add_attribute(lc.ATTR_BY_VAL)
+
+        # define function body
+        builder = lc.Builder.new(func.append_basic_block(''))
+
+        arg = builder.load(func.args[1])
+        e1, e2 = [builder.extract_value(arg, i) for i in range(2)]
+        se1 = builder.fmul(e1, e2)
+        se2 = builder.fdiv(e1, e2)
+        ret = builder.insert_value(lc.Constant.undef(struct_type), se1, 0)
+        ret = builder.insert_value(ret, se2, 1)
+        builder.store(ret, func.args[0])
+        builder.ret_void()
+
+        del builder
+
+        # verify
+        m.verify()
+
+        print(m)
+        # use with ctypes
+        engine = le.EngineBuilder.new(m).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        cfunctype = CFUNCTYPE(TwoDouble, TwoDouble)
+        cfunc = cfunctype(ptr)
+
+        arg = TwoDouble(x=1.321321, y=6.54352)
+        ret = cfunc(arg)
+        print(arg)
+        print(ret)
+
+        self.assertClose(arg.x * arg.y, ret.x)
+        self.assertClose(arg.x / arg.y, ret.y)
+
+    @skip_if_not_64bits
+    def test_two_halfwords(self):
+        '''Arguments smaller or equal to a word is packed into a word.
+
+        Floats structure are not passed on the XMM.
+        Treat it as a i64.
+        '''
+        m = lc.Module.new('test_struct_arg')
+
+        float_type = lc.Type.float()
+        struct_type = lc.Type.struct([float_type, float_type])
+        abi_type = lc.Type.int(64)
+
+        func_type = lc.Type.function(abi_type, [abi_type])
+        func = m.add_function(func_type, name='foo')
+
+        # define function body
+        builder = lc.Builder.new(func.append_basic_block(''))
+
+        arg = func.args[0]
+
+        struct_ptr = builder.alloca(struct_type)
+        struct_int_ptr = builder.bitcast(struct_ptr, lc.Type.pointer(abi_type))
+        builder.store(arg, struct_int_ptr)
+
+        arg = builder.load(struct_ptr)
+
+        e1, e2 = [builder.extract_value(arg, i) for i in range(2)]
+        se1 = builder.fmul(e1, e2)
+        se2 = builder.fdiv(e1, e2)
+        ret = builder.insert_value(lc.Constant.undef(struct_type), se1, 0)
+        ret = builder.insert_value(ret, se2, 1)
+
+        builder.store(ret, struct_ptr)
+        ret = builder.load(struct_int_ptr)
+
+        builder.ret(ret)
+
+        del builder
+
+        # verify
+        m.verify()
+
+        print(m)
+        # use with ctypes
+        engine = le.EngineBuilder.new(m).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        cfunctype = CFUNCTYPE(TwoFloat, TwoFloat)
+        cfunc = cfunctype(ptr)
+
+        arg = TwoFloat(x=1.321321, y=6.54352)
+        ret = cfunc(arg)
+        print(arg)
+        print(ret)
+
+        self.assertClose(arg.x * arg.y, ret.x)
+        self.assertClose(arg.x / arg.y, ret.y)
+
+    #----------------------------------------------------------------------
+    # 32 bits
+
+    @skip_if_not_32bits
+    def test_one_word_register(self):
+        '''Argument is passed by memory.
+        Return value is passed by register.
+        '''
+        m = lc.Module.new('test_struct_arg')
+
+        uint8_type = lc.Type.int(8)
+        struct_type = lc.Type.struct([uint8_type])
+        struct_ptr_type = lc.Type.pointer(struct_type)
+        func_type = lc.Type.function(struct_type, [struct_ptr_type])
+        func = m.add_function(func_type, name='foo')
+
+        # pass structure by value
+        func.args[0].add_attribute(lc.ATTR_BY_VAL)
+
+        # define function body
+        builder = lc.Builder.new(func.append_basic_block(''))
+
+        arg = builder.load(func.args[0])
+        e1 = builder.extract_value(arg, 0)
+        se1 = builder.mul(e1, e1)
+        ret = builder.insert_value(lc.Constant.undef(struct_type), se1, 0)
+
+        builder.ret(ret)
+
+        del builder
+
+        # verify
+        m.verify()
+
+        print(m)
+        # use with ctypes
+        engine = le.EngineBuilder.new(m).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        cfunctype = CFUNCTYPE(OneByte, OneByte)
+        cfunc = cfunctype(ptr)
+
+        arg = OneByte(x=8)
+        ret = cfunc(arg)
+        print(arg)
+        print(ret)
+
+        self.assertEqual(arg.x * arg.x, ret.x)
+
+
+    @skip_if_not_32bits
+    def test_two_floats(self):
+        '''Argument is passed by register.
+        Return in 2 registers
+        '''
+        m = lc.Module.new('test_struct_arg')
+
+        float_type = lc.Type.float()
+        struct_type = lc.Type.struct([float_type, float_type])
+
+        abi_type = lc.Type.int(64)
+        func_type = lc.Type.function(abi_type, [struct_type])
+        func = m.add_function(func_type, name='foo')
+
+        # define function body
+        builder = lc.Builder.new(func.append_basic_block(''))
+
+        out_ptr = builder.alloca(struct_type)
+
+        arg = func.args[0]
+        e1, e2 = [builder.extract_value(arg, i) for i in range(2)]
+        se1 = builder.fmul(e1, e2)
+        se2 = builder.fdiv(e1, e2)
+        ret = builder.insert_value(lc.Constant.undef(struct_type), se1, 0)
+        ret = builder.insert_value(ret, se2, 1)
+        builder.store(ret, out_ptr)
+
+        out_int_ptr = builder.bitcast(out_ptr, lc.Type.pointer(abi_type))
+
+        builder.ret(builder.load(out_int_ptr))
+
+        del builder
+
+        # verify
+        m.verify()
+
+        print(m)
+        # use with ctypes
+        engine = le.EngineBuilder.new(m).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        cfunctype = CFUNCTYPE(TwoFloat, TwoFloat)
+        cfunc = cfunctype(ptr)
+
+        arg = TwoFloat(x=1.321321, y=6.54352)
+        ret = cfunc(arg)
+        print(arg)
+        print(ret)
+
+        self.assertClose(arg.x * arg.y, ret.x)
+        self.assertClose(arg.x / arg.y, ret.y)
+
+    @skip_if_not_32bits
+    def test_bigger_than_two_words(self):
+        '''Pass in memory.
+        '''
+        m = lc.Module.new('test_struct_arg')
+
+        double_type = lc.Type.double()
+        uint8_type = lc.Type.int(8)
+        struct_type = lc.Type.struct([double_type, double_type, uint8_type])
+        struct_ptr_type = lc.Type.pointer(struct_type)
+        func_type = lc.Type.function(lc.Type.void(),
+                                     [struct_ptr_type, struct_ptr_type])
+        func = m.add_function(func_type, name='foo')
+
+        # return value pointer
+        func.args[0].add_attribute(lc.ATTR_STRUCT_RET)
+
+        # pass structure by value
+        func.args[1].add_attribute(lc.ATTR_BY_VAL)
+
+        # define function body
+        builder = lc.Builder.new(func.append_basic_block(''))
+
+        arg = builder.load(func.args[1])
+        e1, e2, e3 = [builder.extract_value(arg, i) for i in range(3)]
+        se1 = builder.fmul(e1, e2)
+        se2 = builder.fdiv(e1, e2)
+        ret = builder.insert_value(lc.Constant.undef(struct_type), se1, 0)
+        ret = builder.insert_value(ret, se2, 1)
+        ret = builder.insert_value(ret, e3, 2)
+        builder.store(ret, func.args[0])
+        builder.ret_void()
+
+        del builder
+
+        # verify
+        m.verify()
+
+        print(m)
+        # use with ctypes
+        engine = le.EngineBuilder.new(m).create()
+        ptr = engine.get_pointer_to_function(func)
+
+        cfunctype = CFUNCTYPE(TwoDoubleOneByte, TwoDoubleOneByte)
+        cfunc = cfunctype(ptr)
+
+        arg = TwoDoubleOneByte(x=1.321321, y=6.54352, z=128)
+        ret = cfunc(arg)
+        print(arg)
+        print(ret)
+
+        self.assertClose(arg.x * arg.y, ret.x)
+        self.assertClose(arg.x / arg.y, ret.y)
+        self.assertEqual(arg.z, ret.z)
+
+tests.append(TestStructMicrosoftABI)
+
+if __name__ == "__main__":
+    unittest.main()

llvm/tests/test_switch.py

@@ -0,0 +1,36 @@
+import unittest
+from llvm.core import (Module, Type, Builder, Constant)
+
+from .support import TestCase, tests
+
+class TestSwitch(TestCase):
+    def test_arg_attr(self):
+        m = Module.new('oifjda')
+        fnty = Type.function(Type.void(), [Type.int()])
+        func = m.add_function(fnty, 'foo')
+        bb = func.append_basic_block('')
+        bbdef = func.append_basic_block('')
+        bbsw1 = func.append_basic_block('')
+        bbsw2 = func.append_basic_block('')
+        bldr = Builder.new(bb)
+
+        swt = bldr.switch(func.args[0], bbdef, n=2)
+        swt.add_case(Constant.int(Type.int(), 0), bbsw1)
+        swt.add_case(Constant.int(Type.int(), 1), bbsw2)
+
+        bldr.position_at_end(bbsw1)
+        bldr.ret_void()
+
+        bldr.position_at_end(bbsw2)
+        bldr.ret_void()
+
+        bldr.position_at_end(bbdef)
+        bldr.ret_void()
+
+        func.verify()
+
+tests.append(TestSwitch)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_target_machines.py

@@ -0,0 +1,62 @@
+import unittest
+import llvm.core as lc
+import llvm.ee as le
+from llvm.core import Type, Builder
+from .support import TestCase, tests, skip_if
+
+# Check PTX backend
+if le.initialize_target('PTX', noraise=True):
+    PTX_ARCH = 'ptx64'
+elif le.initialize_target('NVPTX', noraise=True):
+    PTX_ARCH = 'nvptx64'
+else:
+    PTX_ARCH = None
+
+class TestTargetMachines(TestCase):
+    '''Exercise target machines
+
+    Require PTX backend
+    '''
+    def test_native(self):
+        m, _ = self._build_module()
+        tm = le.EngineBuilder.new(m).select_target()
+
+        self.assertTrue(tm.target_name)
+        self.assertTrue(tm.target_data)
+        self.assertTrue(tm.target_short_description)
+        self.assertTrue(tm.triple)
+        self.assertIn('foo', tm.emit_assembly(m))
+        self.assertTrue(le.get_host_cpu_name())
+
+    @skip_if(not PTX_ARCH, msg='LLVM is not compiled with PTX enabled')
+    def test_ptx(self):
+        arch = PTX_ARCH
+        print(arch)
+        m, func = self._build_module()
+        func.calling_convention = lc.CC_PTX_KERNEL # set calling conv
+        ptxtm = le.TargetMachine.lookup(arch=arch, cpu='sm_20')
+        self.assertTrue(ptxtm.triple)
+        self.assertTrue(ptxtm.cpu)
+        ptxasm = ptxtm.emit_assembly(m)
+        self.assertIn('foo', ptxasm)
+        if arch == 'nvptx64':
+            self.assertIn('.address_size 64', ptxasm)
+        self.assertIn('sm_20', ptxasm)
+
+    def _build_module(self):
+        m = lc.Module.new('TestTargetMachines')
+
+        fnty = Type.function(Type.void(), [])
+        func = m.add_function(fnty, name='foo')
+
+        bldr = Builder.new(func.append_basic_block('entry'))
+        bldr.ret_void()
+        m.verify()
+        return m, func
+
+
+tests.append(TestTargetMachines)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_type_hash.py

@@ -0,0 +1,30 @@
+import unittest
+from llvm.core import Type
+from .support import TestCase, tests
+
+class TestTypeHash(TestCase):
+    def test_scalar_type(self):
+        i32a = Type.int(32)
+        i32b = Type.int(32)
+        i64a = Type.int(64)
+        i64b = Type.int(64)
+        ts = set([i32a, i32b, i64a, i64b])
+        self.assertTrue(len(ts))
+        self.assertTrue(i32a in ts)
+        self.assertTrue(i64b in ts)
+
+    def test_struct_type(self):
+        ta = Type.struct([Type.int(32), Type.float()])
+        tb = Type.struct([Type.int(32), Type.float()])
+        tc = Type.struct([Type.int(32), Type.int(32), Type.float()])
+        ts = set([ta, tb, tc])
+        self.assertTrue(len(ts) == 2)
+        self.assertTrue(ta in ts)
+        self.assertTrue(tb in ts)
+        self.assertTrue(tc in ts)
+
+tests.append(TestTypeHash)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_uses.py

@@ -0,0 +1,43 @@
+import unittest
+from llvm.core import Module, Type, Builder, Constant
+
+from .support import TestCase, tests
+
+class TestUses(TestCase):
+
+    def test_uses(self):
+        m = Module.new('a')
+        t = Type.int()
+        ft = Type.function(t, [t, t, t])
+        f = m.add_function(ft, "func")
+        b = f.append_basic_block('entry')
+        bld = Builder.new(b)
+        tmp1 = bld.add(Constant.int(t, 100), f.args[0], "tmp1")
+        tmp2 = bld.add(tmp1, f.args[1], "tmp2")
+        tmp3 = bld.add(tmp1, f.args[2], "tmp3")
+        bld.ret(tmp3)
+
+        # Testing use count
+        self.assertEqual(f.args[0].use_count, 1)
+        self.assertEqual(f.args[1].use_count, 1)
+        self.assertEqual(f.args[2].use_count, 1)
+        self.assertEqual(tmp1.use_count, 2)
+        self.assertEqual(tmp2.use_count, 0)
+        self.assertEqual(tmp3.use_count, 1)
+
+        # Testing uses
+        self.assert_(f.args[0].uses[0] is tmp1)
+        self.assertEqual(len(f.args[0].uses), 1)
+        self.assert_(f.args[1].uses[0] is tmp2)
+        self.assertEqual(len(f.args[1].uses), 1)
+        self.assert_(f.args[2].uses[0] is tmp3)
+        self.assertEqual(len(f.args[2].uses), 1)
+        self.assertEqual(len(tmp1.uses), 2)
+        self.assertEqual(len(tmp2.uses), 0)
+        self.assertEqual(len(tmp3.uses), 1)
+
+tests.append(TestUses)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/tests/test_volatile.py

@@ -0,0 +1,59 @@
+import unittest
+from llvm.core import Module, Type, Builder
+from .support import TestCase, tests
+
+class TestVolatile(TestCase):
+
+    def test_volatile(self):
+        mod = Module.new('mod')
+        functype = Type.function(Type.void(), [])
+        func = mod.add_function(functype, name='foo')
+        bb = func.append_basic_block('entry')
+        bldr = Builder.new(bb)
+        ptr = bldr.alloca(Type.int())
+
+        # test load inst
+        val = bldr.load(ptr)
+        self.assertFalse(val.is_volatile, "default must be non-volatile")
+        val.set_volatile(True)
+        self.assertTrue(val.is_volatile, "fail to set volatile")
+        val.set_volatile(False)
+        self.assertFalse(val.is_volatile, "fail to unset volatile")
+
+        # test store inst
+        store_inst = bldr.store(val, ptr)
+        self.assertFalse(store_inst.is_volatile, "default must be non-volatile")
+        store_inst.set_volatile(True)
+        self.assertTrue(store_inst.is_volatile, "fail to set volatile")
+        store_inst.set_volatile(False)
+        self.assertFalse(store_inst.is_volatile, "fail to unset volatile")
+
+    def test_volatile_another(self):
+        mod = Module.new('mod')
+        functype = Type.function(Type.void(), [])
+        func = mod.add_function(functype, name='foo')
+        bb = func.append_basic_block('entry')
+        bldr = Builder.new(bb)
+        ptr = bldr.alloca(Type.int())
+
+        # test load inst
+        val = bldr.load(ptr, volatile=True)
+        self.assertTrue(val.is_volatile, "volatile kwarg does not work")
+        val.set_volatile(False)
+        self.assertFalse(val.is_volatile, "fail to unset volatile")
+        val.set_volatile(True)
+        self.assertTrue(val.is_volatile, "fail to set volatile")
+
+        # test store inst
+        store_inst = bldr.store(val, ptr, volatile=True)
+        self.assertTrue(store_inst.is_volatile, "volatile kwarg does not work")
+        store_inst.set_volatile(False)
+        self.assertFalse(store_inst.is_volatile, "fail to unset volatile")
+        store_inst.set_volatile(True)
+        self.assertTrue(store_inst.is_volatile, "fail to set volatile")
+
+tests.append(TestVolatile)
+
+if __name__ == '__main__':
+    unittest.main()
+

llvm/utils/check_intrinsics.py

@@ -0,0 +1,181 @@
+from __future__ import print_function, absolute_import
+import sys
+from llvm.core import Type, Function, Builder, Module
+import llvm.core as lc
+import llvm.ee as le
+import multiprocessing
+from ctypes import *
+
+INTRINSICS = {}
+
+CTYPES_MAP = {
+    Type.int(): c_int32,
+    Type.int(64): c_int64,
+    Type.float(): c_float,
+    Type.double(): c_double,
+}
+
+
+def register(name, retty, *args):
+    def wrap(fn):
+        INTRINSICS[name] = (retty, args), fn
+        return fn
+    return wrap
+
+
+def intr_impl(intrcode, *types):
+    def impl(module, builder, args):
+        intr = Function.intrinsic(module, intrcode, types)
+        r = builder.call(intr, args)
+        return r
+    return impl
+
+
+register("llvm.powi.f64", Type.double(), Type.double(), Type.int())\
+        (intr_impl(lc.INTR_POWI, Type.double()))
+
+register("llvm.powi.f32", Type.float(), Type.float(), Type.int())\
+        (intr_impl(lc.INTR_POWI, Type.float()))
+
+register("llvm.pow.f64", Type.double(), Type.double(), Type.double())\
+        (intr_impl(lc.INTR_POW, Type.double()))
+
+register("llvm.pow.f32", Type.float(), Type.float(), Type.float())\
+        (intr_impl(lc.INTR_POW, Type.float()))
+
+register("llvm.sin.f64", Type.double(), Type.double())\
+        (intr_impl(lc.INTR_SIN, Type.double()))
+
+register("llvm.sin.f32", Type.float(), Type.float())\
+        (intr_impl(lc.INTR_SIN, Type.float()))
+
+register("llvm.cos.f64", Type.double(), Type.double())\
+        (intr_impl(lc.INTR_COS, Type.double()))
+
+register("llvm.cos.f32", Type.float(), Type.float())\
+        (intr_impl(lc.INTR_COS, Type.float()))
+
+register("llvm.log.f64", Type.double(), Type.double())\
+        (intr_impl(lc.INTR_LOG, Type.double()))
+
+register("llvm.log.f32", Type.float(), Type.float())\
+        (intr_impl(lc.INTR_LOG, Type.float()))
+
+register("llvm.log2.f64", Type.double(), Type.double())\
+        (intr_impl(lc.INTR_LOG2, Type.double()))
+
+register("llvm.log2.f32", Type.float(), Type.float())\
+        (intr_impl(lc.INTR_LOG2, Type.float()))
+
+register("llvm.log10.f64", Type.double(), Type.double())\
+        (intr_impl(lc.INTR_LOG10, Type.double()))
+
+register("llvm.log10.f32", Type.float(), Type.float())\
+        (intr_impl(lc.INTR_LOG10, Type.float()))
+
+register("llvm.sqrt.f64", Type.double(), Type.double())\
+        (intr_impl(lc.INTR_SQRT, Type.double()))
+
+register("llvm.sqrt.f32", Type.float(), Type.float())\
+        (intr_impl(lc.INTR_SQRT, Type.float()))
+
+register("llvm.exp.f64", Type.double(), Type.double())\
+        (intr_impl(lc.INTR_EXP, Type.double()))
+
+register("llvm.exp.f32", Type.float(), Type.float())\
+        (intr_impl(lc.INTR_EXP, Type.float()))
+
+register("llvm.exp2.f64", Type.double(), Type.double())\
+        (intr_impl(lc.INTR_EXP2, Type.double()))
+
+register("llvm.exp2.f32", Type.float(), Type.float())\
+        (intr_impl(lc.INTR_EXP2, Type.float()))
+
+register("llvm.fabs.f64", Type.double(), Type.double())\
+        (intr_impl(lc.INTR_FABS, Type.double()))
+
+register("llvm.fabs.f32", Type.float(), Type.float())\
+        (intr_impl(lc.INTR_FABS, Type.float()))
+
+register("llvm.floor.f64", Type.double(), Type.double())\
+        (intr_impl(lc.INTR_FLOOR, Type.double()))
+
+register("llvm.floor.f32", Type.float(), Type.float())\
+        (intr_impl(lc.INTR_FLOOR, Type.float()))
+
+
+def build_test(name):
+    (retty, args), impl = INTRINSICS[name]
+    module = Module.new("test.%s" % name)
+    fn = module.add_function(Type.function(retty, args), name="test_%s" % name)
+    builder = Builder.new(fn.append_basic_block(""))
+    retval = impl(module, builder, fn.args)
+    builder.ret(retval)
+    fn.verify()
+    module.verify()
+    return module, fn
+
+
+def run_test(name):
+    module, fn = build_test(name)
+    eb = le.EngineBuilder.new(module).mcjit(True)
+    engine = eb.create()
+    ptr = engine.get_pointer_to_function(fn)
+
+    argtys = fn.type.pointee.args
+    retty = fn.type.pointee.return_type
+    cargtys = [CTYPES_MAP[a] for a in argtys]
+    cretty = CTYPES_MAP[retty]
+    cfunc = CFUNCTYPE(cretty, *cargtys)(ptr)
+    args = [1] * len(cargtys)
+    cfunc(*args)
+
+
+def spawner(name):
+    print("Testing %s" % name)
+    proc = multiprocessing.Process(target=run_test, args=(name,))
+
+    print('-' * 80)
+    proc.start()
+    proc.join()
+
+    if proc.exitcode != 0:
+        print("FAILED")
+        ok = False
+    else:
+        print("PASSED")
+        ok = True
+    print('=' * 80)
+    print()
+
+    return ok
+
+USAGE = """
+Args: [name]
+
+name: intrinsic name to test
+
+If no name is given, test all intrinsics.
+
+"""
+
+
+def main(argv=()):
+    if len(argv) == 1:
+        intrname = argv[1]
+        spawner(intrname)
+    elif not argv:
+        failed = []
+        for name in sorted(INTRINSICS):
+            if not spawner(name):
+                failed.append(name)
+
+        print("Summary:")
+        for name in failed:
+            print("%s failed" % name)
+    else:
+        print(USAGE)
+
+
+if __name__ == '__main__':
+    main(argv=sys.argv[1:])

llvm/workaround/avx_support.py

@@ -12,7 +12,11 @@ http://software.intel.com/sites/default/files/m/a/b/3/4/d/41604-319433-012a.pdf
 
 """
 
-import sys, os, subprocess
+import sys
+import os
+import subprocess
+import contextlib
+
 
 def detect_avx_support(option='detect'):
     '''Detect AVX support'''
@@ -36,13 +40,25 @@ def detect_unix_like():
     except IOError:
         return False
 
-    for line in info:
+    with contextlib.closing(info):
-        if line.lstrip().startswith('flags'):
+        for line in info:
-            features = line.split()
+            if line.lstrip().startswith('flags'):
-            if 'avx' in features and 'xsave' in features:
+                features = line.split()
-                # enable AVX if flags contain AVX
+                if 'avx' in features and 'xsave' in features:
-                return True
+                    # enable AVX if flags contain AVX
-    return False
+                    return True
+        return False
+
+
+@contextlib.contextmanager
+def _close_popen(popen):
+    if sys.version_info[0] >= 3:
+        with popen:
+            yield
+    else:
+        yield
+        popen.stdout.close()
+
 
 def detect_osx_like():
     try:
@@ -51,9 +67,10 @@ def detect_osx_like():
     except OSError:
         return False
 
-    features = info.stdout.read()
+    with _close_popen(info):
-    features = features.split()
+        features = info.stdout.read().decode('UTF8')
-    return 'AVX1.0' in features and 'OSXSAVE' in features and 'XSAVE' in features
+        features = features.split()
+        return 'AVX1.0' in features and 'OSXSAVE' in features and 'XSAVE' in features
 
 
 if __name__ == '__main__':

llvm_array/__init__.py

@@ -0,0 +1,2 @@
+from __future__ import absolute_import
+from .array import *

llvm_array/array.py

@@ -0,0 +1,257 @@
+# This should be moved to llvmpy
+#
+# There are different array kinds parameterized by eltype and nd
+#
+# Contiguous or Fortran
+# struct {
+#    eltype *data;
+#    intp shape[nd];
+# } contiguous_array(eltype, nd)
+#
+# struct {
+#    eltype *data;
+#    diminfo shape[nd];
+# } strided_array(eltype, nd)
+#
+# struct {
+#    eltype *data;
+#    intp shape[nd];
+#    intp stride[nd];
+# } strided_soa_array(eltype, nd)
+#
+# struct {
+#   intp dim;
+#   intp stride;
+#} diminfo
+#
+# These are for low-level array-routines that need to know the number
+# of dimensions at run-time (not just code-generation time):
+#
+# The first two are recommended
+#
+# struct {
+#   eltype *data;
+#   int32 nd;
+#   intp shape[nd];
+# } contiguous_array_nd(eltype)
+#
+# struct {
+#    eltype *data;
+#    int32 nd;
+#    diminfo shape[nd];
+# } strided_array_nd(eltype)
+#
+#
+# Backward compatible but deprecated:
+# struct {
+#    eltype *data;
+#    int32 nd;
+#    intp shape[nd];
+#    intp stride[nd];
+# } strided_soa_array_nd(eltype)
+#
+#
+# The most general (where the kind of array is stored as well as number
+#                   of dimensions)
+# Rarely needed.
+#
+# struct {
+#   eltype *data;
+#   int16 nd;
+#   int16 dimkind;
+#   ???
+# } array_nd(eltype)
+#
+# where ??? is run-time interpreted based on the dimkind to either:
+# intp shape[nd];  for dimkind = C_CONTIGUOUS or F_CONTIGUOUS
+#
+# diminfo shape[nd]; for dimkind = STRIDED
+#
+# intp shape[ind];
+# intp strides[ind]; dimkind = STRIDED_SOA
+#
+
+import llvm.core as lc
+from llvm.core import Type
+import llvm_cbuilder.shortnames as C
+
+# Different Array Types
+ARRAYBIT = 1<<4
+C_CONTIGUOUS = ARRAYBIT + 0
+F_CONTIGUOUS = ARRAYBIT + 1
+STRIDED = ARRAYBIT + 2
+STRIDED_SOA = ARRAYBIT + 3
+
+HAS_ND = 1<<5
+C_CONTIGUOUS_ND = C_CONTIGUOUS + HAS_ND
+F_CONTIGUOUS_ND = F_CONTIGUOUS + HAS_ND
+STRIDED_ND = STRIDED + HAS_ND
+STRIDED_SOA_ND = STRIDED_SOA + HAS_ND
+
+HAS_DIMKIND = 1<<6
+C_CONTIGUOUS_DK = C_CONTIGUOUS + HAS_DIMKIND
+F_CONTIGUOUS_DK = F_CONTIGUOUS + HAS_DIMKIND
+STRIDED_DK = STRIDED + HAS_DIMKIND
+STRIDED_SOA_DK = STRIDED_SOA + HAS_DIMKIND
+
+array_kinds = (C_CONTIGUOUS, F_CONTIGUOUS, STRIDED, STRIDED_SOA,
+               C_CONTIGUOUS_ND, F_CONTIGUOUS_ND, STRIDED_ND, STRIDED_SOA_DK,
+               C_CONTIGUOUS_DK, F_CONTIGUOUS_DK, STRIDED_DK, STRIDED_SOA_DK)
+
+_invmap = {}
+
+def kind_to_str(kind):
+    global _invmap
+    if not _invmap:
+        for key, value in globals().items():
+            if isinstance(value, int) and value in array_kinds:
+                _invmap[value] = key
+    return _invmap[kind]
+
+def str_to_kind(str):
+    trial = eval(str)
+    if trial not in array_kinds:
+        raise ValueError("Invalid Array Kind")
+    return trial
+
+void_type = C.void
+int32_type = C.int32
+char_type = C.char
+int16_type = C.int16
+intp_type = C.intp
+
+diminfo_type = Type.struct([intp_type,    # shape
+                            intp_type     # stride
+                            ], name='diminfo')
+
+_cache = {}
+# This is the way we define LLVM arrays.
+#  CONTIGUOUS and STRIDED are strongly encouraged...
+def array_type(nd, kind, el_type=char_type):
+    key = (kind, nd, el_type)
+    if _cache.has_key(key):
+        return _cache[key]
+
+    base = kind & (~(HAS_ND | HAS_DIMKIND))
+    if base == C_CONTIGUOUS:
+        dimstr = 'Array_C'
+    elif base == F_CONTIGUOUS:
+        dimstr = 'Array_F'
+    elif base == STRIDED:
+        dimstr = 'Array_S'
+    elif base == STRIDED_SOA:
+        dimstr = 'Array_A'
+    else:
+        raise TypeError("Do not understand Array kind of %d" % kind)
+
+    terms = [Type.pointer(el_type)]        # data
+
+    if (kind & HAS_ND):
+        terms.append(int32_type)           # nd
+        dimstr += '_ND'
+    elif (kind & HAS_DIMKIND):
+        terms.extend([int16_type, int16_type]) # nd, dimkind
+        dimstr += '_DK'
+
+    if base in [C_CONTIGUOUS, F_CONTIGUOUS]:
+        terms.append(Type.array(intp_type, nd))     # shape
+    elif base == STRIDED:
+        terms.append(Type.array(diminfo_type, nd))       # diminfo
+    elif base == STRIDED_SOA:
+        terms.extend([Type.array(intp_type, nd),    # shape
+                      Type.array(intp_type, nd)])   # strides
+
+    ret = Type.struct(terms, name=dimstr)
+    _cache[key] = ret
+    return ret
+
+
+def check_array(arrtyp):
+    if not isinstance(arrtyp, lc.StructType):
+        return None
+    if arrtyp.element_count not in [2, 3, 4, 5]:
+        return None
+
+    # Look through _cache and see if it's there
+    for key, value in _cache.items():
+        if arrtyp is value:
+            return key
+
+    return _raw_check_array(arrtyp)
+
+# Manual check
+def _raw_check_array(arrtyp):
+    a0 = arrtyp.elements[0]
+    a1 = arrtyp.elements[1]
+    if not isinstance(a0, lc.PointerType) or \
+          not (isinstance(a1, lc.ArrayType) or
+               (a1 == int32_type) or (a1 == int16_type)):
+        return None
+
+    data_type = a0.pointee
+
+    if arrtyp.is_literal:
+        c_contig = True
+    else:
+        if arrtyp.name.startswith('Array_F'):
+            c_contig = False
+        else:
+            c_contig = True
+
+
+    if a1 == int32_type:
+        num = 2
+        strided = STRIDED_ND
+        strided_soa = STRIDED_SOA_ND
+        c_contiguous = C_CONTIGUOUS_ND
+        f_contiguous = F_CONTIGUOUS_ND
+    elif a1 == int16_type:
+        if arrtyp.element_count < 3 or arrtyp.elements[2] != int16_type:
+            return None
+        num = 3
+        strided = STRIDED_DK
+        strided_soa = STRIDED_SOA_DK
+        c_contiguous = C_CONTIGUOUS_DK
+        f_contiguous = F_CONTIGUOUS_DK
+    else:
+        num = 1
+        strided = STRIDED
+        strided_soa = STRIDED_SOA
+        c_contiguous = C_CONTIGUOUS
+        f_contiguous = F_CONTIGUOUS
+
+    elcount = num + 2
+    # otherwise we have lc.ArrType as element [1]
+    if arrtyp.element_count not in [num+1,num+2]:
+        return None
+    s1 = arrtyp.elements[num]
+    nd = s1.count
+
+    if arrtyp.element_count == elcount:
+        if not isinstance(arrtyp.elements[num+1], lc.ArrayType):
+            return None
+        s2 = arrtyp.elements[num+1]
+        if s1.element != intp_type or s2.element != intp_type:
+            return None
+        if s1.count != s2.count:
+            return None
+        return strided_soa, nd, data_type
+
+    if s1.element == diminfo_type:
+        return strided, nd, data_type
+    elif s1.element == intp_type:
+        return c_contiguous if c_contig else f_contiguous, nd, data_type
+    else:
+        return None
+
+
+def test():
+    arr = array_type(5, C_CONTIGUOUS)
+    assert check_array(arr) == (C_CONTIGUOUS, 5, char_type)
+    arr = array_type(4, STRIDED)
+    assert check_array(arr) == (STRIDED, 4, char_type)
+    arr = array_type(3, STRIDED_SOA)
+    assert check_array(arr) == (STRIDED_SOA, 3, char_type)
+
+if __name__ == '__main__':
+    test()

llvm_cbuilder/builder.py

@@ -349,7 +349,7 @@ class CBuilder(object):
             elif not isinstance(count, lc.Value):
                 count = self.constant(types.int, count).value
 
-            ptr = self.builder.alloca_array(ty, count, name=name)
+            ptr = self.builder.alloca(ty, size=count, name=name)
             return CArray(self, ptr)
 
     def ret(self, val=None):

llvmpy/README

@@ -1,5 +0,0 @@
-# README
-
-This is a reimplementation of the LLVM binding, aiming to provide a more familiar interface to the C++ API whenever possible.
-
-The implementation uses a custom DSL in python to describe the interface (under binding directory). The DSL serves as input to *gen.py* for generation of the .cpp and .py files for the actual binding.

llvmpy/README.md

@@ -0,0 +1,151 @@
+# README
+
+This is a reimplementation of the LLVM binding, aiming to provide a more 
+familiar interface to the C++ API whenever possible.
+
+The implementation uses a custom DSL in python to describe the interface. 
+The DSL serves as input to *gen/gen.py* for generation of the .cpp and .py files 
+for the actual binding.
+
+# How to Add New Class
+
+Let's use the `llvm::Module` as an example because it should be one of the most
+familiar class in LLVM.
+
+Reference code in https://github.com/llvmpy/llvmpy/blob/master/llvmpy/src/Module.py
+and see LLVM documentation at http://llvm.org/docs/doxygen/html/classllvm_1_1Module.html
+
+1) Import binding helpers
+
+```python
+from binding import *
+# Yes, it is bad practice to import star.
+# I will fix it one day.
+```
+
+2) Import LLVM namespace
+
+```python
+from .namespace import llvm
+```
+
+3) Declare the class
+
+```python
+Module = llvm.Class()
+```
+
+4) Import all the dependencies for the definition
+
+5) Define the class
+
+```python
+@Module
+class Module:
+    ...
+```
+
+## Inside the definition...
+
+5.1) Use the ``_include_`` attribute to add include files.
+
+5.2) Use ``Enum`` to create an enumerator type.
+
+5.3) Make constructor 
+
+Not every class needs to have a binding for the constructor.
+Only add things that will be used.
+
+```python
+new = Constructor(cast(str, StringRef), ref(LLVMContext))
+```
+
+The constructor must be named as "new".
+
+The args to ``Constructor`` are parameters of the signature.
+The first parameter means cast Python string to a StringRef.
+The second parameter means pass LLVMContext object as a value reference.
+
+5.4) Make destructor
+
+Not every class needs to have a binding for the destructor.
+Only add things that will be used.
+If it is always owned by another object, it usually does not need to have one.
+
+```python
+delete = Destructor()
+```
+
+The destructor must be named as "delete".
+
+5.5) Add Simple Methods
+
+```python
+getFunction = Method(ptr(Function), cast(str, StringRef))
+```
+
+The first arg is the return type: a ponter to Function.
+The rest of the args are for the parameters.
+
+Note: ``cast(fromtype, totype)`` can be used as return-type as well.
+In that case, the ``fromtype`` will usually refer to a LLVM object
+and the ``totype`` will refer to the Python object.
+
+5.6) Add custom methods defined in C++
+
+The ``list_functions`` is created as a ``CustomMethod``.
+
+```python
+list_functions = CustomMethod('Module_list_functions', PyObjectPtr)
+```
+
+The first arg is the name that appears in C++.
+The second argument is the return-type.
+The rest of the arguments are parameters.
+
+The definition of ``Module_list_functions`` is located in 
+"include/llvm_binding/extra.h".
+
+5.7) Add custom python method
+
+One can also add custom methods in Python, e.g. ``__str__``.
+
+```python
+@CustomPythonMethod
+def __str__(self):
+    from llvmpy import extra
+    os = extra.make_raw_ostream_for_printing()
+    self.print_(os, None)
+    return os.str()
+```
+
+The body of ``__str__`` is directly copied to the Python output file.
+Thus, it can't reference to anything in current file scope.
+
+# Static Methods
+
+``StaticMethod`` https://github.com/llvmpy/llvmpy/blob/master/llvmpy/src/PassRegistry.py
+
+``CustomStaticMethod`` https://github.com/llvmpy/llvmpy/blob/master/llvmpy/src/Support/TargetRegistry.py
+
+
+# Namespace
+
+``Namespace`` https://github.com/llvmpy/llvmpy/blob/master/llvmpy/src/Support/CodeGen.py
+
+# Functions
+
+``Function`` https://github.com/llvmpy/llvmpy/blob/master/llvmpy/src/Assembly/Parser.py
+
+``CustomFunction`` https://github.com/llvmpy/llvmpy/blob/master/llvmpy/src/Bitcode/ReaderWriter.py
+
+
+
+
+
+# Other Things from the binding.py
+
+https://github.com/llvmpy/llvmpy/blob/master/llvmpy/gen/binding.py
+
+The list of C++ types: https://github.com/llvmpy/llvmpy/blob/master/llvmpy/gen/binding.py#L218
+

llvmpy/capsule.cpp

@@ -1,55 +1,95 @@
 #include <Python.h>
+#include <structmember.h>
 #include <python3adapt.h>
 #include <capsulethunk.h>
 #include <llvm_binding/capsule_context.h>
+#include <llvm_binding/auto_pyobject.h>
+#include <string>
+#include <sstream>
+
+static PyObject* TheAPIModule = NULL;
+static PyObject* TheCapsuleModule = NULL;
+static PyObject* TheCapsuleClass = NULL;
+static PyObject* TheWrapperClass = NULL;
+static PyObject* TheCache = NULL;
+static PyObject* TheAddrDtorDict = NULL;
+static PyObject* TheClassesDict = NULL;
+static PyObject* TheAddrRefCt = NULL;
+static PyObject* ConstantOne = NULL;
+static PyObject* TheDowncastModule = NULL;
 
 static
-PyObject* getName(PyObject* self, PyObject* args) {
+PyObject* GetAPIModule(){
-    PyObject* obj;
+    if (NULL == TheAPIModule)
-    if (!PyArg_ParseTuple(args, "O", &obj)){
+        TheAPIModule = PyImport_ImportModule("llvmpy._api");
-        return NULL;
+    return TheAPIModule;
-    }
-    const char* name = PyCapsule_GetName(obj);
-    if (!name) return NULL;
-
-    return PyString_FromString(name);
 }
 
 static
-PyObject* getPointer(PyObject* self, PyObject* args) {
+PyObject* GetCapsuleModule(){
-    PyObject* obj;
+    if (NULL == TheCapsuleModule)
-    if (!PyArg_ParseTuple(args, "O", &obj)){
+        TheCapsuleModule = PyImport_ImportModule("llvmpy.capsule");
-        return NULL;
+    return TheCapsuleModule;
-    }
-    void* pointer = PyCapsule_GetPointer(obj, PyCapsule_GetName(obj));
-    if (!pointer) return NULL;
-
-    return PyLong_FromVoidPtr(pointer);
 }
 
 static
-PyObject* check(PyObject* self, PyObject* args) {
+PyObject* GetCapsuleClass() {
-    PyObject* obj;
+    if (NULL == TheCapsuleClass)
-    if (!PyArg_ParseTuple(args, "O", &obj)){
+        TheCapsuleClass = PyObject_GetAttrString(GetCapsuleModule(),
-        return NULL;
+                                                 "Capsule");
-    }
+    return TheCapsuleClass;
-    if (PyCapsule_CheckExact(obj)) {
-        Py_RETURN_TRUE;
-    } else {
-        Py_RETURN_FALSE;
-    }
 }
 
-// ------------------
+static
-// PyCapsule Context
+PyObject* GetWrapperClass() {
-// ------------------
+    if (NULL == TheWrapperClass)
+            TheWrapperClass = PyObject_GetAttrString(GetCapsuleModule(),
+                                                     "Wrapper");
+    return TheWrapperClass;
+}
 
 static
-CapsuleContext* getContext(PyObject* self, PyObject* args) {
+PyObject* GetCache() {
-    PyObject* obj;
+    if (NULL == TheCache)
-    if (!PyArg_ParseTuple(args, "O", &obj)) {
+            TheCache = PyObject_GetAttrString(GetCapsuleModule(), "_cache");
-        return NULL;
+    return TheCache;
-    }
+}
+
+static
+PyObject* GetAddrDtorDict() {
+    if (NULL == TheAddrDtorDict)
+            TheAddrDtorDict = PyObject_GetAttrString(GetCapsuleModule(),
+                                                     "_addr2dtor");
+    return TheAddrDtorDict;
+}
+
+static
+PyObject* GetClassesDict() {
+    if (NULL == TheClassesDict)
+        TheClassesDict = PyObject_GetAttrString(GetCapsuleModule(),
+                                                "_pyclasses");
+    return TheClassesDict;
+}
+
+static
+PyObject* GetAddrRefCt() {
+    if (NULL == TheAddrRefCt)
+        TheAddrRefCt = PyObject_GetAttrString(GetCapsuleModule(),
+                                              "_addr2refct");
+    return TheAddrRefCt;
+}
+
+static
+PyObject* GetDowncastModule() {
+    if (NULL == TheDowncastModule)
+        TheDowncastModule = PyObject_GetAttrString(GetAPIModule(),
+                                                   "downcast");
+    return TheDowncastModule;
+
+}
+
+static
+CapsuleContext* GetContext(PyObject *obj) {
     void* context = PyCapsule_GetContext(obj);
     if (!context) {
         PyErr_SetString(PyExc_TypeError, "PyCapsule has no context.");
@@ -59,9 +99,8 @@ CapsuleContext* getContext(PyObject* self, PyObject* args) {
 }
 
 static
-PyObject* getClassName(PyObject* self, PyObject* args) {
+PyObject* GetClassName(PyObject* obj) {
-    CapsuleContext* context = getContext(self, args);
+    CapsuleContext* context = GetContext(obj);
-    //Assert(context->_magic == 0xdead);
     if (!context) {
         return NULL;
     } else {
@@ -69,6 +108,476 @@ PyObject* getClassName(PyObject* self, PyObject* args) {
     }
 }
 
+static
+PyObject* getClassName(PyObject* self, PyObject* args) {
+    PyObject* obj;
+    if (!PyArg_ParseTuple(args, "O", &obj)) {
+        return NULL;
+    }
+    return GetClassName(obj);
+}
+
+static
+PyObject* GetName(PyObject* obj) {
+    const char* name = PyCapsule_GetName(obj);
+    if (!name) return NULL;
+    return PyString_FromString(name);
+}
+
+static
+PyObject* getName(PyObject* self, PyObject* args) {
+    PyObject* obj;
+    if (!PyArg_ParseTuple(args, "O", &obj)){
+        return NULL;
+    }
+    return GetName(obj);
+}
+
+static
+PyObject* GetPointer(PyObject* obj){
+    void *pointer = PyCapsule_GetPointer(obj, PyCapsule_GetName(obj));
+    if (!pointer) return NULL;
+    return PyLong_FromVoidPtr(pointer);
+}
+
+static
+PyObject* getPointer(PyObject* self, PyObject* args) {
+    PyObject* obj;
+    if (!PyArg_ParseTuple(args, "O", &obj)){
+        return NULL;
+    }
+    return GetPointer(obj);
+}
+
+static
+bool Check(PyObject* obj){
+    return PyCapsule_CheckExact(obj);
+}
+
+static
+PyObject* check(PyObject* self, PyObject* args) {
+    PyObject* obj;
+    if (!PyArg_ParseTuple(args, "O", &obj)){
+        return NULL;
+    }
+    if (Check(obj)) {
+        Py_RETURN_TRUE;
+    } else {
+        Py_RETURN_FALSE;
+    }
+}
+
+
+static PyObject* Unwrap(PyObject* obj) {
+    if (PyObject_IsInstance(obj, GetWrapperClass())) {
+        return PyObject_GetAttrString(obj, "_ptr");
+    } else {
+        Py_INCREF(obj);
+        return obj;
+    }
+}
+
+/*
+Unwrap a Wrapper instance into the underlying PyCapsule.
+If `obj` is not a Wrapper instance, returns `obj`.
+*/
+static PyObject* unwrap(PyObject* self, PyObject* args) {
+    PyObject *obj;
+    if (!PyArg_ParseTuple(args, "O", &obj)) {
+        return NULL;
+    }
+    return Unwrap(obj);
+}
+
+static bool HasOwnership(PyObject* obj) {
+    PyObject* addr = GetPointer(obj);
+    PyObject* name = GetName(obj);
+    auto_pyobject nameaddr = PyTuple_Pack(2, name, addr);
+    PyObject* dtor = PyDict_GetItem(GetAddrDtorDict(), *nameaddr);
+    if (!dtor || dtor == Py_None) {
+        return false;
+    } else {
+        return true;
+    }
+}
+
+static PyObject* has_ownership(PyObject* self, PyObject* args) {
+    PyObject *obj;
+    if (!PyArg_ParseTuple(args, "O", &obj)) {
+        return NULL;
+    }
+    if (HasOwnership(obj)) {
+        Py_RETURN_TRUE;
+    } else {
+        Py_RETURN_FALSE;
+    }
+}
+
+static
+void NormalizeString(std::ostream &os, const char* str) {
+    for(; *str; ++str){
+        if (*str == ':') {
+            os << '_';
+            if(*(str + 1) == ':') { ++str; }
+        } else {
+            os << *str;
+        }
+    }
+}
+
+static
+PyObject* WrapCore(PyObject *oldCap, bool owned) {
+    auto_pyobject cap = PyObject_CallFunctionObjArgs(GetCapsuleClass(), oldCap,
+                                                     NULL);
+    auto_pyobject cls = PyObject_CallMethod(*cap, "get_class", "");
+    auto_pyobject addr = GetPointer(oldCap);
+    auto_pyobject name = GetName(oldCap);
+
+    // look up cached object
+    auto_pyobject cache_cls = PyObject_GetItem(GetCache(), *cls);
+    Assert(*cache_cls);
+    int addr_in_cache = PyMapping_HasKey(*cache_cls, *addr);
+
+    PyObject* obj = NULL;
+    if (addr_in_cache) {
+        obj = PyObject_GetItem(*cache_cls, *addr);
+    } else {
+        if (!owned) {
+            auto_pyobject hasDtor = PyObject_CallMethod(*cls, "_has_dtor", "");
+            if (PyObject_IsTrue(*hasDtor)) {
+                auto_pyobject key = PyTuple_Pack(2, *name, *addr);
+                auto_pyobject val = PyObject_GetAttrString(*cls, "_delete_");
+
+                int ok = PyDict_SetItem(GetAddrDtorDict(), *key, *val);
+                Assert(ok != -1);
+            }
+        }
+        obj = PyObject_CallMethod(*cap, "instantiate", "");
+        int ok = PyObject_SetItem(*cache_cls, *addr, obj);
+        Assert(ok != -1);
+    }
+
+    Assert(obj);
+    return obj;
+}
+
+static
+PyObject* Wrap(PyObject* cap, bool owned){
+    if (!Check(cap)) {
+        if (PyList_Check(cap)) {
+            const int N = PyList_Size(cap);
+            PyObject* result = PyList_New(N);
+
+            for (int i = 0; i < N; ++i){
+                PyObject* item = PyList_GetItem(cap, i);
+                if (!item)
+                    return NULL;
+                PyObject* out = Wrap(item, false);
+                if (!out) return NULL;
+                if (-1 == PyList_SetItem(result, i, out))
+                    return NULL;
+            }
+
+            return result;
+        } else {
+            Py_INCREF(cap);
+            return cap;
+        }
+    }
+
+    return WrapCore(cap, owned);
+}
+
+
+static
+PyObject* wrap(PyObject* self, PyObject* args) {
+    PyObject* obj;
+    PyObject* owned = NULL;
+    if (!PyArg_ParseTuple(args, "O|O", &obj, &owned)) {
+        return NULL;
+    }
+    bool ownedFlag = false;
+    if (owned) {
+        ownedFlag = PyObject_IsTrue(owned);
+    }
+    return Wrap(obj, ownedFlag);
+}
+
+static
+PyObject* downcast(PyObject* self, PyObject* args) {
+    PyObject *obj, *cls;
+    if (!PyArg_ParseTuple(args, "OO", &obj, &cls)) {
+        return NULL;
+    }
+
+    auto_pyobject objType = PyObject_Type(obj);
+
+    if (*objType == cls) {
+        Py_INCREF(obj);
+        return obj;
+    }
+
+    PyObject* apiModule = GetAPIModule();
+
+    auto_pyobject fromTy = PyObject_GetAttrString(obj, "_llvm_type_");
+    auto_pyobject toTy = PyObject_GetAttrString(cls, "_llvm_type_");
+
+    std::ostringstream oss;
+
+    auto_pyobject fromTyStr = PyObject_Str(*fromTy);
+    auto_pyobject toTyStr = PyObject_Str(*toTy);
+
+    const char * fromCS = PyString_AsString(*fromTyStr);
+    const char * toCS = PyString_AsString(*toTyStr);
+
+    oss << "downcast_";
+    NormalizeString(oss, fromCS);
+    oss << "_to_";
+    NormalizeString(oss, toCS);
+    std::string fname = oss.str();
+
+    auto_pyobject caster = PyObject_GetAttrString(GetDowncastModule(),
+                                                  fname.c_str());
+
+    if (!caster) {
+        std::ostringstream oss;
+        oss << "Downcast from " << fromCS << " to " << toCS;
+        std::string errmsg = oss.str();
+        PyErr_SetString(PyExc_TypeError, errmsg.c_str());
+        return NULL;
+    }
+
+    auto_pyobject oldObj = Unwrap(obj);
+    auto_pyobject newObj = PyObject_CallFunctionObjArgs(*caster, *oldObj,
+                                                        NULL);
+
+    bool used_to_own = HasOwnership(*oldObj);
+
+    PyObject *result = Wrap(*newObj, !used_to_own);
+
+    int status = PyObject_Not(result);
+    switch(status) {
+    case 0:
+        return result;
+    case 1:
+    default:
+        PyErr_SetString(PyExc_ValueError, "Downcast failed");
+        Py_XDECREF(result);
+        return NULL;
+    }
+}
+
+///////////////////////////////////////////////////////////////////////////////
+
+struct CapsuleObject {
+    PyObject_HEAD;
+    PyObject *capsule;
+};
+
+static
+void Capsule_dealloc(CapsuleObject* self) {
+    Py_XDECREF(self->capsule);
+    Py_TYPE(self)->tp_free((PyObject*)self);
+}
+
+static
+int Capsule_init(CapsuleObject *self, PyObject *args, PyObject *kwds)
+{
+    PyObject *cap;
+    if (!PyArg_ParseTuple(args, "O", &cap)) {
+        return -1;
+    }
+
+    if (!Check(cap)) {
+        PyErr_SetString(PyExc_TypeError, "Expected PyCapsule object");
+        return -1;
+    }
+
+    Py_INCREF(cap);
+    self->capsule = cap;
+
+    PyObject* addr2refct = GetAddrRefCt();
+
+    auto_pyobject ptr = GetPointer(self->capsule);
+    auto_pyobject refct = PyObject_GetItem(addr2refct, *ptr);
+    auto_pyobject inc = PyNumber_InPlaceAdd(*refct, ConstantOne);
+    return PyObject_SetItem(addr2refct, *ptr, *inc);
+}
+
+static
+PyObject* Capsule_getclassname(CapsuleObject* self, void *closure) {
+    return GetClassName(self->capsule);
+}
+
+static
+PyObject* Capsule_getname(CapsuleObject* self, void *closure) {
+    return GetName(self->capsule);
+}
+
+static
+PyObject* Capsule_getpointer(CapsuleObject* self, void *closure) {
+    return GetPointer(self->capsule);
+}
+
+
+static
+PyObject* Capsule_GetClass(CapsuleObject *self){
+    PyObject *pycls = GetClassesDict();
+    auto_pyobject key = GetClassName(self->capsule);
+    return PyDict_GetItem(pycls, *key);    // borrowed reference
+}
+
+static
+PyObject* Capsule_get_class(CapsuleObject* self, PyObject* args) {
+    PyObject* out = Capsule_GetClass(self);
+    Py_XINCREF(out);
+    return out;
+}
+
+static
+PyObject* Capsule_instantiate(CapsuleObject* self, PyObject* args) {
+    return PyObject_CallFunctionObjArgs(Capsule_GetClass(self), self, NULL);
+}
+
+/// Rotate Right
+static unsigned long RotR(unsigned long hash, int offset){
+    if (offset == 0) return hash;
+
+    unsigned long out = hash << (sizeof(hash) * 8 - offset);
+    out |= hash >> offset;
+    return out;
+}
+
+
+/*
+This is called everytime an object is returned from LLVM.
+It derserves to be optimized to reduce unnecessary Python object allocation.
+The following implements a simple hash function that uses XOR and
+right-rotation.
+*/
+static
+long Capsule_hash(CapsuleObject *self) {
+    const char* name = PyCapsule_GetName(self->capsule);
+    void *pointer = PyCapsule_GetPointer(self->capsule, name);
+
+    unsigned long hash = 0xabcd1234 ^ (unsigned long)pointer;
+
+    // The first loop accounts for the different LLVM class name and the
+    // length of the name.
+    for(const char* p = name; *p != '\0'; ++p) {
+        hash ^= *p;
+        hash = RotR(hash, 11);
+    }
+
+    // The second loop accounts for the pointer identity.
+    for(int i = 0; i <sizeof(pointer); ++i) {
+        hash ^= ((unsigned char*)&pointer)[i];
+        hash = RotR(hash, 11);
+    }
+
+    return hash;
+}
+
+static
+bool Capsule_eq(PyObject *self, PyObject *other) {
+    if (PyObject_Type(self) == PyObject_Type(other)) {
+        CapsuleObject *a = (CapsuleObject*)self;
+        CapsuleObject *b = (CapsuleObject*)other;
+
+        void* pa = PyCapsule_GetPointer(a->capsule,
+                                        PyCapsule_GetName(a->capsule));
+        void* pb = PyCapsule_GetPointer(b->capsule,
+                                        PyCapsule_GetName(b->capsule));
+        return pa == pb;
+    }
+    return false;
+}
+
+static
+PyObject* Capsule_richcmp(PyObject *a, PyObject *b, int op) {
+    bool ret = Capsule_eq(a, b);
+    switch(op) {
+    case Py_EQ:             break;
+    case Py_NE: ret = !ret; break;
+    default:
+        return Py_NotImplemented;
+    }
+    if (ret) {
+        Py_RETURN_TRUE;
+    } else {
+        Py_RETURN_FALSE;
+    }
+}
+
+static PyMemberDef Capsule_members[] = {
+    {"capsule", T_OBJECT_EX, offsetof(CapsuleObject, capsule), READONLY, "capsule"},
+    { NULL },
+};
+
+static PyMethodDef Capsule_methods[] = {
+    { "get_class", (PyCFunction)Capsule_get_class, METH_NOARGS,
+      "Get Capsule class"},
+    { "instantiate", (PyCFunction)Capsule_instantiate, METH_NOARGS,
+      "create a new instance"},
+    { NULL },
+};
+
+static PyGetSetDef Capsule_getseters[] = {
+    {"classname", (getter)Capsule_getclassname, NULL, "class name", NULL},
+    {"name", (getter)Capsule_getname, NULL, "name", NULL},
+    {"pointer", (getter)Capsule_getpointer, NULL, "pointer", NULL},
+    { NULL },
+};
+
+static PyTypeObject CapsuleType = {
+#if (PY_MAJOR_VERSION < 3)
+    PyObject_HEAD_INIT(NULL)
+    0,                         /*ob_size*/
+#else
+    PyVarObject_HEAD_INIT(NULL, 0)
+#endif
+    "_capsule.Capsule",        /*tp_name*/
+    sizeof(CapsuleObject),     /*tp_basicsize*/
+    0,                         /*tp_itemsize*/
+    (destructor)Capsule_dealloc,                         /*tp_dealloc*/
+    0,                         /*tp_print*/
+    0,                         /*tp_getattr*/
+    0,                         /*tp_setattr*/
+    0,                         /*tp_compare*/
+    0,                         /*tp_repr*/
+    0,                         /*tp_as_number*/
+    0,                         /*tp_as_sequence*/
+    0,                         /*tp_as_mapping*/
+    (hashfunc)Capsule_hash,    /*tp_hash */
+    0,                         /*tp_call*/
+    0,                         /*tp_str*/
+    0,                         /*tp_getattro*/
+    0,                         /*tp_setattro*/
+    0,                         /*tp_as_buffer*/
+    Py_TPFLAGS_DEFAULT,        /*tp_flags*/
+    "Capsule object",          /* tp_doc */
+    0,		                   /* tp_traverse */
+    0,		                   /* tp_clear */
+    (richcmpfunc)Capsule_richcmp,   /*tp_richcompare */
+    0,		                   /* tp_weaklistoffset */
+    0,		                   /* tp_iter */
+    0,		                   /* tp_iternext */
+    Capsule_methods,           /* tp_methods */
+    Capsule_members,           /* tp_members */
+    Capsule_getseters,                         /* tp_getset */
+    0,                         /* tp_base */
+    0,                         /* tp_dict */
+    0,                         /* tp_descr_get */
+    0,                         /* tp_descr_set */
+    0,                         /* tp_dictoffset */
+    (initproc)Capsule_init,              /* tp_init */
+    0,                         /* tp_alloc */
+    0,                         /* tp_new */
+};
+
+
+///////////////////////////////////////////////////////////////////////////////
+
 
 static PyMethodDef core_methods[] = {
 #define declmethod(func) { #func , ( PyCFunction )func , METH_VARARGS , NULL }
@@ -76,6 +585,10 @@ static PyMethodDef core_methods[] = {
     declmethod(getPointer),
     declmethod(check),
     declmethod(getClassName),
+    declmethod(unwrap),
+    declmethod(wrap),
+    declmethod(has_ownership),
+    declmethod(downcast),
     { NULL },
 #undef declmethod
 };
@@ -106,10 +619,21 @@ extern "C" {
 #else
         PyObject *module = Py_InitModule("_capsule", core_methods);
 #endif
-        if (module == NULL)
+        if (module == NULL){
             INITERROR;
+        }
+
+        if (module) {
+            CapsuleType.tp_new = PyType_GenericNew;
+            if (PyType_Ready(&CapsuleType) < 0) {
+                INITERROR;
+            }
+            Py_INCREF(&CapsuleType);
+            PyModule_AddObject(module, "Capsule", (PyObject*)(&CapsuleType));
+
+            ConstantOne = PyInt_FromLong(1);
+        }
 #if PY_MAJOR_VERSION >= 3
-        
         return module;
 #endif
     }

llvmpy/capsule.py

@@ -1,8 +1,27 @@
-from weakref import WeakKeyDictionary, WeakValueDictionary, ref
+from weakref import WeakValueDictionary
 from collections import defaultdict
 import logging
+from llvmpy._capsule import (unwrap, has_ownership, downcast, wrap,
+                             getClassName, getName, getPointer, Capsule)
+
+
 logger = logging.getLogger(__name__)
 
+NO_DEBUG = False
+
+
+def silent_logger():
+    '''
+    Silent logger for unless we have a error message.
+    '''
+    global NO_DEBUG
+    logger.setLevel(logging.ERROR)
+    NO_DEBUG = True
+
+# comment out the line below to re-enable logging at DEBUG level.
+silent_logger()
+
+
 def set_debug(enabled):
     '''
     Side-effect: configure logger with it is not configured.
@@ -17,75 +36,63 @@ def set_debug(enabled):
     else:
         logger.setLevel(logging.WARNING)
 
-def _capsule_weakref_dtor(item):
-    addr = item.pointer
-    name = item.name
-    _addr2refct[addr] -= 1
-    refct = _addr2refct[addr]
-    assert refct >= 0, "RefCt drop below 0"
-    if refct == 0:
-        dtor = _addr2dtor.pop((name, addr), None)
-        if dtor is not None:
-            logger.debug('Destroy %s %s', name, hex(addr))
-            dtor(item.capsule)
 
-class Capsule(object):
+#class Capsule(object):
-    "Wraps PyCapsule so that we can build weakref of it."
+#    "Wraps PyCapsule so that we can build weakref of it."
+#    from ._capsule import check, getClassName, getName, getPointer
+#    __slots__ = "capsule"
+#
+#    def __init__(self, capsule):
+#        assert Capsule.valid(capsule)
+#        self.capsule = capsule
+#
+#        #weak = WeakRef(self, _capsule_weakref_dtor)
+#        #weak.pointer = self.pointer
+#        #weak.capsule = capsule
+#        #weak.name = self.name
+#        #_capsule2weak[self] = weak
+#        _addr2refct[self.pointer] += 1
+#
+#    @property
+#    def classname(self):
+#        return self.getClassName(self.capsule)
+#
+#    @property
+#    def name(self):
+#        return self.getName(self.capsule)
+#
+#    @property
+#    def pointer(self):
+#        return self.getPointer(self.capsule)
+#
+#    @staticmethod
+#    def valid(capsule):
+#        return Capsule.check(capsule)
+#
+#    def get_class(self):
+#        return _pyclasses[self.classname]
+#
+#    def instantiate(self):
+#        cls = self.get_class()
+#        return cls(self)
+#
+#    def __eq__(self, other):
+#        if isinstance(other, Capsule) and self.pointer == other.pointer:
+#            assert self.name == other.name
+#            return True
+#        else:
+#            return False
+#
+#    def __hash__(self):
+#        return hash((self.pointer, self.name))
+#
+#    def __ne__(self, other):
+#        return not (self == other)
 
-    from ._capsule import check, getClassName, getName, getPointer
 
-    def __init__(self, capsule):
-        assert Capsule.valid(capsule)
-        self.capsule = capsule
-
-        weak = WeakRef(self, _capsule_weakref_dtor)
-        weak.pointer = self.pointer
-        weak.capsule = capsule
-        weak.name = self.name
-        _capsule2weak[self] = weak
-        _addr2refct[self.pointer] += 1
-
-    @property
-    def classname(self):
-        return Capsule.getClassName(self.capsule)
-
-    @property
-    def name(self):
-        return Capsule.getName(self.capsule)
-
-    @property
-    def pointer(self):
-        return Capsule.getPointer(self.capsule)
-
-    @staticmethod
-    def valid(capsule):
-        return Capsule.check(capsule)
-
-    def get_class(self):
-        return _pyclasses[self.classname]
-
-    def instantiate(self):
-        cls = self.get_class()
-        return cls(self)
-
-    def __eq__(self, other):
-        if self.pointer == other.pointer:
-            assert self.name == other.name
-            return True
-        else:
-            return False
-
-    def __hash__(self):
-        return hash((self.pointer, self.name))
-
-    def __ne__(self, other):
-        return not (self == other)
-
-class WeakRef(ref):
-    pass
 
 _addr2refct = defaultdict(lambda: 0)
-_capsule2weak = WeakKeyDictionary()
+#_capsule2weak = WeakKeyDictionary()
 _addr2dtor = {}
 _pyclasses = {}
 
@@ -93,15 +100,16 @@ _pyclasses = {}
 # NOTE: The same 'addr' may appear in multiple class bins.
 _cache = defaultdict(WeakValueDictionary)
 
+
 def release_ownership(old):
     logger.debug('Release %s', old)
-    addr = Capsule.getPointer(old)
+    addr = getPointer(old)
-    name = Capsule.getName(old)
+    name = getName(old)
     if _addr2dtor.get((name, addr)) is None:
-        clsname = Capsule.getClassName(old)
+        clsname = getClassName(old)
         if not _pyclasses[clsname]._has_dtor():
             return
-        # Guard duplicated release
+            # Guard duplicated release
         raise Exception("Already released")
     _addr2dtor[(name, addr)] = None
 
@@ -111,60 +119,86 @@ def obtain_ownership(cap):
     if cls._has_dtor():
         addr = cap.pointer
         name = cap.name
-        assert _addr2dtor[addr] is None
+        assert _addr2dtor[(name, addr)] is None
         _addr2dtor[(name, addr)] = cls._delete_
 
-def has_ownership(cap):
-    addr = Capsule.getPointer(cap)
-    name = Capsule.getName(cap)
-    return _addr2dtor.get((name, addr)) is not None
 
-def wrap(cap, owned=False):
+#def has_ownership(cap):
-    '''Wrap a PyCapsule with the corresponding Wrapper class.
+#    addr = Capsule.getPointer(cap)
-    If `cap` is not a PyCapsule, returns `cap`
+#    name = Capsule.getName(cap)
-    '''
+#    return _addr2dtor.get((name, addr)) is not None
-    if not Capsule.valid(cap):
-        if isinstance(cap, list):
-            return list(map(wrap, cap))
-        return cap     # bypass if cap is not a PyCapsule and not a list
 
-    cap = Capsule(cap)
-    cls = cap.get_class()
-    addr = cap.pointer
-    name = cap.name
-    try: # lookup cached object
-        return _cache[cls][addr]
-    except KeyError:
-        if not owned and cls._has_dtor():
-            _addr2dtor[(name, addr)] = cls._delete_
-        obj = cap.instantiate()
-        _cache[cls][addr] = obj    # cache it
-    return obj
 
-def unwrap(obj):
+#def wrap(cap, owned=False):
+#    '''Wrap a PyCapsule with the corresponding Wrapper class.
+#    If `cap` is not a PyCapsule, returns `cap`
+#    '''
+#    if not Capsule.valid(cap):
+#        if isinstance(cap, list):
+#            return list(map(wrap, cap))
+#        return cap     # bypass if cap is not a PyCapsule and not a list
+#
+#    cap = Capsule(cap)
+#    cls = cap.get_class()
+#    addr = cap.pointer
+#    name = cap.name
+#    # lookup cached object
+#    if cls in _cache and addr in _cache[cls]:
+#        obj = _cache[cls][addr]
+#    else:
+#        if not owned and cls._has_dtor():
+#            _addr2dtor[(name, addr)] = cls._delete_
+#        obj = cap.instantiate()
+#        _cache[cls][addr] = obj    # cache it
+#    return obj
+
+#def unwrap(obj):
     '''Unwrap a Wrapper instance into the underlying PyCapsule.
     If `obj` is not a Wrapper instance, returns `obj`.
     '''
-    if isinstance(obj, Wrapper):
+#    if isinstance(obj, Wrapper):
-        return obj._ptr
+#        return obj._ptr
-    else:
+#    else:
-        return obj
+#        return obj
 
 
 def register_class(clsname):
     def _wrapped(cls):
         _pyclasses[clsname] = cls
         return cls
+
     return _wrapped
 
 
 class Wrapper(object):
-
     __slots__ = '__capsule'
 
     def __init__(self, capsule):
         self.__capsule = capsule
 
+    def __del__(self):
+        if _addr2refct is None:
+            # System is tearing down
+            # No need to free anything
+            return
+
+        item = self.__capsule
+        addr = item.pointer
+        name = item.name
+
+        _addr2refct[addr] -= 1
+        refct = _addr2refct[addr]
+        assert refct >= 0, "RefCt drop below 0"
+        if refct == 0:
+            dtor = _addr2dtor.pop((name, addr), None)
+            if dtor is not None:
+                if not NO_DEBUG:
+                    # Some globals in logger could be removed by python GC
+                    # at interpreter teardown.
+                    # That can cause exception raised and ignored message.
+                    logger.debug('Destroy %s %s', name, hex(addr))
+                dtor(item.capsule)
+
     @property
     def _capsule(self):
         return self.__capsule
@@ -177,10 +211,11 @@ class Wrapper(object):
         return hash(self._capsule)
 
     def __eq__(self, other):
-        return self._capsule == other._capsule
+        if isinstance(other, Wrapper):
+            return self._capsule == other._capsule
 
     def __ne__(self, other):
-        return not(self == other)
+        return not (self == other)
 
     def _downcast(self, newcls):
         return downcast(self, newcls)
@@ -189,25 +224,25 @@ class Wrapper(object):
     def _has_dtor(cls):
         return hasattr(cls, '_delete_')
 
-def downcast(obj, cls):
-    from . import _api
-    if type(obj) is cls:
-        return obj
-    fromty = obj._llvm_type_
-    toty = cls._llvm_type_
-    logger.debug("Downcast %s to %s" , fromty, toty)
-    fname = 'downcast_%s_to_%s' % (fromty, toty)
-    fname = fname.replace('::', '_')
-    try:
-        caster = getattr(_api.downcast, fname)
-    except AttributeError:
-        fmt = "Downcast from %s to %s is not supported"
-        raise TypeError(fmt % (fromty, toty))
-    old = unwrap(obj)
-    new = caster(old)
-    used_to_own = has_ownership(old)
-    res = wrap(new, owned=not used_to_own)
-    if not res:
-        raise ValueError("Downcast failed")
-    return res
 
+#def downcast(obj, cls):
+#    from . import _api
+#
+#    if type(obj) is cls:
+#        return obj
+#    fromty = obj._llvm_type_
+#    toty = cls._llvm_type_
+#    logger.debug("Downcast %s to %s", fromty, toty)
+#    fname = 'downcast_%s_to_%s' % (fromty, toty)
+#    fname = fname.replace('::', '_')
+#    if not hasattr(_api.downcast, fname):
+#        fmt = "Downcast from %s to %s is not supported"
+#        raise TypeError(fmt % (fromty, toty))
+#    caster = getattr(_api.downcast, fname)
+#    old = unwrap(obj)
+#    new = caster(old)
+#    used_to_own = has_ownership(old)
+#    res = wrap(new, owned=not used_to_own)
+#    if not res:
+#        raise ValueError("Downcast failed")
+#    return res

llvmpy/gen/binding.py

@@ -1,5 +1,4 @@
 import inspect, textwrap
-import functools
 import codegen as cg
 import os
 
@@ -8,10 +7,12 @@ namespaces = {}
 
 RESERVED = frozenset(['None'])
 
+
 def makedir(directory):
     if not os.path.exists(directory):
         os.makedirs(directory)
 
+
 class SubModule(object):
     def __init__(self):
         self.methods = []
@@ -247,7 +248,8 @@ class Class(SubModule, _Type):
     def __call__(self, defn):
         assert not self._is_defined
         # process the definition in "defn"
-        self.name = defn.__name__
+        self.name = getattr(defn, '_name_', defn.__name__)
+
         for k, v in defn.__dict__.items():
             if isinstance(v, Method):
                 self.methods.append(v)
@@ -292,6 +294,10 @@ class Class(SubModule, _Type):
         writer.println('@capsule.register_class("%s")' % self.fullname)
         with writer.block('class %(clsname)s(%(bases)s):' % locals()):
             writer.println('_llvm_type_ = "%s"' % self.fullname)
+            if self.bases:
+                writer.println('__slots__ = ()')
+            else:
+                writer.println('__slots__ = "__weakref__"')
             for enum in self.enums:
                 enum.compile_py(writer)
             for meth in self.methods:
@@ -332,7 +338,7 @@ class Class(SubModule, _Type):
         writer.die_if_false(raw, verbose=name)
         ptrty = ptr(self).fullname
         ty = self.fullname
-        fmt = 'typecast< %(ty)s >::from(%(raw)s)'
+        fmt = 'unwrap_as<%(ty)s, %(name)s >::from(%(raw)s)'
         casted = writer.declare(ptrty, fmt % locals())
         writer.die_if_false(casted)
         return casted
@@ -398,6 +404,7 @@ class Enum(object):
                 writer.println(fmt % locals())
         writer.println()
 
+
 class Method(object):
     _kind_ = 'meth'
 
@@ -515,6 +522,7 @@ class Method(object):
                 with writer.block('if len(%s) > %d:' % (unwrapped, i)):
                     writer.release_ownership('%s[%d]' % (unwrapped, i))
 
+
 class CustomMethod(Method):
     def __init__(self, methodname, retty, *argtys):
         super(CustomMethod, self).__init__(retty, *argtys)
@@ -593,6 +601,7 @@ class CustomFunction(Function):
     def fullname(self):
         return self.realname
 
+
 class Destructor(Method):
     _kind_ = 'dtor'
 
@@ -624,6 +633,7 @@ class Constructor(StaticMethod):
         ret = writer.declare(retty.fullname, stmt)
         return ret
 
+
 class ref(_Type):
     def __init__(self, element):
         assert isinstance(element, Class), type(element)
@@ -685,13 +695,16 @@ class ptr(_Type):
         return writer.pycapsule_new(val, self.element.capsule_name,
                                     self.element.fullname)
 
+
 class ownedptr(ptr):
     pass
 
+
 def const(ptr_or_ref):
     ptr_or_ref.const = True
     return ptr_or_ref
 
+
 class cast(_Type):
     format = 'O'
 
@@ -756,6 +769,7 @@ class CustomPythonMethod(object):
         for line in self.sourcelines:
             writer.println(line)
 
+
 class CustomPythonStaticMethod(CustomPythonMethod):
     def compile_py(self, writer):
         writer.println('@staticmethod')
@@ -844,6 +858,7 @@ class Attr(object):
 
 TARGETS_BUILT = os.environ.get('LLVM_TARGETS_BUILT', '').split()
 
+
 def _parse_llvm_version(ver):
     import re
     m = re.compile(r'(\d+)\.(\d+)').match(ver)

llvmpy/gen/codegen.py

@@ -202,7 +202,8 @@ class CppCodeWriter(CodeWriterBase):
 
     def pycapsule_new(self, ptr, name, clsname):
         name_soften = mangle(name)
-        ret = self.call('pycapsule_new', 'PyObject*', ptr, quote(name),
+        cast_to_base = 'cast_to_base<%s >::from(%s)' % (name, ptr)
+        ret = self.call('pycapsule_new', 'PyObject*', cast_to_base, quote(name),
                         quote(clsname))
         with self.block('if (!%(ret)s)' % locals()):
             self.return_null()

llvmpy/gen/gen.py

@@ -9,7 +9,7 @@ extern "C" {
 
 #if (PY_MAJOR_VERSION >= 3)
 
-PyObject *
+PyMODINIT_FUNC
 PyInit_%(module)s(void)
 {
 PyObject *module = create_python_module("%(module)s", meth_%(ns)s);

llvmpy/include/llvm_binding/conversion.h

@@ -256,6 +256,7 @@ template<class Td>
 struct typecast {
     template<class Ts> static
     Td* from(Ts* src) {
+        // check why this is only used in Python3
         return llvm::dyn_cast<Td>(src);
     }
 
@@ -265,3 +266,24 @@ struct typecast {
     }
 };
 
+template<class Td, class Tbase>
+struct unwrap_as {
+    static
+    Td* from(void* src) {
+        Tbase* base = static_cast<Tbase*>(src);
+        return static_cast<Td*>(base);
+    }
+};
+
+template<class Td>
+struct cast_to_base {
+    template<class Ts> static
+    Td* from(Ts* src){
+        return static_cast<Td*>(src);
+    }
+    
+    template<class Ts> static
+    const Td* from(const Ts* src){
+        return static_cast<const Td*>(src);
+    }
+};

llvmpy/include/llvm_binding/extra.h

@@ -1,8 +1,26 @@
 #include <Python.h>
 #include <llvm/ADT/SmallVector.h>
-#include <llvm/Value.h>
+#if LLVM_VERSION_MAJOR >= 3 && LLVM_VERSION_MINOR >= 3
-#include <llvm/DerivedTypes.h>
+    #include <llvm/IR/Value.h>
-#include <llvm/Function.h>
+    #include <llvm/IR/DerivedTypes.h>
+    #include <llvm/IR/Function.h>
+    #include <llvm/IR/Module.h>
+    #include <llvm/IR/Constants.h>
+    #include <llvm/IR/Intrinsics.h>
+    #include <llvm/IR/IRBuilder.h>
+    #if LLVM_VERSION_MINOR >= 4
+        #include <llvm/Support/MemoryObject.h>
+        #include <llvm/MC/MCDisassembler.h>
+    #endif
+#else
+    #include <llvm/Value.h>
+    #include <llvm/DerivedTypes.h>
+    #include <llvm/Function.h>
+    #include <llvm/Module.h>
+    #include <llvm/Constants.h>
+    #include <llvm/Intrinsics.h>
+    #include <llvm/IRBuilder.h>
+#endif
 #include <llvm/Support/raw_ostream.h>
 #include <llvm/Support/FormattedStream.h>
 #include <llvm/Support/MemoryBuffer.h>
@@ -12,20 +30,17 @@
 #include <llvm/ExecutionEngine/ExecutionEngine.h>
 #include <llvm/ExecutionEngine/GenericValue.h>
 #include <llvm/Linker.h>
-#include <llvm/Module.h>
 #include <llvm/Analysis/Verifier.h>
-#include <llvm/Constants.h>
-#include <llvm/Intrinsics.h>
-#include <llvm/IRBuilder.h>
 #include <llvm/PassRegistry.h>
 #include <llvm/Support/Host.h>
 
+#include <llvm/ExecutionEngine/MCJIT.h> // to make MCJIT working
 
 #include "auto_pyobject.h"
 
 namespace extra{
     using namespace llvm;
-    
+
     class raw_svector_ostream_helper: public raw_svector_ostream {
         SmallVectorImpl<char> *SV;
     public:
@@ -56,12 +71,21 @@ namespace extra{
 
 }
 
+static
+PyObject* callwrite(PyObject* self, PyObject* arg)
+{
+    char meth[] = "write";
+    char fmt[] = "O";
+    return PyObject_CallMethod(self, meth, fmt, arg);
+}
+
+
 static
 PyObject* make_raw_ostream_for_printing(PyObject* self, PyObject* args)
 {
     using extra::raw_svector_ostream_helper;
     using llvm::raw_svector_ostream;
-    
+
     if (!PyArg_ParseTuple(args, "")) {
         return NULL;
     }
@@ -75,7 +99,7 @@ PyObject* make_small_vector_from_types(PyObject* self, PyObject* args)
 {
     using llvm::Type;
     typedef llvm::SmallVector<llvm::Type*, 8> SmallVector_Type;
-    
+
     SmallVector_Type* SV = new SmallVector_Type;
     Py_ssize_t size = PyTuple_Size(args);
     for (Py_ssize_t i = 0; i < size; ++i) {
@@ -162,8 +186,8 @@ PyObject* iterator_to_pylist_deref(iterator begin, iterator end,
 {
     PyObject* list = PyList_New(0);
     for(; begin != end; ++begin) {
-        PyObject* cap = pycapsule_new(&*begin, capsuleName, className);
+        auto_pyobject cap = pycapsule_new(&*begin, capsuleName, className);
-        PyList_Append(list, cap);
+        PyList_Append(list, *cap);
     }
     return list;
 }
@@ -174,8 +198,8 @@ PyObject* iterator_to_pylist(iterator begin, iterator end,
 {
     PyObject* list = PyList_New(0);
     for(; begin != end; ++begin) {
-        PyObject* cap = pycapsule_new(*begin, capsuleName, className);
+        auto_pyobject cap = pycapsule_new(*begin, capsuleName, className);
-        PyList_Append(list, cap);
+        PyList_Append(list, *cap);
     }
     return list;
 }
@@ -270,7 +294,12 @@ llvm::ExecutionEngine* ExecutionEngine_create(
     ExecutionEngine *ee = ExecutionEngine::create(M, ForceInterpreter,
                                                   &ErrorStr, OptLevel,
                                                   GVsWithCode);
-    PyFile_WriteString(ErrorStr.c_str(), errout);
+    auto_pyobject buf = PyBytes_FromString(ErrorStr.c_str());
+    if (errout && NULL == callwrite(errout, *buf)){
+        return NULL;
+    }
+
+//    PyFile_WriteString(ErrorStr.c_str(), errout);
     return ee;
 }
 
@@ -292,7 +321,11 @@ llvm::ExecutionEngine* ExecutionEngine_createJIT(
     std::string ErrorStr;
     ExecutionEngine *ee = ExecutionEngine::createJIT(M, &ErrorStr, JMM, OL,
                                                      GCsWithCode, RM, CMM);
-    PyFile_WriteString(ErrorStr.c_str(), errout);
+    auto_pyobject buf = PyBytes_FromString(ErrorStr.c_str());
+    if (errout && NULL == callwrite(errout, *buf)){
+        return NULL;
+    }
+//    PyFile_WriteString(ErrorStr.c_str(), errout);
     return ee;
 }
 
@@ -387,7 +420,7 @@ PyObject* ExecutionEngine_RunFunction(llvm::ExecutionEngine* EE,
             PyErr_SetString(PyExc_RuntimeError, "Failed to index into args?");
             return NULL;
         }
-        
+
         GenericValue* gv = static_cast<GenericValue*>(
                               PyCapsule_GetPointer(obj, GVN));
 
@@ -474,7 +507,7 @@ PyObject* llvm_ParseBitCodeFile(llvm::StringRef Buf, llvm::LLVMContext& Ctx,
         std::string ErrStr;
         M = ParseBitcodeFile(MB, Ctx, &ErrStr);
         auto_pyobject buf = PyBytes_FromString(ErrStr.c_str());
-        if (NULL == PyObject_CallMethod(FObj, "write", "O", *buf)){
+        if (NULL == callwrite(FObj, *buf)){
             return NULL;
         }
 //        if (-1 == PyFile_WriteString(ErrStr.c_str(), FObj)) {
@@ -498,7 +531,7 @@ PyObject* llvm_WriteBitcodeToFile(const llvm::Module *M, PyObject* FObj)
     rso.flush();
     StringRef ref = rso.str();
     auto_pyobject buf = PyBytes_FromStringAndSize(ref.data(), ref.size());
-    return PyObject_CallMethod(FObj, "write", "O", *buf);
+    return callwrite(FObj, *buf);
 }
 
 static
@@ -512,9 +545,13 @@ PyObject* llvm_getBitcodeTargetTriple(llvm::StringRef Buf,
     if (FObj) {
         std::string ErrStr;
         Triple = getBitcodeTargetTriple(MB, Ctx, &ErrStr);
-        if (-1 == PyFile_WriteString(ErrStr.c_str(), FObj)) {
+        auto_pyobject buf = PyBytes_FromString(ErrStr.c_str());
+        if (NULL == callwrite(FObj, *buf)){
             return NULL;
         }
+//        if (-1 == PyFile_WriteString(ErrStr.c_str(), FObj)) {
+//            return NULL;
+//        }
     } else {
         Triple = getBitcodeTargetTriple(MB, Ctx);
     }
@@ -538,11 +575,11 @@ PyObject* TargetMachine_addPassesToEmitFile(
     bool status = TM->addPassesToEmitFile(PM, fso, FTy, disableVerify);
     if (status) {
         StringRef sr = rso.str();
-        PyObject* buf = PyString_FromStringAndSize(sr.data(), sr.size());
+        auto_pyobject buf = PyString_FromStringAndSize(sr.data(), sr.size());
         if (!buf) {
             return NULL;
         }
-        if (-1 == PyFile_WriteObject(buf, Out, Py_PRINT_RAW)){
+        if (-1 == PyFile_WriteObject(*buf, Out, Py_PRINT_RAW)){
             return NULL;
         }
         Py_RETURN_TRUE;
@@ -581,11 +618,17 @@ PyObject* Linker_LinkInModule(llvm::Linker* Linker,
                                PyObject* ErrMsg)
 {
     std::string errmsg;
+#if LLVM_VERSION_MAJOR >= 3 && LLVM_VERSION_MINOR >= 3
+    bool failed = Linker->linkInModule(Mod, &errmsg);
+#else
     bool failed = Linker->LinkInModule(Mod, &errmsg);
+#endif
     if (! failed) {
         Py_RETURN_FALSE;
     } else {
-        if (-1 == PyFile_WriteString(errmsg.c_str(), ErrMsg)) {
+
+        auto_pyobject buf = PyBytes_FromString(errmsg.c_str());
+        if (NULL == callwrite(ErrMsg, *buf)){
             return NULL;
         }
         Py_RETURN_TRUE;
@@ -603,9 +646,13 @@ PyObject* Linker_LinkModules(llvm::Module* Dest,
     if (! failed) {
         Py_RETURN_FALSE;
     } else {
-        if (-1 == PyFile_WriteString(errmsg.c_str(), ErrMsg)) {
+        auto_pyobject buf = PyBytes_FromString(errmsg.c_str());
+        if (NULL == callwrite(ErrMsg, *buf)){
             return NULL;
         }
+//        if (-1 == PyFile_WriteString(errmsg.c_str(), ErrMsg)) {
+//            return NULL;
+//        }
         Py_RETURN_TRUE;
     }
 }
@@ -665,9 +712,14 @@ PyObject* llvm_verifyModule(const llvm::Module& Fn,
     bool failed = llvm::verifyModule(Fn, Action, &errmsg);
 
     if (failed) {
-        if (-1 == PyFile_WriteString(errmsg.c_str(), ErrMsg)) {
+        auto_pyobject buf = PyBytes_FromString(errmsg.c_str());
+        if (NULL == callwrite(ErrMsg, *buf)){
             return NULL;
         }
+
+//        if (-1 == PyFile_WriteString(errmsg.c_str(), ErrMsg)) {
+//            return NULL;
+//        }
         Py_RETURN_TRUE;
     } else {
         Py_RETURN_FALSE;
@@ -790,9 +842,13 @@ PyObject* DynamicLibrary_LoadLibraryPermanently(const char * Filename,
         std::string errmsg;
         failed = DynamicLibrary::LoadLibraryPermanently(Filename, &errmsg);
         if (failed) {
-            if (-1 == PyFile_WriteString(errmsg.c_str(), ErrMsg)) {
+            auto_pyobject buf = PyBytes_FromString(errmsg.c_str());
+            if (NULL == callwrite(ErrMsg, *buf)){
                 return NULL;
             }
+//            if (-1 == PyFile_WriteString(errmsg.c_str(), ErrMsg)) {
+//                return NULL;
+//            }
         }
     } else {
         failed = DynamicLibrary::LoadLibraryPermanently(Filename);
@@ -805,6 +861,27 @@ PyObject* DynamicLibrary_LoadLibraryPermanently(const char * Filename,
     }
 }
 
+static
+PyObject* DynamicLibrary_getPermanentLibrary(const char * Filename,
+                                                PyObject* ErrMsg = 0)
+{
+    using namespace llvm::sys;
+    std::string errmsg;
+    DynamicLibrary dylib = DynamicLibrary::getPermanentLibrary(Filename, &errmsg);
+    if (!dylib.isValid()) {
+        if (ErrMsg) {
+            auto_pyobject buf = PyBytes_FromString(errmsg.c_str());
+            if (!callwrite(ErrMsg, *buf))
+                return NULL;
+        }
+        PyErr_SetString(PyExc_RuntimeError, errmsg.c_str());
+        return NULL;
+    }
+    return pycapsule_new(new DynamicLibrary(dylib),
+                     "llvm::sys::DynamicLibrary",
+                     "llvm::sys::DynamicLibrary");
+}
+
 class PassRegistryEnumerator : public llvm::PassRegistrationListener{
 public:
     PyObject* List;
@@ -814,7 +891,8 @@ public:
     inline virtual void passEnumerate(const llvm::PassInfo * pass_info){
         PyObject* passArg = PyString_FromString(pass_info->getPassArgument());
         PyObject* passName = PyString_FromString(pass_info->getPassName());
-        PyList_Append(List, Py_BuildValue("(OO)", passArg, passName));
+        auto_pyobject pair = Py_BuildValue("(OO)", passArg, passName);
+        PyList_Append(List, *pair);
     }
 };
 
@@ -835,7 +913,12 @@ PyObject* TargetRegistry_lookupTarget(const std::string &Triple,
     std::string error;
     const Target* target = TargetRegistry::lookupTarget(Triple, error);
     if (!target) {
-        PyFile_WriteString(error.c_str(), Error);
+//        PyFile_WriteString(error.c_str(), Error);
+        auto_pyobject buf = PyBytes_FromString(error.c_str());
+        if (NULL == callwrite(Error, *buf)){
+            return NULL;
+        }
+
         Py_RETURN_NONE;
     } else {
         return pycapsule_new(const_cast<Target*>(target), "llvm::Target");
@@ -852,14 +935,18 @@ PyObject* TargetRegistry_lookupTarget(const std::string &Arch,
     std::string error;
     const Target* target = TargetRegistry::lookupTarget(Arch, Triple, error);
     if (!target) {
-        PyFile_WriteString(error.c_str(), Error);
+//        PyFile_WriteString(error.c_str(), Error);
+        auto_pyobject buf = PyBytes_FromString(error.c_str());
+        if (NULL == callwrite(Error, *buf)){
+            return NULL;
+        }
+
         Py_RETURN_NONE;
     } else {
         return pycapsule_new(const_cast<Target*>(target), "llvm::Target");
     }
 }
 
-
 static
 PyObject* TargetRegistry_getClosestTargetForJIT(PyObject* Error)
 {
@@ -867,7 +954,12 @@ PyObject* TargetRegistry_getClosestTargetForJIT(PyObject* Error)
     std::string error;
     const Target* target = TargetRegistry::getClosestTargetForJIT(error);
     if (!target) {
-        PyFile_WriteString(error.c_str(), Error);
+        auto_pyobject buf = PyBytes_FromString(error.c_str());
+        if (NULL == callwrite(Error, *buf)){
+            return NULL;
+        }
+
+//        PyFile_WriteString(error.c_str(), Error);
         Py_RETURN_NONE;
     } else {
         return pycapsule_new(const_cast<Target*>(target), "llvm::Target");
@@ -875,6 +967,66 @@ PyObject* TargetRegistry_getClosestTargetForJIT(PyObject* Error)
 
 }
 
+static
+PyObject* TargetRegistry_targets_list()
+{
+    using namespace llvm;
+
+    return iterator_to_pylist_deref<TargetRegistry::iterator>(
+                  TargetRegistry::begin(), TargetRegistry::end(),
+                  "llvm::Target", "llvm::Target");
+}
+
+#if LLVM_VERSION_MAJOR >= 3 && LLVM_VERSION_MINOR >= 4
+static
+PyObject* MemoryObject_readBytes(const llvm::MemoryObject *mobj,
+                                 uint64_t addr,
+                                 uint64_t size
+                                 )
+{
+    int status;
+    uint8_t *bytes;
+    PyObject* po;
+
+    if(size < 1)
+        goto fail;
+
+    bytes = new uint8_t[size];
+    if(bytes == NULL)
+        goto fail;
+
+    status = mobj->readBytes(addr, size, bytes);
+    if(status != 0) {
+        delete bytes;
+        goto fail;
+    }
+
+    po = PyBytes_FromStringAndSize((const char *) bytes, size);
+    delete bytes;
+
+    return po;
+fail:
+    Py_RETURN_NONE;
+}
+
+static
+PyObject* MCDisassembler_getInstruction(llvm::MCDisassembler *disasm,
+                                        llvm::MCInst &instr,
+                                        const llvm::MemoryObject &region,
+                                        uint64_t address
+                                        )
+{
+    uint64_t size;
+    llvm::MCDisassembler::DecodeStatus status;
+
+    size = 0;
+    status = disasm->getInstruction(instr, size, region, address,
+                                    llvm::nulls(), llvm::nulls());
+    return Py_BuildValue("(i,i)", int(status), size);
+}
+
+#endif /* llvm >= 3.4 */
+
 static
 PyObject* llvm_sys_getHostCPUFeatures(PyObject* Features)
 {
@@ -898,3 +1050,23 @@ PyObject* llvm_sys_getHostCPUFeatures(PyObject* Features)
     }
 }
 
+#if LLVM_VERSION_MAJOR >= 3 && LLVM_VERSION_MINOR >= 3
+static
+PyObject* llvm_sys_isLittleEndianHost()
+{
+    if (llvm::sys::IsLittleEndianHost)
+        Py_RETURN_TRUE;
+    else
+        Py_RETURN_FALSE;
+}
+
+static
+PyObject* llvm_sys_isBigEndianHost()
+{
+    if (llvm::sys::IsBigEndianHost)
+        Py_RETURN_TRUE;
+    else
+        Py_RETURN_FALSE;
+}
+#endif
+

llvmpy/setup.py

@@ -31,7 +31,7 @@ components = ['core', 'analysis', 'scalaropts',
               'interpreter', 'bitreader',
               'bitwriter', 'instrumentation', 'ipa',
               'ipo', 'transformutils', 'asmparser',
-              'linker', 'support', 'vectorize',
+              'linker', 'support', 'vectorize', 'all-targets'
               ]
 
 nvptx = ['nvptx',