Merge branch 'llrt'

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/core.py

@@ -1848,6 +1848,9 @@ class Instruction(User):
     def erase_from_parent(self):
         return self._ptr.eraseFromParent()
 
+    def replace_all_uses_with(self, inst):
+        self._ptr.replaceAllUsesWith(inst)
+
 
 class CallOrInvokeInstruction(Instruction):
     _type_ = api.llvm.CallInst, api.llvm.InvokeInst

llvm/llrt.py

@@ -0,0 +1,77 @@
+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.
+    '''
+    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/test_llvmpy.py

@@ -1416,6 +1416,32 @@ if llvm.version >= (3, 3):
     tests.append(TestMCJIT)
 
 
+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, c_int64
+        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)
+
 class TestArith(TestCase):
     '''
     Test basic arithmetic support with LLVM MCJIT
@@ -1447,10 +1473,12 @@ class TestArith(TestCase):
         inttys = [Type.int(32), Type.int(64)]
         flttys = [Type.float(), Type.double()]
 
-        for ty in inttys:
-            self.func_template(ty, iop)        
-        for ty in flttys:
-            self.func_template(ty, fop)
+        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')
@@ -1466,14 +1494,14 @@ class TestArith(TestCase):
             print('skipped test for div')
             print('known failure due to unresolved external symbol __udivdi3')
             return
-        self.template('udiv', 'fdiv')
+        self.template('udiv', None) # 'fdiv')
 
     def test_rem(self):
         if BITS == 32:
             print('skipped test for rem')
             print('known failure due to unresolved external symbol __umoddi3')
             return
-        self.template('urem', 'frem')
+        self.template('urem', None) # 'frem')
 
 if llvm.version >= (3, 3):
     # MCJIT is broken in 3.2
@@ -1551,6 +1579,8 @@ class TestExact(TestCase):
 
 tests.append(TestExact)
 
+
+
 # ---------------------------------------------------------------------------
 
 def run(verbosity=1):