Move original compiler-rt functions (libgcc replacement) to lib/builtins directory

git-svn-id: https://llvm.org/svn/llvm-project/compiler-rt/trunk@201393 91177308-0d34-0410-b5e6-96231b3b80d8

237 files changed, 13730 insertions, 0 deletions

diff --git a/lib/builtins/CMakeLists.txt b/lib/builtins/CMakeLists.txt
new file mode 100644
index 000000000..81ae21f7d
--- /dev/null
+++ b/lib/builtins/CMakeLists.txt
@@ -0,0 +1,233 @@
+# This directory contains a large amount of C code which provides
+# generic implementations of the core runtime library along with optimized
+# architecture-specific code in various subdirectories.
+
+set(GENERIC_SOURCES
+  absvdi2.c
+  absvsi2.c
+  absvti2.c
+  adddf3.c
+  addsf3.c
+  addvdi3.c
+  addvsi3.c
+  addvti3.c
+  apple_versioning.c
+  ashldi3.c
+  ashlti3.c
+  ashrdi3.c
+  ashrti3.c
+  # FIXME: atomic.c may only be compiled if host compiler understands _Atomic
+  # atomic.c
+  clear_cache.c
+  clzdi2.c
+  clzsi2.c
+  clzti2.c
+  cmpdi2.c
+  cmpti2.c
+  comparedf2.c
+  comparesf2.c
+  ctzdi2.c
+  ctzsi2.c
+  ctzti2.c
+  divdc3.c
+  divdf3.c
+  divdi3.c
+  divmoddi4.c
+  divmodsi4.c
+  divsc3.c
+  divsf3.c
+  divsi3.c
+  divti3.c
+  divxc3.c
+  enable_execute_stack.c
+  eprintf.c
+  extendsfdf2.c
+  ffsdi2.c
+  ffsti2.c
+  fixdfdi.c
+  fixdfsi.c
+  fixdfti.c
+  fixsfdi.c
+  fixsfsi.c
+  fixsfti.c
+  fixunsdfdi.c
+  fixunsdfsi.c
+  fixunsdfti.c
+  fixunssfdi.c
+  fixunssfsi.c
+  fixunssfti.c
+  fixunsxfdi.c
+  fixunsxfsi.c
+  fixunsxfti.c
+  fixxfdi.c
+  fixxfti.c
+  floatdidf.c
+  floatdisf.c
+  floatdixf.c
+  floatsidf.c
+  floatsisf.c
+  floattidf.c
+  floattisf.c
+  floattixf.c
+  floatundidf.c
+  floatundisf.c
+  floatundixf.c
+  floatunsidf.c
+  floatunsisf.c
+  floatuntidf.c
+  floatuntisf.c
+  floatuntixf.c
+  gcc_personality_v0.c
+  int_util.c
+  lshrdi3.c
+  lshrti3.c
+  moddi3.c
+  modsi3.c
+  modti3.c
+  muldc3.c
+  muldf3.c
+  muldi3.c
+  mulodi4.c
+  mulosi4.c
+  muloti4.c
+  mulsc3.c
+  mulsf3.c
+  multi3.c
+  mulvdi3.c
+  mulvsi3.c
+  mulvti3.c
+  mulxc3.c
+  negdf2.c
+  negdi2.c
+  negsf2.c
+  negti2.c
+  negvdi2.c
+  negvsi2.c
+  negvti2.c
+  paritydi2.c
+  paritysi2.c
+  parityti2.c
+  popcountdi2.c
+  popcountsi2.c
+  popcountti2.c
+  powidf2.c
+  powisf2.c
+  powitf2.c
+  powixf2.c
+  subdf3.c
+  subsf3.c
+  subvdi3.c
+  subvsi3.c
+  subvti3.c
+  trampoline_setup.c
+  truncdfsf2.c
+  ucmpdi2.c
+  ucmpti2.c
+  udivdi3.c
+  udivmoddi4.c
+  udivmodsi4.c
+  udivmodti4.c
+  udivsi3.c
+  udivti3.c
+  umoddi3.c
+  umodsi3.c
+  umodti3.c)
+
+set(x86_64_SOURCES
+  x86_64/floatdidf.c
+  x86_64/floatdisf.c
+  x86_64/floatdixf.c
+  x86_64/floatundidf.S
+  x86_64/floatundisf.S
+  x86_64/floatundixf.S
+  ${GENERIC_SOURCES})
+
+set(i386_SOURCES
+  i386/ashldi3.S
+  i386/ashrdi3.S
+  i386/divdi3.S
+  i386/floatdidf.S
+  i386/floatdisf.S
+  i386/floatdixf.S
+  i386/floatundidf.S
+  i386/floatundisf.S
+  i386/floatundixf.S
+  i386/lshrdi3.S
+  i386/moddi3.S
+  i386/muldi3.S
+  i386/udivdi3.S
+  i386/umoddi3.S
+  ${GENERIC_SOURCES})
+
+set(arm_SOURCES
+  arm/adddf3vfp.S
+  arm/addsf3vfp.S
+  arm/aeabi_dcmp.S
+  arm/aeabi_fcmp.S
+  arm/aeabi_idivmod.S
+  arm/aeabi_ldivmod.S
+  arm/aeabi_memcmp.S
+  arm/aeabi_memcpy.S
+  arm/aeabi_memmove.S
+  arm/aeabi_memset.S
+  arm/aeabi_uidivmod.S
+  arm/aeabi_uldivmod.S
+  arm/bswapdi2.S
+  arm/bswapsi2.S
+  arm/comparesf2.S
+  arm/divdf3vfp.S
+  arm/divmodsi4.S
+  arm/divsf3vfp.S
+  arm/divsi3.S
+  arm/eqdf2vfp.S
+  arm/eqsf2vfp.S
+  arm/extendsfdf2vfp.S
+  arm/fixdfsivfp.S
+  arm/fixsfsivfp.S
+  arm/fixunsdfsivfp.S
+  arm/fixunssfsivfp.S
+  arm/floatsidfvfp.S
+  arm/floatsisfvfp.S
+  arm/floatunssidfvfp.S
+  arm/floatunssisfvfp.S
+  arm/gedf2vfp.S
+  arm/gesf2vfp.S
+  arm/gtdf2vfp.S
+  arm/gtsf2vfp.S
+  arm/ledf2vfp.S
+  arm/lesf2vfp.S
+  arm/ltdf2vfp.S
+  arm/ltsf2vfp.S
+  arm/modsi3.S
+  arm/muldf3vfp.S
+  arm/mulsf3vfp.S
+  arm/nedf2vfp.S
+  arm/negdf2vfp.S
+  arm/negsf2vfp.S
+  arm/nesf2vfp.S
+  arm/restore_vfp_d8_d15_regs.S
+  arm/save_vfp_d8_d15_regs.S
+  arm/subdf3vfp.S
+  arm/subsf3vfp.S
+  arm/switch16.S
+  arm/switch32.S
+  arm/switch8.S
+  arm/switchu8.S
+  arm/sync_synchronize.S
+  arm/truncdfsf2vfp.S
+  arm/udivmodsi4.S
+  arm/udivsi3.S
+  arm/umodsi3.S
+  arm/unorddf2vfp.S
+  arm/unordsf2vfp.S
+  ${GENERIC_SOURCES})
+
+if (NOT WIN32)
+  foreach(arch x86_64 i386 arm)
+    if(CAN_TARGET_${arch})
+      add_compiler_rt_static_runtime(clang_rt.${arch} ${arch}
+        SOURCES ${${arch}_SOURCES}
+        CFLAGS "-std=c99")
+    endif()
+  endforeach()
+endif()
diff --git a/lib/builtins/Makefile.mk b/lib/builtins/Makefile.mk
new file mode 100644
index 000000000..3143d91f0
--- /dev/null
+++ b/lib/builtins/Makefile.mk
@@ -0,0 +1,22 @@
+#===- lib/builtins/Makefile.mk -----------------------------*- Makefile -*--===#
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+#===------------------------------------------------------------------------===#
+
+ModuleName := builtins
+SubDirs :=
+
+# Add arch specific optimized implementations.
+SubDirs += i386 ppc x86_64 arm
+
+# Define the variables for this specific directory.
+Sources := $(foreach file,$(wildcard $(Dir)/*.c),$(notdir $(file)))
+ObjNames := $(Sources:%.c=%.o)
+Implementation := Generic
+
+# FIXME: use automatic dependencies?
+Dependencies := $(wildcard $(Dir)/*.h)
diff --git a/lib/builtins/absvdi2.c b/lib/builtins/absvdi2.c
new file mode 100644
index 000000000..682c2355d
--- /dev/null
+++ b/lib/builtins/absvdi2.c
@@ -0,0 +1,29 @@
+/*===-- absvdi2.c - Implement __absvdi2 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ *===----------------------------------------------------------------------===
+ *
+ * This file implements __absvdi2 for the compiler_rt library.
+ *
+ *===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: absolute value */
+
+/* Effects: aborts if abs(x) < 0 */
+
+COMPILER_RT_ABI di_int
+__absvdi2(di_int a)
+{
+    const int N = (int)(sizeof(di_int) * CHAR_BIT);
+    if (a == ((di_int)1 << (N-1)))
+        compilerrt_abort();
+    const di_int t = a >> (N - 1);
+    return (a ^ t) - t;
+}
diff --git a/lib/builtins/absvsi2.c b/lib/builtins/absvsi2.c
new file mode 100644
index 000000000..4812af815
--- /dev/null
+++ b/lib/builtins/absvsi2.c
@@ -0,0 +1,29 @@
+/* ===-- absvsi2.c - Implement __absvsi2 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __absvsi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: absolute value */
+
+/* Effects: aborts if abs(x) < 0 */
+
+COMPILER_RT_ABI si_int
+__absvsi2(si_int a)
+{
+    const int N = (int)(sizeof(si_int) * CHAR_BIT);
+    if (a == (1 << (N-1)))
+        compilerrt_abort();
+    const si_int t = a >> (N - 1);
+    return (a ^ t) - t;
+}
diff --git a/lib/builtins/absvti2.c b/lib/builtins/absvti2.c
new file mode 100644
index 000000000..c1c727798
--- /dev/null
+++ b/lib/builtins/absvti2.c
@@ -0,0 +1,33 @@
+/* ===-- absvti2.c - Implement __absvdi2 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __absvti2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: absolute value */
+
+/* Effects: aborts if abs(x) < 0 */
+
+ti_int
+__absvti2(ti_int a)
+{
+    const int N = (int)(sizeof(ti_int) * CHAR_BIT);
+    if (a == ((ti_int)1 << (N-1)))
+        compilerrt_abort();
+    const ti_int s = a >> (N - 1);
+    return (a ^ s) - s;
+}
+
+#endif
diff --git a/lib/builtins/adddf3.c b/lib/builtins/adddf3.c
new file mode 100644
index 000000000..a55e82d21
--- /dev/null
+++ b/lib/builtins/adddf3.c
@@ -0,0 +1,152 @@
+//===-- lib/adddf3.c - Double-precision addition ------------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements double-precision soft-float addition with the IEEE-754
+// default rounding (to nearest, ties to even).
+//
+//===----------------------------------------------------------------------===//
+
+#define DOUBLE_PRECISION
+#include "fp_lib.h"
+
+ARM_EABI_FNALIAS(dadd, adddf3)
+
+COMPILER_RT_ABI fp_t
+__adddf3(fp_t a, fp_t b) {
+    
+    rep_t aRep = toRep(a);
+    rep_t bRep = toRep(b);
+    const rep_t aAbs = aRep & absMask;
+    const rep_t bAbs = bRep & absMask;
+    
+    // Detect if a or b is zero, infinity, or NaN.
+    if (aAbs - 1U >= infRep - 1U || bAbs - 1U >= infRep - 1U) {
+        
+        // NaN + anything = qNaN
+        if (aAbs > infRep) return fromRep(toRep(a) | quietBit);
+        // anything + NaN = qNaN
+        if (bAbs > infRep) return fromRep(toRep(b) | quietBit);
+        
+        if (aAbs == infRep) {
+            // +/-infinity + -/+infinity = qNaN
+            if ((toRep(a) ^ toRep(b)) == signBit) return fromRep(qnanRep);
+            // +/-infinity + anything remaining = +/- infinity
+            else return a;
+        }
+        
+        // anything remaining + +/-infinity = +/-infinity
+        if (bAbs == infRep) return b;
+        
+        // zero + anything = anything
+        if (!aAbs) {
+            // but we need to get the sign right for zero + zero
+            if (!bAbs) return fromRep(toRep(a) & toRep(b));
+            else return b;
+        }
+        
+        // anything + zero = anything
+        if (!bAbs) return a;
+    }
+    
+    // Swap a and b if necessary so that a has the larger absolute value.
+    if (bAbs > aAbs) {
+        const rep_t temp = aRep;
+        aRep = bRep;
+        bRep = temp;
+    }
+    
+    // Extract the exponent and significand from the (possibly swapped) a and b.
+    int aExponent = aRep >> significandBits & maxExponent;
+    int bExponent = bRep >> significandBits & maxExponent;
+    rep_t aSignificand = aRep & significandMask;
+    rep_t bSignificand = bRep & significandMask;
+    
+    // Normalize any denormals, and adjust the exponent accordingly.
+    if (aExponent == 0) aExponent = normalize(&aSignificand);
+    if (bExponent == 0) bExponent = normalize(&bSignificand);
+    
+    // The sign of the result is the sign of the larger operand, a.  If they
+    // have opposite signs, we are performing a subtraction; otherwise addition.
+    const rep_t resultSign = aRep & signBit;
+    const bool subtraction = (aRep ^ bRep) & signBit;
+    
+    // Shift the significands to give us round, guard and sticky, and or in the
+    // implicit significand bit.  (If we fell through from the denormal path it
+    // was already set by normalize( ), but setting it twice won't hurt
+    // anything.)
+    aSignificand = (aSignificand | implicitBit) << 3;
+    bSignificand = (bSignificand | implicitBit) << 3;
+    
+    // Shift the significand of b by the difference in exponents, with a sticky
+    // bottom bit to get rounding correct.
+    const unsigned int align = aExponent - bExponent;
+    if (align) {
+        if (align < typeWidth) {
+            const bool sticky = bSignificand << (typeWidth - align);
+            bSignificand = bSignificand >> align | sticky;
+        } else {
+            bSignificand = 1; // sticky; b is known to be non-zero.
+        }
+    }
+    
+    if (subtraction) {
+        aSignificand -= bSignificand;
+        
+        // If a == -b, return +zero.
+        if (aSignificand == 0) return fromRep(0);
+        
+        // If partial cancellation occured, we need to left-shift the result
+        // and adjust the exponent:
+        if (aSignificand < implicitBit << 3) {
+            const int shift = rep_clz(aSignificand) - rep_clz(implicitBit << 3);
+            aSignificand <<= shift;
+            aExponent -= shift;
+        }
+    }
+    
+    else /* addition */ {
+        aSignificand += bSignificand;
+        
+        // If the addition carried up, we need to right-shift the result and
+        // adjust the exponent:
+        if (aSignificand & implicitBit << 4) {
+            const bool sticky = aSignificand & 1;
+            aSignificand = aSignificand >> 1 | sticky;
+            aExponent += 1;
+        }
+    }
+    
+    // If we have overflowed the type, return +/- infinity:
+    if (aExponent >= maxExponent) return fromRep(infRep | resultSign);
+    
+    if (aExponent <= 0) {
+        // Result is denormal before rounding; the exponent is zero and we
+        // need to shift the significand.
+        const int shift = 1 - aExponent;
+        const bool sticky = aSignificand << (typeWidth - shift);
+        aSignificand = aSignificand >> shift | sticky;
+        aExponent = 0;
+    }
+    
+    // Low three bits are round, guard, and sticky.
+    const int roundGuardSticky = aSignificand & 0x7;
+    
+    // Shift the significand into place, and mask off the implicit bit.
+    rep_t result = aSignificand >> 3 & significandMask;
+    
+    // Insert the exponent and sign.
+    result |= (rep_t)aExponent << significandBits;
+    result |= resultSign;
+    
+    // Final rounding.  The result may overflow to infinity, but that is the
+    // correct result in that case.
+    if (roundGuardSticky > 0x4) result++;
+    if (roundGuardSticky == 0x4) result += result & 1;
+    return fromRep(result);
+}
diff --git a/lib/builtins/addsf3.c b/lib/builtins/addsf3.c
new file mode 100644
index 000000000..0268324de
--- /dev/null
+++ b/lib/builtins/addsf3.c
@@ -0,0 +1,151 @@
+//===-- lib/addsf3.c - Single-precision addition ------------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements single-precision soft-float addition with the IEEE-754
+// default rounding (to nearest, ties to even).
+//
+//===----------------------------------------------------------------------===//
+
+#define SINGLE_PRECISION
+#include "fp_lib.h"
+
+ARM_EABI_FNALIAS(fadd, addsf3)
+
+fp_t __addsf3(fp_t a, fp_t b) {
+
+    rep_t aRep = toRep(a);
+    rep_t bRep = toRep(b);
+    const rep_t aAbs = aRep & absMask;
+    const rep_t bAbs = bRep & absMask;
+    
+    // Detect if a or b is zero, infinity, or NaN.
+    if (aAbs - 1U >= infRep - 1U || bAbs - 1U >= infRep - 1U) {
+        
+        // NaN + anything = qNaN
+        if (aAbs > infRep) return fromRep(toRep(a) | quietBit);
+        // anything + NaN = qNaN
+        if (bAbs > infRep) return fromRep(toRep(b) | quietBit);
+        
+        if (aAbs == infRep) {
+            // +/-infinity + -/+infinity = qNaN
+            if ((toRep(a) ^ toRep(b)) == signBit) return fromRep(qnanRep);
+            // +/-infinity + anything remaining = +/- infinity
+            else return a;
+        }
+        
+        // anything remaining + +/-infinity = +/-infinity
+        if (bAbs == infRep) return b;
+        
+        // zero + anything = anything
+        if (!aAbs) {
+            // but we need to get the sign right for zero + zero
+            if (!bAbs) return fromRep(toRep(a) & toRep(b));
+            else return b;
+        }
+        
+        // anything + zero = anything
+        if (!bAbs) return a;
+    }
+    
+    // Swap a and b if necessary so that a has the larger absolute value.
+    if (bAbs > aAbs) {
+        const rep_t temp = aRep;
+        aRep = bRep;
+        bRep = temp;
+    }
+    
+    // Extract the exponent and significand from the (possibly swapped) a and b.
+    int aExponent = aRep >> significandBits & maxExponent;
+    int bExponent = bRep >> significandBits & maxExponent;
+    rep_t aSignificand = aRep & significandMask;
+    rep_t bSignificand = bRep & significandMask;
+    
+    // Normalize any denormals, and adjust the exponent accordingly.
+    if (aExponent == 0) aExponent = normalize(&aSignificand);
+    if (bExponent == 0) bExponent = normalize(&bSignificand);
+    
+    // The sign of the result is the sign of the larger operand, a.  If they
+    // have opposite signs, we are performing a subtraction; otherwise addition.
+    const rep_t resultSign = aRep & signBit;
+    const bool subtraction = (aRep ^ bRep) & signBit;
+    
+    // Shift the significands to give us round, guard and sticky, and or in the
+    // implicit significand bit.  (If we fell through from the denormal path it
+    // was already set by normalize( ), but setting it twice won't hurt
+    // anything.)
+    aSignificand = (aSignificand | implicitBit) << 3;
+    bSignificand = (bSignificand | implicitBit) << 3;
+    
+    // Shift the significand of b by the difference in exponents, with a sticky
+    // bottom bit to get rounding correct.
+    const unsigned int align = aExponent - bExponent;
+    if (align) {
+        if (align < typeWidth) {
+            const bool sticky = bSignificand << (typeWidth - align);
+            bSignificand = bSignificand >> align | sticky;
+        } else {
+            bSignificand = 1; // sticky; b is known to be non-zero.
+        }
+    }
+    
+    if (subtraction) {
+        aSignificand -= bSignificand;
+        
+        // If a == -b, return +zero.
+        if (aSignificand == 0) return fromRep(0);
+        
+        // If partial cancellation occured, we need to left-shift the result
+        // and adjust the exponent:
+        if (aSignificand < implicitBit << 3) {
+            const int shift = rep_clz(aSignificand) - rep_clz(implicitBit << 3);
+            aSignificand <<= shift;
+            aExponent -= shift;
+        }
+    }
+    
+    else /* addition */ {
+        aSignificand += bSignificand;
+        
+        // If the addition carried up, we need to right-shift the result and
+        // adjust the exponent:
+        if (aSignificand & implicitBit << 4) {
+            const bool sticky = aSignificand & 1;
+            aSignificand = aSignificand >> 1 | sticky;
+            aExponent += 1;
+        }
+    }
+    
+    // If we have overflowed the type, return +/- infinity:
+    if (aExponent >= maxExponent) return fromRep(infRep | resultSign);
+    
+    if (aExponent <= 0) {
+        // Result is denormal before rounding; the exponent is zero and we
+        // need to shift the significand.
+        const int shift = 1 - aExponent;
+        const bool sticky = aSignificand << (typeWidth - shift);
+        aSignificand = aSignificand >> shift | sticky;
+        aExponent = 0;
+    }
+    
+    // Low three bits are round, guard, and sticky.
+    const int roundGuardSticky = aSignificand & 0x7;
+    
+    // Shift the significand into place, and mask off the implicit bit.
+    rep_t result = aSignificand >> 3 & significandMask;
+    
+    // Insert the exponent and sign.
+    result |= (rep_t)aExponent << significandBits;
+    result |= resultSign;
+    
+    // Final rounding.  The result may overflow to infinity, but that is the
+    // correct result in that case.
+    if (roundGuardSticky > 0x4) result++;
+    if (roundGuardSticky == 0x4) result += result & 1;
+    return fromRep(result);
+}
diff --git a/lib/builtins/addvdi3.c b/lib/builtins/addvdi3.c
new file mode 100644
index 000000000..db45a27f0
--- /dev/null
+++ b/lib/builtins/addvdi3.c
@@ -0,0 +1,36 @@
+/* ===-- addvdi3.c - Implement __addvdi3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __addvdi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a + b */
+
+/* Effects: aborts if a + b overflows */
+
+COMPILER_RT_ABI di_int
+__addvdi3(di_int a, di_int b)
+{
+    di_int s = a + b;
+    if (b >= 0)
+    {
+        if (s < a)
+            compilerrt_abort();
+    }
+    else
+    {
+        if (s >= a)
+            compilerrt_abort();
+    }
+    return s;
+}
diff --git a/lib/builtins/addvsi3.c b/lib/builtins/addvsi3.c
new file mode 100644
index 000000000..81f515cd7
--- /dev/null
+++ b/lib/builtins/addvsi3.c
@@ -0,0 +1,36 @@
+/* ===-- addvsi3.c - Implement __addvsi3 -----------------------------------===
+ *
+ *                    The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __addvsi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a + b */
+
+/* Effects: aborts if a + b overflows */
+
+COMPILER_RT_ABI si_int
+__addvsi3(si_int a, si_int b)
+{
+    si_int s = a + b;
+    if (b >= 0)
+    {
+        if (s < a)
+            compilerrt_abort();
+    }
+    else
+    {
+        if (s >= a)
+            compilerrt_abort();
+    }
+    return s;
+}
diff --git a/lib/builtins/addvti3.c b/lib/builtins/addvti3.c
new file mode 100644
index 000000000..2efcf3b40
--- /dev/null
+++ b/lib/builtins/addvti3.c
@@ -0,0 +1,40 @@
+/* ===-- addvti3.c - Implement __addvti3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __addvti3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: a + b */
+
+/* Effects: aborts if a + b overflows */
+
+ti_int
+__addvti3(ti_int a, ti_int b)
+{
+    ti_int s = a + b;
+    if (b >= 0)
+    {
+        if (s < a)
+            compilerrt_abort();
+    }
+    else
+    {
+        if (s >= a)
+            compilerrt_abort();
+    }
+    return s;
+}
+
+#endif
diff --git a/lib/builtins/apple_versioning.c b/lib/builtins/apple_versioning.c
new file mode 100644
index 000000000..3797a1ab0
--- /dev/null
+++ b/lib/builtins/apple_versioning.c
@@ -0,0 +1,350 @@
+/* ===-- apple_versioning.c - Adds versioning symbols for ld ---------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+
+#if __APPLE__
+  #include <Availability.h>
+  
+  #if __IPHONE_OS_VERSION_MIN_REQUIRED
+    #define NOT_HERE_BEFORE_10_6(sym) 
+    #define NOT_HERE_IN_10_8_AND_EARLIER(sym) \
+        extern const char sym##_tmp61 __asm("$ld$hide$os6.1$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp61 = 0; \
+        extern const char sym##_tmp60 __asm("$ld$hide$os6.0$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp60 = 0; \
+        extern const char sym##_tmp51 __asm("$ld$hide$os5.1$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp51 = 0; \
+        extern const char sym##_tmp50 __asm("$ld$hide$os5.0$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp50 = 0; 
+  #else
+    #define NOT_HERE_BEFORE_10_6(sym) \
+         extern const char sym##_tmp4 __asm("$ld$hide$os10.4$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp4 = 0; \
+        extern const char sym##_tmp5 __asm("$ld$hide$os10.5$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp5 = 0; 
+    #define NOT_HERE_IN_10_8_AND_EARLIER(sym) \
+         extern const char sym##_tmp8 __asm("$ld$hide$os10.8$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp8 = 0; \
+        extern const char sym##_tmp7 __asm("$ld$hide$os10.7$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp7 = 0; \
+        extern const char sym##_tmp6 __asm("$ld$hide$os10.6$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp6 = 0; 
+  #endif 
+
+
+/* Symbols in libSystem.dylib in 10.6 and later, 
+ *  but are in libgcc_s.dylib in earlier versions
+ */
+
+NOT_HERE_BEFORE_10_6(__absvdi2)
+NOT_HERE_BEFORE_10_6(__absvsi2)
+NOT_HERE_BEFORE_10_6(__absvti2)
+NOT_HERE_BEFORE_10_6(__addvdi3)
+NOT_HERE_BEFORE_10_6(__addvsi3)
+NOT_HERE_BEFORE_10_6(__addvti3)
+NOT_HERE_BEFORE_10_6(__ashldi3)
+NOT_HERE_BEFORE_10_6(__ashlti3)
+NOT_HERE_BEFORE_10_6(__ashrdi3)
+NOT_HERE_BEFORE_10_6(__ashrti3)
+NOT_HERE_BEFORE_10_6(__clear_cache)
+NOT_HERE_BEFORE_10_6(__clzdi2)
+NOT_HERE_BEFORE_10_6(__clzsi2)
+NOT_HERE_BEFORE_10_6(__clzti2)
+NOT_HERE_BEFORE_10_6(__cmpdi2)
+NOT_HERE_BEFORE_10_6(__cmpti2)
+NOT_HERE_BEFORE_10_6(__ctzdi2)
+NOT_HERE_BEFORE_10_6(__ctzsi2)
+NOT_HERE_BEFORE_10_6(__ctzti2)
+NOT_HERE_BEFORE_10_6(__divdc3)
+NOT_HERE_BEFORE_10_6(__divdi3)
+NOT_HERE_BEFORE_10_6(__divsc3)
+NOT_HERE_BEFORE_10_6(__divtc3)
+NOT_HERE_BEFORE_10_6(__divti3)
+NOT_HERE_BEFORE_10_6(__divxc3)
+NOT_HERE_BEFORE_10_6(__enable_execute_stack)
+NOT_HERE_BEFORE_10_6(__ffsdi2)
+NOT_HERE_BEFORE_10_6(__ffsti2)
+NOT_HERE_BEFORE_10_6(__fixdfdi)
+NOT_HERE_BEFORE_10_6(__fixdfti)
+NOT_HERE_BEFORE_10_6(__fixsfdi)
+NOT_HERE_BEFORE_10_6(__fixsfti)
+NOT_HERE_BEFORE_10_6(__fixtfdi)
+NOT_HERE_BEFORE_10_6(__fixunsdfdi)
+NOT_HERE_BEFORE_10_6(__fixunsdfsi)
+NOT_HERE_BEFORE_10_6(__fixunsdfti)
+NOT_HERE_BEFORE_10_6(__fixunssfdi)
+NOT_HERE_BEFORE_10_6(__fixunssfsi)
+NOT_HERE_BEFORE_10_6(__fixunssfti)
+NOT_HERE_BEFORE_10_6(__fixunstfdi)
+NOT_HERE_BEFORE_10_6(__fixunsxfdi)
+NOT_HERE_BEFORE_10_6(__fixunsxfsi)
+NOT_HERE_BEFORE_10_6(__fixunsxfti)
+NOT_HERE_BEFORE_10_6(__fixxfdi)
+NOT_HERE_BEFORE_10_6(__fixxfti)
+NOT_HERE_BEFORE_10_6(__floatdidf)
+NOT_HERE_BEFORE_10_6(__floatdisf)
+NOT_HERE_BEFORE_10_6(__floatditf)
+NOT_HERE_BEFORE_10_6(__floatdixf)
+NOT_HERE_BEFORE_10_6(__floattidf)
+NOT_HERE_BEFORE_10_6(__floattisf)
+NOT_HERE_BEFORE_10_6(__floattixf)
+NOT_HERE_BEFORE_10_6(__floatundidf)
+NOT_HERE_BEFORE_10_6(__floatundisf)
+NOT_HERE_BEFORE_10_6(__floatunditf)
+NOT_HERE_BEFORE_10_6(__floatundixf)
+NOT_HERE_BEFORE_10_6(__floatuntidf)
+NOT_HERE_BEFORE_10_6(__floatuntisf)
+NOT_HERE_BEFORE_10_6(__floatuntixf)
+NOT_HERE_BEFORE_10_6(__gcc_personality_v0)
+NOT_HERE_BEFORE_10_6(__lshrdi3)
+NOT_HERE_BEFORE_10_6(__lshrti3)
+NOT_HERE_BEFORE_10_6(__moddi3)
+NOT_HERE_BEFORE_10_6(__modti3)
+NOT_HERE_BEFORE_10_6(__muldc3)
+NOT_HERE_BEFORE_10_6(__muldi3)
+NOT_HERE_BEFORE_10_6(__mulsc3)
+NOT_HERE_BEFORE_10_6(__multc3)
+NOT_HERE_BEFORE_10_6(__multi3)
+NOT_HERE_BEFORE_10_6(__mulvdi3)
+NOT_HERE_BEFORE_10_6(__mulvsi3)
+NOT_HERE_BEFORE_10_6(__mulvti3)
+NOT_HERE_BEFORE_10_6(__mulxc3)
+NOT_HERE_BEFORE_10_6(__negdi2)
+NOT_HERE_BEFORE_10_6(__negti2)
+NOT_HERE_BEFORE_10_6(__negvdi2)
+NOT_HERE_BEFORE_10_6(__negvsi2)
+NOT_HERE_BEFORE_10_6(__negvti2)
+NOT_HERE_BEFORE_10_6(__paritydi2)
+NOT_HERE_BEFORE_10_6(__paritysi2)
+NOT_HERE_BEFORE_10_6(__parityti2)
+NOT_HERE_BEFORE_10_6(__popcountdi2)
+NOT_HERE_BEFORE_10_6(__popcountsi2)
+NOT_HERE_BEFORE_10_6(__popcountti2)
+NOT_HERE_BEFORE_10_6(__powidf2)
+NOT_HERE_BEFORE_10_6(__powisf2)
+NOT_HERE_BEFORE_10_6(__powitf2)
+NOT_HERE_BEFORE_10_6(__powixf2)
+NOT_HERE_BEFORE_10_6(__subvdi3)
+NOT_HERE_BEFORE_10_6(__subvsi3)
+NOT_HERE_BEFORE_10_6(__subvti3)
+NOT_HERE_BEFORE_10_6(__ucmpdi2)
+NOT_HERE_BEFORE_10_6(__ucmpti2)
+NOT_HERE_BEFORE_10_6(__udivdi3)
+NOT_HERE_BEFORE_10_6(__udivmoddi4)
+NOT_HERE_BEFORE_10_6(__udivmodti4)
+NOT_HERE_BEFORE_10_6(__udivti3)
+NOT_HERE_BEFORE_10_6(__umoddi3)
+NOT_HERE_BEFORE_10_6(__umodti3)
+
+
+#if __ppc__
+NOT_HERE_BEFORE_10_6(__gcc_qadd)
+NOT_HERE_BEFORE_10_6(__gcc_qdiv)
+NOT_HERE_BEFORE_10_6(__gcc_qmul)
+NOT_HERE_BEFORE_10_6(__gcc_qsub)
+NOT_HERE_BEFORE_10_6(__trampoline_setup)
+#endif /* __ppc__ */
+
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_compare_exchange)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_compare_exchange_1)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_compare_exchange_2)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_compare_exchange_4)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_compare_exchange_8)
+
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_exchange)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_exchange_1)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_exchange_2)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_exchange_4)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_exchange_8)
+
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_add_1)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_add_2)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_add_4)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_add_8)
+
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_and_1)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_and_2)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_and_4)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_and_8)
+
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_or_1)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_or_2)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_or_4)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_or_8)
+
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_sub_1)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_sub_2)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_sub_4)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_sub_8)
+
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_xor_1)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_xor_2)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_xor_4)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_fetch_xor_8)
+
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_load)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_load_1)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_load_2)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_load_4)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_load_8)
+
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_store)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_store_1)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_store_2)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_store_4)
+NOT_HERE_IN_10_8_AND_EARLIER(__atomic_store_8)
+
+
+#if __arm__ && __DYNAMIC__
+   #define NOT_HERE_UNTIL_AFTER_4_3(sym) \
+        extern const char sym##_tmp1 __asm("$ld$hide$os3.0$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp1 = 0; \
+        extern const char sym##_tmp2 __asm("$ld$hide$os3.1$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp2 = 0; \
+        extern const char sym##_tmp3 __asm("$ld$hide$os3.2$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp3 = 0; \
+        extern const char sym##_tmp4 __asm("$ld$hide$os4.0$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp4 = 0; \
+        extern const char sym##_tmp5 __asm("$ld$hide$os4.1$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp5 = 0; \
+        extern const char sym##_tmp6 __asm("$ld$hide$os4.2$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp6 = 0; \
+        extern const char sym##_tmp7 __asm("$ld$hide$os4.3$_" #sym ); \
+            __attribute__((visibility("default"))) const char sym##_tmp7 = 0; 
+            
+NOT_HERE_UNTIL_AFTER_4_3(__absvdi2)
+NOT_HERE_UNTIL_AFTER_4_3(__absvsi2)
+NOT_HERE_UNTIL_AFTER_4_3(__adddf3)
+NOT_HERE_UNTIL_AFTER_4_3(__adddf3vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__addsf3)
+NOT_HERE_UNTIL_AFTER_4_3(__addsf3vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__addvdi3)
+NOT_HERE_UNTIL_AFTER_4_3(__addvsi3)
+NOT_HERE_UNTIL_AFTER_4_3(__ashldi3)
+NOT_HERE_UNTIL_AFTER_4_3(__ashrdi3)
+NOT_HERE_UNTIL_AFTER_4_3(__bswapdi2)
+NOT_HERE_UNTIL_AFTER_4_3(__bswapsi2)
+NOT_HERE_UNTIL_AFTER_4_3(__clzdi2)
+NOT_HERE_UNTIL_AFTER_4_3(__clzsi2)
+NOT_HERE_UNTIL_AFTER_4_3(__cmpdi2)
+NOT_HERE_UNTIL_AFTER_4_3(__ctzdi2)
+NOT_HERE_UNTIL_AFTER_4_3(__ctzsi2)
+NOT_HERE_UNTIL_AFTER_4_3(__divdc3)
+NOT_HERE_UNTIL_AFTER_4_3(__divdf3)
+NOT_HERE_UNTIL_AFTER_4_3(__divdf3vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__divdi3)
+NOT_HERE_UNTIL_AFTER_4_3(__divsc3)
+NOT_HERE_UNTIL_AFTER_4_3(__divsf3)
+NOT_HERE_UNTIL_AFTER_4_3(__divsf3vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__divsi3)
+NOT_HERE_UNTIL_AFTER_4_3(__eqdf2)
+NOT_HERE_UNTIL_AFTER_4_3(__eqdf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__eqsf2)
+NOT_HERE_UNTIL_AFTER_4_3(__eqsf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__extendsfdf2)
+NOT_HERE_UNTIL_AFTER_4_3(__extendsfdf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__ffsdi2)
+NOT_HERE_UNTIL_AFTER_4_3(__fixdfdi)
+NOT_HERE_UNTIL_AFTER_4_3(__fixdfsi)
+NOT_HERE_UNTIL_AFTER_4_3(__fixdfsivfp)
+NOT_HERE_UNTIL_AFTER_4_3(__fixsfdi)
+NOT_HERE_UNTIL_AFTER_4_3(__fixsfsi)
+NOT_HERE_UNTIL_AFTER_4_3(__fixsfsivfp)
+NOT_HERE_UNTIL_AFTER_4_3(__fixunsdfdi)
+NOT_HERE_UNTIL_AFTER_4_3(__fixunsdfsi)
+NOT_HERE_UNTIL_AFTER_4_3(__fixunsdfsivfp)
+NOT_HERE_UNTIL_AFTER_4_3(__fixunssfdi)
+NOT_HERE_UNTIL_AFTER_4_3(__fixunssfsi)
+NOT_HERE_UNTIL_AFTER_4_3(__fixunssfsivfp)
+NOT_HERE_UNTIL_AFTER_4_3(__floatdidf)
+NOT_HERE_UNTIL_AFTER_4_3(__floatdisf)
+NOT_HERE_UNTIL_AFTER_4_3(__floatsidf)
+NOT_HERE_UNTIL_AFTER_4_3(__floatsidfvfp)
+NOT_HERE_UNTIL_AFTER_4_3(__floatsisf)
+NOT_HERE_UNTIL_AFTER_4_3(__floatsisfvfp)
+NOT_HERE_UNTIL_AFTER_4_3(__floatundidf)
+NOT_HERE_UNTIL_AFTER_4_3(__floatundisf)
+NOT_HERE_UNTIL_AFTER_4_3(__floatunsidf)
+NOT_HERE_UNTIL_AFTER_4_3(__floatunsisf)
+NOT_HERE_UNTIL_AFTER_4_3(__floatunssidfvfp)
+NOT_HERE_UNTIL_AFTER_4_3(__floatunssisfvfp)
+NOT_HERE_UNTIL_AFTER_4_3(__gedf2)
+NOT_HERE_UNTIL_AFTER_4_3(__gedf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__gesf2)
+NOT_HERE_UNTIL_AFTER_4_3(__gesf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__gtdf2)
+NOT_HERE_UNTIL_AFTER_4_3(__gtdf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__gtsf2)
+NOT_HERE_UNTIL_AFTER_4_3(__gtsf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__ledf2)
+NOT_HERE_UNTIL_AFTER_4_3(__ledf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__lesf2)
+NOT_HERE_UNTIL_AFTER_4_3(__lesf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__lshrdi3)
+NOT_HERE_UNTIL_AFTER_4_3(__ltdf2)
+NOT_HERE_UNTIL_AFTER_4_3(__ltdf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__ltsf2)
+NOT_HERE_UNTIL_AFTER_4_3(__ltsf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__moddi3)
+NOT_HERE_UNTIL_AFTER_4_3(__modsi3)
+NOT_HERE_UNTIL_AFTER_4_3(__muldc3)
+NOT_HERE_UNTIL_AFTER_4_3(__muldf3)
+NOT_HERE_UNTIL_AFTER_4_3(__muldf3vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__muldi3)
+NOT_HERE_UNTIL_AFTER_4_3(__mulsc3)
+NOT_HERE_UNTIL_AFTER_4_3(__mulsf3)
+NOT_HERE_UNTIL_AFTER_4_3(__mulsf3vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__mulvdi3)
+NOT_HERE_UNTIL_AFTER_4_3(__mulvsi3)
+NOT_HERE_UNTIL_AFTER_4_3(__nedf2)
+NOT_HERE_UNTIL_AFTER_4_3(__nedf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__negdi2)
+NOT_HERE_UNTIL_AFTER_4_3(__negvdi2)
+NOT_HERE_UNTIL_AFTER_4_3(__negvsi2)
+NOT_HERE_UNTIL_AFTER_4_3(__nesf2)
+NOT_HERE_UNTIL_AFTER_4_3(__nesf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__paritydi2)
+NOT_HERE_UNTIL_AFTER_4_3(__paritysi2)
+NOT_HERE_UNTIL_AFTER_4_3(__popcountdi2)
+NOT_HERE_UNTIL_AFTER_4_3(__popcountsi2)
+NOT_HERE_UNTIL_AFTER_4_3(__powidf2)
+NOT_HERE_UNTIL_AFTER_4_3(__powisf2)
+NOT_HERE_UNTIL_AFTER_4_3(__subdf3)
+NOT_HERE_UNTIL_AFTER_4_3(__subdf3vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__subsf3)
+NOT_HERE_UNTIL_AFTER_4_3(__subsf3vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__subvdi3)
+NOT_HERE_UNTIL_AFTER_4_3(__subvsi3)
+NOT_HERE_UNTIL_AFTER_4_3(__truncdfsf2)
+NOT_HERE_UNTIL_AFTER_4_3(__truncdfsf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__ucmpdi2)
+NOT_HERE_UNTIL_AFTER_4_3(__udivdi3)
+NOT_HERE_UNTIL_AFTER_4_3(__udivmoddi4)
+NOT_HERE_UNTIL_AFTER_4_3(__udivsi3)
+NOT_HERE_UNTIL_AFTER_4_3(__umoddi3)
+NOT_HERE_UNTIL_AFTER_4_3(__umodsi3)
+NOT_HERE_UNTIL_AFTER_4_3(__unorddf2)
+NOT_HERE_UNTIL_AFTER_4_3(__unorddf2vfp)
+NOT_HERE_UNTIL_AFTER_4_3(__unordsf2)
+NOT_HERE_UNTIL_AFTER_4_3(__unordsf2vfp)
+
+NOT_HERE_UNTIL_AFTER_4_3(__divmodsi4)
+NOT_HERE_UNTIL_AFTER_4_3(__udivmodsi4)
+#endif // __arm__ && __DYNAMIC__
+
+       
+       
+
+
+#else /* !__APPLE__ */
+
+extern int avoid_empty_file;
+
+#endif /* !__APPLE__*/
diff --git a/lib/builtins/arm/Makefile.mk b/lib/builtins/arm/Makefile.mk
new file mode 100644
index 000000000..ed2e8323e
--- /dev/null
+++ b/lib/builtins/arm/Makefile.mk
@@ -0,0 +1,20 @@
+#===- lib/builtins/arm/Makefile.mk -------------------------*- Makefile -*--===#
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+#===------------------------------------------------------------------------===#
+
+ModuleName := builtins
+SubDirs := 
+OnlyArchs := armv5 armv6 armv7 armv7k armv7m armv7em armv7s
+
+AsmSources := $(foreach file,$(wildcard $(Dir)/*.S),$(notdir $(file)))
+Sources := $(foreach file,$(wildcard $(Dir)/*.c),$(notdir $(file)))
+ObjNames := $(Sources:%.c=%.o) $(AsmSources:%.S=%.o)
+Implementation := Optimized
+
+# FIXME: use automatic dependencies?
+Dependencies := $(wildcard lib/*.h $(Dir)/*.h)
diff --git a/lib/builtins/arm/adddf3vfp.S b/lib/builtins/arm/adddf3vfp.S
new file mode 100644
index 000000000..4302b6ab3
--- /dev/null
+++ b/lib/builtins/arm/adddf3vfp.S
@@ -0,0 +1,26 @@
+//===-- adddf3vfp.S - Implement adddf3vfp ---------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// double __adddf3vfp(double a, double b) { return a + b; }
+//
+// Adds two double precision floating point numbers using the Darwin
+// calling convention where double arguments are passsed in GPR pairs
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__adddf3vfp)
+	vmov	d6, r0, r1		// move first param from r0/r1 pair into d6
+	vmov	d7, r2, r3		// move second param from r2/r3 pair into d7
+	vadd.f64 d6, d6, d7		
+	vmov	r0, r1, d6		// move result back to r0/r1 pair
+	bx	lr
+END_COMPILERRT_FUNCTION(__adddf3vfp)
diff --git a/lib/builtins/arm/addsf3vfp.S b/lib/builtins/arm/addsf3vfp.S
new file mode 100644
index 000000000..316328abe
--- /dev/null
+++ b/lib/builtins/arm/addsf3vfp.S
@@ -0,0 +1,26 @@
+//===-- addsf3vfp.S - Implement addsf3vfp ---------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern float __addsf3vfp(float a, float b);
+//
+// Adds two single precision floating point numbers using the Darwin
+// calling convention where single arguments are passsed in GPRs
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__addsf3vfp)
+	vmov	s14, r0		// move first param from r0 into float register
+	vmov	s15, r1		// move second param from r1 into float register
+	vadd.f32 s14, s14, s15
+	vmov	r0, s14		// move result back to r0
+	bx	lr
+END_COMPILERRT_FUNCTION(__addsf3vfp)
diff --git a/lib/builtins/arm/aeabi_dcmp.S b/lib/builtins/arm/aeabi_dcmp.S
new file mode 100644
index 000000000..ee2946d2d
--- /dev/null
+++ b/lib/builtins/arm/aeabi_dcmp.S
@@ -0,0 +1,40 @@
+//===-- aeabi_dcmp.S - EABI dcmp* implementation ---------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+// int __aeabi_dcmp{eq,lt,le,ge,gt}(double a, double b) {
+//   int result = __{eq,lt,le,ge,gt}df2(a, b);
+//   if (result {==,<,<=,>=,>} 0) {
+//     return 1;
+//   } else {
+//     return 0;
+//   }
+// }
+
+#define DEFINE_AEABI_DCMP(cond)                            \
+        .syntax unified                          SEPARATOR \
+        .align 2                                 SEPARATOR \
+DEFINE_COMPILERRT_FUNCTION(__aeabi_dcmp ## cond)           \
+        push      { r4, lr }                     SEPARATOR \
+        bl        SYMBOL_NAME(__ ## cond ## df2) SEPARATOR \
+        cmp       r0, #0                         SEPARATOR \
+        b ## cond 1f                             SEPARATOR \
+        mov       r0, #0                         SEPARATOR \
+        pop       { r4, pc }                     SEPARATOR \
+1:                                               SEPARATOR \
+        mov       r0, #1                         SEPARATOR \
+        pop       { r4, pc }                     SEPARATOR \
+END_COMPILERRT_FUNCTION(__aeabi_dcmp ## cond)
+
+DEFINE_AEABI_DCMP(eq)
+DEFINE_AEABI_DCMP(lt)
+DEFINE_AEABI_DCMP(le)
+DEFINE_AEABI_DCMP(ge)
+DEFINE_AEABI_DCMP(gt)
diff --git a/lib/builtins/arm/aeabi_fcmp.S b/lib/builtins/arm/aeabi_fcmp.S
new file mode 100644
index 000000000..ac3f54721
--- /dev/null
+++ b/lib/builtins/arm/aeabi_fcmp.S
@@ -0,0 +1,40 @@
+//===-- aeabi_fcmp.S - EABI fcmp* implementation ---------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+// int __aeabi_fcmp{eq,lt,le,ge,gt}(float a, float b) {
+//   int result = __{eq,lt,le,ge,gt}sf2(a, b);
+//   if (result {==,<,<=,>=,>} 0) {
+//     return 1;
+//   } else {
+//     return 0;
+//   }
+// }
+
+#define DEFINE_AEABI_FCMP(cond)                            \
+        .syntax unified                          SEPARATOR \
+        .align 2                                 SEPARATOR \
+DEFINE_COMPILERRT_FUNCTION(__aeabi_fcmp ## cond)           \
+        push      { r4, lr }                     SEPARATOR \
+        bl        SYMBOL_NAME(__ ## cond ## sf2) SEPARATOR \
+        cmp       r0, #0                         SEPARATOR \
+        b ## cond 1f                             SEPARATOR \
+        mov       r0, #0                         SEPARATOR \
+        pop       { r4, pc }                     SEPARATOR \
+1:                                               SEPARATOR \
+        mov       r0, #1                         SEPARATOR \
+        pop       { r4, pc }                     SEPARATOR \
+END_COMPILERRT_FUNCTION(__aeabi_fcmp ## cond)
+
+DEFINE_AEABI_FCMP(eq)
+DEFINE_AEABI_FCMP(lt)
+DEFINE_AEABI_FCMP(le)
+DEFINE_AEABI_FCMP(ge)
+DEFINE_AEABI_FCMP(gt)
diff --git a/lib/builtins/arm/aeabi_idivmod.S b/lib/builtins/arm/aeabi_idivmod.S
new file mode 100644
index 000000000..ac5dfcc22
--- /dev/null
+++ b/lib/builtins/arm/aeabi_idivmod.S
@@ -0,0 +1,28 @@
+//===-- aeabi_idivmod.S - EABI idivmod implementation ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+// struct { int quot, int rem} __aeabi_idivmod(int numerator, int denominator) {
+//   int rem, quot;
+//   quot = __divmodsi4(numerator, denominator, &rem);
+//   return {quot, rem};
+// }
+
+        .syntax unified
+        .align 2
+DEFINE_COMPILERRT_FUNCTION(__aeabi_idivmod)
+        push    { lr }
+        sub     sp, sp, #4
+        mov     r2, sp
+        bl      SYMBOL_NAME(__divmodsi4)
+        ldr     r1, [sp]
+        add     sp, sp, #4
+        pop     { pc }
+END_COMPILERRT_FUNCTION(__aeabi_idivmod)
diff --git a/lib/builtins/arm/aeabi_ldivmod.S b/lib/builtins/arm/aeabi_ldivmod.S
new file mode 100644
index 000000000..684252571
--- /dev/null
+++ b/lib/builtins/arm/aeabi_ldivmod.S
@@ -0,0 +1,31 @@
+//===-- aeabi_ldivmod.S - EABI ldivmod implementation ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+// struct { int64_t quot, int64_t rem}
+//        __aeabi_ldivmod(int64_t numerator, int64_t denominator) {
+//   int64_t rem, quot;
+//   quot = __divmoddi4(numerator, denominator, &rem);
+//   return {quot, rem};
+// }
+
+        .syntax unified
+        .align 2
+DEFINE_COMPILERRT_FUNCTION(__aeabi_ldivmod)
+        push    {r11, lr}
+        sub     sp, sp, #16
+        add     r12, sp, #8
+        str     r12, [sp]
+        bl      SYMBOL_NAME(__divmoddi4)
+        ldr     r2, [sp, #8]
+        ldr     r3, [sp, #12]
+        add     sp, sp, #16
+        pop     {r11, pc}
+END_COMPILERRT_FUNCTION(__aeabi_ldivmod)
diff --git a/lib/builtins/arm/aeabi_memcmp.S b/lib/builtins/arm/aeabi_memcmp.S
new file mode 100644
index 000000000..fc20825f4
--- /dev/null
+++ b/lib/builtins/arm/aeabi_memcmp.S
@@ -0,0 +1,20 @@
+//===-- aeabi_memcmp.S - EABI memcmp implementation -----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//  void __aeabi_memcmp(void *dest, void *src, size_t n) { memcmp(dest, src, n); }
+
+        .align 2
+DEFINE_COMPILERRT_FUNCTION(__aeabi_memcmp)
+        b       memcmp
+END_COMPILERRT_FUNCTION(__aeabi_memcmp)
+
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_memcmp4, __aeabi_memcmp)
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_memcmp8, __aeabi_memcmp)
diff --git a/lib/builtins/arm/aeabi_memcpy.S b/lib/builtins/arm/aeabi_memcpy.S
new file mode 100644
index 000000000..f6184632e
--- /dev/null
+++ b/lib/builtins/arm/aeabi_memcpy.S
@@ -0,0 +1,20 @@
+//===-- aeabi_memcpy.S - EABI memcpy implementation -----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//  void __aeabi_memcpy(void *dest, void *src, size_t n) { memcpy(dest, src, n); }
+
+        .align 2
+DEFINE_COMPILERRT_FUNCTION(__aeabi_memcpy)
+        b       memcpy
+END_COMPILERRT_FUNCTION(__aeabi_memcpy)
+
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_memcpy4, __aeabi_memcpy)
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_memcpy8, __aeabi_memcpy)
diff --git a/lib/builtins/arm/aeabi_memmove.S b/lib/builtins/arm/aeabi_memmove.S
new file mode 100644
index 000000000..f3d6b0e1a
--- /dev/null
+++ b/lib/builtins/arm/aeabi_memmove.S
@@ -0,0 +1,20 @@
+//===-- aeabi_memmove.S - EABI memmove implementation --------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===---------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//  void __aeabi_memmove(void *dest, void *src, size_t n) { memmove(dest, src, n); }
+
+        .align 2
+DEFINE_COMPILERRT_FUNCTION(__aeabi_memmove)
+        b       memmove
+END_COMPILERRT_FUNCTION(__aeabi_memmove)
+
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_memmove4, __aeabi_memmove)
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_memmove8, __aeabi_memmove)
diff --git a/lib/builtins/arm/aeabi_memset.S b/lib/builtins/arm/aeabi_memset.S
new file mode 100644
index 000000000..43c085e26
--- /dev/null
+++ b/lib/builtins/arm/aeabi_memset.S
@@ -0,0 +1,34 @@
+//===-- aeabi_memset.S - EABI memset implementation -----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//  void __aeabi_memset(void *dest, size_t n, int c) { memset(dest, c, n); }
+//  void __aeabi_memclr(void *dest, size_t n) { __aeabi_memset(dest, n, 0); }
+
+        .align 2
+DEFINE_COMPILERRT_FUNCTION(__aeabi_memset)
+        mov     r3, r1
+        mov     r1, r2
+        mov     r2, r3
+        b       memset
+END_COMPILERRT_FUNCTION(__aeabi_memset)
+
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_memset4, __aeabi_memset)
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_memset8, __aeabi_memset)
+
+DEFINE_COMPILERRT_FUNCTION(__aeabi_memclr)
+        mov     r2, r1
+        mov     r1, #0
+        b       memset
+END_COMPILERRT_FUNCTION(__aeabi_memclr)
+
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_memclr4, __aeabi_memclr)
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_memclr8, __aeabi_memclr)
+
diff --git a/lib/builtins/arm/aeabi_uidivmod.S b/lib/builtins/arm/aeabi_uidivmod.S
new file mode 100644
index 000000000..4c1e818cb
--- /dev/null
+++ b/lib/builtins/arm/aeabi_uidivmod.S
@@ -0,0 +1,29 @@
+//===-- aeabi_uidivmod.S - EABI uidivmod implementation -------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+// struct { unsigned quot, unsigned rem}
+//        __aeabi_uidivmod(unsigned numerator, unsigned denominator) {
+//   unsigned rem, quot;
+//   quot = __udivmodsi4(numerator, denominator, &rem);
+//   return {quot, rem};
+// }
+
+        .syntax unified
+        .align 2
+DEFINE_COMPILERRT_FUNCTION(__aeabi_uidivmod)
+        push    { lr }
+        sub     sp, sp, #4
+        mov     r2, sp
+        bl      SYMBOL_NAME(__udivmodsi4)
+        ldr     r1, [sp]
+        add     sp, sp, #4
+        pop     { pc }
+END_COMPILERRT_FUNCTION(__aeabi_uidivmod)
diff --git a/lib/builtins/arm/aeabi_uldivmod.S b/lib/builtins/arm/aeabi_uldivmod.S
new file mode 100644
index 000000000..62a2113cc
--- /dev/null
+++ b/lib/builtins/arm/aeabi_uldivmod.S
@@ -0,0 +1,31 @@
+//===-- aeabi_uldivmod.S - EABI uldivmod implementation -------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+// struct { uint64_t quot, uint64_t rem}
+//        __aeabi_uldivmod(uint64_t numerator, uint64_t denominator) {
+//   uint64_t rem, quot;
+//   quot = __udivmoddi4(numerator, denominator, &rem);
+//   return {quot, rem};
+// }
+
+        .syntax unified
+        .align 2
+DEFINE_COMPILERRT_FUNCTION(__aeabi_uldivmod)
+        push	{r11, lr}
+        sub	sp, sp, #16
+        add	r12, sp, #8
+        str	r12, [sp]
+        bl	SYMBOL_NAME(__udivmoddi4)
+        ldr	r2, [sp, #8]
+        ldr	r3, [sp, #12]
+        add	sp, sp, #16
+        pop	{r11, pc}
+END_COMPILERRT_FUNCTION(__aeabi_uldivmod)
diff --git a/lib/builtins/arm/bswapdi2.S b/lib/builtins/arm/bswapdi2.S
new file mode 100644
index 000000000..c6e1b79c6
--- /dev/null
+++ b/lib/builtins/arm/bswapdi2.S
@@ -0,0 +1,37 @@
+//===------- bswapdi2 - Implement bswapdi2 --------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern uint64_t __bswapdi2(uint64_t);
+//
+// Reverse all the bytes in a 64-bit integer.
+//
+.align 2
+DEFINE_COMPILERRT_FUNCTION(__bswapdi2)
+#if __ARM_ARCH < 6
+    // before armv6 does not have "rev" instruction
+    // r2 = rev(r0)
+    eor r2, r0, r0, ror #16
+    bic r2, r2, #0xff0000
+    mov r2, r2, lsr #8
+    eor r2, r2, r0, ror #8
+    // r0 = rev(r1)
+    eor r0, r1, r1, ror #16
+    bic r0, r0, #0xff0000
+    mov r0, r0, lsr #8
+    eor r0, r0, r1, ror #8
+#else
+    rev r2, r0  // r2 = rev(r0)
+    rev r0, r1  // r0 = rev(r1)
+#endif
+    mov r1, r2  // r1 = r2 = rev(r0)
+    JMP(lr)
+END_COMPILERRT_FUNCTION(__bswapdi2)
diff --git a/lib/builtins/arm/bswapsi2.S b/lib/builtins/arm/bswapsi2.S
new file mode 100644
index 000000000..5842da263
--- /dev/null
+++ b/lib/builtins/arm/bswapsi2.S
@@ -0,0 +1,29 @@
+//===------- bswapsi2 - Implement bswapsi2 --------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern uint32_t __bswapsi2(uint32_t);
+//
+// Reverse all the bytes in a 32-bit integer.
+//
+.align 2
+DEFINE_COMPILERRT_FUNCTION(__bswapsi2)
+#if __ARM_ARCH < 6
+    // before armv6 does not have "rev" instruction
+ 	eor	r1, r0, r0, ror #16
+ 	bic	r1, r1, #0xff0000
+ 	mov	r1, r1, lsr #8
+ 	eor	r0, r1, r0, ror #8
+#else
+    rev r0, r0
+#endif
+    JMP(lr)
+END_COMPILERRT_FUNCTION(__bswapsi2)
diff --git a/lib/builtins/arm/clzdi2.S b/lib/builtins/arm/clzdi2.S
new file mode 100644
index 000000000..33284cd3a
--- /dev/null
+++ b/lib/builtins/arm/clzdi2.S
@@ -0,0 +1,89 @@
+/* ===-- clzdi2.c - Implement __clzdi2 -------------------------------------===
+ *
+ *               The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements count leading zeros for 64bit arguments.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+#include "../assembly.h"
+
+	.syntax unified
+
+	.text
+	.align	2
+DEFINE_COMPILERRT_FUNCTION(__clzdi2)
+#ifdef __ARM_FEATURE_CLZ
+#ifdef __ARMEB__
+	cmp	r0, 0
+	itee ne
+	clzne	r0, r0
+	clzeq	r0, r1
+	addeq	r0, r0, 32
+#else
+	cmp	r1, 0
+	itee ne
+	clzne	r0, r1
+	clzeq	r0, r0
+	addeq	r0, r0, 32
+#endif
+	JMP(lr)
+#else
+	/* Assumption: n != 0 */
+
+	/*
+	 * r0: n
+	 * r1: upper half of n, overwritten after check
+	 * r1: count of leading zeros in n + 1
+	 * r2: scratch register for shifted r0
+	 */
+#ifdef __ARMEB__
+	cmp	r0, 0
+	moveq	r0, r1
+#else
+	cmp	r1, 0
+	movne	r0, r1
+#endif
+	movne	r1, 1
+	moveq	r1, 33
+
+	/*
+	 * Basic block:
+	 * if ((r0 >> SHIFT) == 0)
+	 *   r1 += SHIFT;
+	 * else
+	 *   r0 >>= SHIFT;
+	 * for descending powers of two as SHIFT.
+	 */
+#define BLOCK(shift) \
+	lsrs	r2, r0, shift; \
+	movne	r0, r2; \
+	addeq	r1, shift \
+
+	BLOCK(16)
+	BLOCK(8)
+	BLOCK(4)
+	BLOCK(2)
+
+	/*
+	 * The basic block invariants at this point are (r0 >> 2) == 0 and
+	 * r0 != 0. This means 1 <= r0 <= 3 and 0 <= (r0 >> 1) <= 1.
+	 *
+	 * r0 | (r0 >> 1) == 0 | (r0 >> 1) == 1 | -(r0 >> 1) | 1 - (r0 >> 1)
+	 * ---+----------------+----------------+------------+--------------
+	 * 1  | 1              | 0              | 0          | 1
+	 * 2  | 0              | 1              | -1         | 0
+	 * 3  | 0              | 1              | -1         | 0
+	 *
+	 * The r1's initial value of 1 compensates for the 1 here.
+	 */
+	sub	r0, r1, r0, lsr #1
+
+	JMP(lr)
+#endif // __ARM_FEATURE_CLZ
+END_COMPILERRT_FUNCTION(__clzdi2)
diff --git a/lib/builtins/arm/clzsi2.S b/lib/builtins/arm/clzsi2.S
new file mode 100644
index 000000000..8081d3284
--- /dev/null
+++ b/lib/builtins/arm/clzsi2.S
@@ -0,0 +1,69 @@
+/* ===-- clzsi2.c - Implement __clzsi2 -------------------------------------===
+ *
+ *               The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements count leading zeros for 32bit arguments.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+#include "../assembly.h"
+
+	.syntax unified
+
+	.text
+	.align	2
+DEFINE_COMPILERRT_FUNCTION(__clzsi2)
+#ifdef __ARM_FEATURE_CLZ
+	clz	r0, r0
+	JMP(lr)
+#else
+	/* Assumption: n != 0 */
+
+	/*
+	 * r0: n
+	 * r1: count of leading zeros in n + 1
+	 * r2: scratch register for shifted r0
+	 */
+	mov	r1, 1
+
+	/*
+	 * Basic block:
+	 * if ((r0 >> SHIFT) == 0)
+	 *   r1 += SHIFT;
+	 * else
+	 *   r0 >>= SHIFT;
+	 * for descending powers of two as SHIFT.
+	 */
+
+#define BLOCK(shift) \
+	lsrs	r2, r0, shift; \
+	movne	r0, r2; \
+	addeq	r1, shift \
+
+	BLOCK(16)
+	BLOCK(8)
+	BLOCK(4)
+	BLOCK(2)
+
+	/*
+	 * The basic block invariants at this point are (r0 >> 2) == 0 and
+	 * r0 != 0. This means 1 <= r0 <= 3 and 0 <= (r0 >> 1) <= 1.
+	 *
+	 * r0 | (r0 >> 1) == 0 | (r0 >> 1) == 1 | -(r0 >> 1) | 1 - (r0 >> 1)
+	 * ---+----------------+----------------+------------+--------------
+	 * 1  | 1              | 0              | 0          | 1
+	 * 2  | 0              | 1              | -1         | 0
+	 * 3  | 0              | 1              | -1         | 0
+	 *
+	 * The r1's initial value of 1 compensates for the 1 here.
+	 */
+	sub	r0, r1, r0, lsr #1
+
+	JMP(lr)
+#endif // __ARM_FEATURE_CLZ
+END_COMPILERRT_FUNCTION(__clzsi2)
diff --git a/lib/builtins/arm/comparesf2.S b/lib/builtins/arm/comparesf2.S
new file mode 100644
index 000000000..ad1b10a7f
--- /dev/null
+++ b/lib/builtins/arm/comparesf2.S
@@ -0,0 +1,148 @@
+//===-- comparesf2.S - Implement single-precision soft-float comparisons --===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the following soft-fp_t comparison routines:
+//
+//   __eqsf2   __gesf2   __unordsf2
+//   __lesf2   __gtsf2
+//   __ltsf2
+//   __nesf2
+//
+// The semantics of the routines grouped in each column are identical, so there
+// is a single implementation for each, with multiple names.
+//
+// The routines behave as follows:
+//
+//   __lesf2(a,b) returns -1 if a < b
+//                         0 if a == b
+//                         1 if a > b
+//                         1 if either a or b is NaN
+//
+//   __gesf2(a,b) returns -1 if a < b
+//                         0 if a == b
+//                         1 if a > b
+//                        -1 if either a or b is NaN
+//
+//   __unordsf2(a,b) returns 0 if both a and b are numbers
+//                           1 if either a or b is NaN
+//
+// Note that __lesf2( ) and __gesf2( ) are identical except in their handling of
+// NaN values.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+.syntax unified
+
+.align 2
+DEFINE_COMPILERRT_FUNCTION(__eqsf2)
+    // Make copies of a and b with the sign bit shifted off the top.  These will
+    // be used to detect zeros and NaNs.
+    mov     r2,         r0, lsl #1
+    mov     r3,         r1, lsl #1
+
+    // We do the comparison in three stages (ignoring NaN values for the time
+    // being).  First, we orr the absolute values of a and b; this sets the Z
+    // flag if both a and b are zero (of either sign).  The shift of r3 doesn't
+    // effect this at all, but it *does* make sure that the C flag is clear for
+    // the subsequent operations.
+    orrs    r12,    r2, r3, lsr #1
+
+    // Next, we check if a and b have the same or different signs.  If they have
+    // opposite signs, this eor will set the N flag.
+    it ne
+    eorsne  r12,    r0, r1
+
+    // If a and b are equal (either both zeros or bit identical; again, we're
+    // ignoring NaNs for now), this subtract will zero out r0.  If they have the
+    // same sign, the flags are updated as they would be for a comparison of the
+    // absolute values of a and b.
+    it pl
+    subspl  r0,     r2, r3
+
+    // If a is smaller in magnitude than b and both have the same sign, place
+    // the negation of the sign of b in r0.  Thus, if both are negative and
+    // a > b, this sets r0 to 0; if both are positive and a < b, this sets
+    // r0 to -1.
+    //
+    // This is also done if a and b have opposite signs and are not both zero,
+    // because in that case the subtract was not performed and the C flag is
+    // still clear from the shift argument in orrs; if a is positive and b
+    // negative, this places 0 in r0; if a is negative and b positive, -1 is
+    // placed in r0.
+    it lo
+    mvnlo   r0,         r1, asr #31
+
+    // If a is greater in magnitude than b and both have the same sign, place
+    // the sign of b in r0.  Thus, if both are negative and a < b, -1 is placed
+    // in r0, which is the desired result.  Conversely, if both are positive
+    // and a > b, zero is placed in r0.
+    it hi
+    movhi   r0,         r1, asr #31
+
+    // If you've been keeping track, at this point r0 contains -1 if a < b and
+    // 0 if a >= b.  All that remains to be done is to set it to 1 if a > b.
+    // If a == b, then the Z flag is set, so we can get the correct final value
+    // into r0 by simply or'ing with 1 if Z is clear.
+    it ne
+    orrne   r0,     r0, #1
+
+    // Finally, we need to deal with NaNs.  If either argument is NaN, replace
+    // the value in r0 with 1.
+    cmp     r2,         #0xff000000
+    ite ls
+    cmpls   r3,         #0xff000000
+    movhi   r0,         #1
+    JMP(lr)
+END_COMPILERRT_FUNCTION(__eqsf2)
+DEFINE_COMPILERRT_FUNCTION_ALIAS(__lesf2, __eqsf2)
+DEFINE_COMPILERRT_FUNCTION_ALIAS(__ltsf2, __eqsf2)
+DEFINE_COMPILERRT_FUNCTION_ALIAS(__nesf2, __eqsf2)
+
+.align 2
+DEFINE_COMPILERRT_FUNCTION(__gtsf2)
+    // Identical to the preceeding except in that we return -1 for NaN values.
+    // Given that the two paths share so much code, one might be tempted to 
+    // unify them; however, the extra code needed to do so makes the code size
+    // to performance tradeoff very hard to justify for such small functions.
+    mov     r2,         r0, lsl #1
+    mov     r3,         r1, lsl #1
+    orrs    r12,    r2, r3, lsr #1
+    it ne
+    eorsne  r12,    r0, r1
+    it pl
+    subspl  r0,     r2, r3
+    it lo
+    mvnlo   r0,         r1, asr #31
+    it hi
+    movhi   r0,         r1, asr #31
+    it ne
+    orrne   r0,     r0, #1
+    cmp     r2,         #0xff000000
+    ite ls
+    cmpls   r3,         #0xff000000
+    movhi   r0,         #-1
+    JMP(lr)
+END_COMPILERRT_FUNCTION(__gtsf2)
+DEFINE_COMPILERRT_FUNCTION_ALIAS(__gesf2, __gtsf2)
+
+.align 2
+DEFINE_COMPILERRT_FUNCTION(__unordsf2)
+    // Return 1 for NaN values, 0 otherwise.
+    mov     r2,         r0, lsl #1
+    mov     r3,         r1, lsl #1
+    mov     r0,         #0
+    cmp     r2,         #0xff000000
+    ite ls
+    cmpls   r3,         #0xff000000
+    movhi   r0,         #1
+    JMP(lr)
+END_COMPILERRT_FUNCTION(__unordsf2)
+
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_fcmpun, __unordsf2)
diff --git a/lib/builtins/arm/divdf3vfp.S b/lib/builtins/arm/divdf3vfp.S
new file mode 100644
index 000000000..ec2f99721
--- /dev/null
+++ b/lib/builtins/arm/divdf3vfp.S
@@ -0,0 +1,26 @@
+//===-- divdf3vfp.S - Implement divdf3vfp ---------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern double __divdf3vfp(double a, double b);
+//
+// Divides two double precision floating point numbers using the Darwin
+// calling convention where double arguments are passsed in GPR pairs
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__divdf3vfp)
+	vmov	d6, r0, r1		// move first param from r0/r1 pair into d6
+	vmov	d7, r2, r3		// move second param from r2/r3 pair into d7
+	vdiv.f64 d5, d6, d7		
+	vmov	r0, r1, d5		// move result back to r0/r1 pair
+	bx	lr
+END_COMPILERRT_FUNCTION(__divdf3vfp)
diff --git a/lib/builtins/arm/divmodsi4.S b/lib/builtins/arm/divmodsi4.S
new file mode 100644
index 000000000..24e3657dd
--- /dev/null
+++ b/lib/builtins/arm/divmodsi4.S
@@ -0,0 +1,61 @@
+/*===-- divmodsi4.S - 32-bit signed integer divide and modulus ------------===//
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ *===----------------------------------------------------------------------===//
+ *
+ * This file implements the __divmodsi4 (32-bit signed integer divide and
+ * modulus) function for the ARM architecture.  A naive digit-by-digit
+ * computation is employed for simplicity.
+ *
+ *===----------------------------------------------------------------------===*/
+
+#include "../assembly.h"
+
+#define ESTABLISH_FRAME    \
+    push   {r4-r7, lr}   ;\
+    add     r7,     sp, #12
+#define CLEAR_FRAME_AND_RETURN \
+    pop    {r4-r7, pc}
+
+.syntax unified
+.align 3
+DEFINE_COMPILERRT_FUNCTION(__divmodsi4)
+#if __ARM_ARCH_EXT_IDIV__
+	tst     r1, r1
+	beq     LOCAL_LABEL(divzero)
+	mov 	r3, r0
+	sdiv	r0, r3, r1
+	mls 	r1, r0, r1, r3
+	str 	r1, [r2]
+	bx  	lr
+LOCAL_LABEL(divzero):
+	mov     r0, #0
+	bx      lr
+#else
+    ESTABLISH_FRAME
+//  Set aside the sign of the quotient and modulus, and the address for the
+//  modulus.
+    eor     r4,     r0, r1
+    mov     r5,     r0
+    mov     r6,     r2
+//  Take the absolute value of a and b via abs(x) = (x^(x >> 31)) - (x >> 31).
+    eor     ip,     r0, r0, asr #31
+    eor     lr,     r1, r1, asr #31
+    sub     r0,     ip, r0, asr #31
+    sub     r1,     lr, r1, asr #31
+//  Unsigned divmod:
+    bl      SYMBOL_NAME(__udivmodsi4)
+//  Apply the sign of quotient and modulus
+    ldr     r1,    [r6]
+    eor     r0,     r0, r4, asr #31
+    eor     r1,     r1, r5, asr #31
+    sub     r0,     r0, r4, asr #31
+    sub     r1,     r1, r5, asr #31
+    str     r1,    [r6]
+    CLEAR_FRAME_AND_RETURN
+#endif
+END_COMPILERRT_FUNCTION(__divmodsi4)
diff --git a/lib/builtins/arm/divsf3vfp.S b/lib/builtins/arm/divsf3vfp.S
new file mode 100644
index 000000000..a4e2f8148
--- /dev/null
+++ b/lib/builtins/arm/divsf3vfp.S
@@ -0,0 +1,26 @@
+//===-- divsf3vfp.S - Implement divsf3vfp ---------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern float __divsf3vfp(float a, float b);
+//
+// Divides two single precision floating point numbers using the Darwin
+// calling convention where single arguments are passsed like 32-bit ints.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__divsf3vfp)
+	vmov	s14, r0		// move first param from r0 into float register
+	vmov	s15, r1		// move second param from r1 into float register
+	vdiv.f32 s13, s14, s15
+	vmov	r0, s13		// move result back to r0
+	bx	lr
+END_COMPILERRT_FUNCTION(__divsf3vfp)
diff --git a/lib/builtins/arm/divsi3.S b/lib/builtins/arm/divsi3.S
new file mode 100644
index 000000000..76b679352
--- /dev/null
+++ b/lib/builtins/arm/divsi3.S
@@ -0,0 +1,52 @@
+/*===-- divsi3.S - 32-bit signed integer divide ---------------------------===//
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ *===----------------------------------------------------------------------===//
+ *
+ * This file implements the __divsi3 (32-bit signed integer divide) function
+ * for the ARM architecture as a wrapper around the unsigned routine.
+ *
+ *===----------------------------------------------------------------------===*/
+
+#include "../assembly.h"
+
+#define ESTABLISH_FRAME \
+    push   {r4, r7, lr}    ;\
+    add     r7,     sp, #4
+#define CLEAR_FRAME_AND_RETURN \
+    pop    {r4, r7, pc}
+
+.syntax unified
+.align 3
+// Ok, APCS and AAPCS agree on 32 bit args, so it's safe to use the same routine.
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_idiv, __divsi3)
+DEFINE_COMPILERRT_FUNCTION(__divsi3)
+#if __ARM_ARCH_EXT_IDIV__
+   tst     r1,r1
+   beq     LOCAL_LABEL(divzero)
+   sdiv    r0, r0, r1
+   bx      lr
+LOCAL_LABEL(divzero):
+   mov     r0,#0
+   bx      lr
+#else
+ESTABLISH_FRAME
+//  Set aside the sign of the quotient.
+    eor     r4,     r0, r1
+//  Take absolute value of a and b via abs(x) = (x^(x >> 31)) - (x >> 31).
+    eor     r2,     r0, r0, asr #31
+    eor     r3,     r1, r1, asr #31
+    sub     r0,     r2, r0, asr #31
+    sub     r1,     r3, r1, asr #31
+//  abs(a) / abs(b)
+    bl      SYMBOL_NAME(__udivsi3)
+//  Apply sign of quotient to result and return.
+    eor     r0,     r0, r4, asr #31
+    sub     r0,     r0, r4, asr #31
+    CLEAR_FRAME_AND_RETURN
+#endif
+END_COMPILERRT_FUNCTION(__divsi3)
diff --git a/lib/builtins/arm/eqdf2vfp.S b/lib/builtins/arm/eqdf2vfp.S
new file mode 100644
index 000000000..198d42278
--- /dev/null
+++ b/lib/builtins/arm/eqdf2vfp.S
@@ -0,0 +1,29 @@
+//===-- eqdf2vfp.S - Implement eqdf2vfp -----------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern int __eqdf2vfp(double a, double b);
+//
+// Returns one iff a == b and neither is NaN.
+// Uses Darwin calling convention where double precision arguments are passsed 
+// like in GPR pairs.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__eqdf2vfp)
+	vmov	d6, r0, r1	// load r0/r1 pair in double register
+	vmov	d7, r2, r3	// load r2/r3 pair in double register
+	vcmp.f64 d6, d7		
+	vmrs	apsr_nzcv, fpscr
+	moveq	r0, #1		// set result register to 1 if equal
+	movne	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__eqdf2vfp)
diff --git a/lib/builtins/arm/eqsf2vfp.S b/lib/builtins/arm/eqsf2vfp.S
new file mode 100644
index 000000000..395187f47
--- /dev/null
+++ b/lib/builtins/arm/eqsf2vfp.S
@@ -0,0 +1,29 @@
+//===-- eqsf2vfp.S - Implement eqsf2vfp -----------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern int __eqsf2vfp(float a, float b);
+//
+// Returns one iff a == b and neither is NaN.
+// Uses Darwin calling convention where single precision arguments are passsed 
+// like 32-bit ints
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__eqsf2vfp)
+	vmov	s14, r0     // move from GPR 0 to float register
+	vmov	s15, r1	    // move from GPR 1 to float register
+	vcmp.f32 s14, s15
+	vmrs	apsr_nzcv, fpscr
+	moveq	r0, #1      // set result register to 1 if equal
+	movne	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__eqsf2vfp)
diff --git a/lib/builtins/arm/extendsfdf2vfp.S b/lib/builtins/arm/extendsfdf2vfp.S
new file mode 100644
index 000000000..ba31f3e0c
--- /dev/null
+++ b/lib/builtins/arm/extendsfdf2vfp.S
@@ -0,0 +1,26 @@
+//===-- extendsfdf2vfp.S - Implement extendsfdf2vfp -----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern double __extendsfdf2vfp(float a);
+//
+// Converts single precision float to double precision result.
+// Uses Darwin calling convention where a single precision parameter is 
+// passed in a GPR and a double precision result is returned in R0/R1 pair.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__extendsfdf2vfp)
+	vmov	s15, r0      // load float register from R0
+	vcvt.f64.f32 d7, s15 // convert single to double
+	vmov	r0, r1, d7   // return result in r0/r1 pair
+	bx	lr
+END_COMPILERRT_FUNCTION(__extendsfdf2vfp)
diff --git a/lib/builtins/arm/fixdfsivfp.S b/lib/builtins/arm/fixdfsivfp.S
new file mode 100644
index 000000000..26adbe389
--- /dev/null
+++ b/lib/builtins/arm/fixdfsivfp.S
@@ -0,0 +1,26 @@
+//===-- fixdfsivfp.S - Implement fixdfsivfp -----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern int __fixdfsivfp(double a);
+//
+// Converts double precision float to a 32-bit int rounding towards zero.
+// Uses Darwin calling convention where a double precision parameter is 
+// passed in GPR register pair.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__fixdfsivfp)
+	vmov	d7, r0, r1    // load double register from R0/R1
+	vcvt.s32.f64 s15, d7  // convert double to 32-bit int into s15
+	vmov	r0, s15	      // move s15 to result register
+	bx	lr
+END_COMPILERRT_FUNCTION(__fixdfsivfp)
diff --git a/lib/builtins/arm/fixsfsivfp.S b/lib/builtins/arm/fixsfsivfp.S
new file mode 100644
index 000000000..c1e0dadb7
--- /dev/null
+++ b/lib/builtins/arm/fixsfsivfp.S
@@ -0,0 +1,26 @@
+//===-- fixsfsivfp.S - Implement fixsfsivfp -----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern int __fixsfsivfp(float a);
+//
+// Converts single precision float to a 32-bit int rounding towards zero.
+// Uses Darwin calling convention where a single precision parameter is 
+// passed in a GPR..
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__fixsfsivfp)
+	vmov	s15, r0        // load float register from R0
+	vcvt.s32.f32 s15, s15  // convert single to 32-bit int into s15
+	vmov	r0, s15	       // move s15 to result register
+	bx	lr
+END_COMPILERRT_FUNCTION(__fixsfsivfp)
diff --git a/lib/builtins/arm/fixunsdfsivfp.S b/lib/builtins/arm/fixunsdfsivfp.S
new file mode 100644
index 000000000..fb91da752
--- /dev/null
+++ b/lib/builtins/arm/fixunsdfsivfp.S
@@ -0,0 +1,27 @@
+//===-- fixunsdfsivfp.S - Implement fixunsdfsivfp -------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern unsigned int __fixunsdfsivfp(double a);
+//
+// Converts double precision float to a 32-bit unsigned int rounding towards 
+// zero. All negative values become zero.
+// Uses Darwin calling convention where a double precision parameter is 
+// passed in GPR register pair.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__fixunsdfsivfp)
+	vmov	d7, r0, r1    // load double register from R0/R1
+	vcvt.u32.f64 s15, d7  // convert double to 32-bit int into s15
+	vmov	r0, s15	      // move s15 to result register
+	bx	lr
+END_COMPILERRT_FUNCTION(__fixunsdfsivfp)
diff --git a/lib/builtins/arm/fixunssfsivfp.S b/lib/builtins/arm/fixunssfsivfp.S
new file mode 100644
index 000000000..f12dff61a
--- /dev/null
+++ b/lib/builtins/arm/fixunssfsivfp.S
@@ -0,0 +1,27 @@
+//===-- fixunssfsivfp.S - Implement fixunssfsivfp -------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern unsigned int __fixunssfsivfp(float a);
+//
+// Converts single precision float to a 32-bit unsigned int rounding towards 
+// zero. All negative values become zero.
+// Uses Darwin calling convention where a single precision parameter is 
+// passed in a GPR..
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__fixunssfsivfp)
+	vmov	s15, r0        // load float register from R0
+	vcvt.u32.f32 s15, s15  // convert single to 32-bit unsigned into s15
+	vmov	r0, s15	       // move s15 to result register
+	bx	lr
+END_COMPILERRT_FUNCTION(__fixunssfsivfp)
diff --git a/lib/builtins/arm/floatsidfvfp.S b/lib/builtins/arm/floatsidfvfp.S
new file mode 100644
index 000000000..ab48933b9
--- /dev/null
+++ b/lib/builtins/arm/floatsidfvfp.S
@@ -0,0 +1,26 @@
+//===-- floatsidfvfp.S - Implement floatsidfvfp ---------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern double __floatsidfvfp(int a);
+//
+// Converts a 32-bit int to a double precision float.
+// Uses Darwin calling convention where a double precision result is 
+// return in GPR register pair.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__floatsidfvfp)
+	vmov	s15, r0        // move int to float register s15
+	vcvt.f64.s32 d7, s15   // convert 32-bit int in s15 to double in d7
+	vmov	r0, r1, d7     // move d7 to result register pair r0/r1
+	bx	lr
+END_COMPILERRT_FUNCTION(__floatsidfvfp)
diff --git a/lib/builtins/arm/floatsisfvfp.S b/lib/builtins/arm/floatsisfvfp.S
new file mode 100644
index 000000000..eb265e990
--- /dev/null
+++ b/lib/builtins/arm/floatsisfvfp.S
@@ -0,0 +1,26 @@
+//===-- floatsisfvfp.S - Implement floatsisfvfp ---------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern float __floatsisfvfp(int a);
+//
+// Converts single precision float to a 32-bit int rounding towards zero.
+// Uses Darwin calling convention where a single precision result is 
+// return in a GPR..
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__floatsisfvfp)
+	vmov	s15, r0	       // move int to float register s15
+	vcvt.f32.s32 s15, s15  // convert 32-bit int in s15 to float in s15
+	vmov	r0, s15        // move s15 to result register
+	bx	lr
+END_COMPILERRT_FUNCTION(__floatsisfvfp)
diff --git a/lib/builtins/arm/floatunssidfvfp.S b/lib/builtins/arm/floatunssidfvfp.S
new file mode 100644
index 000000000..96e2eb21e
--- /dev/null
+++ b/lib/builtins/arm/floatunssidfvfp.S
@@ -0,0 +1,26 @@
+//===-- floatunssidfvfp.S - Implement floatunssidfvfp ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern double __floatunssidfvfp(unsigned int a);
+//
+// Converts a 32-bit int to a double precision float.
+// Uses Darwin calling convention where a double precision result is 
+// return in GPR register pair.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__floatunssidfvfp)
+	vmov	s15, r0        // move int to float register s15
+	vcvt.f64.u32 d7, s15   // convert 32-bit int in s15 to double in d7
+	vmov	r0, r1, d7     // move d7 to result register pair r0/r1
+	bx	lr
+END_COMPILERRT_FUNCTION(__floatunssidfvfp)
diff --git a/lib/builtins/arm/floatunssisfvfp.S b/lib/builtins/arm/floatunssisfvfp.S
new file mode 100644
index 000000000..05471c4a5
--- /dev/null
+++ b/lib/builtins/arm/floatunssisfvfp.S
@@ -0,0 +1,26 @@
+//===-- floatunssisfvfp.S - Implement floatunssisfvfp ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern float __floatunssisfvfp(unsigned int a);
+//
+// Converts single precision float to a 32-bit int rounding towards zero.
+// Uses Darwin calling convention where a single precision result is 
+// return in a GPR..
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__floatunssisfvfp)
+	vmov	s15, r0	       // move int to float register s15
+	vcvt.f32.u32 s15, s15  // convert 32-bit int in s15 to float in s15
+	vmov	r0, s15        // move s15 to result register
+	bx	lr
+END_COMPILERRT_FUNCTION(__floatunssisfvfp)
diff --git a/lib/builtins/arm/gedf2vfp.S b/lib/builtins/arm/gedf2vfp.S
new file mode 100644
index 000000000..aa636aa14
--- /dev/null
+++ b/lib/builtins/arm/gedf2vfp.S
@@ -0,0 +1,29 @@
+//===-- gedf2vfp.S - Implement gedf2vfp -----------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern int __gedf2vfp(double a, double b);
+//
+// Returns one iff a >= b and neither is NaN.
+// Uses Darwin calling convention where double precision arguments are passsed 
+// like in GPR pairs.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__gedf2vfp)
+	vmov 	d6, r0, r1	// load r0/r1 pair in double register
+	vmov 	d7, r2, r3	// load r2/r3 pair in double register
+	vcmp.f64 d6, d7
+	vmrs	apsr_nzcv, fpscr
+	movge	r0, #1      // set result register to 1 if greater than or equal
+	movlt	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__gedf2vfp)
diff --git a/lib/builtins/arm/gesf2vfp.S b/lib/builtins/arm/gesf2vfp.S
new file mode 100644
index 000000000..087ce15f1
--- /dev/null
+++ b/lib/builtins/arm/gesf2vfp.S
@@ -0,0 +1,29 @@
+//===-- gesf2vfp.S - Implement gesf2vfp -----------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern int __gesf2vfp(float a, float b);
+//
+// Returns one iff a >= b and neither is NaN.
+// Uses Darwin calling convention where single precision arguments are passsed 
+// like 32-bit ints
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__gesf2vfp)
+	vmov	s14, r0	    // move from GPR 0 to float register
+	vmov	s15, r1	    // move from GPR 1 to float register
+	vcmp.f32 s14, s15
+	vmrs	apsr_nzcv, fpscr
+	movge	r0, #1      // set result register to 1 if greater than or equal
+	movlt	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__gesf2vfp)
diff --git a/lib/builtins/arm/gtdf2vfp.S b/lib/builtins/arm/gtdf2vfp.S
new file mode 100644
index 000000000..230952a9f
--- /dev/null
+++ b/lib/builtins/arm/gtdf2vfp.S
@@ -0,0 +1,29 @@
+//===-- gtdf2vfp.S - Implement gtdf2vfp -----------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern double __gtdf2vfp(double a, double b);
+//
+// Returns one iff a > b and neither is NaN.
+// Uses Darwin calling convention where double precision arguments are passsed 
+// like in GPR pairs.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__gtdf2vfp)
+	vmov 	d6, r0, r1	// load r0/r1 pair in double register
+	vmov 	d7, r2, r3	// load r2/r3 pair in double register
+	vcmp.f64 d6, d7
+	vmrs	apsr_nzcv, fpscr
+	movgt	r0, #1		// set result register to 1 if equal
+	movle	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__gtdf2vfp)
diff --git a/lib/builtins/arm/gtsf2vfp.S b/lib/builtins/arm/gtsf2vfp.S
new file mode 100644
index 000000000..c1f5db053
--- /dev/null
+++ b/lib/builtins/arm/gtsf2vfp.S
@@ -0,0 +1,29 @@
+//===-- gtsf2vfp.S - Implement gtsf2vfp -----------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern int __gtsf2vfp(float a, float b);
+//
+// Returns one iff a > b and neither is NaN.
+// Uses Darwin calling convention where single precision arguments are passsed 
+// like 32-bit ints
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__gtsf2vfp)
+	vmov	s14, r0		// move from GPR 0 to float register
+	vmov	s15, r1		// move from GPR 1 to float register
+	vcmp.f32 s14, s15
+	vmrs	apsr_nzcv, fpscr
+	movgt	r0, #1		// set result register to 1 if equal
+	movle	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__gtsf2vfp)
diff --git a/lib/builtins/arm/ledf2vfp.S b/lib/builtins/arm/ledf2vfp.S
new file mode 100644
index 000000000..94980a2ef
--- /dev/null
+++ b/lib/builtins/arm/ledf2vfp.S
@@ -0,0 +1,29 @@
+//===-- ledf2vfp.S - Implement ledf2vfp -----------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern double __ledf2vfp(double a, double b);
+//
+// Returns one iff a <= b and neither is NaN.
+// Uses Darwin calling convention where double precision arguments are passsed 
+// like in GPR pairs.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__ledf2vfp)
+	vmov 	d6, r0, r1	// load r0/r1 pair in double register
+	vmov 	d7, r2, r3	// load r2/r3 pair in double register
+	vcmp.f64 d6, d7
+	vmrs	apsr_nzcv, fpscr
+	movls	r0, #1		// set result register to 1 if equal
+	movhi	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__ledf2vfp)
diff --git a/lib/builtins/arm/lesf2vfp.S b/lib/builtins/arm/lesf2vfp.S
new file mode 100644
index 000000000..362352a83
--- /dev/null
+++ b/lib/builtins/arm/lesf2vfp.S
@@ -0,0 +1,29 @@
+//===-- lesf2vfp.S - Implement lesf2vfp -----------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern int __lesf2vfp(float a, float b);
+//
+// Returns one iff a <= b and neither is NaN.
+// Uses Darwin calling convention where single precision arguments are passsed 
+// like 32-bit ints
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__lesf2vfp)
+	vmov	s14, r0     // move from GPR 0 to float register
+	vmov	s15, r1     // move from GPR 1 to float register
+	vcmp.f32 s14, s15
+	vmrs	apsr_nzcv, fpscr
+	movls	r0, #1      // set result register to 1 if equal
+	movhi	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__lesf2vfp)
diff --git a/lib/builtins/arm/ltdf2vfp.S b/lib/builtins/arm/ltdf2vfp.S
new file mode 100644
index 000000000..391ff2924
--- /dev/null
+++ b/lib/builtins/arm/ltdf2vfp.S
@@ -0,0 +1,29 @@
+//===-- ltdf2vfp.S - Implement ltdf2vfp -----------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern double __ltdf2vfp(double a, double b);
+//
+// Returns one iff a < b and neither is NaN.
+// Uses Darwin calling convention where double precision arguments are passsed 
+// like in GPR pairs.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__ltdf2vfp)
+	vmov 	d6, r0, r1	// load r0/r1 pair in double register
+	vmov 	d7, r2, r3	// load r2/r3 pair in double register
+	vcmp.f64 d6, d7
+	vmrs	apsr_nzcv, fpscr
+	movmi	r0, #1		// set result register to 1 if equal
+	movpl	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__ltdf2vfp)
diff --git a/lib/builtins/arm/ltsf2vfp.S b/lib/builtins/arm/ltsf2vfp.S
new file mode 100644
index 000000000..dcaa03561
--- /dev/null
+++ b/lib/builtins/arm/ltsf2vfp.S
@@ -0,0 +1,29 @@
+//===-- ltsf2vfp.S - Implement ltsf2vfp -----------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern int __ltsf2vfp(float a, float b);
+//
+// Returns one iff a < b and neither is NaN.
+// Uses Darwin calling convention where single precision arguments are passsed 
+// like 32-bit ints
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__ltsf2vfp)
+	vmov	s14, r0     // move from GPR 0 to float register
+	vmov	s15, r1     // move from GPR 1 to float register
+	vcmp.f32 s14, s15
+	vmrs	apsr_nzcv, fpscr
+	movmi	r0, #1      // set result register to 1 if equal
+	movpl	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__ltsf2vfp)
diff --git a/lib/builtins/arm/modsi3.S b/lib/builtins/arm/modsi3.S
new file mode 100644
index 000000000..07c5f9381
--- /dev/null
+++ b/lib/builtins/arm/modsi3.S
@@ -0,0 +1,51 @@
+/*===-- modsi3.S - 32-bit signed integer modulus --------------------------===//
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ *===----------------------------------------------------------------------===//
+ *
+ * This file implements the __modsi3 (32-bit signed integer modulus) function
+ * for the ARM architecture as a wrapper around the unsigned routine.
+ *
+ *===----------------------------------------------------------------------===*/
+
+#include "../assembly.h"
+
+#define ESTABLISH_FRAME \
+    push   {r4, r7, lr}    ;\
+    add     r7,     sp, #4
+#define CLEAR_FRAME_AND_RETURN \
+    pop    {r4, r7, pc}
+
+.syntax unified
+.align 3
+DEFINE_COMPILERRT_FUNCTION(__modsi3)
+#if __ARM_ARCH_EXT_IDIV__
+	tst     r1, r1
+	beq     LOCAL_LABEL(divzero)
+	sdiv	r2, r0, r1
+	mls 	r0, r2, r1, r0
+	bx      lr
+LOCAL_LABEL(divzero):
+	mov     r0, #0
+	bx      lr
+#else
+    ESTABLISH_FRAME
+    //  Set aside the sign of the dividend.
+    mov     r4,     r0
+    //  Take absolute value of a and b via abs(x) = (x^(x >> 31)) - (x >> 31).
+    eor     r2,     r0, r0, asr #31
+    eor     r3,     r1, r1, asr #31
+    sub     r0,     r2, r0, asr #31
+    sub     r1,     r3, r1, asr #31
+    //  abs(a) % abs(b)
+    bl     SYMBOL_NAME(__umodsi3)
+    //  Apply sign of dividend to result and return.
+    eor     r0,     r0, r4, asr #31
+    sub     r0,     r0, r4, asr #31
+    CLEAR_FRAME_AND_RETURN
+#endif
+END_COMPILERRT_FUNCTION(__modsi3)
diff --git a/lib/builtins/arm/muldf3vfp.S b/lib/builtins/arm/muldf3vfp.S
new file mode 100644
index 000000000..2c1d70377
--- /dev/null
+++ b/lib/builtins/arm/muldf3vfp.S
@@ -0,0 +1,26 @@
+//===-- muldf3vfp.S - Implement muldf3vfp ---------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern double __muldf3vfp(double a, double b);
+//
+// Multiplies two double precision floating point numbers using the Darwin
+// calling convention where double arguments are passsed in GPR pairs
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__muldf3vfp)
+	vmov 	d6, r0, r1         // move first param from r0/r1 pair into d6
+	vmov 	d7, r2, r3         // move second param from r2/r3 pair into d7
+	vmul.f64 d6, d6, d7		
+	vmov 	r0, r1, d6         // move result back to r0/r1 pair
+	bx	lr
+END_COMPILERRT_FUNCTION(__muldf3vfp)
diff --git a/lib/builtins/arm/mulsf3vfp.S b/lib/builtins/arm/mulsf3vfp.S
new file mode 100644
index 000000000..98af81d23
--- /dev/null
+++ b/lib/builtins/arm/mulsf3vfp.S
@@ -0,0 +1,26 @@
+//===-- mulsf3vfp.S - Implement mulsf3vfp ---------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern float __mulsf3vfp(float a, float b);
+//
+// Multiplies two single precision floating point numbers using the Darwin
+// calling convention where single arguments are passsed like 32-bit ints.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__mulsf3vfp)
+	vmov	s14, r0		// move first param from r0 into float register
+	vmov	s15, r1		// move second param from r1 into float register
+	vmul.f32 s13, s14, s15
+	vmov	r0, s13		// move result back to r0
+	bx	lr
+END_COMPILERRT_FUNCTION(__mulsf3vfp)
diff --git a/lib/builtins/arm/nedf2vfp.S b/lib/builtins/arm/nedf2vfp.S
new file mode 100644
index 000000000..c1a832e1a
--- /dev/null
+++ b/lib/builtins/arm/nedf2vfp.S
@@ -0,0 +1,29 @@
+//===-- nedf2vfp.S - Implement nedf2vfp -----------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern double __nedf2vfp(double a, double b);
+//
+// Returns zero if a and b are unequal and neither is NaN.
+// Uses Darwin calling convention where double precision arguments are passsed 
+// like in GPR pairs.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__nedf2vfp)
+	vmov 	d6, r0, r1	// load r0/r1 pair in double register
+	vmov 	d7, r2, r3	// load r2/r3 pair in double register
+	vcmp.f64 d6, d7		
+	vmrs	apsr_nzcv, fpscr
+	movne	r0, #1		// set result register to 0 if unequal
+	moveq	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__nedf2vfp)
diff --git a/lib/builtins/arm/negdf2vfp.S b/lib/builtins/arm/negdf2vfp.S
new file mode 100644
index 000000000..ad9b87418
--- /dev/null
+++ b/lib/builtins/arm/negdf2vfp.S
@@ -0,0 +1,23 @@
+//===-- negdf2vfp.S - Implement negdf2vfp ---------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern double __negdf2vfp(double a, double b);
+//
+// Returns the negation a double precision floating point numbers using the 
+// Darwin calling convention where double arguments are passsed in GPR pairs.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__negdf2vfp)
+	eor	r1, r1, #-2147483648	// flip sign bit on double in r0/r1 pair
+	bx	lr
+END_COMPILERRT_FUNCTION(__negdf2vfp)
diff --git a/lib/builtins/arm/negsf2vfp.S b/lib/builtins/arm/negsf2vfp.S
new file mode 100644
index 000000000..482eb2640
--- /dev/null
+++ b/lib/builtins/arm/negsf2vfp.S
@@ -0,0 +1,23 @@
+//===-- negsf2vfp.S - Implement negsf2vfp ---------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern float __negsf2vfp(float a);
+//
+// Returns the negation of a single precision floating point numbers using the 
+// Darwin calling convention where single arguments are passsed like 32-bit ints
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__negsf2vfp)
+	eor	r0, r0, #-2147483648	// flip sign bit on float in r0
+	bx	lr
+END_COMPILERRT_FUNCTION(__negsf2vfp)
diff --git a/lib/builtins/arm/nesf2vfp.S b/lib/builtins/arm/nesf2vfp.S
new file mode 100644
index 000000000..127145ec4
--- /dev/null
+++ b/lib/builtins/arm/nesf2vfp.S
@@ -0,0 +1,29 @@
+//===-- nesf2vfp.S - Implement nesf2vfp -----------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern int __nesf2vfp(float a, float b);
+//
+// Returns one iff a != b and neither is NaN.
+// Uses Darwin calling convention where single precision arguments are passsed 
+// like 32-bit ints
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__nesf2vfp)
+	vmov	s14, r0	    // move from GPR 0 to float register
+	vmov	s15, r1	    // move from GPR 1 to float register
+	vcmp.f32 s14, s15
+	vmrs	apsr_nzcv, fpscr
+	movne	r0, #1      // set result register to 1 if unequal
+	moveq	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__nesf2vfp)
diff --git a/lib/builtins/arm/restore_vfp_d8_d15_regs.S b/lib/builtins/arm/restore_vfp_d8_d15_regs.S
new file mode 100644
index 000000000..5d5547415
--- /dev/null
+++ b/lib/builtins/arm/restore_vfp_d8_d15_regs.S
@@ -0,0 +1,35 @@
+//===-- save_restore_regs.S - Implement save/restore* ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// When compiling C++ functions that need to handle thrown exceptions the
+// compiler is required to save all registers and call __Unwind_SjLj_Register
+// in the function prolog.  But when compiling for thumb1, there are
+// no instructions to access the floating point registers, so the
+// compiler needs to add a call to the helper function _save_vfp_d8_d15_regs
+// written in ARM to save the float registers.  In the epilog, the compiler
+// must also add a call to __restore_vfp_d8_d15_regs to restore those registers.
+//
+
+	.text
+	.syntax unified
+
+//
+// Restore registers d8-d15 from stack
+//
+	.align 2
+DEFINE_COMPILERRT_PRIVATE_FUNCTION(__restore_vfp_d8_d15_regs)
+	vldmia	sp!, {d8-d15}           // pop registers d8-d15 off stack
+	bx      lr                      // return to prolog
+END_COMPILERRT_FUNCTION(__restore_vfp_d8_d15_regs)
+
+	// tell linker it can break up file at label boundaries
+	.subsections_via_symbols
diff --git a/lib/builtins/arm/save_vfp_d8_d15_regs.S b/lib/builtins/arm/save_vfp_d8_d15_regs.S
new file mode 100644
index 000000000..4be3ff3a5
--- /dev/null
+++ b/lib/builtins/arm/save_vfp_d8_d15_regs.S
@@ -0,0 +1,35 @@
+//===-- save_restore_regs.S - Implement save/restore* ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// When compiling C++ functions that need to handle thrown exceptions the
+// compiler is required to save all registers and call __Unwind_SjLj_Register
+// in the function prolog.  But when compiling for thumb1, there are
+// no instructions to access the floating point registers, so the
+// compiler needs to add a call to the helper function _save_vfp_d8_d15_regs
+// written in ARM to save the float registers.  In the epilog, the compiler
+// must also add a call to __restore_vfp_d8_d15_regs to restore those registers.
+//
+
+	.text
+	.syntax unified
+
+//
+// Save registers d8-d15 onto stack
+//
+	.align 2
+DEFINE_COMPILERRT_PRIVATE_FUNCTION(__save_vfp_d8_d15_regs)
+	vstmdb	sp!, {d8-d15}           // push registers d8-d15 onto stack
+	bx      lr                      // return to prolog
+END_COMPILERRT_FUNCTION(__save_vfp_d8_d15_regs)
+
+	// tell linker it can break up file at label boundaries
+	.subsections_via_symbols
diff --git a/lib/builtins/arm/softfloat-alias.list b/lib/builtins/arm/softfloat-alias.list
new file mode 100644
index 000000000..cc6a4b3cd
--- /dev/null
+++ b/lib/builtins/arm/softfloat-alias.list
@@ -0,0 +1,21 @@
+#
+# These are soft float functions which can be 
+# aliased to the *vfp functions on arm processors
+# that support floating point instructions.
+#
+___adddf3vfp		___adddf3
+___addsf3vfp		___addsf3
+___divdf3vfp		___divdf3
+___divsf3vfp		___divsf3
+___extendsfdf2vfp	___extendsfdf2
+___fixdfsivfp		___fixdfsi
+___fixsfsivfp		___fixsfsi
+___floatsidfvfp		___floatsidf
+___floatsisfvfp		___floatsisf
+___muldf3vfp		___muldf3
+___mulsf3vfp		___mulsf3
+___subdf3vfp		___subdf3
+___subsf3vfp		___subsf3
+___truncdfsf2vfp	___truncdfsf2
+___floatunssidfvfp	___floatunsidf
+___floatunssisfvfp	___floatunsisf
diff --git a/lib/builtins/arm/subdf3vfp.S b/lib/builtins/arm/subdf3vfp.S
new file mode 100644
index 000000000..747d6596c
--- /dev/null
+++ b/lib/builtins/arm/subdf3vfp.S
@@ -0,0 +1,26 @@
+//===-- subdf3vfp.S - Implement subdf3vfp ---------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern double __subdf3vfp(double a, double b);
+//
+// Returns difference between two double precision floating point numbers using 
+// the Darwin calling convention where double arguments are passsed in GPR pairs
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__subdf3vfp)
+	vmov 	d6, r0, r1         // move first param from r0/r1 pair into d6
+	vmov 	d7, r2, r3         // move second param from r2/r3 pair into d7
+	vsub.f64 d6, d6, d7		
+	vmov 	r0, r1, d6         // move result back to r0/r1 pair
+	bx	lr
+END_COMPILERRT_FUNCTION(__subdf3vfp)
diff --git a/lib/builtins/arm/subsf3vfp.S b/lib/builtins/arm/subsf3vfp.S
new file mode 100644
index 000000000..7cc63cccc
--- /dev/null
+++ b/lib/builtins/arm/subsf3vfp.S
@@ -0,0 +1,27 @@
+//===-- subsf3vfp.S - Implement subsf3vfp ---------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern float __subsf3vfp(float a, float b);
+//
+// Returns the difference between two single precision floating point numbers 
+// using the Darwin calling convention where single arguments are passsed
+// like 32-bit ints.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__subsf3vfp)
+	vmov	s14, r0		// move first param from r0 into float register
+	vmov	s15, r1		// move second param from r1 into float register
+	vsub.f32 s14, s14, s15
+	vmov	r0, s14		// move result back to r0
+	bx	lr
+END_COMPILERRT_FUNCTION(__subsf3vfp)
diff --git a/lib/builtins/arm/switch16.S b/lib/builtins/arm/switch16.S
new file mode 100644
index 000000000..0dedc598a
--- /dev/null
+++ b/lib/builtins/arm/switch16.S
@@ -0,0 +1,46 @@
+//===-- switch.S - Implement switch* --------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// When compiling switch statements in thumb mode, the compiler
+// can use these __switch* helper functions  The compiler emits a blx to
+// the __switch* function followed by a table of displacements for each
+// case statement.  On entry, R0 is the index into the table. The __switch*
+// function uses the return address in lr to find the start of the table.
+// The first entry in the table is the count of the entries in the table.
+// It then uses R0 to index into the table and get the displacement of the
+// address to jump to.  If R0 is greater than the size of the table, it jumps
+// to the last entry in the table. Each displacement in the table is actually
+// the distance from lr to the label, thus making the tables PIC.
+
+
+	.text
+	.syntax unified
+
+//
+// The table contains signed 2-byte sized elements which are 1/2 the distance
+// from lr to the target label.
+//
+	.align 2
+DEFINE_COMPILERRT_PRIVATE_FUNCTION(__switch16)
+	ldrh    ip, [lr, #-1]           // get first 16-bit word in table
+	cmp     r0, ip                  // compare with index
+	add     r0, lr, r0, lsl #1      // compute address of element in table
+	add     ip, lr, ip, lsl #1      // compute address of last element in table
+	ite lo
+	ldrshlo r0, [r0, #1]            // load 16-bit element if r0 is in range
+	ldrshhs r0, [ip, #1]            // load 16-bit element if r0 out of range
+	add     ip, lr, r0, lsl #1      // compute label = lr + element*2
+	bx      ip                      // jump to computed label
+END_COMPILERRT_FUNCTION(__switch16)
+
+	// tell linker it can break up file at label boundaries
+	.subsections_via_symbols
diff --git a/lib/builtins/arm/switch32.S b/lib/builtins/arm/switch32.S
new file mode 100644
index 000000000..64d558eb8
--- /dev/null
+++ b/lib/builtins/arm/switch32.S
@@ -0,0 +1,47 @@
+//===-- switch.S - Implement switch* --------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// When compiling switch statements in thumb mode, the compiler
+// can use these __switch* helper functions  The compiler emits a blx to
+// the __switch* function followed by a table of displacements for each
+// case statement.  On entry, R0 is the index into the table. The __switch*
+// function uses the return address in lr to find the start of the table.
+// The first entry in the table is the count of the entries in the table.
+// It then uses R0 to index into the table and get the displacement of the
+// address to jump to.  If R0 is greater than the size of the table, it jumps
+// to the last entry in the table. Each displacement in the table is actually
+// the distance from lr to the label, thus making the tables PIC.
+
+
+	.text
+	.syntax unified
+
+//
+// The table contains signed 4-byte sized elements which are the distance
+// from lr to the target label.
+//
+	.align 2
+DEFINE_COMPILERRT_PRIVATE_FUNCTION(__switch32)
+	ldr     ip, [lr, #-1]            // get first 32-bit word in table
+	cmp     r0, ip                   // compare with index
+	add     r0, lr, r0, lsl #2       // compute address of element in table
+	add     ip, lr, ip, lsl #2       // compute address of last element in table
+	ite lo
+	ldrlo   r0, [r0, #3]             // load 32-bit element if r0 is in range
+	ldrhs   r0, [ip, #3]             // load 32-bit element if r0 out of range
+	add     ip, lr, r0               // compute label = lr + element
+	bx      ip                       // jump to computed label
+END_COMPILERRT_FUNCTION(__switch32)
+
+	// tell linker it can break up file at label boundaries
+	.subsections_via_symbols
+
diff --git a/lib/builtins/arm/switch8.S b/lib/builtins/arm/switch8.S
new file mode 100644
index 000000000..b5008849a
--- /dev/null
+++ b/lib/builtins/arm/switch8.S
@@ -0,0 +1,45 @@
+//===-- switch.S - Implement switch* --------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// When compiling switch statements in thumb mode, the compiler
+// can use these __switch* helper functions  The compiler emits a blx to
+// the __switch* function followed by a table of displacements for each
+// case statement.  On entry, R0 is the index into the table. The __switch*
+// function uses the return address in lr to find the start of the table.
+// The first entry in the table is the count of the entries in the table.
+// It then uses R0 to index into the table and get the displacement of the
+// address to jump to.  If R0 is greater than the size of the table, it jumps
+// to the last entry in the table. Each displacement in the table is actually
+// the distance from lr to the label, thus making the tables PIC.
+
+
+	.text
+	.syntax unified
+
+//
+// The table contains signed byte sized elements which are 1/2 the distance
+// from lr to the target label.
+//
+	.align 2
+DEFINE_COMPILERRT_PRIVATE_FUNCTION(__switch8)
+	ldrb    ip, [lr, #-1]           // get first byte in table
+	cmp     r0, ip                  // signed compare with index
+	ite lo
+	ldrsblo r0, [lr, r0]            // get indexed byte out of table
+	ldrsbhs r0, [lr, ip]            // if out of range, use last entry in table
+	add     ip, lr, r0, lsl #1      // compute label = lr + element*2
+	bx      ip                      // jump to computed label
+END_COMPILERRT_FUNCTION(__switch8)
+
+	// tell linker it can break up file at label boundaries
+	.subsections_via_symbols
+
diff --git a/lib/builtins/arm/switchu8.S b/lib/builtins/arm/switchu8.S
new file mode 100644
index 000000000..488d4e74e
--- /dev/null
+++ b/lib/builtins/arm/switchu8.S
@@ -0,0 +1,45 @@
+//===-- switch.S - Implement switch* --------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// When compiling switch statements in thumb mode, the compiler
+// can use these __switch* helper functions  The compiler emits a blx to
+// the __switch* function followed by a table of displacements for each
+// case statement.  On entry, R0 is the index into the table. The __switch*
+// function uses the return address in lr to find the start of the table.
+// The first entry in the table is the count of the entries in the table.
+// It then uses R0 to index into the table and get the displacement of the
+// address to jump to.  If R0 is greater than the size of the table, it jumps
+// to the last entry in the table. Each displacement in the table is actually
+// the distance from lr to the label, thus making the tables PIC.
+
+
+	.text
+	.syntax unified
+
+//
+// The table contains unsigned byte sized elements which are 1/2 the distance
+// from lr to the target label.
+//
+	.align 2
+DEFINE_COMPILERRT_PRIVATE_FUNCTION(__switchu8)
+	ldrb    ip, [lr, #-1]           // get first byte in table
+	cmp     r0, ip                  // compare with index
+	ite lo
+	ldrblo  r0, [lr, r0]            // get indexed byte out of table
+	ldrbhs  r0, [lr, ip]            // if out of range, use last entry in table
+	add     ip, lr, r0, lsl #1      // compute label = lr + element*2
+	bx      ip                      // jump to computed label
+END_COMPILERRT_FUNCTION(__switchu8)
+
+	// tell linker it can break up file at label boundaries
+	.subsections_via_symbols
+
diff --git a/lib/builtins/arm/sync_synchronize.S b/lib/builtins/arm/sync_synchronize.S
new file mode 100644
index 000000000..aa18f04fb
--- /dev/null
+++ b/lib/builtins/arm/sync_synchronize.S
@@ -0,0 +1,35 @@
+//===-- sync_synchronize - Implement memory barrier * ----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// When compiling a use of the gcc built-in __sync_synchronize() in thumb1 mode
+// the compiler may emit a call to __sync_synchronize.  
+// On Darwin the implementation jumps to an OS supplied function named 
+// OSMemoryBarrier
+//
+
+	.text
+	.syntax unified
+
+#if __APPLE__
+
+	.align 2
+DEFINE_COMPILERRT_PRIVATE_FUNCTION(__sync_synchronize)
+	stmfd	sp!, {r7, lr}
+	add		r7, sp, #0
+	bl		_OSMemoryBarrier
+	ldmfd	sp!, {r7, pc}
+END_COMPILERRT_FUNCTION(__sync_synchronize)
+
+	// tell linker it can break up file at label boundaries
+	.subsections_via_symbols
+		
+#endif
diff --git a/lib/builtins/arm/truncdfsf2vfp.S b/lib/builtins/arm/truncdfsf2vfp.S
new file mode 100644
index 000000000..55bf324cc
--- /dev/null
+++ b/lib/builtins/arm/truncdfsf2vfp.S
@@ -0,0 +1,26 @@
+//===-- truncdfsf2vfp.S - Implement truncdfsf2vfp -------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern float __truncdfsf2vfp(double a);
+//
+// Converts double precision float to signle precision result.
+// Uses Darwin calling convention where a double precision parameter is 
+// passed in a R0/R1 pair and a signle precision result is returned in R0.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__truncdfsf2vfp)
+	vmov 	d7, r0, r1   // load double from r0/r1 pair
+	vcvt.f32.f64 s15, d7 // convert double to single (trucate precision)
+	vmov 	r0, s15      // return result in r0
+	bx	lr
+END_COMPILERRT_FUNCTION(__truncdfsf2vfp)
diff --git a/lib/builtins/arm/udivmodsi4.S b/lib/builtins/arm/udivmodsi4.S
new file mode 100644
index 000000000..bb5d29c03
--- /dev/null
+++ b/lib/builtins/arm/udivmodsi4.S
@@ -0,0 +1,161 @@
+/*===-- udivmodsi4.S - 32-bit unsigned integer divide and modulus ---------===//
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ *===----------------------------------------------------------------------===//
+ *
+ * This file implements the __udivmodsi4 (32-bit unsigned integer divide and
+ * modulus) function for the ARM 32-bit architecture.
+ *
+ *===----------------------------------------------------------------------===*/
+
+#include "../assembly.h"
+
+	.syntax unified
+
+	.text
+	.p2align 2
+DEFINE_COMPILERRT_FUNCTION(__udivmodsi4)
+#if __ARM_ARCH_EXT_IDIV__
+	tst     r1, r1
+	beq     LOCAL_LABEL(divby0)
+	mov 	r3, r0
+	udiv	r0, r3, r1
+	mls 	r1, r0, r1, r3
+	str 	r1, [r2]
+	bx  	lr
+#else
+	cmp	r1, #1
+	bcc	LOCAL_LABEL(divby0)
+	beq	LOCAL_LABEL(divby1)
+	cmp	r0, r1
+	bcc	LOCAL_LABEL(quotient0)
+	/*
+	 * Implement division using binary long division algorithm.
+	 *
+	 * r0 is the numerator, r1 the denominator.
+	 *
+	 * The code before JMP computes the correct shift I, so that
+	 * r0 and (r1 << I) have the highest bit set in the same position.
+	 * At the time of JMP, ip := .Ldiv0block - 12 * I.
+	 * This depends on the fixed instruction size of block.
+	 *
+	 * block(shift) implements the test-and-update-quotient core.
+	 * It assumes (r0 << shift) can be computed without overflow and
+	 * that (r0 << shift) < 2 * r1. The quotient is stored in r3.
+	 */
+
+#  ifdef __ARM_FEATURE_CLZ
+	clz	ip, r0
+	clz	r3, r1
+	/* r0 >= r1 implies clz(r0) <= clz(r1), so ip <= r3. */
+	sub	r3, r3, ip
+	adr	ip, LOCAL_LABEL(div0block)
+	sub	ip, ip, r3, lsl #2
+	sub	ip, ip, r3, lsl #3
+	mov	r3, #0
+	bx	ip
+#  else
+	str	r4, [sp, #-8]!
+
+	mov	r4, r0
+	adr	ip, LOCAL_LABEL(div0block)
+
+	lsr	r3, r4, #16
+	cmp	r3, r1
+	movhs	r4, r3
+	subhs	ip, ip, #(16 * 12)
+
+	lsr	r3, r4, #8
+	cmp	r3, r1
+	movhs	r4, r3
+	subhs	ip, ip, #(8 * 12)
+
+	lsr	r3, r4, #4
+	cmp	r3, r1
+	movhs	r4, r3
+	subhs	ip, #(4 * 12)
+
+	lsr	r3, r4, #2
+	cmp	r3, r1
+	movhs	r4, r3
+	subhs	ip, ip, #(2 * 12)
+
+	/* Last block, no need to update r3 or r4. */
+	cmp	r1, r4, lsr #1
+	subls	ip, ip, #(1 * 12)
+
+	ldr	r4, [sp], #8	/* restore r4, we are done with it. */
+	mov	r3, #0
+
+	JMP(ip)
+#  endif
+
+#define	IMM	#
+
+#define block(shift) \
+	cmp	r0, r1, lsl IMM shift; \
+	addhs	r3, r3, IMM (1 << shift); \
+	subhs	r0, r0, r1, lsl IMM shift
+
+	block(31)
+	block(30)
+	block(29)
+	block(28)
+	block(27)
+	block(26)
+	block(25)
+	block(24)
+	block(23)
+	block(22)
+	block(21)
+	block(20)
+	block(19)
+	block(18)
+	block(17)
+	block(16)
+	block(15)
+	block(14)
+	block(13)
+	block(12)
+	block(11)
+	block(10)
+	block(9)
+	block(8)
+	block(7)
+	block(6)
+	block(5)
+	block(4)
+	block(3)
+	block(2)
+	block(1)
+LOCAL_LABEL(div0block):
+	block(0)
+
+	str	r0, [r2]
+	mov	r0, r3
+	JMP(lr)
+
+LOCAL_LABEL(quotient0):
+	str	r0, [r2]
+	mov	r0, #0
+	JMP(lr)
+
+LOCAL_LABEL(divby1):
+	mov	r3, #0
+	str	r3, [r2]
+	JMP(lr)
+#endif /* __ARM_ARCH_EXT_IDIV__ */
+
+LOCAL_LABEL(divby0):
+	mov	r0, #0
+#ifdef __ARM_EABI__
+	b	__aeabi_idiv0
+#else
+	JMP(lr)
+#endif
+
+END_COMPILERRT_FUNCTION(__udivmodsi4)
diff --git a/lib/builtins/arm/udivsi3.S b/lib/builtins/arm/udivsi3.S
new file mode 100644
index 000000000..11c1c096e
--- /dev/null
+++ b/lib/builtins/arm/udivsi3.S
@@ -0,0 +1,148 @@
+/*===-- udivsi3.S - 32-bit unsigned integer divide ------------------------===//
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ *===----------------------------------------------------------------------===//
+ *
+ * This file implements the __udivsi3 (32-bit unsigned integer divide)
+ * function for the ARM 32-bit architecture.
+ *
+ *===----------------------------------------------------------------------===*/
+
+#include "../assembly.h"
+
+	.syntax unified
+
+	.text
+	.p2align 2
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_uidiv, __udivsi3)
+DEFINE_COMPILERRT_FUNCTION(__udivsi3)
+#if __ARM_ARCH_EXT_IDIV__
+	tst     r1, r1
+	beq     LOCAL_LABEL(divby0)
+	mov 	r3, r0
+	udiv	r0, r3, r1
+	mls 	r1, r0, r1, r3
+	bx  	lr
+#else
+	cmp	r1, #1
+	bcc	LOCAL_LABEL(divby0)
+	JMPc(lr, eq)
+	cmp	r0, r1
+	movcc	r0, #0
+	JMPc(lr, cc)
+	/*
+	 * Implement division using binary long division algorithm.
+	 *
+	 * r0 is the numerator, r1 the denominator.
+	 *
+	 * The code before JMP computes the correct shift I, so that
+	 * r0 and (r1 << I) have the highest bit set in the same position.
+	 * At the time of JMP, ip := .Ldiv0block - 12 * I.
+	 * This depends on the fixed instruction size of block.
+	 *
+	 * block(shift) implements the test-and-update-quotient core.
+	 * It assumes (r0 << shift) can be computed without overflow and
+	 * that (r0 << shift) < 2 * r1. The quotient is stored in r3.
+	 */
+
+#  ifdef __ARM_FEATURE_CLZ
+	clz	ip, r0
+	clz	r3, r1
+	/* r0 >= r1 implies clz(r0) <= clz(r1), so ip <= r3. */
+	sub	r3, r3, ip
+	adr	ip, LOCAL_LABEL(div0block)
+	sub	ip, ip, r3, lsl #2
+	sub	ip, ip, r3, lsl #3
+	mov	r3, #0
+	bx	ip
+#  else
+	mov	r2, r0
+	adr	ip, LOCAL_LABEL(div0block)
+
+	lsr	r3, r2, #16
+	cmp	r3, r1
+	movhs	r2, r3
+	subhs	ip, ip, #(16 * 12)
+
+	lsr	r3, r2, #8
+	cmp	r3, r1
+	movhs	r2, r3
+	subhs	ip, ip, #(8 * 12)
+
+	lsr	r3, r2, #4
+	cmp	r3, r1
+	movhs	r2, r3
+	subhs	ip, #(4 * 12)
+
+	lsr	r3, r2, #2
+	cmp	r3, r1
+	movhs	r2, r3
+	subhs	ip, ip, #(2 * 12)
+
+	/* Last block, no need to update r2 or r3. */
+	cmp	r1, r2, lsr #1
+	subls	ip, ip, #(1 * 12)
+
+	mov	r3, #0
+
+	JMP(ip)
+#  endif
+
+#define	IMM	#
+
+#define block(shift) \
+	cmp	r0, r1, lsl IMM shift; \
+	addhs	r3, r3, IMM (1 << shift); \
+	subhs	r0, r0, r1, lsl IMM shift
+
+	block(31)
+	block(30)
+	block(29)
+	block(28)
+	block(27)
+	block(26)
+	block(25)
+	block(24)
+	block(23)
+	block(22)
+	block(21)
+	block(20)
+	block(19)
+	block(18)
+	block(17)
+	block(16)
+	block(15)
+	block(14)
+	block(13)
+	block(12)
+	block(11)
+	block(10)
+	block(9)
+	block(8)
+	block(7)
+	block(6)
+	block(5)
+	block(4)
+	block(3)
+	block(2)
+	block(1)
+LOCAL_LABEL(div0block):
+	block(0)
+
+	mov	r0, r3
+	JMP(lr)
+#endif /* __ARM_ARCH_EXT_IDIV__ */
+
+LOCAL_LABEL(divby0):
+	mov	r0, #0
+#ifdef __ARM_EABI__
+	b	__aeabi_idiv0
+#else
+	JMP(lr)
+#endif
+
+END_COMPILERRT_FUNCTION(__udivsi3)
diff --git a/lib/builtins/arm/umodsi3.S b/lib/builtins/arm/umodsi3.S
new file mode 100644
index 000000000..a03afefda
--- /dev/null
+++ b/lib/builtins/arm/umodsi3.S
@@ -0,0 +1,141 @@
+/*===-- umodsi3.S - 32-bit unsigned integer modulus -----------------------===//
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ *===----------------------------------------------------------------------===//
+ *
+ * This file implements the __umodsi3 (32-bit unsigned integer modulus)
+ * function for the ARM 32-bit architecture.
+ *
+ *===----------------------------------------------------------------------===*/
+
+#include "../assembly.h"
+
+	.syntax unified
+
+	.text
+	.p2align 2
+DEFINE_COMPILERRT_FUNCTION(__umodsi3)
+#if __ARM_ARCH_EXT_IDIV__
+	tst     r1, r1
+	beq     LOCAL_LABEL(divby0)
+	mov 	r3, r0
+	udiv	r0, r3, r1
+	mls 	r1, r0, r1, r3
+	str 	r1, [r2]
+	bx  	lr
+#else
+	cmp	r1, #1
+	bcc	LOCAL_LABEL(divby0)
+	moveq	r0, #0
+	JMPc(lr, eq)
+	cmp	r0, r1
+	JMPc(lr, cc)
+	/*
+	 * Implement division using binary long division algorithm.
+	 *
+	 * r0 is the numerator, r1 the denominator.
+	 *
+	 * The code before JMP computes the correct shift I, so that
+	 * r0 and (r1 << I) have the highest bit set in the same position.
+	 * At the time of JMP, ip := .Ldiv0block - 8 * I.
+	 * This depends on the fixed instruction size of block.
+	 *
+	 * block(shift) implements the test-and-update-quotient core.
+	 * It assumes (r0 << shift) can be computed without overflow and
+	 * that (r0 << shift) < 2 * r1. The quotient is stored in r3.
+	 */
+
+#  ifdef __ARM_FEATURE_CLZ
+	clz	ip, r0
+	clz	r3, r1
+	/* r0 >= r1 implies clz(r0) <= clz(r1), so ip <= r3. */
+	sub	r3, r3, ip
+	adr	ip, LOCAL_LABEL(div0block)
+	sub	ip, ip, r3, lsl #3
+	bx	ip
+#  else
+	mov	r2, r0
+	adr	ip, LOCAL_LABEL(div0block)
+
+	lsr	r3, r2, #16
+	cmp	r3, r1
+	movhs	r2, r3
+	subhs	ip, ip, #(16 * 8)
+
+	lsr	r3, r2, #8
+	cmp	r3, r1
+	movhs	r2, r3
+	subhs	ip, ip, #(8 * 8)
+
+	lsr	r3, r2, #4
+	cmp	r3, r1
+	movhs	r2, r3
+	subhs	ip, #(4 * 8)
+
+	lsr	r3, r2, #2
+	cmp	r3, r1
+	movhs	r2, r3
+	subhs	ip, ip, #(2 * 8)
+
+	/* Last block, no need to update r2 or r3. */
+	cmp	r1, r2, lsr #1
+	subls	ip, ip, #(1 * 8)
+
+	JMP(ip)
+#  endif
+
+#define	IMM	#
+
+#define block(shift) \
+	cmp	r0, r1, lsl IMM shift; \
+	subhs	r0, r0, r1, lsl IMM shift
+
+	block(31)
+	block(30)
+	block(29)
+	block(28)
+	block(27)
+	block(26)
+	block(25)
+	block(24)
+	block(23)
+	block(22)
+	block(21)
+	block(20)
+	block(19)
+	block(18)
+	block(17)
+	block(16)
+	block(15)
+	block(14)
+	block(13)
+	block(12)
+	block(11)
+	block(10)
+	block(9)
+	block(8)
+	block(7)
+	block(6)
+	block(5)
+	block(4)
+	block(3)
+	block(2)
+	block(1)
+LOCAL_LABEL(div0block):
+	block(0)
+	JMP(lr)
+#endif /* __ARM_ARCH_EXT_IDIV__ */
+
+LOCAL_LABEL(divby0):
+	mov	r0, #0
+#ifdef __ARM_EABI__
+	b	__aeabi_idiv0
+#else
+	JMP(lr)
+#endif
+
+END_COMPILERRT_FUNCTION(__umodsi3)
diff --git a/lib/builtins/arm/unorddf2vfp.S b/lib/builtins/arm/unorddf2vfp.S
new file mode 100644
index 000000000..ee7de4025
--- /dev/null
+++ b/lib/builtins/arm/unorddf2vfp.S
@@ -0,0 +1,29 @@
+//===-- unorddf2vfp.S - Implement unorddf2vfp ------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern int __unorddf2vfp(double a, double b);
+//
+// Returns one iff a or b is NaN
+// Uses Darwin calling convention where double precision arguments are passsed 
+// like in GPR pairs.
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__unorddf2vfp)
+	vmov 	d6, r0, r1	// load r0/r1 pair in double register
+	vmov 	d7, r2, r3	// load r2/r3 pair in double register
+	vcmp.f64 d6, d7		
+	vmrs	apsr_nzcv, fpscr
+	movvs	r0, #1      // set result register to 1 if "overflow" (any NaNs)
+	movvc	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__unorddf2vfp)
diff --git a/lib/builtins/arm/unordsf2vfp.S b/lib/builtins/arm/unordsf2vfp.S
new file mode 100644
index 000000000..eb419a541
--- /dev/null
+++ b/lib/builtins/arm/unordsf2vfp.S
@@ -0,0 +1,29 @@
+//===-- unordsf2vfp.S - Implement unordsf2vfp -----------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// extern int __unordsf2vfp(float a, float b);
+//
+// Returns one iff a or b is NaN
+// Uses Darwin calling convention where single precision arguments are passsed 
+// like 32-bit ints
+//
+	.syntax unified
+	.align 2
+DEFINE_COMPILERRT_FUNCTION(__unordsf2vfp)
+	vmov	s14, r0     // move from GPR 0 to float register
+	vmov	s15, r1	    // move from GPR 1 to float register
+	vcmp.f32 s14, s15
+	vmrs	apsr_nzcv, fpscr
+	movvs	r0, #1      // set result register to 1 if "overflow" (any NaNs)
+	movvc	r0, #0
+	bx	lr
+END_COMPILERRT_FUNCTION(__unordsf2vfp)
diff --git a/lib/builtins/ashldi3.c b/lib/builtins/ashldi3.c
new file mode 100644
index 000000000..eb4698ac5
--- /dev/null
+++ b/lib/builtins/ashldi3.c
@@ -0,0 +1,43 @@
+/* ====-- ashldi3.c - Implement __ashldi3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __ashldi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a << b */
+
+/* Precondition:  0 <= b < bits_in_dword */
+
+ARM_EABI_FNALIAS(llsl, ashldi3)
+
+COMPILER_RT_ABI di_int
+__ashldi3(di_int a, si_int b)
+{
+    const int bits_in_word = (int)(sizeof(si_int) * CHAR_BIT);
+    dwords input;
+    dwords result;
+    input.all = a;
+    if (b & bits_in_word)  /* bits_in_word <= b < bits_in_dword */
+    {
+        result.s.low = 0;
+        result.s.high = input.s.low << (b - bits_in_word);
+    }
+    else  /* 0 <= b < bits_in_word */
+    {
+        if (b == 0)
+            return a;
+        result.s.low  = input.s.low << b;
+        result.s.high = (input.s.high << b) | (input.s.low >> (bits_in_word - b));
+    }
+    return result.all;
+}
diff --git a/lib/builtins/ashlti3.c b/lib/builtins/ashlti3.c
new file mode 100644
index 000000000..4bd8219b5
--- /dev/null
+++ b/lib/builtins/ashlti3.c
@@ -0,0 +1,45 @@
+/* ===-- ashlti3.c - Implement __ashlti3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __ashlti3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: a << b */
+
+/* Precondition:  0 <= b < bits_in_tword */
+
+ti_int
+__ashlti3(ti_int a, si_int b)
+{
+    const int bits_in_dword = (int)(sizeof(di_int) * CHAR_BIT);
+    twords input;
+    twords result;
+    input.all = a;
+    if (b & bits_in_dword)  /* bits_in_dword <= b < bits_in_tword */
+    {
+        result.s.low = 0;
+        result.s.high = input.s.low << (b - bits_in_dword);
+    }
+    else  /* 0 <= b < bits_in_dword */
+    {
+        if (b == 0)
+            return a;
+        result.s.low  = input.s.low << b;
+        result.s.high = (input.s.high << b) | (input.s.low >> (bits_in_dword - b));
+    }
+    return result.all;
+}
+
+#endif /* __x86_64 */
diff --git a/lib/builtins/ashrdi3.c b/lib/builtins/ashrdi3.c
new file mode 100644
index 000000000..14c878bb7
--- /dev/null
+++ b/lib/builtins/ashrdi3.c
@@ -0,0 +1,44 @@
+/*===-- ashrdi3.c - Implement __ashrdi3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __ashrdi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: arithmetic a >> b */
+
+/* Precondition:  0 <= b < bits_in_dword */
+
+ARM_EABI_FNALIAS(lasr, ashrdi3)
+
+COMPILER_RT_ABI di_int
+__ashrdi3(di_int a, si_int b)
+{
+    const int bits_in_word = (int)(sizeof(si_int) * CHAR_BIT);
+    dwords input;
+    dwords result;
+    input.all = a;
+    if (b & bits_in_word)  /* bits_in_word <= b < bits_in_dword */
+    {
+        /* result.s.high = input.s.high < 0 ? -1 : 0 */
+        result.s.high = input.s.high >> (bits_in_word - 1);
+        result.s.low = input.s.high >> (b - bits_in_word);
+    }
+    else  /* 0 <= b < bits_in_word */
+    {
+        if (b == 0)
+            return a;
+        result.s.high  = input.s.high >> b;
+        result.s.low = (input.s.high << (bits_in_word - b)) | (input.s.low >> b);
+    }
+    return result.all;
+}
diff --git a/lib/builtins/ashrti3.c b/lib/builtins/ashrti3.c
new file mode 100644
index 000000000..ed43641c7
--- /dev/null
+++ b/lib/builtins/ashrti3.c
@@ -0,0 +1,46 @@
+/* ===-- ashrti3.c - Implement __ashrti3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __ashrti3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: arithmetic a >> b */
+
+/* Precondition:  0 <= b < bits_in_tword */
+
+ti_int
+__ashrti3(ti_int a, si_int b)
+{
+    const int bits_in_dword = (int)(sizeof(di_int) * CHAR_BIT);
+    twords input;
+    twords result;
+    input.all = a;
+    if (b & bits_in_dword)  /* bits_in_dword <= b < bits_in_tword */
+    {
+        /* result.s.high = input.s.high < 0 ? -1 : 0 */
+        result.s.high = input.s.high >> (bits_in_dword - 1);
+        result.s.low = input.s.high >> (b - bits_in_dword);
+    }
+    else  /* 0 <= b < bits_in_dword */
+    {
+        if (b == 0)
+            return a;
+        result.s.high  = input.s.high >> b;
+        result.s.low = (input.s.high << (bits_in_dword - b)) | (input.s.low >> b);
+    }
+    return result.all;
+}
+
+#endif /* __x86_64 */
diff --git a/lib/builtins/assembly.h b/lib/builtins/assembly.h
new file mode 100644
index 000000000..1a9f79cb6
--- /dev/null
+++ b/lib/builtins/assembly.h
@@ -0,0 +1,136 @@
+/* ===-- assembly.h - compiler-rt assembler support macros -----------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file defines macros for use in compiler-rt assembler source.
+ * This file is not part of the interface of this library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#ifndef COMPILERRT_ASSEMBLY_H
+#define COMPILERRT_ASSEMBLY_H
+
+#if defined(__POWERPC__) || defined(__powerpc__) || defined(__ppc__)
+#define SEPARATOR @
+#else
+#define SEPARATOR ;
+#endif
+
+#if defined(__APPLE__)
+#define HIDDEN_DIRECTIVE .private_extern
+#define LOCAL_LABEL(name) L_##name
+#define FILE_LEVEL_DIRECTIVE  .subsections_via_symbols
+#define SYMBOL_IS_FUNC(name)
+#else
+#define HIDDEN_DIRECTIVE .hidden
+#define LOCAL_LABEL(name) .L_##name
+#define FILE_LEVEL_DIRECTIVE
+#  if defined(__arm__)
+#  define SYMBOL_IS_FUNC(name) .type name, %function
+#  else
+#  define SYMBOL_IS_FUNC(name) .type name, @function
+#  endif
+#endif
+
+#if defined(__arm__)
+# ifndef __ARM_ARCH
+#  if defined (__ARM_ARCH_7__) || defined (__ARM_ARCH_7A__) || \
+     defined (__ARM_ARCH_7R__) || defined (__ARM_ARCH_7M__) || \
+     defined (__ARM_ARCH_7EM__)
+#  define __ARM_ARCH 7
+#  endif
+# endif
+
+# ifndef __ARM_ARCH
+#  if defined (__ARM_ARCH_6__) || \
+      defined (__ARM_ARCH_6J__) || defined (__ARM_ARCH_6K__) || \
+      defined (__ARM_ARCH_6Z__) || defined (__ARM_ARCH_6ZK__) || \
+      defined (__ARM_ARCH_6ZM__)
+#  define __ARM_ARCH 6
+#  endif
+# endif
+
+# ifndef __ARM_ARCH
+#  if defined (__ARM_ARCH_5__) || defined (__ARM_ARCH_5T__) || \
+      defined (__ARM_ARCH_5TE__) || defined (__ARM_ARCH_5TEJ__)
+#  define __ARM_ARCH 5
+#  endif
+# endif
+
+# ifndef __ARM_ARCH
+# define __ARM_ARCH 4
+# endif
+
+# if defined(__ARM_ARCH_4T__) || __ARM_ARCH >= 5
+# define ARM_HAS_BX
+# endif
+# if !defined(__ARM_FEATURE_CLZ) && \
+     (__ARM_ARCH >= 6 || (__ARM_ARCH == 5 && !defined(__ARM_ARCH_5__)))
+# define __ARM_FEATURE_CLZ
+# endif
+
+# ifdef ARM_HAS_BX
+# define JMP(r)		bx	r
+# define JMPc(r,c)	bx##c	r
+# else
+# define JMP(r)		mov	pc, r
+# define JMPc(r,c)	mov##c	pc, r
+# endif
+#endif
+
+#define GLUE2(a, b) a ## b
+#define GLUE(a, b) GLUE2(a, b)
+#define SYMBOL_NAME(name) GLUE(__USER_LABEL_PREFIX__, name)
+
+#ifdef VISIBILITY_HIDDEN
+#define DECLARE_SYMBOL_VISIBILITY(name)                    \
+  HIDDEN_DIRECTIVE SYMBOL_NAME(name) SEPARATOR
+#else
+#define DECLARE_SYMBOL_VISIBILITY(name)
+#endif
+
+#define DEFINE_COMPILERRT_FUNCTION(name)                   \
+  FILE_LEVEL_DIRECTIVE     SEPARATOR                       \
+  .globl SYMBOL_NAME(name) SEPARATOR                       \
+  SYMBOL_IS_FUNC(SYMBOL_NAME(name)) SEPARATOR              \
+  DECLARE_SYMBOL_VISIBILITY(name)                          \
+  SYMBOL_NAME(name):
+
+#define DEFINE_COMPILERRT_PRIVATE_FUNCTION(name)           \
+  .globl SYMBOL_NAME(name) SEPARATOR                       \
+  SYMBOL_IS_FUNC(SYMBOL_NAME(name)) SEPARATOR              \
+  HIDDEN_DIRECTIVE SYMBOL_NAME(name) SEPARATOR             \
+  SYMBOL_NAME(name):
+
+#define DEFINE_COMPILERRT_PRIVATE_FUNCTION_UNMANGLED(name) \
+  .globl name SEPARATOR                                    \
+  SYMBOL_IS_FUNC(name) SEPARATOR                           \
+  HIDDEN_DIRECTIVE name SEPARATOR                          \
+  name:
+
+#define DEFINE_COMPILERRT_FUNCTION_ALIAS(name, target)     \
+  .globl SYMBOL_NAME(name) SEPARATOR                       \
+  SYMBOL_IS_FUNC(SYMBOL_NAME(name)) SEPARATOR              \
+  .set SYMBOL_NAME(name), SYMBOL_NAME(target) SEPARATOR
+
+#if defined (__ARM_EABI__)
+# define DEFINE_AEABI_FUNCTION_ALIAS(aeabi_name, name)      \
+  DEFINE_COMPILERRT_FUNCTION_ALIAS(aeabi_name, name)
+#else
+# define DEFINE_AEABI_FUNCTION_ALIAS(aeabi_name, name)
+#endif
+
+#ifdef __ELF__
+#define END_COMPILERRT_FUNCTION(name) \
+	.size	SYMBOL_NAME(name), . - SYMBOL_NAME(name)
+#else
+#define END_COMPILERRT_FUNCTION(name)
+#endif
+
+#endif /* COMPILERRT_ASSEMBLY_H */
diff --git a/lib/builtins/atomic.c b/lib/builtins/atomic.c
new file mode 100644
index 000000000..02429a653
--- /dev/null
+++ b/lib/builtins/atomic.c
@@ -0,0 +1,337 @@
+/*===-- atomic.c - Implement support functions for atomic operations.------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ *===----------------------------------------------------------------------===
+ *
+ *  atomic.c defines a set of functions for performing atomic accesses on
+ *  arbitrary-sized memory locations.  This design uses locks that should
+ *  be fast in the uncontended case, for two reasons:
+ * 
+ *  1) This code must work with C programs that do not link to anything
+ *     (including pthreads) and so it should not depend on any pthread
+ *     functions.
+ *  2) Atomic operations, rather than explicit mutexes, are most commonly used
+ *     on code where contended operations are rate.
+ * 
+ *  To avoid needing a per-object lock, this code allocates an array of
+ *  locks and hashes the object pointers to find the one that it should use.
+ *  For operations that must be atomic on two locations, the lower lock is
+ *  always acquired first, to avoid deadlock.
+ *
+ *===----------------------------------------------------------------------===
+ */
+
+#include <stdint.h>
+#include <string.h>
+
+// Clang objects if you redefine a builtin.  This little hack allows us to
+// define a function with the same name as an intrinsic.
+#if __APPLE__
+// mach-o has extra leading underscore
+#pragma redefine_extname __atomic_load_c ___atomic_load
+#pragma redefine_extname __atomic_store_c ___atomic_store
+#pragma redefine_extname __atomic_exchange_c ___atomic_exchange
+#pragma redefine_extname __atomic_compare_exchange_c ___atomic_compare_exchange
+#else
+#pragma redefine_extname __atomic_load_c __atomic_load
+#pragma redefine_extname __atomic_store_c __atomic_store
+#pragma redefine_extname __atomic_exchange_c __atomic_exchange
+#pragma redefine_extname __atomic_compare_exchange_c __atomic_compare_exchange
+#endif
+
+/// Number of locks.  This allocates one page on 32-bit platforms, two on
+/// 64-bit.  This can be specified externally if a different trade between
+/// memory usage and contention probability is required for a given platform.
+#ifndef SPINLOCK_COUNT
+#define SPINLOCK_COUNT (1<<10)
+#endif
+static const long SPINLOCK_MASK = SPINLOCK_COUNT - 1;
+
+////////////////////////////////////////////////////////////////////////////////
+// Platform-specific lock implementation.  Falls back to spinlocks if none is
+// defined.  Each platform should define the Lock type, and corresponding
+// lock() and unlock() functions.
+////////////////////////////////////////////////////////////////////////////////
+#ifdef __FreeBSD__
+#include <errno.h>
+#include <sys/types.h>
+#include <machine/atomic.h>
+#include <sys/umtx.h>
+typedef struct _usem Lock;
+inline static void unlock(Lock *l) {
+  __c11_atomic_store((_Atomic(uint32_t)*)&l->_count, 1, __ATOMIC_RELEASE);
+  __c11_atomic_thread_fence(__ATOMIC_SEQ_CST);
+  if (l->_has_waiters)
+      _umtx_op(l, UMTX_OP_SEM_WAKE, 1, 0, 0);
+}
+inline static void lock(Lock *l) {
+  uint32_t old = 1;
+  while (!__c11_atomic_compare_exchange_weak((_Atomic(uint32_t)*)&l->_count, &old,
+        0, __ATOMIC_ACQUIRE, __ATOMIC_RELAXED)) {
+    _umtx_op(l, UMTX_OP_SEM_WAIT, 0, 0, 0);
+    old = 1;
+  }
+}
+/// locks for atomic operations
+static Lock locks[SPINLOCK_COUNT] = { [0 ...  SPINLOCK_COUNT-1] = {0,1,0} };
+
+#elif defined(__APPLE__)
+#include <libkern/OSAtomic.h>
+typedef OSSpinLock Lock;
+inline static void unlock(Lock *l) {
+  OSSpinLockUnlock(l);
+}
+/// Locks a lock.  In the current implementation, this is potentially
+/// unbounded in the contended case.
+inline static void lock(Lock *l) {  
+  OSSpinLockLock(l);
+}
+static Lock locks[SPINLOCK_COUNT]; // initialized to OS_SPINLOCK_INIT which is 0
+
+#else
+typedef _Atomic(uintptr_t) Lock;
+/// Unlock a lock.  This is a release operation.
+inline static void unlock(Lock *l) {
+  __c11_atomic_store(l, 0, __ATOMIC_RELEASE);
+}
+/// Locks a lock.  In the current implementation, this is potentially
+/// unbounded in the contended case.
+inline static void lock(Lock *l) {
+  uintptr_t old = 0;
+  while (!__c11_atomic_compare_exchange_weak(l, &old, 1, __ATOMIC_ACQUIRE,
+        __ATOMIC_RELAXED))
+    old = 0;
+}
+/// locks for atomic operations
+static Lock locks[SPINLOCK_COUNT];
+#endif
+
+
+/// Returns a lock to use for a given pointer.  
+static inline Lock *lock_for_pointer(void *ptr) {
+  intptr_t hash = (intptr_t)ptr;
+  // Disregard the lowest 4 bits.  We want all values that may be part of the
+  // same memory operation to hash to the same value and therefore use the same
+  // lock.  
+  hash >>= 4;
+  // Use the next bits as the basis for the hash
+  intptr_t low = hash & SPINLOCK_MASK;
+  // Now use the high(er) set of bits to perturb the hash, so that we don't
+  // get collisions from atomic fields in a single object
+  hash >>= 16;
+  hash ^= low;
+  // Return a pointer to the word to use
+  return locks + (hash & SPINLOCK_MASK);
+}
+
+/// Macros for determining whether a size is lock free.  Clang can not yet
+/// codegen __atomic_is_lock_free(16), so for now we assume 16-byte values are
+/// not lock free.
+#define IS_LOCK_FREE_1 __c11_atomic_is_lock_free(1)
+#define IS_LOCK_FREE_2 __c11_atomic_is_lock_free(2)
+#define IS_LOCK_FREE_4 __c11_atomic_is_lock_free(4)
+#define IS_LOCK_FREE_8 __c11_atomic_is_lock_free(8)
+#define IS_LOCK_FREE_16 0
+
+/// Macro that calls the compiler-generated lock-free versions of functions
+/// when they exist.
+#define LOCK_FREE_CASES() \
+  do {\
+  switch (size) {\
+    case 2:\
+      if (IS_LOCK_FREE_2) {\
+        LOCK_FREE_ACTION(uint16_t);\
+      }\
+    case 4:\
+      if (IS_LOCK_FREE_4) {\
+        LOCK_FREE_ACTION(uint32_t);\
+      }\
+    case 8:\
+      if (IS_LOCK_FREE_8) {\
+        LOCK_FREE_ACTION(uint64_t);\
+      }\
+    case 16:\
+      if (IS_LOCK_FREE_16) {\
+        /* FIXME: __uint128_t isn't available on 32 bit platforms.
+        LOCK_FREE_ACTION(__uint128_t);*/\
+      }\
+  }\
+  } while (0)
+
+
+/// An atomic load operation.  This is atomic with respect to the source
+/// pointer only.
+void __atomic_load_c(int size, void *src, void *dest, int model) {
+#define LOCK_FREE_ACTION(type) \
+    *((type*)dest) = __c11_atomic_load((_Atomic(type)*)src, model);\
+    return;
+  LOCK_FREE_CASES();
+#undef LOCK_FREE_ACTION
+  Lock *l = lock_for_pointer(src);
+  lock(l);
+  memcpy(dest, src, size);
+  unlock(l);
+}
+
+/// An atomic store operation.  This is atomic with respect to the destination
+/// pointer only.
+void __atomic_store_c(int size, void *dest, void *src, int model) {
+#define LOCK_FREE_ACTION(type) \
+    __c11_atomic_store((_Atomic(type)*)dest, *(type*)dest, model);\
+    return;
+  LOCK_FREE_CASES();
+#undef LOCK_FREE_ACTION
+  Lock *l = lock_for_pointer(dest);
+  lock(l);
+  memcpy(dest, src, size);
+  unlock(l);
+}
+
+/// Atomic compare and exchange operation.  If the value at *ptr is identical
+/// to the value at *expected, then this copies value at *desired to *ptr.  If
+/// they  are not, then this stores the current value from *ptr in *expected.
+///
+/// This function returns 1 if the exchange takes place or 0 if it fails. 
+int __atomic_compare_exchange_c(int size, void *ptr, void *expected,
+    void *desired, int success, int failure) {
+#define LOCK_FREE_ACTION(type) \
+  return __c11_atomic_compare_exchange_strong((_Atomic(type)*)ptr, (type*)expected,\
+      *(type*)desired, success, failure)
+  LOCK_FREE_CASES();
+#undef LOCK_FREE_ACTION
+  Lock *l = lock_for_pointer(ptr);
+  lock(l);
+  if (memcmp(ptr, expected, size) == 0) {
+    memcpy(ptr, desired, size);
+    unlock(l);
+    return 1;
+  }
+  memcpy(expected, ptr, size);
+  unlock(l);
+  return 0;
+}
+
+/// Performs an atomic exchange operation between two pointers.  This is atomic
+/// with respect to the target address.
+void __atomic_exchange_c(int size, void *ptr, void *val, void *old, int model) {
+#define LOCK_FREE_ACTION(type) \
+    *(type*)old = __c11_atomic_exchange((_Atomic(type)*)ptr, *(type*)val,\
+        model);\
+    return;
+  LOCK_FREE_CASES();
+#undef LOCK_FREE_ACTION
+  Lock *l = lock_for_pointer(ptr);
+  lock(l);
+  memcpy(old, ptr, size);
+  memcpy(ptr, val, size);
+  unlock(l);
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// Where the size is known at compile time, the compiler may emit calls to
+// specialised versions of the above functions.
+////////////////////////////////////////////////////////////////////////////////
+#define OPTIMISED_CASES\
+  OPTIMISED_CASE(1, IS_LOCK_FREE_1, uint8_t)\
+  OPTIMISED_CASE(2, IS_LOCK_FREE_2, uint16_t)\
+  OPTIMISED_CASE(4, IS_LOCK_FREE_4, uint32_t)\
+  OPTIMISED_CASE(8, IS_LOCK_FREE_8, uint64_t)\
+  /* FIXME: __uint128_t isn't available on 32 bit platforms.
+  OPTIMISED_CASE(16, IS_LOCK_FREE_16, __uint128_t)*/\
+
+#define OPTIMISED_CASE(n, lockfree, type)\
+type __atomic_load_##n(type *src, int model) {\
+  if (lockfree)\
+    return __c11_atomic_load((_Atomic(type)*)src, model);\
+  Lock *l = lock_for_pointer(src);\
+  lock(l);\
+  type val = *src;\
+  unlock(l);\
+  return val;\
+}
+OPTIMISED_CASES
+#undef OPTIMISED_CASE
+
+#define OPTIMISED_CASE(n, lockfree, type)\
+void  __atomic_store_##n(type *dest, type val, int model) {\
+  if (lockfree) {\
+    __c11_atomic_store((_Atomic(type)*)dest, val, model);\
+    return;\
+  }\
+  Lock *l = lock_for_pointer(dest);\
+  lock(l);\
+  *dest = val;\
+  unlock(l);\
+  return;\
+}
+OPTIMISED_CASES
+#undef OPTIMISED_CASE
+
+#define OPTIMISED_CASE(n, lockfree, type)\
+type __atomic_exchange_##n(type *dest, type val, int model) {\
+  if (lockfree)\
+    return __c11_atomic_exchange((_Atomic(type)*)dest, val, model);\
+  Lock *l = lock_for_pointer(dest);\
+  lock(l);\
+  type tmp = *dest;\
+  *dest = val;\
+  unlock(l);\
+  return tmp;\
+}
+OPTIMISED_CASES
+#undef OPTIMISED_CASE
+
+#define OPTIMISED_CASE(n, lockfree, type)\
+int __atomic_compare_exchange_##n(type *ptr, type *expected, type desired,\
+    int success, int failure) {\
+  if (lockfree)\
+    return __c11_atomic_compare_exchange_strong((_Atomic(type)*)ptr, expected, desired,\
+        success, failure);\
+  Lock *l = lock_for_pointer(ptr);\
+  lock(l);\
+  if (*ptr == *expected) {\
+    *ptr = desired;\
+    unlock(l);\
+    return 1;\
+  }\
+  *expected = *ptr;\
+  unlock(l);\
+  return 0;\
+}
+OPTIMISED_CASES
+#undef OPTIMISED_CASE
+
+////////////////////////////////////////////////////////////////////////////////
+// Atomic read-modify-write operations for integers of various sizes.
+////////////////////////////////////////////////////////////////////////////////
+#define ATOMIC_RMW(n, lockfree, type, opname, op) \
+type __atomic_fetch_##opname##_##n(type *ptr, type val, int model) {\
+  if (lockfree) \
+    return __c11_atomic_fetch_##opname((_Atomic(type)*)ptr, val, model);\
+  Lock *l = lock_for_pointer(ptr);\
+  lock(l);\
+  type tmp = *ptr;\
+  *ptr = tmp op val;\
+  unlock(l);\
+  return tmp;\
+}
+
+#define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, add, +)
+OPTIMISED_CASES
+#undef OPTIMISED_CASE
+#define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, sub, -)
+OPTIMISED_CASES
+#undef OPTIMISED_CASE
+#define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, and, &)
+OPTIMISED_CASES
+#undef OPTIMISED_CASE
+#define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, or, |)
+OPTIMISED_CASES
+#undef OPTIMISED_CASE
+#define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, xor, ^)
+OPTIMISED_CASES
+#undef OPTIMISED_CASE
diff --git a/lib/builtins/clear_cache.c b/lib/builtins/clear_cache.c
new file mode 100644
index 000000000..7aee28590
--- /dev/null
+++ b/lib/builtins/clear_cache.c
@@ -0,0 +1,71 @@
+/* ===-- clear_cache.c - Implement __clear_cache ---------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __APPLE__
+  #include <libkern/OSCacheControl.h>
+#endif
+#if defined(__NetBSD__) && defined(__arm__)
+  #include <machine/sysarch.h>
+#endif
+
+/*
+ * The compiler generates calls to __clear_cache() when creating 
+ * trampoline functions on the stack for use with nested functions.
+ * It is expected to invalidate the instruction cache for the 
+ * specified range.
+ */
+
+void __clear_cache(void* start, void* end)
+{
+#if __i386__ || __x86_64__
+/*
+ * Intel processors have a unified instruction and data cache
+ * so there is nothing to do
+ */
+#elif defined(__NetBSD__) && defined(__arm__)
+  struct arm_sync_icache_args arg;
+
+  arg.addr = (uintptr_t)start;
+  arg.len = (uintptr_t)end - (uintptr_t)start;
+
+  sysarch(ARM_SYNC_ICACHE, &arg);
+#elif defined(__aarch64__) && !defined(__APPLE__)
+  uint64_t xstart = (uint64_t)(uintptr_t) start;
+  uint64_t xend = (uint64_t)(uintptr_t) end;
+
+  // Get Cache Type Info
+  uint64_t ctr_el0;
+  __asm __volatile("mrs %0, ctr_el0" : "=r"(ctr_el0));
+
+  /*
+   * dc & ic instructions must use 64bit registers so we don't use
+   * uintptr_t in case this runs in an IPL32 environment.
+   */
+  const size_t dcache_line_size = 4 << ((ctr_el0 >> 16) & 15);
+  for (uint64_t addr = xstart; addr < xend; addr += dcache_line_size)
+    __asm __volatile("dc cvau, %0" :: "r"(addr));
+  __asm __volatile("dsb ish");
+
+  const size_t icache_line_size = 4 << ((ctr_el0 >> 0) & 15);
+  for (uint64_t addr = xstart; addr < xend; addr += icache_line_size)
+    __asm __volatile("ic ivau, %0" :: "r"(addr));
+  __asm __volatile("isb sy");
+#else
+    #if __APPLE__
+        /* On Darwin, sys_icache_invalidate() provides this functionality */
+        sys_icache_invalidate(start, end-start);
+    #else
+        compilerrt_abort();
+    #endif
+#endif
+}
+
diff --git a/lib/builtins/clzdi2.c b/lib/builtins/clzdi2.c
new file mode 100644
index 000000000..b9e64da49
--- /dev/null
+++ b/lib/builtins/clzdi2.c
@@ -0,0 +1,29 @@
+/* ===-- clzdi2.c - Implement __clzdi2 -------------------------------------===
+ *
+ *               The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __clzdi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: the number of leading 0-bits */
+
+/* Precondition: a != 0 */
+
+COMPILER_RT_ABI si_int
+__clzdi2(di_int a)
+{
+    dwords x;
+    x.all = a;
+    const si_int f = -(x.s.high == 0);
+    return __builtin_clz((x.s.high & ~f) | (x.s.low & f)) +
+           (f & ((si_int)(sizeof(si_int) * CHAR_BIT)));
+}
diff --git a/lib/builtins/clzsi2.c b/lib/builtins/clzsi2.c
new file mode 100644
index 000000000..25b8ed2c4
--- /dev/null
+++ b/lib/builtins/clzsi2.c
@@ -0,0 +1,53 @@
+/* ===-- clzsi2.c - Implement __clzsi2 -------------------------------------===
+ *
+ *               The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __clzsi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: the number of leading 0-bits */
+
+/* Precondition: a != 0 */
+
+COMPILER_RT_ABI si_int
+__clzsi2(si_int a)
+{
+    su_int x = (su_int)a;
+    si_int t = ((x & 0xFFFF0000) == 0) << 4;  /* if (x is small) t = 16 else 0 */
+    x >>= 16 - t;      /* x = [0 - 0xFFFF] */
+    su_int r = t;       /* r = [0, 16] */
+    /* return r + clz(x) */
+    t = ((x & 0xFF00) == 0) << 3;
+    x >>= 8 - t;       /* x = [0 - 0xFF] */
+    r += t;            /* r = [0, 8, 16, 24] */
+    /* return r + clz(x) */
+    t = ((x & 0xF0) == 0) << 2;
+    x >>= 4 - t;       /* x = [0 - 0xF] */
+    r += t;            /* r = [0, 4, 8, 12, 16, 20, 24, 28] */
+    /* return r + clz(x) */
+    t = ((x & 0xC) == 0) << 1;
+    x >>= 2 - t;       /* x = [0 - 3] */
+    r += t;            /* r = [0 - 30] and is even */
+    /* return r + clz(x) */
+/*     switch (x)
+ *     {
+ *     case 0:
+ *         return r + 2;
+ *     case 1:
+ *         return r + 1;
+ *     case 2:
+ *     case 3:
+ *         return r;
+ *     }
+ */
+    return r + ((2 - x) & -((x & 2) == 0));
+}
diff --git a/lib/builtins/clzti2.c b/lib/builtins/clzti2.c
new file mode 100644
index 000000000..355c20e42
--- /dev/null
+++ b/lib/builtins/clzti2.c
@@ -0,0 +1,33 @@
+/* ===-- clzti2.c - Implement __clzti2 -------------------------------------===
+ *
+ *      	       The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __clzti2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: the number of leading 0-bits */
+
+/* Precondition: a != 0 */
+
+si_int
+__clzti2(ti_int a)
+{
+    twords x;
+    x.all = a;
+    const di_int f = -(x.s.high == 0);
+    return __builtin_clzll((x.s.high & ~f) | (x.s.low & f)) +
+           ((si_int)f & ((si_int)(sizeof(di_int) * CHAR_BIT)));
+}
+
+#endif /* __x86_64 */
diff --git a/lib/builtins/cmpdi2.c b/lib/builtins/cmpdi2.c
new file mode 100644
index 000000000..52634d9c3
--- /dev/null
+++ b/lib/builtins/cmpdi2.c
@@ -0,0 +1,51 @@
+/* ===-- cmpdi2.c - Implement __cmpdi2 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __cmpdi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: if (a <  b) returns 0
+*           if (a == b) returns 1
+*           if (a >  b) returns 2
+*/
+
+COMPILER_RT_ABI si_int
+__cmpdi2(di_int a, di_int b)
+{
+    dwords x;
+    x.all = a;
+    dwords y;
+    y.all = b;
+    if (x.s.high < y.s.high)
+        return 0;
+    if (x.s.high > y.s.high)
+        return 2;
+    if (x.s.low < y.s.low)
+        return 0;
+    if (x.s.low > y.s.low)
+        return 2;
+    return 1;
+}
+
+#ifdef __ARM_EABI__
+/* Returns: if (a <  b) returns -1
+*           if (a == b) returns  0
+*           if (a >  b) returns  1
+*/
+COMPILER_RT_ABI si_int
+__aeabi_lcmp(di_int a, di_int b)
+{
+	return __cmpdi2(a, b) - 1;
+}
+#endif
+
diff --git a/lib/builtins/cmpti2.c b/lib/builtins/cmpti2.c
new file mode 100644
index 000000000..d0aec452a
--- /dev/null
+++ b/lib/builtins/cmpti2.c
@@ -0,0 +1,42 @@
+/* ===-- cmpti2.c - Implement __cmpti2 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __cmpti2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns:  if (a <  b) returns 0
+ *           if (a == b) returns 1
+ *           if (a >  b) returns 2
+ */
+
+si_int
+__cmpti2(ti_int a, ti_int b)
+{
+    twords x;
+    x.all = a;
+    twords y;
+    y.all = b;
+    if (x.s.high < y.s.high)
+        return 0;
+    if (x.s.high > y.s.high)
+        return 2;
+    if (x.s.low < y.s.low)
+        return 0;
+    if (x.s.low > y.s.low)
+        return 2;
+    return 1;
+}
+
+#endif
diff --git a/lib/builtins/comparedf2.c b/lib/builtins/comparedf2.c
new file mode 100644
index 000000000..de67784dc
--- /dev/null
+++ b/lib/builtins/comparedf2.c
@@ -0,0 +1,134 @@
+//===-- lib/comparedf2.c - Double-precision comparisons -----------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// // This file implements the following soft-float comparison routines:
+//
+//   __eqdf2   __gedf2   __unorddf2
+//   __ledf2   __gtdf2
+//   __ltdf2
+//   __nedf2
+//
+// The semantics of the routines grouped in each column are identical, so there
+// is a single implementation for each, and wrappers to provide the other names.
+//
+// The main routines behave as follows:
+//
+//   __ledf2(a,b) returns -1 if a < b
+//                         0 if a == b
+//                         1 if a > b
+//                         1 if either a or b is NaN
+//
+//   __gedf2(a,b) returns -1 if a < b
+//                         0 if a == b
+//                         1 if a > b
+//                        -1 if either a or b is NaN
+//
+//   __unorddf2(a,b) returns 0 if both a and b are numbers
+//                           1 if either a or b is NaN
+//
+// Note that __ledf2( ) and __gedf2( ) are identical except in their handling of
+// NaN values.
+//
+//===----------------------------------------------------------------------===//
+
+#define DOUBLE_PRECISION
+#include "fp_lib.h"
+
+enum LE_RESULT {
+    LE_LESS      = -1,
+    LE_EQUAL     =  0,
+    LE_GREATER   =  1,
+    LE_UNORDERED =  1
+};
+
+enum LE_RESULT __ledf2(fp_t a, fp_t b) {
+    
+    const srep_t aInt = toRep(a);
+    const srep_t bInt = toRep(b);
+    const rep_t aAbs = aInt & absMask;
+    const rep_t bAbs = bInt & absMask;
+    
+    // If either a or b is NaN, they are unordered.
+    if (aAbs > infRep || bAbs > infRep) return LE_UNORDERED;
+    
+    // If a and b are both zeros, they are equal.
+    if ((aAbs | bAbs) == 0) return LE_EQUAL;
+    
+    // If at least one of a and b is positive, we get the same result comparing
+    // a and b as signed integers as we would with a floating-point compare.
+    if ((aInt & bInt) >= 0) {
+        if (aInt < bInt) return LE_LESS;
+        else if (aInt == bInt) return LE_EQUAL;
+        else return LE_GREATER;
+    }
+    
+    // Otherwise, both are negative, so we need to flip the sense of the
+    // comparison to get the correct result.  (This assumes a twos- or ones-
+    // complement integer representation; if integers are represented in a
+    // sign-magnitude representation, then this flip is incorrect).
+    else {
+        if (aInt > bInt) return LE_LESS;
+        else if (aInt == bInt) return LE_EQUAL;
+        else return LE_GREATER;
+    }
+}
+
+enum GE_RESULT {
+    GE_LESS      = -1,
+    GE_EQUAL     =  0,
+    GE_GREATER   =  1,
+    GE_UNORDERED = -1   // Note: different from LE_UNORDERED
+};
+
+enum GE_RESULT __gedf2(fp_t a, fp_t b) {
+    
+    const srep_t aInt = toRep(a);
+    const srep_t bInt = toRep(b);
+    const rep_t aAbs = aInt & absMask;
+    const rep_t bAbs = bInt & absMask;
+    
+    if (aAbs > infRep || bAbs > infRep) return GE_UNORDERED;
+    if ((aAbs | bAbs) == 0) return GE_EQUAL;
+    if ((aInt & bInt) >= 0) {
+        if (aInt < bInt) return GE_LESS;
+        else if (aInt == bInt) return GE_EQUAL;
+        else return GE_GREATER;
+    } else {
+        if (aInt > bInt) return GE_LESS;
+        else if (aInt == bInt) return GE_EQUAL;
+        else return GE_GREATER;
+    }
+}
+
+ARM_EABI_FNALIAS(dcmpun, unorddf2)
+
+int __unorddf2(fp_t a, fp_t b) {
+    const rep_t aAbs = toRep(a) & absMask;
+    const rep_t bAbs = toRep(b) & absMask;
+    return aAbs > infRep || bAbs > infRep;
+}
+
+// The following are alternative names for the preceeding routines.
+
+enum LE_RESULT __eqdf2(fp_t a, fp_t b) {
+    return __ledf2(a, b);
+}
+
+enum LE_RESULT __ltdf2(fp_t a, fp_t b) {
+    return __ledf2(a, b);
+}
+
+enum LE_RESULT __nedf2(fp_t a, fp_t b) {
+    return __ledf2(a, b);
+}
+
+enum GE_RESULT __gtdf2(fp_t a, fp_t b) {
+    return __gedf2(a, b);
+}
+
diff --git a/lib/builtins/comparesf2.c b/lib/builtins/comparesf2.c
new file mode 100644
index 000000000..c1c3a479c
--- /dev/null
+++ b/lib/builtins/comparesf2.c
@@ -0,0 +1,133 @@
+//===-- lib/comparesf2.c - Single-precision comparisons -----------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the following soft-fp_t comparison routines:
+//
+//   __eqsf2   __gesf2   __unordsf2
+//   __lesf2   __gtsf2
+//   __ltsf2
+//   __nesf2
+//
+// The semantics of the routines grouped in each column are identical, so there
+// is a single implementation for each, and wrappers to provide the other names.
+//
+// The main routines behave as follows:
+//
+//   __lesf2(a,b) returns -1 if a < b
+//                         0 if a == b
+//                         1 if a > b
+//                         1 if either a or b is NaN
+//
+//   __gesf2(a,b) returns -1 if a < b
+//                         0 if a == b
+//                         1 if a > b
+//                        -1 if either a or b is NaN
+//
+//   __unordsf2(a,b) returns 0 if both a and b are numbers
+//                           1 if either a or b is NaN
+//
+// Note that __lesf2( ) and __gesf2( ) are identical except in their handling of
+// NaN values.
+//
+//===----------------------------------------------------------------------===//
+
+#define SINGLE_PRECISION
+#include "fp_lib.h"
+
+enum LE_RESULT {
+    LE_LESS      = -1,
+    LE_EQUAL     =  0,
+    LE_GREATER   =  1,
+    LE_UNORDERED =  1
+};
+
+enum LE_RESULT __lesf2(fp_t a, fp_t b) {
+    
+    const srep_t aInt = toRep(a);
+    const srep_t bInt = toRep(b);
+    const rep_t aAbs = aInt & absMask;
+    const rep_t bAbs = bInt & absMask;
+    
+    // If either a or b is NaN, they are unordered.
+    if (aAbs > infRep || bAbs > infRep) return LE_UNORDERED;
+    
+    // If a and b are both zeros, they are equal.
+    if ((aAbs | bAbs) == 0) return LE_EQUAL;
+    
+    // If at least one of a and b is positive, we get the same result comparing
+    // a and b as signed integers as we would with a fp_ting-point compare.
+    if ((aInt & bInt) >= 0) {
+        if (aInt < bInt) return LE_LESS;
+        else if (aInt == bInt) return LE_EQUAL;
+        else return LE_GREATER;
+    }
+    
+    // Otherwise, both are negative, so we need to flip the sense of the
+    // comparison to get the correct result.  (This assumes a twos- or ones-
+    // complement integer representation; if integers are represented in a
+    // sign-magnitude representation, then this flip is incorrect).
+    else {
+        if (aInt > bInt) return LE_LESS;
+        else if (aInt == bInt) return LE_EQUAL;
+        else return LE_GREATER;
+    }
+}
+
+enum GE_RESULT {
+    GE_LESS      = -1,
+    GE_EQUAL     =  0,
+    GE_GREATER   =  1,
+    GE_UNORDERED = -1   // Note: different from LE_UNORDERED
+};
+
+enum GE_RESULT __gesf2(fp_t a, fp_t b) {
+    
+    const srep_t aInt = toRep(a);
+    const srep_t bInt = toRep(b);
+    const rep_t aAbs = aInt & absMask;
+    const rep_t bAbs = bInt & absMask;
+    
+    if (aAbs > infRep || bAbs > infRep) return GE_UNORDERED;
+    if ((aAbs | bAbs) == 0) return GE_EQUAL;
+    if ((aInt & bInt) >= 0) {
+        if (aInt < bInt) return GE_LESS;
+        else if (aInt == bInt) return GE_EQUAL;
+        else return GE_GREATER;
+    } else {
+        if (aInt > bInt) return GE_LESS;
+        else if (aInt == bInt) return GE_EQUAL;
+        else return GE_GREATER;
+    }
+}
+
+ARM_EABI_FNALIAS(fcmpun, unordsf2)
+
+int __unordsf2(fp_t a, fp_t b) {
+    const rep_t aAbs = toRep(a) & absMask;
+    const rep_t bAbs = toRep(b) & absMask;
+    return aAbs > infRep || bAbs > infRep;
+}
+
+// The following are alternative names for the preceeding routines.
+
+enum LE_RESULT __eqsf2(fp_t a, fp_t b) {
+    return __lesf2(a, b);
+}
+
+enum LE_RESULT __ltsf2(fp_t a, fp_t b) {
+    return __lesf2(a, b);
+}
+
+enum LE_RESULT __nesf2(fp_t a, fp_t b) {
+    return __lesf2(a, b);
+}
+
+enum GE_RESULT __gtsf2(fp_t a, fp_t b) {
+    return __gesf2(a, b);
+}
diff --git a/lib/builtins/ctzdi2.c b/lib/builtins/ctzdi2.c
new file mode 100644
index 000000000..db3c6fdc0
--- /dev/null
+++ b/lib/builtins/ctzdi2.c
@@ -0,0 +1,29 @@
+/* ===-- ctzdi2.c - Implement __ctzdi2 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __ctzdi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: the number of trailing 0-bits  */
+
+/* Precondition: a != 0 */
+
+COMPILER_RT_ABI si_int
+__ctzdi2(di_int a)
+{
+    dwords x;
+    x.all = a;
+    const si_int f = -(x.s.low == 0);
+    return __builtin_ctz((x.s.high & f) | (x.s.low & ~f)) +
+              (f & ((si_int)(sizeof(si_int) * CHAR_BIT)));
+}
diff --git a/lib/builtins/ctzsi2.c b/lib/builtins/ctzsi2.c
new file mode 100644
index 000000000..c69486ea4
--- /dev/null
+++ b/lib/builtins/ctzsi2.c
@@ -0,0 +1,57 @@
+/* ===-- ctzsi2.c - Implement __ctzsi2 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __ctzsi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: the number of trailing 0-bits */
+
+/* Precondition: a != 0 */
+
+COMPILER_RT_ABI si_int
+__ctzsi2(si_int a)
+{
+    su_int x = (su_int)a;
+    si_int t = ((x & 0x0000FFFF) == 0) << 4;  /* if (x has no small bits) t = 16 else 0 */
+    x >>= t;           /* x = [0 - 0xFFFF] + higher garbage bits */
+    su_int r = t;       /* r = [0, 16]  */
+    /* return r + ctz(x) */
+    t = ((x & 0x00FF) == 0) << 3;
+    x >>= t;           /* x = [0 - 0xFF] + higher garbage bits */
+    r += t;            /* r = [0, 8, 16, 24] */
+    /* return r + ctz(x) */
+    t = ((x & 0x0F) == 0) << 2;
+    x >>= t;           /* x = [0 - 0xF] + higher garbage bits */
+    r += t;            /* r = [0, 4, 8, 12, 16, 20, 24, 28] */
+    /* return r + ctz(x) */
+    t = ((x & 0x3) == 0) << 1;
+    x >>= t;
+    x &= 3;            /* x = [0 - 3] */
+    r += t;            /* r = [0 - 30] and is even */
+    /* return r + ctz(x) */
+
+/*  The branch-less return statement below is equivalent
+ *  to the following switch statement:
+ *     switch (x)
+ *    {
+ *     case 0:
+ *         return r + 2;
+ *     case 2:
+ *         return r + 1;
+ *     case 1:
+ *     case 3:
+ *         return r;
+ *     }
+ */
+    return r + ((2 - (x >> 1)) & -((x & 1) == 0));
+}
diff --git a/lib/builtins/ctzti2.c b/lib/builtins/ctzti2.c
new file mode 100644
index 000000000..66dc01b7a
--- /dev/null
+++ b/lib/builtins/ctzti2.c
@@ -0,0 +1,33 @@
+/* ===-- ctzti2.c - Implement __ctzti2 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __ctzti2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: the number of trailing 0-bits */
+
+/* Precondition: a != 0 */
+
+si_int
+__ctzti2(ti_int a)
+{
+    twords x;
+    x.all = a;
+    const di_int f = -(x.s.low == 0);
+    return __builtin_ctzll((x.s.high & f) | (x.s.low & ~f)) +
+              ((si_int)f & ((si_int)(sizeof(di_int) * CHAR_BIT)));
+}
+
+#endif
diff --git a/lib/builtins/divdc3.c b/lib/builtins/divdc3.c
new file mode 100644
index 000000000..cfbc498e7
--- /dev/null
+++ b/lib/builtins/divdc3.c
@@ -0,0 +1,60 @@
+/* ===-- divdc3.c - Implement __divdc3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __divdc3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+#include "int_math.h"
+
+/* Returns: the quotient of (a + ib) / (c + id) */
+
+double _Complex
+__divdc3(double __a, double __b, double __c, double __d)
+{
+    int __ilogbw = 0;
+    double __logbw = crt_logb(crt_fmax(crt_fabs(__c), crt_fabs(__d)));
+    if (crt_isfinite(__logbw))
+    {
+        __ilogbw = (int)__logbw;
+        __c = crt_scalbn(__c, -__ilogbw);
+        __d = crt_scalbn(__d, -__ilogbw);
+    }
+    double __denom = __c * __c + __d * __d;
+    double _Complex z;
+    __real__ z = crt_scalbn((__a * __c + __b * __d) / __denom, -__ilogbw);
+    __imag__ z = crt_scalbn((__b * __c - __a * __d) / __denom, -__ilogbw);
+    if (crt_isnan(__real__ z) && crt_isnan(__imag__ z))
+    {
+        if ((__denom == 0.0) && (!crt_isnan(__a) || !crt_isnan(__b)))
+        {
+            __real__ z = crt_copysign(CRT_INFINITY, __c) * __a;
+            __imag__ z = crt_copysign(CRT_INFINITY, __c) * __b;
+        }
+        else if ((crt_isinf(__a) || crt_isinf(__b)) &&
+                 crt_isfinite(__c) && crt_isfinite(__d))
+        {
+            __a = crt_copysign(crt_isinf(__a) ? 1.0 : 0.0, __a);
+            __b = crt_copysign(crt_isinf(__b) ? 1.0 : 0.0, __b);
+            __real__ z = CRT_INFINITY * (__a * __c + __b * __d);
+            __imag__ z = CRT_INFINITY * (__b * __c - __a * __d);
+        }
+        else if (crt_isinf(__logbw) && __logbw > 0.0 &&
+                 crt_isfinite(__a) && crt_isfinite(__b))
+        {
+            __c = crt_copysign(crt_isinf(__c) ? 1.0 : 0.0, __c);
+            __d = crt_copysign(crt_isinf(__d) ? 1.0 : 0.0, __d);
+            __real__ z = 0.0 * (__a * __c + __b * __d);
+            __imag__ z = 0.0 * (__b * __c - __a * __d);
+        }
+    }
+    return z;
+}
diff --git a/lib/builtins/divdf3.c b/lib/builtins/divdf3.c
new file mode 100644
index 000000000..efce6bb4b
--- /dev/null
+++ b/lib/builtins/divdf3.c
@@ -0,0 +1,184 @@
+//===-- lib/divdf3.c - Double-precision division ------------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements double-precision soft-float division
+// with the IEEE-754 default rounding (to nearest, ties to even).
+//
+// For simplicity, this implementation currently flushes denormals to zero.
+// It should be a fairly straightforward exercise to implement gradual
+// underflow with correct rounding.
+//
+//===----------------------------------------------------------------------===//
+
+#define DOUBLE_PRECISION
+#include "fp_lib.h"
+
+ARM_EABI_FNALIAS(ddiv, divdf3)
+
+fp_t __divdf3(fp_t a, fp_t b) {
+    
+    const unsigned int aExponent = toRep(a) >> significandBits & maxExponent;
+    const unsigned int bExponent = toRep(b) >> significandBits & maxExponent;
+    const rep_t quotientSign = (toRep(a) ^ toRep(b)) & signBit;
+    
+    rep_t aSignificand = toRep(a) & significandMask;
+    rep_t bSignificand = toRep(b) & significandMask;
+    int scale = 0;
+    
+    // Detect if a or b is zero, denormal, infinity, or NaN.
+    if (aExponent-1U >= maxExponent-1U || bExponent-1U >= maxExponent-1U) {
+        
+        const rep_t aAbs = toRep(a) & absMask;
+        const rep_t bAbs = toRep(b) & absMask;
+        
+        // NaN / anything = qNaN
+        if (aAbs > infRep) return fromRep(toRep(a) | quietBit);
+        // anything / NaN = qNaN
+        if (bAbs > infRep) return fromRep(toRep(b) | quietBit);
+        
+        if (aAbs == infRep) {
+            // infinity / infinity = NaN
+            if (bAbs == infRep) return fromRep(qnanRep);
+            // infinity / anything else = +/- infinity
+            else return fromRep(aAbs | quotientSign);
+        }
+        
+        // anything else / infinity = +/- 0
+        if (bAbs == infRep) return fromRep(quotientSign);
+        
+        if (!aAbs) {
+            // zero / zero = NaN
+            if (!bAbs) return fromRep(qnanRep);
+            // zero / anything else = +/- zero
+            else return fromRep(quotientSign);
+        }
+        // anything else / zero = +/- infinity
+        if (!bAbs) return fromRep(infRep | quotientSign);
+        
+        // one or both of a or b is denormal, the other (if applicable) is a
+        // normal number.  Renormalize one or both of a and b, and set scale to
+        // include the necessary exponent adjustment.
+        if (aAbs < implicitBit) scale += normalize(&aSignificand);
+        if (bAbs < implicitBit) scale -= normalize(&bSignificand);
+    }
+    
+    // Or in the implicit significand bit.  (If we fell through from the
+    // denormal path it was already set by normalize( ), but setting it twice
+    // won't hurt anything.)
+    aSignificand |= implicitBit;
+    bSignificand |= implicitBit;
+    int quotientExponent = aExponent - bExponent + scale;
+    
+    // Align the significand of b as a Q31 fixed-point number in the range
+    // [1, 2.0) and get a Q32 approximate reciprocal using a small minimax
+    // polynomial approximation: reciprocal = 3/4 + 1/sqrt(2) - b/2.  This
+    // is accurate to about 3.5 binary digits.
+    const uint32_t q31b = bSignificand >> 21;
+    uint32_t recip32 = UINT32_C(0x7504f333) - q31b;
+    
+    // Now refine the reciprocal estimate using a Newton-Raphson iteration:
+    //
+    //     x1 = x0 * (2 - x0 * b)
+    //
+    // This doubles the number of correct binary digits in the approximation
+    // with each iteration, so after three iterations, we have about 28 binary
+    // digits of accuracy.
+    uint32_t correction32;
+    correction32 = -((uint64_t)recip32 * q31b >> 32);
+    recip32 = (uint64_t)recip32 * correction32 >> 31;
+    correction32 = -((uint64_t)recip32 * q31b >> 32);
+    recip32 = (uint64_t)recip32 * correction32 >> 31;
+    correction32 = -((uint64_t)recip32 * q31b >> 32);
+    recip32 = (uint64_t)recip32 * correction32 >> 31;
+    
+    // recip32 might have overflowed to exactly zero in the preceeding
+    // computation if the high word of b is exactly 1.0.  This would sabotage
+    // the full-width final stage of the computation that follows, so we adjust
+    // recip32 downward by one bit.
+    recip32--;
+    
+    // We need to perform one more iteration to get us to 56 binary digits;
+    // The last iteration needs to happen with extra precision.
+    const uint32_t q63blo = bSignificand << 11;
+    uint64_t correction, reciprocal;
+    correction = -((uint64_t)recip32*q31b + ((uint64_t)recip32*q63blo >> 32));
+    uint32_t cHi = correction >> 32;
+    uint32_t cLo = correction;
+    reciprocal = (uint64_t)recip32*cHi + ((uint64_t)recip32*cLo >> 32);
+    
+    // We already adjusted the 32-bit estimate, now we need to adjust the final
+    // 64-bit reciprocal estimate downward to ensure that it is strictly smaller
+    // than the infinitely precise exact reciprocal.  Because the computation
+    // of the Newton-Raphson step is truncating at every step, this adjustment
+    // is small; most of the work is already done.
+    reciprocal -= 2;
+    
+    // The numerical reciprocal is accurate to within 2^-56, lies in the
+    // interval [0.5, 1.0), and is strictly smaller than the true reciprocal
+    // of b.  Multiplying a by this reciprocal thus gives a numerical q = a/b
+    // in Q53 with the following properties:
+    //
+    //    1. q < a/b
+    //    2. q is in the interval [0.5, 2.0)
+    //    3. the error in q is bounded away from 2^-53 (actually, we have a
+    //       couple of bits to spare, but this is all we need).
+    
+    // We need a 64 x 64 multiply high to compute q, which isn't a basic
+    // operation in C, so we need to be a little bit fussy.
+    rep_t quotient, quotientLo;
+    wideMultiply(aSignificand << 2, reciprocal, &quotient, &quotientLo);
+    
+    // Two cases: quotient is in [0.5, 1.0) or quotient is in [1.0, 2.0).
+    // In either case, we are going to compute a residual of the form
+    //
+    //     r = a - q*b
+    //
+    // We know from the construction of q that r satisfies:
+    //
+    //     0 <= r < ulp(q)*b
+    // 
+    // if r is greater than 1/2 ulp(q)*b, then q rounds up.  Otherwise, we
+    // already have the correct result.  The exact halfway case cannot occur.
+    // We also take this time to right shift quotient if it falls in the [1,2)
+    // range and adjust the exponent accordingly.
+    rep_t residual;
+    if (quotient < (implicitBit << 1)) {
+        residual = (aSignificand << 53) - quotient * bSignificand;
+        quotientExponent--;
+    } else {
+        quotient >>= 1;
+        residual = (aSignificand << 52) - quotient * bSignificand;
+    }
+    
+    const int writtenExponent = quotientExponent + exponentBias;
+    
+    if (writtenExponent >= maxExponent) {
+        // If we have overflowed the exponent, return infinity.
+        return fromRep(infRep | quotientSign);
+    }
+    
+    else if (writtenExponent < 1) {
+        // Flush denormals to zero.  In the future, it would be nice to add
+        // code to round them correctly.
+        return fromRep(quotientSign);
+    }
+    
+    else {
+        const bool round = (residual << 1) > bSignificand;
+        // Clear the implicit bit
+        rep_t absResult = quotient & significandMask;
+        // Insert the exponent
+        absResult |= (rep_t)writtenExponent << significandBits;
+        // Round
+        absResult += round;
+        // Insert the sign and return
+        const double result = fromRep(absResult | quotientSign);
+        return result;
+    }
+}
diff --git a/lib/builtins/divdi3.c b/lib/builtins/divdi3.c
new file mode 100644
index 000000000..2c2bcc26d
--- /dev/null
+++ b/lib/builtins/divdi3.c
@@ -0,0 +1,31 @@
+/* ===-- divdi3.c - Implement __divdi3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __divdi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+du_int COMPILER_RT_ABI __udivmoddi4(du_int a, du_int b, du_int* rem);
+
+/* Returns: a / b */
+
+COMPILER_RT_ABI di_int
+__divdi3(di_int a, di_int b)
+{
+    const int bits_in_dword_m1 = (int)(sizeof(di_int) * CHAR_BIT) - 1;
+    di_int s_a = a >> bits_in_dword_m1;           /* s_a = a < 0 ? -1 : 0 */
+    di_int s_b = b >> bits_in_dword_m1;           /* s_b = b < 0 ? -1 : 0 */
+    a = (a ^ s_a) - s_a;                         /* negate if s_a == -1 */
+    b = (b ^ s_b) - s_b;                         /* negate if s_b == -1 */
+    s_a ^= s_b;                                  /*sign of quotient */
+    return (__udivmoddi4(a, b, (du_int*)0) ^ s_a) - s_a;  /* negate if s_a == -1 */
+}
diff --git a/lib/builtins/divmoddi4.c b/lib/builtins/divmoddi4.c
new file mode 100644
index 000000000..2fe2b4818
--- /dev/null
+++ b/lib/builtins/divmoddi4.c
@@ -0,0 +1,27 @@
+/*===-- divmoddi4.c - Implement __divmoddi4 --------------------------------===
+ *
+ *                    The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __divmoddi4 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+extern COMPILER_RT_ABI di_int __divdi3(di_int a, di_int b);
+
+/* Returns: a / b, *rem = a % b  */
+
+COMPILER_RT_ABI di_int
+__divmoddi4(di_int a, di_int b, di_int* rem)
+{
+  di_int d = __divdi3(a,b);
+  *rem = a - (d*b);
+  return d;
+}
diff --git a/lib/builtins/divmodsi4.c b/lib/builtins/divmodsi4.c
new file mode 100644
index 000000000..c7f7b1a7c
--- /dev/null
+++ b/lib/builtins/divmodsi4.c
@@ -0,0 +1,30 @@
+/*===-- divmodsi4.c - Implement __divmodsi4 --------------------------------===
+ *
+ *                    The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __divmodsi4 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+extern COMPILER_RT_ABI si_int __divsi3(si_int a, si_int b);
+
+
+/* Returns: a / b, *rem = a % b  */
+
+COMPILER_RT_ABI si_int
+__divmodsi4(si_int a, si_int b, si_int* rem)
+{
+  si_int d = __divsi3(a,b);
+  *rem = a - (d*b);
+  return d; 
+}
+
+
diff --git a/lib/builtins/divsc3.c b/lib/builtins/divsc3.c
new file mode 100644
index 000000000..caa0c4075
--- /dev/null
+++ b/lib/builtins/divsc3.c
@@ -0,0 +1,60 @@
+/*===-- divsc3.c - Implement __divsc3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __divsc3 for the compiler_rt library.
+ *
+ *===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+#include "int_math.h"
+
+/* Returns: the quotient of (a + ib) / (c + id) */
+
+float _Complex
+__divsc3(float __a, float __b, float __c, float __d)
+{
+    int __ilogbw = 0;
+    float __logbw = crt_logbf(crt_fmaxf(crt_fabsf(__c), crt_fabsf(__d)));
+    if (crt_isfinite(__logbw))
+    {
+        __ilogbw = (int)__logbw;
+        __c = crt_scalbnf(__c, -__ilogbw);
+        __d = crt_scalbnf(__d, -__ilogbw);
+    }
+    float __denom = __c * __c + __d * __d;
+    float _Complex z;
+    __real__ z = crt_scalbnf((__a * __c + __b * __d) / __denom, -__ilogbw);
+    __imag__ z = crt_scalbnf((__b * __c - __a * __d) / __denom, -__ilogbw);
+    if (crt_isnan(__real__ z) && crt_isnan(__imag__ z))
+    {
+        if ((__denom == 0) && (!crt_isnan(__a) || !crt_isnan(__b)))
+        {
+            __real__ z = crt_copysignf(CRT_INFINITY, __c) * __a;
+            __imag__ z = crt_copysignf(CRT_INFINITY, __c) * __b;
+        }
+        else if ((crt_isinf(__a) || crt_isinf(__b)) &&
+                 crt_isfinite(__c) && crt_isfinite(__d))
+        {
+            __a = crt_copysignf(crt_isinf(__a) ? 1 : 0, __a);
+            __b = crt_copysignf(crt_isinf(__b) ? 1 : 0, __b);
+            __real__ z = CRT_INFINITY * (__a * __c + __b * __d);
+            __imag__ z = CRT_INFINITY * (__b * __c - __a * __d);
+        }
+        else if (crt_isinf(__logbw) && __logbw > 0 &&
+                 crt_isfinite(__a) && crt_isfinite(__b))
+        {
+            __c = crt_copysignf(crt_isinf(__c) ? 1 : 0, __c);
+            __d = crt_copysignf(crt_isinf(__d) ? 1 : 0, __d);
+            __real__ z = 0 * (__a * __c + __b * __d);
+            __imag__ z = 0 * (__b * __c - __a * __d);
+        }
+    }
+    return z;
+}
diff --git a/lib/builtins/divsf3.c b/lib/builtins/divsf3.c
new file mode 100644
index 000000000..c91c648fa
--- /dev/null
+++ b/lib/builtins/divsf3.c
@@ -0,0 +1,168 @@
+//===-- lib/divsf3.c - Single-precision division ------------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements single-precision soft-float division
+// with the IEEE-754 default rounding (to nearest, ties to even).
+//
+// For simplicity, this implementation currently flushes denormals to zero.
+// It should be a fairly straightforward exercise to implement gradual
+// underflow with correct rounding.
+//
+//===----------------------------------------------------------------------===//
+
+#define SINGLE_PRECISION
+#include "fp_lib.h"
+
+ARM_EABI_FNALIAS(fdiv, divsf3)
+
+fp_t __divsf3(fp_t a, fp_t b) {
+    
+    const unsigned int aExponent = toRep(a) >> significandBits & maxExponent;
+    const unsigned int bExponent = toRep(b) >> significandBits & maxExponent;
+    const rep_t quotientSign = (toRep(a) ^ toRep(b)) & signBit;
+    
+    rep_t aSignificand = toRep(a) & significandMask;
+    rep_t bSignificand = toRep(b) & significandMask;
+    int scale = 0;
+    
+    // Detect if a or b is zero, denormal, infinity, or NaN.
+    if (aExponent-1U >= maxExponent-1U || bExponent-1U >= maxExponent-1U) {
+        
+        const rep_t aAbs = toRep(a) & absMask;
+        const rep_t bAbs = toRep(b) & absMask;
+        
+        // NaN / anything = qNaN
+        if (aAbs > infRep) return fromRep(toRep(a) | quietBit);
+        // anything / NaN = qNaN
+        if (bAbs > infRep) return fromRep(toRep(b) | quietBit);
+        
+        if (aAbs == infRep) {
+            // infinity / infinity = NaN
+            if (bAbs == infRep) return fromRep(qnanRep);
+            // infinity / anything else = +/- infinity
+            else return fromRep(aAbs | quotientSign);
+        }
+        
+        // anything else / infinity = +/- 0
+        if (bAbs == infRep) return fromRep(quotientSign);
+        
+        if (!aAbs) {
+            // zero / zero = NaN
+            if (!bAbs) return fromRep(qnanRep);
+            // zero / anything else = +/- zero
+            else return fromRep(quotientSign);
+        }
+        // anything else / zero = +/- infinity
+        if (!bAbs) return fromRep(infRep | quotientSign);
+        
+        // one or both of a or b is denormal, the other (if applicable) is a
+        // normal number.  Renormalize one or both of a and b, and set scale to
+        // include the necessary exponent adjustment.
+        if (aAbs < implicitBit) scale += normalize(&aSignificand);
+        if (bAbs < implicitBit) scale -= normalize(&bSignificand);
+    }
+    
+    // Or in the implicit significand bit.  (If we fell through from the
+    // denormal path it was already set by normalize( ), but setting it twice
+    // won't hurt anything.)
+    aSignificand |= implicitBit;
+    bSignificand |= implicitBit;
+    int quotientExponent = aExponent - bExponent + scale;
+    
+    // Align the significand of b as a Q31 fixed-point number in the range
+    // [1, 2.0) and get a Q32 approximate reciprocal using a small minimax
+    // polynomial approximation: reciprocal = 3/4 + 1/sqrt(2) - b/2.  This
+    // is accurate to about 3.5 binary digits.
+    uint32_t q31b = bSignificand << 8;
+    uint32_t reciprocal = UINT32_C(0x7504f333) - q31b;
+    
+    // Now refine the reciprocal estimate using a Newton-Raphson iteration:
+    //
+    //     x1 = x0 * (2 - x0 * b)
+    //
+    // This doubles the number of correct binary digits in the approximation
+    // with each iteration, so after three iterations, we have about 28 binary
+    // digits of accuracy.
+    uint32_t correction;
+    correction = -((uint64_t)reciprocal * q31b >> 32);
+    reciprocal = (uint64_t)reciprocal * correction >> 31;
+    correction = -((uint64_t)reciprocal * q31b >> 32);
+    reciprocal = (uint64_t)reciprocal * correction >> 31;
+    correction = -((uint64_t)reciprocal * q31b >> 32);
+    reciprocal = (uint64_t)reciprocal * correction >> 31;
+    
+    // Exhaustive testing shows that the error in reciprocal after three steps
+    // is in the interval [-0x1.f58108p-31, 0x1.d0e48cp-29], in line with our
+    // expectations.  We bump the reciprocal by a tiny value to force the error
+    // to be strictly positive (in the range [0x1.4fdfp-37,0x1.287246p-29], to
+    // be specific).  This also causes 1/1 to give a sensible approximation
+    // instead of zero (due to overflow).
+    reciprocal -= 2;
+    
+    // The numerical reciprocal is accurate to within 2^-28, lies in the
+    // interval [0x1.000000eep-1, 0x1.fffffffcp-1], and is strictly smaller
+    // than the true reciprocal of b.  Multiplying a by this reciprocal thus
+    // gives a numerical q = a/b in Q24 with the following properties:
+    //
+    //    1. q < a/b
+    //    2. q is in the interval [0x1.000000eep-1, 0x1.fffffffcp0)
+    //    3. the error in q is at most 2^-24 + 2^-27 -- the 2^24 term comes
+    //       from the fact that we truncate the product, and the 2^27 term
+    //       is the error in the reciprocal of b scaled by the maximum
+    //       possible value of a.  As a consequence of this error bound,
+    //       either q or nextafter(q) is the correctly rounded 
+    rep_t quotient = (uint64_t)reciprocal*(aSignificand << 1) >> 32;
+    
+    // Two cases: quotient is in [0.5, 1.0) or quotient is in [1.0, 2.0).
+    // In either case, we are going to compute a residual of the form
+    //
+    //     r = a - q*b
+    //
+    // We know from the construction of q that r satisfies:
+    //
+    //     0 <= r < ulp(q)*b
+    // 
+    // if r is greater than 1/2 ulp(q)*b, then q rounds up.  Otherwise, we
+    // already have the correct result.  The exact halfway case cannot occur.
+    // We also take this time to right shift quotient if it falls in the [1,2)
+    // range and adjust the exponent accordingly.
+    rep_t residual;
+    if (quotient < (implicitBit << 1)) {
+        residual = (aSignificand << 24) - quotient * bSignificand;
+        quotientExponent--;
+    } else {
+        quotient >>= 1;
+        residual = (aSignificand << 23) - quotient * bSignificand;
+    }
+
+    const int writtenExponent = quotientExponent + exponentBias;
+    
+    if (writtenExponent >= maxExponent) {
+        // If we have overflowed the exponent, return infinity.
+        return fromRep(infRep | quotientSign);
+    }
+    
+    else if (writtenExponent < 1) {
+        // Flush denormals to zero.  In the future, it would be nice to add
+        // code to round them correctly.
+        return fromRep(quotientSign);
+    }
+    
+    else {
+        const bool round = (residual << 1) > bSignificand;
+        // Clear the implicit bit
+        rep_t absResult = quotient & significandMask;
+        // Insert the exponent
+        absResult |= (rep_t)writtenExponent << significandBits;
+        // Round
+        absResult += round;
+        // Insert the sign and return
+        return fromRep(absResult | quotientSign);
+    }
+}
diff --git a/lib/builtins/divsi3.c b/lib/builtins/divsi3.c
new file mode 100644
index 000000000..cd19de952
--- /dev/null
+++ b/lib/builtins/divsi3.c
@@ -0,0 +1,39 @@
+/* ===-- divsi3.c - Implement __divsi3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __divsi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+su_int COMPILER_RT_ABI __udivsi3(su_int n, su_int d);
+
+/* Returns: a / b */
+
+ARM_EABI_FNALIAS(idiv, divsi3)
+
+COMPILER_RT_ABI si_int
+__divsi3(si_int a, si_int b)
+{
+    const int bits_in_word_m1 = (int)(sizeof(si_int) * CHAR_BIT) - 1;
+    si_int s_a = a >> bits_in_word_m1;           /* s_a = a < 0 ? -1 : 0 */
+    si_int s_b = b >> bits_in_word_m1;           /* s_b = b < 0 ? -1 : 0 */
+    a = (a ^ s_a) - s_a;                         /* negate if s_a == -1 */
+    b = (b ^ s_b) - s_b;                         /* negate if s_b == -1 */
+    s_a ^= s_b;                                  /* sign of quotient */
+    /*
+     * On CPUs without unsigned hardware division support,
+     *  this calls __udivsi3 (notice the cast to su_int).
+     * On CPUs with unsigned hardware division support,
+     *  this uses the unsigned division instruction.
+     */
+    return ((su_int)a/(su_int)b ^ s_a) - s_a;    /* negate if s_a == -1 */
+}
diff --git a/lib/builtins/divti3.c b/lib/builtins/divti3.c
new file mode 100644
index 000000000..0242c1368
--- /dev/null
+++ b/lib/builtins/divti3.c
@@ -0,0 +1,35 @@
+/* ===-- divti3.c - Implement __divti3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __divti3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+tu_int __udivmodti4(tu_int a, tu_int b, tu_int* rem);
+
+/* Returns: a / b */
+
+ti_int
+__divti3(ti_int a, ti_int b)
+{
+    const int bits_in_tword_m1 = (int)(sizeof(ti_int) * CHAR_BIT) - 1;
+    ti_int s_a = a >> bits_in_tword_m1;           /* s_a = a < 0 ? -1 : 0 */
+    ti_int s_b = b >> bits_in_tword_m1;           /* s_b = b < 0 ? -1 : 0 */
+    a = (a ^ s_a) - s_a;                         /* negate if s_a == -1 */
+    b = (b ^ s_b) - s_b;                         /* negate if s_b == -1 */
+    s_a ^= s_b;                                  /* sign of quotient */
+    return (__udivmodti4(a, b, (tu_int*)0) ^ s_a) - s_a;  /* negate if s_a == -1 */
+}
+
+#endif
diff --git a/lib/builtins/divxc3.c b/lib/builtins/divxc3.c
new file mode 100644
index 000000000..5f240e956
--- /dev/null
+++ b/lib/builtins/divxc3.c
@@ -0,0 +1,63 @@
+/* ===-- divxc3.c - Implement __divxc3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __divxc3 for the compiler_rt library.
+ *
+ */
+
+#if !_ARCH_PPC
+
+#include "int_lib.h"
+#include "int_math.h"
+
+/* Returns: the quotient of (a + ib) / (c + id) */
+
+long double _Complex
+__divxc3(long double __a, long double __b, long double __c, long double __d)
+{
+    int __ilogbw = 0;
+    long double __logbw = crt_logbl(crt_fmaxl(crt_fabsl(__c), crt_fabsl(__d)));
+    if (crt_isfinite(__logbw))
+    {
+        __ilogbw = (int)__logbw;
+        __c = crt_scalbnl(__c, -__ilogbw);
+        __d = crt_scalbnl(__d, -__ilogbw);
+    }
+    long double __denom = __c * __c + __d * __d;
+    long double _Complex z;
+    __real__ z = crt_scalbnl((__a * __c + __b * __d) / __denom, -__ilogbw);
+    __imag__ z = crt_scalbnl((__b * __c - __a * __d) / __denom, -__ilogbw);
+    if (crt_isnan(__real__ z) && crt_isnan(__imag__ z))
+    {
+        if ((__denom == 0) && (!crt_isnan(__a) || !crt_isnan(__b)))
+        {
+            __real__ z = crt_copysignl(CRT_INFINITY, __c) * __a;
+            __imag__ z = crt_copysignl(CRT_INFINITY, __c) * __b;
+        }
+        else if ((crt_isinf(__a) || crt_isinf(__b)) &&
+                 crt_isfinite(__c) && crt_isfinite(__d))
+        {
+            __a = crt_copysignl(crt_isinf(__a) ? 1 : 0, __a);
+            __b = crt_copysignl(crt_isinf(__b) ? 1 : 0, __b);
+            __real__ z = CRT_INFINITY * (__a * __c + __b * __d);
+            __imag__ z = CRT_INFINITY * (__b * __c - __a * __d);
+        }
+        else if (crt_isinf(__logbw) && __logbw > 0 &&
+                 crt_isfinite(__a) && crt_isfinite(__b))
+        {
+            __c = crt_copysignl(crt_isinf(__c) ? 1 : 0, __c);
+            __d = crt_copysignl(crt_isinf(__d) ? 1 : 0, __d);
+            __real__ z = 0 * (__a * __c + __b * __d);
+            __imag__ z = 0 * (__b * __c - __a * __d);
+        }
+    }
+    return z;
+}
+
+#endif
diff --git a/lib/builtins/enable_execute_stack.c b/lib/builtins/enable_execute_stack.c
new file mode 100644
index 000000000..278ca246f
--- /dev/null
+++ b/lib/builtins/enable_execute_stack.c
@@ -0,0 +1,59 @@
+/* ===-- enable_execute_stack.c - Implement __enable_execute_stack ---------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#include <sys/mman.h>
+
+/* #include "config.h"
+ * FIXME: CMake - include when cmake system is ready.
+ * Remove #define HAVE_SYSCONF 1 line.
+ */
+#define HAVE_SYSCONF 1
+
+#ifndef __APPLE__
+#include <unistd.h>
+#endif /* __APPLE__ */
+
+#if __LP64__
+	#define TRAMPOLINE_SIZE 48
+#else
+	#define TRAMPOLINE_SIZE 40
+#endif
+
+/*
+ * The compiler generates calls to __enable_execute_stack() when creating 
+ * trampoline functions on the stack for use with nested functions.
+ * It is expected to mark the page(s) containing the address 
+ * and the next 48 bytes as executable.  Since the stack is normally rw-
+ * that means changing the protection on those page(s) to rwx. 
+ */
+
+void __enable_execute_stack(void* addr)
+{
+
+#if __APPLE__
+	/* On Darwin, pagesize is always 4096 bytes */
+	const uintptr_t pageSize = 4096;
+#elif !defined(HAVE_SYSCONF)
+#error "HAVE_SYSCONF not defined! See enable_execute_stack.c"
+#else
+        const uintptr_t pageSize = sysconf(_SC_PAGESIZE);
+#endif /* __APPLE__ */
+
+	const uintptr_t pageAlignMask = ~(pageSize-1);
+	uintptr_t p = (uintptr_t)addr;
+	unsigned char* startPage = (unsigned char*)(p & pageAlignMask);
+	unsigned char* endPage = (unsigned char*)((p+TRAMPOLINE_SIZE+pageSize) & pageAlignMask);
+	size_t length = endPage - startPage;
+	(void) mprotect((void *)startPage, length, PROT_READ | PROT_WRITE | PROT_EXEC);
+}
+
+
diff --git a/lib/builtins/eprintf.c b/lib/builtins/eprintf.c
new file mode 100644
index 000000000..3626dbf8b
--- /dev/null
+++ b/lib/builtins/eprintf.c
@@ -0,0 +1,34 @@
+/* ===---------- eprintf.c - Implements __eprintf --------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+
+
+#include "int_lib.h"
+#include <stdio.h>
+
+
+/*
+ * __eprintf() was used in an old version of <assert.h>.
+ * It can eventually go away, but it is needed when linking
+ * .o files built with the old <assert.h>.
+ *
+ * It should never be exported from a dylib, so it is marked
+ * visibility hidden.
+ */
+#ifndef _WIN32
+__attribute__((visibility("hidden")))
+#endif
+void __eprintf(const char* format, const char* assertion_expression,
+				const char* line, const char* file)
+{
+	fprintf(stderr, format, assertion_expression, line, file);
+	fflush(stderr);
+	compilerrt_abort();
+}
diff --git a/lib/builtins/extendsfdf2.c b/lib/builtins/extendsfdf2.c
new file mode 100644
index 000000000..91fd2b436
--- /dev/null
+++ b/lib/builtins/extendsfdf2.c
@@ -0,0 +1,137 @@
+//===-- lib/extendsfdf2.c - single -> double conversion -----------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements a fairly generic conversion from a narrower to a wider
+// IEEE-754 floating-point type.  The constants and types defined following the
+// includes below parameterize the conversion.
+//
+// This routine can be trivially adapted to support conversions from 
+// half-precision or to quad-precision. It does not support types that don't
+// use the usual IEEE-754 interchange formats; specifically, some work would be
+// needed to adapt it to (for example) the Intel 80-bit format or PowerPC
+// double-double format.
+//
+// Note please, however, that this implementation is only intended to support
+// *widening* operations; if you need to convert to a *narrower* floating-point
+// type (e.g. double -> float), then this routine will not do what you want it
+// to.
+//
+// It also requires that integer types at least as large as both formats
+// are available on the target platform; this may pose a problem when trying
+// to add support for quad on some 32-bit systems, for example.  You also may
+// run into trouble finding an appropriate CLZ function for wide source types;
+// you will likely need to roll your own on some platforms.
+//
+// Finally, the following assumptions are made:
+//
+// 1. floating-point types and integer types have the same endianness on the
+//    target platform
+//
+// 2. quiet NaNs, if supported, are indicated by the leading bit of the
+//    significand field being set
+//
+//===----------------------------------------------------------------------===//
+
+#include "int_lib.h"
+
+typedef float src_t;
+typedef uint32_t src_rep_t;
+#define SRC_REP_C UINT32_C
+static const int srcSigBits = 23;
+#define src_rep_t_clz __builtin_clz
+
+typedef double dst_t;
+typedef uint64_t dst_rep_t;
+#define DST_REP_C UINT64_C
+static const int dstSigBits = 52;
+
+// End of specialization parameters.  Two helper routines for conversion to and
+// from the representation of floating-point data as integer values follow.
+
+static inline src_rep_t srcToRep(src_t x) {
+    const union { src_t f; src_rep_t i; } rep = {.f = x};
+    return rep.i;
+}
+
+static inline dst_t dstFromRep(dst_rep_t x) {
+    const union { dst_t f; dst_rep_t i; } rep = {.i = x};
+    return rep.f;
+}
+
+// End helper routines.  Conversion implementation follows.
+
+ARM_EABI_FNALIAS(f2d, extendsfdf2)
+
+dst_t __extendsfdf2(src_t a) {
+    
+    // Various constants whose values follow from the type parameters.
+    // Any reasonable optimizer will fold and propagate all of these.
+    const int srcBits = sizeof(src_t)*CHAR_BIT;
+    const int srcExpBits = srcBits - srcSigBits - 1;
+    const int srcInfExp = (1 << srcExpBits) - 1;
+    const int srcExpBias = srcInfExp >> 1;
+    
+    const src_rep_t srcMinNormal = SRC_REP_C(1) << srcSigBits;
+    const src_rep_t srcInfinity = (src_rep_t)srcInfExp << srcSigBits;
+    const src_rep_t srcSignMask = SRC_REP_C(1) << (srcSigBits + srcExpBits);
+    const src_rep_t srcAbsMask = srcSignMask - 1;
+    const src_rep_t srcQNaN = SRC_REP_C(1) << (srcSigBits - 1);
+    const src_rep_t srcNaNCode = srcQNaN - 1;
+    
+    const int dstBits = sizeof(dst_t)*CHAR_BIT;
+    const int dstExpBits = dstBits - dstSigBits - 1;
+    const int dstInfExp = (1 << dstExpBits) - 1;
+    const int dstExpBias = dstInfExp >> 1;
+    
+    const dst_rep_t dstMinNormal = DST_REP_C(1) << dstSigBits;
+    
+    // Break a into a sign and representation of the absolute value
+    const src_rep_t aRep = srcToRep(a);
+    const src_rep_t aAbs = aRep & srcAbsMask;
+    const src_rep_t sign = aRep & srcSignMask;
+    dst_rep_t absResult;
+    
+    if (aAbs - srcMinNormal < srcInfinity - srcMinNormal) {
+        // a is a normal number.
+        // Extend to the destination type by shifting the significand and
+        // exponent into the proper position and rebiasing the exponent.
+        absResult = (dst_rep_t)aAbs << (dstSigBits - srcSigBits);
+        absResult += (dst_rep_t)(dstExpBias - srcExpBias) << dstSigBits;
+    }
+    
+    else if (aAbs >= srcInfinity) {
+        // a is NaN or infinity.
+        // Conjure the result by beginning with infinity, then setting the qNaN
+        // bit (if needed) and right-aligning the rest of the trailing NaN
+        // payload field.
+        absResult = (dst_rep_t)dstInfExp << dstSigBits;
+        absResult |= (dst_rep_t)(aAbs & srcQNaN) << (dstSigBits - srcSigBits);
+        absResult |= aAbs & srcNaNCode;
+    }
+    
+    else if (aAbs) {
+        // a is denormal.
+        // renormalize the significand and clear the leading bit, then insert
+        // the correct adjusted exponent in the destination type.
+        const int scale = src_rep_t_clz(aAbs) - src_rep_t_clz(srcMinNormal);
+        absResult = (dst_rep_t)aAbs << (dstSigBits - srcSigBits + scale);
+        absResult ^= dstMinNormal;
+        const int resultExponent = dstExpBias - srcExpBias - scale + 1;
+        absResult |= (dst_rep_t)resultExponent << dstSigBits;
+    }
+
+    else {
+        // a is zero.
+        absResult = 0;
+    }
+    
+    // Apply the signbit to (dst_t)abs(a).
+    const dst_rep_t result = absResult | (dst_rep_t)sign << (dstBits - srcBits);
+    return dstFromRep(result);
+}
diff --git a/lib/builtins/ffsdi2.c b/lib/builtins/ffsdi2.c
new file mode 100644
index 000000000..a5ac9900f
--- /dev/null
+++ b/lib/builtins/ffsdi2.c
@@ -0,0 +1,33 @@
+/* ===-- ffsdi2.c - Implement __ffsdi2 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __ffsdi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: the index of the least significant 1-bit in a, or
+ * the value zero if a is zero. The least significant bit is index one.
+ */
+
+COMPILER_RT_ABI si_int
+__ffsdi2(di_int a)
+{
+    dwords x;
+    x.all = a;
+    if (x.s.low == 0)
+    {
+        if (x.s.high == 0)
+            return 0;
+        return __builtin_ctz(x.s.high) + (1 + sizeof(si_int) * CHAR_BIT);
+    }
+    return __builtin_ctz(x.s.low) + 1;
+}
diff --git a/lib/builtins/ffsti2.c b/lib/builtins/ffsti2.c
new file mode 100644
index 000000000..27e15d581
--- /dev/null
+++ b/lib/builtins/ffsti2.c
@@ -0,0 +1,37 @@
+/* ===-- ffsti2.c - Implement __ffsti2 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __ffsti2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: the index of the least significant 1-bit in a, or
+ * the value zero if a is zero. The least significant bit is index one.
+ */
+
+si_int
+__ffsti2(ti_int a)
+{
+    twords x;
+    x.all = a;
+    if (x.s.low == 0)
+    {
+        if (x.s.high == 0)
+            return 0;
+        return __builtin_ctzll(x.s.high) + (1 + sizeof(di_int) * CHAR_BIT);
+    }
+    return __builtin_ctzll(x.s.low) + 1;
+}
+
+#endif /* __x86_64 */
diff --git a/lib/builtins/fixdfdi.c b/lib/builtins/fixdfdi.c
new file mode 100644
index 000000000..7665ea5a3
--- /dev/null
+++ b/lib/builtins/fixdfdi.c
@@ -0,0 +1,45 @@
+/* ===-- fixdfdi.c - Implement __fixdfdi -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixdfdi for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: convert a to a signed long long, rounding toward zero. */
+
+/* Assumption: double is a IEEE 64 bit floating point type 
+ *            su_int is a 32 bit integral type
+ *            value in double is representable in di_int (no range checking performed)
+ */
+
+/* seee eeee eeee mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm */
+
+ARM_EABI_FNALIAS(d2lz, fixdfdi)
+
+di_int
+__fixdfdi(double a)
+{
+    double_bits fb;
+    fb.f = a;
+    int e = ((fb.u.s.high & 0x7FF00000) >> 20) - 1023;
+    if (e < 0)
+        return 0;
+    di_int s = (si_int)(fb.u.s.high & 0x80000000) >> 31;
+    dwords r;
+    r.s.high = (fb.u.s.high & 0x000FFFFF) | 0x00100000;
+    r.s.low = fb.u.s.low;
+    if (e > 52)
+        r.all <<= (e - 52);
+    else
+        r.all >>= (52 - e);
+    return (r.all ^ s) - s;
+} 
diff --git a/lib/builtins/fixdfsi.c b/lib/builtins/fixdfsi.c
new file mode 100644
index 000000000..614d032a4
--- /dev/null
+++ b/lib/builtins/fixdfsi.c
@@ -0,0 +1,49 @@
+//===-- lib/fixdfsi.c - Double-precision -> integer conversion ----*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements double-precision to integer conversion for the
+// compiler-rt library.  No range checking is performed; the behavior of this
+// conversion is undefined for out of range values in the C standard.
+//
+//===----------------------------------------------------------------------===//
+
+#define DOUBLE_PRECISION
+#include "fp_lib.h"
+
+#include "int_lib.h"
+
+ARM_EABI_FNALIAS(d2iz, fixdfsi)
+
+int __fixdfsi(fp_t a) {
+    
+    // Break a into sign, exponent, significand
+    const rep_t aRep = toRep(a);
+    const rep_t aAbs = aRep & absMask;
+    const int sign = aRep & signBit ? -1 : 1;
+    const int exponent = (aAbs >> significandBits) - exponentBias;
+    const rep_t significand = (aAbs & significandMask) | implicitBit;
+    
+    // If 0 < exponent < significandBits, right shift to get the result.
+    if ((unsigned int)exponent < significandBits) {
+        return sign * (significand >> (significandBits - exponent));
+    }
+    
+    // If exponent is negative, the result is zero.
+    else if (exponent < 0) {
+        return 0;
+    }
+    
+    // If significandBits < exponent, left shift to get the result.  This shift
+    // may end up being larger than the type width, which incurs undefined
+    // behavior, but the conversion itself is undefined in that case, so
+    // whatever the compiler decides to do is fine.
+    else {
+        return sign * (significand << (exponent - significandBits));
+    }
+}
diff --git a/lib/builtins/fixdfti.c b/lib/builtins/fixdfti.c
new file mode 100644
index 000000000..b110a94b8
--- /dev/null
+++ b/lib/builtins/fixdfti.c
@@ -0,0 +1,45 @@
+/* ===-- fixdfti.c - Implement __fixdfti -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixdfti for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: convert a to a signed long long, rounding toward zero. */
+
+/* Assumption: double is a IEEE 64 bit floating point type 
+ *             su_int is a 32 bit integral type
+ *             value in double is representable in ti_int (no range checking performed)
+ */
+
+/* seee eeee eeee mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm */
+
+ti_int
+__fixdfti(double a)
+{
+    double_bits fb;
+    fb.f = a;
+    int e = ((fb.u.s.high & 0x7FF00000) >> 20) - 1023;
+    if (e < 0)
+        return 0;
+    ti_int s = (si_int)(fb.u.s.high & 0x80000000) >> 31;
+    ti_int r = 0x0010000000000000uLL | (0x000FFFFFFFFFFFFFuLL & fb.u.all);
+    if (e > 52)
+        r <<= (e - 52);
+    else
+        r >>= (52 - e);
+    return (r ^ s) - s;
+}
+
+#endif
diff --git a/lib/builtins/fixsfdi.c b/lib/builtins/fixsfdi.c
new file mode 100644
index 000000000..4f6cfdd7a
--- /dev/null
+++ b/lib/builtins/fixsfdi.c
@@ -0,0 +1,43 @@
+/* ===-- fixsfdi.c - Implement __fixsfdi -----------------------------------===
+ *
+ *                    The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixsfdi for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: convert a to a signed long long, rounding toward zero. */
+
+/* Assumption: float is a IEEE 32 bit floating point type 
+ *             su_int is a 32 bit integral type
+ *             value in float is representable in di_int (no range checking performed)
+ */
+
+/* seee eeee emmm mmmm mmmm mmmm mmmm mmmm */
+
+ARM_EABI_FNALIAS(f2lz, fixsfdi)
+
+COMPILER_RT_ABI di_int
+__fixsfdi(float a)
+{
+    float_bits fb;
+    fb.f = a;
+    int e = ((fb.u & 0x7F800000) >> 23) - 127;
+    if (e < 0)
+        return 0;
+    di_int s = (si_int)(fb.u & 0x80000000) >> 31;
+    di_int r = (fb.u & 0x007FFFFF) | 0x00800000;
+    if (e > 23)
+        r <<= (e - 23);
+    else
+        r >>= (23 - e);
+    return (r ^ s) - s;
+}
diff --git a/lib/builtins/fixsfsi.c b/lib/builtins/fixsfsi.c
new file mode 100644
index 000000000..e3cc42d52
--- /dev/null
+++ b/lib/builtins/fixsfsi.c
@@ -0,0 +1,47 @@
+//===-- lib/fixsfsi.c - Single-precision -> integer conversion ----*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements single-precision to integer conversion for the
+// compiler-rt library.  No range checking is performed; the behavior of this
+// conversion is undefined for out of range values in the C standard.
+//
+//===----------------------------------------------------------------------===//
+
+#define SINGLE_PRECISION
+#include "fp_lib.h"
+
+ARM_EABI_FNALIAS(f2iz, fixsfsi)
+
+COMPILER_RT_ABI int
+__fixsfsi(fp_t a) {
+    // Break a into sign, exponent, significand
+    const rep_t aRep = toRep(a);
+    const rep_t aAbs = aRep & absMask;
+    const int sign = aRep & signBit ? -1 : 1;
+    const int exponent = (aAbs >> significandBits) - exponentBias;
+    const rep_t significand = (aAbs & significandMask) | implicitBit;
+    
+    // If 0 < exponent < significandBits, right shift to get the result.
+    if ((unsigned int)exponent < significandBits) {
+        return sign * (significand >> (significandBits - exponent));
+    }
+    
+    // If exponent is negative, the result is zero.
+    else if (exponent < 0) {
+        return 0;
+    }
+    
+    // If significandBits < exponent, left shift to get the result.  This shift
+    // may end up being larger than the type width, which incurs undefined
+    // behavior, but the conversion itself is undefined in that case, so
+    // whatever the compiler decides to do is fine.
+    else {
+        return sign * (significand << (exponent - significandBits));
+    }
+}
diff --git a/lib/builtins/fixsfti.c b/lib/builtins/fixsfti.c
new file mode 100644
index 000000000..c730ae046
--- /dev/null
+++ b/lib/builtins/fixsfti.c
@@ -0,0 +1,45 @@
+/* ===-- fixsfti.c - Implement __fixsfti -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixsfti for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: convert a to a signed long long, rounding toward zero. */
+
+/* Assumption: float is a IEEE 32 bit floating point type 
+ *             su_int is a 32 bit integral type
+ *             value in float is representable in ti_int (no range checking performed)
+ */
+
+/* seee eeee emmm mmmm mmmm mmmm mmmm mmmm */
+
+ti_int
+__fixsfti(float a)
+{
+    float_bits fb;
+    fb.f = a;
+    int e = ((fb.u & 0x7F800000) >> 23) - 127;
+    if (e < 0)
+        return 0;
+    ti_int s = (si_int)(fb.u & 0x80000000) >> 31;
+    ti_int r = (fb.u & 0x007FFFFF) | 0x00800000;
+    if (e > 23)
+        r <<= (e - 23);
+    else
+        r >>= (23 - e);
+    return (r ^ s) - s;
+}
+
+#endif
diff --git a/lib/builtins/fixunsdfdi.c b/lib/builtins/fixunsdfdi.c
new file mode 100644
index 000000000..9e6371390
--- /dev/null
+++ b/lib/builtins/fixunsdfdi.c
@@ -0,0 +1,47 @@
+/* ===-- fixunsdfdi.c - Implement __fixunsdfdi -----------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixunsdfdi for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: convert a to a unsigned long long, rounding toward zero.
+ *          Negative values all become zero.
+ */
+
+/* Assumption: double is a IEEE 64 bit floating point type 
+ *             du_int is a 64 bit integral type
+ *             value in double is representable in du_int or is negative 
+ *                 (no range checking performed)
+ */
+
+/* seee eeee eeee mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm */
+
+ARM_EABI_FNALIAS(d2ulz, fixunsdfdi)
+
+COMPILER_RT_ABI du_int
+__fixunsdfdi(double a)
+{
+    double_bits fb;
+    fb.f = a;
+    int e = ((fb.u.s.high & 0x7FF00000) >> 20) - 1023;
+    if (e < 0 || (fb.u.s.high & 0x80000000))
+        return 0;
+    udwords r;
+    r.s.high = (fb.u.s.high & 0x000FFFFF) | 0x00100000;
+    r.s.low = fb.u.s.low;
+    if (e > 52)
+        r.all <<= (e - 52);
+    else
+        r.all >>= (52 - e);
+    return r.all;
+}
diff --git a/lib/builtins/fixunsdfsi.c b/lib/builtins/fixunsdfsi.c
new file mode 100644
index 000000000..c6a3c755e
--- /dev/null
+++ b/lib/builtins/fixunsdfsi.c
@@ -0,0 +1,44 @@
+/* ===-- fixunsdfsi.c - Implement __fixunsdfsi -----------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixunsdfsi for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: convert a to a unsigned int, rounding toward zero.
+ *          Negative values all become zero.
+ */
+
+/* Assumption: double is a IEEE 64 bit floating point type 
+ *             su_int is a 32 bit integral type
+ *             value in double is representable in su_int or is negative 
+ *                 (no range checking performed)
+ */
+
+/* seee eeee eeee mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm */
+
+ARM_EABI_FNALIAS(d2uiz, fixunsdfsi)
+
+COMPILER_RT_ABI su_int
+__fixunsdfsi(double a)
+{
+    double_bits fb;
+    fb.f = a;
+    int e = ((fb.u.s.high & 0x7FF00000) >> 20) - 1023;
+    if (e < 0 || (fb.u.s.high & 0x80000000))
+        return 0;
+    return (
+                0x80000000u                      |
+                ((fb.u.s.high & 0x000FFFFF) << 11) |
+                (fb.u.s.low >> 21)
+           ) >> (31 - e);
+}
diff --git a/lib/builtins/fixunsdfti.c b/lib/builtins/fixunsdfti.c
new file mode 100644
index 000000000..fb0336f69
--- /dev/null
+++ b/lib/builtins/fixunsdfti.c
@@ -0,0 +1,47 @@
+/* ===-- fixunsdfti.c - Implement __fixunsdfti -----------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixunsdfti for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: convert a to a unsigned long long, rounding toward zero.
+ *          Negative values all become zero.
+ */
+
+/* Assumption: double is a IEEE 64 bit floating point type 
+ *             tu_int is a 64 bit integral type
+ *             value in double is representable in tu_int or is negative 
+ *                 (no range checking performed)
+ */
+
+/* seee eeee eeee mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm */
+
+tu_int
+__fixunsdfti(double a)
+{
+    double_bits fb;
+    fb.f = a;
+    int e = ((fb.u.s.high & 0x7FF00000) >> 20) - 1023;
+    if (e < 0 || (fb.u.s.high & 0x80000000))
+        return 0;
+    tu_int r = 0x0010000000000000uLL | (fb.u.all & 0x000FFFFFFFFFFFFFuLL);
+    if (e > 52)
+        r <<= (e - 52);
+    else
+        r >>= (52 - e);
+    return r;
+}
+
+#endif
diff --git a/lib/builtins/fixunssfdi.c b/lib/builtins/fixunssfdi.c
new file mode 100644
index 000000000..69d5952e9
--- /dev/null
+++ b/lib/builtins/fixunssfdi.c
@@ -0,0 +1,44 @@
+/* ===-- fixunssfdi.c - Implement __fixunssfdi -----------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixunssfdi for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+/* Returns: convert a to a unsigned long long, rounding toward zero.
+ *          Negative values all become zero.
+ */
+
+/* Assumption: float is a IEEE 32 bit floating point type 
+ *             du_int is a 64 bit integral type
+ *             value in float is representable in du_int or is negative 
+ *                 (no range checking performed)
+ */
+
+/* seee eeee emmm mmmm mmmm mmmm mmmm mmmm */
+
+ARM_EABI_FNALIAS(f2ulz, fixunssfdi)
+
+COMPILER_RT_ABI du_int
+__fixunssfdi(float a)
+{
+    float_bits fb;
+    fb.f = a;
+    int e = ((fb.u & 0x7F800000) >> 23) - 127;
+    if (e < 0 || (fb.u & 0x80000000))
+        return 0;
+    du_int r = (fb.u & 0x007FFFFF) | 0x00800000;
+    if (e > 23)
+        r <<= (e - 23);
+    else
+        r >>= (23 - e);
+    return r;
+}
diff --git a/lib/builtins/fixunssfsi.c b/lib/builtins/fixunssfsi.c
new file mode 100644
index 000000000..e034139ea
--- /dev/null
+++ b/lib/builtins/fixunssfsi.c
@@ -0,0 +1,45 @@
+/* ===-- fixunssfsi.c - Implement __fixunssfsi -----------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixunssfsi for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: convert a to a unsigned int, rounding toward zero.
+ *          Negative values all become zero.
+ */
+
+/* Assumption: float is a IEEE 32 bit floating point type 
+ *             su_int is a 32 bit integral type
+ *             value in float is representable in su_int or is negative 
+ *                 (no range checking performed)
+ */
+
+/* seee eeee emmm mmmm mmmm mmmm mmmm mmmm */
+
+ARM_EABI_FNALIAS(f2uiz, fixunssfsi)
+
+COMPILER_RT_ABI su_int
+__fixunssfsi(float a)
+{
+    float_bits fb;
+    fb.f = a;
+    int e = ((fb.u & 0x7F800000) >> 23) - 127;
+    if (e < 0 || (fb.u & 0x80000000))
+        return 0;
+    su_int r = (fb.u & 0x007FFFFF) | 0x00800000;
+    if (e > 23)
+        r <<= (e - 23);
+    else
+        r >>= (23 - e);
+    return r;
+}
diff --git a/lib/builtins/fixunssfti.c b/lib/builtins/fixunssfti.c
new file mode 100644
index 000000000..8f4c62627
--- /dev/null
+++ b/lib/builtins/fixunssfti.c
@@ -0,0 +1,47 @@
+/* ===-- fixunssfti.c - Implement __fixunssfti -----------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixunssfti for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: convert a to a unsigned long long, rounding toward zero.
+ *          Negative values all become zero.
+ */
+
+/* Assumption: float is a IEEE 32 bit floating point type 
+ *             tu_int is a 64 bit integral type
+ *             value in float is representable in tu_int or is negative 
+ *                 (no range checking performed)
+ */
+
+/* seee eeee emmm mmmm mmmm mmmm mmmm mmmm */
+
+tu_int
+__fixunssfti(float a)
+{
+    float_bits fb;
+    fb.f = a;
+    int e = ((fb.u & 0x7F800000) >> 23) - 127;
+    if (e < 0 || (fb.u & 0x80000000))
+        return 0;
+    tu_int r = (fb.u & 0x007FFFFF) | 0x00800000;
+    if (e > 23)
+        r <<= (e - 23);
+    else
+        r >>= (23 - e);
+    return r;
+}
+
+#endif
diff --git a/lib/builtins/fixunsxfdi.c b/lib/builtins/fixunsxfdi.c
new file mode 100644
index 000000000..6c817d8e1
--- /dev/null
+++ b/lib/builtins/fixunsxfdi.c
@@ -0,0 +1,44 @@
+/* ===-- fixunsxfdi.c - Implement __fixunsxfdi -----------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixunsxfdi for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#if !_ARCH_PPC
+
+#include "int_lib.h"
+
+/* Returns: convert a to a unsigned long long, rounding toward zero.
+ *          Negative values all become zero.
+ */
+
+/* Assumption: long double is an intel 80 bit floating point type padded with 6 bytes
+ *             du_int is a 64 bit integral type
+ *             value in long double is representable in du_int or is negative 
+ *                 (no range checking performed)
+ */
+
+/* gggg gggg gggg gggg gggg gggg gggg gggg | gggg gggg gggg gggg seee eeee eeee eeee |
+ * 1mmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm
+ */
+
+du_int
+__fixunsxfdi(long double a)
+{
+    long_double_bits fb;
+    fb.f = a;
+    int e = (fb.u.high.s.low & 0x00007FFF) - 16383;
+    if (e < 0 || (fb.u.high.s.low & 0x00008000))
+        return 0;
+    return fb.u.low.all >> (63 - e);
+}
+
+#endif
diff --git a/lib/builtins/fixunsxfsi.c b/lib/builtins/fixunsxfsi.c
new file mode 100644
index 000000000..b9da86c4f
--- /dev/null
+++ b/lib/builtins/fixunsxfsi.c
@@ -0,0 +1,44 @@
+/* ===-- fixunsxfsi.c - Implement __fixunsxfsi -----------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixunsxfsi for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#if !_ARCH_PPC
+
+#include "int_lib.h"
+
+/* Returns: convert a to a unsigned int, rounding toward zero.
+ *          Negative values all become zero.
+ */
+
+/* Assumption: long double is an intel 80 bit floating point type padded with 6 bytes
+ *             su_int is a 32 bit integral type
+ *             value in long double is representable in su_int or is negative 
+ *                 (no range checking performed)
+ */
+
+/* gggg gggg gggg gggg gggg gggg gggg gggg | gggg gggg gggg gggg seee eeee eeee eeee |
+ * 1mmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm
+ */
+
+su_int
+__fixunsxfsi(long double a)
+{
+    long_double_bits fb;
+    fb.f = a;
+    int e = (fb.u.high.s.low & 0x00007FFF) - 16383;
+    if (e < 0 || (fb.u.high.s.low & 0x00008000))
+        return 0;
+    return fb.u.low.s.high >> (31 - e);
+}
+
+#endif /* !_ARCH_PPC */
diff --git a/lib/builtins/fixunsxfti.c b/lib/builtins/fixunsxfti.c
new file mode 100644
index 000000000..260bfc01f
--- /dev/null
+++ b/lib/builtins/fixunsxfti.c
@@ -0,0 +1,49 @@
+/* ===-- fixunsxfti.c - Implement __fixunsxfti -----------------------------===
+ *
+ *      	       The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixunsxfti for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: convert a to a unsigned long long, rounding toward zero.
+ *          Negative values all become zero.
+ */
+
+/* Assumption: long double is an intel 80 bit floating point type padded with 6 bytes
+ *             tu_int is a 64 bit integral type
+ *             value in long double is representable in tu_int or is negative 
+ *                 (no range checking performed)
+ */
+
+/* gggg gggg gggg gggg gggg gggg gggg gggg | gggg gggg gggg gggg seee eeee eeee eeee |
+ * 1mmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm
+ */
+
+tu_int
+__fixunsxfti(long double a)
+{
+    long_double_bits fb;
+    fb.f = a;
+    int e = (fb.u.high.s.low & 0x00007FFF) - 16383;
+    if (e < 0 || (fb.u.high.s.low & 0x00008000))
+        return 0;
+    tu_int r = fb.u.low.all;
+    if (e > 63)
+        r <<= (e - 63);
+    else
+        r >>= (63 - e);
+    return r;
+}
+
+#endif
diff --git a/lib/builtins/fixxfdi.c b/lib/builtins/fixxfdi.c
new file mode 100644
index 000000000..9592ce430
--- /dev/null
+++ b/lib/builtins/fixxfdi.c
@@ -0,0 +1,44 @@
+/* ===-- fixxfdi.c - Implement __fixxfdi -----------------------------------===
+ *
+ *      	       The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixxfdi for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#if !_ARCH_PPC
+
+#include "int_lib.h"
+
+/* Returns: convert a to a signed long long, rounding toward zero. */
+
+/* Assumption: long double is an intel 80 bit floating point type padded with 6 bytes
+ *             su_int is a 32 bit integral type
+ *             value in long double is representable in di_int (no range checking performed)
+ */
+
+/* gggg gggg gggg gggg gggg gggg gggg gggg | gggg gggg gggg gggg seee eeee eeee eeee |
+ * 1mmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm
+ */
+
+di_int
+__fixxfdi(long double a)
+{
+    long_double_bits fb;
+    fb.f = a;
+    int e = (fb.u.high.s.low & 0x00007FFF) - 16383;
+    if (e < 0)
+        return 0;
+    di_int s = -(si_int)((fb.u.high.s.low & 0x00008000) >> 15);
+    di_int r = fb.u.low.all;
+    r = (du_int)r >> (63 - e);
+    return (r ^ s) - s;
+}
+
+#endif /* !_ARCH_PPC */
diff --git a/lib/builtins/fixxfti.c b/lib/builtins/fixxfti.c
new file mode 100644
index 000000000..973dc31bf
--- /dev/null
+++ b/lib/builtins/fixxfti.c
@@ -0,0 +1,47 @@
+/* ===-- fixxfti.c - Implement __fixxfti -----------------------------------===
+ *
+ *      	       The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __fixxfti for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: convert a to a signed long long, rounding toward zero. */
+
+/* Assumption: long double is an intel 80 bit floating point type padded with 6 bytes
+ *             su_int is a 32 bit integral type
+ *             value in long double is representable in ti_int (no range checking performed)
+ */
+
+/* gggg gggg gggg gggg gggg gggg gggg gggg | gggg gggg gggg gggg seee eeee eeee eeee |
+ * 1mmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm
+ */
+
+ti_int
+__fixxfti(long double a)
+{
+    long_double_bits fb;
+    fb.f = a;
+    int e = (fb.u.high.s.low & 0x00007FFF) - 16383;
+    if (e < 0)
+        return 0;
+    ti_int s = -(si_int)((fb.u.high.s.low & 0x00008000) >> 15);
+    ti_int r = fb.u.low.all;
+    if (e > 63)
+        r <<= (e - 63);
+    else
+        r >>= (63 - e);
+    return (r ^ s) - s;
+}
+
+#endif /* __x86_64 */
diff --git a/lib/builtins/floatdidf.c b/lib/builtins/floatdidf.c
new file mode 100644
index 000000000..e53fa2580
--- /dev/null
+++ b/lib/builtins/floatdidf.c
@@ -0,0 +1,107 @@
+/*===-- floatdidf.c - Implement __floatdidf -------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ *===----------------------------------------------------------------------===
+ *
+ * This file implements __floatdidf for the compiler_rt library.
+ *
+ *===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: convert a to a double, rounding toward even. */
+
+/* Assumption: double is a IEEE 64 bit floating point type 
+ *             di_int is a 64 bit integral type
+ */
+
+/* seee eeee eeee mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm */
+
+ARM_EABI_FNALIAS(l2d, floatdidf)
+
+#ifndef __SOFT_FP__
+/* Support for systems that have hardware floating-point; we'll set the inexact flag
+ * as a side-effect of this computation.
+ */
+
+COMPILER_RT_ABI double
+__floatdidf(di_int a)
+{
+	static const double twop52 = 0x1.0p52;
+	static const double twop32 = 0x1.0p32;
+	
+	union { int64_t x; double d; } low = { .d = twop52 };
+	
+	const double high = (int32_t)(a >> 32) * twop32;
+	low.x |= a & INT64_C(0x00000000ffffffff);
+	
+	const double result = (high - twop52) + low.d;
+	return result;
+}
+
+#else
+/* Support for systems that don't have hardware floating-point; there are no flags to
+ * set, and we don't want to code-gen to an unknown soft-float implementation.
+ */
+
+COMPILER_RT_ABI double
+__floatdidf(di_int a)
+{
+    if (a == 0)
+        return 0.0;
+    const unsigned N = sizeof(di_int) * CHAR_BIT;
+    const di_int s = a >> (N-1);
+    a = (a ^ s) - s;
+    int sd = N - __builtin_clzll(a);  /* number of significant digits */
+    int e = sd - 1;             /* exponent */
+    if (sd > DBL_MANT_DIG)
+    {
+        /*  start:  0000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQxxxxxxxxxxxxxxxxxx
+         *  finish: 000000000000000000000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQR
+         *                                                12345678901234567890123456
+         *  1 = msb 1 bit
+         *  P = bit DBL_MANT_DIG-1 bits to the right of 1
+         * Q = bit DBL_MANT_DIG bits to the right of 1
+         *  R = "or" of all bits to the right of Q
+        */
+        switch (sd)
+        {
+        case DBL_MANT_DIG + 1:
+            a <<= 1;
+            break;
+        case DBL_MANT_DIG + 2:
+            break;
+        default:
+            a = ((du_int)a >> (sd - (DBL_MANT_DIG+2))) |
+                ((a & ((du_int)(-1) >> ((N + DBL_MANT_DIG+2) - sd))) != 0);
+        };
+        /* finish: */
+        a |= (a & 4) != 0;  /* Or P into R */
+        ++a;  /* round - this step may add a significant bit */
+        a >>= 2;  /* dump Q and R */
+        /* a is now rounded to DBL_MANT_DIG or DBL_MANT_DIG+1 bits */
+        if (a & ((du_int)1 << DBL_MANT_DIG))
+        {
+            a >>= 1;
+            ++e;
+        }
+        /* a is now rounded to DBL_MANT_DIG bits */
+    }
+    else
+    {
+        a <<= (DBL_MANT_DIG - sd);
+        /* a is now rounded to DBL_MANT_DIG bits */
+    }
+    double_bits fb;
+    fb.u.high = ((su_int)s & 0x80000000) |        /* sign */
+                ((e + 1023) << 20)      |        /* exponent */
+                ((su_int)(a >> 32) & 0x000FFFFF); /* mantissa-high */
+    fb.u.low = (su_int)a;                         /* mantissa-low */
+    return fb.f;
+}
+#endif
diff --git a/lib/builtins/floatdisf.c b/lib/builtins/floatdisf.c
new file mode 100644
index 000000000..3e47580ef
--- /dev/null
+++ b/lib/builtins/floatdisf.c
@@ -0,0 +1,80 @@
+/*===-- floatdisf.c - Implement __floatdisf -------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ *===----------------------------------------------------------------------===
+ *
+ * This file implements __floatdisf for the compiler_rt library.
+ *
+ *===----------------------------------------------------------------------===
+ */
+
+/* Returns: convert a to a float, rounding toward even.*/
+
+/* Assumption: float is a IEEE 32 bit floating point type 
+ *             di_int is a 64 bit integral type
+ */ 
+
+/* seee eeee emmm mmmm mmmm mmmm mmmm mmmm */
+
+#include "int_lib.h"
+
+ARM_EABI_FNALIAS(l2f, floatdisf)
+
+COMPILER_RT_ABI float
+__floatdisf(di_int a)
+{
+    if (a == 0)
+        return 0.0F;
+    const unsigned N = sizeof(di_int) * CHAR_BIT;
+    const di_int s = a >> (N-1);
+    a = (a ^ s) - s;
+    int sd = N - __builtin_clzll(a);  /* number of significant digits */
+    int e = sd - 1;             /* exponent */
+    if (sd > FLT_MANT_DIG)
+    {
+        /*  start:  0000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQxxxxxxxxxxxxxxxxxx 
+         *  finish: 000000000000000000000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQR 
+         *                                                12345678901234567890123456 
+         *  1 = msb 1 bit 
+         *  P = bit FLT_MANT_DIG-1 bits to the right of 1 
+         *  Q = bit FLT_MANT_DIG bits to the right of 1   
+         *  R = "or" of all bits to the right of Q 
+         */
+        switch (sd)
+        {
+        case FLT_MANT_DIG + 1:
+            a <<= 1;
+            break;
+        case FLT_MANT_DIG + 2:
+            break;
+        default:
+            a = ((du_int)a >> (sd - (FLT_MANT_DIG+2))) |
+                ((a & ((du_int)(-1) >> ((N + FLT_MANT_DIG+2) - sd))) != 0);
+        };
+        /* finish: */
+        a |= (a & 4) != 0;  /* Or P into R */
+        ++a;  /* round - this step may add a significant bit */
+        a >>= 2;  /* dump Q and R */
+        /* a is now rounded to FLT_MANT_DIG or FLT_MANT_DIG+1 bits */
+        if (a & ((du_int)1 << FLT_MANT_DIG))
+        {
+            a >>= 1;
+            ++e;
+        }
+        /* a is now rounded to FLT_MANT_DIG bits */
+    }
+    else
+    {
+        a <<= (FLT_MANT_DIG - sd);
+        /* a is now rounded to FLT_MANT_DIG bits */
+    }
+    float_bits fb;
+    fb.u = ((su_int)s & 0x80000000) |  /* sign */
+           ((e + 127) << 23)       |  /* exponent */
+           ((su_int)a & 0x007FFFFF);   /* mantissa */
+    return fb.f;
+}
diff --git a/lib/builtins/floatdixf.c b/lib/builtins/floatdixf.c
new file mode 100644
index 000000000..ebf62dba0
--- /dev/null
+++ b/lib/builtins/floatdixf.c
@@ -0,0 +1,46 @@
+/* ===-- floatdixf.c - Implement __floatdixf -------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __floatdixf for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */ 
+
+#if !_ARCH_PPC
+
+#include "int_lib.h"
+
+/* Returns: convert a to a long double, rounding toward even. */
+
+/* Assumption: long double is a IEEE 80 bit floating point type padded to 128 bits
+ *             di_int is a 64 bit integral type
+ */
+
+/* gggg gggg gggg gggg gggg gggg gggg gggg | gggg gggg gggg gggg seee eeee eeee eeee |
+ * 1mmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm
+ */
+
+long double
+__floatdixf(di_int a)
+{
+    if (a == 0)
+        return 0.0;
+    const unsigned N = sizeof(di_int) * CHAR_BIT;
+    const di_int s = a >> (N-1);
+    a = (a ^ s) - s;
+    int clz = __builtin_clzll(a);
+    int e = (N - 1) - clz ;    /* exponent */
+    long_double_bits fb;
+    fb.u.high.s.low = ((su_int)s & 0x00008000) |  /* sign */
+		      (e + 16383);                /* exponent */
+    fb.u.low.all = a << clz;                    /* mantissa */
+    return fb.f;
+}
+
+#endif /* !_ARCH_PPC */
diff --git a/lib/builtins/floatsidf.c b/lib/builtins/floatsidf.c
new file mode 100644
index 000000000..18f378f2a
--- /dev/null
+++ b/lib/builtins/floatsidf.c
@@ -0,0 +1,52 @@
+//===-- lib/floatsidf.c - integer -> double-precision conversion --*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements integer to double-precision conversion for the
+// compiler-rt library in the IEEE-754 default round-to-nearest, ties-to-even
+// mode.
+//
+//===----------------------------------------------------------------------===//
+
+#define DOUBLE_PRECISION
+#include "fp_lib.h"
+
+#include "int_lib.h"
+
+ARM_EABI_FNALIAS(i2d, floatsidf)
+
+fp_t __floatsidf(int a) {
+    
+    const int aWidth = sizeof a * CHAR_BIT;
+    
+    // Handle zero as a special case to protect clz
+    if (a == 0)
+        return fromRep(0);
+    
+    // All other cases begin by extracting the sign and absolute value of a
+    rep_t sign = 0;
+    if (a < 0) {
+        sign = signBit;
+        a = -a;
+    }
+    
+    // Exponent of (fp_t)a is the width of abs(a).
+    const int exponent = (aWidth - 1) - __builtin_clz(a);
+    rep_t result;
+    
+    // Shift a into the significand field and clear the implicit bit.  Extra
+    // cast to unsigned int is necessary to get the correct behavior for
+    // the input INT_MIN.
+    const int shift = significandBits - exponent;
+    result = (rep_t)(unsigned int)a << shift ^ implicitBit;
+    
+    // Insert the exponent
+    result += (rep_t)(exponent + exponentBias) << significandBits;
+    // Insert the sign bit and return
+    return fromRep(result | sign);
+}
diff --git a/lib/builtins/floatsisf.c b/lib/builtins/floatsisf.c
new file mode 100644
index 000000000..83983934f
--- /dev/null
+++ b/lib/builtins/floatsisf.c
@@ -0,0 +1,58 @@
+//===-- lib/floatsisf.c - integer -> single-precision conversion --*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements integer to single-precision conversion for the
+// compiler-rt library in the IEEE-754 default round-to-nearest, ties-to-even
+// mode.
+//
+//===----------------------------------------------------------------------===//
+
+#define SINGLE_PRECISION
+#include "fp_lib.h"
+
+#include "int_lib.h"
+
+ARM_EABI_FNALIAS(i2f, floatsisf)
+
+fp_t __floatsisf(int a) {
+    
+    const int aWidth = sizeof a * CHAR_BIT;
+    
+    // Handle zero as a special case to protect clz
+    if (a == 0)
+        return fromRep(0);
+    
+    // All other cases begin by extracting the sign and absolute value of a
+    rep_t sign = 0;
+    if (a < 0) {
+        sign = signBit;
+        a = -a;
+    }
+    
+    // Exponent of (fp_t)a is the width of abs(a).
+    const int exponent = (aWidth - 1) - __builtin_clz(a);
+    rep_t result;
+    
+    // Shift a into the significand field, rounding if it is a right-shift
+    if (exponent <= significandBits) {
+        const int shift = significandBits - exponent;
+        result = (rep_t)a << shift ^ implicitBit;
+    } else {
+        const int shift = exponent - significandBits;
+        result = (rep_t)a >> shift ^ implicitBit;
+        rep_t round = (rep_t)a << (typeWidth - shift);
+        if (round > signBit) result++;
+        if (round == signBit) result += result & 1;
+    }
+    
+    // Insert the exponent
+    result += (rep_t)(exponent + exponentBias) << significandBits;
+    // Insert the sign bit and return
+    return fromRep(result | sign);
+}
diff --git a/lib/builtins/floattidf.c b/lib/builtins/floattidf.c
new file mode 100644
index 000000000..77749f8d3
--- /dev/null
+++ b/lib/builtins/floattidf.c
@@ -0,0 +1,85 @@
+/* ===-- floattidf.c - Implement __floattidf -------------------------------===
+ *
+ *                    The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __floattidf for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */ 
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: convert a to a double, rounding toward even.*/
+
+/* Assumption: double is a IEEE 64 bit floating point type 
+ *            ti_int is a 128 bit integral type
+ */
+
+/* seee eeee eeee mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm */ 
+
+si_int __clzti2(ti_int a);
+
+double
+__floattidf(ti_int a)
+{
+    if (a == 0)
+        return 0.0;
+    const unsigned N = sizeof(ti_int) * CHAR_BIT;
+    const ti_int s = a >> (N-1);
+    a = (a ^ s) - s;
+    int sd = N - __clzti2(a);  /* number of significant digits */
+    int e = sd - 1;             /* exponent */
+    if (sd > DBL_MANT_DIG)
+    {
+        /* start:  0000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQxxxxxxxxxxxxxxxxxx
+         *  finish: 000000000000000000000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQR
+         *                                               12345678901234567890123456
+         * 1 = msb 1 bit
+         * P = bit DBL_MANT_DIG-1 bits to the right of 1
+         * Q = bit DBL_MANT_DIG bits to the right of 1
+         * R = "or" of all bits to the right of Q
+         */
+        switch (sd)
+        {
+        case DBL_MANT_DIG + 1:
+            a <<= 1;
+            break;
+        case DBL_MANT_DIG + 2:
+            break;
+        default:
+            a = ((tu_int)a >> (sd - (DBL_MANT_DIG+2))) |
+                ((a & ((tu_int)(-1) >> ((N + DBL_MANT_DIG+2) - sd))) != 0);
+        };
+        /* finish: */
+        a |= (a & 4) != 0;  /* Or P into R */
+        ++a;  /* round - this step may add a significant bit */
+        a >>= 2;  /* dump Q and R */
+        /* a is now rounded to DBL_MANT_DIG or DBL_MANT_DIG+1 bits */
+        if (a & ((tu_int)1 << DBL_MANT_DIG))
+        {
+            a >>= 1;
+            ++e;
+        }
+        /* a is now rounded to DBL_MANT_DIG bits */
+    }
+    else
+    {
+        a <<= (DBL_MANT_DIG - sd);
+        /* a is now rounded to DBL_MANT_DIG bits */
+    }
+    double_bits fb;
+    fb.u.s.high = ((su_int)s & 0x80000000) |        /* sign */
+                ((e + 1023) << 20)      |        /* exponent */
+                ((su_int)(a >> 32) & 0x000FFFFF); /* mantissa-high */
+    fb.u.s.low = (su_int)a;                         /* mantissa-low */
+    return fb.f;
+}
+
+#endif
diff --git a/lib/builtins/floattisf.c b/lib/builtins/floattisf.c
new file mode 100644
index 000000000..4776125de
--- /dev/null
+++ b/lib/builtins/floattisf.c
@@ -0,0 +1,84 @@
+/* ===-- floattisf.c - Implement __floattisf -------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __floattisf for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: convert a to a float, rounding toward even. */
+
+/* Assumption: float is a IEEE 32 bit floating point type 
+ *             ti_int is a 128 bit integral type
+ */
+
+/* seee eeee emmm mmmm mmmm mmmm mmmm mmmm */
+
+si_int __clzti2(ti_int a);
+
+float
+__floattisf(ti_int a)
+{
+    if (a == 0)
+        return 0.0F;
+    const unsigned N = sizeof(ti_int) * CHAR_BIT;
+    const ti_int s = a >> (N-1);
+    a = (a ^ s) - s;
+    int sd = N - __clzti2(a);  /* number of significant digits */
+    int e = sd - 1;             /* exponent */
+    if (sd > FLT_MANT_DIG)
+    {
+        /*  start:  0000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQxxxxxxxxxxxxxxxxxx
+        * finish: 000000000000000000000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQR
+        *                                                12345678901234567890123456
+        *  1 = msb 1 bit
+        *  P = bit FLT_MANT_DIG-1 bits to the right of 1
+        *  Q = bit FLT_MANT_DIG bits to the right of 1
+        *  R = "or" of all bits to the right of Q
+        */
+        switch (sd)
+        {
+        case FLT_MANT_DIG + 1:
+            a <<= 1;
+            break;
+        case FLT_MANT_DIG + 2:
+            break;
+        default:
+            a = ((tu_int)a >> (sd - (FLT_MANT_DIG+2))) |
+                ((a & ((tu_int)(-1) >> ((N + FLT_MANT_DIG+2) - sd))) != 0);
+        };
+        /* finish: */
+        a |= (a & 4) != 0;  /* Or P into R */
+        ++a;  /* round - this step may add a significant bit */
+        a >>= 2;  /* dump Q and R */
+        /* a is now rounded to FLT_MANT_DIG or FLT_MANT_DIG+1 bits */
+        if (a & ((tu_int)1 << FLT_MANT_DIG))
+        {
+            a >>= 1;
+            ++e;
+        }
+        /* a is now rounded to FLT_MANT_DIG bits */
+    }
+    else
+    {
+        a <<= (FLT_MANT_DIG - sd);
+        /* a is now rounded to FLT_MANT_DIG bits */
+    }
+    float_bits fb;
+    fb.u = ((su_int)s & 0x80000000) |  /* sign */
+           ((e + 127) << 23)       |  /* exponent */
+           ((su_int)a & 0x007FFFFF);   /* mantissa */
+    return fb.f;
+}
+
+#endif
diff --git a/lib/builtins/floattixf.c b/lib/builtins/floattixf.c
new file mode 100644
index 000000000..3813dc6b7
--- /dev/null
+++ b/lib/builtins/floattixf.c
@@ -0,0 +1,86 @@
+/* ===-- floattixf.c - Implement __floattixf -------------------------------===
+ *
+ *      	       The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __floattixf for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: convert a to a long double, rounding toward even. */
+
+/* Assumption: long double is a IEEE 80 bit floating point type padded to 128 bits
+ *             ti_int is a 128 bit integral type
+ */
+
+/* gggg gggg gggg gggg gggg gggg gggg gggg | gggg gggg gggg gggg seee eeee eeee eeee |
+ * 1mmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm
+ */
+
+si_int __clzti2(ti_int a);
+
+long double
+__floattixf(ti_int a)
+{
+    if (a == 0)
+        return 0.0;
+    const unsigned N = sizeof(ti_int) * CHAR_BIT;
+    const ti_int s = a >> (N-1);
+    a = (a ^ s) - s;
+    int sd = N - __clzti2(a);  /* number of significant digits */
+    int e = sd - 1;             /* exponent */
+    if (sd > LDBL_MANT_DIG)
+    {
+        /*  start:  0000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQxxxxxxxxxxxxxxxxxx
+         *  finish: 000000000000000000000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQR
+         *                                                12345678901234567890123456
+         *  1 = msb 1 bit
+         *  P = bit LDBL_MANT_DIG-1 bits to the right of 1
+         *  Q = bit LDBL_MANT_DIG bits to the right of 1
+         *  R = "or" of all bits to the right of Q
+	      */
+        switch (sd)
+        {
+        case LDBL_MANT_DIG + 1:
+            a <<= 1;
+            break;
+        case LDBL_MANT_DIG + 2:
+            break;
+        default:
+            a = ((tu_int)a >> (sd - (LDBL_MANT_DIG+2))) |
+                ((a & ((tu_int)(-1) >> ((N + LDBL_MANT_DIG+2) - sd))) != 0);
+        };
+        /* finish: */
+        a |= (a & 4) != 0;  /* Or P into R */
+        ++a;  /* round - this step may add a significant bit */
+        a >>= 2;  /* dump Q and R */
+        /* a is now rounded to LDBL_MANT_DIG or LDBL_MANT_DIG+1 bits */
+        if (a & ((tu_int)1 << LDBL_MANT_DIG))
+        {
+            a >>= 1;
+            ++e;
+        }
+        /* a is now rounded to LDBL_MANT_DIG bits */
+    }
+    else
+    {
+        a <<= (LDBL_MANT_DIG - sd);
+        /* a is now rounded to LDBL_MANT_DIG bits */
+    }
+    long_double_bits fb;
+    fb.u.high.s.low = ((su_int)s & 0x8000) |        /* sign */
+                    (e + 16383);                  /* exponent */
+    fb.u.low.all = (du_int)a;                     /* mantissa */
+    return fb.f;
+}
+
+#endif
diff --git a/lib/builtins/floatundidf.c b/lib/builtins/floatundidf.c
new file mode 100644
index 000000000..e52fa0a33
--- /dev/null
+++ b/lib/builtins/floatundidf.c
@@ -0,0 +1,107 @@
+/* ===-- floatundidf.c - Implement __floatundidf ---------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __floatundidf for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+/* Returns: convert a to a double, rounding toward even. */
+
+/* Assumption: double is a IEEE 64 bit floating point type 
+ *             du_int is a 64 bit integral type
+ */
+
+/* seee eeee eeee mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm */
+
+#include "int_lib.h"
+
+ARM_EABI_FNALIAS(ul2d, floatundidf)
+
+#ifndef __SOFT_FP__
+/* Support for systems that have hardware floating-point; we'll set the inexact flag
+ * as a side-effect of this computation.
+ */
+
+
+COMPILER_RT_ABI double
+__floatundidf(du_int a)
+{
+	static const double twop52 = 0x1.0p52;
+	static const double twop84 = 0x1.0p84;
+	static const double twop84_plus_twop52 = 0x1.00000001p84;
+	
+	union { uint64_t x; double d; } high = { .d = twop84 };
+	union { uint64_t x; double d; } low = { .d = twop52 };
+	
+	high.x |= a >> 32;
+	low.x |= a & UINT64_C(0x00000000ffffffff);
+	
+	const double result = (high.d - twop84_plus_twop52) + low.d;
+	return result;
+}
+
+#else
+/* Support for systems that don't have hardware floating-point; there are no flags to
+ * set, and we don't want to code-gen to an unknown soft-float implementation.
+ */ 
+
+COMPILER_RT_ABI double
+__floatundidf(du_int a)
+{
+    if (a == 0)
+        return 0.0;
+    const unsigned N = sizeof(du_int) * CHAR_BIT;
+    int sd = N - __builtin_clzll(a);  /* number of significant digits */
+    int e = sd - 1;             /* exponent */
+    if (sd > DBL_MANT_DIG)
+    {
+        /*  start:  0000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQxxxxxxxxxxxxxxxxxx
+         *  finish: 000000000000000000000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQR
+         *                                                12345678901234567890123456
+         *  1 = msb 1 bit
+         *  P = bit DBL_MANT_DIG-1 bits to the right of 1
+         *  Q = bit DBL_MANT_DIG bits to the right of 1
+         *  R = "or" of all bits to the right of Q
+         */
+        switch (sd)
+        {
+        case DBL_MANT_DIG + 1:
+            a <<= 1;
+            break;
+        case DBL_MANT_DIG + 2:
+            break;
+        default:
+            a = (a >> (sd - (DBL_MANT_DIG+2))) |
+                ((a & ((du_int)(-1) >> ((N + DBL_MANT_DIG+2) - sd))) != 0);
+        };
+        /* finish: */
+        a |= (a & 4) != 0;  /* Or P into R */
+        ++a;  /* round - this step may add a significant bit */
+        a >>= 2;  /* dump Q and R */
+        /* a is now rounded to DBL_MANT_DIG or DBL_MANT_DIG+1 bits */
+        if (a & ((du_int)1 << DBL_MANT_DIG))
+        {
+            a >>= 1;
+            ++e;
+        }
+        /* a is now rounded to DBL_MANT_DIG bits */
+    }
+    else
+    {
+        a <<= (DBL_MANT_DIG - sd);
+        /* a is now rounded to DBL_MANT_DIG bits */
+    }
+    double_bits fb;
+    fb.u.high = ((e + 1023) << 20)      |        /* exponent */
+                ((su_int)(a >> 32) & 0x000FFFFF); /* mantissa-high */
+    fb.u.low = (su_int)a;                         /* mantissa-low  */
+    return fb.f;
+}
+#endif
diff --git a/lib/builtins/floatundisf.c b/lib/builtins/floatundisf.c
new file mode 100644
index 000000000..713a44abc
--- /dev/null
+++ b/lib/builtins/floatundisf.c
@@ -0,0 +1,77 @@
+/*===-- floatundisf.c - Implement __floatundisf ---------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __floatundisf for the compiler_rt library.
+ *
+ *===----------------------------------------------------------------------===
+ */
+
+/* Returns: convert a to a float, rounding toward even. */
+
+/* Assumption: float is a IEEE 32 bit floating point type 
+ *            du_int is a 64 bit integral type
+ */
+
+/* seee eeee emmm mmmm mmmm mmmm mmmm mmmm */
+
+#include "int_lib.h"
+
+ARM_EABI_FNALIAS(ul2f, floatundisf)
+
+COMPILER_RT_ABI float
+__floatundisf(du_int a)
+{
+    if (a == 0)
+        return 0.0F;
+    const unsigned N = sizeof(du_int) * CHAR_BIT;
+    int sd = N - __builtin_clzll(a);  /* number of significant digits */
+    int e = sd - 1;             /* 8 exponent */
+    if (sd > FLT_MANT_DIG)
+    {
+        /*  start:  0000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQxxxxxxxxxxxxxxxxxx
+         *  finish: 000000000000000000000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQR
+         *                                                12345678901234567890123456
+         *  1 = msb 1 bit
+         *  P = bit FLT_MANT_DIG-1 bits to the right of 1
+         *  Q = bit FLT_MANT_DIG bits to the right of 1
+         *  R = "or" of all bits to the right of Q
+         */
+        switch (sd)
+        {
+        case FLT_MANT_DIG + 1:
+            a <<= 1;
+            break;
+        case FLT_MANT_DIG + 2:
+            break;
+        default:
+            a = (a >> (sd - (FLT_MANT_DIG+2))) |
+                ((a & ((du_int)(-1) >> ((N + FLT_MANT_DIG+2) - sd))) != 0);
+        };
+        /* finish: */
+        a |= (a & 4) != 0;  /* Or P into R */
+        ++a;  /* round - this step may add a significant bit */
+        a >>= 2;  /* dump Q and R */
+        /* a is now rounded to FLT_MANT_DIG or FLT_MANT_DIG+1 bits */
+        if (a & ((du_int)1 << FLT_MANT_DIG))
+        {
+            a >>= 1;
+            ++e;
+        }
+        /* a is now rounded to FLT_MANT_DIG bits */
+    }
+    else
+    {
+        a <<= (FLT_MANT_DIG - sd);
+        /* a is now rounded to FLT_MANT_DIG bits */
+    }
+    float_bits fb;
+    fb.u = ((e + 127) << 23)       |  /* exponent */
+           ((su_int)a & 0x007FFFFF);  /* mantissa */
+    return fb.f;
+}
diff --git a/lib/builtins/floatundixf.c b/lib/builtins/floatundixf.c
new file mode 100644
index 000000000..64f7662d9
--- /dev/null
+++ b/lib/builtins/floatundixf.c
@@ -0,0 +1,42 @@
+/* ===-- floatundixf.c - Implement __floatundixf ---------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __floatundixf for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#if !_ARCH_PPC
+
+#include "int_lib.h"
+
+/* Returns: convert a to a long double, rounding toward even. */
+
+/* Assumption: long double is a IEEE 80 bit floating point type padded to 128 bits
+ *             du_int is a 64 bit integral type
+ */
+
+/* gggg gggg gggg gggg gggg gggg gggg gggg | gggg gggg gggg gggg seee eeee eeee eeee |
+ * 1mmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm
+ */
+long double
+__floatundixf(du_int a)
+{
+    if (a == 0)
+        return 0.0;
+    const unsigned N = sizeof(du_int) * CHAR_BIT;
+    int clz = __builtin_clzll(a);
+    int e = (N - 1) - clz ;    /* exponent */
+    long_double_bits fb;
+    fb.u.high.s.low = (e + 16383);              /* exponent */
+    fb.u.low.all = a << clz;                   /* mantissa */
+    return fb.f;
+}
+
+#endif /* _ARCH_PPC */
diff --git a/lib/builtins/floatunsidf.c b/lib/builtins/floatunsidf.c
new file mode 100644
index 000000000..ba6c2cfd2
--- /dev/null
+++ b/lib/builtins/floatunsidf.c
@@ -0,0 +1,41 @@
+//===-- lib/floatunsidf.c - uint -> double-precision conversion ---*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements unsigned integer to double-precision conversion for the
+// compiler-rt library in the IEEE-754 default round-to-nearest, ties-to-even
+// mode.
+//
+//===----------------------------------------------------------------------===//
+
+#define DOUBLE_PRECISION
+#include "fp_lib.h"
+
+#include "int_lib.h"
+
+ARM_EABI_FNALIAS(ui2d, floatunsidf)
+
+fp_t __floatunsidf(unsigned int a) {
+    
+    const int aWidth = sizeof a * CHAR_BIT;
+    
+    // Handle zero as a special case to protect clz
+    if (a == 0) return fromRep(0);
+    
+    // Exponent of (fp_t)a is the width of abs(a).
+    const int exponent = (aWidth - 1) - __builtin_clz(a);
+    rep_t result;
+    
+    // Shift a into the significand field and clear the implicit bit.
+    const int shift = significandBits - exponent;
+    result = (rep_t)a << shift ^ implicitBit;
+    
+    // Insert the exponent
+    result += (rep_t)(exponent + exponentBias) << significandBits;
+    return fromRep(result);
+}
diff --git a/lib/builtins/floatunsisf.c b/lib/builtins/floatunsisf.c
new file mode 100644
index 000000000..e392c0ecf
--- /dev/null
+++ b/lib/builtins/floatunsisf.c
@@ -0,0 +1,49 @@
+//===-- lib/floatunsisf.c - uint -> single-precision conversion ---*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements unsigned integer to single-precision conversion for the
+// compiler-rt library in the IEEE-754 default round-to-nearest, ties-to-even
+// mode.
+//
+//===----------------------------------------------------------------------===//
+
+#define SINGLE_PRECISION
+#include "fp_lib.h"
+
+#include "int_lib.h"
+
+ARM_EABI_FNALIAS(ui2f, floatunsisf)
+
+fp_t __floatunsisf(unsigned int a) {
+    
+    const int aWidth = sizeof a * CHAR_BIT;
+    
+    // Handle zero as a special case to protect clz
+    if (a == 0) return fromRep(0);
+    
+    // Exponent of (fp_t)a is the width of abs(a).
+    const int exponent = (aWidth - 1) - __builtin_clz(a);
+    rep_t result;
+    
+    // Shift a into the significand field, rounding if it is a right-shift
+    if (exponent <= significandBits) {
+        const int shift = significandBits - exponent;
+        result = (rep_t)a << shift ^ implicitBit;
+    } else {
+        const int shift = exponent - significandBits;
+        result = (rep_t)a >> shift ^ implicitBit;
+        rep_t round = (rep_t)a << (typeWidth - shift);
+        if (round > signBit) result++;
+        if (round == signBit) result += result & 1;
+    }
+    
+    // Insert the exponent
+    result += (rep_t)(exponent + exponentBias) << significandBits;
+    return fromRep(result);
+}
diff --git a/lib/builtins/floatuntidf.c b/lib/builtins/floatuntidf.c
new file mode 100644
index 000000000..4c1d3289f
--- /dev/null
+++ b/lib/builtins/floatuntidf.c
@@ -0,0 +1,82 @@
+/* ===-- floatuntidf.c - Implement __floatuntidf ---------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __floatuntidf for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: convert a to a double, rounding toward even. */
+
+/* Assumption: double is a IEEE 64 bit floating point type 
+ *             tu_int is a 128 bit integral type
+ */
+
+/* seee eeee eeee mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm */
+
+si_int __clzti2(ti_int a);
+
+double
+__floatuntidf(tu_int a)
+{
+    if (a == 0)
+        return 0.0;
+    const unsigned N = sizeof(tu_int) * CHAR_BIT;
+    int sd = N - __clzti2(a);  /* number of significant digits */
+    int e = sd - 1;             /* exponent */
+    if (sd > DBL_MANT_DIG)
+    {
+        /*  start:  0000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQxxxxxxxxxxxxxxxxxx
+         *  finish: 000000000000000000000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQR
+         *                                                12345678901234567890123456
+         *  1 = msb 1 bit
+         *  P = bit DBL_MANT_DIG-1 bits to the right of 1
+         *  Q = bit DBL_MANT_DIG bits to the right of 1
+         *  R = "or" of all bits to the right of Q
+	 */
+        switch (sd)
+        {
+        case DBL_MANT_DIG + 1:
+            a <<= 1;
+            break;
+        case DBL_MANT_DIG + 2:
+            break;
+        default:
+            a = (a >> (sd - (DBL_MANT_DIG+2))) |
+                ((a & ((tu_int)(-1) >> ((N + DBL_MANT_DIG+2) - sd))) != 0);
+        };
+        /* finish: */
+        a |= (a & 4) != 0;  /* Or P into R */
+        ++a;  /* round - this step may add a significant bit */
+        a >>= 2;  /* dump Q and R */
+        /* a is now rounded to DBL_MANT_DIG or DBL_MANT_DIG+1 bits */
+        if (a & ((tu_int)1 << DBL_MANT_DIG))
+        {
+            a >>= 1;
+            ++e;
+        }
+        /* a is now rounded to DBL_MANT_DIG bits */
+    }
+    else
+    {
+        a <<= (DBL_MANT_DIG - sd);
+        /* a is now rounded to DBL_MANT_DIG bits */
+    }
+    double_bits fb;
+    fb.u.s.high = ((e + 1023) << 20)      |        /* exponent */
+                ((su_int)(a >> 32) & 0x000FFFFF); /* mantissa-high */
+    fb.u.s.low = (su_int)a;                         /* mantissa-low */
+    return fb.f;
+}
+
+#endif
diff --git a/lib/builtins/floatuntisf.c b/lib/builtins/floatuntisf.c
new file mode 100644
index 000000000..c8da2601b
--- /dev/null
+++ b/lib/builtins/floatuntisf.c
@@ -0,0 +1,81 @@
+/* ===-- floatuntisf.c - Implement __floatuntisf ---------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __floatuntisf for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: convert a to a float, rounding toward even. */
+
+/* Assumption: float is a IEEE 32 bit floating point type 
+ *             tu_int is a 128 bit integral type
+ */
+
+/* seee eeee emmm mmmm mmmm mmmm mmmm mmmm */
+
+si_int __clzti2(ti_int a);
+
+float
+__floatuntisf(tu_int a)
+{
+    if (a == 0)
+        return 0.0F;
+    const unsigned N = sizeof(tu_int) * CHAR_BIT;
+    int sd = N - __clzti2(a);  /* number of significant digits */
+    int e = sd - 1;             /* exponent */
+    if (sd > FLT_MANT_DIG)
+    {
+        /*  start:  0000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQxxxxxxxxxxxxxxxxxx
+         *  finish: 000000000000000000000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQR
+         *                                                12345678901234567890123456
+         *  1 = msb 1 bit
+         *  P = bit FLT_MANT_DIG-1 bits to the right of 1
+         *  Q = bit FLT_MANT_DIG bits to the right of 1
+         *  R = "or" of all bits to the right of Q
+	 */
+        switch (sd)
+        {
+        case FLT_MANT_DIG + 1:
+            a <<= 1;
+            break;
+        case FLT_MANT_DIG + 2:
+            break;
+        default:
+            a = (a >> (sd - (FLT_MANT_DIG+2))) |
+                ((a & ((tu_int)(-1) >> ((N + FLT_MANT_DIG+2) - sd))) != 0);
+        };
+        /* finish: */
+        a |= (a & 4) != 0;  /* Or P into R */
+        ++a;  /* round - this step may add a significant bit */
+        a >>= 2;  /* dump Q and R */
+        /* a is now rounded to FLT_MANT_DIG or FLT_MANT_DIG+1 bits */
+        if (a & ((tu_int)1 << FLT_MANT_DIG))
+        {
+            a >>= 1;
+            ++e;
+        }
+        /* a is now rounded to FLT_MANT_DIG bits */
+    }
+    else
+    {
+        a <<= (FLT_MANT_DIG - sd);
+        /* a is now rounded to FLT_MANT_DIG bits */
+    }
+    float_bits fb;
+    fb.u = ((e + 127) << 23)       |  /* exponent */
+           ((su_int)a & 0x007FFFFF);  /* mantissa */
+    return fb.f;
+}
+
+#endif
diff --git a/lib/builtins/floatuntixf.c b/lib/builtins/floatuntixf.c
new file mode 100644
index 000000000..dbce80f16
--- /dev/null
+++ b/lib/builtins/floatuntixf.c
@@ -0,0 +1,83 @@
+/* ===-- floatuntixf.c - Implement __floatuntixf ---------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __floatuntixf for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: convert a to a long double, rounding toward even. */
+
+/* Assumption: long double is a IEEE 80 bit floating point type padded to 128 bits
+ *             tu_int is a 128 bit integral type
+ */
+
+/* gggg gggg gggg gggg gggg gggg gggg gggg | gggg gggg gggg gggg seee eeee eeee eeee |
+ * 1mmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm | mmmm mmmm mmmm mmmm mmmm mmmm mmmm mmmm
+ */
+
+si_int __clzti2(ti_int a);
+
+long double
+__floatuntixf(tu_int a)
+{
+    if (a == 0)
+        return 0.0;
+    const unsigned N = sizeof(tu_int) * CHAR_BIT;
+    int sd = N - __clzti2(a);  /* number of significant digits */
+    int e = sd - 1;             /* exponent */
+    if (sd > LDBL_MANT_DIG)
+    {
+        /*  start:  0000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQxxxxxxxxxxxxxxxxxx
+         *  finish: 000000000000000000000000000000000000001xxxxxxxxxxxxxxxxxxxxxxPQR
+         *                                                12345678901234567890123456
+         *  1 = msb 1 bit
+         *  P = bit LDBL_MANT_DIG-1 bits to the right of 1
+         *  Q = bit LDBL_MANT_DIG bits to the right of 1
+         *  R = "or" of all bits to the right of Q
+	 */
+        switch (sd)
+        {
+        case LDBL_MANT_DIG + 1:
+            a <<= 1;
+            break;
+        case LDBL_MANT_DIG + 2:
+            break;
+        default:
+            a = (a >> (sd - (LDBL_MANT_DIG+2))) |
+                ((a & ((tu_int)(-1) >> ((N + LDBL_MANT_DIG+2) - sd))) != 0);
+        };
+        /* finish: */
+        a |= (a & 4) != 0;  /* Or P into R */
+        ++a;  /* round - this step may add a significant bit */
+        a >>= 2;  /* dump Q and R */
+        /* a is now rounded to LDBL_MANT_DIG or LDBL_MANT_DIG+1 bits */
+        if (a & ((tu_int)1 << LDBL_MANT_DIG))
+        {
+            a >>= 1;
+            ++e;
+        }
+        /* a is now rounded to LDBL_MANT_DIG bits */
+    }
+    else
+    {
+        a <<= (LDBL_MANT_DIG - sd);
+        /* a is now rounded to LDBL_MANT_DIG bits */
+    }
+    long_double_bits fb;
+    fb.u.high.s.low = (e + 16383);                  /* exponent */
+    fb.u.low.all = (du_int)a;                     /* mantissa */
+    return fb.f;
+}
+
+#endif
diff --git a/lib/builtins/fp_lib.h b/lib/builtins/fp_lib.h
new file mode 100644
index 000000000..661119ae4
--- /dev/null
+++ b/lib/builtins/fp_lib.h
@@ -0,0 +1,144 @@
+//===-- lib/fp_lib.h - Floating-point utilities -------------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a configuration header for soft-float routines in compiler-rt.
+// This file does not provide any part of the compiler-rt interface, but defines
+// many useful constants and utility routines that are used in the
+// implementation of the soft-float routines in compiler-rt.
+//
+// Assumes that float and double correspond to the IEEE-754 binary32 and
+// binary64 types, respectively, and that integer endianness matches floating
+// point endianness on the target platform.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef FP_LIB_HEADER
+#define FP_LIB_HEADER
+
+#include <stdint.h>
+#include <stdbool.h>
+#include <limits.h>
+#include "int_lib.h"
+
+#if defined SINGLE_PRECISION
+
+typedef uint32_t rep_t;
+typedef int32_t srep_t;
+typedef float fp_t;
+#define REP_C UINT32_C
+#define significandBits 23
+
+static inline int rep_clz(rep_t a) {
+    return __builtin_clz(a);
+}
+
+// 32x32 --> 64 bit multiply
+static inline void wideMultiply(rep_t a, rep_t b, rep_t *hi, rep_t *lo) {
+    const uint64_t product = (uint64_t)a*b;
+    *hi = product >> 32;
+    *lo = product;
+}
+
+#elif defined DOUBLE_PRECISION
+
+typedef uint64_t rep_t;
+typedef int64_t srep_t;
+typedef double fp_t;
+#define REP_C UINT64_C
+#define significandBits 52
+
+static inline int rep_clz(rep_t a) {
+#if defined __LP64__
+    return __builtin_clzl(a);
+#else
+    if (a & REP_C(0xffffffff00000000))
+        return __builtin_clz(a >> 32);
+    else 
+        return 32 + __builtin_clz(a & REP_C(0xffffffff));
+#endif
+}
+
+#define loWord(a) (a & 0xffffffffU)
+#define hiWord(a) (a >> 32)
+
+// 64x64 -> 128 wide multiply for platforms that don't have such an operation;
+// many 64-bit platforms have this operation, but they tend to have hardware
+// floating-point, so we don't bother with a special case for them here.
+static inline void wideMultiply(rep_t a, rep_t b, rep_t *hi, rep_t *lo) {
+    // Each of the component 32x32 -> 64 products
+    const uint64_t plolo = loWord(a) * loWord(b);
+    const uint64_t plohi = loWord(a) * hiWord(b);
+    const uint64_t philo = hiWord(a) * loWord(b);
+    const uint64_t phihi = hiWord(a) * hiWord(b);
+    // Sum terms that contribute to lo in a way that allows us to get the carry
+    const uint64_t r0 = loWord(plolo);
+    const uint64_t r1 = hiWord(plolo) + loWord(plohi) + loWord(philo);
+    *lo = r0 + (r1 << 32);
+    // Sum terms contributing to hi with the carry from lo
+    *hi = hiWord(plohi) + hiWord(philo) + hiWord(r1) + phihi;
+}
+
+#else
+#error Either SINGLE_PRECISION or DOUBLE_PRECISION must be defined.
+#endif
+
+#define typeWidth       (sizeof(rep_t)*CHAR_BIT)
+#define exponentBits    (typeWidth - significandBits - 1)
+#define maxExponent     ((1 << exponentBits) - 1)
+#define exponentBias    (maxExponent >> 1)
+
+#define implicitBit     (REP_C(1) << significandBits)
+#define significandMask (implicitBit - 1U)
+#define signBit         (REP_C(1) << (significandBits + exponentBits))
+#define absMask         (signBit - 1U)
+#define exponentMask    (absMask ^ significandMask)
+#define oneRep          ((rep_t)exponentBias << significandBits)
+#define infRep          exponentMask
+#define quietBit        (implicitBit >> 1)
+#define qnanRep         (exponentMask | quietBit)
+
+static inline rep_t toRep(fp_t x) {
+    const union { fp_t f; rep_t i; } rep = {.f = x};
+    return rep.i;
+}
+
+static inline fp_t fromRep(rep_t x) {
+    const union { fp_t f; rep_t i; } rep = {.i = x};
+    return rep.f;
+}
+
+static inline int normalize(rep_t *significand) {
+    const int shift = rep_clz(*significand) - rep_clz(implicitBit);
+    *significand <<= shift;
+    return 1 - shift;
+}
+
+static inline void wideLeftShift(rep_t *hi, rep_t *lo, int count) {
+    *hi = *hi << count | *lo >> (typeWidth - count);
+    *lo = *lo << count;
+}
+
+static inline void wideRightShiftWithSticky(rep_t *hi, rep_t *lo, unsigned int count) {
+    if (count < typeWidth) {
+        const bool sticky = *lo << (typeWidth - count);
+        *lo = *hi << (typeWidth - count) | *lo >> count | sticky;
+        *hi = *hi >> count;
+    }
+    else if (count < 2*typeWidth) {
+        const bool sticky = *hi << (2*typeWidth - count) | *lo;
+        *lo = *hi >> (count - typeWidth) | sticky;
+        *hi = 0;
+    } else {
+        const bool sticky = *hi | *lo;
+        *lo = sticky;
+        *hi = 0;
+    }
+}
+
+#endif // FP_LIB_HEADER
diff --git a/lib/builtins/gcc_personality_v0.c b/lib/builtins/gcc_personality_v0.c
new file mode 100644
index 000000000..8a708cae5
--- /dev/null
+++ b/lib/builtins/gcc_personality_v0.c
@@ -0,0 +1,247 @@
+/* ===-- gcc_personality_v0.c - Implement __gcc_personality_v0 -------------===
+ *
+ *      	       The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ */
+
+#include "int_lib.h"
+
+/*
+ * _Unwind_* stuff based on C++ ABI public documentation
+ * http://refspecs.freestandards.org/abi-eh-1.21.html
+ */
+
+typedef enum {
+    _URC_NO_REASON = 0,
+    _URC_FOREIGN_EXCEPTION_CAUGHT = 1,
+    _URC_FATAL_PHASE2_ERROR = 2,
+    _URC_FATAL_PHASE1_ERROR = 3,
+    _URC_NORMAL_STOP = 4,
+    _URC_END_OF_STACK = 5,
+    _URC_HANDLER_FOUND = 6,
+    _URC_INSTALL_CONTEXT = 7,
+    _URC_CONTINUE_UNWIND = 8
+} _Unwind_Reason_Code;
+
+typedef enum {
+    _UA_SEARCH_PHASE = 1,
+    _UA_CLEANUP_PHASE = 2,
+    _UA_HANDLER_FRAME = 4,
+    _UA_FORCE_UNWIND = 8,
+    _UA_END_OF_STACK = 16
+} _Unwind_Action;
+
+typedef struct _Unwind_Context* _Unwind_Context_t;
+
+struct _Unwind_Exception {
+    uint64_t                exception_class;
+    void                    (*exception_cleanup)(_Unwind_Reason_Code reason, 
+                                                 struct _Unwind_Exception* exc);
+    uintptr_t                private_1;    
+    uintptr_t                private_2;    
+};
+
+extern const uint8_t*    _Unwind_GetLanguageSpecificData(_Unwind_Context_t c);
+extern void              _Unwind_SetGR(_Unwind_Context_t c, int i, uintptr_t n);
+extern void              _Unwind_SetIP(_Unwind_Context_t, uintptr_t new_value);
+extern uintptr_t         _Unwind_GetIP(_Unwind_Context_t context);
+extern uintptr_t         _Unwind_GetRegionStart(_Unwind_Context_t context);
+
+
+/*
+ * Pointer encodings documented at:
+ *   http://refspecs.freestandards.org/LSB_1.3.0/gLSB/gLSB/ehframehdr.html
+ */
+
+#define DW_EH_PE_omit      0xff  /* no data follows */
+
+#define DW_EH_PE_absptr    0x00
+#define DW_EH_PE_uleb128   0x01
+#define DW_EH_PE_udata2    0x02
+#define DW_EH_PE_udata4    0x03
+#define DW_EH_PE_udata8    0x04
+#define DW_EH_PE_sleb128   0x09
+#define DW_EH_PE_sdata2    0x0A
+#define DW_EH_PE_sdata4    0x0B
+#define DW_EH_PE_sdata8    0x0C
+
+#define DW_EH_PE_pcrel     0x10
+#define DW_EH_PE_textrel   0x20
+#define DW_EH_PE_datarel   0x30
+#define DW_EH_PE_funcrel   0x40
+#define DW_EH_PE_aligned   0x50  
+#define DW_EH_PE_indirect  0x80 /* gcc extension */
+
+
+
+/* read a uleb128 encoded value and advance pointer */
+static uintptr_t readULEB128(const uint8_t** data)
+{
+    uintptr_t result = 0;
+    uintptr_t shift = 0;
+    unsigned char byte;
+    const uint8_t* p = *data;
+    do {
+        byte = *p++;
+        result |= (byte & 0x7f) << shift;
+        shift += 7;
+    } while (byte & 0x80);
+    *data = p;
+    return result;
+}
+
+/* read a pointer encoded value and advance pointer */
+static uintptr_t readEncodedPointer(const uint8_t** data, uint8_t encoding)
+{
+    const uint8_t* p = *data;
+    uintptr_t result = 0;
+
+    if ( encoding == DW_EH_PE_omit ) 
+        return 0;
+
+    /* first get value */
+    switch (encoding & 0x0F) {
+        case DW_EH_PE_absptr:
+            result = *((uintptr_t*)p);
+            p += sizeof(uintptr_t);
+            break;
+        case DW_EH_PE_uleb128:
+            result = readULEB128(&p);
+            break;
+        case DW_EH_PE_udata2:
+            result = *((uint16_t*)p);
+            p += sizeof(uint16_t);
+            break;
+        case DW_EH_PE_udata4:
+            result = *((uint32_t*)p);
+            p += sizeof(uint32_t);
+            break;
+        case DW_EH_PE_udata8:
+            result = *((uint64_t*)p);
+            p += sizeof(uint64_t);
+            break;
+        case DW_EH_PE_sdata2:
+            result = *((int16_t*)p);
+            p += sizeof(int16_t);
+            break;
+        case DW_EH_PE_sdata4:
+            result = *((int32_t*)p);
+            p += sizeof(int32_t);
+            break;
+        case DW_EH_PE_sdata8:
+            result = *((int64_t*)p);
+            p += sizeof(int64_t);
+            break;
+        case DW_EH_PE_sleb128:
+        default:
+            /* not supported */
+            compilerrt_abort();
+            break;
+    }
+
+    /* then add relative offset */
+    switch ( encoding & 0x70 ) {
+        case DW_EH_PE_absptr:
+            /* do nothing */
+            break;
+        case DW_EH_PE_pcrel:
+            result += (uintptr_t)(*data);
+            break;
+        case DW_EH_PE_textrel:
+        case DW_EH_PE_datarel:
+        case DW_EH_PE_funcrel:
+        case DW_EH_PE_aligned:
+        default:
+            /* not supported */
+            compilerrt_abort();
+            break;
+    }
+
+    /* then apply indirection */
+    if (encoding & DW_EH_PE_indirect) {
+        result = *((uintptr_t*)result);
+    }
+
+    *data = p;
+    return result;
+}
+
+
+/*
+ * The C compiler makes references to __gcc_personality_v0 in
+ * the dwarf unwind information for translation units that use
+ * __attribute__((cleanup(xx))) on local variables.
+ * This personality routine is called by the system unwinder
+ * on each frame as the stack is unwound during a C++ exception
+ * throw through a C function compiled with -fexceptions.
+ */
+#if __arm__
+// the setjump-longjump based exceptions personality routine has a different name
+_Unwind_Reason_Code __gcc_personality_sj0(int version, _Unwind_Action actions,
+         uint64_t exceptionClass, struct _Unwind_Exception* exceptionObject,
+         _Unwind_Context_t context)
+#else
+_Unwind_Reason_Code __gcc_personality_v0(int version, _Unwind_Action actions,
+         uint64_t exceptionClass, struct _Unwind_Exception* exceptionObject,
+         _Unwind_Context_t context)
+#endif
+{
+    /* Since C does not have catch clauses, there is nothing to do during */
+    /* phase 1 (the search phase). */
+    if ( actions & _UA_SEARCH_PHASE ) 
+        return _URC_CONTINUE_UNWIND;
+        
+    /* There is nothing to do if there is no LSDA for this frame. */
+    const uint8_t* lsda = _Unwind_GetLanguageSpecificData(context);
+    if ( lsda == (uint8_t*) 0 )
+        return _URC_CONTINUE_UNWIND;
+
+    uintptr_t pc = _Unwind_GetIP(context)-1;
+    uintptr_t funcStart = _Unwind_GetRegionStart(context);
+    uintptr_t pcOffset = pc - funcStart;
+
+    /* Parse LSDA header. */
+    uint8_t lpStartEncoding = *lsda++;
+    if (lpStartEncoding != DW_EH_PE_omit) {
+        readEncodedPointer(&lsda, lpStartEncoding); 
+    }
+    uint8_t ttypeEncoding = *lsda++;
+    if (ttypeEncoding != DW_EH_PE_omit) {
+        readULEB128(&lsda);  
+    }
+    /* Walk call-site table looking for range that includes current PC. */
+    uint8_t         callSiteEncoding = *lsda++;
+    uint32_t        callSiteTableLength = readULEB128(&lsda);
+    const uint8_t*  callSiteTableStart = lsda;
+    const uint8_t*  callSiteTableEnd = callSiteTableStart + callSiteTableLength;
+    const uint8_t* p=callSiteTableStart;
+    while (p < callSiteTableEnd) {
+        uintptr_t start = readEncodedPointer(&p, callSiteEncoding);
+        uintptr_t length = readEncodedPointer(&p, callSiteEncoding);
+        uintptr_t landingPad = readEncodedPointer(&p, callSiteEncoding);
+        readULEB128(&p); /* action value not used for C code */
+        if ( landingPad == 0 )
+            continue; /* no landing pad for this entry */
+        if ( (start <= pcOffset) && (pcOffset < (start+length)) ) {
+            /* Found landing pad for the PC.
+             * Set Instruction Pointer to so we re-enter function 
+             * at landing pad. The landing pad is created by the compiler
+             * to take two parameters in registers.
+	     */
+            _Unwind_SetGR(context, __builtin_eh_return_data_regno(0), 
+                                                (uintptr_t)exceptionObject);
+            _Unwind_SetGR(context, __builtin_eh_return_data_regno(1), 0);
+            _Unwind_SetIP(context, funcStart+landingPad);
+            return _URC_INSTALL_CONTEXT;
+        }
+    }
+    
+    /* No landing pad found, continue unwinding. */
+    return _URC_CONTINUE_UNWIND;
+}
+
diff --git a/lib/builtins/i386/Makefile.mk b/lib/builtins/i386/Makefile.mk
new file mode 100644
index 000000000..f3776a02c
--- /dev/null
+++ b/lib/builtins/i386/Makefile.mk
@@ -0,0 +1,20 @@
+#===- lib/builtins/i386/Makefile.mk ------------------------*- Makefile -*--===#
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+#===------------------------------------------------------------------------===#
+
+ModuleName := builtins
+SubDirs := 
+OnlyArchs := i386
+
+AsmSources := $(foreach file,$(wildcard $(Dir)/*.S),$(notdir $(file)))
+Sources := $(foreach file,$(wildcard $(Dir)/*.c),$(notdir $(file)))
+ObjNames := $(Sources:%.c=%.o) $(AsmSources:%.S=%.o)
+Implementation := Optimized
+
+# FIXME: use automatic dependencies?
+Dependencies := $(wildcard lib/*.h $(Dir)/*.h)
diff --git a/lib/builtins/i386/ashldi3.S b/lib/builtins/i386/ashldi3.S
new file mode 100644
index 000000000..a5da1ace1
--- /dev/null
+++ b/lib/builtins/i386/ashldi3.S
@@ -0,0 +1,58 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// di_int __ashldi3(di_int input, int count);
+
+// This routine has some extra memory traffic, loading the 64-bit input via two
+// 32-bit loads, then immediately storing it back to the stack via a single 64-bit
+// store.  This is to avoid a write-small, read-large stall.
+// However, if callers of this routine can be safely assumed to store the argument
+// via a 64-bt store, this is unnecessary memory traffic, and should be avoided.
+// It can be turned off by defining the TRUST_CALLERS_USE_64_BIT_STORES macro.
+
+#ifdef __i386__
+#ifdef __SSE2__
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__ashldi3)
+	movd	  12(%esp),		%xmm2	// Load count
+#ifndef TRUST_CALLERS_USE_64_BIT_STORES
+	movd	   4(%esp),		%xmm0
+	movd	   8(%esp),		%xmm1
+	punpckldq	%xmm1,		%xmm0	// Load input
+#else
+	movq	   4(%esp),		%xmm0	// Load input
+#endif
+	psllq		%xmm2,		%xmm0	// shift input by count
+	movd		%xmm0,		%eax
+	psrlq		$32,		%xmm0
+	movd		%xmm0,		%edx
+	ret
+END_COMPILERRT_FUNCTION(__ashldi3)
+
+#else // Use GPRs instead of SSE2 instructions, if they aren't available.
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__ashldi3)
+	movl	  12(%esp),		%ecx	// Load count
+	movl	   8(%esp),		%edx	// Load high
+	movl	   4(%esp),		%eax	// Load low
+
+	testl		$0x20,		%ecx	// If count >= 32
+	jnz		1f			//    goto 1
+	shldl		%cl, %eax,	%edx	// left shift high by count
+	shll		%cl,		%eax	// left shift low by count
+	ret
+
+1:	movl		%eax,		%edx	// Move low to high
+	xorl		%eax,		%eax	// clear low
+	shll		%cl,		%edx	// shift high by count - 32
+	ret
+END_COMPILERRT_FUNCTION(__ashldi3)
+
+#endif // __SSE2__
+#endif // __i386__
diff --git a/lib/builtins/i386/ashrdi3.S b/lib/builtins/i386/ashrdi3.S
new file mode 100644
index 000000000..ab0fcfcc9
--- /dev/null
+++ b/lib/builtins/i386/ashrdi3.S
@@ -0,0 +1,69 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// di_int __ashrdi3(di_int input, int count);
+
+#ifdef __i386__
+#ifdef __SSE2__
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__ashrdi3)
+	movd	  12(%esp),		%xmm2	// Load count
+	movl	   8(%esp),		%eax
+#ifndef TRUST_CALLERS_USE_64_BIT_STORES
+	movd	   4(%esp),		%xmm0
+	movd	   8(%esp),		%xmm1
+	punpckldq	%xmm1,		%xmm0	// Load input
+#else
+	movq	   4(%esp),		%xmm0	// Load input
+#endif
+
+	psrlq		%xmm2,		%xmm0	// unsigned shift input by count
+	
+	testl		%eax,		%eax	// check the sign-bit of the input
+	jns			1f					// early out for positive inputs
+	
+	// If the input is negative, we need to construct the shifted sign bit
+	// to or into the result, as xmm does not have a signed right shift.
+	pcmpeqb		%xmm1,		%xmm1	// -1ULL
+	psrlq		$58,		%xmm1	// 0x3f
+	pandn		%xmm1,		%xmm2	// 63 - count
+	pcmpeqb		%xmm1,		%xmm1	// -1ULL
+	psubq		%xmm1,		%xmm2	// 64 - count
+	psllq		%xmm2,		%xmm1	// -1 << (64 - count) = leading sign bits
+	por			%xmm1,		%xmm0
+	
+	// Move the result back to the general purpose registers and return
+1:	movd		%xmm0,		%eax
+	psrlq		$32,		%xmm0
+	movd		%xmm0,		%edx
+	ret
+END_COMPILERRT_FUNCTION(__ashrdi3)
+
+#else // Use GPRs instead of SSE2 instructions, if they aren't available.
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__ashrdi3)
+	movl	  12(%esp),		%ecx	// Load count
+	movl	   8(%esp),		%edx	// Load high
+	movl	   4(%esp),		%eax	// Load low
+	
+	testl		$0x20,		%ecx	// If count >= 32
+	jnz			1f					//    goto 1
+
+	shrdl		%cl, %edx,	%eax	// right shift low by count
+	sarl		%cl,		%edx	// right shift high by count
+	ret
+	
+1:	movl		%edx,		%eax	// Move high to low
+	sarl		$31,		%edx	// clear high
+	sarl		%cl,		%eax	// shift low by count - 32
+	ret
+END_COMPILERRT_FUNCTION(__ashrdi3)
+
+#endif // __SSE2__
+#endif // __i386__
diff --git a/lib/builtins/i386/divdi3.S b/lib/builtins/i386/divdi3.S
new file mode 100644
index 000000000..5afec5f9c
--- /dev/null
+++ b/lib/builtins/i386/divdi3.S
@@ -0,0 +1,162 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// di_int __divdi3(di_int a, di_int b);
+
+// result = a / b.
+// both inputs and the output are 64-bit signed integers.
+// This will do whatever the underlying hardware is set to do on division by zero.
+// No other exceptions are generated, as the divide cannot overflow.
+//
+// This is targeted at 32-bit x86 *only*, as this can be done directly in hardware
+// on x86_64.  The performance goal is ~40 cycles per divide, which is faster than
+// currently possible via simulation of integer divides on the x87 unit.
+//
+// Stephen Canon, December 2008
+
+#ifdef __i386__
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__divdi3)
+
+/* This is currently implemented by wrapping the unsigned divide up in an absolute
+   value, then restoring the correct sign at the end of the computation.  This could
+   certainly be improved upon. */
+
+	pushl		%esi
+	movl	 20(%esp),			%edx	// high word of b
+	movl	 16(%esp),			%eax	// low word of b
+	movl		%edx,			%ecx
+	sarl		$31,			%ecx	// (b < 0) ? -1 : 0
+	xorl		%ecx,			%eax
+	xorl		%ecx,			%edx	// EDX:EAX = (b < 0) ? not(b) : b
+	subl		%ecx,			%eax
+	sbbl		%ecx,			%edx	// EDX:EAX = abs(b)
+	movl		%edx,		 20(%esp)
+	movl		%eax,		 16(%esp)	// store abs(b) back to stack
+	movl		%ecx,			%esi	// set aside sign of b
+	
+	movl	 12(%esp),			%edx	// high word of b
+	movl	  8(%esp),			%eax	// low word of b
+	movl		%edx,			%ecx
+	sarl		$31,			%ecx	// (a < 0) ? -1 : 0
+	xorl		%ecx,			%eax
+	xorl		%ecx,			%edx	// EDX:EAX = (a < 0) ? not(a) : a
+	subl		%ecx,			%eax
+	sbbl		%ecx,			%edx	// EDX:EAX = abs(a)
+	movl		%edx,		 12(%esp)
+	movl		%eax,		  8(%esp)	// store abs(a) back to stack
+	xorl		%ecx,			%esi	// sign of result = (sign of a) ^ (sign of b)
+
+	pushl		%ebx
+	movl	 24(%esp),			%ebx	// Find the index i of the leading bit in b.
+	bsrl		%ebx,			%ecx	// If the high word of b is zero, jump to
+	jz			9f						// the code to handle that special case [9].
+	
+	/* High word of b is known to be non-zero on this branch */
+	
+	movl	 20(%esp),			%eax	// Construct bhi, containing bits [1+i:32+i] of b
+	
+	shrl		%cl,			%eax	// Practically, this means that bhi is given by:
+	shrl		%eax					//
+	notl		%ecx					//		bhi = (high word of b) << (31 - i) |
+	shll		%cl,			%ebx	//			  (low word of b) >> (1 + i)
+	orl			%eax,			%ebx	//
+	movl	 16(%esp),			%edx	// Load the high and low words of a, and jump
+	movl	 12(%esp),			%eax	// to [1] if the high word is larger than bhi
+	cmpl		%ebx,			%edx	// to avoid overflowing the upcoming divide.
+	jae			1f						
+		
+	/* High word of a is greater than or equal to (b >> (1 + i)) on this branch */
+	
+	divl		%ebx					// eax <-- qs, edx <-- r such that ahi:alo = bs*qs + r
+
+	pushl		%edi
+	notl		%ecx
+	shrl		%eax
+	shrl		%cl,			%eax	// q = qs >> (1 + i)
+	movl		%eax,			%edi
+	mull	 24(%esp)					// q*blo
+	movl	 16(%esp),			%ebx
+	movl	 20(%esp),			%ecx	// ECX:EBX = a
+	subl		%eax,			%ebx
+	sbbl		%edx,			%ecx	// ECX:EBX = a - q*blo
+	movl	 28(%esp),			%eax
+	imull		%edi,			%eax	// q*bhi
+	subl		%eax,			%ecx	// ECX:EBX = a - q*b
+	sbbl		$0,				%edi	// decrement q if remainder is negative
+	xorl		%edx,			%edx
+	movl		%edi,			%eax
+	
+	addl		%esi,			%eax	// Restore correct sign to result
+	adcl		%esi,			%edx
+	xorl		%esi,			%eax
+	xorl		%esi,			%edx
+	popl		%edi					// Restore callee-save registers
+	popl		%ebx
+	popl		%esi
+	retl								// Return
+
+
+1:	/* High word of a is greater than or equal to (b >> (1 + i)) on this branch */
+	 
+	subl		%ebx,			%edx	// subtract bhi from ahi so that divide will not
+	divl		%ebx					// overflow, and find q and r such that
+										//
+										//		ahi:alo = (1:q)*bhi + r
+										//
+										// Note that q is a number in (31-i).(1+i)
+										// fix point.
+
+	pushl		%edi
+	notl		%ecx
+	shrl		%eax
+	orl			$0x80000000,	%eax
+	shrl		%cl,			%eax	// q = (1:qs) >> (1 + i)
+	movl		%eax,			%edi
+	mull	 24(%esp)					// q*blo
+	movl	 16(%esp),			%ebx
+	movl	 20(%esp),			%ecx	// ECX:EBX = a
+	subl		%eax,			%ebx
+	sbbl		%edx,			%ecx	// ECX:EBX = a - q*blo
+	movl	 28(%esp),			%eax
+	imull		%edi,			%eax	// q*bhi
+	subl		%eax,			%ecx	// ECX:EBX = a - q*b
+	sbbl		$0,				%edi	// decrement q if remainder is negative
+	xorl		%edx,			%edx
+	movl		%edi,			%eax
+	
+	addl		%esi,			%eax	// Restore correct sign to result
+	adcl		%esi,			%edx
+	xorl		%esi,			%eax
+	xorl		%esi,			%edx
+	popl		%edi					// Restore callee-save registers
+	popl		%ebx
+	popl		%esi
+	retl								// Return
+
+	
+9:	/* High word of b is zero on this branch */
+
+	movl	 16(%esp),			%eax	// Find qhi and rhi such that
+	movl	 20(%esp),			%ecx	//
+	xorl		%edx,			%edx	//		ahi = qhi*b + rhi	with	0 ≤ rhi < b
+	divl		%ecx					//
+	movl		%eax,			%ebx	//
+	movl	 12(%esp),			%eax	// Find qlo such that
+	divl		%ecx					//
+	movl		%ebx,			%edx	//		rhi:alo = qlo*b + rlo  with 0 ≤ rlo < b
+	
+	addl		%esi,			%eax	// Restore correct sign to result
+	adcl		%esi,			%edx
+	xorl		%esi,			%eax
+	xorl		%esi,			%edx
+	popl		%ebx					// Restore callee-save registers
+	popl		%esi
+	retl								// Return
+END_COMPILERRT_FUNCTION(__divdi3)
+
+#endif // __i386__
diff --git a/lib/builtins/i386/floatdidf.S b/lib/builtins/i386/floatdidf.S
new file mode 100644
index 000000000..493608886
--- /dev/null
+++ b/lib/builtins/i386/floatdidf.S
@@ -0,0 +1,36 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// double __floatundidf(du_int a);
+
+#ifdef __i386__
+
+#ifndef __ELF__
+.const
+#endif
+.align 4
+twop52: .quad 0x4330000000000000
+twop32: .quad 0x41f0000000000000
+
+#define REL_ADDR(_a)	(_a)-0b(%eax)
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__floatdidf)
+	cvtsi2sd	8(%esp),			%xmm1
+	movss		4(%esp),			%xmm0 // low 32 bits of a
+	calll		0f
+0:	popl		%eax
+	mulsd		REL_ADDR(twop32),	%xmm1 // a_hi as a double (without rounding)
+	movsd		REL_ADDR(twop52),	%xmm2 // 0x1.0p52
+	subsd		%xmm2,				%xmm1 // a_hi - 0x1p52 (no rounding occurs)
+	orpd		%xmm2,				%xmm0 // 0x1p52 + a_lo (no rounding occurs)
+	addsd		%xmm1,				%xmm0 // a_hi + a_lo   (round happens here)
+	movsd		%xmm0,			   4(%esp)
+	fldl	   4(%esp)
+	ret
+END_COMPILERRT_FUNCTION(__floatdidf)
+
+#endif // __i386__
diff --git a/lib/builtins/i386/floatdisf.S b/lib/builtins/i386/floatdisf.S
new file mode 100644
index 000000000..8340331b8
--- /dev/null
+++ b/lib/builtins/i386/floatdisf.S
@@ -0,0 +1,32 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// float __floatdisf(di_int a);
+
+// This routine has some extra memory traffic, loading the 64-bit input via two
+// 32-bit loads, then immediately storing it back to the stack via a single 64-bit
+// store.  This is to avoid a write-small, read-large stall.
+// However, if callers of this routine can be safely assumed to store the argument
+// via a 64-bt store, this is unnecessary memory traffic, and should be avoided.
+// It can be turned off by defining the TRUST_CALLERS_USE_64_BIT_STORES macro.
+
+#ifdef __i386__
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__floatdisf)
+#ifndef TRUST_CALLERS_USE_64_BIT_STORES
+	movd		4(%esp),	%xmm0
+	movd		8(%esp),	%xmm1
+	punpckldq	%xmm1,		%xmm0
+	movq		%xmm0,		4(%esp)
+#endif
+	fildll		4(%esp)
+	fstps		4(%esp)
+	flds		4(%esp)
+	ret
+END_COMPILERRT_FUNCTION(__floatdisf)
+
+#endif // __i386__
diff --git a/lib/builtins/i386/floatdixf.S b/lib/builtins/i386/floatdixf.S
new file mode 100644
index 000000000..eaa34d744
--- /dev/null
+++ b/lib/builtins/i386/floatdixf.S
@@ -0,0 +1,30 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// float __floatdixf(di_int a);
+
+#ifdef __i386__
+
+// This routine has some extra memory traffic, loading the 64-bit input via two
+// 32-bit loads, then immediately storing it back to the stack via a single 64-bit
+// store.  This is to avoid a write-small, read-large stall.
+// However, if callers of this routine can be safely assumed to store the argument
+// via a 64-bt store, this is unnecessary memory traffic, and should be avoided.
+// It can be turned off by defining the TRUST_CALLERS_USE_64_BIT_STORES macro.
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__floatdixf)
+#ifndef TRUST_CALLERS_USE_64_BIT_STORES
+	movd		4(%esp),	%xmm0
+	movd		8(%esp),	%xmm1
+	punpckldq	%xmm1,		%xmm0
+	movq		%xmm0,		4(%esp)
+#endif
+	fildll		4(%esp)
+	ret
+END_COMPILERRT_FUNCTION(__floatdixf)
+
+#endif // __i386__
diff --git a/lib/builtins/i386/floatundidf.S b/lib/builtins/i386/floatundidf.S
new file mode 100644
index 000000000..5b8787ba6
--- /dev/null
+++ b/lib/builtins/i386/floatundidf.S
@@ -0,0 +1,47 @@
+//===-- floatundidf.S - Implement __floatundidf for i386 ------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements __floatundidf for the compiler_rt library.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+// double __floatundidf(du_int a);
+
+#ifdef __i386__
+
+#ifndef __ELF__
+.const
+#endif
+.align 4
+twop52: .quad 0x4330000000000000
+twop84_plus_twop52:
+		.quad 0x4530000000100000
+twop84: .quad 0x4530000000000000
+
+#define REL_ADDR(_a)	(_a)-0b(%eax)
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__floatundidf)
+	movss	8(%esp),						%xmm1 // high 32 bits of a
+	movss	4(%esp),						%xmm0 // low 32 bits of a
+	calll	0f
+0:	popl	%eax
+	orpd	REL_ADDR(twop84),				%xmm1 // 0x1p84 + a_hi (no rounding occurs)
+	subsd	REL_ADDR(twop84_plus_twop52),	%xmm1 // a_hi - 0x1p52 (no rounding occurs)
+	orpd	REL_ADDR(twop52),				%xmm0 // 0x1p52 + a_lo (no rounding occurs)
+	addsd	%xmm1,							%xmm0 // a_hi + a_lo   (round happens here)
+	movsd	%xmm0,						   4(%esp)
+	fldl   4(%esp)
+	ret
+END_COMPILERRT_FUNCTION(__floatundidf)
+
+#endif // __i386__
diff --git a/lib/builtins/i386/floatundisf.S b/lib/builtins/i386/floatundisf.S
new file mode 100644
index 000000000..2253021bc
--- /dev/null
+++ b/lib/builtins/i386/floatundisf.S
@@ -0,0 +1,101 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// float __floatundisf(du_int a);
+
+// Note that there is a hardware instruction, fildll, that does most of what
+// this function needs to do.  However, because of our ia32 ABI, it will take
+// a write-small read-large stall, so the software implementation here is
+// actually several cycles faster.
+
+// This is a branch-free implementation.  A branchy implementation might be
+// faster for the common case if you know something a priori about the input
+// distribution.
+
+/* branch-free x87 implementation - one cycle slower than without x87.
+
+#ifdef __i386__
+
+.const
+.align 3
+
+		.quad	0x43f0000000000000
+twop64:	.quad	0x0000000000000000
+
+#define			TWOp64			twop64-0b(%ecx,%eax,8)
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__floatundisf)
+	movl		8(%esp),		%eax
+	movd		8(%esp),		%xmm1
+	movd		4(%esp),		%xmm0
+	punpckldq	%xmm1,			%xmm0
+	calll		0f
+0:	popl		%ecx
+	sarl		$31,			%eax
+	movq		%xmm0,			4(%esp)
+	fildll		4(%esp)
+	faddl		TWOp64
+	fstps		4(%esp)
+	flds		4(%esp)
+	ret
+END_COMPILERRT_FUNCTION(__floatundisf)
+
+#endif // __i386__
+
+*/
+
+/* branch-free, x87-free implementation - faster at the expense of code size */
+
+#ifdef __i386__
+
+#ifndef __ELF__
+.const
+.align 3
+#else
+.align 8
+#endif
+twop52: .quad 0x4330000000000000
+		.quad 0x0000000000000fff
+sticky: .quad 0x0000000000000000
+		.long 0x00000012
+twelve:	.long 0x00000000
+
+#define			TWOp52			twop52-0b(%ecx)
+#define			STICKY			sticky-0b(%ecx,%eax,8)
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__floatundisf)
+	movl		8(%esp),		%eax
+	movd		8(%esp),		%xmm1
+	movd		4(%esp),		%xmm0
+	punpckldq	%xmm1,			%xmm0
+	
+	calll		0f
+0:	popl		%ecx
+	shrl		%eax					// high 31 bits of input as sint32
+	addl		$0x7ff80000,	%eax
+	sarl		$31,			%eax	// (big input) ? -1 : 0
+	movsd		STICKY,			%xmm1	// (big input) ? 0xfff : 0
+	movl		$12,			%edx
+	andl		%eax,			%edx	// (big input) ? 12 : 0
+	movd		%edx,			%xmm3
+	andpd		%xmm0,			%xmm1	// (big input) ? input & 0xfff : 0
+	movsd		TWOp52,			%xmm2	// 0x1.0p52
+	psrlq		%xmm3,			%xmm0	// (big input) ? input >> 12 : input
+	orpd		%xmm2,			%xmm1	// 0x1.0p52 + ((big input) ? input & 0xfff : input)
+	orpd		%xmm1,			%xmm0	// 0x1.0p52 + ((big input) ? (input >> 12 | input & 0xfff) : input)
+	subsd		%xmm2,			%xmm0	// (double)((big input) ? (input >> 12 | input & 0xfff) : input)
+	cvtsd2ss	%xmm0,			%xmm0	// (float)((big input) ? (input >> 12 | input & 0xfff) : input)
+	pslld		$23,			%xmm3
+	paddd		%xmm3,			%xmm0	// (float)input
+	movd		%xmm0,			4(%esp)
+	flds		4(%esp)
+	ret
+END_COMPILERRT_FUNCTION(__floatundisf)
+
+#endif // __i386__
diff --git a/lib/builtins/i386/floatundixf.S b/lib/builtins/i386/floatundixf.S
new file mode 100644
index 000000000..b728d0697
--- /dev/null
+++ b/lib/builtins/i386/floatundixf.S
@@ -0,0 +1,38 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// long double __floatundixf(du_int a);16
+
+#ifdef __i386__
+
+#ifndef __ELF__
+.const
+#endif
+.align 4
+twop52: .quad 0x4330000000000000
+twop84_plus_twop52_neg:
+		.quad 0xc530000000100000
+twop84: .quad 0x4530000000000000
+
+#define REL_ADDR(_a)	(_a)-0b(%eax)
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__floatundixf)
+	calll	0f
+0:	popl	%eax
+	movss	8(%esp),			%xmm0	// hi 32 bits of input
+	movss	4(%esp),			%xmm1	// lo 32 bits of input
+	orpd	REL_ADDR(twop84),	%xmm0	// 2^84 + hi (as a double)
+	orpd	REL_ADDR(twop52),	%xmm1	// 2^52 + lo (as a double)
+	addsd	REL_ADDR(twop84_plus_twop52_neg),	%xmm0	// hi - 2^52 (no rounding occurs)
+	movsd	%xmm1,				4(%esp)
+	fldl	4(%esp)
+	movsd	%xmm0,				4(%esp)
+	faddl	4(%esp)
+	ret
+END_COMPILERRT_FUNCTION(__floatundixf)
+
+#endif // __i386__
diff --git a/lib/builtins/i386/lshrdi3.S b/lib/builtins/i386/lshrdi3.S
new file mode 100644
index 000000000..dcfc33169
--- /dev/null
+++ b/lib/builtins/i386/lshrdi3.S
@@ -0,0 +1,59 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// di_int __lshrdi3(di_int input, int count);
+
+// This routine has some extra memory traffic, loading the 64-bit input via two
+// 32-bit loads, then immediately storing it back to the stack via a single 64-bit
+// store.  This is to avoid a write-small, read-large stall.
+// However, if callers of this routine can be safely assumed to store the argument
+// via a 64-bt store, this is unnecessary memory traffic, and should be avoided.
+// It can be turned off by defining the TRUST_CALLERS_USE_64_BIT_STORES macro.
+
+#ifdef __i386__
+#ifdef __SSE2__
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__lshrdi3)
+	movd	  12(%esp),		%xmm2	// Load count
+#ifndef TRUST_CALLERS_USE_64_BIT_STORES
+	movd	   4(%esp),		%xmm0
+	movd	   8(%esp),		%xmm1
+	punpckldq	%xmm1,		%xmm0	// Load input
+#else
+	movq	   4(%esp),		%xmm0	// Load input
+#endif
+	psrlq		%xmm2,		%xmm0	// shift input by count
+	movd		%xmm0,		%eax
+	psrlq		$32,		%xmm0
+	movd		%xmm0,		%edx
+	ret
+END_COMPILERRT_FUNCTION(__lshrdi3)
+
+#else // Use GPRs instead of SSE2 instructions, if they aren't available.
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__lshrdi3)
+	movl	  12(%esp),		%ecx	// Load count
+	movl	   8(%esp),		%edx	// Load high
+	movl	   4(%esp),		%eax	// Load low
+	
+	testl		$0x20,		%ecx	// If count >= 32
+	jnz			1f					//    goto 1
+
+	shrdl		%cl, %edx,	%eax	// right shift low by count
+	shrl		%cl,		%edx	// right shift high by count
+	ret
+	
+1:	movl		%edx,		%eax	// Move high to low
+	xorl		%edx,		%edx	// clear high
+	shrl		%cl,		%eax	// shift low by count - 32
+	ret
+END_COMPILERRT_FUNCTION(__lshrdi3)
+
+#endif // __SSE2__
+#endif // __i386__
diff --git a/lib/builtins/i386/moddi3.S b/lib/builtins/i386/moddi3.S
new file mode 100644
index 000000000..79e4fc213
--- /dev/null
+++ b/lib/builtins/i386/moddi3.S
@@ -0,0 +1,166 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// di_int __moddi3(di_int a, di_int b);
+
+// result = remainder of a / b.
+// both inputs and the output are 64-bit signed integers.
+// This will do whatever the underlying hardware is set to do on division by zero.
+// No other exceptions are generated, as the divide cannot overflow.
+//
+// This is targeted at 32-bit x86 *only*, as this can be done directly in hardware
+// on x86_64.  The performance goal is ~40 cycles per divide, which is faster than
+// currently possible via simulation of integer divides on the x87 unit.
+//
+
+// Stephen Canon, December 2008
+
+#ifdef __i386__
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__moddi3)
+
+/* This is currently implemented by wrapping the unsigned modulus up in an absolute
+   value.  This could certainly be improved upon. */
+
+	pushl		%esi
+	movl	 20(%esp),			%edx	// high word of b
+	movl	 16(%esp),			%eax	// low word of b
+	movl		%edx,			%ecx
+	sarl		$31,			%ecx	// (b < 0) ? -1 : 0
+	xorl		%ecx,			%eax
+	xorl		%ecx,			%edx	// EDX:EAX = (b < 0) ? not(b) : b
+	subl		%ecx,			%eax
+	sbbl		%ecx,			%edx	// EDX:EAX = abs(b)
+	movl		%edx,		 20(%esp)
+	movl		%eax,		 16(%esp)	// store abs(b) back to stack
+	
+	movl	 12(%esp),			%edx	// high word of b
+	movl	  8(%esp),			%eax	// low word of b
+	movl		%edx,			%ecx
+	sarl		$31,			%ecx	// (a < 0) ? -1 : 0
+	xorl		%ecx,			%eax
+	xorl		%ecx,			%edx	// EDX:EAX = (a < 0) ? not(a) : a
+	subl		%ecx,			%eax
+	sbbl		%ecx,			%edx	// EDX:EAX = abs(a)
+	movl		%edx,		 12(%esp)
+	movl		%eax,		  8(%esp)	// store abs(a) back to stack
+	movl		%ecx,			%esi	// set aside sign of a
+
+	pushl		%ebx
+	movl	 24(%esp),			%ebx	// Find the index i of the leading bit in b.
+	bsrl		%ebx,			%ecx	// If the high word of b is zero, jump to
+	jz			9f						// the code to handle that special case [9].
+	
+	/* High word of b is known to be non-zero on this branch */
+	
+	movl	 20(%esp),			%eax	// Construct bhi, containing bits [1+i:32+i] of b
+	
+	shrl		%cl,			%eax	// Practically, this means that bhi is given by:
+	shrl		%eax					//
+	notl		%ecx					//		bhi = (high word of b) << (31 - i) |
+	shll		%cl,			%ebx	//			  (low word of b) >> (1 + i)
+	orl			%eax,			%ebx	//
+	movl	 16(%esp),			%edx	// Load the high and low words of a, and jump
+	movl	 12(%esp),			%eax	// to [2] if the high word is larger than bhi
+	cmpl		%ebx,			%edx	// to avoid overflowing the upcoming divide.
+	jae			2f						
+		
+	/* High word of a is greater than or equal to (b >> (1 + i)) on this branch */
+	
+	divl		%ebx					// eax <-- qs, edx <-- r such that ahi:alo = bs*qs + r
+
+	pushl		%edi
+	notl		%ecx
+	shrl		%eax
+	shrl		%cl,			%eax	// q = qs >> (1 + i)
+	movl		%eax,			%edi
+	mull	 24(%esp)					// q*blo
+	movl	 16(%esp),			%ebx
+	movl	 20(%esp),			%ecx	// ECX:EBX = a
+	subl		%eax,			%ebx
+	sbbl		%edx,			%ecx	// ECX:EBX = a - q*blo
+	movl	 28(%esp),			%eax
+	imull		%edi,			%eax	// q*bhi
+	subl		%eax,			%ecx	// ECX:EBX = a - q*b
+	
+	jnc			1f						// if positive, this is the result.
+	addl	 24(%esp),			%ebx	// otherwise
+	adcl	 28(%esp),			%ecx	// ECX:EBX = a - (q-1)*b = result
+1:	movl		%ebx,			%eax
+	movl		%ecx,			%edx
+	
+	addl		%esi,			%eax	// Restore correct sign to result
+	adcl		%esi,			%edx
+	xorl		%esi,			%eax
+	xorl		%esi,			%edx
+	popl		%edi					// Restore callee-save registers
+	popl		%ebx
+	popl		%esi
+	retl								// Return
+
+2:	/* High word of a is greater than or equal to (b >> (1 + i)) on this branch */
+	 
+	subl		%ebx,			%edx	// subtract bhi from ahi so that divide will not
+	divl		%ebx					// overflow, and find q and r such that
+										//
+										//		ahi:alo = (1:q)*bhi + r
+										//
+										// Note that q is a number in (31-i).(1+i)
+										// fix point.
+
+	pushl		%edi
+	notl		%ecx
+	shrl		%eax
+	orl			$0x80000000,	%eax
+	shrl		%cl,			%eax	// q = (1:qs) >> (1 + i)
+	movl		%eax,			%edi
+	mull	 24(%esp)					// q*blo
+	movl	 16(%esp),			%ebx
+	movl	 20(%esp),			%ecx	// ECX:EBX = a
+	subl		%eax,			%ebx
+	sbbl		%edx,			%ecx	// ECX:EBX = a - q*blo
+	movl	 28(%esp),			%eax
+	imull		%edi,			%eax	// q*bhi
+	subl		%eax,			%ecx	// ECX:EBX = a - q*b
+
+	jnc			3f						// if positive, this is the result.
+	addl	 24(%esp),			%ebx	// otherwise
+	adcl	 28(%esp),			%ecx	// ECX:EBX = a - (q-1)*b = result
+3:	movl		%ebx,			%eax
+	movl		%ecx,			%edx
+	
+	addl		%esi,			%eax	// Restore correct sign to result
+	adcl		%esi,			%edx
+	xorl		%esi,			%eax
+	xorl		%esi,			%edx
+	popl		%edi					// Restore callee-save registers
+	popl		%ebx
+	popl		%esi
+	retl								// Return
+	
+9:	/* High word of b is zero on this branch */
+
+	movl	 16(%esp),			%eax	// Find qhi and rhi such that
+	movl	 20(%esp),			%ecx	//
+	xorl		%edx,			%edx	//		ahi = qhi*b + rhi	with	0 ≤ rhi < b
+	divl		%ecx					//
+	movl		%eax,			%ebx	//
+	movl	 12(%esp),			%eax	// Find rlo such that
+	divl		%ecx					//
+	movl		%edx,			%eax	//		rhi:alo = qlo*b + rlo  with 0 ≤ rlo < b
+	popl		%ebx					//
+	xorl		%edx,			%edx	// and return 0:rlo
+
+	addl		%esi,			%eax	// Restore correct sign to result
+	adcl		%esi,			%edx
+	xorl		%esi,			%eax
+	xorl		%esi,			%edx
+	popl		%esi
+	retl								// Return
+END_COMPILERRT_FUNCTION(__moddi3)
+
+#endif // __i386__
diff --git a/lib/builtins/i386/muldi3.S b/lib/builtins/i386/muldi3.S
new file mode 100644
index 000000000..5af1136c1
--- /dev/null
+++ b/lib/builtins/i386/muldi3.S
@@ -0,0 +1,30 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// di_int __muldi3(di_int a, di_int b);
+
+#ifdef __i386__
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__muldi3)
+	pushl	%ebx
+	movl  16(%esp),		%eax	// b.lo
+	movl  12(%esp),		%ecx	// a.hi
+	imull	%eax,		%ecx	// b.lo * a.hi
+	
+	movl   8(%esp),		%edx	// a.lo
+	movl  20(%esp),		%ebx	// b.hi
+	imull	%edx,		%ebx	// a.lo * b.hi
+	
+	mull	%edx				// EDX:EAX = a.lo * b.lo
+	addl	%ecx,		%ebx	// EBX = (a.lo*b.hi + a.hi*b.lo)
+	addl	%ebx,		%edx
+	
+	popl	%ebx
+	retl
+END_COMPILERRT_FUNCTION(__muldi3)
+
+#endif // __i386__
diff --git a/lib/builtins/i386/udivdi3.S b/lib/builtins/i386/udivdi3.S
new file mode 100644
index 000000000..5072b221c
--- /dev/null
+++ b/lib/builtins/i386/udivdi3.S
@@ -0,0 +1,115 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// du_int __udivdi3(du_int a, du_int b);
+
+// result = a / b.
+// both inputs and the output are 64-bit unsigned integers.
+// This will do whatever the underlying hardware is set to do on division by zero.
+// No other exceptions are generated, as the divide cannot overflow.
+//
+// This is targeted at 32-bit x86 *only*, as this can be done directly in hardware
+// on x86_64.  The performance goal is ~40 cycles per divide, which is faster than
+// currently possible via simulation of integer divides on the x87 unit.
+//
+// Stephen Canon, December 2008
+
+#ifdef __i386__
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__udivdi3)
+
+	pushl		%ebx
+	movl	 20(%esp),			%ebx	// Find the index i of the leading bit in b.
+	bsrl		%ebx,			%ecx	// If the high word of b is zero, jump to
+	jz			9f						// the code to handle that special case [9].
+	
+	/* High word of b is known to be non-zero on this branch */
+	
+	movl	 16(%esp),			%eax	// Construct bhi, containing bits [1+i:32+i] of b
+	
+	shrl		%cl,			%eax	// Practically, this means that bhi is given by:
+	shrl		%eax					//
+	notl		%ecx					//		bhi = (high word of b) << (31 - i) |
+	shll		%cl,			%ebx	//			  (low word of b) >> (1 + i)
+	orl			%eax,			%ebx	//
+	movl	 12(%esp),			%edx	// Load the high and low words of a, and jump
+	movl	  8(%esp),			%eax	// to [1] if the high word is larger than bhi
+	cmpl		%ebx,			%edx	// to avoid overflowing the upcoming divide.
+	jae			1f						
+		
+	/* High word of a is greater than or equal to (b >> (1 + i)) on this branch */
+	
+	divl		%ebx					// eax <-- qs, edx <-- r such that ahi:alo = bs*qs + r
+
+	pushl		%edi
+	notl		%ecx
+	shrl		%eax
+	shrl		%cl,			%eax	// q = qs >> (1 + i)
+	movl		%eax,			%edi
+	mull	 20(%esp)					// q*blo
+	movl	 12(%esp),			%ebx
+	movl	 16(%esp),			%ecx	// ECX:EBX = a
+	subl		%eax,			%ebx
+	sbbl		%edx,			%ecx	// ECX:EBX = a - q*blo
+	movl	 24(%esp),			%eax
+	imull		%edi,			%eax	// q*bhi
+	subl		%eax,			%ecx	// ECX:EBX = a - q*b
+	sbbl		$0,				%edi	// decrement q if remainder is negative
+	xorl		%edx,			%edx
+	movl		%edi,			%eax
+	popl		%edi
+	popl		%ebx
+	retl
+
+
+1:	/* High word of a is greater than or equal to (b >> (1 + i)) on this branch */
+	 
+	subl		%ebx,			%edx	// subtract bhi from ahi so that divide will not
+	divl		%ebx					// overflow, and find q and r such that
+										//
+										//		ahi:alo = (1:q)*bhi + r
+										//
+										// Note that q is a number in (31-i).(1+i)
+										// fix point.
+
+	pushl		%edi
+	notl		%ecx
+	shrl		%eax
+	orl			$0x80000000,	%eax
+	shrl		%cl,			%eax	// q = (1:qs) >> (1 + i)
+	movl		%eax,			%edi
+	mull	 20(%esp)					// q*blo
+	movl	 12(%esp),			%ebx
+	movl	 16(%esp),			%ecx	// ECX:EBX = a
+	subl		%eax,			%ebx
+	sbbl		%edx,			%ecx	// ECX:EBX = a - q*blo
+	movl	 24(%esp),			%eax
+	imull		%edi,			%eax	// q*bhi
+	subl		%eax,			%ecx	// ECX:EBX = a - q*b
+	sbbl		$0,				%edi	// decrement q if remainder is negative
+	xorl		%edx,			%edx
+	movl		%edi,			%eax
+	popl		%edi
+	popl		%ebx
+	retl
+
+	
+9:	/* High word of b is zero on this branch */
+
+	movl	 12(%esp),			%eax	// Find qhi and rhi such that
+	movl	 16(%esp),			%ecx	//
+	xorl		%edx,			%edx	//		ahi = qhi*b + rhi	with	0 ≤ rhi < b
+	divl		%ecx					//
+	movl		%eax,			%ebx	//
+	movl	  8(%esp),			%eax	// Find qlo such that
+	divl		%ecx					//
+	movl		%ebx,			%edx	//		rhi:alo = qlo*b + rlo  with 0 ≤ rlo < b
+	popl		%ebx					//
+	retl								// and return qhi:qlo
+END_COMPILERRT_FUNCTION(__udivdi3)
+
+#endif // __i386__
diff --git a/lib/builtins/i386/umoddi3.S b/lib/builtins/i386/umoddi3.S
new file mode 100644
index 000000000..63cefc207
--- /dev/null
+++ b/lib/builtins/i386/umoddi3.S
@@ -0,0 +1,126 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// du_int __umoddi3(du_int a, du_int b);
+
+// result = remainder of a / b.
+// both inputs and the output are 64-bit unsigned integers.
+// This will do whatever the underlying hardware is set to do on division by zero.
+// No other exceptions are generated, as the divide cannot overflow.
+//
+// This is targeted at 32-bit x86 *only*, as this can be done directly in hardware
+// on x86_64.  The performance goal is ~40 cycles per divide, which is faster than
+// currently possible via simulation of integer divides on the x87 unit.
+//
+
+// Stephen Canon, December 2008
+
+#ifdef __i386__
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__umoddi3)
+
+	pushl		%ebx
+	movl	 20(%esp),			%ebx	// Find the index i of the leading bit in b.
+	bsrl		%ebx,			%ecx	// If the high word of b is zero, jump to
+	jz			9f						// the code to handle that special case [9].
+	
+	/* High word of b is known to be non-zero on this branch */
+	
+	movl	 16(%esp),			%eax	// Construct bhi, containing bits [1+i:32+i] of b
+	
+	shrl		%cl,			%eax	// Practically, this means that bhi is given by:
+	shrl		%eax					//
+	notl		%ecx					//		bhi = (high word of b) << (31 - i) |
+	shll		%cl,			%ebx	//			  (low word of b) >> (1 + i)
+	orl			%eax,			%ebx	//
+	movl	 12(%esp),			%edx	// Load the high and low words of a, and jump
+	movl	  8(%esp),			%eax	// to [2] if the high word is larger than bhi
+	cmpl		%ebx,			%edx	// to avoid overflowing the upcoming divide.
+	jae			2f						
+		
+	/* High word of a is greater than or equal to (b >> (1 + i)) on this branch */
+	
+	divl		%ebx					// eax <-- qs, edx <-- r such that ahi:alo = bs*qs + r
+
+	pushl		%edi
+	notl		%ecx
+	shrl		%eax
+	shrl		%cl,			%eax	// q = qs >> (1 + i)
+	movl		%eax,			%edi
+	mull	 20(%esp)					// q*blo
+	movl	 12(%esp),			%ebx
+	movl	 16(%esp),			%ecx	// ECX:EBX = a
+	subl		%eax,			%ebx
+	sbbl		%edx,			%ecx	// ECX:EBX = a - q*blo
+	movl	 24(%esp),			%eax
+	imull		%edi,			%eax	// q*bhi
+	subl		%eax,			%ecx	// ECX:EBX = a - q*b
+	
+	jnc			1f						// if positive, this is the result.
+	addl	 20(%esp),			%ebx	// otherwise
+	adcl	 24(%esp),			%ecx	// ECX:EBX = a - (q-1)*b = result
+1:	movl		%ebx,			%eax
+	movl		%ecx,			%edx
+	
+	popl		%edi
+	popl		%ebx
+	retl
+
+
+2:	/* High word of a is greater than or equal to (b >> (1 + i)) on this branch */
+	 
+	subl		%ebx,			%edx	// subtract bhi from ahi so that divide will not
+	divl		%ebx					// overflow, and find q and r such that
+										//
+										//		ahi:alo = (1:q)*bhi + r
+										//
+										// Note that q is a number in (31-i).(1+i)
+										// fix point.
+
+	pushl		%edi
+	notl		%ecx
+	shrl		%eax
+	orl			$0x80000000,	%eax
+	shrl		%cl,			%eax	// q = (1:qs) >> (1 + i)
+	movl		%eax,			%edi
+	mull	 20(%esp)					// q*blo
+	movl	 12(%esp),			%ebx
+	movl	 16(%esp),			%ecx	// ECX:EBX = a
+	subl		%eax,			%ebx
+	sbbl		%edx,			%ecx	// ECX:EBX = a - q*blo
+	movl	 24(%esp),			%eax
+	imull		%edi,			%eax	// q*bhi
+	subl		%eax,			%ecx	// ECX:EBX = a - q*b
+
+	jnc			3f						// if positive, this is the result.
+	addl	 20(%esp),			%ebx	// otherwise
+	adcl	 24(%esp),			%ecx	// ECX:EBX = a - (q-1)*b = result
+3:	movl		%ebx,			%eax
+	movl		%ecx,			%edx
+	
+	popl		%edi
+	popl		%ebx
+	retl
+
+
+	
+9:	/* High word of b is zero on this branch */
+
+	movl	 12(%esp),			%eax	// Find qhi and rhi such that
+	movl	 16(%esp),			%ecx	//
+	xorl		%edx,			%edx	//		ahi = qhi*b + rhi	with	0 ≤ rhi < b
+	divl		%ecx					//
+	movl		%eax,			%ebx	//
+	movl	  8(%esp),			%eax	// Find rlo such that
+	divl		%ecx					//
+	movl		%edx,			%eax	//		rhi:alo = qlo*b + rlo  with 0 ≤ rlo < b
+	popl		%ebx					//
+	xorl		%edx,			%edx	// and return 0:rlo
+	retl								// 
+END_COMPILERRT_FUNCTION(__umoddi3)
+
+#endif // __i386__
diff --git a/lib/builtins/int_endianness.h b/lib/builtins/int_endianness.h
new file mode 100644
index 000000000..c465a985e
--- /dev/null
+++ b/lib/builtins/int_endianness.h
@@ -0,0 +1,111 @@
+/* ===-- int_endianness.h - configuration header for compiler-rt ------------===
+ *
+ *		       The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file is a configuration header for compiler-rt.
+ * This file is not part of the interface of this library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#ifndef INT_ENDIANNESS_H
+#define INT_ENDIANNESS_H
+
+#if defined(__SVR4) && defined(__sun)
+#include <sys/byteorder.h>
+
+#if defined(_BIG_ENDIAN)
+#define _YUGA_LITTLE_ENDIAN 0
+#define _YUGA_BIG_ENDIAN    1
+#elif defined(_LITTLE_ENDIAN)
+#define _YUGA_LITTLE_ENDIAN 1
+#define _YUGA_BIG_ENDIAN    0
+#else /* !_LITTLE_ENDIAN */
+#error "unknown endianness"
+#endif /* !_LITTLE_ENDIAN */
+
+#endif /* Solaris and AuroraUX. */
+
+/* .. */
+
+#if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__DragonFly__) || defined(__minix)
+#include <sys/endian.h>
+
+#if _BYTE_ORDER == _BIG_ENDIAN
+#define _YUGA_LITTLE_ENDIAN 0
+#define _YUGA_BIG_ENDIAN    1
+#elif _BYTE_ORDER == _LITTLE_ENDIAN
+#define _YUGA_LITTLE_ENDIAN 1
+#define _YUGA_BIG_ENDIAN    0
+#endif /* _BYTE_ORDER */
+
+#endif /* *BSD */
+
+#if defined(__OpenBSD__) || defined(__Bitrig__)
+#include <machine/endian.h>
+
+#if _BYTE_ORDER == _BIG_ENDIAN
+#define _YUGA_LITTLE_ENDIAN 0
+#define _YUGA_BIG_ENDIAN    1
+#elif _BYTE_ORDER == _LITTLE_ENDIAN
+#define _YUGA_LITTLE_ENDIAN 1
+#define _YUGA_BIG_ENDIAN    0
+#endif /* _BYTE_ORDER */
+
+#endif /* OpenBSD and Bitrig. */
+
+/* .. */
+
+/* Mac OSX has __BIG_ENDIAN__ or __LITTLE_ENDIAN__ automatically set by the compiler (at least with GCC) */
+#if defined(__APPLE__) || defined(__ellcc__ )
+
+#ifdef __BIG_ENDIAN__
+#if __BIG_ENDIAN__
+#define _YUGA_LITTLE_ENDIAN 0
+#define _YUGA_BIG_ENDIAN    1
+#endif
+#endif /* __BIG_ENDIAN__ */
+
+#ifdef __LITTLE_ENDIAN__
+#if __LITTLE_ENDIAN__
+#define _YUGA_LITTLE_ENDIAN 1
+#define _YUGA_BIG_ENDIAN    0
+#endif
+#endif /* __LITTLE_ENDIAN__ */
+
+#endif /* Mac OSX */
+
+/* .. */
+
+#if defined(__linux__)
+#include <endian.h>
+
+#if __BYTE_ORDER == __BIG_ENDIAN
+#define _YUGA_LITTLE_ENDIAN 0
+#define _YUGA_BIG_ENDIAN    1
+#elif __BYTE_ORDER == __LITTLE_ENDIAN
+#define _YUGA_LITTLE_ENDIAN 1
+#define _YUGA_BIG_ENDIAN    0
+#endif /* __BYTE_ORDER */
+
+#endif /* GNU/Linux */
+
+#if defined(_WIN32)
+
+#define _YUGA_LITTLE_ENDIAN 1
+#define _YUGA_BIG_ENDIAN    0
+
+#endif /* Windows */
+
+/* . */
+
+#if !defined(_YUGA_LITTLE_ENDIAN) || !defined(_YUGA_BIG_ENDIAN)
+#error Unable to determine endian
+#endif /* Check we found an endianness correctly. */
+
+#endif /* INT_ENDIANNESS_H */
diff --git a/lib/builtins/int_lib.h b/lib/builtins/int_lib.h
new file mode 100644
index 000000000..daeaa9e57
--- /dev/null
+++ b/lib/builtins/int_lib.h
@@ -0,0 +1,56 @@
+/* ===-- int_lib.h - configuration header for compiler-rt  -----------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file is a configuration header for compiler-rt.
+ * This file is not part of the interface of this library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#ifndef INT_LIB_H
+#define INT_LIB_H
+
+/* Assumption: Signed integral is 2's complement. */
+/* Assumption: Right shift of signed negative is arithmetic shift. */
+/* Assumption: Endianness is little or big (not mixed). */
+
+/* ABI macro definitions */
+
+#if __ARM_EABI__
+# define ARM_EABI_FNALIAS(aeabi_name, name)         \
+  void __aeabi_##aeabi_name() __attribute__((alias("__" #name)));
+# define COMPILER_RT_ABI __attribute__((pcs("aapcs")))
+#else
+# define ARM_EABI_FNALIAS(aeabi_name, name)
+# define COMPILER_RT_ABI
+#endif
+
+#if defined(__NetBSD__) && (defined(_KERNEL) || defined(_STANDALONE))
+/*
+ * Kernel and boot environment can't use normal headers,
+ * so use the equivalent system headers.
+ */
+#  include <machine/limits.h>
+#  include <sys/stdint.h>
+#  include <sys/types.h>
+#else
+/* Include the standard compiler builtin headers we use functionality from. */
+#  include <limits.h>
+#  include <stdint.h>
+#  include <stdbool.h>
+#  include <float.h>
+#endif
+
+/* Include the commonly used internal type definitions. */
+#include "int_types.h"
+
+/* Include internal utility function declarations. */
+#include "int_util.h"
+
+#endif /* INT_LIB_H */
diff --git a/lib/builtins/int_math.h b/lib/builtins/int_math.h
new file mode 100644
index 000000000..d6b4bdae1
--- /dev/null
+++ b/lib/builtins/int_math.h
@@ -0,0 +1,67 @@
+/* ===-- int_math.h - internal math inlines ---------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===-----------------------------------------------------------------------===
+ *
+ * This file is not part of the interface of this library.
+ *
+ * This file defines substitutes for the libm functions used in some of the
+ * compiler-rt implementations, defined in such a way that there is not a direct
+ * dependency on libm or math.h. Instead, we use the compiler builtin versions
+ * where available. This reduces our dependencies on the system SDK by foisting
+ * the responsibility onto the compiler.
+ *
+ * ===-----------------------------------------------------------------------===
+ */
+
+#ifndef INT_MATH_H
+#define INT_MATH_H
+
+#ifndef __has_builtin
+#  define  __has_builtin(x) 0
+#endif
+
+#define CRT_INFINITY __builtin_huge_valf()
+
+#define crt_isinf(x) __builtin_isinf((x))
+#define crt_isnan(x) __builtin_isnan((x))
+
+/* Define crt_isfinite in terms of the builtin if available, otherwise provide
+ * an alternate version in terms of our other functions. This supports some
+ * versions of GCC which didn't have __builtin_isfinite.
+ */
+#if __has_builtin(__builtin_isfinite)
+#  define crt_isfinite(x) __builtin_isfinite((x))
+#else
+#  define crt_isfinite(x) \
+  __extension__(({ \
+      __typeof((x)) x_ = (x); \
+      !crt_isinf(x_) && !crt_isnan(x_); \
+    }))
+#endif
+
+#define crt_copysign(x, y) __builtin_copysign((x), (y))
+#define crt_copysignf(x, y) __builtin_copysignf((x), (y))
+#define crt_copysignl(x, y) __builtin_copysignl((x), (y))
+
+#define crt_fabs(x) __builtin_fabs((x))
+#define crt_fabsf(x) __builtin_fabsf((x))
+#define crt_fabsl(x) __builtin_fabsl((x))
+
+#define crt_fmax(x, y) __builtin_fmax((x), (y))
+#define crt_fmaxf(x, y) __builtin_fmaxf((x), (y))
+#define crt_fmaxl(x, y) __builtin_fmaxl((x), (y))
+
+#define crt_logb(x) __builtin_logb((x))
+#define crt_logbf(x) __builtin_logbf((x))
+#define crt_logbl(x) __builtin_logbl((x))
+
+#define crt_scalbn(x, y) __builtin_scalbn((x), (y))
+#define crt_scalbnf(x, y) __builtin_scalbnf((x), (y))
+#define crt_scalbnl(x, y) __builtin_scalbnl((x), (y))
+
+#endif /* INT_MATH_H */
diff --git a/lib/builtins/int_types.h b/lib/builtins/int_types.h
new file mode 100644
index 000000000..fcce390f9
--- /dev/null
+++ b/lib/builtins/int_types.h
@@ -0,0 +1,140 @@
+/* ===-- int_lib.h - configuration header for compiler-rt  -----------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file is not part of the interface of this library.
+ *
+ * This file defines various standard types, most importantly a number of unions
+ * used to access parts of larger types.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#ifndef INT_TYPES_H
+#define INT_TYPES_H
+
+#include "int_endianness.h"
+
+typedef      int si_int;
+typedef unsigned su_int;
+
+typedef          long long di_int;
+typedef unsigned long long du_int;
+
+typedef union
+{
+    di_int all;
+    struct
+    {
+#if _YUGA_LITTLE_ENDIAN
+        su_int low;
+        si_int high;
+#else
+        si_int high;
+        su_int low;
+#endif /* _YUGA_LITTLE_ENDIAN */
+    }s;
+} dwords;
+
+typedef union
+{
+    du_int all;
+    struct
+    {
+#if _YUGA_LITTLE_ENDIAN
+        su_int low;
+        su_int high;
+#else
+        su_int high;
+        su_int low;
+#endif /* _YUGA_LITTLE_ENDIAN */
+    }s;
+} udwords;
+
+#if __x86_64
+
+typedef int      ti_int __attribute__ ((mode (TI)));
+typedef unsigned tu_int __attribute__ ((mode (TI)));
+
+typedef union
+{
+    ti_int all;
+    struct
+    {
+#if _YUGA_LITTLE_ENDIAN
+        du_int low;
+        di_int high;
+#else
+        di_int high;
+        du_int low;
+#endif /* _YUGA_LITTLE_ENDIAN */
+    }s;
+} twords;
+
+typedef union
+{
+    tu_int all;
+    struct
+    {
+#if _YUGA_LITTLE_ENDIAN
+        du_int low;
+        du_int high;
+#else
+        du_int high;
+        du_int low;
+#endif /* _YUGA_LITTLE_ENDIAN */
+    }s;
+} utwords;
+
+static inline ti_int make_ti(di_int h, di_int l) {
+    twords r;
+    r.s.high = h;
+    r.s.low = l;
+    return r.all;
+}
+
+static inline tu_int make_tu(du_int h, du_int l) {
+    utwords r;
+    r.s.high = h;
+    r.s.low = l;
+    return r.all;
+}
+
+#endif /* __x86_64 */
+
+typedef union
+{
+    su_int u;
+    float f;
+} float_bits;
+
+typedef union
+{
+    udwords u;
+    double  f;
+} double_bits;
+
+typedef struct
+{
+#if _YUGA_LITTLE_ENDIAN
+    udwords low;
+    udwords high;
+#else
+    udwords high;
+    udwords low;
+#endif /* _YUGA_LITTLE_ENDIAN */
+} uqwords;
+
+typedef union
+{
+    uqwords     u;
+    long double f;
+} long_double_bits;
+
+#endif /* INT_TYPES_H */
+
diff --git a/lib/builtins/int_util.c b/lib/builtins/int_util.c
new file mode 100644
index 000000000..323e46179
--- /dev/null
+++ b/lib/builtins/int_util.c
@@ -0,0 +1,61 @@
+/* ===-- int_util.c - Implement internal utilities --------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_util.h"
+#include "int_lib.h"
+
+/* NOTE: The definitions in this file are declared weak because we clients to be
+ * able to arbitrarily package individual functions into separate .a files. If
+ * we did not declare these weak, some link situations might end up seeing
+ * duplicate strong definitions of the same symbol.
+ *
+ * We can't use this solution for kernel use (which may not support weak), but
+ * currently expect that when built for kernel use all the functionality is
+ * packaged into a single library.
+ */
+
+#ifdef KERNEL_USE
+
+extern void panic(const char *, ...) __attribute__((noreturn));
+#ifndef _WIN32
+__attribute__((visibility("hidden")))
+#endif
+void compilerrt_abort_impl(const char *file, int line, const char *function) {
+  panic("%s:%d: abort in %s", file, line, function);
+}
+
+#elif __APPLE__
+
+/* from libSystem.dylib */
+extern void __assert_rtn(const char *func, const char *file, 
+                     int line, const char * message) __attribute__((noreturn));
+
+#ifndef _WIN32
+__attribute__((weak))
+__attribute__((visibility("hidden")))
+#endif
+void compilerrt_abort_impl(const char *file, int line, const char *function) {
+  __assert_rtn(function, file, line, "libcompiler_rt abort");
+}
+
+#else
+
+/* Get the system definition of abort() */
+#include <stdlib.h>
+
+#ifndef _WIN32
+__attribute__((weak))
+__attribute__((visibility("hidden")))
+#endif
+void compilerrt_abort_impl(const char *file, int line, const char *function) {
+  abort();
+}
+
+#endif
diff --git a/lib/builtins/int_util.h b/lib/builtins/int_util.h
new file mode 100644
index 000000000..1348b85eb
--- /dev/null
+++ b/lib/builtins/int_util.h
@@ -0,0 +1,29 @@
+/* ===-- int_util.h - internal utility functions ----------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===-----------------------------------------------------------------------===
+ *
+ * This file is not part of the interface of this library.
+ *
+ * This file defines non-inline utilities which are available for use in the
+ * library. The function definitions themselves are all contained in int_util.c
+ * which will always be compiled into any compiler-rt library.
+ *
+ * ===-----------------------------------------------------------------------===
+ */
+
+#ifndef INT_UTIL_H
+#define INT_UTIL_H
+
+/** \brief Trigger a program abort (or panic for kernel code). */
+#define compilerrt_abort() compilerrt_abort_impl(__FILE__, __LINE__, \
+                                                 __FUNCTION__)
+
+void compilerrt_abort_impl(const char *file, int line,
+                           const char *function) __attribute__((noreturn));
+
+#endif /* INT_UTIL_H */
diff --git a/lib/builtins/lshrdi3.c b/lib/builtins/lshrdi3.c
new file mode 100644
index 000000000..6b1ea923b
--- /dev/null
+++ b/lib/builtins/lshrdi3.c
@@ -0,0 +1,43 @@
+/* ===-- lshrdi3.c - Implement __lshrdi3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __lshrdi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: logical a >> b */
+
+/* Precondition:  0 <= b < bits_in_dword */
+
+ARM_EABI_FNALIAS(llsr, lshrdi3)
+
+COMPILER_RT_ABI di_int
+__lshrdi3(di_int a, si_int b)
+{
+    const int bits_in_word = (int)(sizeof(si_int) * CHAR_BIT);
+    udwords input;
+    udwords result;
+    input.all = a;
+    if (b & bits_in_word)  /* bits_in_word <= b < bits_in_dword */
+    {
+        result.s.high = 0;
+        result.s.low = input.s.high >> (b - bits_in_word);
+    }
+    else  /* 0 <= b < bits_in_word */
+    {
+        if (b == 0)
+            return a;
+        result.s.high  = input.s.high >> b;
+        result.s.low = (input.s.high << (bits_in_word - b)) | (input.s.low >> b);
+    }
+    return result.all;
+}
diff --git a/lib/builtins/lshrti3.c b/lib/builtins/lshrti3.c
new file mode 100644
index 000000000..be768143b
--- /dev/null
+++ b/lib/builtins/lshrti3.c
@@ -0,0 +1,45 @@
+/* ===-- lshrti3.c - Implement __lshrti3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __lshrti3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: logical a >> b */
+
+/* Precondition:  0 <= b < bits_in_tword */
+
+ti_int
+__lshrti3(ti_int a, si_int b)
+{
+    const int bits_in_dword = (int)(sizeof(di_int) * CHAR_BIT);
+    utwords input;
+    utwords result;
+    input.all = a;
+    if (b & bits_in_dword)  /* bits_in_dword <= b < bits_in_tword */
+    {
+        result.s.high = 0;
+        result.s.low = input.s.high >> (b - bits_in_dword);
+    }
+    else  /* 0 <= b < bits_in_dword */
+    {
+        if (b == 0)
+            return a;
+        result.s.high  = input.s.high >> b;
+        result.s.low = (input.s.high << (bits_in_dword - b)) | (input.s.low >> b);
+    }
+    return result.all;
+}
+
+#endif /* __x86_64 */ 
diff --git a/lib/builtins/moddi3.c b/lib/builtins/moddi3.c
new file mode 100644
index 000000000..2f3b9cc4f
--- /dev/null
+++ b/lib/builtins/moddi3.c
@@ -0,0 +1,32 @@
+/*===-- moddi3.c - Implement __moddi3 -------------------------------------===
+ *
+ *                    The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __moddi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+COMPILER_RT_ABI du_int __udivmoddi4(du_int a, du_int b, du_int* rem);
+
+/* Returns: a % b */
+
+COMPILER_RT_ABI di_int
+__moddi3(di_int a, di_int b)
+{
+    const int bits_in_dword_m1 = (int)(sizeof(di_int) * CHAR_BIT) - 1;
+    di_int s = b >> bits_in_dword_m1;  /* s = b < 0 ? -1 : 0 */
+    b = (b ^ s) - s;                   /* negate if s == -1 */
+    s = a >> bits_in_dword_m1;         /* s = a < 0 ? -1 : 0 */
+    a = (a ^ s) - s;                   /* negate if s == -1 */
+    di_int r;
+    __udivmoddi4(a, b, (du_int*)&r);
+    return (r ^ s) - s;                /* negate if s == -1 */
+}
diff --git a/lib/builtins/modsi3.c b/lib/builtins/modsi3.c
new file mode 100644
index 000000000..d16213c49
--- /dev/null
+++ b/lib/builtins/modsi3.c
@@ -0,0 +1,25 @@
+/* ===-- modsi3.c - Implement __modsi3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __modsi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+su_int COMPILER_RT_ABI __divsi3(si_int a, si_int b);
+
+/* Returns: a % b */
+
+COMPILER_RT_ABI si_int
+__modsi3(si_int a, si_int b)
+{
+    return a - __divsi3(a, b) * b;
+}
diff --git a/lib/builtins/modti3.c b/lib/builtins/modti3.c
new file mode 100644
index 000000000..752202d45
--- /dev/null
+++ b/lib/builtins/modti3.c
@@ -0,0 +1,36 @@
+/* ===-- modti3.c - Implement __modti3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __modti3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+tu_int __udivmodti4(tu_int a, tu_int b, tu_int* rem);
+
+/*Returns: a % b */
+
+ti_int
+__modti3(ti_int a, ti_int b)
+{
+    const int bits_in_tword_m1 = (int)(sizeof(ti_int) * CHAR_BIT) - 1;
+    ti_int s = b >> bits_in_tword_m1;  /* s = b < 0 ? -1 : 0 */
+    b = (b ^ s) - s;                   /* negate if s == -1 */
+    s = a >> bits_in_tword_m1;         /* s = a < 0 ? -1 : 0 */
+    a = (a ^ s) - s;                   /* negate if s == -1 */
+    ti_int r;
+    __udivmodti4(a, b, (tu_int*)&r);
+    return (r ^ s) - s;                /* negate if s == -1 */
+}
+
+#endif
diff --git a/lib/builtins/muldc3.c b/lib/builtins/muldc3.c
new file mode 100644
index 000000000..5f4a6d16e
--- /dev/null
+++ b/lib/builtins/muldc3.c
@@ -0,0 +1,73 @@
+/* ===-- muldc3.c - Implement __muldc3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __muldc3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+#include "int_math.h"
+
+/* Returns: the product of a + ib and c + id */
+
+double _Complex
+__muldc3(double __a, double __b, double __c, double __d)
+{
+    double __ac = __a * __c;
+    double __bd = __b * __d;
+    double __ad = __a * __d;
+    double __bc = __b * __c;
+    double _Complex z;
+    __real__ z = __ac - __bd;
+    __imag__ z = __ad + __bc;
+    if (crt_isnan(__real__ z) && crt_isnan(__imag__ z))
+    {
+        int __recalc = 0;
+        if (crt_isinf(__a) || crt_isinf(__b))
+        {
+            __a = crt_copysign(crt_isinf(__a) ? 1 : 0, __a);
+            __b = crt_copysign(crt_isinf(__b) ? 1 : 0, __b);
+            if (crt_isnan(__c))
+                __c = crt_copysign(0, __c);
+            if (crt_isnan(__d))
+                __d = crt_copysign(0, __d);
+            __recalc = 1;
+        }
+        if (crt_isinf(__c) || crt_isinf(__d))
+        {
+            __c = crt_copysign(crt_isinf(__c) ? 1 : 0, __c);
+            __d = crt_copysign(crt_isinf(__d) ? 1 : 0, __d);
+            if (crt_isnan(__a))
+                __a = crt_copysign(0, __a);
+            if (crt_isnan(__b))
+                __b = crt_copysign(0, __b);
+            __recalc = 1;
+        }
+        if (!__recalc && (crt_isinf(__ac) || crt_isinf(__bd) ||
+                          crt_isinf(__ad) || crt_isinf(__bc)))
+        {
+            if (crt_isnan(__a))
+                __a = crt_copysign(0, __a);
+            if (crt_isnan(__b))
+                __b = crt_copysign(0, __b);
+            if (crt_isnan(__c))
+                __c = crt_copysign(0, __c);
+            if (crt_isnan(__d))
+                __d = crt_copysign(0, __d);
+            __recalc = 1;
+        }
+        if (__recalc)
+        {
+            __real__ z = CRT_INFINITY * (__a * __c - __b * __d);
+            __imag__ z = CRT_INFINITY * (__a * __d + __b * __c);
+        }
+    }
+    return z;
+}
diff --git a/lib/builtins/muldf3.c b/lib/builtins/muldf3.c
new file mode 100644
index 000000000..c38edba90
--- /dev/null
+++ b/lib/builtins/muldf3.c
@@ -0,0 +1,122 @@
+//===-- lib/muldf3.c - Double-precision multiplication ------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements double-precision soft-float multiplication
+// with the IEEE-754 default rounding (to nearest, ties to even).
+//
+//===----------------------------------------------------------------------===//
+
+#define DOUBLE_PRECISION
+#include "fp_lib.h"
+
+ARM_EABI_FNALIAS(dmul, muldf3)
+
+COMPILER_RT_ABI fp_t
+__muldf3(fp_t a, fp_t b) {
+    
+    const unsigned int aExponent = toRep(a) >> significandBits & maxExponent;
+    const unsigned int bExponent = toRep(b) >> significandBits & maxExponent;
+    const rep_t productSign = (toRep(a) ^ toRep(b)) & signBit;
+    
+    rep_t aSignificand = toRep(a) & significandMask;
+    rep_t bSignificand = toRep(b) & significandMask;
+    int scale = 0;
+    
+    // Detect if a or b is zero, denormal, infinity, or NaN.
+    if (aExponent-1U >= maxExponent-1U || bExponent-1U >= maxExponent-1U) {
+        
+        const rep_t aAbs = toRep(a) & absMask;
+        const rep_t bAbs = toRep(b) & absMask;
+        
+        // NaN * anything = qNaN
+        if (aAbs > infRep) return fromRep(toRep(a) | quietBit);
+        // anything * NaN = qNaN
+        if (bAbs > infRep) return fromRep(toRep(b) | quietBit);
+        
+        if (aAbs == infRep) {
+            // infinity * non-zero = +/- infinity
+            if (bAbs) return fromRep(aAbs | productSign);
+            // infinity * zero = NaN
+            else return fromRep(qnanRep);
+        }
+        
+        if (bAbs == infRep) {
+            // non-zero * infinity = +/- infinity
+            if (aAbs) return fromRep(bAbs | productSign);
+            // zero * infinity = NaN
+            else return fromRep(qnanRep);
+        }
+        
+        // zero * anything = +/- zero
+        if (!aAbs) return fromRep(productSign);
+        // anything * zero = +/- zero
+        if (!bAbs) return fromRep(productSign);
+        
+        // one or both of a or b is denormal, the other (if applicable) is a
+        // normal number.  Renormalize one or both of a and b, and set scale to
+        // include the necessary exponent adjustment.
+        if (aAbs < implicitBit) scale += normalize(&aSignificand);
+        if (bAbs < implicitBit) scale += normalize(&bSignificand);
+    }
+    
+    // Or in the implicit significand bit.  (If we fell through from the
+    // denormal path it was already set by normalize( ), but setting it twice
+    // won't hurt anything.)
+    aSignificand |= implicitBit;
+    bSignificand |= implicitBit;
+    
+    // Get the significand of a*b.  Before multiplying the significands, shift
+    // one of them left to left-align it in the field.  Thus, the product will
+    // have (exponentBits + 2) integral digits, all but two of which must be
+    // zero.  Normalizing this result is just a conditional left-shift by one
+    // and bumping the exponent accordingly.
+    rep_t productHi, productLo;
+    wideMultiply(aSignificand, bSignificand << exponentBits,
+                 &productHi, &productLo);
+    
+    int productExponent = aExponent + bExponent - exponentBias + scale;
+    
+    // Normalize the significand, adjust exponent if needed.
+    if (productHi & implicitBit) productExponent++;
+    else wideLeftShift(&productHi, &productLo, 1);
+    
+    // If we have overflowed the type, return +/- infinity.
+    if (productExponent >= maxExponent) return fromRep(infRep | productSign);
+    
+    if (productExponent <= 0) {
+        // Result is denormal before rounding
+        //
+        // If the result is so small that it just underflows to zero, return
+        // a zero of the appropriate sign.  Mathematically there is no need to
+        // handle this case separately, but we make it a special case to
+        // simplify the shift logic.
+        const unsigned int shift = 1U - (unsigned int)productExponent;
+        if (shift >= typeWidth) return fromRep(productSign);
+        
+        // Otherwise, shift the significand of the result so that the round
+        // bit is the high bit of productLo.
+        wideRightShiftWithSticky(&productHi, &productLo, shift);
+    }
+    
+    else {
+        // Result is normal before rounding; insert the exponent.
+        productHi &= significandMask;
+        productHi |= (rep_t)productExponent << significandBits;
+    }
+    
+    // Insert the sign of the result:
+    productHi |= productSign;
+    
+    // Final rounding.  The final result may overflow to infinity, or underflow
+    // to zero, but those are the correct results in those cases.  We use the
+    // default IEEE-754 round-to-nearest, ties-to-even rounding mode.
+    if (productLo > signBit) productHi++;
+    if (productLo == signBit) productHi += productHi & 1;
+    return fromRep(productHi);
+}
diff --git a/lib/builtins/muldi3.c b/lib/builtins/muldi3.c
new file mode 100644
index 000000000..2dae44c11
--- /dev/null
+++ b/lib/builtins/muldi3.c
@@ -0,0 +1,56 @@
+/* ===-- muldi3.c - Implement __muldi3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __muldi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a * b */
+
+static
+di_int
+__muldsi3(su_int a, su_int b)
+{
+    dwords r;
+    const int bits_in_word_2 = (int)(sizeof(si_int) * CHAR_BIT) / 2;
+    const su_int lower_mask = (su_int)~0 >> bits_in_word_2;
+    r.s.low = (a & lower_mask) * (b & lower_mask);
+    su_int t = r.s.low >> bits_in_word_2;
+    r.s.low &= lower_mask;
+    t += (a >> bits_in_word_2) * (b & lower_mask);
+    r.s.low += (t & lower_mask) << bits_in_word_2;
+    r.s.high = t >> bits_in_word_2;
+    t = r.s.low >> bits_in_word_2;
+    r.s.low &= lower_mask;
+    t += (b >> bits_in_word_2) * (a & lower_mask);
+    r.s.low += (t & lower_mask) << bits_in_word_2;
+    r.s.high += t >> bits_in_word_2;
+    r.s.high += (a >> bits_in_word_2) * (b >> bits_in_word_2);
+    return r.all;
+}
+
+/* Returns: a * b */
+
+ARM_EABI_FNALIAS(lmul, muldi3)
+
+COMPILER_RT_ABI di_int
+__muldi3(di_int a, di_int b)
+{
+    dwords x;
+    x.all = a;
+    dwords y;
+    y.all = b;
+    dwords r;
+    r.all = __muldsi3(x.s.low, y.s.low);
+    r.s.high += x.s.high * y.s.low + x.s.low * y.s.high;
+    return r.all;
+}
diff --git a/lib/builtins/mulodi4.c b/lib/builtins/mulodi4.c
new file mode 100644
index 000000000..0c1b5cdae
--- /dev/null
+++ b/lib/builtins/mulodi4.c
@@ -0,0 +1,58 @@
+/*===-- mulodi4.c - Implement __mulodi4 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __mulodi4 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a * b */
+
+/* Effects: sets *overflow to 1  if a * b overflows */
+
+di_int
+__mulodi4(di_int a, di_int b, int* overflow)
+{
+    const int N = (int)(sizeof(di_int) * CHAR_BIT);
+    const di_int MIN = (di_int)1 << (N-1);
+    const di_int MAX = ~MIN;
+    *overflow = 0; 
+    di_int result = a * b;
+    if (a == MIN)
+    {
+        if (b != 0 && b != 1)
+	    *overflow = 1;
+	return result;
+    }
+    if (b == MIN)
+    {
+        if (a != 0 && a != 1)
+	    *overflow = 1;
+        return result;
+    }
+    di_int sa = a >> (N - 1);
+    di_int abs_a = (a ^ sa) - sa;
+    di_int sb = b >> (N - 1);
+    di_int abs_b = (b ^ sb) - sb;
+    if (abs_a < 2 || abs_b < 2)
+        return result;
+    if (sa == sb)
+    {
+        if (abs_a > MAX / abs_b)
+            *overflow = 1;
+    }
+    else
+    {
+        if (abs_a > MIN / -abs_b)
+            *overflow = 1;
+    }
+    return result;
+}
diff --git a/lib/builtins/mulosi4.c b/lib/builtins/mulosi4.c
new file mode 100644
index 000000000..f3398d1fc
--- /dev/null
+++ b/lib/builtins/mulosi4.c
@@ -0,0 +1,58 @@
+/*===-- mulosi4.c - Implement __mulosi4 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __mulosi4 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a * b */
+
+/* Effects: sets *overflow to 1  if a * b overflows */
+
+si_int
+__mulosi4(si_int a, si_int b, int* overflow)
+{
+    const int N = (int)(sizeof(si_int) * CHAR_BIT);
+    const si_int MIN = (si_int)1 << (N-1);
+    const si_int MAX = ~MIN;
+    *overflow = 0; 
+    si_int result = a * b;
+    if (a == MIN)
+    {
+        if (b != 0 && b != 1)
+	    *overflow = 1;
+	return result;
+    }
+    if (b == MIN)
+    {
+        if (a != 0 && a != 1)
+	    *overflow = 1;
+        return result;
+    }
+    si_int sa = a >> (N - 1);
+    si_int abs_a = (a ^ sa) - sa;
+    si_int sb = b >> (N - 1);
+    si_int abs_b = (b ^ sb) - sb;
+    if (abs_a < 2 || abs_b < 2)
+        return result;
+    if (sa == sb)
+    {
+        if (abs_a > MAX / abs_b)
+            *overflow = 1;
+    }
+    else
+    {
+        if (abs_a > MIN / -abs_b)
+            *overflow = 1;
+    }
+    return result;
+}
diff --git a/lib/builtins/muloti4.c b/lib/builtins/muloti4.c
new file mode 100644
index 000000000..f58dd074e
--- /dev/null
+++ b/lib/builtins/muloti4.c
@@ -0,0 +1,62 @@
+/*===-- muloti4.c - Implement __muloti4 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __muloti4 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: a * b */
+
+/* Effects: sets *overflow to 1  if a * b overflows */
+
+ti_int
+__muloti4(ti_int a, ti_int b, int* overflow)
+{
+    const int N = (int)(sizeof(ti_int) * CHAR_BIT);
+    const ti_int MIN = (ti_int)1 << (N-1);
+    const ti_int MAX = ~MIN;
+    *overflow = 0;
+    ti_int result = a * b;
+    if (a == MIN)
+    {
+        if (b != 0 && b != 1)
+	    *overflow = 1;
+	return result;
+    }
+    if (b == MIN)
+    {
+        if (a != 0 && a != 1)
+	    *overflow = 1;
+        return result;
+    }
+    ti_int sa = a >> (N - 1);
+    ti_int abs_a = (a ^ sa) - sa;
+    ti_int sb = b >> (N - 1);
+    ti_int abs_b = (b ^ sb) - sb;
+    if (abs_a < 2 || abs_b < 2)
+        return result;
+    if (sa == sb)
+    {
+        if (abs_a > MAX / abs_b)
+            *overflow = 1;
+    }
+    else
+    {
+        if (abs_a > MIN / -abs_b)
+            *overflow = 1;
+    }
+    return result;
+}
+
+#endif
diff --git a/lib/builtins/mulsc3.c b/lib/builtins/mulsc3.c
new file mode 100644
index 000000000..6d433fbc4
--- /dev/null
+++ b/lib/builtins/mulsc3.c
@@ -0,0 +1,73 @@
+/* ===-- mulsc3.c - Implement __mulsc3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __mulsc3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+#include "int_math.h"
+
+/* Returns: the product of a + ib and c + id */
+
+float _Complex
+__mulsc3(float __a, float __b, float __c, float __d)
+{
+    float __ac = __a * __c;
+    float __bd = __b * __d;
+    float __ad = __a * __d;
+    float __bc = __b * __c;
+    float _Complex z;
+    __real__ z = __ac - __bd;
+    __imag__ z = __ad + __bc;
+    if (crt_isnan(__real__ z) && crt_isnan(__imag__ z))
+    {
+        int __recalc = 0;
+        if (crt_isinf(__a) || crt_isinf(__b))
+        {
+            __a = crt_copysignf(crt_isinf(__a) ? 1 : 0, __a);
+            __b = crt_copysignf(crt_isinf(__b) ? 1 : 0, __b);
+            if (crt_isnan(__c))
+                __c = crt_copysignf(0, __c);
+            if (crt_isnan(__d))
+                __d = crt_copysignf(0, __d);
+            __recalc = 1;
+        }
+        if (crt_isinf(__c) || crt_isinf(__d))
+        {
+            __c = crt_copysignf(crt_isinf(__c) ? 1 : 0, __c);
+            __d = crt_copysignf(crt_isinf(__d) ? 1 : 0, __d);
+            if (crt_isnan(__a))
+                __a = crt_copysignf(0, __a);
+            if (crt_isnan(__b))
+                __b = crt_copysignf(0, __b);
+            __recalc = 1;
+        }
+        if (!__recalc && (crt_isinf(__ac) || crt_isinf(__bd) ||
+                          crt_isinf(__ad) || crt_isinf(__bc)))
+        {
+            if (crt_isnan(__a))
+                __a = crt_copysignf(0, __a);
+            if (crt_isnan(__b))
+                __b = crt_copysignf(0, __b);
+            if (crt_isnan(__c))
+                __c = crt_copysignf(0, __c);
+            if (crt_isnan(__d))
+                __d = crt_copysignf(0, __d);
+            __recalc = 1;
+        }
+        if (__recalc)
+        {
+            __real__ z = CRT_INFINITY * (__a * __c - __b * __d);
+            __imag__ z = CRT_INFINITY * (__a * __d + __b * __c);
+        }
+    }
+    return z;
+}
diff --git a/lib/builtins/mulsf3.c b/lib/builtins/mulsf3.c
new file mode 100644
index 000000000..861a9ba5f
--- /dev/null
+++ b/lib/builtins/mulsf3.c
@@ -0,0 +1,112 @@
+//===-- lib/mulsf3.c - Single-precision multiplication ------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements single-precision soft-float multiplication
+// with the IEEE-754 default rounding (to nearest, ties to even).
+//
+//===----------------------------------------------------------------------===//
+
+#define SINGLE_PRECISION
+#include "fp_lib.h"
+
+ARM_EABI_FNALIAS(fmul, mulsf3)
+
+COMPILER_RT_ABI fp_t
+__mulsf3(fp_t a, fp_t b) {
+    
+    const unsigned int aExponent = toRep(a) >> significandBits & maxExponent;
+    const unsigned int bExponent = toRep(b) >> significandBits & maxExponent;
+    const rep_t productSign = (toRep(a) ^ toRep(b)) & signBit;
+    
+    rep_t aSignificand = toRep(a) & significandMask;
+    rep_t bSignificand = toRep(b) & significandMask;
+    int scale = 0;
+    
+    // Detect if a or b is zero, denormal, infinity, or NaN.
+    if (aExponent-1U >= maxExponent-1U || bExponent-1U >= maxExponent-1U) {
+        
+        const rep_t aAbs = toRep(a) & absMask;
+        const rep_t bAbs = toRep(b) & absMask;
+        
+        // NaN * anything = qNaN
+        if (aAbs > infRep) return fromRep(toRep(a) | quietBit);
+        // anything * NaN = qNaN
+        if (bAbs > infRep) return fromRep(toRep(b) | quietBit);
+        
+        if (aAbs == infRep) {
+            // infinity * non-zero = +/- infinity
+            if (bAbs) return fromRep(aAbs | productSign);
+            // infinity * zero = NaN
+            else return fromRep(qnanRep);
+        }
+        
+        if (bAbs == infRep) {
+            // non-zero * infinity = +/- infinity
+            if (aAbs) return fromRep(bAbs | productSign);
+            // zero * infinity = NaN
+            else return fromRep(qnanRep);
+        }
+        
+        // zero * anything = +/- zero
+        if (!aAbs) return fromRep(productSign);
+        // anything * zero = +/- zero
+        if (!bAbs) return fromRep(productSign);
+        
+        // one or both of a or b is denormal, the other (if applicable) is a
+        // normal number.  Renormalize one or both of a and b, and set scale to
+        // include the necessary exponent adjustment.
+        if (aAbs < implicitBit) scale += normalize(&aSignificand);
+        if (bAbs < implicitBit) scale += normalize(&bSignificand);
+    }
+    
+    // Or in the implicit significand bit.  (If we fell through from the
+    // denormal path it was already set by normalize( ), but setting it twice
+    // won't hurt anything.)
+    aSignificand |= implicitBit;
+    bSignificand |= implicitBit;
+    
+    // Get the significand of a*b.  Before multiplying the significands, shift
+    // one of them left to left-align it in the field.  Thus, the product will
+    // have (exponentBits + 2) integral digits, all but two of which must be
+    // zero.  Normalizing this result is just a conditional left-shift by one
+    // and bumping the exponent accordingly.
+    rep_t productHi, productLo;
+    wideMultiply(aSignificand, bSignificand << exponentBits,
+                 &productHi, &productLo);
+    
+    int productExponent = aExponent + bExponent - exponentBias + scale;
+    
+    // Normalize the significand, adjust exponent if needed.
+    if (productHi & implicitBit) productExponent++;
+    else wideLeftShift(&productHi, &productLo, 1);
+    
+    // If we have overflowed the type, return +/- infinity.
+    if (productExponent >= maxExponent) return fromRep(infRep | productSign);
+    
+    if (productExponent <= 0) {
+        // Result is denormal before rounding, the exponent is zero and we
+        // need to shift the significand.
+        wideRightShiftWithSticky(&productHi, &productLo, 1U - (unsigned)productExponent);
+    }
+    
+    else {
+        // Result is normal before rounding; insert the exponent.
+        productHi &= significandMask;
+        productHi |= (rep_t)productExponent << significandBits;
+    }
+    
+    // Insert the sign of the result:
+    productHi |= productSign;
+    
+    // Final rounding.  The final result may overflow to infinity, or underflow
+    // to zero, but those are the correct results in those cases.
+    if (productLo > signBit) productHi++;
+    if (productLo == signBit) productHi += productHi & 1;
+    return fromRep(productHi);
+}
diff --git a/lib/builtins/multi3.c b/lib/builtins/multi3.c
new file mode 100644
index 000000000..0b8730f08
--- /dev/null
+++ b/lib/builtins/multi3.c
@@ -0,0 +1,58 @@
+/* ===-- multi3.c - Implement __multi3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+
+ * This file implements __multi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: a * b */
+
+static
+ti_int
+__mulddi3(du_int a, du_int b)
+{
+    twords r;
+    const int bits_in_dword_2 = (int)(sizeof(di_int) * CHAR_BIT) / 2;
+    const du_int lower_mask = (du_int)~0 >> bits_in_dword_2;
+    r.s.low = (a & lower_mask) * (b & lower_mask);
+    du_int t = r.s.low >> bits_in_dword_2;
+    r.s.low &= lower_mask;
+    t += (a >> bits_in_dword_2) * (b & lower_mask);
+    r.s.low += (t & lower_mask) << bits_in_dword_2;
+    r.s.high = t >> bits_in_dword_2;
+    t = r.s.low >> bits_in_dword_2;
+    r.s.low &= lower_mask;
+    t += (b >> bits_in_dword_2) * (a & lower_mask);
+    r.s.low += (t & lower_mask) << bits_in_dword_2;
+    r.s.high += t >> bits_in_dword_2;
+    r.s.high += (a >> bits_in_dword_2) * (b >> bits_in_dword_2);
+    return r.all;
+}
+
+/* Returns: a * b */
+
+ti_int
+__multi3(ti_int a, ti_int b)
+{
+    twords x;
+    x.all = a;
+    twords y;
+    y.all = b;
+    twords r;
+    r.all = __mulddi3(x.s.low, y.s.low);
+    r.s.high += x.s.high * y.s.low + x.s.low * y.s.high;
+    return r.all;
+}
+
+#endif /* __x86_64 */
diff --git a/lib/builtins/mulvdi3.c b/lib/builtins/mulvdi3.c
new file mode 100644
index 000000000..bcc8e659b
--- /dev/null
+++ b/lib/builtins/mulvdi3.c
@@ -0,0 +1,56 @@
+/*===-- mulvdi3.c - Implement __mulvdi3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __mulvdi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a * b */
+
+/* Effects: aborts if a * b overflows */
+
+di_int
+__mulvdi3(di_int a, di_int b)
+{
+    const int N = (int)(sizeof(di_int) * CHAR_BIT);
+    const di_int MIN = (di_int)1 << (N-1);
+    const di_int MAX = ~MIN;
+    if (a == MIN)
+    {
+        if (b == 0 || b == 1)
+            return a * b;
+        compilerrt_abort();
+    }
+    if (b == MIN)
+    {
+        if (a == 0 || a == 1)
+            return a * b;
+        compilerrt_abort();
+    }
+    di_int sa = a >> (N - 1);
+    di_int abs_a = (a ^ sa) - sa;
+    di_int sb = b >> (N - 1);
+    di_int abs_b = (b ^ sb) - sb;
+    if (abs_a < 2 || abs_b < 2)
+        return a * b;
+    if (sa == sb)
+    {
+        if (abs_a > MAX / abs_b)
+            compilerrt_abort();
+    }
+    else
+    {
+        if (abs_a > MIN / -abs_b)
+            compilerrt_abort();
+    }
+    return a * b;
+}
diff --git a/lib/builtins/mulvsi3.c b/lib/builtins/mulvsi3.c
new file mode 100644
index 000000000..d372b2017
--- /dev/null
+++ b/lib/builtins/mulvsi3.c
@@ -0,0 +1,56 @@
+/* ===-- mulvsi3.c - Implement __mulvsi3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __mulvsi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a * b */
+
+/* Effects: aborts if a * b overflows */
+
+si_int
+__mulvsi3(si_int a, si_int b)
+{
+    const int N = (int)(sizeof(si_int) * CHAR_BIT);
+    const si_int MIN = (si_int)1 << (N-1);
+    const si_int MAX = ~MIN;
+    if (a == MIN)
+    {
+        if (b == 0 || b == 1)
+            return a * b;
+        compilerrt_abort();
+    }
+    if (b == MIN)
+    {
+        if (a == 0 || a == 1)
+            return a * b;
+        compilerrt_abort();
+    }
+    si_int sa = a >> (N - 1);
+    si_int abs_a = (a ^ sa) - sa;
+    si_int sb = b >> (N - 1);
+    si_int abs_b = (b ^ sb) - sb;
+    if (abs_a < 2 || abs_b < 2)
+        return a * b;
+    if (sa == sb)
+    {
+        if (abs_a > MAX / abs_b)
+            compilerrt_abort();
+    }
+    else
+    {
+        if (abs_a > MIN / -abs_b)
+            compilerrt_abort();
+    }
+    return a * b;
+}
diff --git a/lib/builtins/mulvti3.c b/lib/builtins/mulvti3.c
new file mode 100644
index 000000000..31f7d2fd3
--- /dev/null
+++ b/lib/builtins/mulvti3.c
@@ -0,0 +1,60 @@
+/* ===-- mulvti3.c - Implement __mulvti3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __mulvti3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: a * b */
+
+/* Effects: aborts if a * b overflows */
+
+ti_int
+__mulvti3(ti_int a, ti_int b)
+{
+    const int N = (int)(sizeof(ti_int) * CHAR_BIT);
+    const ti_int MIN = (ti_int)1 << (N-1);
+    const ti_int MAX = ~MIN;
+    if (a == MIN)
+    {
+        if (b == 0 || b == 1)
+            return a * b;
+        compilerrt_abort();
+    }
+    if (b == MIN)
+    {
+        if (a == 0 || a == 1)
+            return a * b;
+        compilerrt_abort();
+    }
+    ti_int sa = a >> (N - 1);
+    ti_int abs_a = (a ^ sa) - sa;
+    ti_int sb = b >> (N - 1);
+    ti_int abs_b = (b ^ sb) - sb;
+    if (abs_a < 2 || abs_b < 2)
+        return a * b;
+    if (sa == sb)
+    {
+        if (abs_a > MAX / abs_b)
+            compilerrt_abort();
+    }
+    else
+    {
+        if (abs_a > MIN / -abs_b)
+            compilerrt_abort();
+    }
+    return a * b;
+}
+
+#endif
diff --git a/lib/builtins/mulxc3.c b/lib/builtins/mulxc3.c
new file mode 100644
index 000000000..cec057368
--- /dev/null
+++ b/lib/builtins/mulxc3.c
@@ -0,0 +1,77 @@
+/* ===-- mulxc3.c - Implement __mulxc3 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __mulxc3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#if !_ARCH_PPC
+
+#include "int_lib.h"
+#include "int_math.h"
+
+/* Returns: the product of a + ib and c + id */
+
+long double _Complex
+__mulxc3(long double __a, long double __b, long double __c, long double __d)
+{
+    long double __ac = __a * __c;
+    long double __bd = __b * __d;
+    long double __ad = __a * __d;
+    long double __bc = __b * __c;
+    long double _Complex z;
+    __real__ z = __ac - __bd;
+    __imag__ z = __ad + __bc;
+    if (crt_isnan(__real__ z) && crt_isnan(__imag__ z))
+    {
+        int __recalc = 0;
+        if (crt_isinf(__a) || crt_isinf(__b))
+        {
+            __a = crt_copysignl(crt_isinf(__a) ? 1 : 0, __a);
+            __b = crt_copysignl(crt_isinf(__b) ? 1 : 0, __b);
+            if (crt_isnan(__c))
+                __c = crt_copysignl(0, __c);
+            if (crt_isnan(__d))
+                __d = crt_copysignl(0, __d);
+            __recalc = 1;
+        }
+        if (crt_isinf(__c) || crt_isinf(__d))
+        {
+            __c = crt_copysignl(crt_isinf(__c) ? 1 : 0, __c);
+            __d = crt_copysignl(crt_isinf(__d) ? 1 : 0, __d);
+            if (crt_isnan(__a))
+                __a = crt_copysignl(0, __a);
+            if (crt_isnan(__b))
+                __b = crt_copysignl(0, __b);
+            __recalc = 1;
+        }
+        if (!__recalc && (crt_isinf(__ac) || crt_isinf(__bd) ||
+                          crt_isinf(__ad) || crt_isinf(__bc)))
+        {
+            if (crt_isnan(__a))
+                __a = crt_copysignl(0, __a);
+            if (crt_isnan(__b))
+                __b = crt_copysignl(0, __b);
+            if (crt_isnan(__c))
+                __c = crt_copysignl(0, __c);
+            if (crt_isnan(__d))
+                __d = crt_copysignl(0, __d);
+            __recalc = 1;
+        }
+        if (__recalc)
+        {
+            __real__ z = CRT_INFINITY * (__a * __c - __b * __d);
+            __imag__ z = CRT_INFINITY * (__a * __d + __b * __c);
+        }
+    }
+    return z;
+}
+
+#endif
diff --git a/lib/builtins/negdf2.c b/lib/builtins/negdf2.c
new file mode 100644
index 000000000..4e17513a9
--- /dev/null
+++ b/lib/builtins/negdf2.c
@@ -0,0 +1,21 @@
+//===-- lib/negdf2.c - double-precision negation ------------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements double-precision soft-float negation.
+//
+//===----------------------------------------------------------------------===//
+
+#define DOUBLE_PRECISION
+#include "fp_lib.h"
+
+ARM_EABI_FNALIAS(dneg, negdf2)
+
+fp_t __negdf2(fp_t a) {
+    return fromRep(toRep(a) ^ signBit);
+}
diff --git a/lib/builtins/negdi2.c b/lib/builtins/negdi2.c
new file mode 100644
index 000000000..b000dda3b
--- /dev/null
+++ b/lib/builtins/negdi2.c
@@ -0,0 +1,26 @@
+/* ===-- negdi2.c - Implement __negdi2 -------------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __negdi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: -a */
+
+di_int
+__negdi2(di_int a)
+{
+    /* Note: this routine is here for API compatibility; any sane compiler
+     * should expand it inline.
+     */
+    return -a;
+}
diff --git a/lib/builtins/negsf2.c b/lib/builtins/negsf2.c
new file mode 100644
index 000000000..29c17be41
--- /dev/null
+++ b/lib/builtins/negsf2.c
@@ -0,0 +1,22 @@
+//===-- lib/negsf2.c - single-precision negation ------------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements single-precision soft-float negation.
+//
+//===----------------------------------------------------------------------===//
+
+#define SINGLE_PRECISION
+#include "fp_lib.h"
+
+ARM_EABI_FNALIAS(fneg, negsf2)
+
+COMPILER_RT_ABI fp_t
+__negsf2(fp_t a) {
+    return fromRep(toRep(a) ^ signBit);
+}
diff --git a/lib/builtins/negti2.c b/lib/builtins/negti2.c
new file mode 100644
index 000000000..f7e4ad3b9
--- /dev/null
+++ b/lib/builtins/negti2.c
@@ -0,0 +1,30 @@
+/* ===-- negti2.c - Implement __negti2 -------------------------------------===
+ *
+ *      	       The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __negti2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: -a */
+
+ti_int
+__negti2(ti_int a)
+{
+    /* Note: this routine is here for API compatibility; any sane compiler
+     * should expand it inline.
+     */
+    return -a;
+}
+
+#endif
diff --git a/lib/builtins/negvdi2.c b/lib/builtins/negvdi2.c
new file mode 100644
index 000000000..e336ecf28
--- /dev/null
+++ b/lib/builtins/negvdi2.c
@@ -0,0 +1,28 @@
+/* ===-- negvdi2.c - Implement __negvdi2 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __negvdi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: -a */
+
+/* Effects: aborts if -a overflows */
+
+COMPILER_RT_ABI di_int
+__negvdi2(di_int a)
+{
+    const di_int MIN = (di_int)1 << ((int)(sizeof(di_int) * CHAR_BIT)-1);
+    if (a == MIN)
+        compilerrt_abort();
+    return -a;
+}
diff --git a/lib/builtins/negvsi2.c b/lib/builtins/negvsi2.c
new file mode 100644
index 000000000..b9e93fef0
--- /dev/null
+++ b/lib/builtins/negvsi2.c
@@ -0,0 +1,28 @@
+/* ===-- negvsi2.c - Implement __negvsi2 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __negvsi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: -a */
+
+/* Effects: aborts if -a overflows */
+
+COMPILER_RT_ABI si_int
+__negvsi2(si_int a)
+{
+    const si_int MIN = (si_int)1 << ((int)(sizeof(si_int) * CHAR_BIT)-1);
+    if (a == MIN)
+        compilerrt_abort();
+    return -a;
+}
diff --git a/lib/builtins/negvti2.c b/lib/builtins/negvti2.c
new file mode 100644
index 000000000..05df6152f
--- /dev/null
+++ b/lib/builtins/negvti2.c
@@ -0,0 +1,32 @@
+/*===-- negvti2.c - Implement __negvti2 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ *===----------------------------------------------------------------------===
+ *
+ *This file implements __negvti2 for the compiler_rt library.
+ *
+ *===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: -a */
+
+/* Effects: aborts if -a overflows */
+
+ti_int
+__negvti2(ti_int a)
+{
+    const ti_int MIN = (ti_int)1 << ((int)(sizeof(ti_int) * CHAR_BIT)-1);
+    if (a == MIN)
+        compilerrt_abort();
+    return -a;
+}
+
+#endif
diff --git a/lib/builtins/paritydi2.c b/lib/builtins/paritydi2.c
new file mode 100644
index 000000000..2ded54c90
--- /dev/null
+++ b/lib/builtins/paritydi2.c
@@ -0,0 +1,27 @@
+/* ===-- paritydi2.c - Implement __paritydi2 -------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __paritydi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: 1 if number of bits is odd else returns 0 */
+
+si_int COMPILER_RT_ABI __paritysi2(si_int a);
+
+COMPILER_RT_ABI si_int
+__paritydi2(di_int a)
+{
+    dwords x;
+    x.all = a;
+    return __paritysi2(x.s.high ^ x.s.low);
+}
diff --git a/lib/builtins/paritysi2.c b/lib/builtins/paritysi2.c
new file mode 100644
index 000000000..599984663
--- /dev/null
+++ b/lib/builtins/paritysi2.c
@@ -0,0 +1,27 @@
+/* ===-- paritysi2.c - Implement __paritysi2 -------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __paritysi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: 1 if number of bits is odd else returns 0 */
+
+COMPILER_RT_ABI si_int
+__paritysi2(si_int a)
+{
+    su_int x = (su_int)a;
+    x ^= x >> 16;
+    x ^= x >> 8;
+    x ^= x >> 4;
+    return (0x6996 >> (x & 0xF)) & 1;
+}
diff --git a/lib/builtins/parityti2.c b/lib/builtins/parityti2.c
new file mode 100644
index 000000000..a1f47b1d9
--- /dev/null
+++ b/lib/builtins/parityti2.c
@@ -0,0 +1,31 @@
+/* ===-- parityti2.c - Implement __parityti2 -------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __parityti2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */ 
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: 1 if number of bits is odd else returns 0 */
+
+si_int __paritydi2(di_int a);
+
+si_int
+__parityti2(ti_int a)
+{
+    twords x;
+    x.all = a;
+    return __paritydi2(x.s.high ^ x.s.low);
+}
+
+#endif
diff --git a/lib/builtins/popcountdi2.c b/lib/builtins/popcountdi2.c
new file mode 100644
index 000000000..5e8a62f07
--- /dev/null
+++ b/lib/builtins/popcountdi2.c
@@ -0,0 +1,36 @@
+/* ===-- popcountdi2.c - Implement __popcountdi2 ----------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __popcountdi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: count of 1 bits */
+
+COMPILER_RT_ABI si_int
+__popcountdi2(di_int a)
+{
+    du_int x2 = (du_int)a;
+    x2 = x2 - ((x2 >> 1) & 0x5555555555555555uLL);
+    /* Every 2 bits holds the sum of every pair of bits (32) */
+    x2 = ((x2 >> 2) & 0x3333333333333333uLL) + (x2 & 0x3333333333333333uLL);
+    /* Every 4 bits holds the sum of every 4-set of bits (3 significant bits) (16) */
+    x2 = (x2 + (x2 >> 4)) & 0x0F0F0F0F0F0F0F0FuLL;
+    /* Every 8 bits holds the sum of every 8-set of bits (4 significant bits) (8) */
+    su_int x = (su_int)(x2 + (x2 >> 32));
+    /* The lower 32 bits hold four 16 bit sums (5 significant bits). */
+    /*   Upper 32 bits are garbage */
+    x = x + (x >> 16);
+    /* The lower 16 bits hold two 32 bit sums (6 significant bits). */
+    /*   Upper 16 bits are garbage */
+    return (x + (x >> 8)) & 0x0000007F;  /* (7 significant bits) */
+}
diff --git a/lib/builtins/popcountsi2.c b/lib/builtins/popcountsi2.c
new file mode 100644
index 000000000..44544ff49
--- /dev/null
+++ b/lib/builtins/popcountsi2.c
@@ -0,0 +1,33 @@
+/* ===-- popcountsi2.c - Implement __popcountsi2 ---------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __popcountsi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: count of 1 bits */
+
+COMPILER_RT_ABI si_int
+__popcountsi2(si_int a)
+{
+    su_int x = (su_int)a;
+    x = x - ((x >> 1) & 0x55555555);
+    /* Every 2 bits holds the sum of every pair of bits */
+    x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
+    /* Every 4 bits holds the sum of every 4-set of bits (3 significant bits) */
+    x = (x + (x >> 4)) & 0x0F0F0F0F;
+    /* Every 8 bits holds the sum of every 8-set of bits (4 significant bits) */
+    x = (x + (x >> 16));
+    /* The lower 16 bits hold two 8 bit sums (5 significant bits).*/
+    /*    Upper 16 bits are garbage */
+    return (x + (x >> 8)) & 0x0000003F;  /* (6 significant bits) */
+}
diff --git a/lib/builtins/popcountti2.c b/lib/builtins/popcountti2.c
new file mode 100644
index 000000000..95666738f
--- /dev/null
+++ b/lib/builtins/popcountti2.c
@@ -0,0 +1,44 @@
+/* ===-- popcountti2.c - Implement __popcountti2 ----------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __popcountti2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: count of 1 bits */
+
+si_int
+__popcountti2(ti_int a)
+{
+    tu_int x3 = (tu_int)a;
+    x3 = x3 - ((x3 >> 1) & (((tu_int)0x5555555555555555uLL << 64) |
+                                     0x5555555555555555uLL));
+    /* Every 2 bits holds the sum of every pair of bits (64) */
+    x3 = ((x3 >> 2) & (((tu_int)0x3333333333333333uLL << 64) | 0x3333333333333333uLL))
+       + (x3 & (((tu_int)0x3333333333333333uLL << 64) | 0x3333333333333333uLL));
+    /* Every 4 bits holds the sum of every 4-set of bits (3 significant bits) (32) */
+    x3 = (x3 + (x3 >> 4))
+       & (((tu_int)0x0F0F0F0F0F0F0F0FuLL << 64) | 0x0F0F0F0F0F0F0F0FuLL);
+    /* Every 8 bits holds the sum of every 8-set of bits (4 significant bits) (16) */
+    du_int x2 = (du_int)(x3 + (x3 >> 64));
+    /* Every 8 bits holds the sum of every 8-set of bits (5 significant bits) (8) */
+    su_int x = (su_int)(x2 + (x2 >> 32));
+    /* Every 8 bits holds the sum of every 8-set of bits (6 significant bits) (4) */
+    x = x + (x >> 16);
+    /* Every 8 bits holds the sum of every 8-set of bits (7 significant bits) (2) */
+    /* Upper 16 bits are garbage */
+    return (x + (x >> 8)) & 0xFF;  /* (8 significant bits) */
+}
+
+#endif
diff --git a/lib/builtins/powidf2.c b/lib/builtins/powidf2.c
new file mode 100644
index 000000000..ac13b172b
--- /dev/null
+++ b/lib/builtins/powidf2.c
@@ -0,0 +1,34 @@
+/* ===-- powidf2.cpp - Implement __powidf2 ---------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __powidf2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a ^ b */
+
+COMPILER_RT_ABI double
+__powidf2(double a, si_int b)
+{
+    const int recip = b < 0;
+    double r = 1;
+    while (1)
+    {
+        if (b & 1)
+            r *= a;
+        b /= 2;
+        if (b == 0)
+            break;
+        a *= a;
+    }
+    return recip ? 1/r : r;
+}
diff --git a/lib/builtins/powisf2.c b/lib/builtins/powisf2.c
new file mode 100644
index 000000000..0c400ec6d
--- /dev/null
+++ b/lib/builtins/powisf2.c
@@ -0,0 +1,34 @@
+/*===-- powisf2.cpp - Implement __powisf2 ---------------------------------===
+ *
+ *                    The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __powisf2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a ^ b */
+
+COMPILER_RT_ABI float
+__powisf2(float a, si_int b)
+{
+    const int recip = b < 0;
+    float r = 1;
+    while (1)
+    {
+        if (b & 1)
+            r *= a;
+        b /= 2;
+        if (b == 0)
+            break;
+        a *= a;
+    }
+    return recip ? 1/r : r;
+}
diff --git a/lib/builtins/powitf2.c b/lib/builtins/powitf2.c
new file mode 100644
index 000000000..d3b934924
--- /dev/null
+++ b/lib/builtins/powitf2.c
@@ -0,0 +1,38 @@
+/* ===-- powitf2.cpp - Implement __powitf2 ---------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __powitf2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if _ARCH_PPC
+
+/* Returns: a ^ b */
+
+long double
+__powitf2(long double a, si_int b)
+{
+    const int recip = b < 0;
+    long double r = 1;
+    while (1)
+    {
+        if (b & 1)
+            r *= a;
+        b /= 2;
+        if (b == 0)
+            break;
+        a *= a;
+    }
+    return recip ? 1/r : r;
+}
+
+#endif
diff --git a/lib/builtins/powixf2.c b/lib/builtins/powixf2.c
new file mode 100644
index 000000000..f050964d3
--- /dev/null
+++ b/lib/builtins/powixf2.c
@@ -0,0 +1,38 @@
+/* ===-- powixf2.cpp - Implement __powixf2 ---------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __powixf2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#if !_ARCH_PPC
+
+#include "int_lib.h"
+
+/* Returns: a ^ b */
+
+long double
+__powixf2(long double a, si_int b)
+{
+    const int recip = b < 0;
+    long double r = 1;
+    while (1)
+    {
+        if (b & 1)
+            r *= a;
+        b /= 2;
+        if (b == 0)
+            break;
+        a *= a;
+    }
+    return recip ? 1/r : r;
+}
+
+#endif
diff --git a/lib/builtins/ppc/DD.h b/lib/builtins/ppc/DD.h
new file mode 100644
index 000000000..fc3e41cbe
--- /dev/null
+++ b/lib/builtins/ppc/DD.h
@@ -0,0 +1,46 @@
+#ifndef __DD_HEADER
+#define __DD_HEADER
+
+#include "../int_lib.h"
+
+typedef union {
+	long double ld;
+	struct {
+		double hi;
+		double lo;
+	}s;
+}DD;
+
+typedef union { 
+	double d;
+	uint64_t x;
+} doublebits;
+
+#define LOWORDER(xy,xHi,xLo,yHi,yLo) \
+	(((((xHi)*(yHi) - (xy)) + (xHi)*(yLo)) + (xLo)*(yHi)) + (xLo)*(yLo))
+
+static inline double __attribute__((always_inline))
+local_fabs(double x)
+{
+	doublebits result = { .d = x };
+	result.x &= UINT64_C(0x7fffffffffffffff);
+	return result.d;
+}
+
+static inline double __attribute__((always_inline))
+high26bits(double x)
+{
+	doublebits result = { .d = x };
+	result.x &= UINT64_C(0xfffffffff8000000);
+	return result.d;
+}
+
+static inline int __attribute__((always_inline))
+different_sign(double x, double y)
+{
+	doublebits xsignbit = { .d = x }, ysignbit = { .d = y };
+	int result = (int)(xsignbit.x >> 63) ^ (int)(ysignbit.x >> 63);
+	return result;
+}
+
+#endif /* __DD_HEADER */
diff --git a/lib/builtins/ppc/Makefile.mk b/lib/builtins/ppc/Makefile.mk
new file mode 100644
index 000000000..0adc623aa
--- /dev/null
+++ b/lib/builtins/ppc/Makefile.mk
@@ -0,0 +1,20 @@
+#===- lib/builtins/ppc/Makefile.mk -------------------------*- Makefile -*--===#
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+#===------------------------------------------------------------------------===#
+
+ModuleName := builtins
+SubDirs := 
+OnlyArchs := ppc
+
+AsmSources := $(foreach file,$(wildcard $(Dir)/*.S),$(notdir $(file)))
+Sources := $(foreach file,$(wildcard $(Dir)/*.c),$(notdir $(file)))
+ObjNames := $(Sources:%.c=%.o) $(AsmSources:%.S=%.o)
+Implementation := Optimized
+
+# FIXME: use automatic dependencies?
+Dependencies := $(wildcard lib/*.h $(Dir)/*.h)
diff --git a/lib/builtins/ppc/divtc3.c b/lib/builtins/ppc/divtc3.c
new file mode 100644
index 000000000..299128186
--- /dev/null
+++ b/lib/builtins/ppc/divtc3.c
@@ -0,0 +1,96 @@
+/* This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ */
+
+#include "DD.h"
+#include "../int_math.h"
+
+#if !defined(CRT_INFINITY) && defined(HUGE_VAL)
+#define CRT_INFINITY HUGE_VAL
+#endif /* CRT_INFINITY */
+
+#define makeFinite(x) { \
+    (x).s.hi = crt_copysign(crt_isinf((x).s.hi) ? 1.0 : 0.0, (x).s.hi); \
+    (x).s.lo = 0.0;                                                     \
+  }
+
+long double __gcc_qadd(long double, long double);
+long double __gcc_qsub(long double, long double);
+long double __gcc_qmul(long double, long double);
+long double __gcc_qdiv(long double, long double);
+
+long double _Complex
+__divtc3(long double a, long double b, long double c, long double d)
+{
+	DD cDD = { .ld = c };
+	DD dDD = { .ld = d };
+	
+	int ilogbw = 0;
+	const double logbw = crt_logb(crt_fmax(crt_fabs(cDD.s.hi), crt_fabs(dDD.s.hi) ));
+	
+	if (crt_isfinite(logbw))
+	{
+		ilogbw = (int)logbw;
+		
+		cDD.s.hi = crt_scalbn(cDD.s.hi, -ilogbw);
+		cDD.s.lo = crt_scalbn(cDD.s.lo, -ilogbw);
+		dDD.s.hi = crt_scalbn(dDD.s.hi, -ilogbw);
+		dDD.s.lo = crt_scalbn(dDD.s.lo, -ilogbw);
+	}
+	
+	const long double denom = __gcc_qadd(__gcc_qmul(cDD.ld, cDD.ld), __gcc_qmul(dDD.ld, dDD.ld));
+	const long double realNumerator = __gcc_qadd(__gcc_qmul(a,cDD.ld), __gcc_qmul(b,dDD.ld));
+	const long double imagNumerator = __gcc_qsub(__gcc_qmul(b,cDD.ld), __gcc_qmul(a,dDD.ld));
+	
+	DD real = { .ld = __gcc_qdiv(realNumerator, denom) };
+	DD imag = { .ld = __gcc_qdiv(imagNumerator, denom) };
+	
+	real.s.hi = crt_scalbn(real.s.hi, -ilogbw);
+	real.s.lo = crt_scalbn(real.s.lo, -ilogbw);
+	imag.s.hi = crt_scalbn(imag.s.hi, -ilogbw);
+	imag.s.lo = crt_scalbn(imag.s.lo, -ilogbw);
+	
+	if (crt_isnan(real.s.hi) && crt_isnan(imag.s.hi))
+	{
+		DD aDD = { .ld = a };
+		DD bDD = { .ld = b };
+		DD rDD = { .ld = denom };
+		
+		if ((rDD.s.hi == 0.0) && (!crt_isnan(aDD.s.hi) ||
+                                          !crt_isnan(bDD.s.hi)))
+		{
+			real.s.hi = crt_copysign(CRT_INFINITY,cDD.s.hi) * aDD.s.hi;
+			real.s.lo = 0.0;
+			imag.s.hi = crt_copysign(CRT_INFINITY,cDD.s.hi) * bDD.s.hi;
+			imag.s.lo = 0.0;
+		}
+		
+		else if ((crt_isinf(aDD.s.hi) || crt_isinf(bDD.s.hi)) &&
+                         crt_isfinite(cDD.s.hi) && crt_isfinite(dDD.s.hi))
+		{
+			makeFinite(aDD);
+			makeFinite(bDD);
+			real.s.hi = CRT_INFINITY * (aDD.s.hi*cDD.s.hi + bDD.s.hi*dDD.s.hi);
+			real.s.lo = 0.0;
+			imag.s.hi = CRT_INFINITY * (bDD.s.hi*cDD.s.hi - aDD.s.hi*dDD.s.hi);
+			imag.s.lo = 0.0;
+		}
+		
+		else if ((crt_isinf(cDD.s.hi) || crt_isinf(dDD.s.hi)) &&
+                         crt_isfinite(aDD.s.hi) && crt_isfinite(bDD.s.hi))
+		{
+			makeFinite(cDD);
+			makeFinite(dDD);
+			real.s.hi = crt_copysign(0.0,(aDD.s.hi*cDD.s.hi + bDD.s.hi*dDD.s.hi));
+			real.s.lo = 0.0;
+			imag.s.hi = crt_copysign(0.0,(bDD.s.hi*cDD.s.hi - aDD.s.hi*dDD.s.hi));
+			imag.s.lo = 0.0;
+		}
+	}
+	
+	long double _Complex z;
+	__real__ z = real.ld;
+	__imag__ z = imag.ld;
+	
+	return z;
+}
diff --git a/lib/builtins/ppc/fixtfdi.c b/lib/builtins/ppc/fixtfdi.c
new file mode 100644
index 000000000..56e7b3fbf
--- /dev/null
+++ b/lib/builtins/ppc/fixtfdi.c
@@ -0,0 +1,104 @@
+/* This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ */
+
+/* int64_t __fixunstfdi(long double x);
+ * This file implements the PowerPC 128-bit double-double -> int64_t conversion
+ */
+
+#include "DD.h"
+#include "../int_math.h"
+
+uint64_t __fixtfdi(long double input)
+{
+	const DD x = { .ld = input };
+	const doublebits hibits = { .d = x.s.hi };
+	
+	const uint32_t absHighWord = (uint32_t)(hibits.x >> 32) & UINT32_C(0x7fffffff);
+	const uint32_t absHighWordMinusOne = absHighWord - UINT32_C(0x3ff00000);
+	
+	/* If (1.0 - tiny) <= input < 0x1.0p63: */
+	if (UINT32_C(0x03f00000) > absHighWordMinusOne)
+	{
+		/* Do an unsigned conversion of the absolute value, then restore the sign. */
+		const int unbiasedHeadExponent = absHighWordMinusOne >> 20;
+		
+		int64_t result = hibits.x & INT64_C(0x000fffffffffffff); /* mantissa(hi) */
+		result |= INT64_C(0x0010000000000000); /* matissa(hi) with implicit bit */
+		result <<= 10; /* mantissa(hi) with one zero preceeding bit. */
+		
+		const int64_t hiNegationMask = ((int64_t)(hibits.x)) >> 63;
+		
+		/* If the tail is non-zero, we need to patch in the tail bits. */
+		if (0.0 != x.s.lo)
+		{
+			const doublebits lobits = { .d = x.s.lo };
+			int64_t tailMantissa = lobits.x & INT64_C(0x000fffffffffffff);
+			tailMantissa |= INT64_C(0x0010000000000000);
+			
+			/* At this point we have the mantissa of |tail| */
+			/* We need to negate it if head and tail have different signs. */
+			const int64_t loNegationMask = ((int64_t)(lobits.x)) >> 63;
+			const int64_t negationMask = loNegationMask ^ hiNegationMask;
+			tailMantissa = (tailMantissa ^ negationMask) - negationMask;
+			
+			/* Now we have the mantissa of tail as a signed 2s-complement integer */
+			
+			const int biasedTailExponent = (int)(lobits.x >> 52) & 0x7ff;
+			
+			/* Shift the tail mantissa into the right position, accounting for the
+			 * bias of 10 that we shifted the head mantissa by.
+			 */ 
+			tailMantissa >>= (unbiasedHeadExponent - (biasedTailExponent - (1023 - 10)));
+			
+			result += tailMantissa;
+		}
+		
+		result >>= (62 - unbiasedHeadExponent);
+		
+		/* Restore the sign of the result and return */
+		result = (result ^ hiNegationMask) - hiNegationMask;
+		return result;
+		
+	}
+
+	/* Edge cases handled here: */
+	
+	/* |x| < 1, result is zero. */
+	if (1.0 > crt_fabs(x.s.hi))
+		return INT64_C(0);
+	
+	/* x very close to INT64_MIN, care must be taken to see which side we are on. */
+	if (x.s.hi == -0x1.0p63) {
+		
+		int64_t result = INT64_MIN;
+		
+		if (0.0 < x.s.lo)
+		{
+			/* If the tail is positive, the correct result is something other than INT64_MIN.
+			 * we'll need to figure out what it is.
+			 */
+
+			const doublebits lobits = { .d = x.s.lo };
+			int64_t tailMantissa = lobits.x & INT64_C(0x000fffffffffffff);
+			tailMantissa |= INT64_C(0x0010000000000000);
+			
+			/* Now we negate the tailMantissa */
+			tailMantissa = (tailMantissa ^ INT64_C(-1)) + INT64_C(1);
+			
+			/* And shift it by the appropriate amount */
+			const int biasedTailExponent = (int)(lobits.x >> 52) & 0x7ff;
+			tailMantissa >>= 1075 - biasedTailExponent;
+			
+			result -= tailMantissa;
+		}
+		
+		return result;
+	}
+	
+	/* Signed overflows, infinities, and NaNs */
+	if (x.s.hi > 0.0)
+		return INT64_MAX;
+	else
+		return INT64_MIN;
+}
diff --git a/lib/builtins/ppc/fixunstfdi.c b/lib/builtins/ppc/fixunstfdi.c
new file mode 100644
index 000000000..5e6e2cedf
--- /dev/null
+++ b/lib/builtins/ppc/fixunstfdi.c
@@ -0,0 +1,59 @@
+/* This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ */
+
+/* uint64_t __fixunstfdi(long double x); */
+/* This file implements the PowerPC 128-bit double-double -> uint64_t conversion */
+
+#include "DD.h"
+
+uint64_t __fixunstfdi(long double input)
+{
+	const DD x = { .ld = input };
+	const doublebits hibits = { .d = x.s.hi };
+	
+	const uint32_t highWordMinusOne = (uint32_t)(hibits.x >> 32) - UINT32_C(0x3ff00000);
+	
+	/* If (1.0 - tiny) <= input < 0x1.0p64: */
+	if (UINT32_C(0x04000000) > highWordMinusOne)
+	{
+		const int unbiasedHeadExponent = highWordMinusOne >> 20;
+		
+		uint64_t result = hibits.x & UINT64_C(0x000fffffffffffff); /* mantissa(hi) */
+		result |= UINT64_C(0x0010000000000000); /* matissa(hi) with implicit bit */
+		result <<= 11; /* mantissa(hi) left aligned in the int64 field. */
+		
+		/* If the tail is non-zero, we need to patch in the tail bits. */
+		if (0.0 != x.s.lo)
+		{
+			const doublebits lobits = { .d = x.s.lo };
+			int64_t tailMantissa = lobits.x & INT64_C(0x000fffffffffffff);
+			tailMantissa |= INT64_C(0x0010000000000000);
+			
+			/* At this point we have the mantissa of |tail| */
+			
+			const int64_t negationMask = ((int64_t)(lobits.x)) >> 63;
+			tailMantissa = (tailMantissa ^ negationMask) - negationMask;
+			
+			/* Now we have the mantissa of tail as a signed 2s-complement integer */
+			
+			const int biasedTailExponent = (int)(lobits.x >> 52) & 0x7ff;
+			
+			/* Shift the tail mantissa into the right position, accounting for the
+			 * bias of 11 that we shifted the head mantissa by.
+			 */
+			tailMantissa >>= (unbiasedHeadExponent - (biasedTailExponent - (1023 - 11)));
+			
+			result += tailMantissa;
+		}
+		
+		result >>= (63 - unbiasedHeadExponent);
+		return result;
+	}
+	
+	/* Edge cases are handled here, with saturation. */
+	if (1.0 > x.s.hi)
+		return UINT64_C(0);
+	else
+		return UINT64_MAX;
+}
diff --git a/lib/builtins/ppc/floatditf.c b/lib/builtins/ppc/floatditf.c
new file mode 100644
index 000000000..beabdd017
--- /dev/null
+++ b/lib/builtins/ppc/floatditf.c
@@ -0,0 +1,36 @@
+/* This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ */
+
+/* long double __floatditf(long long x); */
+/* This file implements the PowerPC long long -> long double conversion */
+
+#include "DD.h"
+
+long double __floatditf(int64_t a) {
+	
+	static const double twop32 = 0x1.0p32;
+	static const double twop52 = 0x1.0p52;
+	
+	doublebits low  = { .d = twop52 };
+	low.x |= a & UINT64_C(0x00000000ffffffff);	/* 0x1.0p52 + low 32 bits of a. */
+	
+	const double high_addend = (double)((int32_t)(a >> 32))*twop32 - twop52;
+	
+	/* At this point, we have two double precision numbers
+	 * high_addend and low.d, and we wish to return their sum
+	 * as a canonicalized long double:
+	 */
+
+	/* This implementation sets the inexact flag spuriously.
+	 * This could be avoided, but at some substantial cost.
+	*/
+
+	DD result;
+	
+	result.s.hi = high_addend + low.d;
+	result.s.lo = (high_addend - result.s.hi) + low.d;
+	
+	return result.ld;
+	
+}
diff --git a/lib/builtins/ppc/floatunditf.c b/lib/builtins/ppc/floatunditf.c
new file mode 100644
index 000000000..b12e1e738
--- /dev/null
+++ b/lib/builtins/ppc/floatunditf.c
@@ -0,0 +1,41 @@
+/* This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ */
+
+/* long double __floatunditf(unsigned long long x); */
+/* This file implements the PowerPC unsigned long long -> long double conversion */
+
+#include "DD.h"
+
+long double __floatunditf(uint64_t a) {
+	
+	/* Begins with an exact copy of the code from __floatundidf */
+	
+	static const double twop52 = 0x1.0p52;
+	static const double twop84 = 0x1.0p84;
+	static const double twop84_plus_twop52 = 0x1.00000001p84;
+	
+	doublebits high = { .d = twop84 };
+	doublebits low  = { .d = twop52 };
+	
+	high.x |= a >> 32;							/* 0x1.0p84 + high 32 bits of a */
+	low.x |= a & UINT64_C(0x00000000ffffffff);	/* 0x1.0p52 + low 32 bits of a */
+	
+	const double high_addend = high.d - twop84_plus_twop52;
+	
+	/* At this point, we have two double precision numbers
+	 * high_addend and low.d, and we wish to return their sum
+	 * as a canonicalized long double:
+	 */
+
+	/* This implementation sets the inexact flag spuriously. */
+	/* This could be avoided, but at some substantial cost. */
+	
+	DD result;
+	
+	result.s.hi = high_addend + low.d;
+	result.s.lo = (high_addend - result.s.hi) + low.d;
+	
+	return result.ld;
+	
+}
diff --git a/lib/builtins/ppc/gcc_qadd.c b/lib/builtins/ppc/gcc_qadd.c
new file mode 100644
index 000000000..32e16e9d1
--- /dev/null
+++ b/lib/builtins/ppc/gcc_qadd.c
@@ -0,0 +1,76 @@
+/* This file is distributed under the University of Illinois Open Source
+ *  License. See LICENSE.TXT for details.
+ */
+
+/* long double __gcc_qadd(long double x, long double y);
+ * This file implements the PowerPC 128-bit double-double add operation.
+ * This implementation is shamelessly cribbed from Apple's DDRT, circa 1993(!)
+ */
+
+#include "DD.h"
+
+long double __gcc_qadd(long double x, long double y)
+{
+	static const uint32_t infinityHi = UINT32_C(0x7ff00000);
+	
+	DD dst = { .ld = x }, src = { .ld = y };
+	
+	register double A = dst.s.hi, a = dst.s.lo,
+					B = src.s.hi, b = src.s.lo;
+	
+	/* If both operands are zero: */
+	if ((A == 0.0) && (B == 0.0)) {
+		dst.s.hi = A + B;
+		dst.s.lo = 0.0;
+		return dst.ld;
+	}
+	
+	/* If either operand is NaN or infinity: */
+	const doublebits abits = { .d = A };
+	const doublebits bbits = { .d = B };
+	if ((((uint32_t)(abits.x >> 32) & infinityHi) == infinityHi) ||
+		(((uint32_t)(bbits.x >> 32) & infinityHi) == infinityHi)) {
+		dst.s.hi = A + B;
+		dst.s.lo = 0.0;
+		return dst.ld;
+	}
+	
+	/* If the computation overflows: */
+	/* This may be playing things a little bit fast and loose, but it will do for a start. */
+	const double testForOverflow = A + (B + (a + b));
+	const doublebits testbits = { .d = testForOverflow };
+	if (((uint32_t)(testbits.x >> 32) & infinityHi) == infinityHi) {
+		dst.s.hi = testForOverflow;
+		dst.s.lo = 0.0;
+		return dst.ld;
+	}
+	
+	double H, h;
+	double T, t;
+	double W, w;
+	double Y;
+	
+	H = B + (A - (A + B));
+	T = b + (a - (a + b));
+	h = A + (B - (A + B));
+	t = a + (b - (a + b));
+	
+	if (local_fabs(A) <= local_fabs(B))
+		w = (a + b) + h;
+	else
+		w = (a + b) + H;
+	
+	W = (A + B) + w;
+	Y = (A + B) - W;
+	Y += w;
+	
+	if (local_fabs(a) <= local_fabs(b))
+		w = t + Y;
+	else
+		w = T + Y;
+	
+	dst.s.hi = Y = W + w;
+	dst.s.lo = (W - Y) + w;
+	
+	return dst.ld;
+}
diff --git a/lib/builtins/ppc/gcc_qdiv.c b/lib/builtins/ppc/gcc_qdiv.c
new file mode 100644
index 000000000..70aa00b64
--- /dev/null
+++ b/lib/builtins/ppc/gcc_qdiv.c
@@ -0,0 +1,55 @@
+/* This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ */
+
+/* long double __gcc_qdiv(long double x, long double y);
+ * This file implements the PowerPC 128-bit double-double division operation.
+ * This implementation is shamelessly cribbed from Apple's DDRT, circa 1993(!)
+ */
+
+#include "DD.h"
+
+long double __gcc_qdiv(long double a, long double b)
+{	
+	static const uint32_t infinityHi = UINT32_C(0x7ff00000);
+	DD dst = { .ld = a }, src = { .ld = b };
+	
+	register double x = dst.s.hi, x1 = dst.s.lo,
+					y = src.s.hi, y1 = src.s.lo;
+	
+    double yHi, yLo, qHi, qLo;
+    double yq, tmp, q;
+	
+    q = x / y;
+	
+	/* Detect special cases */
+	if (q == 0.0) {
+		dst.s.hi = q;
+		dst.s.lo = 0.0;
+		return dst.ld;
+	}
+	
+	const doublebits qBits = { .d = q };
+	if (((uint32_t)(qBits.x >> 32) & infinityHi) == infinityHi) {
+		dst.s.hi = q;
+		dst.s.lo = 0.0;
+		return dst.ld;
+	}
+	
+    yHi = high26bits(y);
+    qHi = high26bits(q);
+	
+    yq = y * q;
+    yLo = y - yHi;
+    qLo = q - qHi;
+	
+    tmp = LOWORDER(yq, yHi, yLo, qHi, qLo);
+    tmp = (x - yq) - tmp;
+    tmp = ((tmp + x1) - y1 * q) / y;
+    x = q + tmp;
+	
+    dst.s.lo = (q - x) + tmp;
+    dst.s.hi = x;
+	
+    return dst.ld;
+}
diff --git a/lib/builtins/ppc/gcc_qmul.c b/lib/builtins/ppc/gcc_qmul.c
new file mode 100644
index 000000000..fb4c5164c
--- /dev/null
+++ b/lib/builtins/ppc/gcc_qmul.c
@@ -0,0 +1,53 @@
+/* This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ */
+
+/* long double __gcc_qmul(long double x, long double y);
+ * This file implements the PowerPC 128-bit double-double multiply operation.
+ * This implementation is shamelessly cribbed from Apple's DDRT, circa 1993(!)
+ */
+
+#include "DD.h"
+
+long double __gcc_qmul(long double x, long double y)
+{	
+	static const uint32_t infinityHi = UINT32_C(0x7ff00000);
+	DD dst = { .ld = x }, src = { .ld = y };
+	
+	register double A = dst.s.hi, a = dst.s.lo,
+					B = src.s.hi, b = src.s.lo;
+	
+	double aHi, aLo, bHi, bLo;
+    double ab, tmp, tau;
+	
+	ab = A * B;
+	
+	/* Detect special cases */
+	if (ab == 0.0) {
+		dst.s.hi = ab;
+		dst.s.lo = 0.0;
+		return dst.ld;
+	}
+	
+	const doublebits abBits = { .d = ab };
+	if (((uint32_t)(abBits.x >> 32) & infinityHi) == infinityHi) {
+		dst.s.hi = ab;
+		dst.s.lo = 0.0;
+		return dst.ld;
+	}
+	
+	/* Generic cases handled here. */
+    aHi = high26bits(A);
+    bHi = high26bits(B);
+    aLo = A - aHi;
+    bLo = B - bHi;
+	
+    tmp = LOWORDER(ab, aHi, aLo, bHi, bLo);
+    tmp += (A * b + a * B);
+    tau = ab + tmp;
+	
+    dst.s.lo = (ab - tau) + tmp;
+    dst.s.hi = tau;
+	
+    return dst.ld;
+}
diff --git a/lib/builtins/ppc/gcc_qsub.c b/lib/builtins/ppc/gcc_qsub.c
new file mode 100644
index 000000000..c092e24db
--- /dev/null
+++ b/lib/builtins/ppc/gcc_qsub.c
@@ -0,0 +1,76 @@
+/* This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ */
+
+/* long double __gcc_qsub(long double x, long double y);
+ * This file implements the PowerPC 128-bit double-double add operation.
+ * This implementation is shamelessly cribbed from Apple's DDRT, circa 1993(!)
+ */
+
+#include "DD.h"
+
+long double __gcc_qsub(long double x, long double y)
+{
+	static const uint32_t infinityHi = UINT32_C(0x7ff00000);
+	
+	DD dst = { .ld = x }, src = { .ld = y };
+	
+	register double A =  dst.s.hi, a =  dst.s.lo,
+					B = -src.s.hi, b = -src.s.lo;
+	
+	/* If both operands are zero: */
+	if ((A == 0.0) && (B == 0.0)) {
+		dst.s.hi = A + B;
+		dst.s.lo = 0.0;
+		return dst.ld;
+	}
+	
+	/* If either operand is NaN or infinity: */
+	const doublebits abits = { .d = A };
+	const doublebits bbits = { .d = B };
+	if ((((uint32_t)(abits.x >> 32) & infinityHi) == infinityHi) ||
+		(((uint32_t)(bbits.x >> 32) & infinityHi) == infinityHi)) {
+		dst.s.hi = A + B;
+		dst.s.lo = 0.0;
+		return dst.ld;
+	}
+	
+	/* If the computation overflows: */
+	/* This may be playing things a little bit fast and loose, but it will do for a start. */
+	const double testForOverflow = A + (B + (a + b));
+	const doublebits testbits = { .d = testForOverflow };
+	if (((uint32_t)(testbits.x >> 32) & infinityHi) == infinityHi) {
+		dst.s.hi = testForOverflow;
+		dst.s.lo = 0.0;
+		return dst.ld;
+	}
+	
+	double H, h;
+	double T, t;
+	double W, w;
+	double Y;
+	
+	H = B + (A - (A + B));
+	T = b + (a - (a + b));
+	h = A + (B - (A + B));
+	t = a + (b - (a + b));
+	
+	if (local_fabs(A) <= local_fabs(B))
+		w = (a + b) + h;
+	else
+		w = (a + b) + H;
+	
+	W = (A + B) + w;
+	Y = (A + B) - W;
+	Y += w;
+	
+	if (local_fabs(a) <= local_fabs(b))
+		w = t + Y;
+	else
+		w = T + Y;
+	
+	dst.s.hi = Y = W + w;
+	dst.s.lo = (W - Y) + w;
+	
+	return dst.ld;
+}
diff --git a/lib/builtins/ppc/multc3.c b/lib/builtins/ppc/multc3.c
new file mode 100644
index 000000000..738b65a83
--- /dev/null
+++ b/lib/builtins/ppc/multc3.c
@@ -0,0 +1,94 @@
+/* This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ */
+
+#include "DD.h"
+#include "../int_math.h"
+
+#define makeFinite(x) { \
+    (x).s.hi = crt_copysign(crt_isinf((x).s.hi) ? 1.0 : 0.0, (x).s.hi); \
+    (x).s.lo = 0.0;                                                     \
+  }
+
+#define zeroNaN(x) { \
+    if (crt_isnan((x).s.hi)) {                                          \
+      (x).s.hi = crt_copysign(0.0, (x).s.hi);                     \
+      (x).s.lo = 0.0;                                                   \
+    }                                                                   \
+  }
+
+long double __gcc_qadd(long double, long double);
+long double __gcc_qsub(long double, long double);
+long double __gcc_qmul(long double, long double);
+
+long double _Complex
+__multc3(long double a, long double b, long double c, long double d)
+{
+	long double ac = __gcc_qmul(a,c);
+	long double bd = __gcc_qmul(b,d);
+	long double ad = __gcc_qmul(a,d);
+	long double bc = __gcc_qmul(b,c);
+	
+	DD real = { .ld = __gcc_qsub(ac,bd) };
+	DD imag = { .ld = __gcc_qadd(ad,bc) };
+	
+	if (crt_isnan(real.s.hi) && crt_isnan(imag.s.hi))
+	{
+		int recalc = 0;
+		
+		DD aDD = { .ld = a };
+		DD bDD = { .ld = b };
+		DD cDD = { .ld = c };
+		DD dDD = { .ld = d };
+		
+		if (crt_isinf(aDD.s.hi) || crt_isinf(bDD.s.hi))
+		{
+			makeFinite(aDD);
+			makeFinite(bDD);
+			zeroNaN(cDD);
+			zeroNaN(dDD);
+			recalc = 1;
+		}
+		
+		if (crt_isinf(cDD.s.hi) || crt_isinf(dDD.s.hi))
+		{
+			makeFinite(cDD);
+			makeFinite(dDD);
+			zeroNaN(aDD);
+			zeroNaN(bDD);
+			recalc = 1;
+		}
+		
+		if (!recalc)
+		{
+			DD acDD = { .ld = ac };
+			DD bdDD = { .ld = bd };
+			DD adDD = { .ld = ad };
+			DD bcDD = { .ld = bc };
+			
+			if (crt_isinf(acDD.s.hi) || crt_isinf(bdDD.s.hi) ||
+                            crt_isinf(adDD.s.hi) || crt_isinf(bcDD.s.hi))
+			{
+				zeroNaN(aDD);
+				zeroNaN(bDD);
+				zeroNaN(cDD);
+				zeroNaN(dDD);
+				recalc = 1;
+			}
+		}
+		
+		if (recalc)
+		{
+			real.s.hi = CRT_INFINITY * (aDD.s.hi*cDD.s.hi - bDD.s.hi*dDD.s.hi);
+			real.s.lo = 0.0;
+			imag.s.hi = CRT_INFINITY * (aDD.s.hi*dDD.s.hi + bDD.s.hi*cDD.s.hi);
+			imag.s.lo = 0.0;
+		}
+	}
+	
+	long double _Complex z;
+	__real__ z = real.ld;
+	__imag__ z = imag.ld;
+	
+	return z;
+}
diff --git a/lib/builtins/ppc/restFP.S b/lib/builtins/ppc/restFP.S
new file mode 100644
index 000000000..95032897c
--- /dev/null
+++ b/lib/builtins/ppc/restFP.S
@@ -0,0 +1,43 @@
+//===-- restFP.S - Implement restFP ---------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// Helper function used by compiler to restore ppc floating point registers at
+// the end of the function epilog.  This function returns to the address
+// in the LR slot.  So a function epilog must branch (b) not branch and link
+// (bl) to this function.
+// If the compiler wants to restore f27..f31, it does a "b restFP+52"
+//
+// This function should never be exported by a shared library.  Each linkage
+// unit carries its own copy of this function.
+//
+DEFINE_COMPILERRT_PRIVATE_FUNCTION_UNMANGLED(restFP)
+        lfd    f14,-144(r1)
+        lfd    f15,-136(r1)
+        lfd    f16,-128(r1)
+        lfd    f17,-120(r1)
+        lfd    f18,-112(r1)
+        lfd    f19,-104(r1)
+        lfd    f20,-96(r1)
+        lfd    f21,-88(r1)
+        lfd    f22,-80(r1)
+        lfd    f23,-72(r1)
+        lfd    f24,-64(r1)
+        lfd    f25,-56(r1)
+        lfd    f26,-48(r1)
+        lfd    f27,-40(r1)
+        lfd    f28,-32(r1)
+        lfd    f29,-24(r1)
+        lfd    f30,-16(r1)
+        lfd    f31,-8(r1)
+        lwz     r0,8(r1)
+        mtlr	r0
+        blr
diff --git a/lib/builtins/ppc/saveFP.S b/lib/builtins/ppc/saveFP.S
new file mode 100644
index 000000000..72bd459f4
--- /dev/null
+++ b/lib/builtins/ppc/saveFP.S
@@ -0,0 +1,40 @@
+//===-- saveFP.S - Implement saveFP ---------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+//
+// Helper function used by compiler to save ppc floating point registers in
+// function prologs.  This routines also saves r0 in the LR slot.
+// If the compiler wants to save f27..f31, it does a "bl saveFP+52"
+//
+// This function should never be exported by a shared library.  Each linkage
+// unit carries its own copy of this function.
+//
+DEFINE_COMPILERRT_PRIVATE_FUNCTION_UNMANGLED(saveFP)
+	stfd    f14,-144(r1)
+        stfd    f15,-136(r1)
+        stfd    f16,-128(r1)
+        stfd    f17,-120(r1)
+        stfd    f18,-112(r1)
+        stfd    f19,-104(r1)
+        stfd    f20,-96(r1)
+        stfd    f21,-88(r1)
+        stfd    f22,-80(r1)
+        stfd    f23,-72(r1)
+        stfd    f24,-64(r1)
+        stfd    f25,-56(r1)
+        stfd    f26,-48(r1)
+        stfd    f27,-40(r1)
+        stfd    f28,-32(r1)
+        stfd    f29,-24(r1)
+        stfd    f30,-16(r1)
+        stfd    f31,-8(r1)
+        stw      r0,8(r1)
+        blr
diff --git a/lib/builtins/subdf3.c b/lib/builtins/subdf3.c
new file mode 100644
index 000000000..66fb1a54f
--- /dev/null
+++ b/lib/builtins/subdf3.c
@@ -0,0 +1,29 @@
+//===-- lib/adddf3.c - Double-precision subtraction ---------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements double-precision soft-float subtraction with the
+// IEEE-754 default rounding (to nearest, ties to even).
+//
+//===----------------------------------------------------------------------===//
+
+#define DOUBLE_PRECISION
+#include "fp_lib.h"
+
+fp_t COMPILER_RT_ABI __adddf3(fp_t a, fp_t b);
+
+
+ARM_EABI_FNALIAS(dsub, subdf3)
+
+// Subtraction; flip the sign bit of b and add.
+COMPILER_RT_ABI fp_t
+__subdf3(fp_t a, fp_t b) {
+    return __adddf3(a, fromRep(toRep(b) ^ signBit));
+}
+
+/* FIXME: rsub for ARM EABI */
diff --git a/lib/builtins/subsf3.c b/lib/builtins/subsf3.c
new file mode 100644
index 000000000..3659cd8b8
--- /dev/null
+++ b/lib/builtins/subsf3.c
@@ -0,0 +1,28 @@
+//===-- lib/subsf3.c - Single-precision subtraction ---------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements single-precision soft-float subtraction with the
+// IEEE-754 default rounding (to nearest, ties to even).
+//
+//===----------------------------------------------------------------------===//
+
+#define SINGLE_PRECISION
+#include "fp_lib.h"
+
+fp_t COMPILER_RT_ABI __addsf3(fp_t a, fp_t b);
+
+ARM_EABI_FNALIAS(fsub, subsf3)
+
+// Subtraction; flip the sign bit of b and add.
+COMPILER_RT_ABI fp_t
+__subsf3(fp_t a, fp_t b) {
+    return __addsf3(a, fromRep(toRep(b) ^ signBit));
+}
+
+/* FIXME: rsub for ARM EABI */
diff --git a/lib/builtins/subvdi3.c b/lib/builtins/subvdi3.c
new file mode 100644
index 000000000..0f1f924ef
--- /dev/null
+++ b/lib/builtins/subvdi3.c
@@ -0,0 +1,36 @@
+/* ===-- subvdi3.c - Implement __subvdi3 -----------------------------------===
+ *
+ *                The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __subvdi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a - b */
+
+/* Effects: aborts if a - b overflows */
+
+COMPILER_RT_ABI di_int
+__subvdi3(di_int a, di_int b)
+{
+    di_int s = a - b;
+    if (b >= 0)
+    {
+        if (s > a)
+            compilerrt_abort();
+    }
+    else
+    {
+        if (s <= a)
+            compilerrt_abort();
+    }
+    return s;
+}
diff --git a/lib/builtins/subvsi3.c b/lib/builtins/subvsi3.c
new file mode 100644
index 000000000..ec4594c9f
--- /dev/null
+++ b/lib/builtins/subvsi3.c
@@ -0,0 +1,36 @@
+/* ===-- subvsi3.c - Implement __subvsi3 -----------------------------------===
+ *
+ *                The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __subvsi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a - b */
+
+/* Effects: aborts if a - b overflows */
+
+COMPILER_RT_ABI si_int
+__subvsi3(si_int a, si_int b)
+{
+    si_int s = a - b;
+    if (b >= 0)
+    {
+        if (s > a)
+            compilerrt_abort();
+    }
+    else
+    {
+        if (s <= a)
+            compilerrt_abort();
+    }
+    return s;
+}
diff --git a/lib/builtins/subvti3.c b/lib/builtins/subvti3.c
new file mode 100644
index 000000000..b32df5e6c
--- /dev/null
+++ b/lib/builtins/subvti3.c
@@ -0,0 +1,40 @@
+/* ===-- subvti3.c - Implement __subvti3 -----------------------------------===
+ *
+ *      	       The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __subvti3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns: a - b */
+
+/* Effects: aborts if a - b overflows */
+
+ti_int
+__subvti3(ti_int a, ti_int b)
+{
+    ti_int s = a - b;
+    if (b >= 0)
+    {
+        if (s > a)
+            compilerrt_abort();
+    }
+    else
+    {
+        if (s <= a)
+            compilerrt_abort();
+    }
+    return s;
+}
+
+#endif /* __x86_64 */
diff --git a/lib/builtins/trampoline_setup.c b/lib/builtins/trampoline_setup.c
new file mode 100644
index 000000000..e0765b16b
--- /dev/null
+++ b/lib/builtins/trampoline_setup.c
@@ -0,0 +1,47 @@
+/* ===----- trampoline_setup.c - Implement __trampoline_setup -------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+extern void __clear_cache(void* start, void* end);
+
+/*
+ * The ppc compiler generates calls to __trampoline_setup() when creating 
+ * trampoline functions on the stack for use with nested functions.
+ * This function creates a custom 40-byte trampoline function on the stack 
+ * which loads r11 with a pointer to the outer function's locals
+ * and then jumps to the target nested function.
+ */
+
+#if __ppc__ && !defined(__powerpc64__)
+void __trampoline_setup(uint32_t* trampOnStack, int trampSizeAllocated, 
+                                const void* realFunc, void* localsPtr)
+{
+    /* should never happen, but if compiler did not allocate */
+    /* enough space on stack for the trampoline, abort */
+    if ( trampSizeAllocated < 40 )
+        compilerrt_abort();
+    
+    /* create trampoline */
+    trampOnStack[0] = 0x7c0802a6;    /* mflr r0 */
+    trampOnStack[1] = 0x4800000d;    /* bl Lbase */
+    trampOnStack[2] = (uint32_t)realFunc;
+    trampOnStack[3] = (uint32_t)localsPtr;
+    trampOnStack[4] = 0x7d6802a6;    /* Lbase: mflr r11 */
+    trampOnStack[5] = 0x818b0000;    /* lwz    r12,0(r11) */
+    trampOnStack[6] = 0x7c0803a6;    /* mtlr r0 */
+    trampOnStack[7] = 0x7d8903a6;    /* mtctr r12 */
+    trampOnStack[8] = 0x816b0004;    /* lwz    r11,4(r11) */
+    trampOnStack[9] = 0x4e800420;    /* bctr */
+    
+    /* clear instruction cache */
+    __clear_cache(trampOnStack, &trampOnStack[10]);
+}
+#endif /* __ppc__ && !defined(__powerpc64__) */
diff --git a/lib/builtins/truncdfsf2.c b/lib/builtins/truncdfsf2.c
new file mode 100644
index 000000000..61c909a22
--- /dev/null
+++ b/lib/builtins/truncdfsf2.c
@@ -0,0 +1,168 @@
+//===-- lib/truncdfsf2.c - double -> single conversion ------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements a fairly generic conversion from a wider to a narrower
+// IEEE-754 floating-point type in the default (round to nearest, ties to even)
+// rounding mode.  The constants and types defined following the includes below
+// parameterize the conversion.
+//
+// This routine can be trivially adapted to support conversions to 
+// half-precision or from quad-precision. It does not support types that don't
+// use the usual IEEE-754 interchange formats; specifically, some work would be
+// needed to adapt it to (for example) the Intel 80-bit format or PowerPC
+// double-double format.
+//
+// Note please, however, that this implementation is only intended to support
+// *narrowing* operations; if you need to convert to a *wider* floating-point
+// type (e.g. float -> double), then this routine will not do what you want it
+// to.
+//
+// It also requires that integer types at least as large as both formats
+// are available on the target platform; this may pose a problem when trying
+// to add support for quad on some 32-bit systems, for example.
+//
+// Finally, the following assumptions are made:
+//
+// 1. floating-point types and integer types have the same endianness on the
+//    target platform
+//
+// 2. quiet NaNs, if supported, are indicated by the leading bit of the
+//    significand field being set
+//
+//===----------------------------------------------------------------------===//
+
+#include "int_lib.h"
+
+typedef double src_t;
+typedef uint64_t src_rep_t;
+#define SRC_REP_C UINT64_C
+static const int srcSigBits = 52;
+
+typedef float dst_t;
+typedef uint32_t dst_rep_t;
+#define DST_REP_C UINT32_C
+static const int dstSigBits = 23;
+
+// End of specialization parameters.  Two helper routines for conversion to and
+// from the representation of floating-point data as integer values follow.
+
+static inline src_rep_t srcToRep(src_t x) {
+    const union { src_t f; src_rep_t i; } rep = {.f = x};
+    return rep.i;
+}
+
+static inline dst_t dstFromRep(dst_rep_t x) {
+    const union { dst_t f; dst_rep_t i; } rep = {.i = x};
+    return rep.f;
+}
+
+// End helper routines.  Conversion implementation follows.
+
+ARM_EABI_FNALIAS(d2f, truncdfsf2)
+
+COMPILER_RT_ABI dst_t
+__truncdfsf2(src_t a) {
+    
+    // Various constants whose values follow from the type parameters.
+    // Any reasonable optimizer will fold and propagate all of these.
+    const int srcBits = sizeof(src_t)*CHAR_BIT;
+    const int srcExpBits = srcBits - srcSigBits - 1;
+    const int srcInfExp = (1 << srcExpBits) - 1;
+    const int srcExpBias = srcInfExp >> 1;
+    
+    const src_rep_t srcMinNormal = SRC_REP_C(1) << srcSigBits;
+    const src_rep_t significandMask = srcMinNormal - 1;
+    const src_rep_t srcInfinity = (src_rep_t)srcInfExp << srcSigBits;
+    const src_rep_t srcSignMask = SRC_REP_C(1) << (srcSigBits + srcExpBits);
+    const src_rep_t srcAbsMask = srcSignMask - 1;
+    const src_rep_t roundMask = (SRC_REP_C(1) << (srcSigBits - dstSigBits)) - 1;
+    const src_rep_t halfway = SRC_REP_C(1) << (srcSigBits - dstSigBits - 1);
+    
+    const int dstBits = sizeof(dst_t)*CHAR_BIT;
+    const int dstExpBits = dstBits - dstSigBits - 1;
+    const int dstInfExp = (1 << dstExpBits) - 1;
+    const int dstExpBias = dstInfExp >> 1;
+    
+    const int underflowExponent = srcExpBias + 1 - dstExpBias;
+    const int overflowExponent = srcExpBias + dstInfExp - dstExpBias;
+    const src_rep_t underflow = (src_rep_t)underflowExponent << srcSigBits;
+    const src_rep_t overflow = (src_rep_t)overflowExponent << srcSigBits;
+    
+    const dst_rep_t dstQNaN = DST_REP_C(1) << (dstSigBits - 1);
+    const dst_rep_t dstNaNCode = dstQNaN - 1;
+
+    // Break a into a sign and representation of the absolute value
+    const src_rep_t aRep = srcToRep(a);
+    const src_rep_t aAbs = aRep & srcAbsMask;
+    const src_rep_t sign = aRep & srcSignMask;
+    dst_rep_t absResult;
+    
+    if (aAbs - underflow < aAbs - overflow) {
+        // The exponent of a is within the range of normal numbers in the
+        // destination format.  We can convert by simply right-shifting with
+        // rounding and adjusting the exponent.
+        absResult = aAbs >> (srcSigBits - dstSigBits);
+        absResult -= (dst_rep_t)(srcExpBias - dstExpBias) << dstSigBits;
+        
+        const src_rep_t roundBits = aAbs & roundMask;
+        
+        // Round to nearest
+        if (roundBits > halfway)
+            absResult++;
+        
+        // Ties to even
+        else if (roundBits == halfway)
+            absResult += absResult & 1;
+    }
+    
+    else if (aAbs > srcInfinity) {
+        // a is NaN.
+        // Conjure the result by beginning with infinity, setting the qNaN
+        // bit and inserting the (truncated) trailing NaN field.
+        absResult = (dst_rep_t)dstInfExp << dstSigBits;
+        absResult |= dstQNaN;
+        absResult |= aAbs & dstNaNCode;
+    }
+    
+    else if (aAbs > overflow) {
+        // a overflows to infinity.
+        absResult = (dst_rep_t)dstInfExp << dstSigBits;
+    }
+    
+    else {
+        // a underflows on conversion to the destination type or is an exact
+        // zero.  The result may be a denormal or zero.  Extract the exponent
+        // to get the shift amount for the denormalization.
+        const int aExp = aAbs >> srcSigBits;
+        const int shift = srcExpBias - dstExpBias - aExp + 1;
+        
+        const src_rep_t significand = (aRep & significandMask) | srcMinNormal;
+        
+        // Right shift by the denormalization amount with sticky.
+        if (shift > srcSigBits) {
+            absResult = 0;
+        } else {
+            const bool sticky = significand << (srcBits - shift);
+            src_rep_t denormalizedSignificand = significand >> shift | sticky;
+            absResult = denormalizedSignificand >> (srcSigBits - dstSigBits);
+            const src_rep_t roundBits = denormalizedSignificand & roundMask;
+            // Round to nearest
+            if (roundBits > halfway)
+                absResult++;
+            // Ties to even
+            else if (roundBits == halfway)
+                absResult += absResult & 1;
+        }
+    }
+    
+    // Apply the signbit to (dst_t)abs(a).
+    const dst_rep_t result = absResult | sign >> (srcBits - dstBits);
+    return dstFromRep(result);
+    
+}
diff --git a/lib/builtins/ucmpdi2.c b/lib/builtins/ucmpdi2.c
new file mode 100644
index 000000000..40af23613
--- /dev/null
+++ b/lib/builtins/ucmpdi2.c
@@ -0,0 +1,51 @@
+/* ===-- ucmpdi2.c - Implement __ucmpdi2 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __ucmpdi2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns:  if (a <  b) returns 0
+ *           if (a == b) returns 1
+ *           if (a >  b) returns 2
+ */
+
+COMPILER_RT_ABI si_int
+__ucmpdi2(du_int a, du_int b)
+{
+    udwords x;
+    x.all = a;
+    udwords y;
+    y.all = b;
+    if (x.s.high < y.s.high)
+        return 0;
+    if (x.s.high > y.s.high)
+        return 2;
+    if (x.s.low < y.s.low)
+        return 0;
+    if (x.s.low > y.s.low)
+        return 2;
+    return 1;
+}
+
+#ifdef __ARM_EABI__
+/* Returns: if (a <  b) returns -1
+*           if (a == b) returns  0
+*           if (a >  b) returns  1
+*/
+COMPILER_RT_ABI si_int
+__aeabi_ulcmp(di_int a, di_int b)
+{
+	return __ucmpdi2(a, b) - 1;
+}
+#endif
+
diff --git a/lib/builtins/ucmpti2.c b/lib/builtins/ucmpti2.c
new file mode 100644
index 000000000..5466d2174
--- /dev/null
+++ b/lib/builtins/ucmpti2.c
@@ -0,0 +1,42 @@
+/* ===-- ucmpti2.c - Implement __ucmpti2 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __ucmpti2 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Returns:  if (a <  b) returns 0
+ *           if (a == b) returns 1
+ *           if (a >  b) returns 2
+ */
+
+si_int
+__ucmpti2(tu_int a, tu_int b)
+{
+    utwords x;
+    x.all = a;
+    utwords y;
+    y.all = b;
+    if (x.s.high < y.s.high)
+        return 0;
+    if (x.s.high > y.s.high)
+        return 2;
+    if (x.s.low < y.s.low)
+        return 0;
+    if (x.s.low > y.s.low)
+        return 2;
+    return 1;
+}
+
+#endif
diff --git a/lib/builtins/udivdi3.c b/lib/builtins/udivdi3.c
new file mode 100644
index 000000000..6c0303df3
--- /dev/null
+++ b/lib/builtins/udivdi3.c
@@ -0,0 +1,25 @@
+/* ===-- udivdi3.c - Implement __udivdi3 -----------------------------------===
+ *
+ *                    The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __udivdi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+du_int COMPILER_RT_ABI __udivmoddi4(du_int a, du_int b, du_int* rem);
+
+/* Returns: a / b */
+
+COMPILER_RT_ABI du_int
+__udivdi3(du_int a, du_int b)
+{
+    return __udivmoddi4(a, b, 0);
+}
diff --git a/lib/builtins/udivmoddi4.c b/lib/builtins/udivmoddi4.c
new file mode 100644
index 000000000..57282d5b5
--- /dev/null
+++ b/lib/builtins/udivmoddi4.c
@@ -0,0 +1,251 @@
+/* ===-- udivmoddi4.c - Implement __udivmoddi4 -----------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __udivmoddi4 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Effects: if rem != 0, *rem = a % b
+ * Returns: a / b
+ */
+
+/* Translated from Figure 3-40 of The PowerPC Compiler Writer's Guide */
+
+COMPILER_RT_ABI du_int
+__udivmoddi4(du_int a, du_int b, du_int* rem)
+{
+    const unsigned n_uword_bits = sizeof(su_int) * CHAR_BIT;
+    const unsigned n_udword_bits = sizeof(du_int) * CHAR_BIT;
+    udwords n;
+    n.all = a;
+    udwords d;
+    d.all = b;
+    udwords q;
+    udwords r;
+    unsigned sr;
+    /* special cases, X is unknown, K != 0 */
+    if (n.s.high == 0)
+    {
+        if (d.s.high == 0)
+        {
+            /* 0 X
+             * ---
+             * 0 X
+             */
+            if (rem)
+                *rem = n.s.low % d.s.low;
+            return n.s.low / d.s.low;
+        }
+        /* 0 X
+         * ---
+         * K X
+         */
+        if (rem)
+            *rem = n.s.low;
+        return 0;
+    }
+    /* n.s.high != 0 */
+    if (d.s.low == 0)
+    {
+        if (d.s.high == 0)
+        {
+            /* K X
+             * ---
+             * 0 0
+             */ 
+            if (rem)
+                *rem = n.s.high % d.s.low;
+            return n.s.high / d.s.low;
+        }
+        /* d.s.high != 0 */
+        if (n.s.low == 0)
+        {
+            /* K 0
+             * ---
+             * K 0
+             */
+            if (rem)
+            {
+                r.s.high = n.s.high % d.s.high;
+                r.s.low = 0;
+                *rem = r.all;
+            }
+            return n.s.high / d.s.high;
+        }
+        /* K K
+         * ---
+         * K 0
+         */
+        if ((d.s.high & (d.s.high - 1)) == 0)     /* if d is a power of 2 */
+        {
+            if (rem)
+            {
+                r.s.low = n.s.low;
+                r.s.high = n.s.high & (d.s.high - 1);
+                *rem = r.all;
+            }
+            return n.s.high >> __builtin_ctz(d.s.high);
+        }
+        /* K K
+         * ---
+         * K 0
+         */
+        sr = __builtin_clz(d.s.high) - __builtin_clz(n.s.high);
+        /* 0 <= sr <= n_uword_bits - 2 or sr large */
+        if (sr > n_uword_bits - 2)
+        {
+           if (rem)
+                *rem = n.all;
+            return 0;
+        }
+        ++sr;
+        /* 1 <= sr <= n_uword_bits - 1 */
+        /* q.all = n.all << (n_udword_bits - sr); */
+        q.s.low = 0;
+        q.s.high = n.s.low << (n_uword_bits - sr);
+        /* r.all = n.all >> sr; */
+        r.s.high = n.s.high >> sr;
+        r.s.low = (n.s.high << (n_uword_bits - sr)) | (n.s.low >> sr);
+    }
+    else  /* d.s.low != 0 */
+    {
+        if (d.s.high == 0)
+        {
+            /* K X
+             * ---
+             * 0 K
+             */
+            if ((d.s.low & (d.s.low - 1)) == 0)     /* if d is a power of 2 */
+            {
+                if (rem)
+                    *rem = n.s.low & (d.s.low - 1);
+                if (d.s.low == 1)
+                    return n.all;
+                sr = __builtin_ctz(d.s.low);
+                q.s.high = n.s.high >> sr;
+                q.s.low = (n.s.high << (n_uword_bits - sr)) | (n.s.low >> sr);
+                return q.all;
+            }
+            /* K X
+             * ---
+             *0 K
+             */
+            sr = 1 + n_uword_bits + __builtin_clz(d.s.low) - __builtin_clz(n.s.high);
+            /* 2 <= sr <= n_udword_bits - 1
+             * q.all = n.all << (n_udword_bits - sr);
+             * r.all = n.all >> sr;
+             * if (sr == n_uword_bits)
+             * {
+             *     q.s.low = 0;
+             *     q.s.high = n.s.low;
+             *     r.s.high = 0;
+             *     r.s.low = n.s.high;
+             * }
+             * else if (sr < n_uword_bits)  // 2 <= sr <= n_uword_bits - 1
+             * {
+             *     q.s.low = 0;
+             *     q.s.high = n.s.low << (n_uword_bits - sr);
+             *     r.s.high = n.s.high >> sr;
+             *     r.s.low = (n.s.high << (n_uword_bits - sr)) | (n.s.low >> sr);
+             * }
+             * else              // n_uword_bits + 1 <= sr <= n_udword_bits - 1
+             * {
+             *     q.s.low = n.s.low << (n_udword_bits - sr);
+             *     q.s.high = (n.s.high << (n_udword_bits - sr)) |
+             *              (n.s.low >> (sr - n_uword_bits));
+             *     r.s.high = 0;
+             *     r.s.low = n.s.high >> (sr - n_uword_bits);
+             * }
+             */
+            q.s.low =  (n.s.low << (n_udword_bits - sr)) &
+                     ((si_int)(n_uword_bits - sr) >> (n_uword_bits-1));
+            q.s.high = ((n.s.low << ( n_uword_bits - sr))                       &
+                     ((si_int)(sr - n_uword_bits - 1) >> (n_uword_bits-1))) |
+                     (((n.s.high << (n_udword_bits - sr))                     |
+                     (n.s.low >> (sr - n_uword_bits)))                        &
+                     ((si_int)(n_uword_bits - sr) >> (n_uword_bits-1)));
+            r.s.high = (n.s.high >> sr) &
+                     ((si_int)(sr - n_uword_bits) >> (n_uword_bits-1));
+            r.s.low =  ((n.s.high >> (sr - n_uword_bits))                       &
+                     ((si_int)(n_uword_bits - sr - 1) >> (n_uword_bits-1))) |
+                     (((n.s.high << (n_uword_bits - sr))                      |
+                     (n.s.low >> sr))                                         &
+                     ((si_int)(sr - n_uword_bits) >> (n_uword_bits-1)));
+        }
+        else
+        {
+            /* K X
+             * ---
+             * K K
+             */
+            sr = __builtin_clz(d.s.high) - __builtin_clz(n.s.high);
+            /* 0 <= sr <= n_uword_bits - 1 or sr large */
+            if (sr > n_uword_bits - 1)
+            {
+               if (rem)
+                    *rem = n.all;
+                return 0;
+            }
+            ++sr;
+            /* 1 <= sr <= n_uword_bits */
+            /*  q.all = n.all << (n_udword_bits - sr); */
+            q.s.low = 0;
+            q.s.high = n.s.low << (n_uword_bits - sr);
+            /* r.all = n.all >> sr;
+             * if (sr < n_uword_bits)
+             * {
+             *     r.s.high = n.s.high >> sr;
+             *     r.s.low = (n.s.high << (n_uword_bits - sr)) | (n.s.low >> sr);
+             * }
+             * else
+             * {
+             *     r.s.high = 0;
+             *     r.s.low = n.s.high;
+             * }
+             */
+            r.s.high = (n.s.high >> sr) &
+                     ((si_int)(sr - n_uword_bits) >> (n_uword_bits-1));
+            r.s.low = (n.s.high << (n_uword_bits - sr)) |
+                    ((n.s.low >> sr)                  &
+                    ((si_int)(sr - n_uword_bits) >> (n_uword_bits-1)));
+        }
+    }
+    /* Not a special case
+     * q and r are initialized with:
+     * q.all = n.all << (n_udword_bits - sr);
+     * r.all = n.all >> sr;
+     * 1 <= sr <= n_udword_bits - 1
+     */
+    su_int carry = 0;
+    for (; sr > 0; --sr)
+    {
+        /* r:q = ((r:q)  << 1) | carry */
+        r.s.high = (r.s.high << 1) | (r.s.low  >> (n_uword_bits - 1));
+        r.s.low  = (r.s.low  << 1) | (q.s.high >> (n_uword_bits - 1));
+        q.s.high = (q.s.high << 1) | (q.s.low  >> (n_uword_bits - 1));
+        q.s.low  = (q.s.low  << 1) | carry;
+        /* carry = 0;
+         * if (r.all >= d.all)
+         * {
+         *      r.all -= d.all;
+         *      carry = 1;
+         * }
+         */
+        const di_int s = (di_int)(d.all - r.all - 1) >> (n_udword_bits - 1);
+        carry = s & 1;
+        r.all -= d.all & s;
+    }
+    q.all = (q.all << 1) | carry;
+    if (rem)
+        *rem = r.all;
+    return q.all;
+}
diff --git a/lib/builtins/udivmodsi4.c b/lib/builtins/udivmodsi4.c
new file mode 100644
index 000000000..5b49089fc
--- /dev/null
+++ b/lib/builtins/udivmodsi4.c
@@ -0,0 +1,30 @@
+/*===-- udivmodsi4.c - Implement __udivmodsi4 ------------------------------===
+ *
+ *                    The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __udivmodsi4 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+extern su_int COMPILER_RT_ABI __udivsi3(su_int n, su_int d);
+
+
+/* Returns: a / b, *rem = a % b  */
+
+COMPILER_RT_ABI su_int
+__udivmodsi4(su_int a, su_int b, su_int* rem)
+{
+  si_int d = __udivsi3(a,b);
+  *rem = a - (d*b);
+  return d;
+}
+
+
diff --git a/lib/builtins/udivmodti4.c b/lib/builtins/udivmodti4.c
new file mode 100644
index 000000000..f619c7495
--- /dev/null
+++ b/lib/builtins/udivmodti4.c
@@ -0,0 +1,256 @@
+/* ===-- udivmodti4.c - Implement __udivmodti4 -----------------------------===
+ *
+ *                    The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __udivmodti4 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */ 
+
+#include "int_lib.h"
+
+#if __x86_64
+
+/* Effects: if rem != 0, *rem = a % b 
+ * Returns: a / b 
+ */
+
+/* Translated from Figure 3-40 of The PowerPC Compiler Writer's Guide */
+
+tu_int
+__udivmodti4(tu_int a, tu_int b, tu_int* rem)
+{
+    const unsigned n_udword_bits = sizeof(du_int) * CHAR_BIT;
+    const unsigned n_utword_bits = sizeof(tu_int) * CHAR_BIT;
+    utwords n;
+    n.all = a;
+    utwords d;
+    d.all = b;
+    utwords q;
+    utwords r;
+    unsigned sr;
+    /* special cases, X is unknown, K != 0 */
+    if (n.s.high == 0)
+    {
+        if (d.s.high == 0)
+        {
+            /* 0 X
+             * ---
+             * 0 X
+             */
+            if (rem)
+                *rem = n.s.low % d.s.low;
+            return n.s.low / d.s.low;
+        }
+        /* 0 X
+         * ---
+         * K X
+         */
+        if (rem)
+            *rem = n.s.low;
+        return 0;
+    }
+    /* n.s.high != 0 */
+    if (d.s.low == 0)
+    {
+        if (d.s.high == 0)
+        {
+            /* K X
+             * ---
+             * 0 0
+             */
+            if (rem)
+                *rem = n.s.high % d.s.low;
+            return n.s.high / d.s.low;
+        }
+        /* d.s.high != 0 */
+        if (n.s.low == 0)
+        {
+            /* K 0
+             * ---
+             * K 0
+             */
+            if (rem)
+            {
+                r.s.high = n.s.high % d.s.high;
+                r.s.low = 0;
+                *rem = r.all;
+            }
+            return n.s.high / d.s.high;
+        }
+        /* K K
+         * ---
+         * K 0
+         */
+        if ((d.s.high & (d.s.high - 1)) == 0)     /* if d is a power of 2 */
+        {
+            if (rem)
+            {
+                r.s.low = n.s.low;
+                r.s.high = n.s.high & (d.s.high - 1);
+                *rem = r.all;
+            }
+            return n.s.high >> __builtin_ctzll(d.s.high);
+        }
+        /* K K
+         * ---
+         * K 0
+         */
+        sr = __builtin_clzll(d.s.high) - __builtin_clzll(n.s.high);
+        /* 0 <= sr <= n_udword_bits - 2 or sr large */
+        if (sr > n_udword_bits - 2)
+        {
+           if (rem)
+                *rem = n.all;
+            return 0;
+        }
+        ++sr;
+        /* 1 <= sr <= n_udword_bits - 1 */
+        /* q.all = n.all << (n_utword_bits - sr); */
+        q.s.low = 0;
+        q.s.high = n.s.low << (n_udword_bits - sr);
+        /* r.all = n.all >> sr; */
+        r.s.high = n.s.high >> sr;
+        r.s.low = (n.s.high << (n_udword_bits - sr)) | (n.s.low >> sr);
+    }
+    else  /* d.s.low != 0 */
+    {
+        if (d.s.high == 0)
+        {
+            /* K X
+             * ---
+             * 0 K
+             */
+            if ((d.s.low & (d.s.low - 1)) == 0)     /* if d is a power of 2 */
+            {
+                if (rem)
+                    *rem = n.s.low & (d.s.low - 1);
+                if (d.s.low == 1)
+                    return n.all;
+                sr = __builtin_ctzll(d.s.low);
+                q.s.high = n.s.high >> sr;
+                q.s.low = (n.s.high << (n_udword_bits - sr)) | (n.s.low >> sr);
+                return q.all;
+            }
+            /* K X
+             * ---
+             * 0 K
+             */
+            sr = 1 + n_udword_bits + __builtin_clzll(d.s.low)
+                                   - __builtin_clzll(n.s.high);
+            /* 2 <= sr <= n_utword_bits - 1
+             * q.all = n.all << (n_utword_bits - sr);
+             * r.all = n.all >> sr;
+             * if (sr == n_udword_bits)
+             * {
+             *     q.s.low = 0;
+             *     q.s.high = n.s.low;
+             *     r.s.high = 0;
+             *     r.s.low = n.s.high;
+             * }
+             * else if (sr < n_udword_bits)  // 2 <= sr <= n_udword_bits - 1
+             * {
+             *     q.s.low = 0;
+             *     q.s.high = n.s.low << (n_udword_bits - sr);
+             *     r.s.high = n.s.high >> sr;
+             *     r.s.low = (n.s.high << (n_udword_bits - sr)) | (n.s.low >> sr);
+             * }
+             * else              // n_udword_bits + 1 <= sr <= n_utword_bits - 1
+             * {
+             *     q.s.low = n.s.low << (n_utword_bits - sr);
+             *     q.s.high = (n.s.high << (n_utword_bits - sr)) |
+             *              (n.s.low >> (sr - n_udword_bits));
+             *     r.s.high = 0;
+             *     r.s.low = n.s.high >> (sr - n_udword_bits);
+             * }
+             */
+            q.s.low =  (n.s.low << (n_utword_bits - sr)) &
+                     ((di_int)(int)(n_udword_bits - sr) >> (n_udword_bits-1));
+            q.s.high = ((n.s.low << ( n_udword_bits - sr))                        &
+                     ((di_int)(int)(sr - n_udword_bits - 1) >> (n_udword_bits-1))) |
+                     (((n.s.high << (n_utword_bits - sr))                       |
+                     (n.s.low >> (sr - n_udword_bits)))                         &
+                     ((di_int)(int)(n_udword_bits - sr) >> (n_udword_bits-1)));
+            r.s.high = (n.s.high >> sr) &
+                     ((di_int)(int)(sr - n_udword_bits) >> (n_udword_bits-1));
+            r.s.low =  ((n.s.high >> (sr - n_udword_bits))                        &
+                     ((di_int)(int)(n_udword_bits - sr - 1) >> (n_udword_bits-1))) |
+                     (((n.s.high << (n_udword_bits - sr))                       |
+                     (n.s.low >> sr))                                           &
+                     ((di_int)(int)(sr - n_udword_bits) >> (n_udword_bits-1)));
+        }
+        else
+        {
+            /* K X
+             * ---
+             * K K
+             */
+            sr = __builtin_clzll(d.s.high) - __builtin_clzll(n.s.high);
+            /*0 <= sr <= n_udword_bits - 1 or sr large */
+            if (sr > n_udword_bits - 1)
+            {
+               if (rem)
+                    *rem = n.all;
+                return 0;
+            }
+            ++sr;
+            /* 1 <= sr <= n_udword_bits */
+            /* q.all = n.all << (n_utword_bits - sr); */
+            q.s.low = 0;
+            q.s.high = n.s.low << (n_udword_bits - sr);
+            /* r.all = n.all >> sr;
+             * if (sr < n_udword_bits)
+             * {
+             *     r.s.high = n.s.high >> sr;
+             *     r.s.low = (n.s.high << (n_udword_bits - sr)) | (n.s.low >> sr);
+             * }
+             * else
+             * {
+             *     r.s.high = 0;
+             *     r.s.low = n.s.high;
+             * }
+             */
+            r.s.high = (n.s.high >> sr) &
+                     ((di_int)(int)(sr - n_udword_bits) >> (n_udword_bits-1));
+            r.s.low = (n.s.high << (n_udword_bits - sr)) |
+                    ((n.s.low >> sr)                   &
+                    ((di_int)(int)(sr - n_udword_bits) >> (n_udword_bits-1)));
+        }
+    }
+    /* Not a special case
+     * q and r are initialized with:
+     * q.all = n.all << (n_utword_bits - sr);
+     * r.all = n.all >> sr;
+     * 1 <= sr <= n_utword_bits - 1
+     */
+    su_int carry = 0;
+    for (; sr > 0; --sr)
+    {
+        /* r:q = ((r:q)  << 1) | carry */
+        r.s.high = (r.s.high << 1) | (r.s.low  >> (n_udword_bits - 1));
+        r.s.low  = (r.s.low  << 1) | (q.s.high >> (n_udword_bits - 1));
+        q.s.high = (q.s.high << 1) | (q.s.low  >> (n_udword_bits - 1));
+        q.s.low  = (q.s.low  << 1) | carry;
+        /* carry = 0;
+         * if (r.all >= d.all)
+         * {
+         *     r.all -= d.all;
+         *      carry = 1;
+         * }
+         */
+        const ti_int s = (ti_int)(d.all - r.all - 1) >> (n_utword_bits - 1);
+        carry = s & 1;
+        r.all -= d.all & s;
+    }
+    q.all = (q.all << 1) | carry;
+    if (rem)
+        *rem = r.all;
+    return q.all;
+}
+
+#endif /* __x86_64 */
diff --git a/lib/builtins/udivsi3.c b/lib/builtins/udivsi3.c
new file mode 100644
index 000000000..5d0140cc3
--- /dev/null
+++ b/lib/builtins/udivsi3.c
@@ -0,0 +1,66 @@
+/* ===-- udivsi3.c - Implement __udivsi3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __udivsi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a / b */
+
+/* Translated from Figure 3-40 of The PowerPC Compiler Writer's Guide */
+
+ARM_EABI_FNALIAS(uidiv, udivsi3)
+
+/* This function should not call __divsi3! */
+COMPILER_RT_ABI su_int
+__udivsi3(su_int n, su_int d)
+{
+    const unsigned n_uword_bits = sizeof(su_int) * CHAR_BIT;
+    su_int q;
+    su_int r;
+    unsigned sr;
+    /* special cases */
+    if (d == 0)
+        return 0; /* ?! */
+    if (n == 0)
+        return 0;
+    sr = __builtin_clz(d) - __builtin_clz(n);
+    /* 0 <= sr <= n_uword_bits - 1 or sr large */
+    if (sr > n_uword_bits - 1)  /* d > r */
+        return 0;
+    if (sr == n_uword_bits - 1)  /* d == 1 */
+        return n;
+    ++sr;
+    /* 1 <= sr <= n_uword_bits - 1 */
+    /* Not a special case */
+    q = n << (n_uword_bits - sr);
+    r = n >> sr;
+    su_int carry = 0;
+    for (; sr > 0; --sr)
+    {
+        /* r:q = ((r:q)  << 1) | carry */
+        r = (r << 1) | (q >> (n_uword_bits - 1));
+        q = (q << 1) | carry;
+        /* carry = 0;
+         * if (r.all >= d.all)
+         * {
+         *      r.all -= d.all;
+         *      carry = 1;
+         * }
+         */
+        const si_int s = (si_int)(d - r - 1) >> (n_uword_bits - 1);
+        carry = s & 1;
+        r -= d & s;
+    }
+    q = (q << 1) | carry;
+    return q;
+}
diff --git a/lib/builtins/udivti3.c b/lib/builtins/udivti3.c
new file mode 100644
index 000000000..d9e1bb46f
--- /dev/null
+++ b/lib/builtins/udivti3.c
@@ -0,0 +1,29 @@
+/* ===-- udivti3.c - Implement __udivti3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __udivti3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+tu_int __udivmodti4(tu_int a, tu_int b, tu_int* rem);
+
+/* Returns: a / b */
+
+tu_int
+__udivti3(tu_int a, tu_int b)
+{
+    return __udivmodti4(a, b, 0);
+}
+
+#endif /* __x86_64 */
diff --git a/lib/builtins/umoddi3.c b/lib/builtins/umoddi3.c
new file mode 100644
index 000000000..3541ab6e9
--- /dev/null
+++ b/lib/builtins/umoddi3.c
@@ -0,0 +1,27 @@
+/* ===-- umoddi3.c - Implement __umoddi3 -----------------------------------===
+ *
+ *                    The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __umoddi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+du_int COMPILER_RT_ABI __udivmoddi4(du_int a, du_int b, du_int* rem);
+
+/* Returns: a % b */
+
+COMPILER_RT_ABI du_int
+__umoddi3(du_int a, du_int b)
+{
+    du_int r;
+    __udivmoddi4(a, b, &r);
+    return r;
+}
diff --git a/lib/builtins/umodsi3.c b/lib/builtins/umodsi3.c
new file mode 100644
index 000000000..aae741d89
--- /dev/null
+++ b/lib/builtins/umodsi3.c
@@ -0,0 +1,25 @@
+/* ===-- umodsi3.c - Implement __umodsi3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __umodsi3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+/* Returns: a % b */
+
+su_int COMPILER_RT_ABI __udivsi3(su_int a, su_int b);
+
+COMPILER_RT_ABI su_int
+__umodsi3(su_int a, su_int b)
+{
+    return a - __udivsi3(a, b) * b;
+}
diff --git a/lib/builtins/umodti3.c b/lib/builtins/umodti3.c
new file mode 100644
index 000000000..8ebe7f0dc
--- /dev/null
+++ b/lib/builtins/umodti3.c
@@ -0,0 +1,31 @@
+/* ===-- umodti3.c - Implement __umodti3 -----------------------------------===
+ *
+ *                     The LLVM Compiler Infrastructure
+ *
+ * This file is dual licensed under the MIT and the University of Illinois Open
+ * Source Licenses. See LICENSE.TXT for details.
+ *
+ * ===----------------------------------------------------------------------===
+ *
+ * This file implements __umodti3 for the compiler_rt library.
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#include "int_lib.h"
+
+#if __x86_64
+
+tu_int __udivmodti4(tu_int a, tu_int b, tu_int* rem);
+
+/* Returns: a % b */
+
+tu_int
+__umodti3(tu_int a, tu_int b)
+{
+    tu_int r;
+    __udivmodti4(a, b, &r);
+    return r;
+}
+
+#endif
diff --git a/lib/builtins/x86_64/Makefile.mk b/lib/builtins/x86_64/Makefile.mk
new file mode 100644
index 000000000..83848dddd
--- /dev/null
+++ b/lib/builtins/x86_64/Makefile.mk
@@ -0,0 +1,20 @@
+#===- lib/builtins/x86_64/Makefile.mk ----------------------*- Makefile -*--===#
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+#===------------------------------------------------------------------------===#
+
+ModuleName := builtins
+SubDirs := 
+OnlyArchs := x86_64 x86_64h
+
+AsmSources := $(foreach file,$(wildcard $(Dir)/*.S),$(notdir $(file)))
+Sources := $(foreach file,$(wildcard $(Dir)/*.c),$(notdir $(file)))
+ObjNames := $(Sources:%.c=%.o) $(AsmSources:%.S=%.o)
+Implementation := Optimized
+
+# FIXME: use automatic dependencies?
+Dependencies := $(wildcard lib/*.h $(Dir)/*.h)
diff --git a/lib/builtins/x86_64/floatdidf.c b/lib/builtins/x86_64/floatdidf.c
new file mode 100644
index 000000000..388404e5e
--- /dev/null
+++ b/lib/builtins/x86_64/floatdidf.c
@@ -0,0 +1,16 @@
+/* This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ */
+
+/* double __floatdidf(di_int a); */
+
+#ifdef __x86_64__
+
+#include "../int_lib.h"
+
+double __floatdidf(int64_t a)
+{
+	return (double)a;
+}
+
+#endif /* __x86_64__ */
diff --git a/lib/builtins/x86_64/floatdisf.c b/lib/builtins/x86_64/floatdisf.c
new file mode 100644
index 000000000..96c3728e9
--- /dev/null
+++ b/lib/builtins/x86_64/floatdisf.c
@@ -0,0 +1,14 @@
+/* This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ */
+
+#ifdef __x86_64__
+
+#include "../int_lib.h"
+
+float __floatdisf(int64_t a)
+{
+	return (float)a;
+}
+
+#endif /* __x86_64__ */
diff --git a/lib/builtins/x86_64/floatdixf.c b/lib/builtins/x86_64/floatdixf.c
new file mode 100644
index 000000000..c01193a82
--- /dev/null
+++ b/lib/builtins/x86_64/floatdixf.c
@@ -0,0 +1,16 @@
+/* This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ */
+
+/* long double __floatdixf(di_int a); */
+
+#ifdef __x86_64__
+
+#include "../int_lib.h"
+
+long double __floatdixf(int64_t a)
+{
+	return (long double)a;
+}
+
+#endif /* __i386__ */
diff --git a/lib/builtins/x86_64/floatundidf.S b/lib/builtins/x86_64/floatundidf.S
new file mode 100644
index 000000000..1df3d7479
--- /dev/null
+++ b/lib/builtins/x86_64/floatundidf.S
@@ -0,0 +1,44 @@
+//===-- floatundidf.S - Implement __floatundidf for x86_64 ----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements __floatundidf for the compiler_rt library.
+//
+//===----------------------------------------------------------------------===//
+
+#include "../assembly.h"
+
+// double __floatundidf(du_int a);
+
+#ifdef __x86_64__
+
+#ifndef __ELF__
+.const
+#endif
+.align 4
+twop52: .quad 0x4330000000000000
+twop84_plus_twop52:
+		.quad 0x4530000000100000
+twop84: .quad 0x4530000000000000
+
+#define REL_ADDR(_a)	(_a)(%rip)
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__floatundidf)
+	movd	%edi,							%xmm0 // low 32 bits of a
+	shrq	$32,							%rdi  // high 32 bits of a
+	orq		REL_ADDR(twop84),				%rdi  // 0x1p84 + a_hi (no rounding occurs)
+	orpd	REL_ADDR(twop52),				%xmm0 // 0x1p52 + a_lo (no rounding occurs)
+	movd	%rdi,							%xmm1
+	subsd	REL_ADDR(twop84_plus_twop52),	%xmm1 // a_hi - 0x1p52 (no rounding occurs)
+	addsd	%xmm1,							%xmm0 // a_hi + a_lo   (round happens here)
+	ret
+END_COMPILERRT_FUNCTION(__floatundidf)
+
+#endif // __x86_64__
diff --git a/lib/builtins/x86_64/floatundisf.S b/lib/builtins/x86_64/floatundisf.S
new file mode 100644
index 000000000..7dd5c329b
--- /dev/null
+++ b/lib/builtins/x86_64/floatundisf.S
@@ -0,0 +1,34 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// float __floatundisf(du_int a);
+
+#ifdef __x86_64__
+
+#ifndef __ELF__
+.literal4
+#endif
+two: .single 2.0
+
+#define REL_ADDR(_a)	(_a)(%rip)
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__floatundisf)
+	movq		$1,			%rsi
+	testq		%rdi,		%rdi
+	js			1f
+	cvtsi2ssq	%rdi,		%xmm0
+	ret
+	
+1:	andq		%rdi,		%rsi
+	shrq		%rdi
+	orq			%rsi,		%rdi
+	cvtsi2ssq	%rdi,		%xmm0
+	mulss	REL_ADDR(two),	%xmm0
+	ret
+END_COMPILERRT_FUNCTION(__floatundisf)
+
+#endif // __x86_64__
diff --git a/lib/builtins/x86_64/floatundixf.S b/lib/builtins/x86_64/floatundixf.S
new file mode 100644
index 000000000..b05954a0b
--- /dev/null
+++ b/lib/builtins/x86_64/floatundixf.S
@@ -0,0 +1,64 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// long double __floatundixf(du_int a);
+
+#ifdef __x86_64__
+
+#ifndef __ELF__
+.const
+#endif
+.align 4
+twop64: .quad 0x43f0000000000000
+
+#define REL_ADDR(_a)	(_a)(%rip)
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__floatundixf)
+	movq	%rdi,	 -8(%rsp)
+	fildq	-8(%rsp)
+	test	%rdi,		%rdi
+	js		1f
+	ret
+1:	faddl	REL_ADDR(twop64)
+	ret
+END_COMPILERRT_FUNCTION(__floatundixf)
+
+#endif // __x86_64__
+
+
+/* Branch-free implementation is ever so slightly slower, but more beautiful.
+   It is likely superior for inlining, so I kept it around for future reference.
+
+#ifdef __x86_64__
+
+.const
+.align 4
+twop52: .quad 0x4330000000000000
+twop84_plus_twop52_neg:
+		.quad 0xc530000000100000
+twop84: .quad 0x4530000000000000
+
+#define REL_ADDR(_a)	(_a)(%rip)
+
+.text
+.align 4
+DEFINE_COMPILERRT_FUNCTION(__floatundixf)
+	movl	%edi,				%esi			// low 32 bits of input
+	shrq	$32,				%rdi			// hi 32 bits of input
+	orq		REL_ADDR(twop84),	%rdi			// 2^84 + hi (as a double)
+	orq		REL_ADDR(twop52),	%rsi			// 2^52 + lo (as a double)
+	movq	%rdi,			 -8(%rsp)
+	movq	%rsi,			-16(%rsp)
+	fldl	REL_ADDR(twop84_plus_twop52_neg)	
+	faddl	-8(%rsp)	// hi - 2^52 (as double extended, no rounding occurs)
+	faddl	-16(%rsp)	// hi + lo (as double extended)
+	ret
+END_COMPILERRT_FUNCTION(__floatundixf)
+
+#endif // __x86_64__
+
+*/