vendor/compiler-rt/compiler-rt-trunk-r300422

21 files changed, 659 insertions, 77 deletions

diff --git a/lib/builtins/CMakeLists.txt b/lib/builtins/CMakeLists.txt
index 3cf78616ac07..161487e703d7 100644
--- a/lib/builtins/CMakeLists.txt
+++ b/lib/builtins/CMakeLists.txt
@@ -71,6 +71,7 @@ set(GENERIC_SOURCES
   extendsfdf2.c
   extendhfsf2.c
   ffsdi2.c
+  ffssi2.c
   ffsti2.c
   fixdfdi.c
   fixdfsi.c
@@ -132,6 +133,7 @@ set(GENERIC_SOURCES
   negvdi2.c
   negvsi2.c
   negvti2.c
+  os_version_check.c
   paritydi2.c
   paritysi2.c
   parityti2.c
@@ -174,15 +176,6 @@ if(COMPILER_RT_HAS_ATOMIC_KEYWORD AND NOT COMPILER_RT_EXCLUDE_ATOMIC_BUILTIN)
     atomic.c)
 endif()
 
-set(MSVC_SOURCES
- divsc3.c
- divdc3.c
- divxc3.c
- mulsc3.c
- muldc3.c
- mulxc3.c)
-
-
 if(APPLE)
   set(GENERIC_SOURCES
     ${GENERIC_SOURCES}
@@ -262,9 +255,9 @@ else () # MSVC
       x86_64/floatdidf.c
       x86_64/floatdisf.c
       x86_64/floatdixf.c
-      ${MSVC_SOURCES})
+      ${GENERIC_SOURCES})
   set(x86_64h_SOURCES ${x86_64_SOURCES})
-  set(i386_SOURCES ${MSVC_SOURCES})
+  set(i386_SOURCES ${GENERIC_SOURCES})
   set(i686_SOURCES ${i386_SOURCES})
 endif () # if (NOT MSVC)
 
@@ -302,6 +295,13 @@ set(arm_SOURCES
   arm/umodsi3.S
   ${GENERIC_SOURCES})
 
+set(thumb1_SOURCES
+  arm/divsi3.S
+  arm/udivsi3.S
+  arm/comparesf2.S
+  arm/addsf3.S
+  ${GENERIC_SOURCES})
+
 set(arm_EABI_SOURCES
   arm/aeabi_cdcmp.S
   arm/aeabi_cdcmpeq_check_nan.c
@@ -320,6 +320,7 @@ set(arm_EABI_SOURCES
   arm/aeabi_memset.S
   arm/aeabi_uidivmod.S
   arm/aeabi_uldivmod.S)
+
 set(arm_Thumb1_JT_SOURCES
   arm/switch16.S
   arm/switch32.S
@@ -401,6 +402,10 @@ elseif(NOT WIN32)
     ${arm_SOURCES}
     ${arm_EABI_SOURCES}
     ${arm_Thumb1_SOURCES})
+
+  set(thumb1_SOURCES
+    ${thumb1_SOURCES}
+    ${arm_EABI_SOURCES})
 endif()
 
 set(aarch64_SOURCES
@@ -431,7 +436,7 @@ set(armv7k_SOURCES ${arm_SOURCES})
 set(arm64_SOURCES ${aarch64_SOURCES})
 
 # macho_embedded archs
-set(armv6m_SOURCES ${GENERIC_SOURCES})
+set(armv6m_SOURCES ${thumb1_SOURCES})
 set(armv7m_SOURCES ${arm_SOURCES})
 set(armv7em_SOURCES ${arm_SOURCES})
 
@@ -486,7 +491,7 @@ else ()
       # Needed for clear_cache on debug mode, due to r7's usage in inline asm.
       # Release mode already sets it via -O2/3, Debug mode doesn't.
       if (${arch} STREQUAL "armhf")
-        list(APPEND BUILTIN_CFLAGS -fomit-frame-pointer)
+        list(APPEND BUILTIN_CFLAGS -fomit-frame-pointer -DCOMPILER_RT_ARMHF_TARGET)
       endif()
 
       add_compiler_rt_runtime(clang_rt.builtins
diff --git a/lib/builtins/README.txt b/lib/builtins/README.txt
index ad36e4e5279a..b3d083614ee0 100644
--- a/lib/builtins/README.txt
+++ b/lib/builtins/README.txt
@@ -45,6 +45,7 @@ si_int __ctzsi2(si_int a);  // count trailing zeros
 si_int __ctzdi2(di_int a);  // count trailing zeros
 si_int __ctzti2(ti_int a);  // count trailing zeros
 
+si_int __ffssi2(si_int a);  // find least significant 1 bit
 si_int __ffsdi2(di_int a);  // find least significant 1 bit
 si_int __ffsti2(ti_int a);  // find least significant 1 bit
 
diff --git a/lib/builtins/arm/addsf3.S b/lib/builtins/arm/addsf3.S
new file mode 100644
index 000000000000..362b5c147ea6
--- /dev/null
+++ b/lib/builtins/arm/addsf3.S
@@ -0,0 +1,277 @@
+/*===-- addsf3.S - Adds two single precision floating pointer numbers-----===//
+ *
+ *                     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 __addsf3 (single precision floating pointer number
+ * addition with the IEEE-754 default rounding (to nearest, ties to even)
+ * function for the ARM Thumb1 ISA.
+ *
+ *===----------------------------------------------------------------------===*/
+
+#include "../assembly.h"
+#define significandBits 23
+#define typeWidth 32
+
+	.syntax unified
+	.text
+  .thumb
+  .p2align 2
+
+DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_fadd, __addsf3)
+
+DEFINE_COMPILERRT_THUMB_FUNCTION(__addsf3)
+  push {r4, r5, r6, r7, lr}
+  // Get the absolute value of a and b.
+  lsls r2, r0, #1
+  lsls r3, r1, #1
+  lsrs r2, r2, #1  /* aAbs */
+  beq  LOCAL_LABEL(a_zero_nan_inf)
+  lsrs r3, r3, #1  /* bAbs */
+  beq  LOCAL_LABEL(zero_nan_inf)
+
+  // Detect if a or b is infinity or Nan.
+  lsrs r6, r2, #(significandBits)
+  lsrs r7, r3, #(significandBits)
+  cmp  r6, #0xFF
+  beq  LOCAL_LABEL(zero_nan_inf)
+  cmp  r7, #0xFF
+  beq  LOCAL_LABEL(zero_nan_inf)
+
+  // Swap Rep and Abs so that a and aAbs has the larger absolute value.
+  cmp r2, r3
+  bhs LOCAL_LABEL(no_swap)
+  movs r4, r0
+  movs r5, r2
+  movs r0, r1
+  movs r2, r3
+  movs r1, r4
+  movs r3, r5
+LOCAL_LABEL(no_swap):
+
+  // Get the significands and shift them to give us round, guard and sticky.
+  lsls r4, r0, #(typeWidth - significandBits)
+  lsrs r4, r4, #(typeWidth - significandBits - 3) /* aSignificand << 3 */
+  lsls r5, r1, #(typeWidth - significandBits)
+  lsrs r5, r5, #(typeWidth - significandBits - 3) /* bSignificand << 3 */
+
+  // Get the implicitBit.
+  movs r6, #1
+  lsls r6, r6, #(significandBits + 3)
+
+  // Get aExponent and set implicit bit if necessary.
+  lsrs r2, r2, #(significandBits)
+  beq LOCAL_LABEL(a_done_implicit_bit)
+  orrs r4, r6
+LOCAL_LABEL(a_done_implicit_bit):
+
+  // Get bExponent and set implicit bit if necessary.
+  lsrs r3, r3, #(significandBits)
+  beq LOCAL_LABEL(b_done_implicit_bit)
+  orrs r5, r6
+LOCAL_LABEL(b_done_implicit_bit):
+
+  // Get the difference in exponents.
+  subs r6, r2, r3
+  beq LOCAL_LABEL(done_align)
+
+  // If b is denormal, then a must be normal as align > 0, and we only need to
+  // right shift bSignificand by (align - 1) bits.
+  cmp  r3, #0
+  bne  1f
+  subs r6, r6, #1
+1:
+
+  // No longer needs bExponent. r3 is dead here.
+  // Set sticky bits of b: sticky = bSignificand << (typeWidth - align).
+  movs r3, #(typeWidth)
+  subs r3, r3, r6
+  movs r7, r5
+  lsls r7, r3
+  beq 1f
+  movs r7, #1
+1:
+
+  // bSignificand = bSignificand >> align | sticky;
+  lsrs r5, r6
+  orrs r5, r7
+  bne LOCAL_LABEL(done_align)
+  movs r5, #1 //  sticky; b is known to be non-zero.
+
+LOCAL_LABEL(done_align):
+  // isSubtraction = (aRep ^ bRep) >> 31;
+  movs r7, r0
+  eors r7, r1
+  lsrs r7, #31
+  bne LOCAL_LABEL(do_substraction)
+
+  // Same sign, do Addition.
+
+  // aSignificand += bSignificand;
+  adds r4, r4, r5
+
+  // Check carry bit.
+  movs r6, #1
+  lsls r6, r6, #(significandBits + 3 + 1)
+  movs r7, r4
+  ands r7, r6
+  beq LOCAL_LABEL(form_result)
+  // If the addition carried up, we need to right-shift the result and
+  // adjust the exponent.
+  movs r7, r4
+  movs r6, #1
+  ands r7, r6 // sticky = aSignificand & 1;
+  lsrs r4, #1
+  orrs r4, r7  // result Significand
+  adds r2, #1  // result Exponent
+  // If we have overflowed the type, return +/- infinity.
+  cmp  r2, 0xFF
+  beq  LOCAL_LABEL(ret_inf)
+
+LOCAL_LABEL(form_result):
+  // Shift the sign, exponent and significand into place.
+  lsrs r0, #(typeWidth - 1)
+  lsls r0, #(typeWidth - 1) // Get Sign.
+  lsls r2, #(significandBits)
+  orrs r0, r2
+  movs r1, r4
+  lsls r4, #(typeWidth - significandBits - 3)
+  lsrs r4, #(typeWidth - significandBits)
+  orrs r0, r4
+
+  // Final rounding.  The result may overflow to infinity, but that is the
+  // correct result in that case.
+  // roundGuardSticky = aSignificand & 0x7;
+  movs r2, #0x7
+  ands r1, r2
+  // if (roundGuardSticky > 0x4) result++;
+
+  cmp r1, #0x4
+  blt LOCAL_LABEL(done_round)
+  beq 1f
+  adds r0, #1
+  pop {r4, r5, r6, r7, pc}
+1:
+
+  // if (roundGuardSticky == 0x4) result += result & 1;
+  movs r1, r0
+  lsrs r1, #1
+  bcc  LOCAL_LABEL(done_round)
+  adds r0, r0, #1
+LOCAL_LABEL(done_round):
+  pop {r4, r5, r6, r7, pc}
+
+LOCAL_LABEL(do_substraction):
+  subs r4, r4, r5 // aSignificand -= bSignificand;
+  beq  LOCAL_LABEL(ret_zero)
+  movs r6, r4
+  cmp  r2, 0
+  beq  LOCAL_LABEL(form_result) // if a's exp is 0, no need to normalize.
+  // If partial cancellation occured, we need to left-shift the result
+  // and adjust the exponent:
+  lsrs r6, r6, #(significandBits + 3)
+  bne LOCAL_LABEL(form_result)
+
+  push {r0, r1, r2, r3}
+  movs r0, r4
+  bl   __clzsi2
+  movs r5, r0
+  pop {r0, r1, r2, r3}
+  // shift = rep_clz(aSignificand) - rep_clz(implicitBit << 3);
+  subs r5, r5, #(typeWidth - significandBits - 3 - 1)
+  // aSignificand <<= shift; aExponent -= shift;
+  lsls r4, r5
+  subs  r2, r2, r5
+  bgt LOCAL_LABEL(form_result)
+
+  // Do normalization if aExponent <= 0.
+  movs r6, #1
+  subs r6, r6, r2 // 1 - aExponent;
+  movs r2, #0 // aExponent = 0;
+  movs r3, #(typeWidth) // bExponent is dead.
+  subs r3, r3, r6
+  movs r7, r4
+  lsls r7, r3  // stickyBit = (bool)(aSignificant << (typeWidth - align))
+  beq 1f
+  movs r7, #1
+1:
+  lsrs r4, r6 /* aSignificand >> shift */
+  orrs r4, r7
+  b LOCAL_LABEL(form_result)
+
+LOCAL_LABEL(ret_zero):
+  movs r0, #0
+  pop {r4, r5, r6, r7, pc}
+
+
+LOCAL_LABEL(a_zero_nan_inf):
+  lsrs r3, r3, #1
+
+LOCAL_LABEL(zero_nan_inf):
+  // Here  r2 has aAbs, r3 has bAbs
+  movs r4, #0xFF
+  lsls r4, r4, #(significandBits) // Make +inf.
+
+  cmp r2, r4
+  bhi LOCAL_LABEL(a_is_nan)
+  cmp r3, r4
+  bhi LOCAL_LABEL(b_is_nan)
+
+  cmp r2, r4
+  bne LOCAL_LABEL(a_is_rational)
+  // aAbs is INF.
+  eors r1, r0 // aRep ^ bRep.
+  movs r6, #1
+  lsls r6, r6, #(typeWidth - 1) // get sign mask.
+  cmp r1, r6 // if they only differ on sign bit, it's -INF + INF
+  beq LOCAL_LABEL(a_is_nan)
+  pop {r4, r5, r6, r7, pc}
+
+LOCAL_LABEL(a_is_rational):
+  cmp r3, r4
+  bne LOCAL_LABEL(b_is_rational)
+  movs r0, r1
+  pop {r4, r5, r6, r7, pc}
+
+LOCAL_LABEL(b_is_rational):
+  // either a or b or both are zero.
+  adds r4, r2, r3
+  beq  LOCAL_LABEL(both_zero)
+  cmp r2, #0 // is absA 0 ?
+  beq LOCAL_LABEL(ret_b)
+  pop {r4, r5, r6, r7, pc}
+
+LOCAL_LABEL(both_zero):
+  ands r0, r1 // +0 + -0 = +0
+  pop {r4, r5, r6, r7, pc}
+
+LOCAL_LABEL(ret_b):
+  movs r0, r1
+
+LOCAL_LABEL(ret):
+  pop {r4, r5, r6, r7, pc}
+
+LOCAL_LABEL(b_is_nan):
+  movs r0, r1
+LOCAL_LABEL(a_is_nan):
+  movs r1, #1
+  lsls r1, r1, #(significandBits -1) // r1 is quiet bit.
+  orrs r0, r1
+  pop {r4, r5, r6, r7, pc}
+
+LOCAL_LABEL(ret_inf):
+  movs r4, #0xFF
+  lsls r4, r4, #(significandBits)
+  orrs r0, r4
+  lsrs r0, r0, #(significandBits)
+  lsls r0, r0, #(significandBits)
+  pop {r4, r5, r6, r7, pc}
+
+
+END_COMPILERRT_FUNCTION(__addsf3)
+
+NO_EXEC_STACK_DIRECTIVE
diff --git a/lib/builtins/arm/aeabi_cdcmp.S b/lib/builtins/arm/aeabi_cdcmp.S
index 8008f5fca262..b67814d9f20b 100644
--- a/lib/builtins/arm/aeabi_cdcmp.S
+++ b/lib/builtins/arm/aeabi_cdcmp.S
@@ -30,6 +30,19 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cdcmpeq)
         push {r0-r3, lr}
         bl __aeabi_cdcmpeq_check_nan
         cmp r0, #1
+#if __ARM_ARCH_ISA_THUMB == 1
+        beq 1f
+        // NaN has been ruled out, so __aeabi_cdcmple can't trap
+        mov r0, sp
+        ldm r0, {r0-r3}
+        bl __aeabi_cdcmple
+        pop {r0-r3, pc}
+1:
+        // Z = 0, C = 1
+        movs r0, #0xF
+        lsls r0, r0, #31
+        pop {r0-r3, pc}
+#else
         pop {r0-r3, lr}
 
         // NaN has been ruled out, so __aeabi_cdcmple can't trap
@@ -37,6 +50,7 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cdcmpeq)
 
         msr CPSR_f, #APSR_C
         JMP(lr)
+#endif
 END_COMPILERRT_FUNCTION(__aeabi_cdcmpeq)
 
 
@@ -59,6 +73,28 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cdcmple)
 
         bl __aeabi_dcmplt
         cmp r0, #1
+#if __ARM_ARCH_ISA_THUMB == 1
+        bne 1f
+        // Z = 0, C = 0
+        movs r0, #1
+        lsls r0, r0, #1
+        pop {r0-r3, pc}
+1:
+        mov r0, sp
+        ldm r0, {r0-r3}
+        bl __aeabi_dcmpeq
+        cmp r0, #1
+        bne 2f
+        // Z = 1, C = 1
+        movs r0, #2
+        lsls r0, r0, #31
+        pop {r0-r3, pc}
+2:
+        // Z = 0, C = 1
+        movs r0, #0xF
+        lsls r0, r0, #31
+        pop {r0-r3, pc}
+#else
         moveq ip, #0
         beq 1f
 
@@ -72,6 +108,7 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cdcmple)
         msr CPSR_f, ip
         pop {r0-r3}
         POP_PC()
+#endif
 END_COMPILERRT_FUNCTION(__aeabi_cdcmple)
 
 // int __aeabi_cdrcmple(double a, double b) {
diff --git a/lib/builtins/arm/aeabi_cfcmp.S b/lib/builtins/arm/aeabi_cfcmp.S
index 274baf7aecf2..e37aa3d06c4e 100644
--- a/lib/builtins/arm/aeabi_cfcmp.S
+++ b/lib/builtins/arm/aeabi_cfcmp.S
@@ -30,6 +30,19 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cfcmpeq)
         push {r0-r3, lr}
         bl __aeabi_cfcmpeq_check_nan
         cmp r0, #1
+#if __ARM_ARCH_ISA_THUMB == 1
+        beq 1f
+        // NaN has been ruled out, so __aeabi_cfcmple can't trap
+        mov r0, sp
+        ldm r0, {r0-r3}
+        bl __aeabi_cfcmple
+        pop {r0-r3, pc}
+1:
+        // Z = 0, C = 1
+        movs r0, #0xF
+        lsls r0, r0, #31
+        pop {r0-r3, pc}
+#else
         pop {r0-r3, lr}
 
         // NaN has been ruled out, so __aeabi_cfcmple can't trap
@@ -37,6 +50,7 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cfcmpeq)
 
         msr CPSR_f, #APSR_C
         JMP(lr)
+#endif
 END_COMPILERRT_FUNCTION(__aeabi_cfcmpeq)
 
 
@@ -59,6 +73,28 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cfcmple)
 
         bl __aeabi_fcmplt
         cmp r0, #1
+#if __ARM_ARCH_ISA_THUMB == 1
+        bne 1f
+        // Z = 0, C = 0
+        movs r0, #1
+        lsls r0, r0, #1
+        pop {r0-r3, pc}
+1:
+        mov r0, sp
+        ldm r0, {r0-r3}
+        bl __aeabi_fcmpeq
+        cmp r0, #1
+        bne 2f
+        // Z = 1, C = 1
+        movs r0, #2
+        lsls r0, r0, #31
+        pop {r0-r3, pc}
+2:
+        // Z = 0, C = 1
+        movs r0, #0xF
+        lsls r0, r0, #31
+        pop {r0-r3, pc}
+#else
         moveq ip, #0
         beq 1f
 
@@ -72,6 +108,7 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cfcmple)
         msr CPSR_f, ip
         pop {r0-r3}
         POP_PC()
+#endif
 END_COMPILERRT_FUNCTION(__aeabi_cfcmple)
 
 // int __aeabi_cfrcmple(float a, float b) {
diff --git a/lib/builtins/arm/aeabi_dcmp.S b/lib/builtins/arm/aeabi_dcmp.S
index 43e439268d9a..51539c0ac813 100644
--- a/lib/builtins/arm/aeabi_dcmp.S
+++ b/lib/builtins/arm/aeabi_dcmp.S
@@ -26,10 +26,10 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_dcmp ## cond)           \
         bl        SYMBOL_NAME(__ ## cond ## df2) SEPARATOR \
         cmp       r0, #0                         SEPARATOR \
         b ## cond 1f                             SEPARATOR \
-        mov       r0, #0                         SEPARATOR \
+        movs      r0, #0                         SEPARATOR \
         pop       { r4, pc }                     SEPARATOR \
 1:                                               SEPARATOR \
-        mov       r0, #1                         SEPARATOR \
+        movs      r0, #1                         SEPARATOR \
         pop       { r4, pc }                     SEPARATOR \
 END_COMPILERRT_FUNCTION(__aeabi_dcmp ## cond)
 
diff --git a/lib/builtins/arm/aeabi_idivmod.S b/lib/builtins/arm/aeabi_idivmod.S
index b43ea699058d..0164b15dca18 100644
--- a/lib/builtins/arm/aeabi_idivmod.S
+++ b/lib/builtins/arm/aeabi_idivmod.S
@@ -26,7 +26,7 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_idivmod)
         push    {r0, r1, lr}
         bl      SYMBOL_NAME(__divsi3)
         pop     {r1, r2, r3} // now r0 = quot, r1 = num, r2 = denom
-        muls    r2, r2, r0   // r2 = quot * denom
+        muls    r2, r0, r2   // r2 = quot * denom
         subs    r1, r1, r2
         JMP     (r3)
 #else
diff --git a/lib/builtins/arm/aeabi_ldivmod.S b/lib/builtins/arm/aeabi_ldivmod.S
index 3dae14ef07ec..038ae5d723a3 100644
--- a/lib/builtins/arm/aeabi_ldivmod.S
+++ b/lib/builtins/arm/aeabi_ldivmod.S
@@ -23,23 +23,23 @@
         .syntax unified
         .p2align 2
 DEFINE_COMPILERRT_FUNCTION(__aeabi_ldivmod)
-        push    {r11, lr}
+        push    {r6, lr}
         sub     sp, sp, #16
-        add     r12, sp, #8
-        str     r12, [sp]
+        add     r6, sp, #8
+        str     r6, [sp]
 #if defined(__MINGW32__)
-        mov     r12, r0
-        mov     r0, r2
-        mov     r2, r12
-        mov     r12, r1
-        mov     r1, r3
-        mov     r3, r12
+        movs    r6, r0
+        movs    r0, r2
+        movs    r2, r6
+        movs    r6, r1
+        movs    r1, r3
+        movs    r3, r6
 #endif
         bl      SYMBOL_NAME(__divmoddi4)
         ldr     r2, [sp, #8]
         ldr     r3, [sp, #12]
         add     sp, sp, #16
-        pop     {r11, pc}
+        pop     {r6, pc}
 END_COMPILERRT_FUNCTION(__aeabi_ldivmod)
 
 NO_EXEC_STACK_DIRECTIVE
diff --git a/lib/builtins/arm/aeabi_memset.S b/lib/builtins/arm/aeabi_memset.S
index 48edd89705be..633f592279b5 100644
--- a/lib/builtins/arm/aeabi_memset.S
+++ b/lib/builtins/arm/aeabi_memset.S
@@ -26,7 +26,7 @@ DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_memset8, __aeabi_memset)
 
 DEFINE_COMPILERRT_FUNCTION(__aeabi_memclr)
         mov     r2, r1
-        mov     r1, #0
+        movs    r1, #0
         b       memset
 END_COMPILERRT_FUNCTION(__aeabi_memclr)
 
diff --git a/lib/builtins/arm/aeabi_uidivmod.S b/lib/builtins/arm/aeabi_uidivmod.S
index 7098bc6ff92e..a627fc740a02 100644
--- a/lib/builtins/arm/aeabi_uidivmod.S
+++ b/lib/builtins/arm/aeabi_uidivmod.S
@@ -29,7 +29,7 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_uidivmod)
         push    {r0, r1, lr}
         bl      SYMBOL_NAME(__aeabi_uidiv)
         pop     {r1, r2, r3}
-        muls    r2, r2, r0 // r2 = quot * denom
+        muls    r2, r0, r2 // r2 = quot * denom
         subs    r1, r1, r2
         JMP     (r3)
 LOCAL_LABEL(case_denom_larger):
diff --git a/lib/builtins/arm/aeabi_uldivmod.S b/lib/builtins/arm/aeabi_uldivmod.S
index bc26e5674ca0..be343b6bc826 100644
--- a/lib/builtins/arm/aeabi_uldivmod.S
+++ b/lib/builtins/arm/aeabi_uldivmod.S
@@ -23,23 +23,23 @@
         .syntax unified
         .p2align 2
 DEFINE_COMPILERRT_FUNCTION(__aeabi_uldivmod)
-        push	{r11, lr}
+        push	{r6, lr}
         sub	sp, sp, #16
-        add	r12, sp, #8
-        str	r12, [sp]
+        add	r6, sp, #8
+        str	r6, [sp]
 #if defined(__MINGW32__)
-        mov     r12, r0
-        mov     r0, r2
-        mov     r2, r12
-        mov     r12, r1
-        mov     r1, r3
-        mov     r3, r12
+        movs    r6, r0
+        movs    r0, r2
+        movs    r2, r6
+        movs    r6, r1
+        movs    r1, r3
+        movs    r3, r6
 #endif
         bl	SYMBOL_NAME(__udivmoddi4)
         ldr	r2, [sp, #8]
         ldr	r3, [sp, #12]
         add	sp, sp, #16
-        pop	{r11, pc}
+        pop	{r6, pc}
 END_COMPILERRT_FUNCTION(__aeabi_uldivmod)
 
 NO_EXEC_STACK_DIRECTIVE
diff --git a/lib/builtins/arm/comparesf2.S b/lib/builtins/arm/comparesf2.S
index a5b7e36487d6..ef7091bf3c8f 100644
--- a/lib/builtins/arm/comparesf2.S
+++ b/lib/builtins/arm/comparesf2.S
@@ -69,7 +69,7 @@ DEFINE_COMPILERRT_FUNCTION(__eqsf2)
     // the subsequent operations.
 #if __ARM_ARCH_ISA_THUMB == 1
     lsrs    r6,     r3, #1
-    orrs    r6,     r2, r6
+    orrs    r6,     r2
 #else
     orrs    r12,    r2, r3, lsr #1
 #endif
@@ -194,7 +194,7 @@ DEFINE_COMPILERRT_FUNCTION(__gtsf2)
     lsls    r2,        r0, #1
     lsls    r3,        r1, #1
     lsrs    r6,        r3, #1
-    orrs    r6,        r2, r6
+    orrs    r6,        r2
     beq     1f
     movs    r6,        r0
     eors    r6,        r1
diff --git a/lib/builtins/arm/udivsi3.S b/lib/builtins/arm/udivsi3.S
index fcc472b4f3d9..b97b3080bff4 100644
--- a/lib/builtins/arm/udivsi3.S
+++ b/lib/builtins/arm/udivsi3.S
@@ -37,7 +37,16 @@ DEFINE_COMPILERRT_FUNCTION(__udivsi3)
 	beq     LOCAL_LABEL(divby0)
 	udiv	r0, r0, r1
 	bx  	lr
-#else
+
+LOCAL_LABEL(divby0):
+	mov     r0, #0
+#  ifdef __ARM_EABI__
+	b       __aeabi_idiv0
+#  else
+	JMP(lr)
+#  endif
+
+#else /* ! __ARM_ARCH_EXT_IDIV__ */
 	cmp	r1, #1
 	bcc	LOCAL_LABEL(divby0)
 #if __ARM_ARCH_ISA_THUMB == 1
@@ -186,9 +195,12 @@ LOCAL_LABEL(skip_1):
 LOCAL_LABEL(divby0):
 	movs	r0, #0
 #      if defined(__ARM_EABI__)
+	push {r7, lr}
 	bl	__aeabi_idiv0 // due to relocation limit, can't use b.
-#      endif
+	pop  {r7, pc}
+#      else
 	JMP(lr)
+#      endif
 
 
 #if __ARM_ARCH_ISA_THUMB == 1
@@ -252,16 +264,6 @@ LOCAL_LABEL(div0block):
 	JMP(lr)
 #endif /* __ARM_ARCH_EXT_IDIV__ */
 
-#if __ARM_ARCH_EXT_IDIV__
-LOCAL_LABEL(divby0):
-        mov     r0, #0
-#  ifdef __ARM_EABI__
-        b       __aeabi_idiv0
-#  else
-        JMP(lr)
-#  endif
-#endif
-
 END_COMPILERRT_FUNCTION(__udivsi3)
 
 NO_EXEC_STACK_DIRECTIVE
diff --git a/lib/builtins/clear_cache.c b/lib/builtins/clear_cache.c
index 4c2ac3b1eb05..25570fc2157a 100644
--- a/lib/builtins/clear_cache.c
+++ b/lib/builtins/clear_cache.c
@@ -82,10 +82,6 @@ uintptr_t GetCurrentProcess(void);
   #endif
 #endif
 
-#if defined(__linux__) && defined(__arm__)
-  #include <asm/unistd.h>
-#endif
-
 /*
  * The compiler generates calls to __clear_cache() when creating 
  * trampoline functions on the stack for use with nested functions.
@@ -94,7 +90,7 @@ uintptr_t GetCurrentProcess(void);
  */
 
 void __clear_cache(void *start, void *end) {
-#if __i386__ || __x86_64__
+#if __i386__ || __x86_64__ || defined(_M_IX86) || defined(_M_X64)
 /*
  * Intel processors have a unified instruction and data cache
  * so there is nothing to do
@@ -108,6 +104,15 @@ void __clear_cache(void *start, void *end) {
 
         sysarch(ARM_SYNC_ICACHE, &arg);
     #elif defined(__linux__)
+    /*
+     * We used to include asm/unistd.h for the __ARM_NR_cacheflush define, but
+     * it also brought many other unused defines, as well as a dependency on
+     * kernel headers to be installed.
+     *
+     * This value is stable at least since Linux 3.13 and should remain so for
+     * compatibility reasons, warranting it's re-definition here.
+     */
+    #define __ARM_NR_cacheflush 0x0f0002
          register int start_reg __asm("r0") = (int) (intptr_t) start;
          const register int end_reg __asm("r1") = (int) (intptr_t) end;
          const register int flags __asm("r2") = 0;
diff --git a/lib/builtins/cpu_model.c b/lib/builtins/cpu_model.c
index 9a3737020a4e..5ff6baf43876 100644
--- a/lib/builtins/cpu_model.c
+++ b/lib/builtins/cpu_model.c
@@ -27,6 +27,10 @@
 #include <intrin.h>
 #endif
 
+#ifndef __has_attribute
+#define __has_attribute(attr) 0
+#endif
+
 enum VendorSignatures {
   SIG_INTEL = 0x756e6547 /* Genu */,
   SIG_AMD = 0x68747541 /* Auth */
@@ -720,14 +724,17 @@ static unsigned getAvailableFeatures(unsigned int ECX, unsigned int EDX,
   return Features;
 }
 
-#ifdef HAVE_INIT_PRIORITY
-#define CONSTRUCTOR_PRIORITY (101)
+#if defined(HAVE_INIT_PRIORITY)
+#define CONSTRUCTOR_ATTRIBUTE __attribute__((__constructor__ 101))
+#elif __has_attribute(__constructor__)
+#define CONSTRUCTOR_ATTRIBUTE __attribute__((__constructor__))
 #else
-#define CONSTRUCTOR_PRIORITY
+// FIXME: For MSVC, we should make a function pointer global in .CRT$X?? so that
+// this runs during initialization.
+#define CONSTRUCTOR_ATTRIBUTE
 #endif
 
-int __cpu_indicator_init(void)
-    __attribute__((constructor CONSTRUCTOR_PRIORITY));
+int __cpu_indicator_init(void) CONSTRUCTOR_ATTRIBUTE;
 
 struct __processor_model {
   unsigned int __cpu_vendor;
@@ -742,7 +749,7 @@ struct __processor_model {
    the priority set.  However, it still runs after ifunc initializers and
    needs to be called explicitly there.  */
 
-int __attribute__((constructor CONSTRUCTOR_PRIORITY))
+int CONSTRUCTOR_ATTRIBUTE
 __cpu_indicator_init(void) {
   unsigned int EAX, EBX, ECX, EDX;
   unsigned int MaxLeaf = 5;
diff --git a/lib/builtins/divtc3.c b/lib/builtins/divtc3.c
index 04693df471ff..16e538ba4a33 100644
--- a/lib/builtins/divtc3.c
+++ b/lib/builtins/divtc3.c
@@ -17,7 +17,7 @@
 
 /* Returns: the quotient of (a + ib) / (c + id) */
 
-COMPILER_RT_ABI long double _Complex
+COMPILER_RT_ABI Lcomplex
 __divtc3(long double __a, long double __b, long double __c, long double __d)
 {
     int __ilogbw = 0;
@@ -29,31 +29,31 @@ __divtc3(long double __a, long double __b, long double __c, long double __d)
         __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))
+    Lcomplex z;
+    COMPLEX_REAL(z) = crt_scalbnl((__a * __c + __b * __d) / __denom, -__ilogbw);
+    COMPLEX_IMAGINARY(z) = crt_scalbnl((__b * __c - __a * __d) / __denom, -__ilogbw);
+    if (crt_isnan(COMPLEX_REAL(z)) && crt_isnan(COMPLEX_IMAGINARY(z)))
     {
         if ((__denom == 0.0) && (!crt_isnan(__a) || !crt_isnan(__b)))
         {
-            __real__ z = crt_copysignl(CRT_INFINITY, __c) * __a;
-            __imag__ z = crt_copysignl(CRT_INFINITY, __c) * __b;
+            COMPLEX_REAL(z) = crt_copysignl(CRT_INFINITY, __c) * __a;
+            COMPLEX_IMAGINARY(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 : 0.0, __a);
             __b = crt_copysignl(crt_isinf(__b) ? 1.0 : 0.0, __b);
-            __real__ z = CRT_INFINITY * (__a * __c + __b * __d);
-            __imag__ z = CRT_INFINITY * (__b * __c - __a * __d);
+            COMPLEX_REAL(z) = CRT_INFINITY * (__a * __c + __b * __d);
+            COMPLEX_IMAGINARY(z) = CRT_INFINITY * (__b * __c - __a * __d);
         }
         else if (crt_isinf(__logbw) && __logbw > 0.0 &&
                  crt_isfinite(__a) && crt_isfinite(__b))
         {
             __c = crt_copysignl(crt_isinf(__c) ? 1.0 : 0.0, __c);
             __d = crt_copysignl(crt_isinf(__d) ? 1.0 : 0.0, __d);
-            __real__ z = 0.0 * (__a * __c + __b * __d);
-            __imag__ z = 0.0 * (__b * __c - __a * __d);
+            COMPLEX_REAL(z) = 0.0 * (__a * __c + __b * __d);
+            COMPLEX_IMAGINARY(z) = 0.0 * (__b * __c - __a * __d);
         }
     }
     return z;
diff --git a/lib/builtins/ffssi2.c b/lib/builtins/ffssi2.c
new file mode 100644
index 000000000000..e5180eff5e08
--- /dev/null
+++ b/lib/builtins/ffssi2.c
@@ -0,0 +1,29 @@
+/* ===-- ffssi2.c - Implement __ffssi2 -------------------------------------===
+ *
+ *                     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 __ffssi2 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
+__ffssi2(si_int a)
+{
+    if (a == 0)
+    {
+        return 0;
+    }
+    return __builtin_ctz(a) + 1;
+}
diff --git a/lib/builtins/int_lib.h b/lib/builtins/int_lib.h
index 39eee18d9149..8a202dde70f1 100644
--- a/lib/builtins/int_lib.h
+++ b/lib/builtins/int_lib.h
@@ -32,7 +32,11 @@
 #if __ARM_EABI__
 # define ARM_EABI_FNALIAS(aeabi_name, name)         \
   void __aeabi_##aeabi_name() __attribute__((alias("__" #name)));
-# define COMPILER_RT_ABI __attribute__((pcs("aapcs")))
+# ifdef COMPILER_RT_ARMHF_TARGET
+#   define COMPILER_RT_ABI
+# else
+#   define COMPILER_RT_ABI __attribute__((pcs("aapcs")))
+# endif
 #else
 # define ARM_EABI_FNALIAS(aeabi_name, name)
 # define COMPILER_RT_ABI
diff --git a/lib/builtins/os_version_check.c b/lib/builtins/os_version_check.c
new file mode 100644
index 000000000000..74ade2f5b966
--- /dev/null
+++ b/lib/builtins/os_version_check.c
@@ -0,0 +1,178 @@
+/* ===-- os_version_check.c - OS version checking  -------------------------===
+ *
+ *                     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 function __isOSVersionAtLeast, used by
+ * Objective-C's @available
+ *
+ * ===----------------------------------------------------------------------===
+ */
+
+#ifdef __APPLE__
+
+#include <CoreFoundation/CoreFoundation.h>
+#include <dispatch/dispatch.h>
+#include <TargetConditionals.h>
+#include <dlfcn.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+/* These three variables hold the host's OS version. */
+static int32_t GlobalMajor, GlobalMinor, GlobalSubminor;
+static dispatch_once_t DispatchOnceCounter;
+
+/* Find and parse the SystemVersion.plist file. */
+static void parseSystemVersionPList(void *Unused) {
+  (void)Unused;
+  /* Load CoreFoundation dynamically */
+  const void *NullAllocator = dlsym(RTLD_DEFAULT, "kCFAllocatorNull");
+  if (!NullAllocator)
+    return;
+  const CFAllocatorRef kCFAllocatorNull =
+      *(const CFAllocatorRef *)NullAllocator;
+  typeof(CFDataCreateWithBytesNoCopy) *CFDataCreateWithBytesNoCopyFunc =
+      (typeof(CFDataCreateWithBytesNoCopy) *)dlsym(
+          RTLD_DEFAULT, "CFDataCreateWithBytesNoCopy");
+  if (!CFDataCreateWithBytesNoCopyFunc)
+    return;
+  typeof(CFPropertyListCreateWithData) *CFPropertyListCreateWithDataFunc =
+      (typeof(CFPropertyListCreateWithData) *)dlsym(
+          RTLD_DEFAULT, "CFPropertyListCreateWithData");
+  /* CFPropertyListCreateWithData was introduced only in macOS 10.6+, so it
+   * will be NULL on earlier OS versions. */
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wdeprecated-declarations"
+  typeof(CFPropertyListCreateFromXMLData) *CFPropertyListCreateFromXMLDataFunc =
+      (typeof(CFPropertyListCreateFromXMLData) *)dlsym(
+          RTLD_DEFAULT, "CFPropertyListCreateFromXMLData");
+#pragma clang diagnostic pop
+  /* CFPropertyListCreateFromXMLDataFunc is deprecated in macOS 10.10, so it
+   * might be NULL in future OS versions. */
+  if (!CFPropertyListCreateWithDataFunc && !CFPropertyListCreateFromXMLDataFunc)
+    return;
+  typeof(CFStringCreateWithCStringNoCopy) *CFStringCreateWithCStringNoCopyFunc =
+      (typeof(CFStringCreateWithCStringNoCopy) *)dlsym(
+          RTLD_DEFAULT, "CFStringCreateWithCStringNoCopy");
+  if (!CFStringCreateWithCStringNoCopyFunc)
+    return;
+  typeof(CFDictionaryGetValue) *CFDictionaryGetValueFunc =
+      (typeof(CFDictionaryGetValue) *)dlsym(RTLD_DEFAULT,
+                                            "CFDictionaryGetValue");
+  if (!CFDictionaryGetValueFunc)
+    return;
+  typeof(CFGetTypeID) *CFGetTypeIDFunc =
+      (typeof(CFGetTypeID) *)dlsym(RTLD_DEFAULT, "CFGetTypeID");
+  if (!CFGetTypeIDFunc)
+    return;
+  typeof(CFStringGetTypeID) *CFStringGetTypeIDFunc =
+      (typeof(CFStringGetTypeID) *)dlsym(RTLD_DEFAULT, "CFStringGetTypeID");
+  if (!CFStringGetTypeIDFunc)
+    return;
+  typeof(CFStringGetCString) *CFStringGetCStringFunc =
+      (typeof(CFStringGetCString) *)dlsym(RTLD_DEFAULT, "CFStringGetCString");
+  if (!CFStringGetCStringFunc)
+    return;
+  typeof(CFRelease) *CFReleaseFunc =
+      (typeof(CFRelease) *)dlsym(RTLD_DEFAULT, "CFRelease");
+  if (!CFReleaseFunc)
+    return;
+
+  char *PListPath = "/System/Library/CoreServices/SystemVersion.plist";
+
+#if TARGET_OS_SIMULATOR
+  char *PListPathPrefix = getenv("IPHONE_SIMULATOR_ROOT");
+  if (!PListPathPrefix)
+    return;
+  char FullPath[strlen(PListPathPrefix) + strlen(PListPath) + 1];
+  strcpy(FullPath, PListPathPrefix);
+  strcat(FullPath, PListPath);
+  PListPath = FullPath;
+#endif
+  FILE *PropertyList = fopen(PListPath, "r");
+  if (!PropertyList)
+    return;
+
+  /* Dynamically allocated stuff. */
+  CFDictionaryRef PListRef = NULL;
+  CFDataRef FileContentsRef = NULL;
+  UInt8 *PListBuf = NULL;
+
+  fseek(PropertyList, 0, SEEK_END);
+  long PListFileSize = ftell(PropertyList);
+  if (PListFileSize < 0)
+    goto Fail;
+  rewind(PropertyList);
+
+  PListBuf = malloc((size_t)PListFileSize);
+  if (!PListBuf)
+    goto Fail;
+
+  size_t NumRead = fread(PListBuf, 1, (size_t)PListFileSize, PropertyList);
+  if (NumRead != (size_t)PListFileSize)
+    goto Fail;
+
+  /* Get the file buffer into CF's format. We pass in a null allocator here *
+   * because we free PListBuf ourselves */
+  FileContentsRef = (*CFDataCreateWithBytesNoCopyFunc)(
+      NULL, PListBuf, (CFIndex)NumRead, kCFAllocatorNull);
+  if (!FileContentsRef)
+    goto Fail;
+
+  if (CFPropertyListCreateWithDataFunc)
+    PListRef = (*CFPropertyListCreateWithDataFunc)(
+        NULL, FileContentsRef, kCFPropertyListImmutable, NULL, NULL);
+  else
+    PListRef = (*CFPropertyListCreateFromXMLDataFunc)(
+        NULL, FileContentsRef, kCFPropertyListImmutable, NULL);
+  if (!PListRef)
+    goto Fail;
+
+  CFStringRef ProductVersion = (*CFStringCreateWithCStringNoCopyFunc)(
+      NULL, "ProductVersion", kCFStringEncodingASCII, kCFAllocatorNull);
+  if (!ProductVersion)
+    goto Fail;
+  CFTypeRef OpaqueValue = (*CFDictionaryGetValueFunc)(PListRef, ProductVersion);
+  (*CFReleaseFunc)(ProductVersion);
+  if (!OpaqueValue ||
+      (*CFGetTypeIDFunc)(OpaqueValue) != (*CFStringGetTypeIDFunc)())
+    goto Fail;
+
+  char VersionStr[32];
+  if (!(*CFStringGetCStringFunc)((CFStringRef)OpaqueValue, VersionStr,
+                                 sizeof(VersionStr), kCFStringEncodingUTF8))
+    goto Fail;
+  sscanf(VersionStr, "%d.%d.%d", &GlobalMajor, &GlobalMinor, &GlobalSubminor);
+
+Fail:
+  if (PListRef)
+    (*CFReleaseFunc)(PListRef);
+  if (FileContentsRef)
+    (*CFReleaseFunc)(FileContentsRef);
+  free(PListBuf);
+  fclose(PropertyList);
+}
+
+int32_t __isOSVersionAtLeast(int32_t Major, int32_t Minor, int32_t Subminor) {
+  /* Populate the global version variables, if they haven't already. */
+  dispatch_once_f(&DispatchOnceCounter, NULL, parseSystemVersionPList);
+
+  if (Major < GlobalMajor) return 1;
+  if (Major > GlobalMajor) return 0;
+  if (Minor < GlobalMinor) return 1;
+  if (Minor > GlobalMinor) return 0;
+  return Subminor <= GlobalSubminor;
+}
+
+#else
+
+/* Silence an empty translation unit warning. */
+typedef int unused;
+
+#endif
diff --git a/lib/builtins/x86_64/floatdidf.c b/lib/builtins/x86_64/floatdidf.c
index 388404e5e089..dead0ed42c65 100644
--- a/lib/builtins/x86_64/floatdidf.c
+++ b/lib/builtins/x86_64/floatdidf.c
@@ -4,7 +4,7 @@
 
 /* double __floatdidf(di_int a); */
 
-#ifdef __x86_64__
+#if defined(__x86_64__) || defined(_M_X64)
 
 #include "../int_lib.h"
 
diff --git a/lib/builtins/x86_64/floatdisf.c b/lib/builtins/x86_64/floatdisf.c
index 96c3728e92c3..99d5621c6327 100644
--- a/lib/builtins/x86_64/floatdisf.c
+++ b/lib/builtins/x86_64/floatdisf.c
@@ -2,7 +2,7 @@
  * License. See LICENSE.TXT for details.
  */
 
-#ifdef __x86_64__
+#if defined(__x86_64__) || defined(_M_X64)
 
 #include "../int_lib.h"