1 files changed, 109 insertions, 4 deletions

diff --git a/lib/Headers/arm_acle.h b/lib/Headers/arm_acle.h
index a0fd6894ab96..814df2c3d782 100644
--- a/lib/Headers/arm_acle.h
+++ b/lib/Headers/arm_acle.h
@@ -66,6 +66,41 @@ static __inline__ void __attribute__((always_inline, nodebug)) __yield(void) {
 }
 #endif
 
+#if __ARM_32BIT_STATE
+#define __dbg(t) __builtin_arm_dbg(t)
+#endif
+
+/* 8.5 Swap */
+static __inline__ uint32_t __attribute__((always_inline, nodebug))
+  __swp(uint32_t x, volatile uint32_t *p) {
+  uint32_t v;
+  do v = __builtin_arm_ldrex(p); while (__builtin_arm_strex(x, p));
+  return v;
+}
+
+/* 8.6 Memory prefetch intrinsics */
+/* 8.6.1 Data prefetch */
+#define __pld(addr) __pldx(0, 0, 0, addr)
+
+#if __ARM_32BIT_STATE
+#define __pldx(access_kind, cache_level, retention_policy, addr) \
+  __builtin_arm_prefetch(addr, access_kind, 1)
+#else
+#define __pldx(access_kind, cache_level, retention_policy, addr) \
+  __builtin_arm_prefetch(addr, access_kind, cache_level, retention_policy, 1)
+#endif
+
+/* 8.6.2 Instruction prefetch */
+#define __pli(addr) __plix(0, 0, addr)
+
+#if __ARM_32BIT_STATE
+#define __plix(cache_level, retention_policy, addr) \
+  __builtin_arm_prefetch(addr, 0, 0)
+#else
+#define __plix(cache_level, retention_policy, addr) \
+  __builtin_arm_prefetch(addr, 0, cache_level, retention_policy, 0)
+#endif
+
 /* 8.7 NOP */
 static __inline__ void __attribute__((always_inline, nodebug)) __nop(void) {
   __builtin_arm_nop();
@@ -73,6 +108,32 @@ static __inline__ void __attribute__((always_inline, nodebug)) __nop(void) {
 
 /* 9 DATA-PROCESSING INTRINSICS */
 /* 9.2 Miscellaneous data-processing intrinsics */
+/* ROR */
+static __inline__ uint32_t __attribute__((always_inline, nodebug))
+  __ror(uint32_t x, uint32_t y) {
+  y %= 32;
+  if (y == 0)  return x;
+  return (x >> y) | (x << (32 - y));
+}
+
+static __inline__ uint64_t __attribute__((always_inline, nodebug))
+  __rorll(uint64_t x, uint32_t y) {
+  y %= 64;
+  if (y == 0)  return x;
+  return (x >> y) | (x << (64 - y));
+}
+
+static __inline__ unsigned long __attribute__((always_inline, nodebug))
+  __rorl(unsigned long x, uint32_t y) {
+#if __SIZEOF_LONG__ == 4
+  return __ror(x, y);
+#else
+  return __rorll(x, y);
+#endif
+}
+
+
+/* CLZ */
 static __inline__ uint32_t __attribute__((always_inline, nodebug))
   __clz(uint32_t t) {
   return __builtin_clz(t);
@@ -85,13 +146,10 @@ static __inline__ unsigned long __attribute__((always_inline, nodebug))
 
 static __inline__ uint64_t __attribute__((always_inline, nodebug))
   __clzll(uint64_t t) {
-#if __SIZEOF_LONG_LONG__ == 8
   return __builtin_clzll(t);
-#else
-  return __builtin_clzl(t);
-#endif
 }
 
+/* REV */
 static __inline__ uint32_t __attribute__((always_inline, nodebug))
   __rev(uint32_t t) {
   return __builtin_bswap32(t);
@@ -111,6 +169,53 @@ static __inline__ uint64_t __attribute__((always_inline, nodebug))
   return __builtin_bswap64(t);
 }
 
+/* REV16 */
+static __inline__ uint32_t __attribute__((always_inline, nodebug))
+  __rev16(uint32_t t) {
+  return __ror(__rev(t), 16);
+}
+
+static __inline__ unsigned long __attribute__((always_inline, nodebug))
+  __rev16l(unsigned long t) {
+    return __rorl(__revl(t), sizeof(long) / 2);
+}
+
+static __inline__ uint64_t __attribute__((always_inline, nodebug))
+  __rev16ll(uint64_t t) {
+  return __rorll(__revll(t), 32);
+}
+
+/* REVSH */
+static __inline__ int16_t __attribute__((always_inline, nodebug))
+  __revsh(int16_t t) {
+  return __builtin_bswap16(t);
+}
+
+/* RBIT */
+static __inline__ uint32_t __attribute__((always_inline, nodebug))
+  __rbit(uint32_t t) {
+  return __builtin_arm_rbit(t);
+}
+
+static __inline__ uint64_t __attribute__((always_inline, nodebug))
+  __rbitll(uint64_t t) {
+#if __ARM_32BIT_STATE
+  return (((uint64_t) __builtin_arm_rbit(t)) << 32) |
+    __builtin_arm_rbit(t >> 32);
+#else
+  return __builtin_arm_rbit64(t);
+#endif
+}
+
+static __inline__ unsigned long __attribute__((always_inline, nodebug))
+  __rbitl(unsigned long t) {
+#if __SIZEOF_LONG__ == 4
+  return __rbit(t);
+#else
+  return __rbitll(t);
+#endif
+}
+
 /*
  * 9.4 Saturating intrinsics
  *