add fallback rpcc implementation

- use on arm, arm64 and any new platform
- use faster integer math instead of double
- use similar scale as rdtsc so that timeouts work

common.h

@@ -410,7 +410,35 @@ typedef char env_var_t[MAX_PATH];
 typedef char* env_var_t;
 #define readenv(p, n) ((p)=getenv(n))
 #endif
+
+#if !defined(RPCC_DEFINED) && !defined(OS_WINDOWS)
+#ifdef _POSIX_MONOTONIC_CLOCK
+#if defined(__GNUC_PREREQ) && __GLIBC_PREREQ(2, 17) // don't require -lrt
+#define USE_MONOTONIC
+#elif defined(OS_ANDROID)
+#define USE_MONOTONIC
+#endif
+#endif
+/* use similar scale as x86 rdtsc for timeouts to work correctly */
+static inline unsigned long long rpcc(void){
+#ifdef USE_MONOTONIC
+  struct timespec ts;
+  clock_gettime(CLOCK_MONOTONIC, &ts);
+  return (unsigned long long)ts.tv_sec * 1000000000ull + ts.tv_nsec;
+#else
+  struct timeval tv;
+  gettimeofday(&tv,NULL);
+  return (unsigned long long)tv.tv_sec * 1000000000ull + tv.tv_usec * 1000;
+#endif
+}
+#define RPCC_DEFINED
+#define RPCC64BIT
+#endif // !RPCC_DEFINED
+
+#ifndef RPCC_DEFINED
+#error "rpcc() implementation is missing for your platform"
 #endif
+#endif // !ASSEMBLER
 
 #ifdef OS_LINUX
 #include "common_linux.h"

common_alpha.h

@@ -89,6 +89,7 @@ static __inline unsigned int rpcc(void){
 
   return r0;
 }
+#define RPCC_DEFINED
 
 
 #define HALT 	ldq	$0, 0($0)

common_arm.h

@@ -72,16 +72,6 @@ static void __inline blas_lock(volatile BLASULONG *address){
 }
 
 
-static inline unsigned long long rpcc(void){
-  unsigned long long ret=0;
-  double v;
-  struct timeval tv;
-  gettimeofday(&tv,NULL);
-  v=(double) tv.tv_sec + (double) tv.tv_usec * 1e-6;
-  ret = (unsigned long long) ( v * 1000.0d );
-  return ret;
-}
-
 static inline int blas_quickdivide(blasint x, blasint y){
   return x / y;
 }

common_arm64.h

@@ -71,16 +71,6 @@ static void __inline blas_lock(volatile BLASULONG *address){
 }
 
 
-static inline unsigned long long rpcc(void){
-  unsigned long long ret=0;
-  double v;
-  struct timeval tv;
-  gettimeofday(&tv,NULL);
-  v=(double) tv.tv_sec + (double) tv.tv_usec * 1e-6;
-  ret = (unsigned long long) ( v * 1000.0d );
-  return ret;
-}
-
 static inline int blas_quickdivide(blasint x, blasint y){
   return x / y;
 }

common_ia64.h

@@ -75,6 +75,7 @@ static __inline unsigned long rpcc(void) {
   __asm__ __volatile__ ("mov %0=ar.itc" : "=r"(clocks));
   return clocks;
 }
+#define RPCC_DEFINED
 
 
 static __inline unsigned long stmxcsr(void){
@@ -103,6 +104,7 @@ static __inline void blas_lock(volatile unsigned long *address){
 static __inline unsigned int rpcc(void) {
   return __getReg(_IA64_REG_AR_ITC);
 }
+#define RPCC_DEFINED
 
 static __inline unsigned int stmxcsr(void) {
   return __getReg(_IA64_REG_AR_FPSR);

common_mips64.h

@@ -118,6 +118,7 @@ static inline unsigned int rpcc(void){
 #endif
   return ret;
 }
+#define RPCC_DEFINED
 
 #if defined(LOONGSON3A) || defined(LOONGSON3B)
 #ifndef NO_AFFINITY

common_power.h

@@ -103,6 +103,7 @@ static inline unsigned long rpcc(void){
 #endif
 
 }
+#define RPCC_DEFINED
 
 #ifdef __64BIT__
 #define RPCC64BIT

common_sparc.h

@@ -66,6 +66,7 @@ static __inline unsigned long rpcc(void){
 
   return clocks;
 };
+#define RPCC_DEFINED
 
 #ifdef __64BIT__
 #define RPCC64BIT

common_x86.h

@@ -73,6 +73,7 @@ static __inline unsigned long long rpcc(void){
 
   return ((unsigned long long)a + ((unsigned long long)d << 32));
 };
+#define RPCC_DEFINED
 
 static __inline unsigned long getstackaddr(void){
   unsigned long addr;

common_x86_64.h

@@ -82,6 +82,7 @@ static __inline BLASULONG rpcc(void){
 
   return ((BLASULONG)a + ((BLASULONG)d << 32));
 }
+#define RPCC_DEFINED
 
 #define RPCC64BIT