From e12cf1123e8784ce6fe9d2ac14526331fbe2c555 Mon Sep 17 00:00:00 2001
From: Grazvydas Ignotas <notasas@gmail.com>
Date: Sun, 16 Aug 2015 17:27:25 +0200
Subject: [PATCH] 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        | 28 ++++++++++++++++++++++++++++
 common_alpha.h  |  1 +
 common_arm.h    | 10 ----------
 common_arm64.h  | 10 ----------
 common_ia64.h   |  2 ++
 common_mips64.h |  1 +
 common_power.h  |  1 +
 common_sparc.h  |  1 +
 common_x86.h    |  1 +
 common_x86_64.h |  1 +
 10 files changed, 36 insertions(+), 20 deletions(-)

diff --git a/common.h b/common.h
index 320adadcb..5998b5608 100644
--- a/common.h
+++ b/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"
diff --git a/common_alpha.h b/common_alpha.h
index 845fb316a..86f58966a 100644
--- a/common_alpha.h
+++ b/common_alpha.h
@@ -89,6 +89,7 @@ static __inline unsigned int rpcc(void){
 
   return r0;
 }
+#define RPCC_DEFINED
 
 
 #define HALT 	ldq	$0, 0($0)
diff --git a/common_arm.h b/common_arm.h
index 2dabd4d7f..7e0c02306 100644
--- a/common_arm.h
+++ b/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;
 }
diff --git a/common_arm64.h b/common_arm64.h
index aa310c5f2..cc08fa75b 100644
--- a/common_arm64.h
+++ b/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;
 }
diff --git a/common_ia64.h b/common_ia64.h
index 8e92b5992..d1f210749 100644
--- a/common_ia64.h
+++ b/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);
diff --git a/common_mips64.h b/common_mips64.h
index 7cd86b375..bc1a52fb4 100644
--- a/common_mips64.h
+++ b/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
diff --git a/common_power.h b/common_power.h
index e9b5cb630..3b9471a17 100644
--- a/common_power.h
+++ b/common_power.h
@@ -103,6 +103,7 @@ static inline unsigned long rpcc(void){
 #endif
 
 }
+#define RPCC_DEFINED
 
 #ifdef __64BIT__
 #define RPCC64BIT
diff --git a/common_sparc.h b/common_sparc.h
index 87ef75276..8a16e3d3a 100644
--- a/common_sparc.h
+++ b/common_sparc.h
@@ -66,6 +66,7 @@ static __inline unsigned long rpcc(void){
 
   return clocks;
 };
+#define RPCC_DEFINED
 
 #ifdef __64BIT__
 #define RPCC64BIT
diff --git a/common_x86.h b/common_x86.h
index 99a723fd7..9506716ce 100644
--- a/common_x86.h
+++ b/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;
diff --git a/common_x86_64.h b/common_x86_64.h
index efb902416..3a02beefb 100644
--- a/common_x86_64.h
+++ b/common_x86_64.h
@@ -82,6 +82,7 @@ static __inline BLASULONG rpcc(void){
 
   return ((BLASULONG)a + ((BLASULONG)d << 32));
 }
+#define RPCC_DEFINED
 
 #define RPCC64BIT