From d1baf14a64d93062763f9899fa9c2d7e4bad62a3 Mon Sep 17 00:00:00 2001
From: Xianyi Zhang <traits.zhang@gmail.com>
Date: Fri, 11 Nov 2011 17:49:41 +0000
Subject: [PATCH] Enable thread affinity on Loongson 3B. Fixed the bug of
 reading cycle counter.

In Loongson 3A and 3B, the CPU core increases the counter in every 2 cycles by default.
---
 Makefile.system |  2 ++
 common_mips64.h | 28 +++++++++++++++++++++-------
 2 files changed, 23 insertions(+), 7 deletions(-)

diff --git a/Makefile.system b/Makefile.system
index 84f41a78f..985f95084 100644
--- a/Makefile.system
+++ b/Makefile.system
@@ -591,9 +591,11 @@ endif
 
 ifneq ($(ARCH), x86_64)
 ifneq ($(ARCH), x86)
+ifneq ($(CORE), LOONGSON3B)
 NO_AFFINITY = 1
 endif
 endif
+endif
 
 ifdef NO_AFFINITY
 CCOMMON_OPT	+= -DNO_AFFINITY
diff --git a/common_mips64.h b/common_mips64.h
index 15f947eb8..5db96c4aa 100644
--- a/common_mips64.h
+++ b/common_mips64.h
@@ -101,13 +101,15 @@ static void INLINE blas_lock(volatile unsigned long *address){
 
 static inline unsigned int rpcc(void){
   unsigned long ret;
-#if defined(LOONGSON3A) 
-  unsigned long long tmp;
-  __asm__ __volatile__("dmfc0 %0, $25, 1": "=r"(tmp):: "memory");
-  ret=tmp;
-#elif defined(LOONGSON3B)
-  //Temp Implementation.
-  return 1;
+#if defined(LOONGSON3A) || defined(LOONGSON3B)
+  //  unsigned long long tmp;
+  //__asm__ __volatile__("dmfc0 %0, $25, 1": "=r"(tmp):: "memory");
+  //ret=tmp;
+  __asm__ __volatile__(".set push \n"
+                       ".set mips32r2\n"
+                       "rdhwr %0, $2\n"
+                       ".set pop": "=r"(ret):: "memory");
+
 #else
   __asm__ __volatile__(".set   push    \n"                                     
           ".set   mips32r2\n"                                                  
@@ -117,6 +119,18 @@ static inline unsigned int rpcc(void){
   return ret;
 }
 
+//#if defined(LOONGSON3A) || defined(LOONGSON3B)
+static inline int WhereAmI(void){
+  int ret=0;
+  __asm__ __volatile__(".set push \n"
+                       ".set mips32r2\n"
+                       "rdhwr %0, $0\n"
+                       ".set pop": "=r"(ret):: "memory");
+  return ret;
+
+}
+//#endif
+
 static inline int blas_quickdivide(blasint x, blasint y){
   return x / y;
 }