Replace vpermpd with vpermilpd

[platform/upstream/openblas.git] / common_x86_64.h

diff --git a/common_x86_64.h b/common_x86_64.h
index f27c1e9..c05998d 100644 (file)
--- a/common_x86_64.h
+++ b/common_x86_64.h
@@ -129,12 +129,13 @@ static __inline void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx){
   *ecx=cpuinfo[2];
   *edx=cpuinfo[3];
 #else
-        __asm__ __volatile__("cpuid"
+        __asm__ __volatile__("mov $0, %%ecx;"
+                            "cpuid"
                             : "=a" (*eax),
                             "=b" (*ebx),
                             "=c" (*ecx),
                             "=d" (*edx)
-                            : "0" (op), "c"(0));
+                            : "0" (op));
 #endif
 }
 
@@ -210,7 +211,7 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
        
   y = blas_quick_divide_table[y];
 
-  __asm__ __volatile__  ("mull %0" :"=d" (result) :"a"(x), "0" (y));
+  __asm__ __volatile__  ("mull %0" :"=d" (result), "+a"(x) : "0" (y));
 
   return result;
 }
@@ -276,7 +277,7 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 #ifdef ASSEMBLER
 
 #if defined(PILEDRIVER) || defined(BULLDOZER) || defined(STEAMROLLER) || defined(EXCAVATOR)
-//Enable some optimazation for barcelona.
+//Enable some optimization for barcelona.
 #define BARCELONA_OPTIMIZATION
 #endif