fixed rdtscp assembly

configure.in

@@ -4740,7 +4740,7 @@ pac_cv_posix_clock_realtime=no)
         pac_cv_linux86_cycle,[
             AC_TRY_COMPILE(,[
                 long long var, *var_ptr=&var;
-	        __asm__ __volatile__("rdtscp" : "=A" (*var_ptr));
+		__asm__ __volatile__("rdtscp; shl $32, %%rdx; or %%rdx, %%rax" : "=a" (*var_ptr) : : "ecx", "rdx");
             ],pac_cv_linux86_cycle=rdtscp,
                 AC_TRY_COMPILE(,[
 	            long long var, *var_ptr=&var;

src/mpid/ch3/channels/sshm/configure.in

@@ -135,7 +135,7 @@ dnl gcc checking
         pac_cv_linux86_cycle,[
             AC_TRY_COMPILE(,[
                 long long var, *var_ptr=&var;
-	        __asm__ __volatile__("rdtscp" : "=A" (*var_ptr));
+		__asm__ __volatile__("rdtscp; shl $32, %%rdx; or %%rdx, %%rax" : "=a" (*var_ptr) : : "ecx", "rdx");
             ],pac_cv_linux86_cycle=rdtscp,
                 AC_TRY_COMPILE(,[
 	            long long var, *var_ptr=&var;

src/mpid/ch3/channels/sshm/src/ch3i_progress.h

@@ -18,7 +18,8 @@
    the rdtscp instruction which is synchronizing, we use this when we
    can. */
 #ifdef GCC_X86_CYCLE_RDTSCP
-#define MPID_CPU_TICK(var_ptr) __asm__ __volatile__("rdtscp" : "=A" (*var_ptr))
+#define MPID_CPU_TICK(var_ptr)                                                                          \
+    __asm__ __volatile__("rdtscp; shl $32, %%rdx; or %%rdx, %%rax" : "=a" (*var_ptr) : : "ecx", "rdx")
 #elif defined(GCC_X86_CYCLE_CPUID_RDTSC)
 /* Here we have to save the ebx register for when the compiler is
    generating position independent code (e.g., when it's generating

src/mpid/ch3/channels/ssm/configure.in

@@ -176,7 +176,7 @@ dnl gcc checking
         pac_cv_linux86_cycle,[
             AC_TRY_COMPILE(,[
                 long long var, *var_ptr=&var;
-	        __asm__ __volatile__("rdtscp" : "=A" (*var_ptr));
+		__asm__ __volatile__("rdtscp; shl $32, %%rdx; or %%rdx, %%rax" : "=a" (*var_ptr) : : "ecx", "rdx");
             ],pac_cv_linux86_cycle=rdtscp,
                 AC_TRY_COMPILE(,[
 	            long long var, *var_ptr=&var;

src/mpid/ch3/channels/ssm/include/ch3i_progress.h

@@ -30,7 +30,8 @@
    the rdtscp instruction which is synchronizing, we use this when we
    can. */
 #ifdef GCC_X86_CYCLE_RDTSCP
-#define MPID_CPU_TICK(var_ptr) __asm__ __volatile__("rdtscp" : "=A" (*var_ptr))
+#define MPID_CPU_TICK(var_ptr)                                                                          \
+    __asm__ __volatile__("rdtscp; shl $32, %%rdx; or %%rdx, %%rax" : "=a" (*var_ptr) : : "ecx", "rdx")
 #elif defined(GCC_X86_CYCLE_CPUID_RDTSC)
 /* Here we have to save the ebx register for when the compiler is
    generating position independent code (e.g., when it's generating