Re: [Mingw-users] Trouble with rdtsc on amd64 in 32-bit environment

[PerrimanSmythe <com...@ma...>]

Gisle Vanem-2 wrote:
> 
> "PerrimanSmythe" <com...@ma...> wrote:
> 
>> The problem is, that when I'm reading the tsc value using the rdtsc
>> instruction, I get only the lower 32 bits of timestamp. I've tried using
>> different styles of calling the rdtsc I've found on the internet, but the
>> problem persists. The current code I use is as follows:
>> 
>> unsigned long long rdtsc()
>> {
>> unsigned long long temp;
>> __asm__ __volatile__
>> (
>> "cpuid \n"
>> "rdtsc \n"
>> "leal %0, %%ecx \n"
>> "movl %%eax, (%%ecx) \n"
>> "movl %%edx, 4(%%ecx)" :: "m"(temp) : "eax", "ebx", "ecx", "edx"
>> );
>> return temp;
>> }
> 
> Here is one I'm using with success for many years:
> 
> /*
>  * Return CPU timestamp value. This code is originally by
>  *   Tom Burgess <Tom...@bc...> and
>  *   Douglas Eleveld <del...@dd...>
>  */
> extern __inline__ uint64 get_rdtsc (void)
> {
>   register uint64 tsc;
>   __asm__ __volatile__ (
>             ".byte 0x0F, 0x31;"   /* rdtsc opcode */
>           : "=A" (tsc) );
>   return (tsc);
> }
> 
> 
> --gv
> 
> 

Thank you for you reply. Unfortunately, it doesn't work this way either. I
only get the lower 32 bits of tsc, and if my code runs for more than 2^32
cycles, and that's about 1.5 seconds, it overflows and I get wrong results.
I also tried exactly the same code on the core 2 duo notebook with 32-bit
Vista OS (this one is not the one I mentioned in first post) with the
exactly same MinGW I have on my PC, and it works just fine!
-- 
View this message in context: http://www.nabble.com/Trouble-with-rdtsc-on-amd64-in-32-bit-environment-tp25843286p25845204.html
Sent from the MinGW - User mailing list archive at Nabble.com.