ACK: [PATCH 09/11] fwts: Use gettimeofday for cpu benchmark

[Colin Ian King]

On 21/05/15 10:34, Jeremy Kerr wrote:
> Currently, we use time() to measure out one second of time for a CPU
> benchmark. This means we have to wait around one second for a new time
> value before performing the test, and means we have to assume that we've
> been running for about one second.
> 
> This change uses gettimeofday() instead; this way, we don't need to wait
> before starting a test, and get millisecond resolution when calculating
> the performance value.
> 
> Signed-off-by: Jeremy Kerr <jk at ozlabs.org>
> 
> ---
>  src/lib/src/fwts_cpu.c |   26 ++++++++++++++------------
>  1 file changed, 14 insertions(+), 12 deletions(-)
> 
> diff --git a/src/lib/src/fwts_cpu.c b/src/lib/src/fwts_cpu.c
> index a7cfd3d..743de63 100644
> --- a/src/lib/src/fwts_cpu.c
> +++ b/src/lib/src/fwts_cpu.c
> @@ -31,11 +31,11 @@
>  #include <sys/types.h>
>  #include <sys/syscall.h>
>  #include <sys/wait.h>
> +#include <sys/time.h>
>  #include <signal.h>
>  #include <fcntl.h>
>  #include <math.h>
>  #include <sched.h>
> -#include <time.h>
>  
>  #include <linux/perf_event.h>
>  
> @@ -373,12 +373,12 @@ int fwts_cpu_benchmark(
>  	const int cpu,		/* CPU we want to measure performance */
>  	fwts_cpu_benchmark_result *result)
>  {
> +	struct timeval start, end, duration;
>  	unsigned long long perfctr_result;
>  	fwts_cpu_benchmark_result tmp;
>  	cpu_set_t mask, oldset;
>  	int perfctr, ncpus, rc;
>  	static bool warned;
> -	time_t current;
>  	bool perf_ok;
>  
>  	ncpus = fwts_cpu_enumerate();
> @@ -418,24 +418,24 @@ int fwts_cpu_benchmark(
>  		return FWTS_ERROR;
>  	}
>  
> -	/* Wait until we get a new second */
> -	current = time(NULL);
> -	while (current == time(NULL))
> -		sched_yield();
> -
>  	if (perf_ok)
>  		perf_start_counter(perfctr);
> -
> -	current = time(NULL);
> +	gettimeofday(&start, NULL);
>  
>  	/*
>  	 * And burn some CPU cycles and get a bogo-compute like
>  	 * loop count measure of CPU performance.
>  	 */
> -	do {
> +	for (;;) {
>  		fwts_cpu_burn_cycles();
> +
>  		tmp.loops++;
> -	} while (current == time(NULL));
> +
> +		gettimeofday(&end, NULL);
> +		timersub(&end, &start, &duration);
> +		if (duration.tv_sec >= 1)
> +			break;
> +	}
>  
>  	if (perf_ok)
>  		perf_stop_counter(perfctr);
> @@ -448,7 +448,9 @@ int fwts_cpu_benchmark(
>  	if (perf_ok) {
>  		rc = perf_read_counter(perfctr, &perfctr_result);
>  		if (rc == FWTS_OK) {
> -			tmp.cycles = perfctr_result;
> +			tmp.cycles = perfctr_result /
> +				((1.0 * duration.tv_usec / 1000000) +
> +				 duration.tv_sec);
>  			tmp.cycles_valid = true;
>  		} else {
>  			fwts_log_warning(fw, "failed to read perf counters");
> 
Good improvement.

Acked-by: Colin Ian King <colin.king at canonical.com>