[Intel-gfx] [igt-dev] [PATCH i-g-t 1/1] intel-gpu-top: Support for client stats

[Petri Latvala]

On Fri, Sep 04, 2020 at 02:06:07PM +0100, Tvrtko Ursulin wrote:
> From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> 
> Adds support for per-client engine busyness stats i915 exports in sysfs
> and produces output like the below:
> 
> ==========================================================================
> intel-gpu-top -  935/ 935 MHz;    0% RC6; 14.73 Watts;     1097 irqs/s
> 
>       IMC reads:     1401 MiB/s
>      IMC writes:        4 MiB/s
> 
>           ENGINE      BUSY                                 MI_SEMA MI_WAIT
>      Render/3D/0   63.73% |███████████████████           |      3%      0%
>        Blitter/0    9.53% |██▊                           |      6%      0%
>          Video/0   39.32% |███████████▊                  |     16%      0%
>          Video/1   15.62% |████▋                         |      0%      0%
>   VideoEnhance/0    0.00% |                              |      0%      0%
> 
>   PID            NAME     RCS          BCS          VCS         VECS
>  4084        gem_wsim |█████▌     ||█          ||           ||           |
>  4086        gem_wsim |█▌         ||           ||███        ||           |
> ==========================================================================
> 
> Apart from the existing physical engine utilization it now also shows
> utilization per client and per engine class.
> 
> v2:
>  * Version to match removal of global enable_stats toggle.
>  * Plus various fixes.
> 
> v3:
>  * Support brief backward jumps in client stats.
> 
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> ---
>  tools/intel_gpu_top.c | 539 +++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 528 insertions(+), 11 deletions(-)
> 
> diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
> index cae01c25b920..9eac569e75de 100644
> --- a/tools/intel_gpu_top.c
> +++ b/tools/intel_gpu_top.c
> @@ -679,23 +679,347 @@ static void pmu_sample(struct engines *engines)
>  	}
>  }
>  
> +enum client_status {
> +	FREE = 0, /* mbz */
> +	ALIVE,
> +	PROBE
> +};
> +
> +struct clients;
> +
> +struct client {
> +	struct clients *clients;
> +
> +	enum client_status status;
> +	unsigned int id;
> +	unsigned int pid;
> +	char name[128];
> +	unsigned int samples;
> +	unsigned long total;
> +	struct engines *engines;
> +	unsigned long *val;
> +	uint64_t *last;
> +};
> +
> +struct engine_class {
> +	unsigned int class;
> +	const char *name;
> +	unsigned int num_engines;
> +};
> +
> +struct clients {
> +	unsigned int num_classes;
> +	struct engine_class *class;
> +
> +	unsigned int num_clients;
> +	struct client *client;
> +};
> +
> +#define for_each_client(clients, c, tmp) \
> +	for ((tmp) = (clients)->num_clients, c = (clients)->client; \
> +	     (tmp > 0); (tmp)--, (c)++)
> +
> +static struct clients *init_clients(void)
> +{
> +	struct clients *clients = malloc(sizeof(*clients));
> +
> +	return memset(clients, 0, sizeof(*clients));
> +}
> +
> +#define SYSFS_CLIENTS "/sys/class/drm/card0/clients"

Now that intel_gpu_top supports device selection, this path works
every time only 60% of the time, right?



-- 
Petri Latvala