On Fri, Sep 04, 2020 at 02:06:07PM +0100, Tvrtko Ursulin wrote:
> From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
>
> 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@xxxxxxxxx>
> ---
> 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
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
