[igt-dev] [RFC PATCH v10 3/6] lib/igt_gt: use for_each_engine2 to loop through engines

[Chris Wilson]

Quoting Andi Shyti (2019-03-05 13:16:08)
> 'for_each_engine2()' defines a loop through the gpu engines.
> 
> It can work with both active and pre-defined engines.
> 
> In case we are looping through active engines (i.e. the running
> kernel has the query and get/setparam ioctls), the
> intel_active_engines2 points to an array that contains only the
> list engines dynamically allocated after having interrogated the
> driver.
> 
> While, if we are looping through pre-defined engines,
> intel_active_engines2 points to the 'intel_execution_engines2'
> array and works exactly as for_each_engine() but using the
> new 'struct intel_execution_engine2' data type.
> 
> Signed-off-by: Andi Shyti <andi.shyti at intel.com>
> ---
>  lib/igt_gt.h | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/lib/igt_gt.h b/lib/igt_gt.h
> index f4bd6c22a81a..f8bdaa62899f 100644
> --- a/lib/igt_gt.h
> +++ b/lib/igt_gt.h
> @@ -30,6 +30,8 @@
>  
>  #include "i915_drm.h"
>  
> +#include "i915/gem_engine_topology.h"
> +
>  void igt_require_hang_ring(int fd, int ring);
>  
>  typedef struct igt_hang {
> @@ -86,6 +88,12 @@ extern const struct intel_execution_engine {
>              e__++) \
>                 for_if (gem_ring_has_physical_engine(fd__, flags__ = e__->exec_id | e__->flags))
>  
> +#define for_each_engine2(fd, ctx) \
> +               for (struct intel_execution_engine2 *e2__ = \
> +                    gem_set_context_get_engines(fd, ctx); e2__->name; e2__++) \
> +                       for_if (gem_has_engine_topology() || \
> +                               gem_has_engine(fd, e2__->class, e2__->instance))
> +

for (struct intel_engine_iter it = intel_engine_iter(fd, ctx); \
     ((e__) = it.pos < it.nengine ? &it.engines[it.pos] : NULL); \
     intel_engine_iter_next(&it))

struct intel_engine_iter {
	int pos;
	int nengine;
	struct intel_execution_engine2 *engines;
};

Allocate the array everytime as it may change for different (fd, ctx).
In intel_engine_iter_next() remember to free that array when pos >=
nengine.

#define for_each_engine2(fd, ctx, e__)

Is that future-proof enough?
-Chris