fresnel-fourier iter5b-β Phase 7 fix-forward commit D: destination_* for vaapi-copy late-surface flow

Phase 7 empirical: all 5 libva codecs returned all-zero because
CreateContext's surfaces_ids[] walk was a no-op for ffmpeg-vaapi-copy
which passes surfaces_count=0 to vaCreateContext (per the iter6
comment at context.c:262). Surfaces existed in driver_data's
surface_heap but weren't in the param array → destination_* stayed
at the zero initialization from CreateSurfaces2 β → BeginPicture's
surface_bind_slot saw destination_planes_count=0 → no data
assignment → copy_surface_to_image read all-zero.

Fix: cache the format-uniform CAPTURE geometry in driver_data
(fmt_valid, fmt_planes_count, fmt_buffers_count, fmt_format_height,
fmt_sizes[], fmt_bytesperlines[]). Populate at CreateContext after
v4l2_get_format(CAPTURE). Walk surface_heap (not just surfaces_ids[])
to fill every existing surface. Add lazy-fill in CreateSurfaces2 for
surfaces created AFTER CreateContext. Invalidate cache in
DestroyContext.

New helper: surface_fill_format_uniform(driver_data, surface_object).
Idempotent on destination_planes_count != 0.

Signed-off-by: claude-noether <claude-noether@reauktion.de>
This commit is contained in:
claude-noether
2026-05-12 18:52:33 +00:00
parent 7055b14f5e
commit 70196f8065
4 changed files with 135 additions and 34 deletions
+44 -34
View File
@@ -54,7 +54,6 @@ VAStatus RequestCreateContext(VADriverContextP context, VAConfigID config_id,
struct request_data *driver_data = context->pDriverData;
struct object_config *config_object;
struct object_context *context_object = NULL;
struct object_surface *surface_object;
struct video_format *video_format;
unsigned int destination_sizes[VIDEO_MAX_PLANES];
unsigned int destination_bytesperlines[VIDEO_MAX_PLANES];
@@ -65,7 +64,7 @@ VAStatus RequestCreateContext(VADriverContextP context, VAConfigID config_id,
VAContextID id;
VAStatus status;
unsigned int output_type, capture_type;
unsigned int i, j;
unsigned int j;
bool found;
int rc;
@@ -196,7 +195,7 @@ VAStatus RequestCreateContext(VADriverContextP context, VAConfigID config_id,
/*
* Compute format-uniform destination_* values. Same for all
* surfaces of this format; written once per surface here, never
* surfaces of this format; written once per surface, never
* changed by BeginPicture's slot acquisition.
*/
if (video_format->v4l2_buffers_count == 1) {
@@ -207,39 +206,44 @@ VAStatus RequestCreateContext(VADriverContextP context, VAConfigID config_id,
}
/*
* Walk surfaces_ids and populate the format-uniform fields on each
* object_surface. CreateSurfaces2 (β) left these zeroed.
* iter5b-β Commit D: cache the format-uniform CAPTURE geometry
* in driver_data. CreateSurfaces2 calls AFTER this CreateContext
* (ffmpeg vaapi-copy late-surface-allocation case) will lazy-fill
* via surface_fill_format_uniform(); the surface_heap walk below
* fills surfaces that pre-existed when CreateContext fired.
*/
for (i = 0; i < (unsigned int)surfaces_count; i++) {
surface_object = SURFACE(driver_data, surfaces_ids[i]);
if (surface_object == NULL) {
status = VA_STATUS_ERROR_INVALID_SURFACE;
goto error;
}
surface_object->destination_planes_count = destination_planes_count;
surface_object->destination_buffers_count =
video_format->v4l2_buffers_count;
driver_data->fmt_planes_count = destination_planes_count;
driver_data->fmt_buffers_count = video_format->v4l2_buffers_count;
driver_data->fmt_format_height = format_height;
for (j = 0; j < destination_planes_count; j++) {
driver_data->fmt_sizes[j] = destination_sizes[j];
driver_data->fmt_bytesperlines[j] =
destination_bytesperlines[j];
}
driver_data->fmt_valid = true;
if (video_format->v4l2_buffers_count == 1) {
for (j = 0; j < destination_planes_count; j++) {
surface_object->destination_offsets[j] =
j > 0 ? destination_sizes[j - 1] : 0;
surface_object->destination_sizes[j] =
destination_sizes[j];
surface_object->destination_bytesperlines[j] =
destination_bytesperlines[0];
}
} else if (video_format->v4l2_buffers_count == destination_planes_count) {
for (j = 0; j < destination_planes_count; j++) {
surface_object->destination_offsets[j] = 0;
surface_object->destination_sizes[j] =
destination_sizes[j];
surface_object->destination_bytesperlines[j] =
destination_bytesperlines[j];
}
} else {
status = VA_STATUS_ERROR_ALLOCATION_FAILED;
goto error;
/*
* Walk the surface_heap (not just surfaces_ids[]) to populate
* destination_* on every existing surface. Pre-Commit-D we walked
* surfaces_ids[], which is empty for ffmpeg vaapi-copy consumers
* that call vaCreateContext with surfaces_count=0 — those surfaces
* exist in the heap but aren't in the param array. Walking the
* heap catches both flows. Late-created surfaces (after this
* CreateContext) fill via surface_fill_format_uniform in
* CreateSurfaces2's per-surface init.
*/
{
struct object_surface *surface_iter;
int heap_iter;
surface_iter = (struct object_surface *)
object_heap_first(&driver_data->surface_heap,
&heap_iter);
while (surface_iter != NULL) {
surface_fill_format_uniform(driver_data, surface_iter);
surface_iter = (struct object_surface *)
object_heap_next(&driver_data->surface_heap,
&heap_iter);
}
}
@@ -500,7 +504,13 @@ VAStatus RequestDestroyContext(VADriverContextP context, VAContextID context_id)
* β doesn't have a last_output_{width,height,pixelformat} cache
* (those fields are deleted). Each CreateContext is a fresh
* S_FMT(OUTPUT) cycle.
*
* Commit D: invalidate the format-uniform cache so a CreateSurfaces2
* call between DestroyContext and the next CreateContext doesn't
* lazy-fill with stale geometry from the now-torn-down session.
* The next CreateContext re-populates the cache.
*/
driver_data->fmt_valid = false;
return VA_STATUS_SUCCESS;
}
+21
View File
@@ -90,7 +90,28 @@ struct request_data {
* which is naturally one-shot per context cycle; no caching is
* required. DestroyContext + next CreateContext rebuild from
* scratch.
*
* iter5b-β Commit D: cache the format-uniform CAPTURE-side
* geometry from v4l2_get_format so CreateSurfaces2 can populate
* a newly-created surface's destination_* fields without
* re-querying the device. Set by CreateContext after the
* v4l2_get_format(CAPTURE) call; consumed by both:
* 1. CreateContext's surface_heap walk (fills surfaces that
* pre-exist when CreateContext fires);
* 2. CreateSurfaces2's per-surface init (fills surfaces
* created AFTER CreateContext, e.g. ffmpeg vaapi-copy
* pool dynamics where the consumer passes surfaces_count=0
* to vaCreateContext and creates surfaces lazily).
*
* fmt_valid is true once CreateContext has populated the cache;
* CreateSurfaces2 only lazy-fills when fmt_valid is true.
*/
bool fmt_valid;
unsigned int fmt_format_height;
unsigned int fmt_planes_count;
unsigned int fmt_buffers_count;
unsigned int fmt_sizes[VIDEO_MAX_PLANES];
unsigned int fmt_bytesperlines[VIDEO_MAX_PLANES];
};
VAStatus VA_DRIVER_INIT_FUNC(VADriverContextP context);
+49
View File
@@ -115,6 +115,44 @@ void surface_unbind_slot(struct request_data *driver_data,
surface_object->current_slot = NULL;
}
/*
* iter5b-β Commit D: fill format-uniform destination_* on a surface
* from driver_data's CAPTURE-format cache. Idempotent: no-op if
* destination_planes_count is non-zero already.
*/
void surface_fill_format_uniform(struct request_data *driver_data,
struct object_surface *surface_object)
{
unsigned int j;
if (!driver_data->fmt_valid)
return;
if (surface_object->destination_planes_count != 0)
return;
surface_object->destination_planes_count = driver_data->fmt_planes_count;
surface_object->destination_buffers_count = driver_data->fmt_buffers_count;
if (driver_data->fmt_buffers_count == 1) {
for (j = 0; j < driver_data->fmt_planes_count; j++) {
surface_object->destination_offsets[j] =
j > 0 ? driver_data->fmt_sizes[j - 1] : 0;
surface_object->destination_sizes[j] =
driver_data->fmt_sizes[j];
surface_object->destination_bytesperlines[j] =
driver_data->fmt_bytesperlines[0];
}
} else if (driver_data->fmt_buffers_count == driver_data->fmt_planes_count) {
for (j = 0; j < driver_data->fmt_planes_count; j++) {
surface_object->destination_offsets[j] = 0;
surface_object->destination_sizes[j] =
driver_data->fmt_sizes[j];
surface_object->destination_bytesperlines[j] =
driver_data->fmt_bytesperlines[j];
}
}
}
VAStatus RequestCreateSurfaces2(VADriverContextP context, unsigned int format,
unsigned int width, unsigned int height,
VASurfaceID *surfaces_ids,
@@ -173,6 +211,17 @@ VAStatus RequestCreateSurfaces2(VADriverContextP context, unsigned int format,
surface_object->request_fd = -1;
/*
* iter5b-β Commit D: if CreateContext has already populated
* the format-uniform cache (driver_data->fmt_valid), fill
* the new surface's destination_* immediately. This covers
* the case where a consumer creates more surfaces AFTER
* CreateContext. The first batch of surfaces (created before
* CreateContext) gets filled by CreateContext's surface_heap
* walk; this lazy-fill handles late arrivals.
*/
surface_fill_format_uniform(driver_data, surface_object);
surfaces_ids[i] = id;
}
+21
View File
@@ -165,6 +165,27 @@ VAStatus RequestExportSurfaceHandle(VADriverContextP context,
VASurfaceID surface_id, uint32_t mem_type,
uint32_t flags, void *descriptor);
/*
* iter5b-β Commit D: populate a surface's format-uniform destination_*
* fields (planes_count, buffers_count, offsets, sizes, bytesperlines)
* from driver_data's cached CAPTURE-side geometry. Idempotent: skip
* if already filled (destination_planes_count != 0). Caller must
* ensure driver_data->fmt_valid is true (CreateContext has run).
*
* Called by:
* - context.c::RequestCreateContext after v4l2_get_format(CAPTURE)
* populates the cache; walks the surface_heap and fills every
* existing surface (covers surfaces created before CreateContext,
* including the ffmpeg vaapi-copy case where surfaces_count=0 is
* passed but surfaces exist in the heap from earlier
* CreateSurfaces2 calls).
* - surface.c::RequestCreateSurfaces2 after surface allocation,
* covering the case where CreateContext fired before this
* CreateSurfaces2 call (fmt cache is valid, fill immediately).
*/
void surface_fill_format_uniform(struct request_data *driver_data,
struct object_surface *surface_object);
/*
* Iter2 Fix 3: bind / unbind a CAPTURE-pool slot to an object_surface.
* Called from picture.c::RequestBeginPicture (acquire+bind) and