marfrit/libva-v4l2-request-fourier
1
0
Fork 1
Code Issues 2 Pull Requests Actions Packages Projects Releases Wiki Activity

ampere-av1 Phase 3 fix: wire reference_frame_ts[] from VAAPI ref_frame_map[]

Browse Source 
Phase 2.1 first hardware test on ampere passed frame 1 (IDR) bit-exact
vs kdirect but frame 2 (inter) diverged starting at byte 64897. Root
cause: reference_frame_ts[] left at zero — kernel can't cross-reference
prior CAPTURE buffers without timestamps.

Fix: in av1_set_controls (which has driver_data), iterate VAAPI's
ref_frame_map[8] (VASurfaceIDs), look up each via SURFACE(driver_data,
ref_id), and pull v4l2_timeval_to_ns(&ref_surface->timestamp) into the
V4L2 ctrl. VA_INVALID_SURFACE entries stay at calloc-zero. Mirrors the
vp9.c:614-628 pattern scaled to AV1's 8 ref slots.

surface_object->timestamp itself is populated in picture.c::EndPicture
from context_object->timestamp_counter at QBUF time on the OUTPUT
buffer — already in place from iter1 baseline.

Verification on ampere (/tmp/test_av1.ivf 208x208, 2 frames):
  Frame 1 + 2 libva sha 029ee72c214b37c1 == kdirect 029ee72c214b37c1
  → 100% byte-identical, kdirect was Phase 0-verified bit-perfect

order_hints[] still zero — VAAPI doesn't expose per-ref POC; observed
not load-bearing on the 208x208 smoke vector. Multi-tile + film_grain
stress vectors are next (av1-1-b8-23-film_grain-50.ivf).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
Claude (noether)
2026-05-17 10:28:32 +00:00
parent 85bcddb5ad
commit 5fb7e36955
1 changed files with 32 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/av1.c
+32 -4
View File
@@ -53,11 +53,14 @@
#include "av1.h"
#include "context.h"
#include "object_heap.h"
#include "request.h"
#include "surface.h"
#include "utils.h"
#include "v4l2.h"
#include <va/va.h>
#include <linux/videodev2.h>
#include <linux/v4l2-controls.h>
@@ -359,11 +362,13 @@ static void av1_fill_frame(VADecPictureParameterBufferAV1 *picture,
}
/* ---- reference frames + order hints ---- */
for (i = 0; i < BACKEND_AV1_TOTAL_REFS_PER_FRAME; i++) {
/* VAAPI doesn't expose order_hints[]; leave zero. */
/* reference_frame_ts[] is filled by the orchestrator (av1_set_controls)
* which has driver_data for the SURFACE() lookup. order_hints[] not
* exposed per-ref by VAAPI — leave zero. ref_frame_idx[7] is the
* index map from spec-defined ref slots (LAST..ALTREF) into
* ref_frame_map[8] (the surface IDs). */
for (i = 0; i < BACKEND_AV1_TOTAL_REFS_PER_FRAME; i++)
ctrl->order_hints[i] = 0;
ctrl->reference_frame_ts[i] = 0;
}
for (i = 0; i < BACKEND_AV1_REFS_PER_FRAME; i++)
ctrl->ref_frame_idx[i] = picture->ref_frame_idx[i];
@@ -534,6 +539,29 @@ int av1_set_controls(struct request_data *driver_data,
av1_fill_sequence(picture, &sequence);
av1_fill_frame(picture, &frame);
/*
* Phase 2.1 + frame-2 divergence fix: wire reference_frame_ts[].
* VAAPI exposes ref_frame_map[8] as VASurfaceIDs; the kernel needs
* v4l2-style timestamps to cross-reference the corresponding
* CAPTURE buffers (set on the OUTPUT buffer at QBUF time per
* picture.c::EndPicture, via surface_object->timestamp). Mirrors
* the vp9.c:614-628 pattern, scaled to AV1's 8 ref slots.
*
* VA_INVALID_SURFACE entries stay at the calloc'd zero timestamp
* (kernel reads zero, doesn't try to dereference).
*/
for (i = 0; i < BACKEND_AV1_TOTAL_REFS_PER_FRAME; i++) {
VASurfaceID ref_id = picture->ref_frame_map[i];
struct object_surface *ref_surface;
if (ref_id == VA_INVALID_SURFACE)
continue;
ref_surface = SURFACE(driver_data, ref_id);
if (ref_surface)
frame.reference_frame_ts[i] =
v4l2_timeval_to_ns(&ref_surface->timestamp);
}
if (driver_data->has_av1_film_grain)
av1_fill_film_grain(picture, &film_grain);