Hi!

The code in ensure_quant_matrix() won't create a iq_matrix when vol_hdr.load_intra_quant_mat or priv->vol_hdr.load_non_intra_quant_mat is null. I think this should be modified, and the iq_matrix should be allocated in all cases, because theses flags that will tell the hardware if the iq_matrices should be loaded or not is in the struct itself (for example in vdpau-driver/src/vdpau-decode.c, translate_VAIQMatrixBufferMPEG4())

Suggested patch inspired from libavcodec vaapi_mpeg4.c:

diff --git a/gst-libs/gst/vaapi/gstvaapidecoder_mpeg4.c b/gst-libs/gst/vaapi/gstvaapidecoder_mpeg4.c
index b461877..d9e1012 100644
--- a/gst-libs/gst/vaapi/gstvaapidecoder_mpeg4.c
+++ b/gst-libs/gst/vaapi/gstvaapidecoder_mpeg4.c
@@ -242,7 +242,9 @@ ensure_quant_matrix(GstVaapiDecoderMpeg4 *decoder, GstVaapiPicture *picture)
     GstVaapiDecoderMpeg4Private * const priv = &decoder->priv;
     VAIQMatrixBufferMPEG4 *iq_matrix;
 
-    if (!priv->vol_hdr.load_intra_quant_mat && !priv->vol_hdr.load_non_intra_quant_mat) {
+    /* Fill in VAIQMatrixBufferMPEG4 */
+    /* Only the first inverse quantisation method uses the weighting matrices */
+    if (!priv->vol_hdr.quant_type) {
             return GST_VAAPI_DECODER_STATUS_SUCCESS;
     }