International organisation for standardisation organisation internationale de normalisation

5.13.6Wavefront parallel processing

5.13.6.1.1.1.1.1.1JCTVC-H0226 On derivation of luma QP predictor for Wavefront Parallel Processing [S. Shimada, K. Kazui, J. Koyama, A. Nakagawa (Fujitsu)]

The BoG agreed that there was an issue, and recommended, as a solution, to inherit the QP prediction value at the left edge from the slice header in which the LCU belongs (as is done for independent tiles). Decision: Adopted as recommended by BoG.

5.13.6.1.1.1.1.1.2JCTVC-H0328 AHG4: Crosscheck for Samsung simplification on WPP in JCTVC-H0349 [C.-Y. Tsai, Y.-W. Huang (MediaTek)] [late]
5.13.6.1.1.1.1.1.3JCTVC-H0513 On parallel processing schemes [Y. -K. Wang, M. Coban (Qualcomm)]

The BoG recommended to disallow the combination of WPP and entropy slices. Decision: Agreed.

The BoG could not reach consensus regarding recommending to disallow the combination of tiles and entropy slices, and recommended further discussion. Decision: In Track A, it was agreed to disallow the combination of tiles and entropy slices.

5.13.6.1.1.1.1.1.4JCTVC-H0517 On wavefront parallel processing [Y. -K. Wang, M. Coban (Qualcomm)]

The BoG recommended that the WPP entropy coding synchronization point be fixed to be the second LCU in the above LCU row. H0349 recommended the same idea. This will require the removal of the entropy_coding_synchro syntax element and an accompanying modification to the decoding process. The proponents are asked to work together to propose how the entropy_coding_synchro syntax element would be removed. Decision: Agreed (subject to producing the specific details).

The BoG recommended for the proponents to evaluate the overhead due to the addition of entry points as an alternative.

The BoG recommended CABAC flushing at the end of each LCU row to be mandated. The idea is also included in H0349. This adoption will require the removal of the cabac_istate_reset_flag syntax element and an accompanying modification to the decoding process. Decision: Agreed (subject to producing the specific details).

The BoG recognized a need to investigate whether the value of having byte alignment is worth the 0.6% BD bit rate average loss (across all sequences) that is incurred for an alternative approach of mandating byte alignment at the end of each LCU row.

5.13.6.1.1.1.1.1.5JCTVC-H0349 Simplification on WPP [T. Lee, J. Choi, J. Park (Samsung)]

The BoG recommended adoption of the ideas in this proposal, which were identical to those proposed in items 2 and 3a of H0517. See actions noted above.