11:56tomeu: pH5: so I'm going to start by figuring out the run length table, then the symbol map and then see how the encoding goes
11:56tomeu: on the s905d3, which I guess could be different
11:58pH5: tomeu: \o/ let me know what you learn, maybe the rlt mechanism is similar
12:05pH5: something is different with tiling on v8. the output_image_tile_y_size value chosen by the blob is different and it again sets some bits in the unused regions
12:10tomeu: the tile thing could be due to different parameters that currently are hardcoded
12:21tomeu: actually, the zrl table seems to be constant with VIV_VX_ENABLE_NON_ZERO_BALANCE=0
12:22tomeu: so I will move to the symbol table
15:09pH5: tomeu: thank you, using the correct input/accumulation buffer depths fixes the tiling issue
15:09tomeu: yay!
15:12pH5: I've also seen the blob set unused8 = 0x3 for input sizes that are an exact multiple of 8, but just not doing that doesn't seem to make a difference.
15:13tomeu: yeah, I have been in that situation a few times...