V9958 documentation unclear

I don’t know what you refer to by Mode I and Mode II, but if you refer to the mode0 and mode1 pins on the chip, those are the I/O port selection inputs, you don’t need to worry about them from a software perspective. If you refer to Sprite Mode I and II, then the TMS9918 compatible modes use Sprite Mode I and all of the new screen modes (screen 4 aka graphic 3 and up) in the V9938 use Sprite Mode II.

The V9958 DAC is indeed 15-bit (5 bits per R, G, B component, plus 1 transparency bit) which means that it can output 32 shades per colour component, up to a theoretical maximum of 32^3 = 32768.

However the palette registers of the V9958 still only allows you to specify 3 bits per R, G, B component (compatible with the 9-bit V9938), allowing you to specify 512 colours max for palette colours. The larger 15 bits output colour space is only available in the YJK modes (screen 10-12) which use a special encoding that more or less separates luminance from colour, where the luminance (Y) can be specified per-pixel and the colour information (JK) is shared between 4 adjacent pixels.

Either way, because of that encoding, it can still only specify up to 19268 colours out of the 32768 possible. The colour with the slightly reduced resolution is blue (which is the least visible to the human eye). For more information read my The YJK screen modes article on the MAP.

The Sega VDP and V99x8 series are both derived from the YM2220 which is Yamaha’s TMS9918 clone, so they have commonalities, however all the new screen modes of the V99x8 are quite different, e.g. the V99x8 introduces bitmap modes which the Sega VDP does not have, while the Sega VDPs has a more powerful pattern mode and in later versions (SG-1000 Mark III and MegaDrive) dropped support for the TMS9918 compatible modes.

I've found. On the other hand, these bits do not react in the same way according to the type of VDP. So better to leave them set.

You are right, since I often see more defects on red in compressed images, I thought that it was on this color that we reserved less space to store its values.

@Grauw,
I am referring to page 29 of V9958 manual. CPU can select A0 and A1, is that related to superimpose?
Page 9 is mentioning colors 2^16 = 65536.
Short summary mentioned 256 colors max simultaneously.
Also 192.68 colors.

Blue is correct, see also color specification for screen 8. 3 bits green, 3 bits red and 2 bits blue, which makes total of 8 bits.

No, A0 and A1 are CPU address lines, they select the I/O port.

It’s phrased confusingly, but what they mean is that the Y, J and K components combined are 16 bits (4 bits for Y, 6 bits for J and K). On page 8 (bottom right) you see that in YJK+YAE mode they also say the colour specification is 2^17, one additional bit for the Y component.

However this does not give you 2^16 or 2^17 colours to display, because the formula adds Y (unsigned 5-bit), J and K (signed 6-bit) in such a way that the resulting colour component value can be lower than 0 or higher than 31. In these cases, the colour component value clips and is clamped to either 0 or 31.

On the Interactive YJK colour space visualisation on the MAP YJK screen modes article, turn on the “Clip range” checkbox to see this clipping visualised. There are 16384 (2^14) un-clipped colours, and due to the clamping to the 0-31 range you get an additional 2884 unique colours totalling 19268.

p.s. The A bit in YJK+YAE mode is not "alpha" (transparency), but "attribute"; whether the four Y bits are interpreted as an YJK luminance value, or as a 16-bit palette colour code (like in screen 5 / graphic 4). This allows the developer to use palette colours to e.g. to draw menus on top of a photo, or cover up colour spill artefacts. In ways you can see this as a screen 5 image overlaid on an YJK image.

In YJK mode you can’t superimpose the image (except the borders). In YJK+YAE mode you can superimpose the image, by setting the A bit and specifying palette colour 0 in the Y bits (the TP flag in register 0 must not be set).

This refers to the V9938 screen 8 / graphic 7. In YJK mode it can exceed that number to 19268.

You put the dot in a strange place for the second time now, so just to make sure you didn’t misread the colour count; there are 19 thousand 268 colours, not 192 and 68/100ths.

That is the case for screen 8. For YJK mode it’s more like 5 bits red, 5 bits green, “4 bits” blue. In YJK+YAE mode blue resolution is further reduced to “3 bits”.

Note I put quotation marks around those, because in isolation blue still uses all 32 discrete output voltages. However the precise value is offset by 0-1 (YJK) or 0-3 (YJK+YAE) depending on the red and green components. E.g. in YJK (20, 20, 17) and (20, 21, 18) can be represented, but (20, 20, 16) and (20, 21, 17) can not.