[Vm-dev] [OpenSmalltalk/opensmalltalk-vm] BitBlt rgbMul is broken for depth 1 and not correctly rounded above depth 4 (Issue #651)

[Nicolas Cellier]

If we wanted to handle 16bits depth without split (thus 6 channels in parallel on a 32bits word), that would get a bit tricky due to the dead-bit.
We would need two different masks, because shifting a mask does not produce the other one:

    lowMask :=  2r00000011111000000111110000011111. "green2-red1-blue1"
    highMask := 2r01111100000111110000001111100000.	"red2-blue2-green1"
    highWordShift := 27.
    doubleGroupMask := 2r0000001111100000011111000001111100000011111000000111110000011111. "highMask << highWordShift + lowMask"
    doubleWord1 := word1 bitAnd: highMask.
    doubleWord2 := word2 bitAnd: highMask.
    doubleWord1 := doubleWord1 << highWordShift + (word1 bitAnd: lowMask).
    doubleWord2 := doubleWord2 << highWordShift + (word2 bitAnd: lowMask).

Then the shifts for accessing each component in double word would be tricky, either 10 or 11 in the loop (0 10 21 32 42 53).

    extraShift := 2r10110.
    shift := 0.
    0 to: 5 do: [:i |
        doubleWordMul := doubleWordMul + (((doubleWordSrc >> shift bitAnd: channelMask) * (doubleWordSrc >> shift bitAnd: channelMask) + half) << shift).
        shift := shift + (2 * nBits) + (extraShift >> i bitAnd: 1)].

The rest should work unchanged.

-- 
Reply to this email directly or view it on GitHub:
https://github.com/OpenSmalltalk/opensmalltalk-vm/issues/651#issuecomment-1227847296
You are receiving this because you are subscribed to this thread.

Message ID: <OpenSmalltalk/opensmalltalk-vm/issues/651/1227847296 at github.com>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20220825/3fec40d2/attachment-0001.html>