I suppose it could be useful, faster method of multiplying matrix :
https://science.slashdot.org/story/22/10/13/2259214/deepmind-breaks-50-year-math-record-using-ai-new-record-falls-a-week-later (https://science.slashdot.org/story/22/10/13/2259214/deepmind-breaks-50-year-math-record-using-ai-new-record-falls-a-week-later)
Cool, I read the news somewhere else. Now I need to restudy the math though, I wasn't even aware there was an older method to do things faster like the Strassen algorithm mentioned.
But anyway, on CPC some simple 3d dot transformations could be done with a lot of precalcs and not even touching the matrix math let alone muls. But maybe useful in some other stuff.
It's 25 years ago since I did this for the last time, but
IIRC a 3x3 matrix was multiplied with a 1x3 matrix for rotating a 3D object.
I don't know where it is necessary to multiply 4x4 matrices especially on the CPC.
This new improvement from yesterday is even for 5x5 matrices.
Any chance for a set of very fast z80 vector matrix operations? That is beyond my capability, but I'd love to use them.
Quote from: genesis8 on 09:41, 14 October 22I suppose it could be useful, faster method of multiplying matrix :
https://science.slashdot.org/story/22/10/13/2259214/deepmind-breaks-50-year-math-record-using-ai-new-record-falls-a-week-later (https://science.slashdot.org/story/22/10/13/2259214/deepmind-breaks-50-year-math-record-using-ai-new-record-falls-a-week-later)
Now I can finally start writing my JPEG viewer for the CPC ;D