[Libre-soc-bugs] [Bug 1155] O(n^2) multiplication REMAP mode(s)

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=1155

--- Comment #28 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Jacob Lifshay from comment #27)
> (In reply to Luke Kenneth Casson Leighton from comment #25)
> > thoughts?
> 
> triangles are cool, needs further thought. I'm thinking we'd want a way to
> do new_x = (x + y) % x_sz for diagonal matrix stuff:

yep. see comment #0 3rd option
except it was just going to be sum (x+y).

> new indexes:
> 0 1 2 3 -- rotated right 0
> 5 6 7 4 -- rotated right 1
> A B 8 9 -- rotated right 2

ok *matrix* rotated, iinteresting, like it.

ok so there are also some AV patterns to cover, which are:

|   | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |    |   | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |
8 | 9 | a |
| 0 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | a |    | 0 |           | 0 | 1 | 2 | 3 | 4 |
5 | 6 | 7 |
| 1 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | a |    | 1 |           | 0 | 1 | 2 | 3 | 4 |
5 | 6 | 7 |
| 2 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 |    | 2 |       | 0 | 1 | 2 | 3 | 4 | 5 |
6 | 7 |
| 3 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | -> | 3 |       | 0 | 1 | 2 | 3 | 4 | 5 |
6 | 7 |
| 4 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |    | 4 |   | 0 | 1 | 2 | 3 | 4 | 5 | 6 |
7 |
| 5 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |    | 5 |   | 0 | 1 | 2 | 3 | 4 | 5 | 6 |
7 |
| 6 | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |    | 6 | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |
| 7 | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |    | 7 | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |

which is a "skip=0110" mode with x=7,y=7,z=2 which gives the (twice)
repeated row. i.e where it was *believed* that skip could be dropped,
actually it can't.

hm.

|0:5   |6:11  | 12:17   | 18:20   | 21:23   |24:27 |28:29  |30:31| Mode  |
|----- |----- | ------- | ------- | ------  |------|------ |---- | ----- |
|xdimsz|ydimsz| zdimsz  | submode2|sk1/invxy|offset|submode|0b11 |bigmul |

5 bits for submodes. sk1 to give that repetition. invxy for mirroring.
(like Indexed Mode which has mirroring)

let's see what is possible:

* submode[0]: standard matrix (x*xdimsz + y) or bigmul (x+y)
* submode[1]: square or triangle

* submode2[0] 0b00 modulo then add offset  0b01 add offset then modulo
* submode2[1:2] triange 0b00 x+y   0b01 y+z   0b10 z+x   0b11 RSVD
                square  0b00 x,y,z 0b01 y,x,z 0b10 z,x,y 0b11 z,y,x

i am reasonably certain there are better uses for last 2 bits of submode2
offset is signed. modulo it puts bigger numbers out first

-- 
You are receiving this mail because:
You are on the CC list for the bug.