amiga: 68k optimized alias functions

[BSzili]

Added R_Alias_clip_left, R_Alias_clip_right, R_Alias_clip_top and R_Alias_clip_bottom asm implementations. R_Alias_clip_left and R_Alias_clip_top has a small issue where they clip most polygons 1 pixel too early, leaving a seam at the edge of the screen. It's not very noticeable except with large models. The Quake port I use as a base has the same issue, but it's even less noticeable due to everything being brown. There's probably a rounding error somewhere, but I couldn't see any obvious problems. I'll look into it later. I also ported over R_AliasTransformVector and R_AliasTransformFinalVert that have slightly better implementation than the compiler output.

[sezero]

This is in. I applied two follow-up patches to it.

[sezero]

@BSzili: I noticed this attached patch in tyrquake, will apply soon: patch.txt It touches the C-only versions and not the x86-asm versions.

If you based your m68k asm on the C-only routines, you may need to revise?

[BSzili]

Yes, a quick fix for 68k version should be something like this:

--- r_aclip68k-old.s    2022-04-15 22:01:12.419783500 +0200
+++ r_aclip68k.s        2023-09-14 17:48:53.283761200 +0200
@@ -59,7 +59,7 @@
                move.l  (a1)+,d0
                move.l  (a1)+,d1
                move.l  REFDEF_ALIASVRECT+VRECT_X(a3),d4
-               cmp.l   d1,d3
+               cmp.l   d0,d2
                blt.b   .cont
                sub.l   d2,d4
                fmove.l d4,fp0
@@ -156,7 +156,7 @@
                move.l  (a1)+,d0
                move.l  (a1)+,d1
                move.l  REFDEF_ALIASVRECTRIGHT(a3),d4
-               cmp.l   d1,d3
+               cmp.l   d0,d2
                blt.b   .cont
                sub.l   d2,d4
                fmove.l d4,fp0

[sezero]

OK, will apply shortly along with the C-only version. Thanks!

[sezero]

Patch is in, along with other things.