On 04/24/2012 10:12 AM, Eric Anholt wrote:
On Tue, 24 Apr 2012 00:34:00 -0700, Kenneth Graunke<[email protected]>
wrote:
A little analysis shows that the worst-case value for "nr" is 16:
- base_mrf = 2 ... 2
- header present (say gen == 5) ... 4
- SIMD16 mode: += 4 * reg_width ... 12
- source_depth_to_render_target ... 14
- dest_depth_reg ... 16
This resulted in us setting base_mrf to 2 and mlen to 14. In other
words, we'd try to use m2..m16. But m16 doesn't exist pre-Gen6. Also,
nr is "where you'd put the next arg if there was one", right? so nr of
16 should be fine, since you've only loaded up to m15?
Drat. You're right...
m2 m3 m4 m5 m6 m7 m8 m9 m10 m11 m12 m13 m14 m15
header rrrrrr gggggg bbbbbb aaaaaaa o-Depth dstDepth
That does fit. My assertions in the previous patch are off by one.
Back to the drawing board...
_______________________________________________
mesa-dev mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
