Conversation

Replying to
You could burn instructions to shift and mask on the low order address bits, but a faster path would be to interleave 8 independent textures at different bit positions in the words, and have specialized texture mapping routines with a constant shift and mask. I don't recall ever\
2
4
203
considering that arrangement, but there may have been relevant consumer systems where doing it for 2 or 4 bit textures packed in bytes would have been helpful. Figuring out how to best exploit chained vector instructions for texture mapping on the Cray would be phase two.
6
3
185
This Tweet is from a suspended account. Learn more
Show replies
Show replies