Closed atlanswer closed 3 years ago
Switch implemented.
inv_scale
now.(idx + 0.5) * inv_scale - 0.5
was implemented, but the result is rounded, which is not necessary in the algorithm.The output image finally looks reasonable now.
The linear assembly implementation was completed without optimization. And the fixed-point multiplication in C implementation was resolved. 512 x 1.5 cycle count comparison: | C float approach | C int approach | Linear assembly |
---|---|---|---|
1654442635 | 1027521855 | 60202045 |
Resolves #5.