Because the [-1, 1) lerp can handle a higher sampling rate in t,
decrease the t step to ensure everything is covered. The
limit needs to be trimmed by a 1/65536 to allow the older lerps
to run properly.
Remove the scaled test cases. They no longer make sense.
Change-Id: Ida841d65ebe794ba5d0ae67dfb145268764ec6a2
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/448101
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Using the extra bit improves the difference rate from .38% to .19%.
I think this uses the full resolution of all the values, and there
is no more accuracy by trying to get more bits in use. The maximum
difference is still 1.
This gives the same answer with the neon_vqrdmulhq_s16 and
ssse3_mm_mulhrs_epi16 instructions.
Change-Id: Ie801b68aead408c5e1845aaf94ea660121e414b9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/447719
Reviewed-by: Brian Osman <brianosman@google.com>
Study that compares scaled integer lerp to float lerp. This
shows an error rate of 0.38% with a max difference of 1.
Study to compare scaled values the would result as the
intermediate values of a bilerp. This shows an error rate
of 24% with a max difference of 1.
Change-Id: Id27b6baa9ce56efedb30f490a3f84ccf808315fa
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/447183
Reviewed-by: Brian Osman <brianosman@google.com>
