Age | Commit message (Collapse) | Author |
|
Integer divisions replaced with floating point multiplications.
Code is still bitexact (calculations produce the same results
for all the possible range of input values).
This optimized function is now almost 3x faster than original
C variant and improves encoding performance somewhat.
|
|
Almost 2x faster than original C variant.
|
|
Is approximately 10x faster than original C variant.
|
|
Is approximately 6x faster than original C variant.
|
|
|
|
Is approximately 4x faster than original C variant.
|
|
Is approximately 4x faster than original C variant.
|
|
Is approximately 4x faster than original C variant.
|
|
ARM NEON optimizations need the use of 16-bit data types and
benefit from 16-byte alignemnt, so we want to have macro
WITH_SIMD defined in the code.
Also automated tests are updated to work correctly on ARM.
And the required stubs for ARM optimizations added (just
copied from 'jsimd_none.c').
|
|
|