... | ... | @@ -549,6 +549,7 @@ Should also consider: |
|
|
|
|
|
- vector constants, at least at Cmm level
|
|
|
- replicating a scalar to a vector
|
|
|
- FMA: fused multiply add, this is supported by NEON and AVX however software fallback may not be possible with the same precision. Tricky.
|
|
|
- AVX also suppports a bunch of interesting things:
|
|
|
|
|
|
- permute, shuffle, "blend", masked moves.
|
... | ... | |