P4
Currently API "VectorMask.toVector()" is implemented with broadcast and blend optimizations:
```
@ForceInline
AbstractVector<E> toVectorTemplate() {
AbstractSpecies<E> vsp = this.vspecies();
Vector<E> zero = vsp.broadcast(0L);
Vector<E> mone = vsp.broadcast(-1L);
return (AbstractVector)zero.blend(mone, this);
}
```
This can be optimized by doing hotspot intrinsify directly. This is a cheap operation on some architectures like AArch64. No code should be generated on NEON, while a single `cpy` is needed on SVE. As the fallback implementation, we can use the original blend pattern if architectures do not support the direct vector instruction for this API.
```
@ForceInline
AbstractVector<E> toVectorTemplate() {
AbstractSpecies<E> vsp = this.vspecies();
Vector<E> zero = vsp.broadcast(0L);
Vector<E> mone = vsp.broadcast(-1L);
return (AbstractVector)zero.blend(mone, this);
}
```
This can be optimized by doing hotspot intrinsify directly. This is a cheap operation on some architectures like AArch64. No code should be generated on NEON, while a single `cpy` is needed on SVE. As the fallback implementation, we can use the original blend pattern if architectures do not support the direct vector instruction for this API.