Minor fixes.

After some thinking and discussions, we would like to propose that false value is represented as TR::Int64 type and value 0. The true value will be represented as TR::Int64 value -1.

This will make it easier to splat a -1 value to all mask elements on platforms that don't have Mask registers. On those platforms, all mask operations happen bit-wise and therefore true lanes should contain all 1s.

It should also help platforms that have Mask registers since the passed long value will represent the whole mask.

If other values are passed, the behavior is undefined.

On the second thought, the child type can be TR::Int8. Codegen can always do b2l (usually noop) if needed. We would like to keep the opcode's semantics to a minimum.

I support that thought. It seemed a bit unnatural to require a 64-bit value to pass in true or false otherwise. Even for projects that represent true and false as byte (8-bit).

Updated according to the comments above.

Jenkins build ppc64le

jenkins build plinux

Improved opcode description.

Improved commit message

Fixed typo

Jenkins build all

Build failures seem unrelated to me.