>The vector extension was implemented in industry based on the draft
Not widely if that's what you meant by "in industry".
Most (like SiFive) did indeed wait for ratification (in case of last minute changes), before announcing their cores.
Pre-standard V chips do of course exist, and this is a good thing (there's better testing in using actual hardware implementations).
They have to use the opcode space reserved for custom extensions, and obviously cannot claim V compatibility. No overlap with actual V opcode space, and therefore no compatibility issues.
These are, nevertheless, fine for embedded uses in which the vendor has control over the whole stack.
It does also mean they have to take on the associated cost of having to maintain it, unless they choose not to actually use the custom extension at all, a choice that realistically will actually suffice for most deployments.
Not widely if that's what you meant by "in industry".
Most (like SiFive) did indeed wait for ratification (in case of last minute changes), before announcing their cores.
Pre-standard V chips do of course exist, and this is a good thing (there's better testing in using actual hardware implementations).
They have to use the opcode space reserved for custom extensions, and obviously cannot claim V compatibility. No overlap with actual V opcode space, and therefore no compatibility issues.
These are, nevertheless, fine for embedded uses in which the vendor has control over the whole stack.
It does also mean they have to take on the associated cost of having to maintain it, unless they choose not to actually use the custom extension at all, a choice that realistically will actually suffice for most deployments.