Eccelerators.Tokenization

src/Unigram/ArrayUnigramStorage.lvt

namespace Eccelerators.Tokenization.Unigram

/**
 * Dependency-free array workspace for the compatibility segmenter.
 * Owns WORDS 64-bit cells; larger hardware designs may inject a RAM provider.
 * Consumers still initialize live cells and serialize access under IUnigramStorage.
 */
component ArrayUnigramStorage<WORDS: int> : IUnigramStorage
{
	static assert WORDS > 0, "invalid unigram storage capacity"

	// Owned fallback 64-bit workspace cells.
	data: logic[WORDS, 64]

	/**
	 * Returns true for 64-bit word addresses 0..WORDS-1.
	 */
	override fn IsValidAddress(address: int) bool
	{
		return address >= 0 && address < WORDS
	}

	/**
	 * Returns the stored word, or zero outside 0..WORDS-1.
	 * Consumers must not use this fallback to bypass IUnigramStorage preconditions.
	 */
	override fn Read(address: int) logic[64]
	{
		if (address < 0 || address >= WORDS) {
			return 0
		}

		return data[address]
	}

	/**
	 * Stores all 64 bits at a valid word address; ignores out-of-range writes.
	 * Consumers are required to validate capacity before using this provider.
	 */
	override fn Write(address: int, value: logic[64])
	{
		if (address >= 0 && address < WORDS) {
			data[address] = value
		}
	}
}