I am very curious, so we do need to store meta data for each of the values right? (e.g., which base, number of bytes used, etc.) Is it still worth it to do the compression?
This comment was marked helpful 0 times.
RICEric22
It seems that (and the paper suggested a few slides earlier) the number of bytes used in compression is different for each base, so those that are compressed to 1 byte goes to base 1, those compressed to 2 bytes go to base 2, etc., so the corresponding compression values for each base don't need to be stored. Otherwise, each pattern in the cache line can probably have start / end tags. From this you can infer the size of the compression, and thus the base that was used.
I am very curious, so we do need to store meta data for each of the values right? (e.g., which base, number of bytes used, etc.) Is it still worth it to do the compression?
This comment was marked helpful 0 times.
It seems that (and the paper suggested a few slides earlier) the number of bytes used in compression is different for each base, so those that are compressed to 1 byte goes to base 1, those compressed to 2 bytes go to base 2, etc., so the corresponding compression values for each base don't need to be stored. Otherwise, each pattern in the cache line can probably have start / end tags. From this you can infer the size of the compression, and thus the base that was used.
This comment was marked helpful 0 times.