El mar, 17-06-2008 a las 11:06 +0800, Bean escribió: > AFAIK, lzma decompresser don't need intermediate buffers, it operates > directly on the input and output buffer, for example, the lzmadecoder > in coreboot: I think you didn't see the intermediate buffer because it is hidden in the Probs field passed through the decoderstate struct. I've googled my way through the net to find what the hell "probs" mean, and I still don't understand ten contiguous lines of the LZMA code. What I could find is that
http://qa.coreboot.org/docs/doxygen/lzma_8c-source.html provides the decoder with about 32KiB of "scratchpad" memory and will reject decompressing files which "require" (as somehow indicated by the first 5 bytes of the LZMA stream) more than that. I don't know if that space is used for dictionaries, trees, forests or a Matrix full of virtual monkeys hitting keys at random until the decompressed stream is achieved, but its size seems to impose constraints at the time of compression (probably a maximum -n level) that merit investigation. PS: have you tested the performance of other compression algorithms like gzip-deflate (much simpler) or bzip2 (near LZMA for small files but with less memory requisites)?
signature.asc
Description: Esta parte del mensaje está firmada digitalmente
_______________________________________________ Grub-devel mailing list Grub-devel@gnu.org http://lists.gnu.org/mailman/listinfo/grub-devel
