Adding a temporal energy floor to prevent extremely small values when there's
only one bin in the band.

Using the number of unallocated bytes for coarse quantization and VBR
instead the total number of bytes.

Also a minor cleanup to the recombining/division

Also made the new table an unsigned char array.

Should eventually allow using 8-bit values for the pulse->bits table.

When the pitch flag gets corrupted, make sure that pitch is actually
allowed instead of generating a bogus pitch index and crashing.

Timothy B. Terriberry authored 14 years ago and Jean-Marc Valin committed 14 years ago

Change ec_{enc|dec}_bit_prob to take probabilities in units of 1/65536 instead of 1/256. This allows them to use a single 16x16->32 multiply instead of a 24x8->32 multiply. Also change the time-frequency resolution flag coding to ensure that "0" is always the most-probable symbol (i.e., that prob("1")<50%), as that's where all the rounding error accumulates.

Timothy B. Terriberry authored 14 years ago and Jean-Marc Valin committed 14 years ago

Provide direct implementations ec_{enc|dec}_bit_prob() that do not require a division instead of using the normal entropy coder path. This should be exactly equivalent to the existing code.