Hi,
> +typedef struct Model {
> + int cum_prob[MODEL_MAX_SYMS + 1];
> + int weights[MODEL_MAX_SYMS + 1];
> + int idx2sym[MODEL_MAX_SYMS + 1];
> + int sym2idx[MODEL_MAX_SYMS + 1];
> + int num_syms;
> + int thr_weight, threshold;
> +} Model;
How high are the weights typically? Same for the LUT values of idx2sym
and sym2idx. num_syms seems to be set to 8 or 2, but there is probably
something I missed in between.
Basically, I am wondering if a smaller type could be used.
> +static void arith_normalise(ArithCoder *c)
[...]
> +static int arith_get_bit(ArithCoder *c)
[...]
> +static int arith_get_bits(ArithCoder *c, int bits)
[...]
> +static int arith_get_number(ArithCoder *c, int mod_val)
I guess you are not interested in this part, but probably those (and
mostly those) could benefit from some av_inline, but most compilers
probably get it right on their own.
> + memset(neighbours, src[-1], 4);
Something like AV_WN32A(neighbours, src[-1]*0x01010101) maybe? I guess
other parts would be more in need than this but still...
I haven't checked even remotely if there is any issue with incorrect
bitstreams, which would probably be a more interesting review.
--
Christophe
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
