[PATCH 1/4] core/conv: Add optimized Viterbi decoding

[Holger Hans Peter Freyther]

On Tue, Apr 29, 2014 at 12:12:09AM -0400, Thomas Tsou wrote:

Hi,

> Add a separate, faster convolution decoding implementation for rates
> up to N=4 and constraint lengths of K=5 and K=7, which covers the
> most GSM code uses. The decoding algorithm exploits the symmetric
> structure of the Viterbi add-compare-select (ACS) operation - commonly
> known as the ACS butterfly. This shift-register optimization can be
> found in the well-known text by Dave Forney.

I am not knowledgable enough to comment on the actual viterbi things
so I will focus on the things around it.

> +/* Aligned Memory Allocator
> + *     SSE requires 16-byte memory alignment. We store relevant trellis values
> + *     (accumulated sums, outputs, and path decisions) as 16 bit signed integers
> + *     so the allocated memory is casted as such.
> + */
> +#define SSE_ALIGN	16
> +
> +static int16_t *vdec_malloc(size_t n)
> +{
> +#ifdef HAVE_SSE3
> +	return (int16_t *) memalign(SSE_ALIGN, sizeof(int16_t) * n);
> +#else
> +	return (int16_t *) malloc(sizeof(int16_t) * n);
> +#endif
> +}

argh, it would be nice if you could use talloc here but then we would
need to play games and align pointers ourselves. Maybe change the API
to at least have a 'ctx' similar to other talloc API?

> +static void free_trellis(struct vtrellis *trellis)
> +{
> +	if (!trellis)
> +		return;
> +
> +	free(trellis->vals);
> +	free(trellis->outputs);
> +	free(trellis->sums);
> +	free(trellis);


Can you use talloc here?

> +		_traceback_rec(dec, state, out, len);

_ is reserved for the system. We might want to avoid using that.

cheers

	holger