[lwip-devel] pbuf_alloc optimization

[Charles Landau]

Looking at pbuf_alloc in LWIP 1.3, it could be made faster and smaller 
with some judicious inlining:

pbuf.h:

static inline
struct pbuf * pbuf_alloc(pbuf_layer l, u16_t size, pbuf_type type)
{
  unsigned int offset = 0;
  switch (layer) {
  case PBUF_TRANSPORT:
    offset += PBUF_TRANSPORT_HLEN;
  case PBUF_IP:
    offset += PBUF_IP_HLEN;
  etc.
  }
  switch (type) {
  case PBUF_POOL:
    return pbuf_alloc_pool(offset, size);
  case PBUF_RAM:
    return pbuf_alloc_ram(offset, size);
  case PBUF_ROM:
  case PBUF_REF:
    return pbuf_alloc_romref(offset, size, type);
  etc.
  }
}

Since the layer and type parameters are always constant, the compiler 
can optimize the call into a single call of pbuf_alloc_<type>, 
eliminating the two switch's. I believe gcc is smart enough to do so.