← Back to team overview

rohc team mailing list archive

Re: Patch to optimize WLSB memory allocation (Was: Support of RFC4996 RoHC-TCP)


 Hello Didier,

 Find my answers bellow in the text.

Sorry for the subject of my initial e-mail, I used the last one by facility ...

 Best Regards,


Le 24/05/2012 21:39, Didier Barvaux a écrit :

    Attached is a patch to optimize wlbc in current branch. Only one
allocation of memroy, not two.

Thank you for the patch. Please find hereafter some inline
comments/questions on your patch.

And, I changed the subject of the email thread to better describe its

  struct c_window
+	uint16_t is_used;    /**<  Whether the window entry is used
or not */ uint16_t sn;     /**<  The Sequence Number (SN) associated
with the entry (used to acknowledge the entry) */
  	uint32_t value;  /**<  The value stored in the window entry */
-	bool is_used;    /**<  Whether the window entry is used or
not */ };

Why changing the position and the type of is_used?

==> Just to align members of the structure, and compact
   the size (16+16+32) rather than (16+(16)+32+32) when not compacted.

  	size_t bits;
  	/// Shift parameter (see 4.5.2 in the RFC 3095)
  	rohc_lsb_shift_t p;
+	/// @brief The window in which numerous previous values of
the encoded value
+	///        are stored to help recreate the value
+	struct c_window window[1];

Why not use flexible arrays from ISO C99?

==> Because I am not sure all old compilers understand and support that, especially older cross-compilers I used. This way is not the best, but I am sure that it is very easy to compile.

You may find documentation in:

==> Thank you for this link.

-	wlsb = malloc(sizeof(struct c_wlsb));
+	wlsb = malloc(sizeof(struct c_wlsb)+(sizeof(struct
c_window)*(window_width-1))); if(wlsb == NULL)
  		rohc_debugf(0, "cannot allocate memory for the W-LSB
object\n"); goto error;
-	bzero(wlsb, sizeof(struct c_wlsb));

OK. At first, I was not sure about the bzero() removal, but all
required struct members are initialized, so that's OK.

==> It's why I suppress the bzero initialisation.


Follow ups