Show HN: Learn C and its lower levels interactively, in the browser

But more importantly, every MVP is going to lack certain features that some folks deem essential. But when we focus on missing features, or worse, miss the point of a project ("you built X but what I really want is Y"), we discourage them or others from trying/sharing future experiments.

Hyperbole; it's applicable in specific circumstances.

There are ways to code without restrict to get the same performance.

Sample code:

  struct node {
    struct node *next, *prev;
  };

  /* Original: pred and succ must not overlap, but this is not expressed. */

  void insert_after_A(struct node *pred, struct node *succ)
  {
    succ->prev = pred;
    succ->next = pred->next;
    pred->next->prev = succ;
    pred->next = succ;
  }

  /* Optimize with restrict: pred and succ do not overlap. */

  void insert_after_B(struct node *restrict pred, struct node *restrict succ)
  {
    succ->prev = pred;
    succ->next = pred->next;
    pred->next->prev = succ;
    pred->next = succ;
  }

  /* Optimize by hand in "load-store style". */

  void insert_after_C(struct node *pred, struct node *succ)
  {
    struct node *oldsucc = pred->next;
    succ->prev = pred;
    succ->next = aft;
    oldsucc->prev = succ;
    pred->next = succ;
  }

  insert_after_A:
          movl  4(%esp), %eax    ;; %eax = pred
          movl  8(%esp), %edx    ;; %edx = succ
          movl  (%eax), %ecx     ;; %ecx = pred->next
          movl  %eax, 4(%edx)    ;; {%edx:succ}->prev = pred
          movl  %ecx, (%edx)     ;; {%edx:succ}->next = {%ecx:pred}->next
          movl  (%eax), %ecx     ;; %ecx = pred->next !!! deja vu: this load was done a few instructions back!
          movl  %edx, (%eax)     ;; {%eax:next}->next = {%edx:succ} 
          movl  %edx, 4(%ecx)    ;; {%ecx:pred->next}->prev = {%edx:succ} 
          ret

  insert_after_B:
          movl  4(%esp), %edx
          movl  8(%esp), %eax
          movl  (%edx), %ecx
          movl  %edx, 4(%eax)
          movl  %eax, (%edx)
          movl  %ecx, (%eax)
          movl  %eax, 4(%ecx)
          ret

  insert_after_C:
          movl  4(%esp), %edx
          movl  8(%esp), %eax
          movl  (%edx), %ecx
          movl  %edx, 4(%eax)
          movl  %ecx, (%eax)
          movl  %eax, 4(%ecx)
          movl  %eax, (%edx)
          ret

Function C works by caching the pred->next value in a local variable and referring to that.

None of the assignments through the structure type can possibly affect the value of aft; the structures cannot overlap with the local variable. (This is an implicit non-overlap restriction similar to what restrict expresses for the two arguments.)

Once we establish aft, all of the pointers involved in the function are local variables; so none of the local->memb = val assignments raise any suspicion that the value of local has been overwritten, requiring it to be reloaded from memory. We code five accesses and got five (plus the two to load the arguments from the stack, making seven).

Function B has the disadvantage that the behavior becomes undefined if pred and succ are pointers to the same node. Function C has no such problem. Even though the code is just as good as for B, the behavior is defined for overlapping pred and succ.

restrict is C trying to keep up with Fortran. What are situations when we can't use this type of load-store coding to reduce memory operations? Why, array processing!

Well, of course we can take the same approach in array processing; but the problem is that array processing is automatically unrolled by the compiler. Unrolling is hampered in some situations when we don't know whether the arrays overlap. If we simply introduce local variables into the loop, it still won't be un...