Differences

This shows you the differences between two versions of the page.

--- dev:questions_ulam [2015/05/31 07:21] – [Q: How to change the size of the MFM simulator?] ackley
+++ dev:questions_ulam [2015/06/01 18:11] (current) – [Q: Do smaller grid geometries make the simulator run faster?] xychen
@@ Line 145: / Line 145: @@
 **Thank you**! Now I understand that ''ws.getPick(typeIndex)'' returns an event window index. ''ws.getPick(typeIndex)'' can randomly pick one atom of this type. When I want to select an atom of this type but also has the maximum value of a certain data member, I should use the for loop to check each atom of this type and get this maximum atom. Like here I want to find the ''Request ATOM'' which has the max density value within this EventWindow.
-**A: 31-May-2015 12:51:43AM-0600:** To select an atom based on anything other than its type, ''WindowServices'' can help with the event window indexing, but in general you'll need to write a loop explicitly, rather than using ''scan()''.  But even if so there's some help for doing fair picking among maxima and minima.  (Rest of answer pending!)
+**A: 31-May-2015 12:51:43AM-0600:** To select an atom based on anything other than its type, in general you'll need to write a loop explicitly, rather than using ''scan()'' -- although ''WindowServices'' can help with the event window indexing.
+But even when you do need to loop, we have ''SelectorServices'' to provide some help for simpler tasks like finding and picking fairly among maxima and minima.  Here are two ''SelectorServices'' examples.  The first, ''SSDemo1'', uses ''SelectorServices'' in a relatively general way, while the second, ''SSDemo2'', is a bit optimized for shorter code.
+<code - SSDemo1.ulam>
+/**
+   SSDemo1s share the highest Score among their connected group.
+ */
+element SSDemo1 {
+  // Typedefs
+  typedef Unsigned(8) Score;
+  // Utilities
+  EventWindow ew;
+  Random random;
+  DebugUtils du;
+  // Data members
+  Score val;
+  Void behave() {
+    if (val == 0) // randomize initial vals
+      val = random.bits(val.sizeof);
+    WindowServices ws;    // Scanning support
+    SelectorServices ss;  // Selection support
+    ws.reset(1,4); // Scan all but us (for this example)
+    ss.reset();
+    for (Int idx = ws.next(); idx >= 0; idx = ws.next()) {
+      Atom a = ew[idx];
+      if (a as SSDemo1)
+        ss.maximize(idx, (Int) a.val);
+    }
+    // This code is fairly general, but cumbersome.  Compare SSDemo2
+    if (ss.selectionMade()) {
+      Int sidx = ss.getSelectedKey(); // The site that maximized
+      SSDemo1 f = (SSDemo1) ew[sidx]; // which we know is a SSDemo1
+      if (f.val > 0)                  // ..if they've been initted
+        val = f.val;                  // ..pick up their val
+      du.printContext();              // and report for debugging.
+    }
+  }
+}
+</code>
+<code - SSDemo2.ulam>
+/**
+   SSDemo2s share the highest Score among their connected group.
+ */
+element SSDemo2 {
+  // Typedefs
+  typedef Unsigned(8) Score;
+  // Utilities
+  EventWindow ew;
+  Random random;
+  DebugUtils du;
+  // Data members
+  Score val;
+  Void behave() {
+    if (val == 0) // randomize initial vals
+      val = random.bits(val.sizeof);
+    WindowServices ws;    // Scanning support
+    SelectorServices ss;  // Selection support
+    ws.reset(0,4); // Scan everyone including us
+    ss.reset();
+    for (Int idx = ws.next(); idx >= 0; idx = ws.next()) {
+      Atom a = ew[idx];
+      if (a as SSDemo2)
+        ss.maximize((Int) a.val);  // Don't need the idx this way..
+    }
+    // We scanned ourselves, so we know ss.selectionMade() will be
+    // true.  So we just take the chosen value (even if it was ours).
+    val = (Score) ss.getSelectedValue();
+    du.printContext();
+  }
+}
+</code>
 ==== Q: How to change the size of the MFM simulator? ====
@@ Line 168: / Line 259: @@
 **Thank you!** That's better than drawing ''Walls'' by hand. The right side is the resized MFM simulator by {2C2}. And isn't the smaller simulator running faster?
-{{:dev:mfms_signalrequest_2.9.png?nolink&400 |}}{{:dev:mfms_resize.png?nolink&400 |}}
-**A: 31-May-2015 01:06:01AM-0600: ** It depends somewhat on your hardware -- in particular, how many //real// cores you have, but yes, in general a smaller grid geometry -- fewer and smaller tiles, down to some limit -- will typically be faster than more and larger tiles.  Any single tile geometry will be running single-threaded, though, so in some cases multiple smaller tiles will be faster than one bigger one -- you just need to experiment.   Also, if you type **''a''** a few times after typing **''i''** in the simulator, it will display the AER it is producing, so you can get a rough sense of what's faster and slower.
+**A: 31-May-2015 01:06:01AM-0600: ** It depends somewhat on your hardware -- in particular, how many //real// cores you have, but yes, in general a smaller grid geometry -- fewer and smaller tiles, down to some limit -- will typically be faster than more and larger tiles.  Any single tile geometry will be running single-threaded, though, so in some cases multiple smaller tiles will be faster than one bigger one -- you'll need to experiment some, if maximizing AER is crucial.   Also, if you type **''a''** a few times after typing **''i''** in the simulator, it will display the AER it is producing, so you can get a rough sense of what's faster and slower.
 On a separate point, note that the "edge of the grid" is somewhat different that a ring of ''Wall''s.  For example, non-existent sites return ''false'' from the ''EventWindow isLive'' method, but sites containing ''Wall''s return true.  For consistency you might want to draw a ring of ''Wall'' around the boundary of even a smaller grid.
+{{:dev:mfms_signalrequest_2.9.png?nolink&400 |}}{{:dev:mfms_resize.png?nolink&400 |}}