neighbour based algorithm for threads

It does seem to affect since sim falls into a "stable" configuration
where cells rarely change quite quickly.  I'm thinking of making a
better colouring algorithm to see if I can get better images.

src/base.h

@@ -30,6 +30,18 @@ typedef double f64;
 // 64 byte programs
 #define SIZEOF_PROGRAM (1LU << 6)
 
+#define NUM_PROGRAMS_POW_2 17
+#define NUM_PROGRAMS       (1LU << NUM_PROGRAMS_POW_2)
+#define SIMULATION_SIZE    (SIZEOF_PROGRAM * NUM_PROGRAMS)
+#define THREAD_POOL        4
+
+// GUI stuff
+#define WIDTH  800
+#define HEIGHT 600
+// Our grid will be of length sqrt(NUM_PROGRAMS).  This is the same as
+// 1LU<<(NUM_PROGRAMS_POW_2/2).
+static const size_t GRID_WIDTH = 1LU << (NUM_PROGRAMS_POW_2 / 2);
+
 #endif
 
 /* Copyright (C) 2026 Aryadev Chavali

src/main.c

@@ -17,12 +17,6 @@
 #include "program_iter.h"
 #include "simulation.h"
 
-#define WIDTH  800
-#define HEIGHT 600
-
-// Our grid will be of length sqrt(NUM_PROGRAMS).  This is the same as
-// 1LU<<(NUM_PROGRAMS_POW_2/2).
-static const size_t GRID_WIDTH = 1LU << (NUM_PROGRAMS_POW_2 / 2);
 static const size_t CELL_WIDTH = WIDTH / GRID_WIDTH;
 
 static const char *VALID_OPS = "<>{}-+.,[]";

src/simulation.c

@@ -21,33 +21,71 @@ bool any_threads_using(const struct ThreadState *const states, u64 n)
   return false;
 }
 
+void get_neighbours(u64 index, u64 ret[4])
+{
+  memset(ret, 0xFF, sizeof(*ret) * 4);
+  u64 x   = index / GRID_WIDTH;
+  u64 y   = index % GRID_WIDTH;
+  u64 ptr = 0;
+  if (x > 0)
+  {
+    ret[ptr++] = ((x - 1) * GRID_WIDTH) + y;
+  }
+  if (x < GRID_WIDTH - 1)
+  {
+    ret[ptr++] = ((x + 1) * GRID_WIDTH) + y;
+  }
+  if (y > 0)
+  {
+    ret[ptr++] = (x * GRID_WIDTH) + (y + 1);
+  }
+  if (y < GRID_WIDTH - 1)
+  {
+    ret[ptr++] = (x * GRID_WIDTH) + (y - 1);
+  }
+}
+
 void thread_pick(struct ThreadState *state)
 {
   struct Simulation *sim = state->sim;
+  u64 p1 = 0, p2 = 0;
   while (true)
   {
-    u64 p1 = rand() % (SIMULATION_SIZE / SIZEOF_PROGRAM);
+    p1 = rand() % (SIMULATION_SIZE / SIZEOF_PROGRAM);
     if (any_threads_using(sim->states, p1))
     {
       continue;
     }
-    // TODO: Instead of picking a RANDOM p2, why not choose a neighbour of p1?
-    // This way programs that are likely to replicate are replicating closer to
-    // themselves, and thus have a higher chance of further replication.
-    u64 p2 = rand() % (SIMULATION_SIZE / SIZEOF_PROGRAM);
-    while (p1 * 8 <= ((p2 * 8) + SIZEOF_PROGRAM) &&
-           p2 * 8 <= ((p1 * 8) + SIZEOF_PROGRAM))
+    u64 neighbours[4];
+    get_neighbours(p1, neighbours);
+
+    u64 p2 = -1;
+    for (u64 i = 0; i < 4; ++i)
+    {
+      if (neighbours[i] == UINT64_MAX ||
+          any_threads_using(sim->states, neighbours[i]))
+        continue;
+      p2 = neighbours[i];
+      break;
+    }
+
+    if (p2 < GRID_WIDTH)
+    {
+      break;
+    }
+
+    // Otherwise pick randomly.
+    p2 = rand() % (SIMULATION_SIZE / SIZEOF_PROGRAM);
+    while ((p1 * 8 <= ((p2 * 8) + SIZEOF_PROGRAM) &&
+            p2 * 8 <= ((p1 * 8) + SIZEOF_PROGRAM)) ||
+           any_threads_using(sim->states, p2))
     {
       p2 = rand() % (SIMULATION_SIZE / SIZEOF_PROGRAM);
     }
-    if (any_threads_using(sim->states, p2))
-    {
-      continue;
-    }
-    state->p1 = p1;
-    state->p2 = p2;
     break;
   }
+  state->p1 = p1;
+  state->p2 = p2;
 }
 
 const struct timespec THREAD_DEFAULT_SLEEP = {.tv_sec = 1};

src/simulation.h

@@ -12,12 +12,6 @@
 #include <stdatomic.h>
 #include <threads.h>
 
-#define NUM_PROGRAMS_POW_2 10
-#define NUM_PROGRAMS       (1LU << NUM_PROGRAMS_POW_2)
-#define SIMULATION_SIZE    (SIZEOF_PROGRAM * NUM_PROGRAMS)
-
-#define THREAD_POOL 8
-
 struct ThreadState
 {
   u64 p1, p2;