misc/random: add mp_rand_in_range32()

removes awkward back-and-forth int -> double -> int conversion.
the last step also broke uniformity, though it likely doesn't
matter much in mpv's playlist shuffle context.

misc/random.c

@@ -22,6 +22,8 @@
 
 #include <libavutil/random_seed.h>
 
+#include "common/common.h"
+#include "misc/mp_assert.h"
 #include "osdep/timer.h"
 #include "random.h"
 
@@ -84,3 +86,17 @@ double mp_rand_next_double(mp_rand_state *s)
 {
     return (mp_rand_next(s) >> 11) * 0x1.0p-53;
 }
+
+// <https://web.archive.org/web/20250321082025/
+// https://www.pcg-random.org/posts/bounded-rands.html#bitmask-with-rejection-unbiased-apples-method>
+uint32_t mp_rand_in_range32(mp_rand_state *s, uint32_t min, uint32_t max)
+{
+    mp_assert(min < max);
+    uint32_t range = max - min;
+    uint32_t mask = mp_round_next_power_of_2(range) - 1;
+    uint32_t ret;
+    do {
+        ret = mp_rand_next(s) & mask;
+    } while (ret >= range);
+    return min + ret;
+}

misc/random.h

@@ -46,3 +46,9 @@ uint64_t mp_rand_next(mp_rand_state *s);
  * distribution, and update the state accordingly.
  */
 double mp_rand_next_double(mp_rand_state *s);
+
+/*
+ * Return a 32 bit integer in the range [min, max) with uniform distribution.
+ * Caller should ensure `min < max`.
+ */
+uint32_t mp_rand_in_range32(mp_rand_state *s, uint32_t min, uint32_t max);