Add the largest float representable as integer (signed/unsigned) to F…

…32 testing (execution) set (gpuweb#3942)

Co-authored-by: Peter McNeeley <[email protected]>

src/webgpu/util/math.ts

@@ -961,6 +961,17 @@ export function scalarF32Range(
   counts.neg_norm = counts.neg_norm === undefined ? counts.pos_norm : counts.neg_norm;
   counts.neg_sub = counts.neg_sub === undefined ? counts.pos_sub : counts.neg_sub;
 
+  let special_pos: number[] = [];
+  // The first interior point for 'pos_norm' is at 3. Because we have two special values we start allowing these
+  // special values as soon as they will fit as interior values.
+  if (counts.pos_norm >= 4) {
+    special_pos = [
+      // Largest float as signed integer
+      0x4effffff,
+      // Largest float as unsigned integer
+      0x4f7fffff,
+    ];
+  }
   // Generating bit fields first and then converting to f32, so that the spread across the possible f32 values is more
   // even. Generating against the bounds of f32 values directly results in the values being extremely biased towards the
   // extremes, since they are so much larger.
@@ -980,7 +991,14 @@ export function scalarF32Range(
       kBit.f32.positive.subnormal.max,
       counts.pos_sub
     ),
-    ...linearRange(kBit.f32.positive.min, kBit.f32.positive.max, counts.pos_norm),
+    ...[
+      ...linearRange(
+        kBit.f32.positive.min,
+        kBit.f32.positive.max,
+        counts.pos_norm - special_pos.length
+      ),
+      ...special_pos,
+    ].sort((n1, n2) => n1 - n2),
   ].map(Math.trunc);
   return bit_fields.map(reinterpretU32AsF32);
 }