Refactor abs without logical branching (#4497)

Co-authored-by: Francisco Giordano <fg@frang.io>
Co-authored-by: Hadrien Croubois <hadrien.croubois@gmail.com>
Co-authored-by: ernestognw <ernestognw@gmail.com>

contracts/utils/math/SignedMath.sol

@@ -36,8 +36,15 @@ library SignedMath {
      */
     function abs(int256 n) internal pure returns (uint256) {
         unchecked {
-            // must be unchecked in order to support `n = type(int256).min`
-            return uint256(n >= 0 ? n : -n);
+            // Formula from the "Bit Twiddling Hacks" by Sean Eron Anderson.
+            // Since `n` is a signed integer, the generated bytecode will use the SAR opcode to perform the right shift,
+            // taking advantage of the most significant (or "sign" bit) in two's complement representation.
+            // This opcode adds new most significant bits set to the value of the previous most significant bit. As a result,
+            // the mask will either be `bytes(0)` (if n is positive) or `~bytes32(0)` (if n is negative).
+            int256 mask = n >> 255;
+
+            // A `bytes(0)` mask leaves the input unchanged, while a `~bytes32(0)` mask complements it.
+            return uint256((n + mask) ^ mask);
         }
     }
 }