diff --git a/src/cpu.rs b/src/cpu.rs
index acd08a9..13ee486 100644
--- a/src/cpu.rs
+++ b/src/cpu.rs
@@ -548,8 +548,9 @@ mod tests {
     #[test_case(0x00C0, 0x7f, 0, 1; "overflow 2")]
     #[test_case(0x00FF, 0x01, 1, 1; "overflow with half carry")]
     #[test_case(0x000A, 0x0C, 1, 0; "half carry")]
-    #[test_case(0x0018, -0xC, 1, 0; "negative half carry")]
-    #[test_case(0x000E, -0x0F, 1, 1; "underflow with half carry")]
+    #[test_case(0x0018, -0x4, 1, 0; "negative half carry")]
+    #[test_case(0x000F, -0x10, 1, 1; "underflow with half carry")]
+    #[test_case(0x000E, -0x0F, 0, 1; "underflow with no half carry")]
     fn test_load_effective_address_flags(
         starting_sp: u16,
         add_value: i8,
diff --git a/src/cpu/run/arithutil.rs b/src/cpu/run/arithutil.rs
index f7943a7..8ad7fca 100644
--- a/src/cpu/run/arithutil.rs
+++ b/src/cpu/run/arithutil.rs
@@ -24,6 +24,22 @@ impl CarryingAdd<u8, u16> for u16 {
     }
 }
 
+impl CarryingAdd<i8, u16> for u16 {
+    fn add_with_carry(self, rhs: i8) -> (u16, bool, bool) {
+        if rhs >= 0 {
+            return self.add_with_carry(rhs.unsigned_abs());
+        }
+
+        let (total, carry) = self.overflowing_sub(rhs.unsigned_abs().into());
+        // This is a special case of half carry. Though we could probably twiddle the bits to make one
+        // unified half carry check, I'm gonna trust the guy who has debugged a real gameboy.
+        // https://stackoverflow.com/a/7261149/1103734
+        let half_carry = total & 0xF <= self & 0xF;
+
+        (total, half_carry, carry)
+    }
+}
+
 /// `did_8_bit_half_carry` checks whether or not the given addition would have done an 8 bit half carry.
 // NOTE: These generics are not as generic as they could be. The Into<u16> is just a shortcut because we
 //       only use up to u16s
@@ -54,12 +70,40 @@ mod tests {
         assert_eq!((expected, half_carry, full_carry), a.add_with_carry(b));
     }
 
-    #[test_case(0x0000, 0x00, 0, false, false; "no addition")]
-    #[test_case(0x00FF, 0x10, 271, false, true; "overflow")]
-    #[test_case(0x00C0, 0x7F, 319, false, true; "overflow 2")]
-    #[test_case(0x00FF, 0x01, 256, true, true; "overflow with half carry")]
-    #[test_case(0x0F0A, 0x0C, 3862, true, false; "half carry")]
-    fn test_u8_u16_carrying_add(a: u16, b: u8, expected: u16, half_carry: bool, full_carry: bool) {
-        assert_eq!((expected, half_carry, full_carry), a.add_with_carry(b));
+    macro_rules! u16_u8_positive_cases {
+        ($test: item) => {
+            #[test_case(0x0000, 0x00, 0, false, false; "no addition")]
+            #[test_case(0x00FF, 0x10, 271, false, true; "overflow")]
+            #[test_case(0x00C0, 0x7F, 319, false, true; "overflow 2")]
+            #[test_case(0x00FF, 0x01, 256, true, true; "overflow with half carry")]
+            #[test_case(0x0F0A, 0x0C, 3862, true, false; "half carry")]
+            $test
+        };
+    }
+
+    u16_u8_positive_cases! {
+        fn test_u8_u16_carrying_add(a: u16, b: u8, expected: u16, half_carry: bool, full_carry: bool) {
+            assert_eq!((expected, half_carry, full_carry), a.add_with_carry(b));
+        }
+    }
+
+    // our positive testcases should work just the same
+    u16_u8_positive_cases! {
+        // negative numbers, though, have their own cases to consider
+        //
+        // this first testcase is one that I generated by taking the difference of these two lists
+        // (there were many more examples but I picked one)
+        //         In [1]: for sp in range(0xFFFF):
+        //    ...:     for n in range(1, 0xFF):
+        //    ...:         a.append((sp, n))
+        //    ...:         if ((sp - n) & 0xF) <= (sp & 0xF):
+        //    ...:             c.append((sp, n))
+        #[test_case(0x1D21, -0x46, 0x1CDB, false, false; "negative; no carry")]
+        #[test_case(0x000F, -0x10, 0xFFFF, true, true; "negative; underflow")]
+        #[test_case(0x000F, -0x7F, 0xFF90, true, true; "negative; underflow 2")]
+        #[test_case(0x249D, -0x08, 0x2495, true, false; "negative; half carry")]
+        fn test_i8_u16_carrying_add(a: u16, b: i8, expected: u16, half_carry: bool, full_carry: bool) {
+            assert_eq!((expected, half_carry, full_carry), a.add_with_carry(b));
+        }
     }
 }
diff --git a/src/cpu/run/load16.rs b/src/cpu/run/load16.rs
index e229703..68ced79 100644
--- a/src/cpu/run/load16.rs
+++ b/src/cpu/run/load16.rs
@@ -31,26 +31,10 @@ impl InstructionRunner<SixteenBitLoadInstruction> for SixteenBitLoadRunner {
                     .registers
                     .get_single_16bit_register(register::SingleSixteenBit::StackPointer);
 
-                let (new_sp, half_carry, carry) = if offset >= 0 {
-                    let unsigned_offset = u8::try_from(offset)
-                        .expect("failed to put positive (or zero) 8 bit value into unsigned eight bit, which should always work...");
-
-                    current_sp.add_with_carry(unsigned_offset)
-                } else {
-                    let sixteen_bit_offset = u16::try_from(i16::from(offset).abs())
-                        .expect("failed to convert an abs'd 16 bit value to a u16");
-                    // TODO: This could be less gross but it's theo nly place we do it...
-                    let (new_sp, underflow) = current_sp.overflowing_sub(sixteen_bit_offset);
-                    // https://robdor.com/2016/08/10/gameboy-emulator-half-carry-flag/
-                    let half_carry =
-                        ((current_sp & 0xf) + (sixteen_bit_offset & 0xf)) & 0x10 == 0x10;
-
-                    (new_sp, half_carry, underflow)
-                };
+                let (new_sp, half_carry, carry) = current_sp.add_with_carry(offset);
 
                 let half_carry_bit = if half_carry { 1 } else { 0 };
                 let carry_bit = if carry { 1 } else { 0 };
-
                 processor
                     .registers
                     .set_flag_bit(register::Flag::Carry, carry_bit);