Optimize mul_cmp

Author: Raimo33

src/modinv32_impl.h

@@ -49,8 +49,10 @@ static int secp256k1_modinv32_mul_cmp_30(const secp256k1_modinv32_signed30 *a, i
         VERIFY_CHECK(bm.v[i] >> 30 == 0);
     }
     for (i = 8; i >= 0; --i) {
-        if (am.v[i] < bm.v[i]) return -1;
-        if (am.v[i] > bm.v[i]) return 1;
+        diff = (am.v[i] > bm.v[i]) - (am.v[i] < bm.v[i]);
+        if (diff != 0) {
+            return diff;
+        }
     }
     return 0;
 }
@@ -77,6 +79,8 @@ static void secp256k1_modinv32_normalize_30(secp256k1_modinv32_signed30 *r, int3
     VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(r, 9, &modinfo->modulus, 1) < 0); /* r < modulus */
 #endif
 
+    /* TODO: parallelize */
+
     /* In a first step, add the modulus if the input is negative, and then negate if requested.
      * This brings r from range (-2*modulus,modulus) to range (-modulus,modulus). As all input
      * limbs are in range (-2^30,2^30), this cannot overflow an int32_t. Note that the right

src/modinv64_impl.h

@@ -65,8 +65,10 @@ static int secp256k1_modinv64_mul_cmp_62(const secp256k1_modinv64_signed62 *a, i
         VERIFY_CHECK(bm.v[i] >> 62 == 0);
     }
     for (i = 4; i >= 0; --i) {
-        if (am.v[i] < bm.v[i]) return -1;
-        if (am.v[i] > bm.v[i]) return 1;
+        int diff = (am.v[i] > bm.v[i]) - (am.v[i] < bm.v[i]);
+        if (diff != 0) {
+            return diff;
+        }
     }
     return 0;
 }
@@ -101,6 +103,8 @@ static void secp256k1_modinv64_normalize_62(secp256k1_modinv64_signed62 *r, int6
     VERIFY_CHECK(secp256k1_modinv64_mul_cmp_62(r, 5, &modinfo->modulus, 1) < 0); /* r < modulus */
 #endif
 
+    /* TODO: parallelize */
+
     /* In a first step, add the modulus if the input is negative, and then negate if requested.
      * This brings r from range (-2*modulus,modulus) to range (-modulus,modulus). As all input
      * limbs are in range (-2^62,2^62), this cannot overflow an int64_t. Note that the right