Binary op blockers

[VirdhatchaniKN]

Blocker issues discussed :

• Binary Bitwise ops
• Binary Shift operators
• LCM, GCD

Binary Bitwise ops

---

Op Requirement :

• ElemwiseBinary/ELEMWISE_BINARY_BITWISEAND
• ElemwiseBinary/ELEMWISE_BINARY_BITWISEOR
• ElemwiseBinary/ELEMWISE_BINARY_BITWISEXOR

Issues :

Unary ops implemented #9437 , #9436 These ops were directly implemented using their respective operators ( ~ , & , | ) . But the same method cannot be used in binary as these operators are not available outside the kernel.

Binary Shift operators

---

Op Requirement :

• ElemwiseBinary/ELEMWISE_BINARY_LEFT_SHIFT
• ElemwiseBinary/ELEMWISE_BINARY_RIGHT_SHIFT

Issues :

Initial Implementation done in #8636 This implementation requires the involvement of a lot of ops, which may lead to performance issues.

LCM, GCD

---

Op Requirement :

• ElemwiseBinary/ELEMWISE_BINARY_LCM
• ElemwiseBinary/ELEMWISE_BINARY_GCD

Issues :

Implementing one of these ops, can make it easier for the other implementation as their are dependant .We have come up two approaches for GCD

1. Using for loop

   def custom_gcd(a, b):
   def gcd(x, y):
       min_val = x if x < y else y
       for i in range(min_val, 0, -1):
           if x % i == 0 and y % i == 0:
               return i
       return 1
   
   return np.array([gcd(int(x), int(y)) for x, y in zip(a, b)])

2. Recursive method

   def custom_gcd(a, b):
   def gcd(x, y):
       if y == 0:
           return x
       else:
           return gcd(y, x % y)
   
   return np.array([gcd(x, y) for x, y in zip(a, b)])

3. Nested for loops

But these methods would involve repetitive loops/calling of a function.

[umadevimcw]

@rtawfik01 We are in need of your inputs on above mentioned ops