Implement ListInsert in the LLVM Backend

[Thirumalai-Shaktivel]

No description provided.

[Thirumalai-Shaktivel]

@czgdp1807
I'm planning to implement in this way:

int main() {
  int a[] = {1, 3, 4, 0};
  for (int i = 3; i > 1; i--) {
    a[i] = a[i-1];
  }
  a[1] = 2;
  for (int i = 0; i <=3; i ++) {
    std::cout << a[i] << " ";
  }
  return 0;
}

[Thirumalai-Shaktivel]

Can you please provide some explanation:

• What does get_pointer_to_current_end_point work? (How does it actually calculate the value?)
• How to get the List size?

llvm::Value* arg_size = builder->CreateMul(llvm::ConstantInt::get(context, llvm::APInt(32, type_size)), src_capacity);

Does this work as intended?

• The following retrieves the value at the specific pos in the list, right?
  llvm_utils->create_gep(list, pos)

[czgdp1807]

Well, it needs the following steps IMO,

1. First resize if needed. Since, n is always pointing to the index of the next incoming element you can simply call LLVMList::resize_if_needed.
2. Now start shifting element to the right by one step (you need to generate loop in LLVM for that) starting from the n - 1 position till pos.
3. Now call LLVMList::write_item with list, pos and value.

That should be it.

[Thirumalai-Shaktivel]

Yup, I had the same thought!

[czgdp1807]

What does get_pointer_to_current_end_point do? (How does it actually calculate the value?)

It points to the index of the next incoming element. Say you have 4 elements in the list then its value will be 4 (0, 1, 2, 3 will be the indices of the already existing elements and 4 will be the index of the next incoming list).

How to get the List size?

If you are planning to use size for resizing for inserting the new element then you don't need it. resize_if_needed already does that. capacity is the maximum element of elements a list can accommodate before next doubling + 1, and n (current end point) is the number of elements currently in the list.

[Thirumalai-Shaktivel]

Thank you, I got it!

[czgdp1807]

Well doing the second step in #932 (comment) from n - 1 to pos (in reverse direction) might not be cache efficient. Starting from pos should be better. You just need to store pos + 1's value in a temporary variable. Something like the following,

tmp1 = l[pos]
for i in range(pos, n):
    tmp2 = l[i + 1]
    l[i + 1] = tmp1
    tmp1 = tmp2

I tested the above approach but not so sure. You can see the code here and try it out for yourself before implementing the same logic in LPython (https://ideone.com/JSDhuh, -100 acts like a garbage value).

[Thirumalai-Shaktivel]

I have implemented it as mentioned, but I got stuck in handling the loop, Can you please review and help debug the issue here? The while loop goes to an infinite loop!

[Thirumalai-Shaktivel]

@czgdp1807, this PR is ready, Can you please review it?

[czgdp1807]

Will review tomorrow.

[Thirumalai-Shaktivel]

Thank you very much for your guidance and review.

[certik]

Fine to merge.

We can create some nicer API to create loops in LLVM in some easier way.

[Thirumalai-Shaktivel]

Thank you!

[Thirumalai-Shaktivel]

Fine to merge.

We can create some nicer API to create loops in LLVM in some easier way.

How do I implement API for loops, or if possible, if statements?

[certik]

You implement some nice C++ utility classes or functions, that make using loops/if statements in LLVM easier.

Perhaps something with C++ lambda functions: if(<lambda condition>, <lamba then>, optional <lambda else>), and you just fill in the condition and then / else, and the if function generates the correct boilerplate.

I created a new issue #938 to discuss further.