Add promoted const_args to --fold-constant PASS

[YanzhouTang]

As we discussed in Issue #152 , after args being promoted to constant, they are not folded in the next --fold-constant PASS.

This PR add this feature based on simple_loop case.

1. neura.gep: its left operand represents base_pointer, we fold this operand if it is a constant.
2. neura.store: its right operand represents destination address, we fold this operand if it is a constant.

After leveraging --promote-func-arg-to-const and --fold-constant, the final operations are reduced to 20, and II = 5 (equals to RecII = 5)

[tancheng]

the final operations are reduced to 20, and II = 5 (equals to RecII = 5)

What is the original #ops and II?

[YanzhouTang]

the final operations are reduced to 20, and II = 5 (equals to RecII = 5)

What is the original #ops and II?

Original ops are 33, and original II = 7

[tancheng]

the final operations are reduced to 20, and II = 5 (equals to RecII = 5)

What is the original #ops and II?

Original ops are 33, and original II = 7

Cool, then next step is https://github.com/coredac/dataflow/blob/main/test/neura/for_loop/relu_test.mlir, IIRC, its II is 11, let's see whether ur optimization can help it, plz try before and after ur optimization on this test, and include // CHECK the updated II in this test. And we will optimize it further with other optimization if necessary.

After above done, let's move to your complex kernel.

[tancheng]

@YanzhouTang any blocker~?

[YanzhouTang]

@YanzhouTang any blocker~?

Sorry I'm busy with some school matters ddl yesterday. 😭 I've updated a version that rename the lhs_value and rhs_value, as well as test.mlir. I'll go on relu kernel optimization today.

[tancheng]

Why the Github Action stops... Any idea?

[YanzhouTang]

Why the Github Action stops... Any idea?

I found some testcases failed due to renaming in my local version. I'm fixing these cases one by one and I'll make a new commit after all tetscase clean