[opt] Slightly simplify algebraic simplification

[xumingkuan]

Related issue = a continuation of #2332

val_cast_to_uint64() looks cleaner than two if-branches with the same body.

[Click here for the format server]

---

[xumingkuan]

I'm not sure if we need a function to cast integer values to int64/uint64 (i.e., handle signed/unsigned cases together)...

[k-ye]

I'm not sure if we need a function to cast integer values to int64/uint64 (i.e., handle signed/unsigned cases together)...

We already have val_int() and val_uint()?

We do need to add the const version of these functions. For now they return a mutable reference to the underlying data..

[xumingkuan]

I'm not sure if we need a function to cast integer values to int64/uint64 (i.e., handle signed/unsigned cases together)...

We already have val_int() and val_uint()?

These 2 functions only handles signed/unsigned versions separately. I wonder if we want to use only one function when we don't care if the value of the TypedConstant is signed or not.

We do need to add the const version of these functions. For now they return a mutable reference to the underlying data..

They are already const, I think.

[k-ye]

I'm not sure if we need a function to cast integer values to int64/uint64 (i.e., handle signed/unsigned cases together)...

We already have val_int() and val_uint()?

These 2 functions only handles signed/unsigned versions separately. I wonder if we want to use only one function when we don't care if the value of the TypedConstant is signed or not.

Right, sounds like it could be useful

We do need to add the const version of these functions. For now they return a mutable reference to the underlying data..

They are already const, I think.

Ah, i was talking about these:

taichi/taichi/ir/type.h

Lines 402 to 411 in c824a8b

	int32 &val_int32();
	float32 &val_float32();
	int64 &val_int64();
	float64 &val_float64();
	int8 &val_int8();
	int16 &val_int16();
	uint8 &val_uint8();
	uint16 &val_uint16();
	uint32 &val_uint32();
	uint64 &val_uint64();

[xumingkuan]

We do need to add the const version of these functions. For now they return a mutable reference to the underlying data..

They are already const, I think.

Ah, i was talking about these:

taichi/taichi/ir/type.h

Lines 402 to 411 in c824a8b

	int32 &val_int32();
	float32 &val_float32();
	int64 &val_int64();
	float64 &val_float64();
	int8 &val_int8();
	int16 &val_int16();
	uint8 &val_uint8();
	uint16 &val_uint16();
	uint32 &val_uint32();
	uint64 &val_uint64();

Right, I agree that we need these, but maybe copying these functions will make this file too long?

[k-ye]

Right, I agree that we need these, but maybe copying these functions will make this file too long?

That shouldn't be a concern...

[xumingkuan]

This PR should be ready for review now :)