fix: remove unnecessary cast in bit-shift

[guipublic]

Description

Problem*

Resolves #6788

Summary*

Since bit shift are now forced to be done with u8 rhs operand, we do not need to cast it to unsigned type anymore.

Additional Context

Documentation*

Check one:

• No documentation needed.
• Documentation included in this PR.
• [For Experimental Features] Documentation to be submitted in a separate PR.

PR Checklist*

• I have tested the changes locally.
• I have formatted the changes with Prettier and/or cargo fmt on default settings.

[github-actions]

Changes to circuit sizes

Generated at commit: 48b78007243bd8c3f0764161f681339378a9ebdf, compared to commit: 5c00a79d2c93056d07330c350bf7b6efbf81d477

🧾 Summary (10% most significant diffs)

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
binary_operator_overloading	-158 ✅	-56.03%	-198 ✅	-4.44%
bit_shifts_runtime	-478 ✅	-62.24%	-601 ✅	-13.19%

---

Full diff report 👇

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
u16_support	71 (-78)	-52.35%	2,894 (-97)	-3.24%
u128	517 (-120)	-18.84%	4,478 (-152)	-3.28%
binary_operator_overloading	124 (-158)	-56.03%	4,265 (-198)	-4.44%
bit_shifts_runtime	290 (-478)	-62.24%	3,956 (-601)	-13.19%

[github-actions]

Peak Memory Sample

Program	Peak Memory
keccak256	78.48M
workspace	123.79M
regression_4709	422.91M
ram_blowup_regression	1.58G
private-kernel-tail	201.60M
private-kernel-reset	716.87M
private-kernel-inner	291.68M
parity-root	171.94M

[github-actions]

Execution Sample

Program	Execution Time	%
sha256_regression	0.612s	-3%
regression_4709	0.389s	-2%
ram_blowup_regression	4.481s	-2%
rollup-base-public	21.592s	-1%
rollup-base-private	19.617s	-1%
private-kernel-tail	0.695s	0%
private-kernel-reset	1.485s	-5%
private-kernel-inner	0.969s	-1%
parity-root	0.524s	-3%

[github-actions]

Compilation Sample

Program	Compilation Time	%
sha256_regression	1.315s	-6%
regression_4709	0.771s	-5%
ram_blowup_regression	15.123s	-2%
rollup-base-public	117.561s	4%
rollup-base-private	94.722s	-3%
private-kernel-tail	1.065s	0%
private-kernel-reset	7.260s	-12%
private-kernel-inner	2.107s	-3%
parity-root	0.728s	-2%
noir-contracts	88.609s	0%

[TomAFrench]

I'm surprised that we don't automatically remove noop casts when we're inserting them into the dfg. Can you add this to the simplify function?

[jfecher]

There is:

} else if *dst_typ == dfg.type_of_value(value) {
   SimplifiedTo(value)

already in simplify_cast. We should investigate why this isn't sufficient (are we returning before that else if?).

I wonder if I'm reading the description of this PR wrong. It seems the rhs was casted to a u8 before, but now it is an unsigned type to begin with (of any size?). Are we calling pow with a non-u8 rhs now, like a u32?

[guipublic]

Cast are properly simplified, the issue was that the number of bits to shift was casted to the type of the number to be shifted. This is because originally:

1. the shift operation required operands to be of the same type
2. In case of signed operands, we were converting the number of bits to unsigned value for internal computation (the power operation)

When we shifted to u8 operand, the unsigned cast was left, which does not add any constraint per-se but change the type of the operand which indirectly affected the power operation (in the to_bits() function).