Optimize away zero-length MCP/MCPI instructions

[Dnreikronos]

Description

MCP with length=0 does nothing at runtime in FuelVM: no panic, no memory change. It resets $err to 0, but every instruction does that anyway. So we can safely strip these out.

This adds two match arms to remove_redundant_ops():

• MCP(_, _, len) where len is $zero (set by constant propagation)
• MCPI(_, _, imm) where the immediate is 0

Since this runs after constant_propagate() and dce(), any register known to be zero is already $zero by the time we get here. Leftover setup instructions (like MOVI len_reg, 0) become dead code, and DCE cleans them up on the next round.

The IR-level path (compile_mem_copy_bytes) already skips zero-length copies at compile time. This catches the cases that only resolve to zero after constant propagation at the ASM level (inline ASM paths, Vec push of empty slices, etc.).

Closes #7624

Checklist

• I have linked to any relevant issues.
• I have commented my code, particularly in hard-to-understand areas.
• I have updated the documentation where relevant (Forc, Sway book, etc.).
• I have added tests that prove my fix is effective or that my feature works.
• I have added (or requested a) parsing test in core/trybuild, if needed.
• I have added the necessary Breaking* or New Feature labels, if relevant.
• I have done my best to ensure that my PR adheres to the Fuel Labs Code Review Standards.
• I have requested a review from the relevant team or maintainers.

Test plan

• mcp_zero_len.sw IR generation test: checks that mcp shows up in the IR but is gone from final ASM (not: mcp filecheck directive)
• All 78 IR generation tests still pass
• Manually confirmed with forc build --asm final on a script using asm { mcp dst src zero; }, no mcp in output

[cursor]

PR Summary

Low Risk
Localized compiler peephole change for documented no-op copies; behavior preserved with tests and lower gas on affected paths.

Overview
Extends Fuel VM remove_redundant_ops so zero-length memory copies are dropped from generated assembly: MCP when the length register is $zero (after constant propagation) and MCPI when the immediate length is 0. These are no-ops at runtime, so removing them shrinks code and saves gas on paths that only become zero-length at the ASM layer (inline asm, empty Vec copies, etc.).

Adds mcp_zero_len.sw to assert copies still appear in IR but not in final ASM, and refreshes vec_encoding_decoding snapshot gas/artifact sizes slightly lower.

^{Reviewed by Cursor Bugbot for commit 8d92e1a. Bugbot is set up for automated code reviews on this repo. Configure here.}

[codspeed-hq]

Merging this PR will not alter performance

✅ 25 untouched benchmarks

---

_{Comparing Dnreikronos:Dnreikronos/optimize-mcp-zero-length (8d92e1a) with master (b556c08)}

[xunilrj]

@Dnreikronos You may need to run cargo fmt and update snapshot tests cargo r -p test -r -- -k snapshot; cargo insta accept

[Dnreikronos]

@Dnreikronos You may need to run cargo fmt and update snapshot tests cargo r -p test -r -- -k snapshot; cargo insta accept

Adjusted!

[fuel-cla-bot]

Thanks for the contribution! Before we can merge this, we need @Dnreikronos to sign the Fuel Labs Contributor License Agreement.

[Dnreikronos]

@xunilrj @ironcev

I'm just trying to sign the "Fuel Labs Contributor License Agreement"

[ironcev]

@Dnreikronos We currently have an issue with the CLA signing. I hope it will be solved soon. Thank you for your contribution and please bear with us until the verification step is fixed.

[Voxelot]

@Dnreikronos the CLA bot should be back in working order. Appreciate your patience and your contribution!

[ironcev]

👍

[ironcev]

@Dnreikronos Did you try to sign the CLA?

[Dnreikronos]

@Dnreikronos Did you try to sign the CLA?

Yes, but I still get the same error:

[Dnreikronos]

Now worked, @ironcev :)