chore(benchmarks): add benchmarks for niop opcode in all operation modes

[rymnc]

Added

A new benchmarking group for niop operations in the alu.rs file. The changes add comprehensive benchmarks for various operations (ADD, SUB, MUL, EXP, SLL, XNOR) across different operand widths (U8, U16, U32)

Additions to benchmarking:

• A new niop benchmarking group was added to evaluate the performance of narrow integer operations (niop) with different operand widths (U8, U16, U32).
• Introduced the niop_bench macro to simplify the creation of benchmarks for operations like ADD, SUB, MUL, EXP, SLL, and XNOR.
• Benchmarks iterate over generated operand pairs for each width, ensuring a variety of test cases for each operation.

Breaking

• Used a new fuel-vm 0.64.0 release which brings breaking changes: https://github.com/FuelLabs/fuel-vm/releases/tag/v0.64.0
• The default gas cost has been changed according to new optimizations and a new benchmark machine for a local node.
• Receipts now are behind an Arc pointer to avoid unnecessary cloning. It affects transaction states and preconfirmations.
• The deploy_contract_with_id now works with [u8] instead of the Contract type.

Changed

Actualized benchmarks with the performance of the network. With this change we should be able to run benchmarks on nightly bases.

Added several optimizations to the code base:

• Executor now reuses the memory instance between blocks production and validation avoiding allocations.
• Receipts are not cloned anymore from the VM. Instead, they are extracted from the VM and stored ina s shared pointer.
• The gas usage for script is discovered in a faster way.
• Create transaction doesn't require a cloning of the contract bytecode anymore to calculate its ContractId.
• Serialization and deserialization of the bytes became much faster for canonical, postcard, bincode codecs.

[MitchTurner]

s there somewhere I can read up on what niop is? I've been out of the loop on it.

edit: The description helps haha:

A new niop benchmarking group was added to evaluate the performance of narrow integer operations (niop) with different operand widths (U8, U16, U32).

Still unfamiliar with what it's for, but this helps me approximate :)

[MitchTurner]

nit: I think this would be more readable as niop_benchmark_group. It's confusing passing niop to something named niop_bench

[rymnc]

addressed in ee93d38

[rymnc]

https://docs.fuel.network/docs/specs/fuel-vm/instruction-set/#niop-narrow-integer-operation

[xgreenx]

Hmm, how we convert results of these benchmark into the actual cost? Because we have only one gas cost which we need to fill, but we have 18 benchmarks for it. I see how we can use it in sanity checks, but I don't understand how does it work for the collect binary=D

[rymnc]

Hmm, how we convert results of these benchmark into the actual cost? Because we have only one gas cost which we need to fill, but we have 18 benchmarks for it. I see how we can use it in sanity checks, but I don't understand how does it work for the collect binary=D

checking

[rymnc]

Hmm, how we convert results of these benchmark into the actual cost? Because we have only one gas cost which we need to fill, but we have 18 benchmarks for it. I see how we can use it in sanity checks, but I don't understand how does it work for the collect binary=D

the collect binary groups benchmark groups and takes the average from them i believe.

fuel-core/benches/src/bin/collect.rs

Lines 588 to 605 in ee93d38

	fn into_relative_costs(self, mut costs: Costs) -> Costs {
	let baseline = self.get_baseline();
	let State {
	mut ids, groups, ..
	} = self;
	let relative = groups.into_iter().filter_map(|(name, samples)| {
	let relative = samples
	.into_iter()
	.find(|sample| sample.starts_with(&name) && ids.contains_key(sample))
	.and_then(|sample| ids.remove(&sample))
	.map(|mean| Cost::Relative(map_to_ratio(baseline, mean)))?;
	Some((name, relative))
	});
	costs.0.extend(relative);
	costs
	}