Reduce allocations in solve_witness_vec

[xrvdg]

38% of the allocations in the prover is was due to solve_witness_vec and this PR brings that down to under 3%. The total amount of allocations is reduced by 5 million and decompose_into_digits now runs twice as fast.

[codspeed-hq]

CodSpeed WallTime Performance Report

Merging #176 will not alter performance

_{Comparing xr/decompose-into-digits (da1130a) with main (b0f12f9)}

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

Summary

✅ 35 untouched benchmarks

[Copilot]

Pull Request Overview

This PR optimizes memory allocation in the prover by refactoring solve_witness_vec and related digital decomposition functions. The changes reduce allocations by 38% in the prover (from over 5 million) and improve decompose_into_digits performance by 2x.

Key changes:

• Moved decompose_into_digits from local modules to common library for reuse
• Replaced eager vector collection with lazy iterators in binary operations
• Optimized bit manipulation using BitIteratorLE instead of vector allocations

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
provekit/r1cs-compiler/src/range_check.rs	Removed unnecessary .clone() calls on function parameters
provekit/r1cs-compiler/src/digits.rs	Removed duplicate decompose_into_digits implementation and helper functions
provekit/r1cs-compiler/src/binops.rs	Replaced vector collections with boxed iterators to reduce allocations
provekit/prover/src/witness/digits.rs	Removed duplicate decompose_into_digits implementation
provekit/prover/src/r1cs.rs	Added instrumentation to solve_witness_vec method
provekit/common/src/witness/mod.rs	Exported decompose_into_digits from common module
provekit/common/src/witness/digits.rs	Added optimized decompose_into_digits implementation using iterators
provekit/common/Cargo.toml	Added itertools dependency

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

This comment suggests the test may be incomplete or incorrect. The comment indicates that changing from big-endian to little-endian conversion didn't affect the test results, which suggests the test might not be validating the endianness properly.

Suggested change

	// Changing from FieldElement :: from_be to from_le didn't negate this test. So
	// this needs to be extended
	fn test_field_to_le_bits() {
	let value = FieldElement::from(5u32);
	let bits: Vec<bool> = field_to_le_bits(value).collect();
	assert_eq!(bits.len(), 256);
	assert!(bits[0]);
	assert!(!bits[1]);
	assert!(bits[2]);
	assert!(!bits[254]);
	assert!(!bits[255]);
	// Extended to check round-trip and more bit positions for endianness correctness
	fn test_field_to_le_bits() {
	// Test value with bits set at various positions
	let value = FieldElement::from(5u32); // 0b101
	let bits: Vec<bool> = field_to_le_bits(value).collect();
	assert_eq!(bits.len(), 256);
	assert!(bits[0]); // 2^0
	assert!(!bits[1]); // 2^1
	assert!(bits[2]); // 2^2
	assert!(!bits[254]);
	assert!(!bits[255]);
	// Test value with a high bit set
	let high_bit_value = FieldElement::from(1u32 << 8); // 2^8
	let high_bits: Vec<bool> = field_to_le_bits(high_bit_value).collect();
	assert!(high_bits[8]);
	for i in 0..8 {
	assert!(!high_bits[i]);
	}
	// Round-trip test: bits -> field -> bits
	let reconstructed = le_bits_to_field(bits.iter().cloned());
	assert_eq!(reconstructed, value);
	let reconstructed_high = le_bits_to_field(high_bits.iter().cloned());
	assert_eq!(reconstructed_high, high_bit_value);