perf: optimize shuffle array element iteration with slice-based append

[andygrove]

Summary

Optimizes primitive type array iteration in native shuffle by using bulk operations instead of per-element iteration.

Key optimizations:

• Non-nullable path: Uses append_slice() for optimal memcpy-style copy
• Nullable path: Uses pointer iteration with efficient null bitset reading (reads 64 bits at a time)

Supported types: i8, i16, i32, i64, f32, f64, date32, timestamp

Benchmark Results

Benchmark for converting SparkUnsafeArray (10K elements) to Arrow array:

Type	Baseline	Optimized	Speedup
i32/no_nulls	6.08µs	0.65µs	9.3x
i32/with_nulls	22.49µs	16.21µs	1.39x
i64/no_nulls	6.15µs	1.22µs	5x
i64/with_nulls	16.41µs	16.41µs	1x
f64/no_nulls	8.05µs	1.22µs	6.6x
f64/with_nulls	16.52µs	16.21µs	1.02x
date32/no_nulls	~6µs*	0.66µs	~9x
timestamp/no_nulls	~6µs*	1.21µs	~5x

*Baseline estimated from similar types

Why such dramatic improvement for non-nullable?

The original code appended elements one by one using index-based access:

for idx in 0..array.get_num_elements() {
    builder.append_value(array.get_int(idx));  // get_int does: offset + idx * 4
}

The optimized code uses slice-based bulk append:

let slice = unsafe {
    std::slice::from_raw_parts(self.element_offset as *const i32, num_elements)
};
builder.append_slice(slice);  // Single memcpy-style operation

Test plan

• All Rust tests pass (118 tests)
• Native shuffle test suite passes (16 tests)
• Clippy clean
• Added dedicated benchmark (native/core/benches/array_conversion.rs)

🤖 Generated with Claude Code

[andygrove]

@sqlbenchmark run tpch

[andygrove]

@sqlbenchmark run tpch

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 60.02%. Comparing base (f09f8af) to head (fe54548).
⚠️ Report is 858 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff              @@
##               main    #3222      +/-   ##
============================================
+ Coverage     56.12%   60.02%   +3.89%     
- Complexity      976     1429     +453     
============================================
  Files           119      170      +51     
  Lines         11743    15746    +4003     
  Branches       2251     2602     +351     
============================================
+ Hits           6591     9451    +2860     
- Misses         4012     4976     +964     
- Partials       1140     1319     +179     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[andygrove]

@sqlbenchmark run tpch

[sqlbenchmark]

Comet TPC-H Benchmark Results

Commit: fe54548 - fix: address clippy warnings in benchmark
Scale Factor: SF100
Iterations: 1

Query Times

Query	Time (s)	Query	Time (s)
Q1	10.84	Q12	6.83
Q2	5.88	Q13	6.76
Q3	9.72	Q14	3.56
Q4	11.25	Q15	7.27
Q5	19.02	Q16	4.94
Q6	2.50	Q17	32.00
Q7	11.88	Q18	33.60
Q8	24.26	Q19	6.73
Q9	37.46	Q20	6.87
Q10	10.51	Q21	45.81
Q11	4.26	Q22	5.03

Total Time: 306.97 seconds

Spark Configuration

Setting	Value
Spark Master	local[*]
Driver Memory	32G
Driver Cores	8
Executor Memory	32G
Executor Cores	8
Off-Heap Enabled	true
Off-Heap Size	24g
Shuffle Manager	CometShuffleManager
Comet Replace SMJ	true

---

Automated benchmark run by dfbench

[sqlbenchmark]

Comet TPC-H Benchmark Results

Commit: fe54548 - fix: address clippy warnings in benchmark
Scale Factor: SF100
Iterations: 1

Query Times

Query	Time (s)	Query	Time (s)
Q1	10.75	Q12	6.77
Q2	5.99	Q13	6.88
Q3	9.64	Q14	3.56
Q4	11.48	Q15	7.15
Q5	19.00	Q16	4.98
Q6	2.59	Q17	31.85
Q7	11.91	Q18	33.46
Q8	24.44	Q19	6.69
Q9	37.66	Q20	6.57
Q10	10.38	Q21	45.80
Q11	4.30	Q22	4.88

Total Time: 306.72 seconds

Spark Configuration

Setting	Value
Spark Master	local[*]
Driver Memory	32G
Driver Cores	8
Executor Memory	32G
Executor Cores	8
Off-Heap Enabled	true
Off-Heap Size	24g
Shuffle Manager	CometShuffleManager
Comet Replace SMJ	true

---

Automated benchmark run by dfbench

[sqlbenchmark]

Comet TPC-H Benchmark Results

Commit: fe54548 - fix: address clippy warnings in benchmark
Scale Factor: SF100
Iterations: 1

Query Times

Query	Time (s)	Query	Time (s)
Q1	10.77	Q12	6.80
Q2	5.86	Q13	6.81
Q3	9.62	Q14	3.52
Q4	11.69	Q15	7.17
Q5	18.82	Q16	4.49
Q6	2.53	Q17	31.85
Q7	11.90	Q18	33.30
Q8	24.14	Q19	6.76
Q9	37.42	Q20	6.54
Q10	10.38	Q21	46.16
Q11	4.26	Q22	4.98

Total Time: 305.78 seconds

Spark Configuration

Setting	Value
Spark Master	local[*]
Driver Memory	32G
Driver Cores	8
Executor Memory	32G
Executor Cores	8
Off-Heap Enabled	true
Off-Heap Size	24g
Shuffle Manager	CometShuffleManager
Comet Replace SMJ	true

---

Automated benchmark run by dfbench

[andygrove]

note that I did not expect any improvements in tpc-h since it does not use arrays

[andygrove]

replaced with #3289 so I can work on all optimizations in a single PR for now