Benchmark: mat4 array caching

Script Preparation code:

function multiply(out, a, b) {
		let b0 = b[0]
		let b1 = b[1]
		let b2 = b[2]
		let b3 = b[3]
		
		out[0] = b0 * a[0] + b1 * a[4] + b2 * a[8]  + b3 * a[12]
		out[1] = b0 * a[1] + b1 * a[5] + b2 * a[9]  + b3 * a[13]
		out[2] = b0 * a[2] + b1 * a[6] + b2 * a[10] + b3 * a[14]
		out[3] = b0 * a[3] + b1 * a[7] + b2 * a[11] + b3 * a[15]

		b0 = b[4]
		b1 = b[5]
		b2 = b[6]
		b3 = b[7]
		
		out[4] = b0 * a[0] + b1 * a[4] + b2 * a[8]  + b3 * a[12]
		out[5] = b0 * a[1] + b1 * a[5] + b2 * a[9]  + b3 * a[13]
		out[6] = b0 * a[2] + b1 * a[6] + b2 * a[10] + b3 * a[14]
		out[7] = b0 * a[3] + b1 * a[7] + b2 * a[11] + b3 * a[15]

		b0 = b[8]
		b1 = b[9]
		b2 = b[10]
		b3 = b[11]

		out[8]  = b0 * a[0] + b1 * a[4] + b2 * a[8]  + b3 * a[12]
		out[9]  = b0 * a[1] + b1 * a[5] + b2 * a[9]  + b3 * a[13]
		out[10] = b0 * a[2] + b1 * a[6] + b2 * a[10] + b3 * a[14]
		out[11] = b0 * a[3] + b1 * a[7] + b2 * a[11] + b3 * a[15]

		b0 = b[12]
		b1 = b[13]
		b2 = b[14]
		b3 = b[15]

		out[12] = b0 * a[0] + b1 * a[4] + b2 * a[8]  + b3 * a[12]
		out[13] = b0 * a[1] + b1 * a[5] + b2 * a[9]  + b3 * a[13]
		out[14] = b0 * a[2] + b1 * a[6] + b2 * a[10] + b3 * a[14]
		out[15] = b0 * a[3] + b1 * a[7] + b2 * a[11] + b3 * a[15]
		
		return out
	}

function multiplyCached(out, a, b) {
  let a00 = a[0],
    a01 = a[1],
    a02 = a[2],
    a03 = a[3];
  let a10 = a[4],
    a11 = a[5],
    a12 = a[6],
    a13 = a[7];
  let a20 = a[8],
    a21 = a[9],
    a22 = a[10],
    a23 = a[11];
  let a30 = a[12],
    a31 = a[13],
    a32 = a[14],
    a33 = a[15];

  // Cache only the current line of the second matrix
  let b0 = b[0],
    b1 = b[1],
    b2 = b[2],
    b3 = b[3];
  out[0] = b0 * a00 + b1 * a10 + b2 * a20 + b3 * a30;
  out[1] = b0 * a01 + b1 * a11 + b2 * a21 + b3 * a31;
  out[2] = b0 * a02 + b1 * a12 + b2 * a22 + b3 * a32;
  out[3] = b0 * a03 + b1 * a13 + b2 * a23 + b3 * a33;

  b0 = b[4];
  b1 = b[5];
  b2 = b[6];
  b3 = b[7];
  out[4] = b0 * a00 + b1 * a10 + b2 * a20 + b3 * a30;
  out[5] = b0 * a01 + b1 * a11 + b2 * a21 + b3 * a31;
  out[6] = b0 * a02 + b1 * a12 + b2 * a22 + b3 * a32;
  out[7] = b0 * a03 + b1 * a13 + b2 * a23 + b3 * a33;

  b0 = b[8];
  b1 = b[9];
  b2 = b[10];
  b3 = b[11];
  out[8] = b0 * a00 + b1 * a10 + b2 * a20 + b3 * a30;
  out[9] = b0 * a01 + b1 * a11 + b2 * a21 + b3 * a31;
  out[10] = b0 * a02 + b1 * a12 + b2 * a22 + b3 * a32;
  out[11] = b0 * a03 + b1 * a13 + b2 * a23 + b3 * a33;

  b0 = b[12];
  b1 = b[13];
  b2 = b[14];
  b3 = b[15];
  out[12] = b0 * a00 + b1 * a10 + b2 * a20 + b3 * a30;
  out[13] = b0 * a01 + b1 * a11 + b2 * a21 + b3 * a31;
  out[14] = b0 * a02 + b1 * a12 + b2 * a22 + b3 * a32;
  out[15] = b0 * a03 + b1 * a13 + b2 * a23 + b3 * a33;
  return out;
}

Tests:

With cache
let result = [0, 0, 0, 0] multiplyCached(result, [123.12, 131.31, 143.12, 0.123], [143.12, 0.123, 123.12, 131.31])
Without cache
let result = [0, 0, 0, 0] multiply(result, [123.12, 131.31, 143.12, 0.123], [143.12, 0.123, 123.12, 131.31])

Rendered benchmark preparation results:

Suite status: <idle, ready to run>

Previous results

Test case name	Result
With cache
Without cache

Fastest: N/A

Slowest: N/A

Latest run results:

No previous run results

This benchmark does not have any results yet. Be the first one to run it!

Autogenerated LLM Summary (model llama3.2:3b, generated one year ago):

LLMs can make mistakes. Check important info.

A challenging benchmarking problem!

After analyzing the provided code, I'll attempt to break down and explain it.

**Code Analysis**

The code appears to be a JavaScript function `multiplyCached` (or `multiply`) that multiplies two arrays: `[a00, ..., a33]` and `[b0, ..., b3]`. The multiplication is done using the standard matrix multiplication algorithm. The difference between `multiplyCached` and `multiply` lies in the caching mechanism.

In `multiplyCached`, there are 16 iterations (from `out[0]` to `out[15]`) that perform the same calculation: multiplying each row of `[a00, ..., a33]` with each column of `[b0, ..., b3]`. These calculations seem to be stored in cache (`b0` to `b15`) and reused across iterations.

In contrast, `multiply` does not use caching. Each iteration performs the full calculation from scratch.

**Benchmarking**

The benchmarking data suggests that `multiplyCached` outperforms `multiply`, with a significant difference in executions per second (6051233.5 vs 4273649.5).

This outcome is likely due to the caching mechanism in `multiplyCached`. By reusing previously computed values, it avoids redundant calculations and reduces the computational overhead.

**Conclusion**

The provided code demonstrates an optimization technique using caching to improve the performance of matrix multiplication. The benchmarking data supports this claim, showing a substantial speedup for the cached version (`multiplyCached`) compared to the non-cached version (`multiply`).

Related benchmarks:

mat4 array caching (version: 0)

Comparing performance of: With cache vs Without cache

Created: 4 years ago by: Guest

Jump to the latest result

With cache

Without cache

Suite status: <idle, ready to run>

Fastest: N/A

Slowest: N/A

No previous run results

Autogenerated LLM Summary (model llama3.2:3b, generated one year ago):