Benchmark: Buffer Access - MeasureThat.net

Script Preparation code:

var bench1buffer;
var bench2buffer;
var bench3buffer;
{
    const stride = 10;
    bench1buffer = new ArrayBuffer(stride * 1000);
    /*
      0 u16 id
      2 f32 x
      6 f32 y
    */
    for (let i = 0; i < 1000; i++) {
        const struct = new DataView(bench1buffer, i * stride, stride);
        struct.setUint16(0, i);
        struct.setFloat32(2, i);
        struct.setFloat32(6, i * 2);
    }
}
{
    const stride = 12;
    bench2buffer = new ArrayBuffer(stride * 1000);
    /*
      0 u16 id
      4 f32 x
      8 f32 y
    */
    const floats = new Float32Array(bench2buffer);
    const u16s = new Uint16Array(bench2buffer);
    for (let i = 0; i < 1000; i++) {
        u16s[i * stride] = i;
        floats[i * stride + 4] = i;
        floats[i * stride + 8] = i * 2;
    }
}
{
    bench3buffer = {
        id: new Uint16Array(1000),
        x: new Float32Array(1000),
        y: new Float32Array(1000),
    };
    for (let i = 0; i < 1000; i++) {
        bench3buffer.id[i] = i;
        bench3buffer.x[i] = i;
        bench3buffer.y[i] = i * 2;
    }
}

Tests:

DataView AoS
const stride = 10; /* 0 u16 id 2 f32 x 6 f32 y */ let sum = 0; for (let i = 0; i < 1000; i++) { const struct = new DataView(bench1buffer, i * stride, stride); const id = struct.getUint16(0); const x = struct.getFloat32(2); const y = struct.getFloat32(6); sum += id + x * y; }
DataView byte-aligned AoS
const stride = 12; /* 0 u16 id 4 f32 x 8 f32 y */ let sum = 0; for (let i = 0; i < 1000; i++) { const struct = new DataView(bench2buffer, i * stride, stride); const id = struct.getUint16(0); const x = struct.getFloat32(4); const y = struct.getFloat32(8); sum += id + x * y; }
TypedArray AoS
const stride = 10; /* 0 u16 id 2 f32 x 6 f32 y */ let floats = new Float32Array(bench1buffer); let u32s = new Uint32Array(bench1buffer); let sum = 0; for (let i = 0; i < 1000; i++) { const offset = i * stride; const id = floats[offset]; const x = floats[offset + 2]; const y = floats[offset + 6]; sum += id + x * y; }
TypedArray byte-aligned AoS
const stride = 12; /* 0 u16 id 4 f32 x 8 f32 y */ let floats = new Float32Array(bench2buffer); let u32s = new Uint32Array(bench2buffer); let sum = 0; for (let i = 0; i < 1000; i++) { const offset = i * stride; const id = floats[offset]; const x = floats[offset + 4]; const y = floats[offset + 8]; sum += id + x * y; }
TypedArray SoA
let sum = 0; for (let i = 0; i < 1000; i++) { const id = bench3buffer.id[i]; const x = bench3buffer.x[i]; const y = bench3buffer.y[i]; sum += id + x * y; }

Rendered benchmark preparation results:

Suite status: <idle, ready to run>

Previous results

Test case name	Result
DataView AoS
DataView byte-aligned AoS
TypedArray AoS
TypedArray byte-aligned AoS
TypedArray SoA

Fastest: N/A

Slowest: N/A

Latest run results:

Run details: (Test run date: 2 years ago)

User agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/98.0.4758.109 Safari/537.36

Browser/OS: Chrome 98 on Mac OS X 10.15.7

View result in a separate tab

Test name	Executions per second
DataView AoS	3758.1 Ops/sec
DataView byte-aligned AoS	3909.9 Ops/sec
TypedArray AoS	129978.1 Ops/sec
TypedArray byte-aligned AoS	145009.3 Ops/sec
TypedArray SoA	5237.6 Ops/sec

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

LLMs can make mistakes. Check important info.

Let's break down the provided benchmark and its options.

**Benchmark Overview**

The benchmark is designed to compare the performance of different data access patterns: Array Of Arrays (AoA), Array Of Structures (SoA), and two variations of TypedArray access patterns.

**Data Structure**

The benchmark creates three data structures:

1. `bench3buffer.id`, `bench3buffer.x`, and `bench3buffer.y` - These are the arrays that will be used to test SoA.
2. The first 1000 elements of a 32-bit integer array (`floats`) in both TypedArray AoA and TypedArray SoA patterns.
3. A 32-bit integer array (`u32s`) in both TypedArray AoA and TypedArray SoA patterns.

**Access Patterns**

The benchmark measures the execution time for each access pattern:

1. **TypedArray byte-aligned AoS**: Accessing elements using contiguous memory blocks, where each element is aligned to a 4-byte boundary.
2. **TypedArray AoA**: Accessing elements using two separate arrays, `floats` and `u32s`, with each array containing the x and y coordinates of an object.
3. **TypedArray SoA**: Accessing elements using a single array, where each element is a struct containing the id, x, and y coordinates.

**Browser Results**

The latest benchmark results show that:

1. TypedArray byte-aligned AoS outperforms all other patterns with an execution time of approximately 145009 executions per second.
2. TypedArray AoA is slower than TypedArray SoA but faster than TypedArray byte-aligned AoS, with an execution time of around 129978 executions per second.
3. TypedArray SoA has the highest execution time among all three patterns, with approximately 5237 executions per second.

**Performance Implications**

These results suggest that when working with large arrays or complex data structures, using contiguous memory access (TypedArray byte-aligned AoS) can lead to significant performance gains compared to other patterns. However, if the data structure is already optimized for contiguous memory access, it may be unnecessary to use a TypedArray pattern.

In summary, this benchmark highlights the importance of considering data access patterns and memory layout when optimizing code performance.

Related benchmarks:

Buffer Access (version: 0)

Comparing performance of: DataView AoS vs DataView byte-aligned AoS vs TypedArray AoS vs TypedArray byte-aligned AoS vs TypedArray SoA

Created: 2 years ago by: Guest

Jump to the latest result

DataView AoS

DataView byte-aligned AoS

TypedArray AoS

TypedArray byte-aligned AoS

TypedArray SoA

Suite status: <idle, ready to run>

Experimental features:

Fastest: N/A

Slowest: N/A

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