Skip to content

adaptive-wasm-loading.md

Latest commit

 

History

History
81 lines (59 loc) · 4.33 KB

adaptive-wasm-loading.md

File metadata and controls

81 lines (59 loc) · 4.33 KB
layout default-layout
title About Adaptive WebAssembly (Wasm) Loading
keywords Dynamsoft Barcode Reader, FAQ, JavaScript, tech basic, wasm, loading
description How to enable -SIMD-Pthread Wasm for accelerated deep learning computation?
needAutoGenerateSidebar false

About Adaptive WebAssembly (Wasm) Loading

What is adaptive Wasm loading?

Dynamsoft Barcode Reader(JavaScript) includes three optimized WebAssembly (Wasm) variants — -Baseline, -Pthread, and -SIMD-Pthread — which can be dynamically loaded based on the runtime environment.
This adaptive loading mechanism ensures the SDK automatically selects the most compatible and highest-performing Wasm module available in each browser, further improving performance in modern environments.

Comparison of the three Wasm variants

Feature Baseline Wasm SIMD Wasm SIMD + Pthread Wasm
Parallelism Single-threaded Single Instruction Multiple Data (CPU vector instruction set) Multi-threaded(via Web Workers + SharedArrayBuffer) + SIMD
Performance Characteristics Simple, limited by single-core performance Leverages CPU vectorized parallelism to speed up data processing Combines SIMD vectorization and multi-core acceleration for maximum performance
Compatibility Supported in all Wasm environments Requires browser support for Wasm SIMD instruction set Requires browser support for both Wasm SIMD and Wasm threads (cross-origin isolation)
Minimum Supported Browser Versions Chrome 78+
Edge 79+
Safari 14.5+
Firefox 68+		 Chrome 91+
Edge 91+
Safari 16.4+
Firefox 89+		 Chrome 91+
Edge 91+
Safari 16.4+
Firefox 89+
Wasm Size 5588 KB 6974 KB 8225 KB
Transfer Size 2.11 MB 2.55 MB 2.81 MB

How to enable -SIMD-Pthread Wasm for accelerated deep learning computation?

To unlock multi-threaded performance with the -SIMD-Pthread Wasm variant, configure your server to enable cross-origin isolation by adding the following HTTP headers to all responses:

Cross-Origin-Opener-Policy: same-origin
Cross-Origin-Embedder-Policy: require-corp

Then, ensure that all SDK resources are served under the same origin.

Once correctly configured, the SDK will automatically detect the environment and load the -SIMD-Pthread Wasm to leverage multi-core acceleration.

Tip

You can verify whether your site is correctly isolated by checking the browser console for cross-origin isolation status or by calling window.crossOriginIsolated in DevTools.

How to manually specify which Wasm variant to load?

By default, the SDK automatically determines the most suitable WebAssembly (Wasm) variant to load based on the browser’s capabilities.
However, developers can manually override this behavior and explicitly specify which Wasm module to load through the wasmLoadOptions property.

Example

Dynamsoft.Core.CoreModule.wasmLoadOptions = {
    wasmType: "ml-simd-pthread",
    pthreadPoolSize: 5,
};

Supported Wasm Types

type WasmType =
  | "baseline"         // Basic single-threaded variant
  | "ml-simd"          // SIMD-optimized variant
  | "ml-simd-pthread"  // Multi-threaded + SIMD optimized variant
  | "auto";            // Automatically select based on environment

Note

Setting wasmType to "auto" (default) allows the SDK to automatically choose the optimal Wasm based on runtime capability detection.

When using "ml-simd-pthread", ensure that cross-origin isolation is properly configured as described in How to enable -SIMD-Pthread Wasm for accelerated deep learning computation.

If the specified Wasm variant is not supported in the current browser, the SDK will gracefully fall back to a compatible variant.

Why isn’t -SIMD-Pthread Wasm enabled by default on iOS?

Due to iOS’s strict memory allocation and management limitations, loading -SIMD-Pthread Wasm can sometimes lead to “out of memory” errors on older devices or iOS versions. To ensure stability and compatibility, the SDK does not load -SIMD-Pthread Wasm by default on iOS. Instead, it automatically falls back to the most suitable Baseline or SIMD variant depending on the environment.