Compute and Workgroups

Compute shaders have run small WGSL experiments since WGSL Types and culled a shelf in Indirect Draws and Bundles. This section covers how a dispatch is divided among the GPU's cores, how invocations share memory and wait for each other, what goes wrong when they do not, and how results return to JavaScript. Timings come from timestamp queries on the GTX 1650 of a machine shared with other work, so compare ratios rather than absolute times.

Subsections

• Compute Pipelines
• Workgroup Sizes
• Invocation IDs
• Workgroup Memory
• Barriers
• Atomics
• Race Conditions
• Subgroups
• Subgroup Size
• Reading Results
• First Compute Shader