Chapter 10 of OpenGL super classic (7th Edition) calculation shader

---

Preface

utilize GPU Of computing power
Unlike other shaders , It can synchronize and share data with each other
No fixed input or output , Do not connect with any other fixed pipeline process , High flexibility , It has functions that other processes do not have

One 、 Use calculation shaders

The calculation shader must be a separate program , You cannot combine other shaders （ The vertices 、 Surface subdivision 、 The geometric 、 fragment ）

1. Execute calculation shader

sketch ： Calculation shaders are executed in the workgroup , There is and only one global workgroup sent to opengl perform , There are multiple local workgroups under the global workgroup , Each local workgroup has its own work item （3D block ）, Each work item can be set in the layout qualifier of the calculation shader
glDispatchCompute() and glDispatchComputeIndirect()： Set the number of local workgroups
glGetIntegeri_v()： Get the working status of the local workgroup , for example ： The maximum size of the global workgroup （ Number of local working groups ）, The volume of a local working group
glGetIntegeriv()： Get the volume of a local workgroup

Calculate the input and output of the shader ：
sketch ： No built-in output , It is a single process pipeline , Built in variables can determine the location of local workgroups and work items （ Indexes ）
gl_WorkGroupID： The index of the local workgroup in the global workgroup , for example ：gl_WorkGroupID.x、gl_WorkGroupID.y、gl_WorkGroupID.z
gl_LocalInvocationID： Index of work items in local workgroups , for example ：gl_LocalInvocationID.x、gl_LocalInvocationID.y、gl_LocalInvocationID.z
gl_NumWorkGroups： Total number of calls to local workgroups
gl_WorkGroupSize： Total number of calls to work items
gl_LocalInvocationIndex： Put... In the work item 3D Variable to 1D Indexes
have access to imageStore、atomics As shader output , Write to shader storage block , for example ： Image inversion

2. Compute shader communication

sketch ： Computational shaders can declare variable communication through a limited number of shared memory qualifiers , The access speed is very fast
Sync ： adopt barrier() function , Pay attention to competitive conditions and barrier() Impact on competitive conditions
Example ： Parallel prefix and 、 Application of depth of field

Compute shader cluster ：
sketch ： An individual's behavior is based only on his perception of other members of the group around him , There is no cooperation between the members on a certain decision , for example ： There is no leader in the fish school
Example ： Clustered paper planes