GPU process - detailed architecture overview

Startup

the entry point for the GPU process

Loop

MainMessageLoop

is used to process events for a particular thread
puts the incoming messages, tasks to a queue
pops a task from the queue and starts it
strong relationship with the IPC communication framework
has task reentrancy protection

second task cannot be started until first task finishes

IPC

framework which is used for inter-process communication
connects directly to the MainMessageLoop
provides communication channels through which the messages can be sent
message creating, sending and receiving
asynchronous message handling

TaskAnnotator

all incoming tasks are going through a TaskAnnotator which annotates the task before the execution
implements common debug annotations for posted tasks. This includes data such as task origins, queueing durations and memory usage
runs a previously queued task

GPU Command Buffer

implements IPC methods (Receive, Send)
handles messages (Set/Get buffer, Flush, Create/Destroy Images, etc) and sends commands

gfx::GLSurface

encapsulates a surface that can be rendered to with GL, hiding platform specific management

gfx::GLContext

encapsulates an OpenGL context, hiding platform specific management

GLES2 Decoder

decodes GLES2 commands from Command Buffer
calls GL methods

Shader, Texture, …, Draw elements

calls actual OpenGL functions
compiles and executes shader codes
manipulates textures (bind, remove, setTarget, etc)
other calls manage the GLContext’s state

Swap buffers

handles frame swaps
if the buffer is offscreen then it copies the rendered frame to another framebuffer

Sources: