2.1 KiB
The tinygrad framework has four pieces
- a PyTorch like frontend.
- a scheduler which breaks the compute into kernels.
- a lowering engine which converts ASTs into code that can run on the accelerator.
- an execution engine which can run that code.
Frontend
Everything in Tensor is syntactic sugar around function.py, where the forwards and backwards passes are implemented for the different mlops. There's about 25 of them, implemented using about 20 basic ops. Those basic ops go on to construct a graph of:
::: tinygrad.lazy.LazyBuffer options: show_source: false
The LazyBuffer graph specifies the compute in terms of low level tinygrad ops. Not all LazyBuffers will actually become realized. There's two types of LazyBuffers, base and view. base contains compute into a contiguous buffer, and view is a view (specified by a ShapeTracker). Inputs to a base can be either base or view, inputs to a view can only be a single base.
Scheduling
The scheduler converts the graph of LazyBuffers into a list of ScheduleItem. One ScheduleItem is one kernel on the GPU, and the scheduler is responsible for breaking the large compute graph into subgraphs that can fit in a kernel. ast specifies what compute to run, and bufs specifies what buffers to run it on.
::: tinygrad.ops.ScheduleItem
Lowering
The code in realize lowers ScheduleItem to ExecItem with
::: tinygrad.engine.realize.lower_schedule
There's a ton of complexity hidden behind this, see the codegen/ directory.
First we lower the AST to UOps, which is a linear list of the compute to be run. This is where the BEAM search happens. The UOps can be changed by CompilerOptions.
::: tinygrad.device.CompilerOptions
Then we render the UOps into code, then we compile the code to binary.
Execution
Creating ExecItem, which has a run method
::: tinygrad.engine.realize.ExecItem options: members: true
Lists of ExecItem can be condensed into a single ExecItem with the Graph API (rename to Queue?)