Expand description
Pre-compiled execution plan for kernel execution.
ExecutionPlan separates one-time preparation (kernel compilation, buffer
allocation) from fast repeated execution.
┌─────────────────────────────────────────────────────────┐
│ PREPARATION (one-time) │
│ Schedule → instantiate → compile_kernels → build() │
│ ↓ │
│ ExecutionPlan │
└─────────────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────────────┐
│ EXECUTION (fast path) │
│ dependency-ordered PreparedOp execution │
└─────────────────────────────────────────────────────────┘§Example
ⓘ
let plan = tensor.prepare()?;
plan.execute()?;
let output = plan.output_buffer();Structs§
- Execution
Plan - Pre-compiled execution plan for a computation graph.
- Execution
Plan Builder - Builder for creating ExecutionPlan from schedule data.
- Prepared
Buffer View - Prepared zero-copy buffer view operation.
- Prepared
Copy - Prepared buffer-to-buffer copy operation.
- Prepared
Custom Function - Prepared custom runtime function operation.
- Prepared
Kernel - A pre-compiled kernel ready for execution.
- Runtime
Var - Bound description for one
DefineVarconsumed by a kernel.
Enums§
- Prepared
Op - Prepared execution item.
Functions§
- collect_
runtime_ vars - Walk
rootand collect bounds for every reachableDefineVar.