pub struct ExtensionExecutionContext<'a, B: TensorBackend> { /* private fields */ }Expand description
Backend and cache state passed to one extension execution.
Implementations§
Source§impl<'a, B: TensorBackend> ExtensionExecutionContext<'a, B>
impl<'a, B: TensorBackend> ExtensionExecutionContext<'a, B>
Sourcepub fn new(backend: &'a mut B, caches: &'a mut ExtensionCacheStore) -> Self
pub fn new(backend: &'a mut B, caches: &'a mut ExtensionCacheStore) -> Self
Build a context from externally-owned backend and cache state.
Sourcepub fn backend_mut(&mut self) -> &mut B
pub fn backend_mut(&mut self) -> &mut B
Borrow the backend mutably for extension execution.
Sourcepub fn caches(&self) -> &ExtensionCacheStore
pub fn caches(&self) -> &ExtensionCacheStore
Borrow the extension runtime cache store.
Sourcepub fn caches_mut(&mut self) -> &mut ExtensionCacheStore
pub fn caches_mut(&mut self) -> &mut ExtensionCacheStore
Borrow the extension runtime cache store mutably.
Sourcepub fn execute_core_exec_program_unsegmented(
&mut self,
program: &ExecProgram,
inputs: Vec<Tensor>,
) -> Result<Vec<Tensor>>where
B: 'static,
pub fn execute_core_exec_program_unsegmented(
&mut self,
program: &ExecProgram,
inputs: Vec<Tensor>,
) -> Result<Vec<Tensor>>where
B: 'static,
Execute a core-only execution program one instruction at a time.
This is for extension runtimes that lower their own operation into a
temporary ExecProgram containing only core tensor ops. Nested
ExecOp::Extension instructions are rejected so extension dispatch
cannot bypass the owning runtime registry.
§Examples
use tenferro_cpu::CpuBackend;
use tenferro_ops::dim_expr::DimExpr;
use tenferro_runtime::extension::{ExecInstruction, ExecOp, ExecProgram};
use tenferro_runtime::{DType, ExtensionCacheStore, ExtensionExecutionContext, Tensor};
let program = ExecProgram {
instructions: vec![ExecInstruction {
op: ExecOp::Add,
input_slots: vec![0, 1],
output_slots: vec![2],
dtype: DType::F64,
output_shapes: vec![vec![]].into(),
output_extents: vec![vec![]].into(),
last_use: vec![true, true],
}],
input_slots: vec![0, 1],
output_slots: vec![2],
n_slots: 3,
};
let lhs = Tensor::from_vec_col_major(vec![], vec![1.0_f64]).unwrap();
let rhs = Tensor::from_vec_col_major(vec![], vec![2.0_f64]).unwrap();
let mut backend = CpuBackend::new();
let mut caches = ExtensionCacheStore::new();
let mut ctx = ExtensionExecutionContext::new(&mut backend, &mut caches);
let outputs = ctx
.execute_core_exec_program_unsegmented(&program, vec![lhs, rhs])
.unwrap();
assert_eq!(outputs[0].as_slice::<f64>().unwrap(), &[3.0]);Sourcepub fn parts_mut(&mut self) -> (&mut B, &mut ExtensionCacheStore)
pub fn parts_mut(&mut self) -> (&mut B, &mut ExtensionCacheStore)
Borrow backend and extension cache store as disjoint mutable parts.
Trait Implementations§
Auto Trait Implementations§
impl<'a, B> Freeze for ExtensionExecutionContext<'a, B>
impl<'a, B> !RefUnwindSafe for ExtensionExecutionContext<'a, B>
impl<'a, B> Send for ExtensionExecutionContext<'a, B>where
B: Send,
impl<'a, B> Sync for ExtensionExecutionContext<'a, B>where
B: Sync,
impl<'a, B> Unpin for ExtensionExecutionContext<'a, B>
impl<'a, B> UnsafeUnpin for ExtensionExecutionContext<'a, B>
impl<'a, B> !UnwindSafe for ExtensionExecutionContext<'a, B>
Blanket Implementations§
Source§impl<T> BorrowMut<T> for Twhere
T: ?Sized,
impl<T> BorrowMut<T> for Twhere
T: ?Sized,
Source§fn borrow_mut(&mut self) -> &mut T
fn borrow_mut(&mut self) -> &mut T
Mutably borrows from an owned value. Read more