Skip to main content

get_or_compile_kernel

svod_runtime::kernel_cache

Function get_or_compile_kernel 

Source 
pub fn get_or_compile_kernel<F, E>(
    ast_id: u64,
    device: &str,
    compile_fn: F,
) -> Result<Arc<CachedKernel>, E>
where
    F: FnOnce() -> Result<CachedKernel, E>,
Expand description

Get or compile a kernel by UOp ID and device.

Thread-safe: if multiple threads call this with the same key concurrently, exactly one will compile the kernel, and all others will receive a clone of the Arc to that kernel.

§Arguments

  • ast_id - The UOp ID of the kernel AST (from hash consing)
  • device - Device string (e.g., “CPU”, “CUDA:0”)
  • compile_fn - Function to compile the kernel if not cached

§Returns

Arc to the cached kernel (either from cache or freshly compiled).

§Errors

Returns error if compilation fails