Struct CudaTranspiler 

pub struct CudaTranspiler {
    pub shared_vars: HashMap<String, SharedVarInfo>,
    /* private fields */
}

CUDA code transpiler.

Fields

shared_vars: HashMap<String, SharedVarInfo>

Variables that are SharedTile or SharedArray types.

Implementations

impl CudaTranspiler

pub fn new(config: StencilConfig) -> Self

Create a new transpiler with stencil configuration.

pub fn new_generic() -> Self

Create a new transpiler without stencil configuration.

pub fn with_mode(mode: ValidationMode) -> Self

Create a new transpiler with a specific validation mode.

pub fn for_ring_kernel() -> Self

Create a transpiler configured for ring kernel handler transpilation.

pub fn set_validation_mode(&mut self, mode: ValidationMode)

Set the validation mode.

pub fn shared_memory(&self) -> &SharedMemoryConfig

Get the shared memory configuration.

pub fn transpile_stencil(&mut self, func: &ItemFn) -> Result<String>

Transpile a stencil kernel function.

pub fn transpile_generic_kernel(&mut self, func: &ItemFn) -> Result<String>

Transpile a generic (non-stencil) kernel function.

This generates a __global__ kernel without stencil-specific preamble. The kernel code can use thread_idx_x(), block_idx_x() etc. to access CUDA thread indices directly.

pub fn transpile_ring_kernel( &mut self, handler: &ItemFn, config: &RingKernelConfig, ) -> Result<String>

Transpile a handler function into a persistent ring kernel.

This wraps the handler body in a persistent message-processing loop with control block integration, queue operations, and HLC support.