Module dsl 

Rust DSL functions for writing CUDA kernels.

This module provides Rust functions that map to CUDA intrinsics during transpilation. These functions have CPU fallback implementations for testing but are transpiled to the corresponding CUDA operations when used in kernel code.

Thread/Block Index Access

use ringkernel_cuda_codegen::dsl::*;

fn my_kernel(...) {
    let tx = thread_idx_x();  // -> threadIdx.x
    let bx = block_idx_x();   // -> blockIdx.x
    let idx = bx * block_dim_x() + tx;  // Global thread index
}

Thread Synchronization

sync_threads();  // -> __syncthreads()

Functions

block_dim_x

Get the block dimension (x dimension). Transpiles to: blockDim.x

block_dim_y

Get the block dimension (y dimension). Transpiles to: blockDim.y

block_dim_z

Get the block dimension (z dimension). Transpiles to: blockDim.z

block_idx_x

Get the block index within a grid (x dimension). Transpiles to: blockIdx.x

block_idx_y

Get the block index within a grid (y dimension). Transpiles to: blockIdx.y

block_idx_z

Get the block index within a grid (z dimension). Transpiles to: blockIdx.z

grid_dim_x

Get the grid dimension (x dimension). Transpiles to: gridDim.x

grid_dim_y

Get the grid dimension (y dimension). Transpiles to: gridDim.y

grid_dim_z

Get the grid dimension (z dimension). Transpiles to: gridDim.z

sync_threads

Synchronize all threads in a block. Transpiles to: __syncthreads()

thread_fence

Thread memory fence. Transpiles to: __threadfence()

thread_fence_block

Block-level memory fence. Transpiles to: __threadfence_block()

thread_idx_x

Get the thread index within a block (x dimension). Transpiles to: threadIdx.x

thread_idx_y

Get the thread index within a block (y dimension). Transpiles to: threadIdx.y

thread_idx_z

Get the thread index within a block (z dimension). Transpiles to: threadIdx.z