Crate ringkernel_ir 

RingKernel Intermediate Representation (IR)

This crate provides a unified IR for GPU code generation across multiple backends (CUDA, WGSL, MSL). The IR is SSA-based and captures GPU-specific operations.

Architecture

Rust DSL → IR → Backend-specific lowering → CUDA/WGSL/MSL

Example

use ringkernel_ir::{IrBuilder, IrType, Dimension};

let mut builder = IrBuilder::new("saxpy");

// Define parameters
let x = builder.parameter("x", IrType::Ptr(Box::new(IrType::F32)));
let y = builder.parameter("y", IrType::Ptr(Box::new(IrType::F32)));
let a = builder.parameter("a", IrType::F32);
let n = builder.parameter("n", IrType::I32);

// Get thread index
let idx = builder.thread_id(Dimension::X);

// Bounds check
let in_bounds = builder.lt(idx, n);
builder.if_then(in_bounds, |b| {
    let x_val = b.load(x, idx);
    let y_val = b.load(y, idx);
    let result = b.add(b.mul(a, x_val), y_val);
    b.store(y, idx, result);
});

let ir = builder.build();

Re-exports

pub use lower_cuda::lower_to_cuda;

pub use lower_cuda::lower_to_cuda_with_config;

pub use lower_cuda::CudaLowering;

pub use lower_cuda::CudaLoweringConfig;

pub use lower_cuda::LoweringError;

pub use lower_msl::lower_to_msl;

pub use lower_msl::lower_to_msl_with_config;

pub use lower_msl::MslLowering;

pub use lower_msl::MslLoweringConfig;

pub use lower_msl::MslLoweringError;

pub use lower_wgsl::lower_to_wgsl;

pub use lower_wgsl::lower_to_wgsl_with_config;

pub use lower_wgsl::WgslLowering;

pub use lower_wgsl::WgslLoweringConfig;

pub use lower_wgsl::WgslLoweringError;

pub use optimize::optimize;

pub use optimize::run_constant_folding;

pub use optimize::run_dce;

pub use optimize::AlgebraicSimplification;

pub use optimize::ConstantFolding;

pub use optimize::DeadBlockElimination;

pub use optimize::DeadCodeElimination;

pub use optimize::OptimizationPass;

pub use optimize::OptimizationResult;

pub use optimize::PassManager;

Modules

lower_cuda

IR to CUDA lowering pass.

lower_msl

IR to MSL (Metal Shading Language) lowering pass.

lower_wgsl

IR to WGSL lowering pass.

optimize

Optimization passes for the IR.

Structs

BackendCapabilities

Backend-specific capabilities.

Block

A basic block containing IR nodes.

BlockId

Unique identifier for IR blocks.

Capabilities

Set of capabilities required or available.

Instruction

An IR instruction that produces a value.

IrBuilder

Builder for constructing IR modules.

IrBuilderScope

Scoped builder for structured control flow.

IrModule

A complete IR module representing a GPU kernel.

IrPrinter

IR pretty printer.

KernelConfig

Kernel configuration.

Parameter

A function parameter.

ValidationResult

Result of validation.

Validator

IR validator.

Value

An IR value with type information.

ValueId

Unique identifier for IR values.

VectorType

Vector types.

Enums

AtomicOp

Atomic operations.

BinaryOp

Binary operations.

CapabilityFlag

Capability flags for GPU features.

CastKind

Cast kinds.

CompareOp

Comparison operations.

ConstantValue

Constant values.

Dimension

Dimension for GPU indexing.

IrError

IR errors.

IrNode

IR node representing an operation.

IrType

IR type.

KernelMode

Kernel execution mode.

MathOp

Math operations (intrinsics).

MemoryScope

Memory scope for fences.

ScalarType

Scalar types supported in IR.

Terminator

Block terminator instructions.

UnaryOp

Unary operations.

ValidationLevel

Validation strictness level.

WarpReduceOp

Warp reduce operations.

WarpShuffleOp

Warp shuffle operations.

WarpVoteOp

Warp vote operations.

Type Aliases

IrResult

IR result type.