Module wgsl_forge 

Deterministic WGSL generation, validation, certification, and tuning.

WGSL Forge treats shader semantics and hardware scheduling as separate typed inputs. Tuning is therefore allowed to change work distribution without changing the mathematical operation being certified.

Re-exports

pub use backend::resolve_execution_backend;

pub use cache::ManifestCache;

pub use dispatch::caps;

pub use dispatch::coopmat_usable;

pub use dispatch::fft_f32;

pub use dispatch::gemm_cpu_f64;

pub use dispatch::gemm_f32;

pub use dispatch::gemm_f32_tc;

pub use dispatch::gemm_f32_tc_coopmat;

pub use dispatch::gemm_f32_tc_reduced;

pub use dispatch::gemm_f64;

pub use dispatch::gemm_f64_df64;

pub use dispatch::gemv_cpu_f64;

pub use dispatch::gemv_f32;

pub use dispatch::gemv_f64;

pub use dispatch::pairwise_sq_dist_cpu_f64;

pub use dispatch::pairwise_sq_dist_f64;

pub use dispatch::ComputeCaps;

pub use dispatch::GEMM_GPU_THRESHOLD;

pub use emit::decode_spirv_words;

pub use emit::emit_shader;

pub use emit::matmul_tc_wgsl;

pub use emit::GeneratedShader;

pub use emit::TargetBackend;

pub use ir::BufferAccess;

pub use ir::BufferElement;

pub use ir::BufferSpec;

pub use ir::BuiltinKernel;

pub use ir::KernelSpec;

pub use ir::Op;

pub use ir::P64GpuWords64;

pub use ir::ScalarType;

pub use ir::SharedLen;

pub use ir::SharedMemorySpec;

pub use manifest::AdapterIdentity;

pub use manifest::CertificationManifest;

pub use manifest::HardwareProfile;

pub use manifest::TimingSource;

pub use manifest::TimingSummary;

pub use manifest::TuningManifest;

pub use manifest::ValidationLevel;

pub use oracle::candidate_evaluation;

pub use oracle::certify_builtin;

pub use oracle::compare_f32;

pub use oracle::dft_cpu;

pub use oracle::evaluate_builtin;

pub use oracle::evaluate_ffn;

pub use oracle::evaluate_fft;

pub use oracle::evaluate_matmul_tc;

pub use oracle::evaluate_p64;

pub use oracle::evaluate_topk;

pub use oracle::ffn_cpu;

pub use oracle::ffn_tensors;

pub use oracle::fft_inputs;

pub use oracle::matmul_cpu;

pub use oracle::p64_project_cpu;

pub use oracle::p64_records;

pub use oracle::topk_cpu;

pub use oracle::topk_inputs;

pub use oracle::AffineParams;

pub use oracle::ComparisonReport;

pub use oracle::FfnParams;

pub use oracle::FftParams;

pub use oracle::GpuEvaluation;

pub use oracle::OracleCase;

pub use oracle::OracleTolerance;

pub use oracle::TopKParams;

pub use oracle::evaluate_affine_cuda;

pub use oracle::evaluate_ffn_cuda;

pub use oracle::evaluate_topk_cuda;

pub use roofline::roofline_for;

pub use roofline::RooflineBound;

pub use roofline::RooflineEstimate;

pub use runtime::ForgeRuntime;

pub use schedule::AdapterConstraints;

pub use schedule::Schedule;

pub use schedule::ScheduleSpace;

pub use tune::tune_with;

pub use tune::CandidateEvaluation;

pub use tune::CandidateFailure;

pub use tune::CandidateResult;

pub use tune::TuningConfig;

pub use tune::TuningResult;

pub use validate::validate_native;

pub use validate::validate_wgsl;

pub use validate::ValidationReport;

Modules

audio

Forge-native audio kernels (GPU compute with a CPU oracle + parity certify).

backend

Automatic native→fallback backend selection (plan §2).

cache

calibration

W10 — Forge calibration pipeline (the training-related upgrade).

dispatch

Capability-aware “best path on this machine” compute dispatcher, keystoned on GEMM.

emit

execute

graph_ops

Native compute-graph op-node kernels — WGSL templates emitted directly from a graph node (not delegated to a legacy BuiltinKernel emitter, since these op-classes have no legacy standalone kernel). Each carries an exact CPU oracle and a GPU differential certify, exactly like the rest of the forge.

ir

manifest

oracle

GPU-forge differential oracle: CPU references, the numeric comparator, the GpuEvaluation evidence bundle, and the per-kernel evaluators that certify emitted shaders against those references.

physics

Certified physics forge kernels — complete, deterministic compute shaders with exact CPU oracles, naga validation, and GPU certification.

roofline

Simple roofline estimates per kernel (plan §6 / §10).

runtime

Consumer-facing runtime for the certified WGSL Forge.

schedule

tune

validate

Enums

ForgeError

Constants

CUDARC_API_VERSION

cudarc crate API version the cross-backend (CUDA) oracle is built against. Folded into the tuning/certification cache key (plan §8) so reuse is invalidated when the CUDA toolchain surface changes.

FORGE_SCHEMA_VERSION

NAGA_API_VERSION

WGPU_API_VERSION

Functions

generate_builtin