srdatalog.ir.codegen

Codegen backends — IIR Render registries.

Each subpackage is one codegen that consumes IIR ops and produces target-specific text. Codegens are not dialects: they have no ops, no lowerings, no rewrites. They are renderers + target configuration.

See docs/stage3a_execution_plan.md §3 for the file layout contract.

Currently registered:

• cuda: NVIDIA CUDA C++ emission.

Planned (per docs/stage3a_execution_plan.md §10 deferrals):

• cpp_tbb: CPU C++ with TBB parallel-for.

• cpp_omp: CPU C++ with OpenMP.

• hip: AMD HIP.

• metal: Apple Metal shading language.

• sycl: SYCL kernels.

Subpackages

• srdatalog.ir.codegen.cuda
  • srdatalog.ir.codegen.cuda.build
    • srdatalog.ir.codegen.cuda.build.cache
    • srdatalog.ir.codegen.cuda.build.compiler
    • srdatalog.ir.codegen.cuda.build.compiler_ninja
    • srdatalog.ir.codegen.cuda.build.loader
  • srdatalog.ir.codegen.cuda.render
    • srdatalog.ir.codegen.cuda.render.d2l
    • srdatalog.ir.codegen.cuda.render.iir_cf
    • srdatalog.ir.codegen.cuda.render.parallel_data
    • srdatalog.ir.codegen.cuda.render.sorted_array
  • srdatalog.ir.codegen.cuda.api
  • srdatalog.ir.codegen.cuda.batchfile
  • srdatalog.ir.codegen.cuda.complete_runner
  • srdatalog.ir.codegen.cuda.context
  • srdatalog.ir.codegen.cuda.emit
  • srdatalog.ir.codegen.cuda.envelope
  • srdatalog.ir.codegen.cuda.helpers
  • srdatalog.ir.codegen.cuda.main_file
  • srdatalog.ir.codegen.cuda.materialized
  • srdatalog.ir.codegen.cuda.orchestrator
  • srdatalog.ir.codegen.cuda.pipeline_utils
  • srdatalog.ir.codegen.cuda.plugin
  • srdatalog.ir.codegen.cuda.runner
  • srdatalog.ir.codegen.cuda.schema
  • srdatalog.ir.codegen.cuda.view_slots