[Refactor] update get_c_pointers to c_abi_spec

[sjfeng1999]

Motivation

Technical Details

Test Plan

Test Result

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[Copilot]

Pull request overview

This PR refactors FlyDSL’s host-side argument packing from the legacy __get_c_pointers__ / _reusable_slot_spec approach to a unified __c_abi_spec__ protocol, and updates tests/kernels to use the new torch-native tensor wrapper (from_torch_tensor). It also introduces a C++ MemRefSpec binding to centralize memref-type + cache-signature derivation and reduce duplicated Python-side state.

Changes:

• Replace get_c_pointers with c_abi_spec and migrate scalar/stream/struct/tensor argument implementations to __c_abi_spec__.
• Introduce MemRefSpec (C++/nanobind) and refactor tensor JIT args into MemRefJitArg + concrete TorchTensorJitArg / DLTensorJitArg.
• Update unit/system/kernel tests and some kernels to prefer flyc.from_torch_tensor(...) over flyc.from_dlpack(...).

Reviewed changes

Copilot reviewed 28 out of 28 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
tests/unit/test_universal_atomic.py	Switch tensor adaptation to from_torch_tensor for atomic tests.
tests/unit/test_tensor_cache_signature.py	Update cache-signature tests to TorchTensorJitArg / __c_abi_spec__ terminology and behavior.
tests/unit/test_struct.py	Migrate struct ABI assertions from pointer lists to slot fills via c_abi_spec.
tests/unit/test_math_ops.py	Switch tensor adaptation to from_torch_tensor for math-op launches.
tests/unit/test_for_auto_iter_args.py	Switch output tensor adaptation to from_torch_tensor.
tests/unit/test_callstate_dispatch.py	Update dispatch/packing tests to the new __c_abi_spec__ slot-fill model.
tests/system/test_while_e2e.py	Switch output tensor adaptation to from_torch_tensor.
tests/system/test_inline_compare_scf_if.py	Switch tensor adaptation to from_torch_tensor for control-flow system tests.
tests/system/test_ifexp_e2e.py	Switch output tensor adaptation to from_torch_tensor.
tests/system/test_if_liveout_minimal.py	Switch output tensor adaptation to from_torch_tensor.
tests/system/test_for_auto_iter_args_e2e.py	Switch output tensor adaptation to from_torch_tensor.
tests/system/test_control_flow_compile.py	Switch tensor adaptation to from_torch_tensor for compile tests.
tests/system/test_closure_freevars_mismatch.py	Switch output tensor adaptation to from_torch_tensor.
tests/kernels/test_vec_add.py	Switch tensor adaptation to from_torch_tensor in benchmark test.
tests/kernels/test_fp8_gemm_rowscale.py	Switch static-weight tensor wrappers to from_torch_tensor.
python/flydsl/expr/typing.py	Replace stream C-pointer extraction with __c_abi_spec__ fill protocol.
python/flydsl/expr/struct.py	Replace struct __get_c_pointers__ with recursive __c_abi_spec__ slot fills; add struct-type helpers to exports.
python/flydsl/expr/numeric.py	Replace scalar pointer extraction with __c_abi_spec__ fill protocol for numeric types.
python/flydsl/compiler/protocol.py	Replace get_c_pointers with c_abi_spec in compiler protocol layer.
python/flydsl/compiler/jit_function.py	Rework CallState construction to use c_abi_spec fills; remove old fallback dispatch generation.
python/flydsl/compiler/jit_executor.py	Centralize CallState implementation and switch CompiledArtifact dispatch to slot-fill based packing.
python/flydsl/compiler/jit_argument.py	Major refactor: introduce MemRefJitArg, TorchTensorJitArg, DLTensorJitArg, and new from_torch_tensor.
python/flydsl/compiler/init.py	Export from_torch_tensor from the compiler package.
lib/Bindings/Python/MemRefSpec.h	New C++ class binding for memref type + cache signature + dynamic-dim indices.
lib/Bindings/Python/FlyExtension.cpp	Bind MemRefSpec into _mlirDialectsFly; slim down DLTensorAdaptor binding API.
lib/Bindings/Python/DLTensorAdaptor.h	Simplify adaptor to metadata-only + dtype reporting; remove dynamic-mark and memref-desc building.
kernels/qk_norm_rope_quant.py	Switch static tensor wrappers to from_torch_tensor.
kernels/fused_rope_cache_kernel.py	Switch fallback wrapping to from_torch_tensor.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.