refactor: prefix every remaining public macro with FF_

CLAUDE.md

@@ -48,7 +48,7 @@ impl/kernels ─ impl/cpu ─ api/cpu + src/lib-cpu ┐
  Namespace `ff::<FF_DEVICE>::<module>` via the `FF_NAMESPACE_BEGIN` macros —
  do **not** hard-code `ff::cpu`.
 - **`include/fastfields/impl/cpu/`**, **`impl/cuda/`** — the loops over
- elements (thread pool / OpenMP; `__global__` kernels + `CUHOST` launchers).
+ elements (thread pool / OpenMP; `__global__` kernels + `FF_CUHOST` launchers).
  Header-only, templated, dynamic sizes. `ff::cpu::…` / `ff::cuda::…`.
 - **`include/fastfields/api/cpu/`**, **`api/cuda/`** + **`src/lib-cpu/`**,
  **`src/lib-cuda/`** — the dtype-dispatch boundary. Public symbols take

MIGRATION.md

@@ -135,7 +135,7 @@ CPU path against a brute-force / reference implementation, as `test_distance.cpp
  spurious extra arg (no matching overload); triggered by `restrict::loop`.
 10. **cpu-impl/{resize,restrict,splinc}.h** — wrong include prefix `"lib/…"` →
  `"kernels/…"`; impl namespace was plain `ff::<module>` but the kernels live in
- `ff::cpu::` (`FF_DEVICE`) so it must be `FF_NAMESPACE_BEGIN(FF)/(FF_DEVICE)/(<module>)`
+ `ff::cpu::` (`FF_DEVICE`) so it must be `FF_NAMESPACE_BEGIN(FF_NS)/(FF_DEVICE)/(<module>)`
  like distance; `index2offset_nd<ndim>()` runtime-ndim → dynamic overload;
  `jf::has_atomic_add` → `has_atomic_add`.
 

include/fastfields/api/checks.h

@@ -4,7 +4,7 @@
 #include "fastfields/core/dlpack.h"
 #include "fastfields/core/defines.h"
 
-FF_NAMESPACE_BEGIN(FF)
+FF_NAMESPACE_BEGIN(FF_NS)
 
 /**
  * Assert that a set of DLTensors all live on the same device.
@@ -55,6 +55,6 @@ inline void require_same_device(const DLTensor & ref, const DLTensor & t, const
  require_same_device(ref, rest...);
 }
 
-FF_NAMESPACE_END(FF)
+FF_NAMESPACE_END(FF_NS)
 
 #endif // FF_LIB_CHECKS

include/fastfields/api/cpu/pushpull_dispatch.h

@@ -23,7 +23,7 @@
 #include "fastfields/impl/kernels/utils.h"
 #include "fastfields/impl/cpu/pushpull.h"
 
-FF_NAMESPACE_BEGIN(FF)
+FF_NAMESPACE_BEGIN(FF_NS)
 FF_NAMESPACE_BEGIN(FF_DEVICE)
 
 // reduce/accumulation type used by the sampling kernels. Match jitfields
@@ -34,7 +34,7 @@ typedef double reduce_t;
  * DISPATCH *
  ***********************************************************************/
 
-#define PP_DTYPE(D, I, B, FN, args...) \
+#define FF_PP_DTYPE(D, I, B, FN, args...) \
  switch (code) { \
  case kDLFloat: switch (bits) { \
  case 32: return (use_32bits ? FN<D,I,B,float, int32_t>(args) \
@@ -45,25 +45,25 @@ typedef double reduce_t;
  }; default: break; \
  }
 
-// The template argument fed to PP_DTYPE is FF_BOUND_<NAME> / FF_SPLINE_<NAME>
+// The template argument fed to FF_PP_DTYPE is FF_BOUND_<NAME> / FF_SPLINE_<NAME>
 // (kernels/bounds.h, kernels/spline.h) -- the condition/order itself when it
 // is statically compiled, `Dynamic` otherwise per BOUNDFLAGS/SPLINEFLAGS. The
 // switch labels stay exhaustive on the *runtime* value either way; only the
 // instantiated template argument collapses onto the shared Dynamic path.
-#define PP_BOUND(D, I, FN, args...) \
+#define FF_PP_BOUND(D, I, FN, args...) \
  switch (bnd) { \
- case bound_t::Zero: PP_DTYPE(D,I,FF_BOUND_ZERO, FN,args); break; \
- case bound_t::Replicate: PP_DTYPE(D,I,FF_BOUND_REPLICATE,FN,args); break; \
- case bound_t::DCT1: PP_DTYPE(D,I,FF_BOUND_DCT1, FN,args); break; \
- case bound_t::DCT2: PP_DTYPE(D,I,FF_BOUND_DCT2, FN,args); break; \
- case bound_t::DST1: PP_DTYPE(D,I,FF_BOUND_DST1, FN,args); break; \
- case bound_t::DST2: PP_DTYPE(D,I,FF_BOUND_DST2, FN,args); break; \
- case bound_t::DFT: PP_DTYPE(D,I,FF_BOUND_DFT, FN,args); break; \
- case bound_t::NoCheck: PP_DTYPE(D,I,FF_BOUND_NOCHECK, FN,args); break; \
+ case bound_t::Zero: FF_PP_DTYPE(D,I,FF_BOUND_ZERO, FN,args); break; \
+ case bound_t::Replicate: FF_PP_DTYPE(D,I,FF_BOUND_REPLICATE,FN,args); break; \
+ case bound_t::DCT1: FF_PP_DTYPE(D,I,FF_BOUND_DCT1, FN,args); break; \
+ case bound_t::DCT2: FF_PP_DTYPE(D,I,FF_BOUND_DCT2, FN,args); break; \
+ case bound_t::DST1: FF_PP_DTYPE(D,I,FF_BOUND_DST1, FN,args); break; \
+ case bound_t::DST2: FF_PP_DTYPE(D,I,FF_BOUND_DST2, FN,args); break; \
+ case bound_t::DFT: FF_PP_DTYPE(D,I,FF_BOUND_DFT, FN,args); break; \
+ case bound_t::NoCheck: FF_PP_DTYPE(D,I,FF_BOUND_NOCHECK, FN,args); break; \
  default: throw std::invalid_argument("Unsupported boundary condition");\
  }
 
-// There used to be a second, hand-duplicated PP_ORDER (behind `-DFF_TEST_SPARSE`)
+// There used to be a second, hand-duplicated FF_PP_ORDER (behind `-DFF_TEST_SPARSE`)
 // that hard-coded a *covering* subset of the order x bound matrix -- literally
 // rejecting (throwing) most bound/order combinations at runtime -- purely to
 // keep the test build's compile time down. That is now redundant with, and
@@ -72,36 +72,36 @@ typedef double reduce_t;
 // instantiation (the actual compile-cost win FF_TEST_SPARSE was chasing),
 // while every combination stays fully *functional* (just via the Dynamic
 // runtime path instead of a dedicated static one) rather than throwing.
-// There is therefore only one PP_ORDER/PP_BOUND now; which combinations are
+// There is therefore only one FF_PP_ORDER/FF_PP_BOUND now; which combinations are
 // statically instantiated and which share Dynamic is entirely a BOUNDFLAGS/
 // SPLINEFLAGS *build-time* choice (Makefile: a sparser default for the `test`
 // target, the full static matrix for the library), not a code-level branch.
-#define PP_ORDER(D, FN, args...) \
+#define FF_PP_ORDER(D, FN, args...) \
  switch (spl) { \
- case spline_t::Nearest: PP_BOUND(D,FF_SPLINE_NEAREST, FN,args); break; \
- case spline_t::Linear: PP_BOUND(D,FF_SPLINE_LINEAR, FN,args); break; \
- case spline_t::Quadratic: PP_BOUND(D,FF_SPLINE_QUADRATIC, FN,args); break; \
- case spline_t::Cubic: PP_BOUND(D,FF_SPLINE_CUBIC, FN,args); break; \
- case spline_t::FourthOrder: PP_BOUND(D,FF_SPLINE_FOURTHORDER, FN,args); break; \
- case spline_t::FifthOrder: PP_BOUND(D,FF_SPLINE_FIFTHORDER, FN,args); break; \
- case spline_t::SixthOrder: PP_BOUND(D,FF_SPLINE_SIXTHORDER, FN,args); break; \
- case spline_t::SeventhOrder: PP_BOUND(D,FF_SPLINE_SEVENTHORDER,FN,args); break; \
+ case spline_t::Nearest: FF_PP_BOUND(D,FF_SPLINE_NEAREST, FN,args); break; \
+ case spline_t::Linear: FF_PP_BOUND(D,FF_SPLINE_LINEAR, FN,args); break; \
+ case spline_t::Quadratic: FF_PP_BOUND(D,FF_SPLINE_QUADRATIC, FN,args); break; \
+ case spline_t::Cubic: FF_PP_BOUND(D,FF_SPLINE_CUBIC, FN,args); break; \
+ case spline_t::FourthOrder: FF_PP_BOUND(D,FF_SPLINE_FOURTHORDER, FN,args); break; \
+ case spline_t::FifthOrder: FF_PP_BOUND(D,FF_SPLINE_FIFTHORDER, FN,args); break; \
+ case spline_t::SixthOrder: FF_PP_BOUND(D,FF_SPLINE_SIXTHORDER, FN,args); break; \
+ case spline_t::SeventhOrder: FF_PP_BOUND(D,FF_SPLINE_SEVENTHORDER,FN,args); break; \
  default: throw std::invalid_argument("Unsupported spline order"); \
  }
 
-#define DISPATCH_PP(FN, args...) \
+#define FF_DISPATCH_PP(FN, args...) \
 { \
  switch (ndim) { \
- case 1: PP_ORDER(1, FN, args); break; \
- case 2: PP_ORDER(2, FN, args); break; \
- case 3: PP_ORDER(3, FN, args); break; \
+ case 1: FF_PP_ORDER(1, FN, args); break; \
+ case 2: FF_PP_ORDER(2, FN, args); break; \
+ case 3: FF_PP_ORDER(3, FN, args); break; \
  default: throw std::invalid_argument("Only 1D, 2D and 3D are supported"); \
  }; \
  throw std::invalid_argument("Unsupported data type"); \
 }
 
 
 FF_NAMESPACE_END(FF_DEVICE)
-FF_NAMESPACE_END(FF)
+FF_NAMESPACE_END(FF_NS)
 
 #endif // FF_CPU_PUSHPULL_DISPATCH

include/fastfields/api/cuda/pushpull_dispatch.h

@@ -23,7 +23,7 @@
 #include "fastfields/impl/kernels/utils.h"
 #include "fastfields/impl/cuda/pushpull.h"
 
-FF_NAMESPACE_BEGIN(FF)
+FF_NAMESPACE_BEGIN(FF_NS)
 FF_NAMESPACE_BEGIN(FF_DEVICE)
 
 // reduce/accumulation type used by the sampling kernels. Match jitfields
@@ -34,7 +34,7 @@ typedef double reduce_t;
  * DISPATCH *
  ***********************************************************************/
 
-#define PP_DTYPE(D, I, B, FN, args...) \
+#define FF_PP_DTYPE(D, I, B, FN, args...) \
  switch (code) { \
  case kDLFloat: switch (bits) { \
  case 32: return (use_32bits ? FN<D,I,B,float, int32_t>(args) \
@@ -45,52 +45,52 @@ typedef double reduce_t;
  }; default: break; \
  }
 
-// The template argument fed to PP_DTYPE is FF_BOUND_<NAME> / FF_SPLINE_<NAME>
+// The template argument fed to FF_PP_DTYPE is FF_BOUND_<NAME> / FF_SPLINE_<NAME>
 // (kernels/bounds.h, kernels/spline.h) -- the condition/order itself when it
 // is statically compiled, `Dynamic` otherwise per BOUNDFLAGS/SPLINEFLAGS. The
 // switch labels stay exhaustive on the *runtime* value either way; only the
 // instantiated template argument collapses onto the shared Dynamic path.
 // This is what actually keeps ptxas's memory bounded -- see BOUNDFLAGS/
 // SPLINEFLAGS in the Makefile.
-#define PP_BOUND(D, I, FN, args...) \
+#define FF_PP_BOUND(D, I, FN, args...) \
  switch (bnd) { \
- case bound_t::Zero: PP_DTYPE(D,I,FF_BOUND_ZERO, FN,args); break; \
- case bound_t::Replicate: PP_DTYPE(D,I,FF_BOUND_REPLICATE,FN,args); break; \
- case bound_t::DCT1: PP_DTYPE(D,I,FF_BOUND_DCT1, FN,args); break; \
- case bound_t::DCT2: PP_DTYPE(D,I,FF_BOUND_DCT2, FN,args); break; \
- case bound_t::DST1: PP_DTYPE(D,I,FF_BOUND_DST1, FN,args); break; \
- case bound_t::DST2: PP_DTYPE(D,I,FF_BOUND_DST2, FN,args); break; \
- case bound_t::DFT: PP_DTYPE(D,I,FF_BOUND_DFT, FN,args); break; \
- case bound_t::NoCheck: PP_DTYPE(D,I,FF_BOUND_NOCHECK, FN,args); break; \
+ case bound_t::Zero: FF_PP_DTYPE(D,I,FF_BOUND_ZERO, FN,args); break; \
+ case bound_t::Replicate: FF_PP_DTYPE(D,I,FF_BOUND_REPLICATE,FN,args); break; \
+ case bound_t::DCT1: FF_PP_DTYPE(D,I,FF_BOUND_DCT1, FN,args); break; \
+ case bound_t::DCT2: FF_PP_DTYPE(D,I,FF_BOUND_DCT2, FN,args); break; \
+ case bound_t::DST1: FF_PP_DTYPE(D,I,FF_BOUND_DST1, FN,args); break; \
+ case bound_t::DST2: FF_PP_DTYPE(D,I,FF_BOUND_DST2, FN,args); break; \
+ case bound_t::DFT: FF_PP_DTYPE(D,I,FF_BOUND_DFT, FN,args); break; \
+ case bound_t::NoCheck: FF_PP_DTYPE(D,I,FF_BOUND_NOCHECK, FN,args); break; \
  default: throw std::invalid_argument("Unsupported boundary condition");\
  }
 
-#define PP_ORDER(D, FN, args...) \
+#define FF_PP_ORDER(D, FN, args...) \
  switch (spl) { \
- case spline_t::Nearest: PP_BOUND(D,FF_SPLINE_NEAREST, FN,args); break; \
- case spline_t::Linear: PP_BOUND(D,FF_SPLINE_LINEAR, FN,args); break; \
- case spline_t::Quadratic: PP_BOUND(D,FF_SPLINE_QUADRATIC, FN,args); break; \
- case spline_t::Cubic: PP_BOUND(D,FF_SPLINE_CUBIC, FN,args); break; \
- case spline_t::FourthOrder: PP_BOUND(D,FF_SPLINE_FOURTHORDER, FN,args); break; \
- case spline_t::FifthOrder: PP_BOUND(D,FF_SPLINE_FIFTHORDER, FN,args); break; \
- case spline_t::SixthOrder: PP_BOUND(D,FF_SPLINE_SIXTHORDER, FN,args); break; \
- case spline_t::SeventhOrder: PP_BOUND(D,FF_SPLINE_SEVENTHORDER,FN,args); break; \
+ case spline_t::Nearest: FF_PP_BOUND(D,FF_SPLINE_NEAREST, FN,args); break; \
+ case spline_t::Linear: FF_PP_BOUND(D,FF_SPLINE_LINEAR, FN,args); break; \
+ case spline_t::Quadratic: FF_PP_BOUND(D,FF_SPLINE_QUADRATIC, FN,args); break; \
+ case spline_t::Cubic: FF_PP_BOUND(D,FF_SPLINE_CUBIC, FN,args); break; \
+ case spline_t::FourthOrder: FF_PP_BOUND(D,FF_SPLINE_FOURTHORDER, FN,args); break; \
+ case spline_t::FifthOrder: FF_PP_BOUND(D,FF_SPLINE_FIFTHORDER, FN,args); break; \
+ case spline_t::SixthOrder: FF_PP_BOUND(D,FF_SPLINE_SIXTHORDER, FN,args); break; \
+ case spline_t::SeventhOrder: FF_PP_BOUND(D,FF_SPLINE_SEVENTHORDER,FN,args); break; \
  default: throw std::invalid_argument("Unsupported spline order"); \
  }
 
-#define DISPATCH_PP(FN, args...) \
+#define FF_DISPATCH_PP(FN, args...) \
 { \
  switch (ndim) { \
- case 1: PP_ORDER(1, FN, args); break; \
- case 2: PP_ORDER(2, FN, args); break; \
- case 3: PP_ORDER(3, FN, args); break; \
+ case 1: FF_PP_ORDER(1, FN, args); break; \
+ case 2: FF_PP_ORDER(2, FN, args); break; \
+ case 3: FF_PP_ORDER(3, FN, args); break; \
  default: throw std::invalid_argument("Only 1D, 2D and 3D are supported"); \
  }; \
  throw std::invalid_argument("Unsupported data type"); \
 }
 
 
 FF_NAMESPACE_END(FF_DEVICE)
-FF_NAMESPACE_END(FF)
+FF_NAMESPACE_END(FF_NS)
 
 #endif // FF_CUDA_PUSHPULL_DISPATCH

include/fastfields/api/cuda/stream.h

@@ -18,7 +18,7 @@
 #include <cstdint>
 #include "fastfields/core/cuda_switch.h"
 
-FF_NAMESPACE_BEGIN(FF)
+FF_NAMESPACE_BEGIN(FF_NS)
 FF_NAMESPACE_BEGIN(FF_DEVICE)
 
 // intptr_t -> cudaStream_t (0 == the default stream).
@@ -28,6 +28,6 @@ static inline cudaStream_t _reg_stream(intptr_t stream)
 }
 
 FF_NAMESPACE_END(FF_DEVICE)
-FF_NAMESPACE_END(FF)
+FF_NAMESPACE_END(FF_NS)
 
 #endif // FF_CUDA_STREAM

include/fastfields/api/distance.h

@@ -4,7 +4,7 @@
 #include <cstdint>
 #include "fastfields/core/defines.h"
 
-FF_NAMESPACE_BEGIN(FF)
+FF_NAMESPACE_BEGIN(FF_NS)
 
 #ifndef FF_LIB_BOUND_SPLINE_T
 #define FF_LIB_BOUND_SPLINE_T
@@ -175,6 +175,6 @@ void dt_mesh(
  intptr_t stream = 0
 );
 
-FF_NAMESPACE_END(FF)
+FF_NAMESPACE_END(FF_NS)
 
 #endif // FF_LIB_DISTANCE

include/fastfields/api/posdef.h

@@ -4,7 +4,7 @@
 #include <cstdint>
 #include "fastfields/core/defines.h"
 
-FF_NAMESPACE_BEGIN(FF)
+FF_NAMESPACE_BEGIN(FF_NS)
 
 /**
  * Compact symmetric ("Sym") positive-definite matrix operations.
@@ -79,6 +79,6 @@ void sym_invert_(
  intptr_t stream = 0
 );
 
-FF_NAMESPACE_END(FF)
+FF_NAMESPACE_END(FF_NS)
 
 #endif // FF_LIB_POSDEF

include/fastfields/api/pushpull.h

@@ -6,7 +6,7 @@
 
 #ifndef FF_LIB_BOUND_SPLINE_T
 #define FF_LIB_BOUND_SPLINE_T
-FF_NAMESPACE_BEGIN(FF)
+FF_NAMESPACE_BEGIN(FF_NS)
 
 FF_NAMESPACE_BEGIN(bound_t)
 using T = int8_t;
@@ -34,10 +34,10 @@ static constexpr T SixthOrder = 6;
 static constexpr T SeventhOrder = 7;
 FF_NAMESPACE_END(spline_t)
 
-FF_NAMESPACE_END(FF)
+FF_NAMESPACE_END(FF_NS)
 #endif // FF_LIB_BOUND_SPLINE_T
 
-FF_NAMESPACE_BEGIN(FF)
+FF_NAMESPACE_BEGIN(FF_NS)
 
 /**
  * @brief Sample ("pull") a spline-encoded volume at arbitrary coordinates.
@@ -227,6 +227,6 @@ void grad_backward(
  intptr_t stream = 0
 );
 
-FF_NAMESPACE_END(FF)
+FF_NAMESPACE_END(FF_NS)
 
 #endif // FF_LIB_PUSHPULL

include/fastfields/api/reg_field.h

@@ -6,7 +6,7 @@
 
 #ifndef FF_LIB_BOUND_SPLINE_T
 #define FF_LIB_BOUND_SPLINE_T
-FF_NAMESPACE_BEGIN(FF)
+FF_NAMESPACE_BEGIN(FF_NS)
 
 FF_NAMESPACE_BEGIN(bound_t)
 using T = int8_t;
@@ -34,10 +34,10 @@ static constexpr T SixthOrder = 6;
 static constexpr T SeventhOrder = 7;
 FF_NAMESPACE_END(spline_t)
 
-FF_NAMESPACE_END(FF)
+FF_NAMESPACE_END(FF_NS)
 #endif // FF_LIB_BOUND_SPLINE_T
 
-FF_NAMESPACE_BEGIN(FF)
+FF_NAMESPACE_BEGIN(FF_NS)
 
 /**
  * @brief Apply a spatial regulariser operator to a multi-channel field.
@@ -375,6 +375,6 @@ void field_relax_rls(
  intptr_t stream = 0
 );
 
-FF_NAMESPACE_END(FF)
+FF_NAMESPACE_END(FF_NS)
 
 #endif // FF_LIB_REG_FIELD

include/fastfields/api/reg_flow.h

@@ -6,7 +6,7 @@
 
 #ifndef FF_LIB_BOUND_SPLINE_T
 #define FF_LIB_BOUND_SPLINE_T
-FF_NAMESPACE_BEGIN(FF)
+FF_NAMESPACE_BEGIN(FF_NS)
 
 FF_NAMESPACE_BEGIN(bound_t)
 using T = int8_t;
@@ -34,10 +34,10 @@ static constexpr T SixthOrder = 6;
 static constexpr T SeventhOrder = 7;
 FF_NAMESPACE_END(spline_t)
 
-FF_NAMESPACE_END(FF)
+FF_NAMESPACE_END(FF_NS)
 #endif // FF_LIB_BOUND_SPLINE_T
 
-FF_NAMESPACE_BEGIN(FF)
+FF_NAMESPACE_BEGIN(FF_NS)
 
 /**
  * @brief Apply a spatial regulariser operator to a vector flow field.
@@ -396,6 +396,6 @@ void flow_relax_rls(
  intptr_t stream = 0
 );
 
-FF_NAMESPACE_END(FF)
+FF_NAMESPACE_END(FF_NS)
 
 #endif // FF_LIB_REG_FLOW