From 6284c1d2825ea171d35b5fde8801ed5a640be2ac Mon Sep 17 00:00:00 2001
From: Frederik Hennig <frederik.hennig@fau.de>
Date: Thu, 6 Mar 2025 14:52:10 +0100
Subject: [PATCH] Some cleanup to the pystencils runtime headers
- Remove ancient `PyStencilsField.h`
- move half-precision and HIP-specific integer typedefs to `pystencils_runtime/*.h` directory
- remove `INFINITY` macros since they cause redefinition errors with nvcc and hipcc
---
src/pystencils/backend/platforms/cuda.py | 2 +-
src/pystencils/include/PyStencilsField.h | 19 -------------------
.../half.h} | 0
.../hip.h} | 4 ----
src/pystencils/jit/gpu_cupy.py | 4 ++--
src/pystencils/types/types.py | 4 ++--
6 files changed, 5 insertions(+), 28 deletions(-)
delete mode 100644 src/pystencils/include/PyStencilsField.h
rename src/pystencils/include/{half_precision.h => pystencils_runtime/half.h} (100%)
rename src/pystencils/include/{gpu_defines.h => pystencils_runtime/hip.h} (56%)
diff --git a/src/pystencils/backend/platforms/cuda.py b/src/pystencils/backend/platforms/cuda.py
index e896fc2bb..d630594ba 100644
--- a/src/pystencils/backend/platforms/cuda.py
+++ b/src/pystencils/backend/platforms/cuda.py
@@ -188,7 +188,7 @@ class CudaPlatform(GenericGpu):
@property
def required_headers(self) -> set[str]:
- return {'"gpu_defines.h"'}
+ return {'"pystencils_runtime/hip.h"'} # TODO: move to HipPlatform once it is introduced
def materialize_iteration_space(
self, body: PsBlock, ispace: IterationSpace
diff --git a/src/pystencils/include/PyStencilsField.h b/src/pystencils/include/PyStencilsField.h
deleted file mode 100644
index 3055cae23..000000000
--- a/src/pystencils/include/PyStencilsField.h
+++ /dev/null
@@ -1,19 +0,0 @@
-#pragma once
-
-extern "C++" {
-#ifdef __CUDA_ARCH__
-template <typename DTYPE_T, std::size_t DIMENSION> struct PyStencilsField {
- DTYPE_T *data;
- DTYPE_T shape[DIMENSION];
- DTYPE_T stride[DIMENSION];
-};
-#else
-#include <array>
-
-template <typename DTYPE_T, std::size_t DIMENSION> struct PyStencilsField {
- DTYPE_T *data;
- std::array<DTYPE_T, DIMENSION> shape;
- std::array<DTYPE_T, DIMENSION> stride;
-};
-#endif
-}
diff --git a/src/pystencils/include/half_precision.h b/src/pystencils/include/pystencils_runtime/half.h
similarity index 100%
rename from src/pystencils/include/half_precision.h
rename to src/pystencils/include/pystencils_runtime/half.h
diff --git a/src/pystencils/include/gpu_defines.h b/src/pystencils/include/pystencils_runtime/hip.h
similarity index 56%
rename from src/pystencils/include/gpu_defines.h
rename to src/pystencils/include/pystencils_runtime/hip.h
index 67e7722e9..4bf4917f8 100644
--- a/src/pystencils/include/gpu_defines.h
+++ b/src/pystencils/include/pystencils_runtime/hip.h
@@ -1,9 +1,5 @@
#pragma once
-#define POS_INFINITY __int_as_float(0x7f800000)
-#define INFINITY POS_INFINITY
-#define NEG_INFINITY __int_as_float(0xff800000)
-
#ifdef __HIPCC_RTC__
typedef __hip_uint8_t uint8_t;
typedef __hip_int8_t int8_t;
diff --git a/src/pystencils/jit/gpu_cupy.py b/src/pystencils/jit/gpu_cupy.py
index 1c771a427..d45abf878 100644
--- a/src/pystencils/jit/gpu_cupy.py
+++ b/src/pystencils/jit/gpu_cupy.py
@@ -250,8 +250,8 @@ class CupyJit(JitBase):
headers = self._runtime_headers
headers |= kfunc.required_headers
- if '"half_precision.h"' in headers:
- headers.remove('"half_precision.h"')
+ if '"pystencils_runtime/half.h"' in headers:
+ headers.remove('"pystencils_runtime/half.h"')
if cp.cuda.runtime.is_hip:
headers.add("<hip/hip_fp16.h>")
else:
diff --git a/src/pystencils/types/types.py b/src/pystencils/types/types.py
index 825ac1d5d..8dea97ca4 100644
--- a/src/pystencils/types/types.py
+++ b/src/pystencils/types/types.py
@@ -661,7 +661,7 @@ class PsIeeeFloatType(PsScalarType):
@property
def required_headers(self) -> set[str]:
if self._width == 16:
- return {'"half_precision.h"'}
+ return {'"pystencils_runtime/half.h"'}
else:
return set()
@@ -672,7 +672,7 @@ class PsIeeeFloatType(PsScalarType):
match self.width:
case 16:
- return f"((half) {value})" # see include/half_precision.h
+ return f"((half) {value})" # see include/pystencils_runtime/half.h
case 32:
return f"{value}f"
case 64:
--
GitLab