start introduction of Target.HIP

src/pystencils/backend/emission/base_printer.py

@@ -383,7 +383,7 @@ class BasePrinter(ABC):
         from ...codegen import GpuKernel
         
         sig_parts = [self._func_prefix] if self._func_prefix is not None else []
-        if isinstance(func, GpuKernel) and func.target == Target.CUDA:
+        if isinstance(func, GpuKernel) and func.target.is_gpu():
             sig_parts.append("__global__")
         sig_parts += ["void", func.name, f"({params_str})"]
         signature = " ".join(sig_parts)

src/pystencils/codegen/config.py

@@ -593,12 +593,14 @@ class CreateKernelConfig(ConfigBase):
         """Returns either the user-specified JIT compiler, or infers one from the target if none is given."""
         jit: JitBase | None = self.get_option("jit")
 
+        target = self.get_target()
+
         if jit is None:
-            if self.get_target().is_cpu():
+            if target.is_cpu():
                 from ..jit import LegacyCpuJit
 
                 return LegacyCpuJit()
-            elif self.get_target() == Target.CUDA:
+            elif target == Target.CUDA:
                 try:
                     from ..jit.gpu_cupy import CupyJit
                     
@@ -609,7 +611,7 @@ class CreateKernelConfig(ConfigBase):
 
                     return no_jit
 
-            elif self.get_target() == Target.SYCL:
+            elif target == Target.SYCL or target == Target.HIP:
                 from ..jit import no_jit
 
                 return no_jit

src/pystencils/codegen/driver.py

@@ -398,7 +398,7 @@ class DefaultKernelCreationDriver:
         return kernel_ast
 
     def _get_gpu_indexing(self) -> GpuIndexing | None:
-        if self._target != Target.CUDA:
+        if not self._target.is_gpu():
             return None
 
         from .gpu_indexing import dim3
@@ -441,7 +441,7 @@ class DefaultKernelCreationDriver:
             omit_range_check: bool = gpu_opts.get_option("omit_range_check")
 
             match self._target:
-                case Target.CUDA:
+                case Target.CUDA | Target.HIP:
                     from ..backend.platforms import CudaPlatform
 
                     thread_mapping = (

src/pystencils/codegen/target.py

@@ -30,6 +30,7 @@ class Target(Flag):
     _GPU = auto()
 
     _CUDA = auto()
+    _HIP = auto()
 
     _SYCL = auto()
 
@@ -86,6 +87,12 @@ class Target(Flag):
     Generate a CUDA kernel for a generic Nvidia GPU.
     """
 
+    HIP = _GPU | _HIP
+    """Generic HIP GPU target.
+    
+    Generate a HIP kernel for generic AMD or NVidia GPUs.
+    """
+
     GPU = CUDA
     """Alias for `Target.CUDA`, for backward compatibility."""