numba-cuda 0.19.0__py3-none-any.whl → 0.20.0__py3-none-any.whl

numba_cuda/numba/cuda/core/interpreter.py

@@ -1,3 +1,6 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: BSD-2-Clause
+
 import builtins
 import collections
 import dis
@@ -5,28 +8,29 @@ import operator
 import logging
 import textwrap
 
-from numba.core import errors, ir, config
+from numba.core import errors, ir
+from numba.cuda.core import config
+from numba.cuda.errors import UnsupportedBytecodeError
 from numba.core.errors import (
     NotDefinedError,
-    UnsupportedBytecodeError,
     error_extras,
 )
 from numba.cuda.core import ir_utils
-from numba.core.utils import (
+from numba.cuda.utils import (
     PYVERSION,
     BINOPS_TO_OPERATORS,
     INPLACE_BINOPS_TO_OPERATORS,
-    _lazy_pformat,
 )
+from numba.cuda.utils import _lazy_pformat
 from numba.core.byteflow import Flow, AdaptDFA, AdaptCFA, BlockKind
-from numba.core.unsafe import eh
+from numba.cuda.core.unsafe import eh
 from numba.cpython.unsafe.tuple import unpack_single_tuple
 
 
 if PYVERSION in ((3, 12), (3, 13)):
     # Operands for CALL_INTRINSIC_1
     from numba.core.byteflow import CALL_INTRINSIC_1_Operand as ci1op
-elif PYVERSION in ((3, 10), (3, 11)):
+elif PYVERSION in ((3, 9), (3, 10), (3, 11)):
     pass
 else:
     raise NotImplementedError(PYVERSION)
@@ -1392,7 +1396,7 @@ class Interpreter(object):
                     if entry.start < self.last_active_offset
                 ]
             )
-        elif PYVERSION in ((3, 10), (3, 11)):
+        elif PYVERSION in ((3, 9), (3, 10), (3, 11)):
             pass
         else:
             raise NotImplementedError(PYVERSION)
@@ -1407,7 +1411,10 @@ class Interpreter(object):
         if PYVERSION in ((3, 11), (3, 12), (3, 13)):
             # Insert end of try markers
             self._end_try_blocks()
-        elif PYVERSION in ((3, 10),):
+        elif PYVERSION in (
+            (3, 9),
+            (3, 10),
+        ):
             pass
         else:
             raise NotImplementedError(PYVERSION)
@@ -1431,7 +1438,7 @@ class Interpreter(object):
         peepholes = []
         if PYVERSION in ((3, 11), (3, 12), (3, 13)):
             peepholes.append(peep_hole_split_at_pop_block)
-        if PYVERSION in ((3, 10), (3, 11), (3, 12), (3, 13)):
+        if PYVERSION in ((3, 9), (3, 10), (3, 11), (3, 12), (3, 13)):
             peepholes.append(peep_hole_list_to_tuple)
         peepholes.append(peep_hole_delete_with_exit)
         if PYVERSION in ((3, 10), (3, 11), (3, 12), (3, 13)):
@@ -1589,7 +1596,10 @@ class Interpreter(object):
             if newtryblk is not None:
                 if newtryblk is not tryblk:
                     self._insert_try_block_begin()
-        elif PYVERSION in ((3, 10),):
+        elif PYVERSION in (
+            (3, 9),
+            (3, 10),
+        ):
             while self.syntax_blocks:
                 if offset >= self.syntax_blocks[-1].exit:
                     self.syntax_blocks.pop()
@@ -1826,7 +1836,10 @@ class Interpreter(object):
                     if inst.offset >= top.exit:
                         self.current_block.append(ir.PopBlock(loc=self.loc))
                         self.syntax_blocks.pop()
-        elif PYVERSION in ((3, 10),):
+        elif PYVERSION in (
+            (3, 9),
+            (3, 10),
+        ):
             pass
         else:
             raise NotImplementedError(PYVERSION)
@@ -2023,7 +2036,7 @@ class Interpreter(object):
             target = self.get(container)
             expr = ir.Expr.getitem(target, index=index, loc=self.loc)
             self.store(expr, res)
-    elif PYVERSION in ((3, 10), (3, 11)):
+    elif PYVERSION in ((3, 9), (3, 10), (3, 11)):
         pass
     else:
         raise NotImplementedError(PYVERSION)
@@ -2049,7 +2062,7 @@ class Interpreter(object):
                 target=target, index=index, value=value, loc=self.loc
             )
             self.current_block.append(stmt)
-    elif PYVERSION in ((3, 10), (3, 11)):
+    elif PYVERSION in ((3, 9), (3, 10), (3, 11)):
         pass
     else:
         raise NotImplementedError(PYVERSION)
@@ -2342,7 +2355,7 @@ class Interpreter(object):
             dstname = self.code_locals[oparg2]
             self.store(value=self.get(value2), name=dstname)
 
-    elif PYVERSION in ((3, 10), (3, 11), (3, 12)):
+    elif PYVERSION in ((3, 9), (3, 10), (3, 11), (3, 12)):
         pass
     else:
         raise NotImplementedError(PYVERSION)
@@ -2360,7 +2373,7 @@ class Interpreter(object):
                 undef = ir.Expr.undef(loc=self.loc)
                 self.store(undef, name=res)
 
-    elif PYVERSION in ((3, 10), (3, 11)):
+    elif PYVERSION in ((3, 9), (3, 10), (3, 11)):
         pass
     else:
         raise NotImplementedError(PYVERSION)
@@ -2400,7 +2413,7 @@ class Interpreter(object):
         item = self.get(item)
         if PYVERSION in ((3, 12), (3, 13)):
             attr = self.code_names[inst.arg >> 1]
-        elif PYVERSION in ((3, 10), (3, 11)):
+        elif PYVERSION in ((3, 9), (3, 10), (3, 11)):
             attr = self.code_names[inst.arg]
         else:
             raise NotImplementedError(PYVERSION)
@@ -2436,7 +2449,10 @@ class Interpreter(object):
             value = self.get_global_value(name)
             gl = ir.Global(name, value, loc=self.loc)
             self.store(gl, res)
-    elif PYVERSION in ((3, 10),):
+    elif PYVERSION in (
+        (3, 9),
+        (3, 10),
+    ):
 
         def op_LOAD_GLOBAL(self, inst, res):
             name = self.code_names[inst.arg]
@@ -2464,7 +2480,10 @@ class Interpreter(object):
                 value = self.get_closure_value(idx)
                 gl = ir.FreeVar(idx, name, value, loc=self.loc)
             self.store(gl, res)
-    elif PYVERSION in ((3, 10),):
+    elif PYVERSION in (
+        (3, 9),
+        (3, 10),
+    ):
 
         def op_LOAD_DEREF(self, inst, res):
             n_cellvars = len(self.code_cellvars)
@@ -2491,7 +2510,10 @@ class Interpreter(object):
             name = self.func_id.func.__code__._varname_from_oparg(inst.arg)
             value = self.get(value)
             self.store(value=value, name=name)
-    elif PYVERSION in ((3, 10),):
+    elif PYVERSION in (
+        (3, 9),
+        (3, 10),
+    ):
 
         def op_STORE_DEREF(self, inst, value):
             n_cellvars = len(self.code_cellvars)
@@ -2543,7 +2565,7 @@ class Interpreter(object):
                         if ex.target == end
                     ]
                 )
-        elif PYVERSION in ((3, 10), (3, 11)):
+        elif PYVERSION in ((3, 9), (3, 10), (3, 11)):
             pass
         else:
             raise NotImplementedError(PYVERSION)
@@ -3104,7 +3126,7 @@ class Interpreter(object):
             self.store(ir.Expr.cast(self.get(retval), loc=self.loc), castval)
             ret = ir.Return(self.get(castval), loc=self.loc)
             self.current_block.append(ret)
-    elif PYVERSION in ((3, 10), (3, 11)):
+    elif PYVERSION in ((3, 9), (3, 10), (3, 11)):
         pass
     else:
         raise NotImplementedError(PYVERSION)
@@ -3114,7 +3136,7 @@ class Interpreter(object):
         def op_TO_BOOL(self, inst, val, res):
             self.store(self.get(val), res)  # TODO: just a lazy hack
 
-    elif PYVERSION in ((3, 10), (3, 11), (3, 12)):
+    elif PYVERSION in ((3, 9), (3, 10), (3, 11), (3, 12)):
         pass
     else:
         raise NotImplementedError(PYVERSION)
@@ -3125,7 +3147,7 @@ class Interpreter(object):
             # TODO: fifth lowest bit now indicates a forced version to bool.
         elif PYVERSION in ((3, 12),):
             op = dis.cmp_op[inst.arg >> 4]
-        elif PYVERSION in ((3, 10), (3, 11)):
+        elif PYVERSION in ((3, 9), (3, 10), (3, 11)):
             op = dis.cmp_op[inst.arg]
         else:
             raise NotImplementedError(PYVERSION)
@@ -3256,7 +3278,7 @@ class Interpreter(object):
 
         def op_POP_JUMP_IF_NOT_NONE(self, inst, pred):
             self._jump_if_none(inst, pred, False)
-    elif PYVERSION in ((3, 10), (3, 11)):
+    elif PYVERSION in ((3, 9), (3, 10), (3, 11)):
         pass
     else:
         raise NotImplementedError(PYVERSION)
@@ -3426,7 +3448,10 @@ class Interpreter(object):
                 assert 0, "unreachable"
             self.store(gl, res)
 
-    elif PYVERSION in ((3, 10),):
+    elif PYVERSION in (
+        (3, 9),
+        (3, 10),
+    ):
 
         def op_LOAD_CLOSURE(self, inst, res):
             n_cellvars = len(self.code_cellvars)
@@ -3576,7 +3601,7 @@ class Interpreter(object):
                 return
             else:
                 raise NotImplementedError(operand)
-    elif PYVERSION in ((3, 10), (3, 11)):
+    elif PYVERSION in ((3, 9), (3, 10), (3, 11)):
         pass
     else:
         raise NotImplementedError(PYVERSION)
@@ -3586,7 +3611,7 @@ if PYVERSION in ((3, 12), (3, 13)):
 
     class INTRINSIC_STOPITERATION_ERROR(AssertionError):
         pass
-elif PYVERSION in ((3, 10), (3, 11)):
+elif PYVERSION in ((3, 9), (3, 10), (3, 11)):
     pass
 else:
     raise NotImplementedError(PYVERSION)

numba_cuda/numba/cuda/core/ir_utils.py

@@ -1,7 +1,6 @@
-#
-# Copyright (c) 2017 Intel Corporation
+# SPDX-FileCopyrightText: Copyright (c) 2017 Intel Corporation
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: BSD-2-Clause
-#
 
 import numpy
 import math
@@ -12,7 +11,9 @@ import warnings
 
 import numba
 from numba.core.extending import _Intrinsic
-from numba.core import types, typing, ir, analysis, postproc, rewrites, config
+from numba.core import types, ir, analysis
+from numba.cuda import typing
+from numba.cuda.core import postproc, rewrites, config
 from numba.core.typing.templates import signature
 from numba.core.analysis import (
     compute_live_map,
@@ -249,12 +250,7 @@ def mk_range_block(typemap, start, stop, step, calltypes, scope, loc):
     range_call_assign = ir.Assign(range_call, range_call_var, loc)
     # iter_var = getiter(range_call_var)
     iter_call = ir.Expr.getiter(range_call_var, loc)
-    if config.USE_LEGACY_TYPE_SYSTEM:
-        calltype_sig = signature(
-            types.range_iter64_type, types.range_state64_type
-        )
-    else:
-        calltype_sig = signature(types.range_iter_type, types.range_state_type)
+    calltype_sig = signature(types.range_iter64_type, types.range_state64_type)
     calltypes[iter_call] = calltype_sig
     iter_var = ir.Var(scope, mk_unique_var("$iter_var"), loc)
     typemap[iter_var.name] = types.iterators.RangeIteratorType(types.intp)
@@ -333,10 +329,7 @@ def mk_loop_header(typemap, phi_var, calltypes, scope, loc):
         types.intp, types.boolean
     )
     iternext_call = ir.Expr.iternext(phi_var, loc)
-    if config.USE_LEGACY_TYPE_SYSTEM:
-        range_iter_type = types.range_iter64_type
-    else:
-        range_iter_type = types.range_iter_type
+    range_iter_type = types.range_iter64_type
     calltypes[iternext_call] = signature(
         types.containers.Pair(types.intp, types.boolean), range_iter_type
     )
@@ -813,8 +806,6 @@ def has_no_side_effect(rhs, lives, call_table):
     """Returns True if this expression has no side effects that
     would prevent re-ordering.
     """
-    from numba.parfors import array_analysis, parfor
-    from numba.misc.special import prange
 
     if isinstance(rhs, ir.Expr) and rhs.op == "call":
         func_name = rhs.func.name
@@ -827,11 +818,7 @@ def has_no_side_effect(rhs, lives, call_table):
             or call_list == ["stencil", numba]
             or call_list == ["log", numpy]
             or call_list == ["dtype", numpy]
-            or call_list == [array_analysis.wrap_index]
-            or call_list == [prange]
-            or call_list == ["prange", numba]
             or call_list == ["pndindex", numba]
-            or call_list == [parfor.internal_prange]
             or call_list == ["ceil", math]
             or call_list == [max]
             or call_list == [int]
@@ -1894,7 +1881,7 @@ def compile_to_numba_ir(
     if typingctx and other typing inputs are available and update typemap and
     calltypes.
     """
-    from numba.core import typed_passes
+    from numba.cuda.core import typed_passes
 
     # mk_func can be actual function or make_function node, or a njit function
     if hasattr(mk_func, "code"):
@@ -1976,7 +1963,8 @@ def get_ir_of_code(glbls, fcode):
         fcode, func_env, func_arg, func_clo, glbls
     )
 
-    from numba.core import compiler
+    from numba.cuda import compiler
+    from numba.cuda.core.compiler import StateDict
 
     ir = compiler.run_frontend(f)
 
@@ -1985,7 +1973,7 @@ def get_ir_of_code(glbls, fcode):
     # for example, Raise nodes need to become StaticRaise before type inference
     class DummyPipeline(object):
         def __init__(self, f_ir):
-            self.state = compiler.StateDict()
+            self.state = StateDict()
             self.state.typingctx = None
             self.state.targetctx = None
             self.state.args = None
@@ -1998,10 +1986,10 @@ def get_ir_of_code(glbls, fcode):
     rewrites.rewrite_registry.apply("before-inference", state)
     # call inline pass to handle cases like stencils and comprehensions
     swapped = {}  # TODO: get this from diagnostics store
-    import numba.core.inline_closurecall
+    from numba.cuda.core.inline_closurecall import InlineClosureCallPass
 
-    inline_pass = numba.core.inline_closurecall.InlineClosureCallPass(
-        ir, numba.core.cpu.ParallelOptions(False), swapped
+    inline_pass = InlineClosureCallPass(
+        ir, numba.cuda.core.options.ParallelOptions(False), swapped
     )
     inline_pass.run()
 
@@ -2014,8 +2002,8 @@ def get_ir_of_code(glbls, fcode):
     # added to create valid IR.
 
     # rebuild IR in SSA form
-    from numba.core.untyped_passes import ReconstructSSA
-    from numba.core.typed_passes import PreLowerStripPhis
+    from numba.cuda.core.untyped_passes import ReconstructSSA
+    from numba.cuda.core.typed_passes import PreLowerStripPhis
 
     reconstruct_ssa = ReconstructSSA()
     phistrip = PreLowerStripPhis()
@@ -2495,7 +2483,7 @@ def legalize_single_scope(blocks):
     return len({blk.scope for blk in blocks.values()}) == 1
 
 
-def check_and_legalize_ir(func_ir, flags: "numba.core.compiler.Flags"):
+def check_and_legalize_ir(func_ir, flags: "numba.core.flags.Flags"):
     """
     This checks that the IR presented is legal
     """

numba_cuda/numba/cuda/core/options.py

@@ -0,0 +1,262 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: BSD-2-Clause
+
+"""
+Defines CUDA Options for use in the CUDA target
+"""
+
+from abc import ABCMeta, abstractmethod
+
+
+class AbstractOptionValue(metaclass=ABCMeta):
+    """Abstract base class for custom option values."""
+
+    @abstractmethod
+    def encode(self) -> str:
+        """Returns an encoding of the values"""
+        ...
+
+    def __repr__(self) -> str:
+        return f"{self.__class__.__name__}({self.encode()})"
+
+
+class FastMathOptions(AbstractOptionValue):
+    """
+    Options for controlling fast math optimization.
+    """
+
+    def __init__(self, value):
+        # https://releases.llvm.org/7.0.0/docs/LangRef.html#fast-math-flags
+        valid_flags = {
+            "fast",
+            "nnan",
+            "ninf",
+            "nsz",
+            "arcp",
+            "contract",
+            "afn",
+            "reassoc",
+        }
+
+        if isinstance(value, FastMathOptions):
+            self.flags = value.flags.copy()
+        elif value is True:
+            self.flags = {"fast"}
+        elif value is False:
+            self.flags = set()
+        elif isinstance(value, set):
+            invalid = value - valid_flags
+            if invalid:
+                raise ValueError("Unrecognized fastmath flags: %s" % invalid)
+            self.flags = value
+        elif isinstance(value, dict):
+            invalid = set(value.keys()) - valid_flags
+            if invalid:
+                raise ValueError("Unrecognized fastmath flags: %s" % invalid)
+            self.flags = {v for v, enable in value.items() if enable}
+        else:
+            msg = "Expected fastmath option(s) to be either a bool, dict or set"
+            raise ValueError(msg)
+
+    def __bool__(self):
+        return bool(self.flags)
+
+    __nonzero__ = __bool__
+
+    def encode(self) -> str:
+        return str(self.flags)
+
+    def __eq__(self, other):
+        if type(other) is type(self):
+            return self.flags == other.flags
+        return NotImplemented
+
+
+class ParallelOptions(AbstractOptionValue):
+    """
+    Options for controlling auto parallelization.
+    """
+
+    __slots__ = (
+        "enabled",
+        "comprehension",
+        "reduction",
+        "inplace_binop",
+        "setitem",
+        "numpy",
+        "stencil",
+        "fusion",
+        "prange",
+    )
+
+    def __init__(self, value):
+        if isinstance(value, bool):
+            self.enabled = value
+            self.comprehension = value
+            self.reduction = value
+            self.inplace_binop = value
+            self.setitem = value
+            self.numpy = value
+            self.stencil = value
+            self.fusion = value
+            self.prange = value
+        elif isinstance(value, dict):
+            self.enabled = True
+            self.comprehension = value.pop("comprehension", True)
+            self.reduction = value.pop("reduction", True)
+            self.inplace_binop = value.pop("inplace_binop", True)
+            self.setitem = value.pop("setitem", True)
+            self.numpy = value.pop("numpy", True)
+            self.stencil = value.pop("stencil", True)
+            self.fusion = value.pop("fusion", True)
+            self.prange = value.pop("prange", True)
+            if value:
+                msg = "Unrecognized parallel options: %s" % value.keys()
+                raise NameError(msg)
+        elif isinstance(value, ParallelOptions):
+            self.enabled = value.enabled
+            self.comprehension = value.comprehension
+            self.reduction = value.reduction
+            self.inplace_binop = value.inplace_binop
+            self.setitem = value.setitem
+            self.numpy = value.numpy
+            self.stencil = value.stencil
+            self.fusion = value.fusion
+            self.prange = value.prange
+        else:
+            msg = "Expect parallel option to be either a bool or a dict"
+            raise ValueError(msg)
+
+    def _get_values(self):
+        """Get values as dictionary."""
+        return {k: getattr(self, k) for k in self.__slots__}
+
+    def __eq__(self, other):
+        if type(other) is type(self):
+            return self._get_values() == other._get_values()
+        return NotImplemented
+
+    def encode(self) -> str:
+        return ", ".join(f"{k}={v}" for k, v in self._get_values().items())
+
+
+class InlineOptions(AbstractOptionValue):
+    """
+    Options for controlling inlining
+    """
+
+    def __init__(self, value):
+        ok = False
+        if isinstance(value, str):
+            if value in ("always", "never"):
+                ok = True
+        else:
+            ok = hasattr(value, "__call__")
+
+        if ok:
+            self._inline = value
+        else:
+            msg = (
+                "kwarg 'inline' must be one of the strings 'always' or "
+                "'never', or it can be a callable that returns True/False. "
+                "Found value %s" % value
+            )
+            raise ValueError(msg)
+
+    @property
+    def is_never_inline(self):
+        """
+        True if never inline
+        """
+        return self._inline == "never"
+
+    @property
+    def is_always_inline(self):
+        """
+        True if always inline
+        """
+        return self._inline == "always"
+
+    @property
+    def has_cost_model(self):
+        """
+        True if a cost model is provided
+        """
+        return not (self.is_always_inline or self.is_never_inline)
+
+    @property
+    def value(self):
+        """
+        The raw value
+        """
+        return self._inline
+
+    def __eq__(self, other):
+        if type(other) is type(self):
+            return self.value == other.value
+        return NotImplemented
+
+    def encode(self) -> str:
+        return repr(self._inline)
+
+
+class TargetOptions:
+    """Target options maps user options from decorators to the
+    ``numba.core.compiler.Flags`` used by lowering and target context.
+    """
+
+    class Mapping:
+        def __init__(self, flag_name, apply=lambda x: x):
+            self.flag_name = flag_name
+            self.apply = apply
+
+    def finalize(self, flags, options):
+        """Subclasses can override this method to make target specific
+        customizations of default flags.
+
+        Parameters
+        ----------
+        flags : Flags
+        options : dict
+        """
+        pass
+
+    @classmethod
+    def parse_as_flags(cls, flags, options):
+        """Parse target options defined in ``options`` and set ``flags``
+        accordingly.
+
+        Parameters
+        ----------
+        flags : Flags
+        options : dict
+        """
+        opt = cls()
+        opt._apply(flags, options)
+        opt.finalize(flags, options)
+        return flags
+
+    def _apply(self, flags, options):
+        # Find all Mapping instances in the class
+        mappings = {}
+        cls = type(self)
+        for k in dir(cls):
+            v = getattr(cls, k)
+            if isinstance(v, cls.Mapping):
+                mappings[k] = v
+
+        used = set()
+        for k, mapping in mappings.items():
+            if k in options:
+                v = mapping.apply(options[k])
+                setattr(flags, mapping.flag_name, v)
+                used.add(k)
+
+        unused = set(options) - used
+        if unused:
+            # Unread options?
+            m = (
+                f"Unrecognized options: {unused}. "
+                f"Known options are {mappings.keys()}"
+            )
+            raise KeyError(m)