numba-cuda 0.19.1__py3-none-any.whl → 0.20.1__py3-none-any.whl

numba_cuda/numba/cuda/tests/cudapy/test_ssa.py

@@ -0,0 +1,453 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: BSD-2-Clause
+"""
+Tests for SSA reconstruction
+"""
+
+import sys
+import copy
+import logging
+
+import numpy as np
+
+from numba import types, cuda
+from numba.cuda import jit
+from numba.core import errors
+
+from numba.extending import overload
+from numba.cuda.tests.support import override_config
+from numba.cuda.testing import CUDATestCase, skip_on_cudasim
+
+
+_DEBUG = False
+
+if _DEBUG:
+    # Enable debug logger on SSA reconstruction
+    ssa_logger = logging.getLogger("numba.cuda.core.ssa")
+    ssa_logger.setLevel(level=logging.DEBUG)
+    ssa_logger.addHandler(logging.StreamHandler(sys.stderr))
+
+
+class SSABaseTest(CUDATestCase):
+    """
+    This class comes from numba tests, but has been modified to work with CUDA kernels.
+    Return values were replaced by output arrays, and tuple returns assign to elements of the output array.
+    """
+
+    def check_func(self, func, result_array, *args):
+        # For CUDA kernels, we need to create output arrays and call with [1,1] launch config
+        # Create GPU array with same shape as expected result array
+        gpu_result_array = cuda.to_device(np.zeros_like(result_array))
+
+        # Call the CUDA kernel
+        func[1, 1](gpu_result_array, *copy.deepcopy(args))
+        gpu_result = gpu_result_array.copy_to_host()
+
+        # Call the original Python function for expected result
+        cpu_result = np.zeros_like(result_array)
+        func.py_func(cpu_result, *copy.deepcopy(args))
+
+        # Compare all results
+        np.testing.assert_array_equal(gpu_result, cpu_result)
+
+
+class TestSSA(SSABaseTest):
+    """
+    Contains tests to help isolate problems in SSA
+    """
+
+    def test_argument_name_reused(self):
+        @jit
+        def foo(result, x):
+            x += 1
+            result[0] = x
+
+        self.check_func(foo, np.array([124.0]), 123)
+
+    def test_if_else_redefine(self):
+        @jit
+        def foo(result, x, y):
+            z = x * y
+            if x < y:
+                z = x
+            else:
+                z = y
+            result[0] = z
+
+        self.check_func(foo, np.array([2.0]), 3, 2)
+        self.check_func(foo, np.array([2.0]), 2, 3)
+
+    def test_sum_loop(self):
+        @jit
+        def foo(result, n):
+            c = 0
+            for i in range(n):
+                c += i
+            result[0] = c
+
+        self.check_func(foo, np.array([0.0]), 0)
+        self.check_func(foo, np.array([45.0]), 10)
+
+    def test_sum_loop_2vars(self):
+        @jit
+        def foo(result, n):
+            c = 0
+            d = n
+            for i in range(n):
+                c += i
+                d += n
+            result[0] = c
+            result[1] = d
+
+        self.check_func(foo, np.array([0.0, 0.0]), 0)
+        self.check_func(foo, np.array([45.0, 110.0]), 10)
+
+    def test_sum_2d_loop(self):
+        @jit
+        def foo(result, n):
+            c = 0
+            for i in range(n):
+                for j in range(n):
+                    c += j
+                c += i
+            result[0] = c
+
+        self.check_func(foo, np.array([0.0]), 0)
+        self.check_func(foo, np.array([495.0]), 10)
+
+    def check_undefined_var(self, should_warn):
+        @jit
+        def foo(result, n):
+            if n:
+                if n > 0:
+                    c = 0
+                result[0] = c
+            else:
+                # variable c is not defined in this branch
+                c += 1
+                result[0] = c
+
+        if should_warn:
+            with self.assertWarns(errors.NumbaWarning) as warns:
+                # n=1 so we won't actually run the branch with the uninitialized
+                self.check_func(foo, np.array([0]), 1)
+            self.assertIn(
+                "Detected uninitialized variable c", str(warns.warning)
+            )
+        else:
+            self.check_func(foo, np.array([0]), 1)
+
+        with self.assertRaises(UnboundLocalError):
+            result = np.array([0])
+            foo.py_func(result, 0)
+
+    @skip_on_cudasim(
+        "Numba variable warnings are not supported in the simulator"
+    )
+    def test_undefined_var(self):
+        with override_config("ALWAYS_WARN_UNINIT_VAR", 0):
+            self.check_undefined_var(should_warn=False)
+        with override_config("ALWAYS_WARN_UNINIT_VAR", 1):
+            self.check_undefined_var(should_warn=True)
+
+    def test_phi_propagation(self):
+        @jit
+        def foo(result, actions):
+            n = 1
+
+            i = 0
+            ct = 0
+            while n > 0 and i < len(actions):
+                n -= 1
+
+                while actions[i]:
+                    if actions[i]:
+                        if actions[i]:
+                            n += 10
+                        actions[i] -= 1
+                    else:
+                        if actions[i]:
+                            n += 20
+                        actions[i] += 1
+
+                    ct += n
+                ct += n
+            result[0] = ct
+            result[1] = n
+
+        self.check_func(foo, np.array([1, 2]), np.array([1, 2]))
+
+    def test_unhandled_undefined(self):
+        @cuda.jit
+        def function1(arg1, arg2, arg3, arg4, arg5):
+            # This function is auto-generated.
+            if arg1:
+                var1 = arg2
+                var2 = arg3
+                var3 = var2
+                var4 = arg1
+                return
+            else:
+                if arg2:
+                    if arg4:
+                        var5 = arg4  # noqa: F841
+                        return
+                    else:
+                        var6 = var4
+                        return
+                    return var6
+                else:
+                    if arg5:
+                        if var1:
+                            if arg5:
+                                var1 = var6
+                                return
+                            else:
+                                var7 = arg2  # noqa: F841
+                                return arg2
+                            return
+                        else:
+                            if var2:
+                                arg5 = arg2
+                                return arg1
+                            else:
+                                var6 = var3
+                                return var4
+                            return
+                        return
+                    else:
+                        var8 = var1
+                        return
+                    return var8
+                var9 = var3  # noqa: F841
+                var10 = arg5  # noqa: F841
+                return var1
+
+        NONE_SENTINEL = 99
+
+        @cuda.jit
+        def function1_caller(result, arg1, arg2, arg3, arg4, arg5):
+            retval = function1(arg1, arg2, arg3, arg4, arg5)
+            if retval is None:
+                result[0] = NONE_SENTINEL
+            else:
+                result[0] = retval
+
+        # The argument values is not critical for re-creating the bug
+        # because the bug is in compile-time.
+
+        expect = function1.py_func(2, 3, 6, 0, 7)
+        if expect is None:
+            expect = NONE_SENTINEL
+        result = np.zeros(1, dtype=np.int64)
+        function1_caller[1, 1](result, 2, 3, 6, 0, 7)
+        got = result[0]
+        self.assertEqual(expect, got)
+
+
+class TestReportedSSAIssues(SSABaseTest):
+    # Tests from issues
+    # https://github.com/numba/numba/issues?q=is%3Aopen+is%3Aissue+label%3ASSA
+
+    def test_issue2194(self):
+        @jit
+        def foo(result, V):
+            s = np.uint32(1)
+
+            for i in range(s):
+                V[i] = 1
+            for i in range(s, 1):
+                pass
+            result[0] = V[0]
+
+        V = np.empty(1)
+        self.check_func(foo, np.array([1.0]), V)
+
+    def test_issue3094(self):
+        @jit
+        def foo(result, pred):
+            if pred:
+                x = 1
+            else:
+                x = 0
+            result[0] = x
+
+        self.check_func(foo, np.array([0]), False)
+
+    def test_issue3931(self):
+        @jit
+        def foo(result, arr):
+            for i in range(1):
+                arr = arr.reshape(3 * 2)
+                arr = arr.reshape(3, 2)
+            # Copy result array elements
+            for i in range(arr.shape[0]):
+                for j in range(arr.shape[1]):
+                    result[i, j] = arr[i, j]
+
+        result_gpu = np.zeros((3, 2))
+        self.check_func(foo, result_gpu, np.zeros((3, 2)))
+
+    def test_issue3976(self):
+        def overload_this(a):
+            return 42
+
+        @jit
+        def foo(result, a):
+            if a:
+                s = 5
+                s = overload_this(s)
+            else:
+                s = 99
+
+            result[0] = s
+
+        @overload(overload_this)
+        def ol(a):
+            return overload_this
+
+        self.check_func(foo, np.array([42]), True)
+
+    def test_issue3979(self):
+        @jit
+        def foo(result, A, B):
+            x = A[0]
+            y = B[0]
+            for i in A:
+                x = i
+            for i in B:
+                y = i
+            result[0] = x
+            result[1] = y
+
+        self.check_func(
+            foo, np.array([2, 4]), np.array([1, 2]), np.array([3, 4])
+        )
+
+    def test_issue5219(self):
+        def overload_this(a, b=None):
+            if isinstance(b, tuple):
+                b = b[0]
+            return b
+
+        @overload(overload_this)
+        def ol(a, b=None):
+            b_is_tuple = isinstance(b, (types.Tuple, types.UniTuple))
+
+            def impl(a, b=None):
+                if b_is_tuple is True:
+                    b = b[0]
+                return b
+
+            return impl
+
+        @jit
+        def test_tuple(result, a, b):
+            result[0] = overload_this(a, b)
+
+        self.check_func(test_tuple, np.array([2]), 1, (2,))
+
+    def test_issue5223(self):
+        @jit
+        def bar(result, x):
+            if len(x) == 5:
+                for i in range(len(x)):
+                    result[i] = x[i]
+            else:
+                # Manual copy since .copy() not available in CUDA
+                for i in range(len(x)):
+                    result[i] = x[i] + 1
+
+        a = np.ones(5)
+        a.flags.writeable = False
+        expected = np.ones(5)  # Since len(a) == 5, it should return unchanged
+        self.check_func(bar, expected, a)
+
+    def test_issue5243(self):
+        @jit
+        def foo(result, q, lin):
+            stencil_val = 0.0  # noqa: F841
+            stencil_val = q[0, 0]  # noqa: F841
+            result[0] = lin[0]
+
+        lin = np.array([0.1, 0.6, 0.3])
+        self.check_func(foo, np.array([0.1]), np.zeros((2, 2)), lin)
+
+    def test_issue5482_missing_variable_init(self):
+        # Test error that lowering fails because variable is missing
+        # a definition before use.
+        @jit
+        def foo(result, x, v, n):
+            for i in range(n):
+                if i == 0:
+                    if i == x:
+                        pass
+                    else:
+                        problematic = v
+                else:
+                    if i == x:
+                        pass
+                    else:
+                        problematic = problematic + v
+            result[0] = problematic
+
+        self.check_func(foo, np.array([10]), 1, 5, 3)
+
+    def test_issue5493_unneeded_phi(self):
+        # Test error that unneeded phi is inserted because variable does not
+        # have a dominance definition.
+        data = (np.ones(2), np.ones(2))
+        A = np.ones(1)
+        B = np.ones(1)
+
+        @jit
+        def foo(res, m, n, data):
+            if len(data) == 1:
+                v0 = data[0]
+            else:
+                v0 = data[0]
+                # Unneeded PHI node for `problematic` would be placed here
+                for _ in range(1, len(data)):
+                    v0[0] += A[0]
+
+            for t in range(1, m):
+                for idx in range(n):
+                    t = B
+
+                    if idx == 0:
+                        if idx == n - 1:
+                            pass
+                        else:
+                            res[0] = t[0]
+                    else:
+                        if idx == n - 1:
+                            pass
+                        else:
+                            res[0] += t[0]
+
+        self.check_func(foo, np.array([10]), 10, 10, data)
+
+    def test_issue5623_equal_statements_in_same_bb(self):
+        def foo(pred, stack):
+            i = 0
+            c = 1
+
+            if pred is True:
+                stack[i] = c
+                i += 1
+                stack[i] = c
+                i += 1
+
+        python = np.array([0, 666])
+        foo(True, python)
+
+        nb = np.array([0, 666])
+
+        # Convert to CUDA kernel
+        foo_cuda = jit(foo)
+        foo_cuda[1, 1](True, nb)
+
+        expect = np.array([1, 1])
+
+        np.testing.assert_array_equal(python, expect)
+        np.testing.assert_array_equal(nb, expect)

numba_cuda/numba/cuda/tests/cudapy/test_sync.py

@@ -4,7 +4,7 @@
 import numpy as np
 from numba import cuda, int32, float32
 from numba.cuda.testing import skip_on_cudasim, unittest, CUDATestCase
-from numba.core.config import ENABLE_CUDASIM
+from numba.cuda.core.config import ENABLE_CUDASIM
 
 
 def useless_syncthreads(ary):