numba-cuda 0.0.1__py3-none-any.whl → 0.0.13__py3-none-any.whl

numba_cuda/numba/cuda/tests/data/warn.cu

@@ -0,0 +1,7 @@
+extern "C" __device__
+int bar(int* out, int a) {
+  // Explicitly placed to generate a warning for testing the NVRTC program log
+  int unused;
+  *out = a * 2;
+  return 0;
+}

numba_cuda/numba/cuda/tests/doc_examples/__init__.py

@@ -0,0 +1,6 @@
+from numba.cuda.tests import load_testsuite
+import os
+
+
+def load_tests(loader, tests, pattern):
+    return load_testsuite(loader, os.path.dirname(__file__))

numba_cuda/numba/cuda/tests/doc_examples/ffi/functions.cu

@@ -0,0 +1,49 @@
+// magictoken.ex_mul_f32_f32.begin
+// Foreign function example: multiplication of a pair of floats
+
+extern "C" __device__ int
+mul_f32_f32(
+  float* return_value,
+  float x,
+  float y)
+{
+  // Compute result and store in caller-provided slot
+  *return_value = x * y;
+
+  // Signal that no Python exception occurred
+  return 0;
+}
+// magictoken.ex_mul_f32_f32.end
+
+
+// magictoken.ex_sum_reduce_proto.begin
+extern "C"
+__device__ int
+sum_reduce(
+  float* return_value,
+  float* array,
+  int n
+);
+// magictoken.ex_sum_reduce_proto.end
+
+
+// Performs a simple reduction on an array passed by pointer using the
+// ffi.from_buffer() method. Implements the prototype above.
+extern "C"
+__device__ int
+sum_reduce(
+  float* return_value,
+  float* array,
+  int n
+)
+{
+  double sum = 0.0;
+
+  for (size_t i = 0; i < n; ++i) {
+    sum += array[i];
+  }
+
+  *return_value = (float)sum;
+
+  return 0;
+}

numba_cuda/numba/cuda/tests/doc_examples/test_cg.py

@@ -0,0 +1,77 @@
+# Contents in this file are referenced from the sphinx-generated docs.
+# "magictoken" is used for markers as beginning and ending of example text.
+
+import unittest
+from numba.cuda.testing import (CUDATestCase, skip_on_cudasim,
+                                skip_if_cudadevrt_missing, skip_unless_cc_60,
+                                skip_if_mvc_enabled)
+
+
+@skip_if_cudadevrt_missing
+@skip_unless_cc_60
+@skip_if_mvc_enabled('CG not supported with MVC')
+@skip_on_cudasim("cudasim doesn't support cuda import at non-top-level")
+class TestCooperativeGroups(CUDATestCase):
+    def test_ex_grid_sync(self):
+        # magictoken.ex_grid_sync_kernel.begin
+        from numba import cuda, int32
+        import numpy as np
+
+        sig = (int32[:,::1],)
+
+        @cuda.jit(sig)
+        def sequential_rows(M):
+            col = cuda.grid(1)
+            g = cuda.cg.this_grid()
+
+            rows = M.shape[0]
+            cols = M.shape[1]
+
+            for row in range(1, rows):
+                opposite = cols - col - 1
+                # Each row's elements are one greater than the previous row
+                M[row, col] = M[row - 1, opposite] + 1
+                # Wait until all threads have written their column element,
+                # and that the write is visible to all other threads
+                g.sync()
+        # magictoken.ex_grid_sync_kernel.end
+
+        # magictoken.ex_grid_sync_data.begin
+        # Empty input data
+        A = np.zeros((1024, 1024), dtype=np.int32)
+        # A somewhat arbitrary choice (one warp), but generally smaller block sizes
+        # allow more blocks to be launched (noting that other limitations on
+        # occupancy apply such as shared memory size)
+        blockdim = 32
+        griddim = A.shape[1] // blockdim
+        # magictoken.ex_grid_sync_data.end
+
+        # Skip this test if the grid size used in the example is too large for
+        # a cooperative launch on the current GPU
+        mb = sequential_rows.overloads[sig].max_cooperative_grid_blocks(blockdim)
+        if mb < griddim:
+            self.skipTest('Device does not support a large enough coop grid')
+
+        # magictoken.ex_grid_sync_launch.begin
+        # Kernel launch - this is implicitly a cooperative launch
+        sequential_rows[griddim, blockdim](A)
+
+        # What do the results look like?
+        # print(A)
+        #
+        # [[   0    0    0 ...    0    0    0]
+        #  [   1    1    1 ...    1    1    1]
+        #  [   2    2    2 ...    2    2    2]
+        #  ...
+        #  [1021 1021 1021 ... 1021 1021 1021]
+        #  [1022 1022 1022 ... 1022 1022 1022]
+        #  [1023 1023 1023 ... 1023 1023 1023]]
+        # magictoken.ex_grid_sync_launch.end
+
+        # Sanity check - are the results what we expect?
+        reference = np.tile(np.arange(1024), (1024, 1)).T
+        np.testing.assert_equal(A, reference)
+
+
+if __name__ == '__main__':
+    unittest.main()

numba_cuda/numba/cuda/tests/doc_examples/test_cpu_gpu_compat.py

@@ -0,0 +1,76 @@
+import unittest
+
+from numba.cuda.testing import CUDATestCase, skip_on_cudasim
+from numba.tests.support import captured_stdout
+import numpy as np
+
+
+@skip_on_cudasim("cudasim doesn't support cuda import at non-top-level")
+class TestCpuGpuCompat(CUDATestCase):
+    """
+    Test compatibility of CPU and GPU functions
+    """
+
+    def setUp(self):
+        # Prevent output from this test showing up when running the test suite
+        self._captured_stdout = captured_stdout()
+        self._captured_stdout.__enter__()
+        super().setUp()
+
+    def tearDown(self):
+        # No exception type, value, or traceback
+        self._captured_stdout.__exit__(None, None, None)
+        super().tearDown()
+
+    def test_ex_cpu_gpu_compat(self):
+        # ex_cpu_gpu_compat.import.begin
+        from math import pi
+
+        import numba
+        from numba import cuda
+        # ex_cpu_gpu_compat.import.end
+
+        # ex_cpu_gpu_compat.allocate.begin
+        X = cuda.to_device([1, 10, 234])
+        Y = cuda.to_device([2, 2, 4014])
+        Z = cuda.to_device([3, 14, 2211])
+        results = cuda.to_device([0.0, 0.0, 0.0])
+        # ex_cpu_gpu_compat.allocate.end
+
+        # ex_cpu_gpu_compat.define.begin
+        @numba.jit
+        def business_logic(x, y, z):
+            return 4 * z * (2 * x - (4 * y) / 2 * pi)
+        # ex_cpu_gpu_compat.define.end
+
+        # ex_cpu_gpu_compat.cpurun.begin
+        print(business_logic(1, 2, 3))  # -126.79644737231007
+        # ex_cpu_gpu_compat.cpurun.end
+
+        # ex_cpu_gpu_compat.usegpu.begin
+        @cuda.jit
+        def f(res, xarr, yarr, zarr):
+            tid = cuda.grid(1)
+            if tid < len(xarr):
+                # The function decorated with numba.jit may be directly reused
+                res[tid] = business_logic(xarr[tid], yarr[tid], zarr[tid])
+        # ex_cpu_gpu_compat.usegpu.end
+
+        # ex_cpu_gpu_compat.launch.begin
+        f.forall(len(X))(results, X, Y, Z)
+        print(results)
+        # [-126.79644737231007, 416.28324559588634, -218912930.2987788]
+        # ex_cpu_gpu_compat.launch.end
+
+        expect = [
+            business_logic(x, y, z) for x, y, z in zip(X, Y, Z)
+        ]
+
+        np.testing.assert_equal(
+            expect,
+            results.copy_to_host()
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()

numba_cuda/numba/cuda/tests/doc_examples/test_ffi.py

@@ -0,0 +1,82 @@
+# Contents in this file are referenced from the sphinx-generated docs.
+# "magictoken" is used for markers as beginning and ending of example text.
+
+import unittest
+from numba.cuda.testing import (CUDATestCase, skip_on_cudasim)
+from numba.tests.support import skip_unless_cffi
+
+
+@skip_unless_cffi
+@skip_on_cudasim("cudasim doesn't support cuda import at non-top-level")
+class TestFFI(CUDATestCase):
+    def test_ex_linking_cu(self):
+        # magictoken.ex_linking_cu.begin
+        from numba import cuda
+        import numpy as np
+        import os
+
+        # Declaration of the foreign function
+        mul = cuda.declare_device('mul_f32_f32', 'float32(float32, float32)')
+
+        # Path to the source containing the foreign function
+        # (here assumed to be in a subdirectory called "ffi")
+        basedir = os.path.dirname(os.path.abspath(__file__))
+        functions_cu = os.path.join(basedir, 'ffi', 'functions.cu')
+
+        # Kernel that links in functions.cu and calls mul
+        @cuda.jit(link=[functions_cu])
+        def multiply_vectors(r, x, y):
+            i = cuda.grid(1)
+
+            if i < len(r):
+                r[i] = mul(x[i], y[i])
+
+        # Generate random data
+        N = 32
+        np.random.seed(1)
+        x = np.random.rand(N).astype(np.float32)
+        y = np.random.rand(N).astype(np.float32)
+        r = np.zeros_like(x)
+
+        # Run the kernel
+        multiply_vectors[1, 32](r, x, y)
+
+        # Sanity check - ensure the results match those expected
+        np.testing.assert_array_equal(r, x * y)
+        # magictoken.ex_linking_cu.end
+
+    def test_ex_from_buffer(self):
+        from numba import cuda
+        import os
+
+        basedir = os.path.dirname(os.path.abspath(__file__))
+        functions_cu = os.path.join(basedir, 'ffi', 'functions.cu')
+
+        # magictoken.ex_from_buffer_decl.begin
+        signature = 'float32(CPointer(float32), int32)'
+        sum_reduce = cuda.declare_device('sum_reduce', signature)
+        # magictoken.ex_from_buffer_decl.end
+
+        # magictoken.ex_from_buffer_kernel.begin
+        import cffi
+        ffi = cffi.FFI()
+
+        @cuda.jit(link=[functions_cu])
+        def reduction_caller(result, array):
+            array_ptr = ffi.from_buffer(array)
+            result[()] = sum_reduce(array_ptr, len(array))
+        # magictoken.ex_from_buffer_kernel.end
+
+        import numpy as np
+        x = np.arange(10).astype(np.float32)
+        r = np.ndarray((), dtype=np.float32)
+
+        reduction_caller[1, 1](r, x)
+
+        expected = np.sum(x)
+        actual = r[()]
+        np.testing.assert_allclose(expected, actual)
+
+
+if __name__ == '__main__':
+    unittest.main()

numba_cuda/numba/cuda/tests/doc_examples/test_laplace.py

@@ -0,0 +1,155 @@
+import unittest
+
+from numba.cuda.testing import (CUDATestCase, skip_if_cudadevrt_missing,
+                                skip_on_cudasim, skip_unless_cc_60,
+                                skip_if_mvc_enabled)
+from numba.tests.support import captured_stdout
+
+
+@skip_if_cudadevrt_missing
+@skip_unless_cc_60
+@skip_if_mvc_enabled('CG not supported with MVC')
+@skip_on_cudasim("cudasim doesn't support cuda import at non-top-level")
+class TestLaplace(CUDATestCase):
+    """
+    Test simple vector addition
+    """
+
+    def setUp(self):
+        # Prevent output from this test showing up when running the test suite
+        self._captured_stdout = captured_stdout()
+        self._captured_stdout.__enter__()
+        super().setUp()
+
+    def tearDown(self):
+        # No exception type, value, or traceback
+        self._captured_stdout.__exit__(None, None, None)
+        super().tearDown()
+
+    def test_ex_laplace(self):
+
+        # set True to regenerate the figures that
+        # accompany this example
+        plot = False
+
+        # ex_laplace.import.begin
+        import numpy as np
+        from numba import cuda
+        # ex_laplace.import.end
+
+        # ex_laplace.allocate.begin
+        # Use an odd problem size.
+        # This is so there can be an element truly in the "middle" for symmetry.
+        size = 1001
+        data = np.zeros(size)
+
+        # Middle element is made very hot
+        data[500] = 10000
+        buf_0 = cuda.to_device(data)
+
+        # This extra array is used for synchronization purposes
+        buf_1 = cuda.device_array_like(buf_0)
+
+        niter = 10000
+        # ex_laplace.allocate.end
+
+        if plot:
+            import matplotlib.pyplot as plt
+            fig, ax = plt.subplots(figsize=(16 * 0.66, 9 * 0.66))
+            plt.plot(
+                np.arange(len(buf_0)),
+                buf_0.copy_to_host(),
+                lw=3,
+                marker="*",
+                color='black'
+            )
+
+            plt.title('Initial State', fontsize=24)
+            plt.xlabel('Position', fontsize=24)
+            plt.ylabel('Temperature', fontsize=24)
+
+            ax.set_xticks(ax.get_xticks(), fontsize=16)
+            ax.set_yticks(ax.get_yticks(), fontsize=16)
+            plt.xlim(0, len(data))
+            plt.ylim(0, 10001)
+            plt.savefig('laplace_initial.svg')
+
+        # ex_laplace.kernel.begin
+        @cuda.jit
+        def solve_heat_equation(buf_0, buf_1, timesteps, k):
+            i = cuda.grid(1)
+
+            # Don't continue if our index is outside the domain
+            if i >= len(buf_0):
+                return
+
+            # Prepare to do a grid-wide synchronization later
+            grid = cuda.cg.this_grid()
+
+            for step in range(timesteps):
+                # Select the buffer from the previous timestep
+                if (step % 2) == 0:
+                    data = buf_0
+                    next_data = buf_1
+                else:
+                    data = buf_1
+                    next_data = buf_0
+
+                # Get the current temperature associated with this point
+                curr_temp = data[i]
+
+                # Apply formula from finite difference equation
+                if i == 0:
+                    # Left wall is held at T = 0
+                    next_temp = curr_temp + k * (data[i + 1] - (2 * curr_temp))
+                elif i == len(data) - 1:
+                    # Right wall is held at T = 0
+                    next_temp = curr_temp + k * (data[i - 1] - (2 * curr_temp))
+                else:
+                    # Interior points are a weighted average of their neighbors
+                    next_temp = curr_temp + k * (
+                        data[i - 1] - (2 * curr_temp) + data[i + 1]
+                    )
+
+                # Write new value to the next buffer
+                next_data[i] = next_temp
+
+                # Wait for every thread to write before moving on
+                grid.sync()
+        # ex_laplace.kernel.end
+
+        # ex_laplace.launch.begin
+        solve_heat_equation.forall(len(data))(
+            buf_0, buf_1, niter, 0.25
+        )
+        # ex_laplace.launch.end
+
+        results = buf_1.copy_to_host()
+        if plot:
+            fig, ax = plt.subplots(figsize=(16 * 0.66, 9 * 0.66))
+            plt.plot(
+                np.arange(len(results)),
+                results, lw=3,
+                marker="*",
+                color='black'
+            )
+            plt.title(f"T = {niter}", fontsize=24)
+            plt.xlabel('Position', fontsize=24)
+            plt.ylabel('Temperature', fontsize=24)
+
+            ax.set_xticks(ax.get_xticks(), fontsize=16)
+            ax.set_yticks(ax.get_yticks(), fontsize=16)
+
+            plt.ylim(0, max(results))
+            plt.xlim(0, len(results))
+            plt.savefig('laplace_final.svg')
+
+        # Integral over the domain should be equal to its initial value.
+        # Note that this should match the initial value of data[500] above, but
+        # we don't assign it to a variable because that would make the example
+        # code look a bit oddly verbose.
+        np.testing.assert_allclose(results.sum(), 10000)
+
+
+if __name__ == "__main__":
+    unittest.main()

numba_cuda/numba/cuda/tests/doc_examples/test_matmul.py

@@ -0,0 +1,173 @@
+"""
+Matrix multiplication example via `cuda.jit`.
+
+Reference: https://stackoverflow.com/a/64198479/13697228 by @RobertCrovella
+
+Contents in this file are referenced from the sphinx-generated docs.
+"magictoken" is used for markers as beginning and ending of example text.
+"""
+import unittest
+from numba.cuda.testing import CUDATestCase, skip_on_cudasim
+from numba.tests.support import captured_stdout
+
+
+@skip_on_cudasim("cudasim doesn't support cuda import at non-top-level")
+class TestMatMul(CUDATestCase):
+    """
+    Text matrix multiplication using simple, shared memory/square, and shared
+    memory/nonsquare cases.
+    """
+
+    def setUp(self):
+        # Prevent output from this test showing up when running the test suite
+        self._captured_stdout = captured_stdout()
+        self._captured_stdout.__enter__()
+        super().setUp()
+
+    def tearDown(self):
+        # No exception type, value, or traceback
+        self._captured_stdout.__exit__(None, None, None)
+        super().tearDown()
+
+    def test_ex_matmul(self):
+        """Test of matrix multiplication on various cases."""
+        # magictoken.ex_import.begin
+        from numba import cuda, float32
+        import numpy as np
+        import math
+        # magictoken.ex_import.end
+
+        # magictoken.ex_matmul.begin
+        @cuda.jit
+        def matmul(A, B, C):
+            """Perform square matrix multiplication of C = A * B."""
+            i, j = cuda.grid(2)
+            if i < C.shape[0] and j < C.shape[1]:
+                tmp = 0.
+                for k in range(A.shape[1]):
+                    tmp += A[i, k] * B[k, j]
+                C[i, j] = tmp
+        # magictoken.ex_matmul.end
+
+        # magictoken.ex_run_matmul.begin
+        x_h = np.arange(16).reshape([4, 4])
+        y_h = np.ones([4, 4])
+        z_h = np.zeros([4, 4])
+
+        x_d = cuda.to_device(x_h)
+        y_d = cuda.to_device(y_h)
+        z_d = cuda.to_device(z_h)
+
+        threadsperblock = (16, 16)
+        blockspergrid_x = math.ceil(z_h.shape[0] / threadsperblock[0])
+        blockspergrid_y = math.ceil(z_h.shape[1] / threadsperblock[1])
+        blockspergrid = (blockspergrid_x, blockspergrid_y)
+
+        matmul[blockspergrid, threadsperblock](x_d, y_d, z_d)
+        z_h = z_d.copy_to_host()
+        print(z_h)
+        print(x_h @ y_h)
+        # magictoken.ex_run_matmul.end
+
+        # magictoken.ex_fast_matmul.begin
+        # Controls threads per block and shared memory usage.
+        # The computation will be done on blocks of TPBxTPB elements.
+        # TPB should not be larger than 32 in this example
+        TPB = 16
+
+        @cuda.jit
+        def fast_matmul(A, B, C):
+            """
+            Perform matrix multiplication of C = A * B using CUDA shared memory.
+
+            Reference: https://stackoverflow.com/a/64198479/13697228 by @RobertCrovella
+            """
+            # Define an array in the shared memory
+            # The size and type of the arrays must be known at compile time
+            sA = cuda.shared.array(shape=(TPB, TPB), dtype=float32)
+            sB = cuda.shared.array(shape=(TPB, TPB), dtype=float32)
+
+            x, y = cuda.grid(2)
+
+            tx = cuda.threadIdx.x
+            ty = cuda.threadIdx.y
+            bpg = cuda.gridDim.x    # blocks per grid
+
+            # Each thread computes one element in the result matrix.
+            # The dot product is chunked into dot products of TPB-long vectors.
+            tmp = float32(0.)
+            for i in range(bpg):
+                # Preload data into shared memory
+                sA[ty, tx] = 0
+                sB[ty, tx] = 0
+                if y < A.shape[0] and (tx + i * TPB) < A.shape[1]:
+                    sA[ty, tx] = A[y, tx + i * TPB]
+                if x < B.shape[1] and (ty + i * TPB) < B.shape[0]:
+                    sB[ty, tx] = B[ty + i * TPB, x]
+
+                # Wait until all threads finish preloading
+                cuda.syncthreads()
+
+                # Computes partial product on the shared memory
+                for j in range(TPB):
+                    tmp += sA[ty, j] * sB[j, tx]
+
+                # Wait until all threads finish computing
+                cuda.syncthreads()
+            if y < C.shape[0] and x < C.shape[1]:
+                C[y, x] = tmp
+        # magictoken.ex_fast_matmul.end
+
+        # magictoken.ex_run_fast_matmul.begin
+        x_h = np.arange(16).reshape([4, 4])
+        y_h = np.ones([4, 4])
+        z_h = np.zeros([4, 4])
+
+        x_d = cuda.to_device(x_h)
+        y_d = cuda.to_device(y_h)
+        z_d = cuda.to_device(z_h)
+
+        threadsperblock = (TPB, TPB)
+        blockspergrid_x = math.ceil(z_h.shape[0] / threadsperblock[0])
+        blockspergrid_y = math.ceil(z_h.shape[1] / threadsperblock[1])
+        blockspergrid = (blockspergrid_x, blockspergrid_y)
+
+        fast_matmul[blockspergrid, threadsperblock](x_d, y_d, z_d)
+        z_h = z_d.copy_to_host()
+        print(z_h)
+        print(x_h @ y_h)
+        # magictoken.ex_run_fast_matmul.end
+
+        # fast_matmul test(s)
+        msg = "fast_matmul incorrect for shared memory, square case."
+        self.assertTrue(np.all(z_h == x_h @ y_h), msg=msg)
+
+        # magictoken.ex_run_nonsquare.begin
+        x_h = np.arange(115).reshape([5, 23])
+        y_h = np.ones([23, 7])
+        z_h = np.zeros([5, 7])
+
+        x_d = cuda.to_device(x_h)
+        y_d = cuda.to_device(y_h)
+        z_d = cuda.to_device(z_h)
+
+        threadsperblock = (TPB, TPB)
+        grid_y_max = max(x_h.shape[0], y_h.shape[0])
+        grid_x_max = max(x_h.shape[1], y_h.shape[1])
+        blockspergrid_x = math.ceil(grid_x_max / threadsperblock[0])
+        blockspergrid_y = math.ceil(grid_y_max / threadsperblock[1])
+        blockspergrid = (blockspergrid_x, blockspergrid_y)
+
+        fast_matmul[blockspergrid, threadsperblock](x_d, y_d, z_d)
+        z_h = z_d.copy_to_host()
+        print(z_h)
+        print(x_h @ y_h)
+        # magictoken.ex_run_nonsquare.end
+
+        # nonsquare fast_matmul test(s)
+        msg = "fast_matmul incorrect for shared memory, non-square case."
+        self.assertTrue(np.all(z_h == x_h @ y_h), msg=msg)
+
+
+if __name__ == '__main__':
+    unittest.main()