Newer
Older
import sympy as sp
import pystencils
from pystencils.backends.cuda_backend import CudaBackend
from pystencils.backends.opencl_backend import OpenClBackend
from pystencils.opencl.opencljit import make_python_function
try:
import pyopencl as cl
HAS_OPENCL = True
except Exception:
HAS_OPENCL = False
z, y, x = pystencils.fields("z, y, x: [2d]")
assignments = pystencils.AssignmentCollection({
z[0, 0]: x[0, 0] * sp.log(x[0, 0] * y[0, 0])
})
print(assignments)
ast = pystencils.create_kernel(assignments, target='gpu')
print(ast)
code = pystencils.show_code(ast, custom_backend=CudaBackend())
print(code)
opencl_code = pystencils.show_code(ast, custom_backend=OpenClBackend())
assert "__global double * RESTRICT const _data_x" in str(opencl_code)
assert "__global double * RESTRICT" in str(opencl_code)
assert "get_local_id(0)" in str(opencl_code)
@pytest.mark.skipif(not HAS_OPENCL, reason="Test requires pyopencl")
def test_opencl_jit_fixed_size():
z, y, x = pystencils.fields("z, y, x: [20,30]")
assignments = pystencils.AssignmentCollection({
z[0, 0]: x[0, 0] * sp.log(x[0, 0] * y[0, 0])
})
print(assignments)
ast = pystencils.create_kernel(assignments, target='gpu')
print(ast)
code = pystencils.show_code(ast, custom_backend=CudaBackend())
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
print(code)
opencl_code = pystencils.show_code(ast, custom_backend=OpenClBackend())
print(opencl_code)
cuda_kernel = ast.compile()
assert cuda_kernel is not None
import pycuda.gpuarray as gpuarray
x_cpu = np.random.rand(20, 30)
y_cpu = np.random.rand(20, 30)
z_cpu = np.random.rand(20, 30)
x = gpuarray.to_gpu(x_cpu)
y = gpuarray.to_gpu(y_cpu)
z = gpuarray.to_gpu(z_cpu)
cuda_kernel(x=x, y=y, z=z)
result_cuda = z.get()
import pyopencl.array as array
ctx = cl.create_some_context(0)
queue = cl.CommandQueue(ctx)
x = array.to_device(queue, x_cpu)
y = array.to_device(queue, y_cpu)
z = array.to_device(queue, z_cpu)
opencl_kernel = make_python_function(ast, queue, ctx)
assert opencl_kernel is not None
opencl_kernel(x=x, y=y, z=z)
result_opencl = z.get(queue)
assert np.allclose(result_cuda, result_opencl)
@pytest.mark.skipif(not HAS_OPENCL, reason="Test requires pyopencl")
def test_opencl_jit():
z, y, x = pystencils.fields("z, y, x: [2d]")
assignments = pystencils.AssignmentCollection({
z[0, 0]: x[0, 0] * sp.log(x[0, 0] * y[0, 0])
})
print(assignments)
ast = pystencils.create_kernel(assignments, target='gpu')
print(ast)
code = pystencils.show_code(ast, custom_backend=CudaBackend())
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
print(code)
opencl_code = pystencils.show_code(ast, custom_backend=OpenClBackend())
print(opencl_code)
cuda_kernel = ast.compile()
assert cuda_kernel is not None
import pycuda.gpuarray as gpuarray
x_cpu = np.random.rand(20, 30)
y_cpu = np.random.rand(20, 30)
z_cpu = np.random.rand(20, 30)
x = gpuarray.to_gpu(x_cpu)
y = gpuarray.to_gpu(y_cpu)
z = gpuarray.to_gpu(z_cpu)
cuda_kernel(x=x, y=y, z=z)
result_cuda = z.get()
import pyopencl.array as array
ctx = cl.create_some_context(0)
queue = cl.CommandQueue(ctx)
x = array.to_device(queue, x_cpu)
y = array.to_device(queue, y_cpu)
z = array.to_device(queue, z_cpu)
opencl_kernel = make_python_function(ast, queue, ctx)
assert opencl_kernel is not None
opencl_kernel(x=x, y=y, z=z)
result_opencl = z.get(queue)
assert np.allclose(result_cuda, result_opencl)
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
@pytest.mark.skipif(not HAS_OPENCL, reason="Test requires pyopencl")
def test_opencl_jit_with_parameter():
z, y, x = pystencils.fields("z, y, x: [2d]")
a = sp.Symbol('a')
assignments = pystencils.AssignmentCollection({
z[0, 0]: x[0, 0] * sp.log(x[0, 0] * y[0, 0]) + a
})
print(assignments)
ast = pystencils.create_kernel(assignments, target='gpu')
print(ast)
code = pystencils.show_code(ast, custom_backend=CudaBackend())
print(code)
opencl_code = pystencils.show_code(ast, custom_backend=OpenClBackend())
print(opencl_code)
cuda_kernel = ast.compile()
assert cuda_kernel is not None
import pycuda.gpuarray as gpuarray
x_cpu = np.random.rand(20, 30)
y_cpu = np.random.rand(20, 30)
z_cpu = np.random.rand(20, 30)
x = gpuarray.to_gpu(x_cpu)
y = gpuarray.to_gpu(y_cpu)
z = gpuarray.to_gpu(z_cpu)
cuda_kernel(x=x, y=y, z=z, a=5.)
result_cuda = z.get()
import pyopencl.array as array
ctx = cl.create_some_context(0)
queue = cl.CommandQueue(ctx)
x = array.to_device(queue, x_cpu)
y = array.to_device(queue, y_cpu)
z = array.to_device(queue, z_cpu)
opencl_kernel = make_python_function(ast, queue, ctx)
assert opencl_kernel is not None
opencl_kernel(x=x, y=y, z=z, a=5.)
result_opencl = z.get(queue)
assert np.allclose(result_cuda, result_opencl)