def bench(arrayLength, dtype, innerRepeat, outerRepeat):
import numpy as np
setup = """
a = np.full(arrayLength, 1, dtype=real_t);
b = np.full(arrayLength, 2, dtype=real_t);
c = np.full(arrayLength, 3, dtype=real_t);
alpha = real_t(42)
"""
times = {
"copy": 'copy(a)',
"scale": 'scale(a, alpha)',
"add": 'add(a, b)',
"triad": 'triad(a, b, alpha)'
}
namespace = {
"copy": copy,
"scale": scale,
"add": add,
"triad": triad,
"np": np,
"arrayLength": arrayLength,
"real_t": dtype
}
state = state_o(namespace, innerRepeat, outerRepeat)
results = benchmark(times, inspect.cleandoc(setup), state)
return [("COPY", results["copy"]), ("SCALE", results["scale"]),
("ADD", results["add"]), ("TRIAD", results["triad"])]
def bench(arrayLength, dtype, innerRepeat, outerRepeat):
setup = """
a = init(real_t(1), arrayLength);
b = init(real_t(2), arrayLength);
c = init(real_t(3), arrayLength);
alpha = real_t(42);
"""
times = {
"copy": 'copy(a, b)',
"scale": 'scale(a, b, alpha)',
"add": 'add(a, b, c)',
"triad": 'triad(a, b, c, alpha)'
}
namespace = {
"copy": copy,
"scale": scale,
"add": add,
"triad": triad,
"init": init,
"arrayLength": arrayLength,
"real_t": dtype
}
state = state_o(namespace, innerRepeat, outerRepeat)
results = benchmark(times, inspect.cleandoc(setup), state)
return [
("COPY", results["copy"]),
("SCALE", results["scale"]),
("ADD", results["add"]),
("TRIAD", results["triad"])
]
def bench(arrayLength, dtype, innerRepeat, outerRepeat):
import numpy as np
setup = """
a = np.full(arrayLength, 1, dtype=real_t);
b = np.full(arrayLength, 2, dtype=real_t);
c = np.full(arrayLength, 3, dtype=real_t);
alpha = real_t(42)
"""
times = {
"copy": 'copy(a, b)',
"scale": 'scale(a, b, alpha)',
"add": 'add(a, b, c)',
"triad": 'triad(a, b, c, alpha)'
}
namespace = {
"copy": copy,
"scale": scale,
"add": add,
"triad": triad,
"np": np,
"arrayLength": arrayLength,
"real_t": dtype
}
state = state_o(namespace, innerRepeat, outerRepeat)
results = benchmark(times, inspect.cleandoc(setup), state)
logger.debug("NUMBA Threading Layer: {}".format(npyufunc.parallel.threading_layer()))
logger.debug("NUMBA N Threads: {}".format(npyufunc.parallel.get_thread_count()))
return [
("COPY", results["copy"]),
("SCALE", results["scale"]),
("ADD", results["add"]),
("TRIAD", results["triad"])
]
def bench(arrayLength, dtype, innerRepeat, outerRepeat):
import numpy as np
setup = """
a = np.full(arrayLength, 1, dtype=real_t);
b = np.full(arrayLength, 2, dtype=real_t);
c = np.full(arrayLength, 3, dtype=real_t);
alpha = real_t(42)
stream_a = cuda.stream()
stream_b = cuda.stream()
stream_c = cuda.stream()
d_a = cuda.to_device(a, stream_a)
d_b = cuda.to_device(b, stream_b)
d_c = cuda.to_device(c, stream_c)
n_threads = 256
n_blocks = np.ceil(arrayLength / n_threads).astype("int32")
"""
times = {
"copy": 'copy[n_blocks, n_threads](d_a, d_b)',
"scale": 'scale(a, b, alpha)',
"add": 'add(a, b, c)',
"triad": 'triad(a, b, c, alpha)'
}
namespace = {
"np": np,
"cuda": cuda,
"copy": copy,
"scale": scale,
"add": add,
"triad": triad,
"arrayLength": arrayLength,
"real_t": dtype
}
state = state_o(namespace, innerRepeat, outerRepeat)
results = benchmark(times, inspect.cleandoc(setup), state)
return [("COPY", results["copy"]), ("SCALE", results["scale"]),
("ADD", results["add"]), ("TRIAD", results["triad"])]