def run_timings(k, mean=0, sd=1, a=-2, b=2, seed=101):
n = 10**k
mean = np.ones(n) * mean
sd = np.ones(n) * sd
a = np.ones(n) * a
b = np.ones(n) * b
start = cuda.Event()
end = cuda.Event()
gpu_curand_init(seed)
start.record()
gpu_trunc_norm(n, mean, sd, a, b)
end.record()
end.synchronize()
gpu_time = start.time_till(end) * 1e-3
gpu_curand_deinit()
set_seed(seed)
start.record()
cpu_trunc_norm(n, mean, sd, a, b)
end.record()
end.synchronize()
cpu_time = start.time_till(end) * 1e-3
return (gpu_time, cpu_time)
def run_timings(k, mean = 0, sd = 1, a = -2, b = 2, seed = 101):
n = 10 ** k
mean = np.ones(n) * mean
sd = np.ones(n) * sd
a = np.ones(n) * a
b = np.ones(n) * b
start = cuda.Event()
end = cuda.Event()
gpu_curand_init(seed)
start.record()
gpu_trunc_norm(n, mean, sd, a, b)
end.record()
end.synchronize()
gpu_time = start.time_till(end) * 1e-3
gpu_curand_deinit()
set_seed(seed)
start.record()
cpu_trunc_norm(n, mean, sd, a, b)
end.record()
end.synchronize()
cpu_time = start.time_till(end) * 1e-3
return (gpu_time, cpu_time)
def run_diagnostics(n, mean=0, sd=1, a=-2, b=2, seed=101):
true_mean = trunc_norm_mean(mean, sd, a, b)
mean = np.ones(n) * mean
sd = np.ones(n) * sd
a = np.ones(n) * a
b = np.ones(n) * b
gpu_curand_init(seed)
gpu_mean = np.mean(gpu_trunc_norm(n, mean, sd, a, b))
gpu_curand_deinit()
set_seed(seed)
cpu_mean = np.mean(cpu_trunc_norm(n, mean, sd, a, b))
return (gpu_mean, cpu_mean, true_mean)
def run_diagnostics(n, mean = 0, sd = 1, a = -2, b = 2, seed = 101):
true_mean = trunc_norm_mean(mean, sd, a, b)
mean = np.ones(n) * mean
sd = np.ones(n) * sd
a = np.ones(n) * a
b = np.ones(n) * b
gpu_curand_init(seed)
gpu_mean = np.mean(gpu_trunc_norm(n, mean, sd, a, b))
gpu_curand_deinit()
set_seed(seed)
cpu_mean = np.mean(cpu_trunc_norm(n, mean, sd, a, b))
return (gpu_mean, cpu_mean, true_mean)
