From f1e3f7a0f8a531a388da74ba8732349db016521a Mon Sep 17 00:00:00 2001 From: Jacob Freck Date: Thu, 8 Mar 2018 12:35:29 -0800 Subject: [PATCH] Bug: remove unnecessary example (#417) --- examples/src/main/python/gpu/nubma_example.py | 67 ------------------- 1 file changed, 67 deletions(-) delete mode 100644 examples/src/main/python/gpu/nubma_example.py diff --git a/examples/src/main/python/gpu/nubma_example.py b/examples/src/main/python/gpu/nubma_example.py deleted file mode 100644 index 689d772a..00000000 --- a/examples/src/main/python/gpu/nubma_example.py +++ /dev/null @@ -1,67 +0,0 @@ -import numpy as np -import numba -import math -import time -from numba import cuda, vectorize, jit -from pyspark.sql import SparkSession - -@cuda.jit('(float32[:], float32[:])') -def func(inp, out): - i = cuda.grid(1) - if i < out.size: - out[i] = inp[i] ** 2 - - -def gpu_work(xs): - inp = np.asarray(list(xs), dtype=np.float32) - out = np.zeros_like(inp) - block_size = 32 * 4 - grid_size = (inp.size + block_size - 1) - func[grid_size, block_size](inp, out) - return out - -spark = SparkSession\ - .builder\ - .getOrCreate() - -sc = spark.sparkContext - -# detect if there is an available GPU -if cuda.is_available(): - rdd = sc.parallelize(list(range(100))) - print("Partitions", rdd.getNumPartitions()) - - rdd = rdd.mapPartitions(gpu_work) - print(rdd.collect()) - - - # compare cpu and gpu run times - - @vectorize(["float32(float32, float32)", "float64(float64, float64)"], target='cpu') - def cpu_some_trig(x, y): - return math.cos(x) + math.sin(y) - - @vectorize(["float32(float32, float32)", "float64(float64, float64)"], target='cuda') - def cuda_some_trig(x, y): - return math.cos(x) + math.sin(y) - - - nelem = 10 ** 8 - - xs = np.random.random(nelem).astype(np.float32) - ys = np.random.random(nelem).astype(np.float32) - - start_time = time.time() - cpu_some_trig(xs, ys) - cpu_elapsed_time = time.time() - start_time - - start_time = time.time() - cuda_some_trig(xs, ys) - gpu_elapsed_time = time.time() - start_time - - print("CPU time: {}".format(cpu_elapsed_time)) - print("GPU time: {}".format(gpu_elapsed_time)) - print("GPU performed {}% faster than CPU".format((cpu_elapsed_time-gpu_elapsed_time)*100)) - -else: - print("No CUDA devices detected. Are you running this on a GPU enabled VM?")