import tensorflow as tf
from numpy.testing.utils import nulp_diff
import time

n = 1024
v1 = tf.Variable(tf.ones_initializer((n, n)))
v2 = tf.Variable(tf.ones_initializer((n, n)))
op = v1.assign(tf.matmul(v1, v2)/n).op
norm = tf.reduce_sum(v1)

sess = tf.Session()
sess.run(tf.initialize_all_variables())
sess.run(op)  # pre-warming
start_time = time.time()
sess.run(tf.initialize_all_variables())
num_iters = 10
start = time.time()
for i in range(num_iters):
  sess.run(op)

elapsed = time.time()-start
error = nulp_diff(n**2, sess.run(norm))
total_ops = num_iters*(n**3+n**2*(n-1))
print("%.2f G ops/sec, error %.2f"%(total_ops/elapsed/10**9,error))