<< 13/22 >>
First Last

matmul + add

XLAなし: 222.77 GFLOPS
XLAあり: 207.17 GFLOPS (7.5%遅い)

 def gen_graph():
     N = 1024
     a = tf.Variable(tf.random_normal([N, N]), name='a')
     b = tf.Variable(tf.random_normal([N, N]), name='b')
     c = tf.Variable(tf.random_normal([N, N]), name='c')
     r = tf.matmul(a, b)
     r = tf.add(r, c, name='result')
     return r