Question

我以前在Tensorflow 1.2中使用以下命令：

export TF_XLA_FLAGS='--dump_ir_before_passes=true --dump_temp_products_to=./tmp'

用于在Tensorflow中转储LLVM IR。但是，在Tensorflow 1.3中删除了此标志link_to_the_flag_definition的定义文件，我现在想知道如何获得LLVM IR转储？

这是一个方便的测试文件：

import tensorflow as tf
import numpy as np
import os
from tensorflow.python.client import timeline
import json

run_metadata = tf.RunMetadata()
sess = tf.Session()
jit_scope = tf.contrib.compiler.jit.experimental_jit_scope
x = tf.placeholder(np.float32, shape=[1000000])
y = tf.placeholder(np.float32, shape=[1000000])
c = tf.constant(0.1)

with jit_scope():
  z = tf.add(tf.scalar_mul(0.1,x), y)
  ix = np.ones((1000000), dtype=np.float32)
  iy = np.ones((1000000), dtype=np.float32)
  sess.run(z,
           feed_dict={x: ix, y: iy},
           options=tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE),
           run_metadata=run_metadata)
  trace = timeline.Timeline(step_stats=run_metadata.step_stats)
  with open('timeline.ctf.json', 'w') as trace_file:
    trace_file.write(trace.generate_chrome_trace_format())

Answer 1

我在这里https://github.com/tensorflow/tensorflow/issues/11462找到了标志function convertToOHLC(data) { var parsedData = JSON.parse(data), pointStart = parsedData[0].date, range = [], low, high, ranges = [], dataOHLC = [], interval = 60 * 1000; parsedData.sort(function(a, b) { return a.date - b.date }); $.each(parsedData, function(i, el) { if (pointStart + interval < el.date) { ranges.push(range.slice()); range = []; range.push(el); pointStart = pointStart + interval; } else { range.push(el); } if (i === parsedData.length - 1) { ranges.push(range); } }); $.each(ranges, function(i, range) { low = range[0].price; high = range[0].price; $.each(range, function(i, el) { low = Math.min(low, el.price); high = Math.max(high, el.price); }); dataOHLC.push({ x: range[0].date + 30 * 1000, open: Number(range[0].price), high: high, low: low, close: Number(range[range.length - 1].price) }); }); return dataOHLC }。它是在Tensorflow 1.3中添加的。

如何转储tensorflow XLA LLVM IR？

1 个答案: