better profiler and benchmark

[panyx0718]

1. fix fluid_benchmark train_parallel() wrong trainer_id (always 0 trainer_id in train_parallel()).
2. fix profiler int value overflow
3. more flexible profiler python API

try it with

#!/bin/bash

PADDLE_TRAINING_ROLE=PSERVER PADDLE_PSERVER_PORT=7164 PADDLE_PSERVER_IPS=127.0.0.1 PADDLE_TRAINERS=2 PADDLE_CURRENT_IP=127.0.0.1 PADDLE_TRAINER_ID=0 python ../Paddle/benchmark/fluid/fluid_benchmark.py --model resnet --device GPU --update_method pserver --iterations=10000 &

sleep 15

CUDA_VISIBLE_DEVICES=0,1 PADDLE_TRAINING_ROLE=TRAINER PADDLE_PSERVER_PORT=7164 PADDLE_PSERVER_IPS=127.0.0.1 PADDLE_TRAINERS=2 PADDLE_CURRENT_IP=127.0.0.1 PADDLE_TRAINER_ID=0 python ../Paddle/benchmark/fluid/fluid_benchmark.py --model resnet --device GPU --update_method pserver --iterations=10000 --gpus 2 --profile &

CUDA_VISIBLE_DEVICES=4,5 PADDLE_TRAINING_ROLE=TRAINER PADDLE_PSERVER_PORT=7164 PADDLE_PSERVER_IPS=127.0.0.1 PADDLE_TRAINERS=2 PADDLE_CURRENT_IP=127.0.0.1 PADDLE_TRAINER_ID=1 python ../Paddle/benchmark/fluid/fluid_benchmark.py --model resnet --device GPU --update_method pserver --iterations=10000 --gpus 2 --profile &

[typhoonzero]

LGTM! Thanks for this fix!