V0XNIHILI

AWS EC2 LaTeX

Setup

[ec2-user ~]$ wget http://ftp.jaist.ac.jp/pub/CTAN/systems/texlive/tlnet/install-tl-unx.tar.gz
[ec2-user ~]$ tar xzf install-tl-unx.tar.gz
[ec2-user ~]$ cd install-tl-2017xxxx
[ec2-user ~]$ sudo ./install-tl

V0XNIHILI

Multicore Python

Below you can find some sample template code which you can use to run multicore Python computations.

# =================================================================================================
# Imports
# =================================================================================================

import multiprocessing as mp

V0XNIHILI

for y in range(0, 1000):
    for m in range(1,13):
        max_day = 32

        # Month length correction
        if m == 11 or m == 9 or m == 6 or m == 4:
            max_day = 31
        # Leap year correction
        elif m == 2:
            y_actual = y + 2000

V0XNIHILI

sysctl -n machdep.cpu.brand_string

# Example output: Intel(R) Core(TM) i7-4870HQ CPU @ 2.50GHz

V0XNIHILI

On my 2014 MacBook Pro (2.5 GHz), this vectorized version is about 250 times faster when used for a large numbers of triangles compared to the original algorithm. You can also parallelize this code over multiple threads, but due to the large overhead of creating a thread pool, this is only really useful when you have a very large number of rays.

I also tried to vectorize the rays, instead of using a loop. I did this by using:

np.reshape(np.array(np.meshgrid(np.arange(0, len(ray_direction), 1), np.arange(0, len(triangles_vertices), 1))), (2, resulting_array_length))

to create all combinations of indices of the rays and the triangles. However, using this process, combined with using np.take() to actually create these combinations is about twice as slow as the code below.

V0XNIHILI

void divide (float a, float b, float *c)
{
#pragma HLS interface s_axilite port=a
#pragma HLS interface s_axilite port=b
#pragma HLS interface s_axilite port=c
#pragma HLS interface ap_ctrl_none port=return
	*c = a / b;
}

V0XNIHILI

from typing import List, Any, Callable

def zip_with (func: Callable[[Any, Any], Any], list1: List[Any], list2: List[Any]) -> Any:
    """
    >>> zip_with (lambda x, y: x + y, [1, 2, 3], [4, 5, 6])
    [5, 7, 9]
    """
    
    return [func (x, y) for x, y in zip (list1, list2)]
  

V0XNIHILI

import torch
import torch.nn as nn
import torchvision
from torch.utils.data import Dataset

class SequentialMNIST(Dataset):

    def __init__(self, MNIST_dataset: torchvision.datasets.MNIST):
        self.MNIST_dataset = MNIST_dataset

V0XNIHILI

V0XNIHILI

"""Data taken from paper: Computing's Energy Problem (and what we can do about it) - Mark Horowitz

All output energies are in pJ (pico Joule)
"""

import math

E_INT_ADD_PER_BIT = (0.03 / 8 + 0.1 / 32) / 2