aforren1

Reducing Chrome’s Presentation Latency on Windows

Alex Forrence
email: alex.forrence@gmail.com

Intro

For this project, I improved presentation latency in Chrome on Windows by introducing waitable swap chains. Waitable swap chains allow us to both reduce latency by shortening the queue of pending frames and aligning frame production for optimal consumption by the operating system, and reduce latency spikes by dropping frames under load.

Initial proposal:

aforren1

import moderngl as mgl
import glfw
import numpy as np
import glm
import mapbox_earcut as earcut
from timeit import default_timer
from xxhash import xxh32

vao_cache = {}

aforren1

Usually 10648 draw calls (22 Dim) at 60fps

aforren1

# Speech 2

## Gratitude

Great job to <maid of honor>! I was tearing up there at the end.

I'd like to say a few words of thanks:

First and foremost to Natasha and Matt, for inviting all of us to share in their special day 

aforren1

https://ci.appveyor.com/api/buildjobs/9lty5fdd9a8y133i/artifacts/dist/ratcave-0.8.0-cp37-cp37m-win_amd64.whl

aforren1

Install raspbian (https://www.raspberrypi.org/downloads/raspbian/)

run `sudo apt update && sudo apt upgrade --yes`

Turn on desktop GL in `sudo raspi-config`
(sudo apt install mesa-utils & run glxgears for verification)

edit /boot/config.txt and add `gpu_freq=500` (overclocks the GPU from 400->500) and `gpu_mem=320` (gives the GPU more memory)

aforren1

# Install preemptive kernel
apt install build-essential git linux-image-rt-amd64 -y # check with `uname -r`

# Install miniconda (update later via `conda update conda`
wget http://repo.continuum.io/miniconda/Miniconda3-3.7.0-Linux-x86_64.sh -O ~/miniconda.sh
bash ~/miniconda.sh -b -p $HOME/miniconda # accept license, add to path
echo 'export PATH="$HOME/miniconda3/bin:$PATH"' >> ~/.bashrc

# update audio settings

aforren1

class Experiment(Machine):
    def is_line_crossed(self): return self.line_crossed
    
    def reset_line(self): self.line_crossed = False
    
    def done_countdown(self): 
        self.countdown_done = True
        print "Done the countdown"
        
    def start_countdown(self):

aforren1

IDE

• rstudio

Reproducible research tools

• packrat - roughly equivalent to python's virtualenv (isolated package env)
• drake/remake - like GNU make (I think drake is better maintained at this point, but few examples)
• RStudio's projects - at least drops you in the right working directory & ties in with packrat
• Make a package! - higher startup cost, but user installation is trivial
• rocker (R through Docker)

aforren1

We'll stick to 32-bit installation (to make iohub happy?)

1. Install python 2.7.12 x86 (make sure to add python to your PATH var, just for ease-of-use)

• We use this version solely for compatibility with the nidaqmx package (2.7.14 being compatible when it comes out)
• Ideally, psychopy will eventually support 3.x, which is actively developed

2. Install wheels for numpy and scipy from http://www.lfd.uci.edu/~gohlke/pythonlibs/

• for instance, use numpy‑1.12.1+mkl‑cp27‑cp27m‑win32.whl because we're using python 2.7 on 32 bit
• python -m pip install <path_to_numpy_whl>.whl
• We install these two globally because they're cumbersome to keep re-installing and rather large