w1th0utnam3

Maximizing performance on modern architectures with data-level parallelism

Abstract

In the solution process of problems discretized using FEM, the assembly of element level tensors to the global matrix often contributes a significant amount of computational time relative to the overall process. The goal of this GSoC 2018 project is to improve the performance of the FEniCS framework in the assembly phase. To achieve this, the FEniCS Form Compiler (FFC) should be enhanced such that the generated code can fully utilize data-level

w1th0utnam3

# Dockerfile to build the FEniCS-X development libraries
#   with support for Loopy and Intel OpenCL drivers
#
# Authors:
# Jack S. Hale <[email protected]>
# Lizao Li <[email protected]>
# Garth N. Wells <[email protected]>
# Jan Blechta <[email protected]>
# Fabian Löschner <[email protected]>

w1th0utnam3

import dolfin
import dolfin.cpp
from dolfin import *
from dolfin.la import PETScMatrix, PETScVector, PETScKrylovSolver
from dolfin.jit.jit import ffc_jit

import numba as nb
import numpy as np

import time

w1th0utnam3

import dolfin
import dolfin.cpp
from dolfin import *
from dolfin.la import PETScMatrix, PETScVector, PETScKrylovSolver
from dolfin.jit.jit import ffc_jit

import numba as nb
import numpy as np

import cffi

w1th0utnam3

void tabulate_tensor_A(
    double* restrict A, 
    const double* const* w,
    const double* restrict coordinate_dofs,
    int cell_orientation)
{
    // Precomputed values of basis functions and precomputations
    // FE* dimensions: [entities][points][dofs]
    // PI* dimensions: [entities][dofs][dofs] or [entities][dofs]
    // PM* dimensions: [entities][dofs][dofs]

w1th0utnam3

from ufl import *
import ffc.compiler

# Define cell type we solve on
cell = tetrahedron
d = cell.geometric_dimension()

# The finite element function space
element = VectorElement("Lagrange", cell, 1, dim=d)

w1th0utnam3

GPU Passthrough for VM on Ubuntu

Main tutorial: https://ycnrg.org/vga-passthrough-with-ovmf-vfio/

Other resources:

• https://wiki.archlinux.org/index.php/PCI_passthrough_via_OVMF
• https://github.com/sk1080/nvidia-kvm-patcher

Reset issue After powering off the Windows VM, it cannot be started again without a host power-cycle. Might be related to Windows not properly resetting the device and Linux being unable to power-cycle it by itself.

w1th0utnam3

Links

Workflow:

• Node version manager on Windows: https://github.com/coreybutler/nvm-windows
• Node version manager on Linux/macOS: https://github.com/nvm-sh/nvm
• Docker and nodejs: https://hackernoon.com/a-better-way-to-develop-node-js-with-docker-cd29d3a0093

React and TypeScript:

• Create a TypeScript based React app: https://create-react-app.dev/docs/adding-typescript
• React + TypeScript cheatsheet: https://github.com/typescript-cheatsheets/react-typescript-cheatsheet

w1th0utnam3

use glutin::event_loop::{ControlFlow, EventLoop};
use glutin::window::WindowBuilder;

fn main() {
    let el = EventLoop::new();
    let wb = WindowBuilder::new();
    let windowed_context = glutin::ContextBuilder::new()
        .build_windowed(wb, &el)
        .unwrap();

w1th0utnam3

import matplotlib as mpl

def draw_loglog_slope(fig, ax, origin, width_inches, slope, inverted=False, color=None, polygon_kwargs=None, label=True, labelcolor=None, label_kwargs=None, zorder=None):
    """
    This function draws slopes or "convergence triangles" into loglog plots.

    @param fig: The figure
    @param ax: The axes object to draw to
    @param origin: The 2D origin (usually lower-left corner) coordinate of the triangle
    @param width_inches: The width in inches of the triangle