Skip to content

Szalinski: A Tool for Synthesizing Structured CAD Models with Equality Saturation and Inverse Transformations

License

Notifications You must be signed in to change notification settings

uwplse/szalinski

Repository files navigation

PLDI 2020 AEC, Paper# 471

NOTE: In the rest of this document, we refer to our tool as Szalinski. In our submission, we called it Albatross for anonymity.

Goals of the artifact

In our paper, we evaluated the following about Szalinski (Section 7):

  1. End-to-End: we ran Szalinski on the flat CSG outputs of a mesh decompiler (Reincarnate). The results are in Table 2.

  2. Scalability: we evaluated Szalinski on a large dataset of models scraped from a popular online repository (Thingiverse). The results are in Figure 14 (first three box plots).

  3. Sensitivity: we evaluated the usefulness of Szalinski's two main features: CAD rewrites and Inverse Transformations. The results are in Figure 14 (last two box plots).

In support of these results, this artifact reproduces Table 2 and Figure 14. In addition, it also generates the output programs in Figure 15 that are used in the case studies.

This document contains the following parts:

  • System requirements

  • Getting started

  • How to run Szalinski

    • Reproducing Table 2 (takes < 5 minutes)
    • Reproducing Figure 14 (takes approx. 1.5 hour)
    • Reproducing Figure 15 (takes < 5 minutes)
    • Validation
  • Reusability

    • How to set up Szalinski on a different machine (this is also how we set up the VM)
    • Description of the code and how to modify it
  • Notes and remarks

System requirements

  • We provide the artifact as a virtual machine image. To open it you need virtual box version 6.1.2, which can be downloaded here.

  • Specs of the machine where we ran the VM: Intel i7-8700K (12 threads @ 4.7GHz), 32GiB RAM

Getting started

  • Please download the .ova file here and open it with Virtual Box by going to File -> import appliance and giving the path to the .ova file and clicking on continue. In the next window that pops up, click on Import. It should take a few minutes to import.

NOTE: When you import the .ova file, we recommend that you increase the CPU count as much as you can afford. Similarly, when running make, we recommend adding make -jN where N is the number of CPUs you allocated to the VM.

  • Next, please open the virtual machine image in virtual box by clicking on the green Start button.

  • Login is automatic, but in case needed, the password is: pldi2020.

  • The terminal should be open at startup. The project repository is already cloned. Navigate to the szalinski directory. All the required packages are already installed and Szalinki is already compiled for you, ready to be run.

  • To allow a quick verification of our artifact, we provided pre-generated data and results in the VM. You can therefore skip the make commands in the instructions and directly view the results (see below on how to do that).

  • As a next step, you can verify that the results are indeed generated from the data we provided. To do so, delete the results (out/aec-table2/table2.csv, out/fig14.pdf) and run the make commands as explained below.

  • To run the tool yourself entirely from scratch, first delete the entire out directory and follow the instructions below.

Running the tools

Reproducing Table 2

Navigate to the directory that contains the Makefile and type make out/aec-table2/table2.csv. This should take about 3 minutes. This will reproduce Table 2 from the paper. To view the content of the table, type cat out/aec-table2/table2.csv | column -t -s, and compare the numbers with Table 2 in the paper.

NOTE:

  • We have significantly improved Szalinski since the PLDI deadline. As a result, for several case studies, the numbers in the last three columns of the table are lower (hence better in this case) than what is reported in the paper.
  • We suspect that different versions of OpenSCAD use different triangulation algorithms for compiling to mesh. The version supported by Ubuntu 19.10 (the VM) is different from the version we used during the deadline because we ran it on a MacOS. Due to this, the numbers in the #Tri column may vary in this artifact.

Reproducing Figure 14

We have included in the repo the 2,127 examples from Thingiverse that we evaluated on in the paper. The remainder of the 12,939 scraped from Thingiverse were either malformed or used features not supported by Szalinski. The script (scripts/scrape-thingiverse.sh) scrapes models under the customizable category, from the first 500 pages.

NOTE: Running this part takes about an hour. We recommend first reproducing Figure 15 and Table 2, both of which take much less time.

Navigate to the directory that contains the Makefile and type make out/fig14.pdf. Open the generated file in a pdf viewer and compare with Figure 14 in the paper.

Reproducing Figure 15 programs

Navigate to the directory that contains the Makefile and type make aec-fig15. This should take less than a minute. Then look in the out/aec-fig15 directory. The optimized programs generated by Szalinski are in the files with extensions normal.csexp.opt. There should be 6 such files. Open them and compare the content with the programs listed in Figure 15 of the paper.

NOTE:

  • The programs in the paper are sugared and represented more compactly for space.
  • MapI found in the artifact results corresponds to Tabulate in the paper.
  • When comparing the results generated by the artifact to the programs in Figure 15 of the paper, it is most important to check that the high-level structure in terms of Fold and MapI synthesized by the artifact matches that reported in the paper.

Validation

Section 6 of our paper describes Szalinski's validation process. We use OpenSCAD to compile CSG programs to meshes and use CGAL to compute the Hausdorff distance between two meshes.

To validate the programs in Figure 15, run make out/aec-fig15/hausdorff. This should terminate in less than 3 minutes. It should show you the names of the 6 examples in Figure 15 and the corresponding Hausdorff distances which are close to zero.

We have also validated all our other results reported in the paper. However, our experience indicates that OpenSCAD's compilation step is often very slow. Therefore, the other commands mentioned in the instruction for reproducing the results do not perform validation by default.

You can validate any example from our evaluation by typing: make out/dir_name/example_name.normal.diff, where dir_name can be aec-table2, aec_fig15 or thingiverse, and example_name is the name of whatever example you choose. Then open the generated .diff file and check that the Hausdorff distance is within some epsilon of 0.

NOTE: For many example, CGAL crashes or is slow at computing the Hausdorff distance. For these, we recommend a manual validation if you are interested. In order to validate an example, type the following: make out/dir_name/example_name.diff.scad. You can open the generated .scad file in OpenSCAD (already installed in the VM). In OpenSCAD, click on the Render button (the second button from the right) in the toolbar. You should either see nothing rendered or some residual thin walls that are artifacts of rounding error prevalent in OpenSCAD.

Reusability

Here we provide instructions on how to start using Szalinski including installation and changing the rules and features of the Caddy language.

Setup instructions

Following are the steps for setting up Szalinski from scratch on a different machine that runs Ubuntu 19.10.

  • Install rust. Type curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh in the terminal and follow the subsequent instructions. The version we used is 1.41.0. Seehttps://www.rust-lang.org/tools/install for more information.

  • Make sure you configure your current shell by typing: source $HOME/.cargo/env (the Rust installation will prompt you to do this).

  • Install make by typing: sudo apt-get install make

  • Install g++ by typing: sudo apt-get install g++

  • Install jq by typing: sudo apt-get install jq

  • Install CGAL by typing sudo apt-get install libcgal-dev

  • Install OpenSCAD by typing sudo apt-get install openscad

  • Install git by typing sudo apt install git

  • Install pip by typing sudo apt install python3-pip and then install numpy by typing pip3 install numpy and matplotlib by typing pip3 install matplotlib

  • We have made a github release for the PLDI AEC from where you can get the source.

  • Navigate to the project directory where the Makefile is and run the tool as described above.

Changing Caddy and modifying the rules

  • The Caddy language is defined in cad.rs in the src directory. A simple feature you can add is support for a new primitive or new transformations. You can also change the costs of various language constructs. The definition of the cost function starts at line 267.

  • As we described in the paper, to verify the correctness of Szalinski, we evaluate Caddy programs to flat Core Caddy and pretty print to CSG. This code is in eval.rs.

  • solve.rs and permute.rs contains code that solves for first and second degree polynomials in Cartesian and Spherical coordinates, and performs partitioning and permutations of lists.

  • The rewrites rules are in rules.rs. Syntactic rewrites are written using the rw! macro. Each rewrite has a name, a left hand side, and a right hand side. You can add / remove rules to see how that affects the final Caddy output of Szalinski. For example, if you comment out the rules for inverse transformations, they will not be propagated and eliminated, and therefore the quality of Szalinski's output will not be as good.

Notes and remarks

Szalinski is implemented in Rust. As mentioned in Section 6 of the paper, it uses OpenSCAD to compile CSG programs to triangular meshes, and CGAL to compute the Hausdorff distance between two meshes.