Skip to content

Latest commit

 

History

History

io

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Mesh I/O benchmark

Two C++ operations use the same dragon geometry ladder, from 58K to 1.026M triangles:

  • read_obj: warmed file-path read of triangulated, position-only OBJ into indexed float32 coordinates and int32 triangles.
  • read_stl: warmed binary STL file-path read into exactly welded indexed float32 coordinates and int32 triangles.

The timed region includes opening and reading the file, parsing it, exact STL vertex welding, and constructing the final indexed output arrays. Output destruction happens after the clock stops. An untimed warm-up puts the file in the OS page cache and removes storage-device variance.

Every warm-up and timed sample validates point/face counts and an order-independent signature of the oriented triangle geometry over float bits, with signed zero normalized. A parser that merely returns STL triangle soup does not satisfy the contract. Raw rows retain every timed sample and signature component; the run report records source, binary, corpus, and list hashes.

Prepare the two representations:

python corpus/prepare_io.py

The OBJ dispatch threshold has a separate generated small-file ladder:

python corpus/prepare_obj_small.py

corpus/single/io-obj-smoke.txt covers the supported fixed-triangle face spellings and the shared rejection behavior for a non-triangle face.

Build the full C++ field:

TRUEFORM_DIR=../../trueform cpp/build.sh \
  -DHAVE_RAPIDOBJ=ON -DHAVE_TINYOBJ=ON -DHAVE_STL_READER=ON \
  -DHAVE_IGL=ON -DHAVE_CGAL=ON -DHAVE_VTK=ON

Run with process-wide mimalloc through the root runner:

python run.py --op read_obj --track cpp --corpus io-obj --samples 7
python run.py --op read_stl --track cpp --corpus io-stl --samples 7

The publication run repeats every library in three fresh, order-varied processes and writes a provenance report alongside the raw data:

python cpp/io/run_ladder.py \
  --output-root results/raw/io_YYYYMMDD_final/mimalloc

The port-ready fixed-OBJ candidate and reusable tf::io::mapped_file live in cpp/io/local; cpp/io/local/PORT_NOTES.md is the authoritative copy and cross-platform validation checklist.

On macOS the runner injects /opt/homebrew/opt/mimalloc/lib/libmimalloc.dylib with DYLD_INSERT_LIBRARIES. Override the path with MIMALLOC_LIBRARY.