Seeking suggestions to optimize Clipper2 performance in WebAssembly

[ErikSom]

Hello everyone,

I'm working on porting Clipper2 to WebAssembly and aiming to unlock its full potential. While Clipper2 should ideally perform at least 30% faster than Clipper1, my current benchmarks in a WASM environment are not reflecting this expected improvement.

My build process:

C libs:

emcmake cmake ../ \
-DCMAKE_BUILD_TYPE=Release \
-DCLIPPER2_HI_PRECISION=OFF

WASM:

# Common flags
COMMON_FLAGS="-Iclipper2/CPP/Clipper2Lib/include -DUSINGZ --bind -s MODULARIZE=1 -s WASM_BIGINT -s ALLOW_MEMORY_GROWTH=1 -s EXIT_RUNTIME=0"

# Development build flags
DEV_FLAGS="-g3 -s ASSERTIONS=2 --source-map-base http://localhost:11009/ -s DISABLE_EXCEPTION_CATCHING=0 -s DEMANGLE_SUPPORT=1 -s SAFE_HEAP=1 -O0"

# Production build flags
PROD_FLAGS="-O3 -s DISABLE_EXCEPTION_CATCHING=1"

# Select flags based on build type
if [ "$BUILD_TYPE" = "dev" ]; then
    FLAGS="$COMMON_FLAGS $DEV_FLAGS"
    echo "Building Development Version"
else
    FLAGS="$COMMON_FLAGS $PROD_FLAGS"
    echo "Building Production Version"
fi

# build ES6
echo "Building ES6"
em++ $FLAGS -s EXPORT_ES6=1 -s NO_FILESYSTEM=1 -s ENVIRONMENT='web' -s clipper2-wasm/clipper.bindings.cpp clipper2/CPP/build/libClipper2Z.a -o clipper2-wasm/dist/es/clipper2z.js -s EXPORT_NAME="Clipper2Z" --post-js clipper2-wasm/glue-stub-z.js --embind-emit-tsd clipper2z.d.ts

WASM Benchmark results:

Clipper1: 2901 ms
Clipper2: 2988 ms

C Benchmark results:

Clipper1: 1058 ms
Clipper2: 761 ms

Encountered Issue

Despite the theoretical performance improvements in Clipper2, my WASM port is not outperforming the existing Clipper1 port. The native C benchmarks show the expected performance gain, but this isn't translating into WASM.

Request for Community Input

I'm looking for suggestions on:

• Enhancing the WebAssembly compilation process for better performance.
• Tuning and profiling techniques specific to WASM.
• Insights into the discrepancy between native C and WASM performance.

Any feedback or collaborative ideas would be greatly appreciated. Here's the link to my project for reference: https://github.com/ErikSom/Clipper2-WASM/.

Thank you all in advance for your help!

[ErikSom]

Here is a link to my benchmark code:
https://github.com/ErikSom/Clipper2-WASM/tree/main/clipper2-wasm/examples/benchmark

And the calls in:
https://github.com/ErikSom/Clipper2-WASM/blob/c97b96530ec1116280117ce230d897b4cb232ba0/clipper2-wasm/examples/benchmark/testPolyOperation.js#L30

Have the same call count / instructions as the Clipper1 lib as seen here:
https://github.com/xaviergonz/js-angusj-clipper/blob/380839193ce6c0d196cdfa7c7d44844bf3c2eaa9/src/clipFunctions.ts#L173

[ErikSom]

I ended up figuring out the culprit - I was comparing double operations with in64 operations, internally Clipper2 transforms doubles to ints so it was doing an additional transformation.

These are my new results:

clipper1 Intersection EvenOdd 5000 ops 5000 points: 2898.5 ms
clipper2 Intersection EvenOdd 5000 ops 5000 points: 2061.5 ms
28.88% improvement

clipper1 Union EvenOdd 5000 ops 5000 points: 3812.699999988079 ms
clipper2 Union EvenOdd 5000 ops 5000 points: 3094.800000011921 ms
18.83% improvement

clipper1 Difference EvenOdd 5000 ops 5000 points: 3291 ms
clipper2 Difference EvenOdd 5000 ops 5000 points: 2554.699999988079 ms
22.37% improvement

clipper1 Xor EvenOdd 5000 ops 5000 points: 4045.400000035763 ms
clipper2 Xor EvenOdd 5000 ops 5000 points: 3576 ms
11.60% improvement

clipper1 Intersection EvenOdd 10000 ops 100 points: 185 ms
clipper2 Intersection EvenOdd 10000 ops 100 points: 106.40000003576279 ms
42.49% improvement

clipper1 Union EvenOdd 10000 ops 100 points: 236.39999997615814 ms
clipper2 Union EvenOdd 10000 ops 100 points: 141.89999997615814 ms
39.97% improvement

clipper1 Difference EvenOdd 10000 ops 100 points: 209.19999998807907 ms
clipper2 Difference EvenOdd 10000 ops 100 points: 123.40000003576279 ms
41.01% improvement

clipper1 Xor EvenOdd 10000 ops 100 points: 256.60000002384186 ms
clipper2 Xor EvenOdd 10000 ops 100 points: 156.4000000357628 ms
39.05% improvement