#!/usr/bin/env bash
# Builds wasm/matmul.c two ways:
#   - single-threaded ES module -> wasm/dist/matmul.{js,wasm}
#     (naive + blocked kernels; loaded through Vite's module worker)
#   - threaded (pthreads) module -> public/matmul/matmul_mt.{js,wasm}
#     (blocked+SIMD parallelized over rows; loaded by public/matmul/worker.js,
#      a classic worker, since it spawns nested pthread workers)
# Requires: emsdk (expected at ~/emsdk, or already on PATH).
set -euo pipefail

cd "$(dirname "$0")"

if ! command -v emcc >/dev/null 2>&1; then
  source ~/emsdk/emsdk_env.sh
fi

mkdir -p dist ../public/matmul

# Single-threaded ES module.
emcc -O3 -msimd128 matmul.c \
  -sMODULARIZE=1 \
  -sEXPORT_ES6=1 \
  -sEXPORT_NAME=createMatmulModule \
  -sENVIRONMENT=web,worker \
  -sALLOW_MEMORY_GROWTH=1 \
  -sEXPORTED_FUNCTIONS=_matmul_naive,_matmul_blocked,_malloc,_free \
  -sEXPORTED_RUNTIME_METHODS=HEAPF64 \
  -o dist/matmul.js

cp matmul.d.ts dist/matmul.d.ts

# Threaded (pthreads) module. Classic (non-ES6) module so a classic worker can
# importScripts it and spawn pthread workers. Fixed shared memory (growth is
# costly with shared memory); pool must cover max threads + margin.
emcc -O3 -msimd128 -pthread -DMATMUL_MT matmul.c \
  -sMODULARIZE=1 \
  -sEXPORT_NAME=createMatmulMT \
  -sENVIRONMENT=web,worker,node \
  -sPTHREAD_POOL_SIZE=20 \
  -sINITIAL_MEMORY=512MB \
  -sEXPORTED_FUNCTIONS=_matmul_blocked_mt,_malloc,_free \
  -sEXPORTED_RUNTIME_METHODS=HEAPF64 \
  -o ../public/matmul/matmul_mt.js

echo "Built wasm/dist/matmul.{js,wasm,d.ts} + public/matmul/matmul_mt.{js,wasm}"