Experiment with autovectorization

codecs/mozjpeg_enc/Makefile

@@ -2,14 +2,14 @@ CODEC_URL := https://github.com/mozilla/mozjpeg/archive/v3.3.1.tar.gz
 CODEC_DIR := node_modules/mozjpeg
 CODEC_OUT_RELATIVE := .libs/libjpeg.a rdswitch.o
 CODEC_OUT := $(addprefix $(CODEC_DIR)/, $(CODEC_OUT_RELATIVE))
-OUT_JS := mozjpeg_enc.js
+OUT_JS := mozjpeg_enc_simd.js
 OUT_WASM := $(OUT_JS:.js=.wasm)
 
 CFLAGS += -msimd128 -msse --em-config /src/.emscripten
 CXXFLAGS += -msimd128 -msse --em-config /src/.emscripten
 
-#override CFLAGS += "-msimd128 -msse"
+#CFLAGS += --em-config /src/.emscripten
-#override CXXFLAGS += "-msimd128 -msse"
+#CXXFLAGS += --em-config /src/.emscripten
 
 .PHONY: all clean
 
@@ -28,6 +28,7 @@ all: $(OUT_JS)
 		--closure 1 \
 		-s ALLOW_MEMORY_GROWTH=1 \
 		-s MODULARIZE=1 \
+		-s EXPORT_ES6=1 \
 		-s 'EXPORT_NAME="$(basename $(@F))"' \
 		-o $@ \
 		$+

codecs/mozjpeg_enc/bench.d8.js

@@ -0,0 +1,62 @@
+import mozjpeg from "./mozjpeg_enc.js";
+import mozjpegSimd from "./mozjpeg_enc_simd.js";
+
+const width = 2048;
+const height = 2048;
+const size = width * height * 4;
+const data = new Uint8ClampedArray(size);
+for(let i = 0; i < size; i++) {
+	data[i] = Math.random() * 256;
+}
+
+mozjpeg({
+	onRuntimeInitialized() {
+		const start = performance.now();
+		const result = this.encode(data, width, height, {
+		  quality: 75,
+		  baseline: false,
+		  arithmetic: false,
+		  progressive: true,
+		  optimize_coding: true,
+		  smoothing: 0,
+		  color_space: 3,
+		  quant_table: 3,
+		  trellis_multipass: false,
+		  trellis_opt_zero: false,
+		  trellis_opt_table: false,
+		  trellis_loops: 1,
+		  auto_subsample: true,
+		  chroma_subsample: 2,
+		  separate_chroma_quality: false,
+		  chroma_quality: 75,
+		});
+		const end = performance.now();
+		console.log("No SIMD", end - start);
+	}
+})
+
+mozjpegSimd({
+	onRuntimeInitialized() {
+		const start = performance.now();
+		const result = this.encode(data, width, height, {
+		  quality: 75,
+		  baseline: false,
+		  arithmetic: false,
+		  progressive: true,
+		  optimize_coding: true,
+		  smoothing: 0,
+		  color_space: 3,
+		  quant_table: 3,
+		  trellis_multipass: false,
+		  trellis_opt_zero: false,
+		  trellis_opt_table: false,
+		  trellis_loops: 1,
+		  auto_subsample: true,
+		  chroma_subsample: 2,
+		  separate_chroma_quality: false,
+		  chroma_quality: 75,
+		});
+		const end = performance.now();
+		console.log("SIMD", end - start);
+	}
+})

codecs/mozjpeg_enc/mozjpeg_enc.cpp

@@ -1,7 +1,5 @@
 #include <emscripten/bind.h>
 #include <emscripten/val.h>
-#include <wasm_simd128.h>
-#include <xmmintrin.h>
 #include <inttypes.h>
 #include <setjmp.h>
 #include <stdio.h>

codecs/mozjpeg_enc/mozjpeg_enc.js

@@ -1,7 +1,7 @@
 
 var mozjpeg_enc = (function() {
-  var _scriptDir = typeof document !== 'undefined' && document.currentScript ? document.currentScript.src : undefined;
+  var _scriptDir = import.meta.url;
-  if (typeof __filename !== 'undefined') _scriptDir = _scriptDir || __filename;
+  
   return (
 function(mozjpeg_enc) {
   mozjpeg_enc = mozjpeg_enc || {};
@@ -63,10 +63,4 @@ d.run=Cb;if(d.preInit)for("function"==typeof d.preInit&&(d.preInit=[d.preInit]);
 }
 );
 })();
-if (typeof exports === 'object' && typeof module === 'object')
+export default mozjpeg_enc;
-      module.exports = mozjpeg_enc;
-    else if (typeof define === 'function' && define['amd'])
-      define([], function() { return mozjpeg_enc; });
-    else if (typeof exports === 'object')
-      exports["mozjpeg_enc"] = mozjpeg_enc;
-