Skip to content

Move optimization pipeline from JS to Wasm #1234

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Apr 21, 2020
Merged

Move optimization pipeline from JS to Wasm #1234

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3814cb1
Move optimization logic to Wasm
dcodeIO Apr 20, 2020
1308c1b
Merge branch 'master' into optimize-in-wasm
dcodeIO Apr 21, 2020
ead4a0c
regenerate fixtures
dcodeIO Apr 21, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
223 changes: 11 additions & 212 deletions cli/asc.js
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -635,26 +635,22 @@ exports.main = function main(argv, options, callback) {
if (args.trapMode === "clamp") {
stats.optimizeCount++;
stats.optimizeTime += measure(() => {
module.runPasses([ "trap-mode-clamp" ]);
module.runPass("trap-mode-clamp");
});
} else if (args.trapMode === "js") {
stats.optimizeCount++;
stats.optimizeTime += measure(() => {
module.runPasses([ "trap-mode-js" ]);
module.runPass("trap-mode-js");
});
} else if (args.trapMode !== "allow") {
module.dispose();
return callback(Error("Unsupported trap mode"));
}

// Implicitly run costly non-LLVM optimizations on -O3 or -Oz
// see: https://github.com/WebAssembly/binaryen/pull/1596
if (optimizeLevel >= 3 || shrinkLevel >= 2) optimizeLevel = 4;

module.setOptimizeLevel(optimizeLevel);
module.setShrinkLevel(shrinkLevel);
module.setDebugInfo(args.debug);

// Optimize the module
const debugInfo = args.debug;
const usesARC = args.runtime == "half" || args.runtime == "full";
const converge = args.converge;
const runPasses = [];
if (args.runPasses) {
if (typeof args.runPasses === "string") {
Expand All @@ -668,213 +664,16 @@ exports.main = function main(argv, options, callback) {
}
}

function doOptimize() {
const hasARC = args.runtime == "half" || args.runtime == "full";
const passes = [];
function add(pass) { passes.push(pass); }

if (optimizeLevel >= 2 && shrinkLevel === 0) {
// tweak inlining options when speed more preferable than size
module.setAlwaysInlineMaxSize(12);
module.setFlexibleInlineMaxSize(70);
module.setOneCallerInlineMaxSize(200);
} else {
// tweak inlining options when size matters
optimizeLevel === 0 && shrinkLevel >= 0
? module.setAlwaysInlineMaxSize(2)
: module.setAlwaysInlineMaxSize(4); // default: 2
module.setFlexibleInlineMaxSize(65); // default: 20
module.setOneCallerInlineMaxSize(80); // default: 15
}

// Optimize the module if requested
if (optimizeLevel > 0 || shrinkLevel > 0) {
// Binaryen's default passes with Post-AssemblyScript passes added.
// see: Binaryen/src/pass.cpp

// PassRunner::addDefaultGlobalOptimizationPrePasses
add("duplicate-function-elimination");
add("remove-unused-module-elements"); // differs

// PassRunner::addDefaultFunctionOptimizationPasses
if (optimizeLevel >= 3 || shrinkLevel >= 1) {
add("ssa-nomerge");
}
if (optimizeLevel >= 3) {
add("flatten"); // differs
add("simplify-locals-notee-nostructure"); // differs
add("vacuum"); // differs
add("code-folding"); // differs
add("flatten");
add("local-cse");
add("reorder-locals"); // differs
}
if (optimizeLevel >= 2 || shrinkLevel >= 1) { // differs
add("rse");
add("vacuum");
}
if (hasARC) { // differs
if (optimizeLevel < 3) {
add("flatten");
}
add("post-assemblyscript");
}
add("optimize-instructions"); // differs
add("inlining"); // differs
add("dce");
add("remove-unused-brs");
add("remove-unused-names");
add("inlining-optimizing"); // differs
if (optimizeLevel >= 2 || shrinkLevel >= 1) {
add("pick-load-signs");
add("simplify-globals-optimizing"); // differs
}
if (optimizeLevel >= 3 || shrinkLevel >= 2) {
add("precompute-propagate");
} else {
add("precompute");
}
add("vacuum"); // differs
// this will be done later (1)
// if (optimizeLevel >= 2 || shrinkLevel >= 2) {
// add("code-pushing");
// }
if (optimizeLevel >= 3 && shrinkLevel <= 1) { // differs
add("licm");
}
add("simplify-locals-nostructure");
add("vacuum");
add("reorder-locals");
add("remove-unused-brs");
// if (optimizeLevel >= 3 || shrinkLevel >= 2) { // do it later
// add("merge-locals");
// }
add("coalesce-locals");
add("simplify-locals");
add("vacuum");
add("reorder-locals");
add("coalesce-locals");
add("reorder-locals");
if (optimizeLevel >= 3 || shrinkLevel >= 1) { // differs
add("merge-locals");
}
add("vacuum");
if (optimizeLevel >= 3 || shrinkLevel >= 1) {
add("code-folding");
}
if (optimizeLevel >= 2 || shrinkLevel >= 1) { // differs
add("simplify-globals-optimizing");
}
add("merge-blocks");
add("remove-unused-brs");
add("remove-unused-names");
add("merge-blocks");
// make this later & move to (2)
// if (optimizeLevel >= 3 || shrinkLevel >= 2) {
// add("precompute-propagate");
// } else {
// add("precompute");
// }
if (optimizeLevel >= 3) {
add("optimize-instructions");
}
if (optimizeLevel >= 2 || shrinkLevel >= 1) {
add("rse");
}
add("vacuum");
// PassRunner::addDefaultGlobalOptimizationPostPasses
if (optimizeLevel >= 2 || shrinkLevel >= 1) {
add("simplify-globals-optimizing"); // differs
add("dae-optimizing");
}
if (optimizeLevel >= 2 || shrinkLevel >= 2) {
add("inlining-optimizing");
}
if (module.getLowMemoryUnused()) {
if (optimizeLevel >= 3 || shrinkLevel >= 1) {
add("optimize-added-constants-propagate");
} else {
add("optimize-added-constants");
}
}
// "duplicate-function-elimination" will better done later
// add("duplicate-function-elimination");
add("duplicate-import-elimination");
if (optimizeLevel >= 2 || shrinkLevel >= 2) {
add("simplify-globals-optimizing");
} else {
add("simplify-globals");
add("vacuum"); // differs
}
// moved from (2)
// it works better after globals optimizations like simplify-globals, inlining-optimizing and etc
if (optimizeLevel >= 2 || shrinkLevel >= 1) { // differs
add("precompute-propagate");
} else {
add("precompute");
}
// replace indirect calls with direct, reduce arity and
// inline this calls if possible
add("directize"); // differs
add("dae-optimizing"); // differs
add("inlining-optimizing"); // differs
// ARC finalization should be done exactly after inlining for better release/retain reduction
if (hasARC) { // differs
add("post-assemblyscript-finalize");
}
if (optimizeLevel >= 2 || shrinkLevel >= 1) { // differs
add("rse");
// move some code after early return which potentially could reduce computations
// do this after CFG cleanup (originally it was done before)
// moved from (1)
add("code-pushing");
if (optimizeLevel >= 3) {
// this quite expensive so do this only for highest opt level
add("simplify-globals");
add("vacuum");
// replace indirect calls with direct and inline if possible again.
add("inlining-optimizing");
add("directize");
add("dae-optimizing");
add("precompute-propagate");
add("vacuum");
add("merge-locals");
add("coalesce-locals");
add("simplify-locals-nostructure");
add("vacuum");
add("inlining-optimizing");
add("precompute-propagate");
}
add("remove-unused-brs");
add("remove-unused-names");
add("vacuum");
add("optimize-instructions");
add("simplify-globals-optimizing");
}
// remove unused elements of table and pack / reduce memory
add("duplicate-function-elimination"); // differs
add("remove-unused-nonfunction-module-elements"); // differs
add("memory-packing");
add("remove-unused-module-elements"); // differs
// It seems stack-ir unuseful for our needs.
// if (optimizeLevel >= 3 || shrinkLevel >= 1) { // differs. was optimizeLevel >= 2
// add("generate-stack-ir");
// add("optimize-stack-ir");
// }
}

// Append additional passes if requested and execute
module.runPasses(passes.concat(runPasses));
}

stats.optimizeTime += measure(() => {
stats.optimizeCount++;
doOptimize();
if (args.converge) {
module.optimize(optimizeLevel, shrinkLevel, debugInfo, usesARC);
module.runPasses(runPasses);
if (converge) {
let last = module.toBinary();
do {
stats.optimizeCount++;
doOptimize();
module.optimize(optimizeLevel, shrinkLevel, debugInfo, usesARC);
module.runPasses(runPasses);
let next = module.toBinary();
if (next.output.length >= last.output.length) {
if (next.output.length > last.output.length) {
Expand Down
Loading