Add a simple tuple optimization pass (#5937)

In some cases tuples are obviously not needed, such as when they are only used
in local operations and make/extract. Such tuples are not used as return values or
in control flow structures, so we might as well lower them to individual locals per
lane, which other passes can optimize a lot better.

I believe LLVM does the same with its own tuples: it lowers them as much as
possible, leaving only necessary ones.

Fixes #5923

1 files changed, 9 insertions, 0 deletions

diff --git a/src/passes/pass.cpp b/src/passes/pass.cpp
index 0d47cd78e..15e8c1f26 100644
--- a/src/passes/pass.cpp
+++ b/src/passes/pass.cpp
@@ -480,6 +480,9 @@ void PassRegistry::registerPasses() {
   registerPass("trap-mode-js",
                "replace trapping operations with js semantics",
                createTrapModeJS);
+  registerPass("tuple-optimization",
+               "optimize trivial tuples away",
+               createTupleOptimizationPass);
   registerPass("type-merging",
                "merge types to their supertypes where possible",
                createTypeMergingPass);
@@ -558,6 +561,12 @@ void PassRunner::addDefaultFunctionOptimizationPasses() {
   if (options.optimizeLevel >= 2 || options.shrinkLevel >= 2) {
     addIfNoDWARFIssues("code-pushing");
   }
+  if (wasm->features.hasMultivalue()) {
+    // Optimize tuples before local opts (as splitting tuples can help local
+    // opts), but also not too early, as we want to be after
+    // optimize-instructions at least (which can remove tuple-related things).
+    addIfNoDWARFIssues("tuple-optimization");
+  }
   // don't create if/block return values yet, as coalesce can remove copies that
   // that could inhibit
   addIfNoDWARFIssues("simplify-locals-nostructure");