autodrop must be run before we optimize in asm2wasm, as otherwise its input is not yet valid

then after finalizeCalls, we must autodrop again to drop things that finalizeCalls changed

7 files changed, 114 insertions, 80 deletions

diff --git a/src/asm2wasm.h b/src/asm2wasm.h
index 10d065400..7cc671597 100644
--- a/src/asm2wasm.h
+++ b/src/asm2wasm.h
@@ -523,7 +523,10 @@ void Asm2WasmBuilder::processAsm(Ref ast) {
     for (unsigned i = 1; i < body->size(); i++) {
       if (body[i][0] == DEFUN) numFunctions++;
     }
-    optimizingBuilder = make_unique<OptimizingIncrementalModuleBuilder>(&wasm, numFunctions);
+    optimizingBuilder = make_unique<OptimizingIncrementalModuleBuilder>(&wasm, numFunctions, [&](PassRunner& passRunner) {
+      // run autodrop first, before optimizations
+      passRunner.add<AutoDrop>();
+    });
   }
 
   // first pass - do almost everything, but function imports and indirect calls
@@ -786,7 +789,7 @@ void Asm2WasmBuilder::processAsm(Ref ast) {
   };
   PassRunner passRunner(&wasm);
   passRunner.add<FinalizeCalls>(this);
-  passRunner.add<AutoDrop>();
+  passRunner.add<AutoDrop>(); // FinalizeCalls may cause us to require additional drops
   if (optimize) passRunner.add("vacuum"); // autodrop can add some garbage
   passRunner.run();
 
diff --git a/src/passes/pass.cpp b/src/passes/pass.cpp
index 1e7e85ccb..437b023bc 100644
--- a/src/passes/pass.cpp
+++ b/src/passes/pass.cpp
@@ -209,6 +209,9 @@ void PassRunner::run() {
 }
 
 void PassRunner::runFunction(Function* func) {
+  if (debug) {
+    std::cerr << "[PassRunner] running passes on function " << func->name << std::endl;
+  }
   for (auto* pass : passes) {
     runPassOnFunction(pass, func);
   }
diff --git a/src/wasm-module-building.h b/src/wasm-module-building.h
index ead074991..43cc493d1 100644
--- a/src/wasm-module-building.h
+++ b/src/wasm-module-building.h
@@ -73,6 +73,7 @@ static std::mutex debug;
 class OptimizingIncrementalModuleBuilder {
   Module* wasm;
   uint32_t numFunctions;
+  std::function<void (PassRunner&)> addPrePasses;
   Function* endMarker;
   std::atomic<Function*>* list;
   uint32_t nextFunction; // only used on main thread
@@ -86,8 +87,8 @@ class OptimizingIncrementalModuleBuilder {
 public:
   // numFunctions must be equal to the number of functions allocated, or higher. Knowing
   // this bounds helps avoid locking.
-  OptimizingIncrementalModuleBuilder(Module* wasm, Index numFunctions)
-      : wasm(wasm), numFunctions(numFunctions), endMarker(nullptr), list(nullptr), nextFunction(0),
+  OptimizingIncrementalModuleBuilder(Module* wasm, Index numFunctions, std::function<void (PassRunner&)> addPrePasses)
+      : wasm(wasm), numFunctions(numFunctions), addPrePasses(addPrePasses), endMarker(nullptr), list(nullptr), nextFunction(0),
         numWorkers(0), liveWorkers(0), activeWorkers(0), availableFuncs(0), finishedFuncs(0),
         finishing(false) {
     if (numFunctions == 0) {
@@ -201,6 +202,7 @@ private:
 
   void optimizeFunction(Function* func) {
     PassRunner passRunner(wasm);
+    addPrePasses(passRunner);
     passRunner.addDefaultFunctionOptimizationPasses();
     passRunner.runFunction(func);
   }
diff --git a/test/emcc_hello_world.fromasm b/test/emcc_hello_world.fromasm
index c0b245dd9..ab579533e 100644
--- a/test/emcc_hello_world.fromasm
+++ b/test/emcc_hello_world.fromasm
@@ -2729,21 +2729,23 @@
               )
             )
           )
-          (call_indirect $FUNCSIG$iiii
-            (get_local $0)
-            (i32.sub
-              (get_local $1)
-              (get_local $2)
-            )
-            (i32.const 1)
-            (i32.add
-              (i32.and
-                (i32.load offset=40
-                  (get_local $0)
+          (drop
+            (call_indirect $FUNCSIG$iiii
+              (get_local $0)
+              (i32.sub
+                (get_local $1)
+                (get_local $2)
+              )
+              (i32.const 1)
+              (i32.add
+                (i32.and
+                  (i32.load offset=40
+                    (get_local $0)
+                  )
+                  (i32.const 7)
                 )
-                (i32.const 7)
+                (i32.const 2)
               )
-              (i32.const 2)
             )
           )
         )
@@ -7119,13 +7121,15 @@
                                   )
                                   (i32.const 0)
                                 )
-                                (call $___fwritex
-                                  (get_local $5)
-                                  (i32.sub
-                                    (get_local $75)
+                                (drop
+                                  (call $___fwritex
                                     (get_local $5)
+                                    (i32.sub
+                                      (get_local $75)
+                                      (get_local $5)
+                                    )
+                                    (get_local $0)
                                   )
-                                  (get_local $0)
                                 )
                               )
                               (if
@@ -7234,17 +7238,19 @@
                                     )
                                     (i32.const 0)
                                   )
-                                  (call $___fwritex
-                                    (get_local $1)
-                                    (select
-                                      (i32.const 9)
-                                      (get_local $15)
-                                      (i32.gt_s
-                                        (get_local $15)
+                                  (drop
+                                    (call $___fwritex
+                                      (get_local $1)
+                                      (select
                                         (i32.const 9)
+                                        (get_local $15)
+                                        (i32.gt_s
+                                          (get_local $15)
+                                          (i32.const 9)
+                                        )
                                       )
+                                      (get_local $0)
                                     )
-                                    (get_local $0)
                                   )
                                 )
                                 (set_local $1
@@ -7461,17 +7467,19 @@
                                       )
                                       (i32.const 0)
                                     )
-                                    (call $___fwritex
-                                      (get_local $1)
-                                      (select
-                                        (get_local $8)
-                                        (get_local $15)
-                                        (i32.gt_s
-                                          (get_local $15)
+                                    (drop
+                                      (call $___fwritex
+                                        (get_local $1)
+                                        (select
                                           (get_local $8)
+                                          (get_local $15)
+                                          (i32.gt_s
+                                            (get_local $15)
+                                            (get_local $8)
+                                          )
                                         )
+                                        (get_local $0)
                                       )
-                                      (get_local $0)
                                     )
                                   )
                                   (if
diff --git a/test/emcc_hello_world.fromasm.imprecise b/test/emcc_hello_world.fromasm.imprecise
index 07ee8d33f..32e34c0e9 100644
--- a/test/emcc_hello_world.fromasm.imprecise
+++ b/test/emcc_hello_world.fromasm.imprecise
@@ -2723,21 +2723,23 @@
               )
             )
           )
-          (call_indirect $FUNCSIG$iiii
-            (get_local $0)
-            (i32.sub
-              (get_local $1)
-              (get_local $2)
-            )
-            (i32.const 1)
-            (i32.add
-              (i32.and
-                (i32.load offset=40
-                  (get_local $0)
+          (drop
+            (call_indirect $FUNCSIG$iiii
+              (get_local $0)
+              (i32.sub
+                (get_local $1)
+                (get_local $2)
+              )
+              (i32.const 1)
+              (i32.add
+                (i32.and
+                  (i32.load offset=40
+                    (get_local $0)
+                  )
+                  (i32.const 7)
                 )
-                (i32.const 7)
+                (i32.const 2)
               )
-              (i32.const 2)
             )
           )
         )
@@ -7113,13 +7115,15 @@
                                   )
                                   (i32.const 0)
                                 )
-                                (call $___fwritex
-                                  (get_local $5)
-                                  (i32.sub
-                                    (get_local $75)
+                                (drop
+                                  (call $___fwritex
                                     (get_local $5)
+                                    (i32.sub
+                                      (get_local $75)
+                                      (get_local $5)
+                                    )
+                                    (get_local $0)
                                   )
-                                  (get_local $0)
                                 )
                               )
                               (if
@@ -7228,17 +7232,19 @@
                                     )
                                     (i32.const 0)
                                   )
-                                  (call $___fwritex
-                                    (get_local $1)
-                                    (select
-                                      (i32.const 9)
-                                      (get_local $15)
-                                      (i32.gt_s
-                                        (get_local $15)
+                                  (drop
+                                    (call $___fwritex
+                                      (get_local $1)
+                                      (select
                                         (i32.const 9)
+                                        (get_local $15)
+                                        (i32.gt_s
+                                          (get_local $15)
+                                          (i32.const 9)
+                                        )
                                       )
+                                      (get_local $0)
                                     )
-                                    (get_local $0)
                                   )
                                 )
                                 (set_local $1
@@ -7455,17 +7461,19 @@
                                       )
                                       (i32.const 0)
                                     )
-                                    (call $___fwritex
-                                      (get_local $1)
-                                      (select
-                                        (get_local $8)
-                                        (get_local $15)
-                                        (i32.gt_s
-                                          (get_local $15)
+                                    (drop
+                                      (call $___fwritex
+                                        (get_local $1)
+                                        (select
                                           (get_local $8)
+                                          (get_local $15)
+                                          (i32.gt_s
+                                            (get_local $15)
+                                            (get_local $8)
+                                          )
                                         )
+                                        (get_local $0)
                                       )
-                                      (get_local $0)
                                     )
                                   )
                                   (if
diff --git a/test/unit.fromasm b/test/unit.fromasm
index bc5281ed5..f48dfcdb2 100644
--- a/test/unit.fromasm
+++ b/test/unit.fromasm
@@ -491,10 +491,15 @@
     )
   )
   (func $smallIf
-    (if
-      (i32.const 2)
-      (call $lb
-        (i32.const 3)
+    (block $do-once$0
+      (drop
+        (if
+          (i32.const 2)
+          (call $lb
+            (i32.const 3)
+          )
+          (br $do-once$0)
+        )
       )
     )
   )
diff --git a/test/unit.fromasm.imprecise b/test/unit.fromasm.imprecise
index 945d6cc8e..06ab40915 100644
--- a/test/unit.fromasm.imprecise
+++ b/test/unit.fromasm.imprecise
@@ -473,10 +473,15 @@
     )
   )
   (func $smallIf
-    (if
-      (i32.const 2)
-      (call $lb
-        (i32.const 3)
+    (block $do-once$0
+      (drop
+        (if
+          (i32.const 2)
+          (call $lb
+            (i32.const 3)
+          )
+          (br $do-once$0)
+        )
       )
     )
   )