[Wasm] Fix emitting setup code too late.

Setup code should *always* be emitted before emitting of the own computation
code, e.g. the setup code must be emitted before compiling a data layout since
this compilation may temporarily introduce boolean variables and emit code
using them into the returned blocks, however, these blocks may be placed
between setup and teardown code which may represent the lifespan of another
boolean variable which then overrides the value of the other one (our DSL is
able to use the same bit again since the `Variable` object is already distroyed
and thus its bit is free to use).
We had this issue previously and thus introduced the `referenced_bits_` field
for `PrimitiveExpr`s, however, beeing able to return blocks of Wasm code and
place them somewhere else circumvents our former solution. Maybe we can extend
the idea of memorizing referenced bits to blocks as well until they are
attached or destroyed.

src/backend/WasmOperator.cpp

@@ -421,18 +421,22 @@ void Scan::execute(const Match<Scan> &M, setup_t setup, pipeline_t pipeline, tea
     oss << table.name << "_mem";
     Ptr<void> base_address = Module::Get().get_global<void*>(oss.str().c_str());
 
+    /*----- Emit setup code *before* compiling data layout to not overwrite its temporary boolean variables. -----*/
+    setup();
+
     /*----- Compile data layout to generate sequential load from table. -----*/
     auto [inits, loads, jumps] = compile_load_sequential(schema, base_address, table.layout(),
                                                          table.schema(M.scan.alias()), tuple_id);
 
     /*----- Generate the loop for the actual scan, with the pipeline emitted into the loop body. -----*/
-    setup();
     inits.attach_to_current();
     WHILE (tuple_id < num_rows) {
         loads.attach_to_current();
         pipeline();
         jumps.attach_to_current();
     }
+
+    /*----- Emit teardown code. -----*/
     teardown();
 }
 
@@ -1867,6 +1871,9 @@ void OrderedGrouping::execute(const Match<OrderedGrouping> &M, setup_t setup, pi
         auto S = CodeGenContext::Get().scoped_environment(); // create scoped environment for this function
         auto &env = CodeGenContext::Get().env();
 
+        /*----- Emit setup code *before* possibly introducing temporary boolean variables to not overwrite them. -----*/
+        setup();
+
         /*----- Add computed group tuple to current environment. ----*/
         for (auto &e : M.grouping.schema().deduplicate()) {
             try {
@@ -1913,8 +1920,9 @@ void OrderedGrouping::execute(const Match<OrderedGrouping> &M, setup_t setup, pi
         }
 
         /*----- Resume pipeline. -----*/
-        setup();
         pipeline();
+
+        /*----- Emit teardown code. -----*/
         teardown();
     }
 }
@@ -2427,6 +2435,9 @@ void Aggregation::execute(const Match<Aggregation> &M, setup_t setup, pipeline_t
     }
     aggregation_child_pipeline(); // call child function
 
+    /*----- Emit setup code *before* possibly introducing temporary boolean variables to not overwrite them. -----*/
+    setup();
+
     /*----- Add computed aggregates tuple to current environment. ----*/
     auto &env = CodeGenContext::Get().env();
     for (auto &e : M.aggregation.schema().deduplicate()) {
@@ -2462,8 +2473,9 @@ void Aggregation::execute(const Match<Aggregation> &M, setup_t setup, pipeline_t
     }
 
     /*----- Resume pipeline. -----*/
-    setup();
     pipeline();
+
+    /*----- Emit teardown code. -----*/
     teardown();
 }
 

src/backend/WasmUtil.cpp

@@ -2136,19 +2136,23 @@ void Buffer<IsGlobal>::resume_pipeline(param_t tuple_schema_)
             Ptr<void> base_address = PARAMETER(0);
             U32 size = PARAMETER(1);
 
+            /*----- Emit setup code *before* compiling data layout to not overwrite its temporary boolean variables. -*/
+            setup_();
+
             /*----- Compile data layout to generate sequential load from buffer. -----*/
             Var<U32> load_tuple_id; // default initialized to 0
             auto [load_inits, loads, load_jumps] =
                 compile_load_sequential(tuple_schema, base_address, layout_, schema_, load_tuple_id);
 
             /*----- Generate loop for loading entire buffer, with the pipeline emitted into the loop body. -----*/
-            setup_();
             load_inits.attach_to_current();
             WHILE (load_tuple_id < size) {
                 loads.attach_to_current();
                 pipeline_();
                 load_jumps.attach_to_current();
             }
+
+            /*----- Emit teardown code. -----*/
             teardown_();
         }
         resume_pipeline_ = std::move(resume_pipeline);
@@ -2187,19 +2191,23 @@ void Buffer<IsGlobal>::resume_pipeline_inline(param_t tuple_schema_) const
         pred = env.extract_predicate().is_true_and_not_null();
     U32 num_tuples = pred ? Select(*pred, size, 0U) : size;
 
+    /*----- Emit setup code *before* compiling data layout to not overwrite its temporary boolean variables. -----*/
+    setup_();
+
     /*----- Compile data layout to generate sequential load from buffer. -----*/
     Var<U32> load_tuple_id(0); // explicitly (re-)set tuple ID to 0
     auto [load_inits, loads, load_jumps] =
         compile_load_sequential(tuple_schema, base_address, layout_, schema_, load_tuple_id);
 
     /*----- Generate loop for loading entire buffer, with the pipeline emitted into the loop body. -----*/
-    setup_();
     load_inits.attach_to_current();
     WHILE (load_tuple_id < num_tuples) {
         loads.attach_to_current();
         pipeline_();
         load_jumps.attach_to_current();
     }
+
+    /*----- Emit teardown code. -----*/
     teardown_();
 }