[Bug] [opt] Fix compilation crash when there is a container statement after an unconditional continue

taichi/ir/control_flow_graph.h

@@ -65,6 +65,7 @@ class ControlFlowGraph {
   void reaching_definition_analysis(bool after_lower_access);
 
   void simplify_graph();
+  // This pass cannot eliminate container statements properly for now.
   bool unreachable_code_elimination();
   bool store_to_load_forwarding(bool after_lower_access);
 };

taichi/ir/ir.cpp

@@ -665,7 +665,7 @@ void Block::insert(std::unique_ptr<Stmt> &&stmt, int location) {
 
 void Block::insert(VecStatement &&stmt, int location) {
   if (location == -1) {
-    location = (int)statements.size() - 1;
+    location = (int)statements.size();
   }
   for (int i = 0; i < stmt.size(); i++) {
     insert(std::move(stmt[i]), location + i);

taichi/transforms/cfg_optimization.cpp

@@ -12,8 +12,6 @@ void cfg_optimization(IRNode *root, bool after_lower_access) {
   while (true) {
     bool modified = false;
     cfg->simplify_graph();
-    if (cfg->unreachable_code_elimination())
-      modified = true;
     if (cfg->store_to_load_forwarding(after_lower_access))
       modified = true;
     if (!modified)

taichi/transforms/ir_printer.cpp

@@ -250,7 +250,8 @@ class IRPrinter : public IRVisitor {
 
   void visit(ContinueStmt *stmt) override {
     if (stmt->scope) {
-      print("{} continue (scope={})", stmt->name(), stmt->name());
+      print("{} continue (scope={})", stmt->name(),
+            stmt->scope->name());
     } else {
       print("{} continue", stmt->name());
     }

taichi/transforms/simplify.cpp

@@ -958,7 +958,13 @@ class BasicBlockSimplify : public IRVisitor {
   }
 
   void visit(ContinueStmt *stmt) override {
-    return;
+    if (stmt != stmt->parent->back()) {
+      const int location = stmt->parent->locate(stmt);
+      while (location + 1 < (int)stmt->parent->size()) {
+        stmt->parent->erase(location + 1);
+      }
+      throw IRModified();
+    }
   }
 
   static bool is_global_write(Stmt *stmt) {
@@ -1201,6 +1207,8 @@ bool simplify(IRNode *root, Kernel *kernel) {
     else
       break;
   }
+  if (modified)
+    fix_block_parents(root);
   return modified;
 }
 

tests/python/test_continue.py

@@ -82,3 +82,67 @@ def run():
     xs = x.to_numpy()
     for i in range(n):
         assert xs[i] == 0
+
+
+@ti.all_archs
+def test_kernel_continue_in_nested_if():
+    x = ti.var(ti.i32, shape=n)
+
+    @ti.kernel
+    def run(a: ti.i32):
+        for i in range(1):
+            if a:
+                if a:
+                    continue
+            if a:
+                if a:
+                    continue
+            x[i] = i
+
+    x[0] = 1
+    run(1)
+    assert x[0] == 1
+    run(0)
+    assert x[0] == 0
+
+
+@ti.all_archs
+def test_kernel_continue_in_nested_if_2():
+    x = ti.var(ti.i32, shape=n)
+
+    @ti.kernel
+    def run(a: ti.i32):
+        for i in range(1):
+            if a:
+                if a:
+                    continue
+            if a:
+                continue
+            x[i] = i
+
+    x[0] = 1
+    run(1)
+    assert x[0] == 1
+    run(0)
+    assert x[0] == 0
+
+
+@ti.all_archs
+def test_kernel_continue_in_nested_if_3():
+    x = ti.var(ti.i32, shape=n)
+
+    @ti.kernel
+    def run(a: ti.i32):
+        for i in range(1):
+            if a:
+                continue
+            if a:
+                if a:
+                    continue
+            x[i] = i
+
+    x[0] = 1
+    run(1)
+    assert x[0] == 1
+    run(0)
+    assert x[0] == 0