Introduce a temporary boolean constant for benchmarking

README.md

@@ -1,7 +1,7 @@
 <div align="center">
   <img width="500px" src="https://github.com/yuanming-hu/taichi/raw/master/misc/logo.png">
   <h3> <a href="https://taichi.readthedocs.io/en/latest/"> Docs </a> | <a href="https://taichi.readthedocs.io/en/latest/hello.html"> Tutorial </a> | <a href="https://github.com/yuanming-hu/difftaichi"> DiffTaichi </a> | <a href="https://github.com/yuanming-hu/taichi/tree/master/examples"> Examples </a> | <a href="https://taichi.readthedocs.io/en/latest/contributor_guide.html"> Contribute </a> | <a href="https://forum.taichi.graphics/"> Forum </a> </h3>
-</div>        
+</div>
 
 | **Documentations** | **Chat** | taichi-nightly | taichi-nightly-cuda-10-0 | taichi-nightly-cuda-10-1 |
 |:-----|:-----|:----|:----|:----|
@@ -46,7 +46,7 @@ python3 -m pip install taichi-nightly-cuda-10-1
       - Experimental support for automatically differentiating through conditional global load/stores (by **Yuanming Hu**)
    - **Bug fixes**
       - Fixed stack traceback printing on OS X (#610) (by **Yuanming Hu**)
-   - **CLI** 
+   - **CLI**
       - `ti format` now cover all files from upstream/master to the working tree (#629) (by **Ye Kuang**)
       - `ti test` now uses `argparse` for better customizability (#601) (by **彭于斌**)
    - **OpenGL backend**

examples/mgpcg_advanced.py

@@ -36,15 +36,14 @@ def __init__(self):
                              shape=(self.N_gui, self.N_gui))  # image buffer
 
         indices = ti.ijk if self.dim == 3 else ti.ij
-        self.grid = ti.root.pointer(indices,
-                                    [self.N_tot // 4]).dense(indices, 4).place(
-                                        self.x, self.p, self.Ap)
+        self.grid = ti.root.pointer(indices, [self.N_tot // 4]).dense(
+            indices, 4).place(self.x, self.p, self.Ap)
 
         for l in range(self.n_mg_levels):
             self.grid = ti.root.pointer(indices,
                                         [self.N_tot // (4 * 2**l)]).dense(
-                                            indices, 4).place(
-                                                self.r[l], self.z[l])
+                                            indices,
+                                            4).place(self.r[l], self.z[l])
 
         ti.root.place(self.alpha, self.beta, self.sum)
 
@@ -55,8 +54,8 @@ def init(self):
                     (self.N_ext, self.N_tot - self.N_ext), ) * self.dim)):
             self.r[0][I] = 1.0
             for k in ti.static(range(self.dim)):
-                self.r[0][I] *= ti.sin(2.0 * np.pi *
-                                       (I[k] - self.N_ext) * 2.0 / self.N_tot)
+                self.r[0][I] *= ti.sin(2.0 * np.pi * (I[k] - self.N_ext) *
+                                       2.0 / self.N_tot)
             self.z[0][I] = 0.0
             self.Ap[I] = 0.0
             self.p[I] = 0.0

taichi/program/compile_config.cpp

@@ -2,6 +2,8 @@
 
 TLANG_NAMESPACE_BEGIN
 
+const bool advanced_optimization = true;
+
 CompileConfig::CompileConfig() {
   arch = host_arch();
   simd_width = default_simd_width(arch);

taichi/program/compile_config.h

@@ -3,6 +3,10 @@
 
 TLANG_NAMESPACE_BEGIN
 
+// TODO(xumingkuan): Temporary variable for benchmarking.
+// TODO(xumingkuan): Will be removed in the future.
+extern const bool advanced_optimization;
+
 struct CompileConfig {
   Arch arch;
   bool debug;

taichi/transforms/simplify.cpp

@@ -667,8 +667,10 @@ class BasicBlockSimplify : public IRVisitor {
   }
 
   void visit(LinearizeStmt *stmt) override {
-    //    if (is_done(stmt))
-    //      return;
+    if (!advanced_optimization) {
+      if (is_done(stmt))
+        return;
+    }
 
     if (stmt->inputs.size() && stmt->inputs.back()->is<IntegerOffsetStmt>()) {
       auto previous_offset = stmt->inputs.back()->as<IntegerOffsetStmt>();
@@ -681,7 +683,25 @@ class BasicBlockSimplify : public IRVisitor {
       offset_stmt->as<IntegerOffsetStmt>()->input = stmt;
       throw IRModified();
     }
-    //    set_done(stmt);
+    if (!advanced_optimization) {
+      for (int i = 0; i < current_stmt_id; i++) {
+        auto &bstmt = block->statements[i];
+        if (stmt->ret_type == bstmt->ret_type) {
+          auto &bstmt_data = *bstmt;
+          if (typeid(bstmt_data) == typeid(*stmt)) {
+            auto bstmt_ = bstmt->as<LinearizeStmt>();
+            if (identical_vectors(bstmt_->inputs, stmt->inputs) &&
+                identical_vectors(bstmt_->strides, stmt->strides)) {
+              stmt->replace_with(bstmt.get());
+              stmt->parent->erase(current_stmt_id);
+              throw IRModified();
+            }
+          }
+        }
+      }
+      set_done(stmt);
+      return;
+    }
 
     // Lower into a series of adds and muls.
     auto sum = Stmt::make<ConstStmt>(LaneAttribute<TypedConstant>(0));
@@ -1014,7 +1034,8 @@ void simplify(IRNode *root) {
 
 void full_simplify(IRNode *root, const CompileConfig &config) {
   constant_fold(root);
-  alg_simp(root, config);
+  if (advanced_optimization)
+    alg_simp(root, config);
   die(root);
   simplify(root);
 }