removed common.h/cpp; added compile_config.h/cpp; cleanups

taichi/arch.cpp

@@ -48,4 +48,15 @@ bool arch_use_host_memory(Arch arch) {
   return arch_is_cpu(arch);
 }
 
+int default_simd_width(Arch arch) {
+  if (arch == Arch::x64) {
+    return 8;
+  } else if (arch == Arch::cuda) {
+    return 32;
+  } else {
+    TI_NOT_IMPLEMENTED;
+    return -1;
+  }
+}
+
 TLANG_NAMESPACE_END

taichi/arch.h

@@ -1,8 +1,7 @@
 #pragma once
 
 #include <string>
-#include <taichi/common/util.h>
-#include <taichi/common.h>
+#include "taichi/lang_util.h"
 
 TLANG_NAMESPACE_BEGIN
 
@@ -22,4 +21,6 @@ bool arch_is_gpu(Arch arch);
 
 bool arch_use_host_memory(Arch arch);
 
+int default_simd_width(Arch arch);
+
 TLANG_NAMESPACE_END

taichi/codegen/codegen_llvm_cuda.cpp

@@ -86,7 +86,6 @@ class CodeGenLLVMCUDA : public CodeGenLLVM {
     for (auto &task : offloaded_local) {
       task.cuda_func = cuda_module->lookup_function(task.name);
     }
-    auto prog = this->prog;
     return [offloaded_local, cuda_module](Context context) {
       for (auto task : offloaded_local) {
         TI_DEBUG("Launching kernel {}<<<{}, {}>>>", task.name, task.grid_dim,

taichi/codegen/codegen_metal.h

@@ -1,6 +1,5 @@
 #pragma once
 
-#include <taichi/common.h>
 #include <taichi/constants.h>
 #include <taichi/platform/metal/metal_data_types.h>
 #include <taichi/platform/metal/metal_kernel_util.h>

taichi/codegen/codegen_opengl.h

@@ -1,8 +1,7 @@
 #pragma once
 
-#include <taichi/common.h>
-#include <taichi/constants.h>
-#include <taichi/lang_util.h>
+#include "taichi/constants.h"
+#include "taichi/lang_util.h"
 
 #include <string>
 #include <unordered_map>

taichi/compile_config.cpp

@@ -0,0 +1,33 @@
+#include "taichi/compile_config.h"
+
+TLANG_NAMESPACE_BEGIN
+
+CompileConfig::CompileConfig() {
+  arch = Arch::x64;
+  simd_width = default_simd_width(arch);
+  external_optimization_level = 3;
+  print_ir = false;
+  print_accessor_ir = false;
+  use_llvm = true;
+  print_struct_llvm_ir = false;
+  print_kernel_llvm_ir = false;
+  print_kernel_llvm_ir_optimized = false;
+  demote_dense_struct_fors = true;
+  max_vector_width = 8;
+  debug = false;
+  lazy_compilation = true;
+  serial_schedule = false;
+  simplify_before_lower_access = true;
+  lower_access = true;
+  simplify_after_lower_access = true;
+  default_fp = DataType::f32;
+  default_ip = DataType::i32;
+  verbose_kernel_launches = false;
+  enable_profiler = false;
+  default_cpu_block_dim = 0;  // 0 = adaptive
+  default_gpu_block_dim = 64;
+  verbose = true;
+  fast_math = true;
+}
+
+TLANG_NAMESPACE_END

taichi/compile_config.h

@@ -0,0 +1,40 @@
+#include "taichi/lang_util.h"
+#include "taichi/arch.h"
+
+TLANG_NAMESPACE_BEGIN
+
+struct CompileConfig {
+  Arch arch;
+  bool debug;
+  int simd_width;
+  bool lazy_compilation;
+  int external_optimization_level;
+  int max_vector_width;
+  bool print_ir;
+  bool print_accessor_ir;
+  bool serial_schedule;
+  bool simplify_before_lower_access;
+  bool lower_access;
+  bool simplify_after_lower_access;
+  bool demote_dense_struct_fors;
+  bool use_llvm;
+  bool print_struct_llvm_ir;
+  bool print_kernel_llvm_ir;
+  bool print_kernel_llvm_ir_optimized;
+  bool verbose_kernel_launches;
+  bool enable_profiler;
+  bool verbose;
+  bool fast_math;
+  bool use_unified_memory;
+  DataType default_fp;
+  DataType default_ip;
+  std::string extra_flags;
+  int default_cpu_block_dim;
+  int default_gpu_block_dim;
+
+  CompileConfig();
+};
+
+extern CompileConfig default_compile_config;
+
+TLANG_NAMESPACE_END

taichi/constants.h

@@ -5,6 +5,7 @@
 constexpr int taichi_max_num_indices = 8;
 constexpr int taichi_max_num_args = 8;
 constexpr int taichi_max_num_snodes = 1024;
+constexpr int taichi_max_gpu_block_dim = 1024;
 constexpr std::size_t taichi_global_tmp_buffer_size = 1024 * 1024;
 constexpr int taichi_max_num_mem_requests = 1024 * 64;
 constexpr std::size_t taichi_page_size = 4096;

taichi/context.h

@@ -3,7 +3,7 @@
 #include "constants.h"
 
 #if defined(TI_RUNTIME_HOST)
-#include "common.h"
+#include "taichi/constants.h"
 
 namespace taichi::Tlang {
 using namespace taichi;

taichi/extension.h

@@ -1,8 +1,8 @@
 #pragma once
 
+#include "taichi/arch.h"
+
 #include <string>
-#include <taichi/common.h>
-#include <taichi/arch.h>
 
 TLANG_NAMESPACE_BEGIN
 

taichi/ir/ir.cpp

@@ -222,7 +222,7 @@ FrontendForStmt::FrontendForStmt(const ExprGroup &loop_var,
   if (cfg.arch == Arch::cuda) {
     vectorize = 1;
     parallelize = 1;
-    TI_ASSERT(block_dim <= max_gpu_block_dim);
+    TI_ASSERT(block_dim <= taichi_max_gpu_block_dim);
   } else {
     // cpu
     if (block_dim == 0)

taichi/ir/ir.h

@@ -4,11 +4,12 @@
 
 #include <atomic>
 #include <unordered_map>
-#include <taichi/common/util.h>
-#include <taichi/common/bit.h>
+#include "taichi/common/util.h"
+#include "taichi/common/bit.h"
 #include "taichi/lang_util.h"
-#include "snode.h"
-#include "expr.h"
+#include "taichi/ir/snode.h"
+#include "taichi/ir/expr.h"
+#include "taichi/compile_config.h"
 #include "taichi/llvm/llvm_fwd.h"
 
 TLANG_NAMESPACE_BEGIN

taichi/ir/snode.cpp

@@ -60,32 +60,6 @@ SNode &SNode::create_node(std::vector<Index> indices,
   return new_node;
 }
 
-void SNode::clear_data() {
-  if (clear_func == nullptr) {
-    if (clear_kernel == nullptr) {
-      clear_kernel = &kernel([&]() {
-        current_ast_builder().insert(Stmt::make<ClearAllStmt>(this, false));
-      });
-    }
-    (*(Kernel *)clear_kernel)();
-  } else {
-    clear_func(0);
-  }
-}
-
-void SNode::clear_data_and_deactivate() {
-  if (clear_func == nullptr) {
-    if (clear_and_deactivate_kernel == nullptr) {
-      clear_and_deactivate_kernel = &kernel([&]() {
-        current_ast_builder().insert(Stmt::make<ClearAllStmt>(this, true));
-      });
-    }
-    (*(Kernel *)clear_and_deactivate_kernel)();
-  } else {
-    clear_func(1);
-  }
-}
-
 void SNode::lazy_grad() {
   if (this->type == SNodeType::place)
     return;
@@ -223,8 +197,6 @@ SNode::SNode(int depth, SNodeType t) : depth(depth), type(t) {
   num_active_indices = 0;
   std::memset(taken_bits, 0, sizeof(taken_bits));
   std::memset(physical_index_position, -1, sizeof(physical_index_position));
-  access_func = nullptr;
-  stat_func = nullptr;
   parent = nullptr;
   _verbose = false;
   _multi_threaded = false;
@@ -234,7 +206,6 @@ SNode::SNode(int depth, SNodeType t) : depth(depth), type(t) {
   _morton = false;
   _bitmasked = false;
 
-  clear_func = nullptr;
   clear_kernel = nullptr;
   clear_and_deactivate_kernel = nullptr;