update

paddle/fluid/lite/core/CMakeLists.txt

@@ -8,3 +8,4 @@ cc_library(scope_lite SRCS scope.cc)
 
 cc_test(test_scope_lite SRCS scope_test.cc DEPS scope_lite)
 cc_test(test_kernel_lite SRCS kernel_test.cc DEPS target_wrapper_x86)
+cc_test(test_op_lite SRCS op_lite_test.cc DEPS op_lite)

paddle/fluid/lite/core/kernel.h

@@ -26,6 +26,8 @@
 namespace paddle {
 namespace lite {
 
+// An base with virtual functions to unify all the kernel implementation on
+// different targets.
 class KernelBase {
  public:
   virtual void Run() = 0;
@@ -45,8 +47,12 @@ class KernelBase {
     return param_.get<Param>();
   }
 
- protected:
+  virtual TargetType target() const = 0;
+  virtual PrecisionType precision() const = 0;
+
   virtual ~KernelBase() = default;
+
+ protected:
   core::any_context_t context_;
   mutable operators::param_t param_;
 };

paddle/fluid/lite/core/kernel_test.cc

@@ -28,6 +28,10 @@ class SomeKernel : public OpKernel<TARGET(kHost), PRECISION(kFloat)> {
     LOG(INFO) << param<operators::FcParam>().in_num_col_dims;
     test_code = param<operators::FcParam>().in_num_col_dims;
   }
+
+  TargetType target() const override { return TARGET(kHost); }
+  PrecisionType precision() const override { return PRECISION(kFloat); }
+
 };
 
 TEST(Kernel, test) {

paddle/fluid/lite/core/memory.h

@@ -28,9 +28,6 @@ static void* TargetMalloc(TargetType target, size_t size) {
     case static_cast<int>(TargetType::kCUDA):
       data = TargetWrapper<TARGET(kCUDA)>::Malloc(size);
       break;
-    case static_cast<int>(TargetType::kARM):
-      data = TargetWrapper<TARGET(kARM)>::Malloc(size);
-      break;
     case static_cast<int>(TargetType::kHost):
       data = TargetWrapper<TARGET(kHost)>::Malloc(size);
       break;
@@ -48,9 +45,6 @@ static void TargetFree(TargetType target, void* data) {
     case static_cast<int>(TargetType::kCUDA):
       TargetWrapper<TARGET(kX86)>::Free(data);
       break;
-    case static_cast<int>(TargetType::kARM):
-      TargetWrapper<TARGET(kX86)>::Free(data);
-      break;
     default:
       LOG(FATAL) << "Unknown type";
   }

paddle/fluid/lite/core/op_lite.cc

@@ -12,4 +12,36 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "paddle/fluid/lite/core/op_lite.h"
 #include "op_lite.h"
+#include "paddle/fluid/lite/core/op_registry.h"
+
+namespace paddle {
+namespace lite {
+
+std::vector<std::unique_ptr<KernelBase>> OpLite::CreateKernels(
+    const std::vector<OpLite::Place> &places) {
+  std::vector<std::unique_ptr<KernelBase>> kernels;
+  CHECK(!op_type_.empty()) << "op_type_ should be set first";
+
+  for (auto place : places) {
+    kernels.emplace_back(KernelRegistry::Global().Create(op_type_, place.target,
+                                                         place.precision));
+  }
+
+  return kernels;
+}
+
+void OpLite::PickKernel(const std::vector<OpLite::Place> &valid_places,
+                        OpLite::KernelStrategy kernel_strategy) {
+  switch (kernel_strategy) {
+    case KernelStrategy::kStatic:
+      StaticPickKernel(valid_places);
+      break;
+    default:
+      LOG(FATAL) << "unsupported kernel strategy";
+  }
+}
+
+}  // namespace lite
+}  // namespace paddle

paddle/fluid/lite/core/op_lite.h

@@ -70,7 +70,15 @@ class OpLite : public Registry {
   // Inference the outputs' shape.
   virtual bool InferShape() const { return true; }
   // Run this operator.
-  virtual bool Run() = 0;
+  virtual bool Run() {
+    CHECK(kernel_);
+    SyncInputEvents();
+
+    kernel_->Run();
+
+    RecordOutputEvents();
+    return true;
+  }
   // Build the operator, attach it with the runtime environment.
   virtual bool Build(const framework::OpDesc &opdesc, lite::Scope *scope) = 0;
   // Human-readable information.
@@ -79,21 +87,31 @@ class OpLite : public Registry {
   const Place &kernel_place() const { return kernel_place_; }
 
  protected:
+  void PickKernel(const std::vector<Place> &valid_places,
+                  KernelStrategy kernel_strategy = KernelStrategy::kStatic);
+
   // Specify the kernel to run by default. This will specify the value of
   // `kernel_place_`.
   virtual void StaticPickKernel(const std::vector<Place> &valid_targets) = 0;
 
-  void PickKernel(const std::vector<Place> &valid_places,
-                  KernelStrategy kernel_strategy = KernelStrategy::kStatic);
+  // Wait until all the inputs' events are ready.
+  void SyncInputEvents() {}
+
+  // Record the output events, and that will tell all the dependent operators
+  // some inputs are ready.
+  void RecordOutputEvents() {}
 
   // Create all the kernels for the valid targets.
-  void CreateKernels();
+  std::vector<std::unique_ptr<KernelBase>> CreateKernels(
+      const std::vector<Place> &places);
 
   virtual ~OpLite() = default;
 
  protected:
   std::unique_ptr<OpContext> op_context_;
   Place kernel_place_;
+  std::unique_ptr<KernelBase> kernel_;
+  std::string op_type_;
 };
 
 }  // namespace lite

paddle/fluid/lite/core/op_lite_test.cc

+#include <gtest/gtest.h>
+#include "paddle/fluid/lite/core/op_lite.h"
+
+namespace paddle {
+namespace lite {
+
+TEST(OpLite, test) {
+
+
+}
+
+}  // namespace lite
+}  // namespace paddle

paddle/fluid/lite/core/op_registry.h

@@ -59,7 +59,6 @@ class KernelRegistry final {
               KernelRegistryForTarget<TARGET(kCUDA), PRECISION(kInt8)> *,   //
               KernelRegistryForTarget<TARGET(kX86), PRECISION(kFloat)> *,   //
               KernelRegistryForTarget<TARGET(kX86), PRECISION(kInt8)> *,    //
-              KernelRegistryForTarget<TARGET(kARM), PRECISION(kFloat)> *,   //
               KernelRegistryForTarget<TARGET(kHost), PRECISION(kFloat)> *   //
               >;
 
@@ -77,7 +76,6 @@ registries_[0].set<kernel_target_t *>(
                *>(&KernelRegistryForTarget<TARGET(target__),                 \
                                            PRECISION(precision__)>::Global());
     // Currently, just register 2 kernel targets.
-    INIT_FOR(kARM, kFloat);
     INIT_FOR(kHost, kFloat);
 #undef INIT_FOR
   }
@@ -97,6 +95,42 @@ registries_[0].set<kernel_target_t *>(
         ->Register(name, std::move(creator));
   }
 
+  template <TargetType Target, PrecisionType Precision>
+  std::unique_ptr<KernelBase> Create(const std::string &op_type) {
+    using kernel_registor_t = KernelRegistryForTarget<Target, Precision>;
+    return registries_[GetKernelOffset<Target, Precision>()]
+        .template get<kernel_registor_t *>()
+        ->Create(op_type);
+  }
+
+  std::unique_ptr<KernelBase> Create(const std::string &op_type,
+                                     TargetType target,
+                                     PrecisionType precision) {
+#define CREATE_KERNEL(target__)                                    \
+  switch (precision) {                                             \
+    case PRECISION(kFloat):                                        \
+      return Create<TARGET(target__), PRECISION(kFloat)>(op_type); \
+    default:                                                       \
+      CHECK(false) << "not supported kernel place yet";            \
+  }
+
+    switch (target) {
+      case TARGET(kHost): {
+        CREATE_KERNEL(kHost);
+      } break;
+      case TARGET(kX86): {
+        CREATE_KERNEL(kX86);
+      } break;
+      case TARGET(kCUDA): {
+        CREATE_KERNEL(kCUDA);
+      } break;
+      default:
+        CHECK(false) << "not supported kernel place";
+    }
+
+#undef CREATE_KERNEL
+  }
+
   // Get a kernel registry offset in all the registries.
   template <TargetType Target, PrecisionType Precision>
   static constexpr int GetKernelOffset() {

paddle/fluid/lite/core/target_wrapper.h

@@ -18,7 +18,7 @@
 namespace paddle {
 namespace lite {
 
-enum class TargetType { kHost = 0, kX86, kCUDA, kARM, kLastAsPlaceHolder };
+enum class TargetType { kHost = 0, kX86, kCUDA, kLastAsPlaceHolder };
 // Some helper macro to get a specific TargetType.
 #define TARGET(item__) paddle::lite::TargetType::item__
 #define TARGET_VAL(item__) static_cast<int>(TARGET(item__))

paddle/fluid/lite/core/types.h

@@ -22,8 +22,7 @@ namespace lite {
 namespace core {
 
 using any_context_t = variant<Context<TARGET(kX86)>,  //
-                              Context<TARGET(kCUDA)>,  //
-                              Context<TARGET(kARM)>    //
+                              Context<TARGET(kCUDA)>  //
                               >;
 
 }  // namespace core