merge error fix

lite/api/paddle_lite_factory_helper.h

@@ -19,7 +19,50 @@
 #pragma once
 
 // some platform-independent defintion
-#include "lite/utils/macros.h"
+// #include "lite/utils/macros.h"
+// just copy lite/utils/macros.h here temperarily by mlu, since it affects
+// compilation of demos
+// BEGIN by MLU
+#ifndef DISALLOW_COPY_AND_ASSIGN
+#define DISALLOW_COPY_AND_ASSIGN(class__) \
+  class__(const class__&) = delete;       \
+  class__& operator=(const class__&) = delete;
+#endif
+
+#define LITE_UNIMPLEMENTED CHECK(false) << "Not Implemented";
+
+#if defined(_WIN32)
+#define UNUSED
+#define __builtin_expect(EXP, C) (EXP)
+#else
+#define UNUSED __attribute__((unused))
+#endif
+
+/*
+#ifndef LIKELY
+#define LIKELY(x) __builtin_expect(!!(x), 1)
+#endif
+
+#ifndef UNLIKELY
+//#define UNLIKELY(x) __built_expect(!!(x), 0)
+#define UNLIKELY(x) (x)
+#endif
+ */
+
+#ifdef __CUDACC__
+#define HOSTDEVICE __host__ __device__
+#define DEVICE __device__
+#define HOST __host__
+#else
+#define HOSTDEVICE
+#define DEVICE
+#define HOST
+#endif
+
+#if defined(__FLT_MAX__)
+#define FLT_MAX __FLT_MAX__
+#endif  // __FLT_MAX__
+// END by MLU
 
 #define USE_LITE_OP(op_type__)       \
   extern int touch_op_##op_type__(); \

lite/core/mir/mlu_postprocess_pass.cc

@@ -570,11 +570,11 @@ void MLUPostprocessPass::Apply(const std::unique_ptr<SSAGraph>& graph) {
 // arg_in and arg_out are assumed to be NHWC which user should be aware of.
 // Thus here we change these args' layout to NHWC
 #ifdef LITE_WITH_MLU
-  if (lite::DeviceInfo::Global().InputLayout() == DATALAYOUT(kNHWC)) {
+  if (lite::TargetWrapperMlu::InputLayout() == DATALAYOUT(kNHWC)) {
     ModifyLayout(graph.get());
   }
 
-  if (lite::DeviceInfo::Global().UseFirstConv()) {
+  if (lite::TargetWrapperMlu::UseFirstConv()) {
     GatherAndModifyFirstConvNodes(graph.get());
   }
 #endif

lite/kernels/mlu/bridges/graph.h

@@ -120,7 +120,8 @@ class Graph {
       output_addrs_[i] = output_tensors_[i]->mlu_data();
     }
 
-#if PRINT_HW_TIME thread_local float hw_time;
+#if PRINT_HW_TIME
+    thread_local float hw_time;
     CNRT_CALL(cnrtPlaceNotifier(notifier_start_, que));
 #endif
     CNML_CALL(cnmlComputeFusionOpForward_V3(fusion_op_,

lite/kernels/mlu/subgraph_compute.h

@@ -84,32 +84,6 @@ class SubgraphEngine : public subgraph::Engine {
     return true;
   }
 
-  int Build() {
-    // In order to attach all of the ops of the block desc, we need to build
-    // the original program firstly.
-    BuildOriginProgram();
-    // Run InferShape() of all of ops, and convert Paddle ops to MLU IR graph
-    build_device_program_status_ = BuildDeviceProgram();
-    return build_device_program_status_;
-  }
-
-  int Launch() {
-    // Rebuild device program when the shapes of input tensors have been
-    // changed.
-    if (subgraph::CHECK_SUCCESS(build_device_program_status_) &&
-        subgraph::CHECK_REBUILD_WHEN_SHAPE_CHANGED(
-            build_device_program_status_) &&
-        InputShapeChanged()) {
-      Build();
-    }
-    if (subgraph::CHECK_FAILED(build_device_program_status_)) {
-      LaunchOriginProgram();
-    } else {
-      LaunchDeviceProgram();
-    }
-    return 0;
-  }
-
  protected:
   int BuildDeviceProgram() override {
     int status = 0;