[RUNTIME][OPENCL] delay device check (#1657)

docs/deploy/aws_fpga.md

@@ -108,10 +108,10 @@ python run.py
 Synthesis
 ---------
 
-- Run synthesis with the following script. `XCL_EMULATION_MODE` must be set to 1 at this stage.
+- Run synthesis with the following script.
 
 ```bash
-export XCL_EMULATION_MODE=1
+unset XCL_EMULATION_MODE
 export XCL_TARGET=hw
 
 python build.py

src/runtime/opencl/opencl_module.cc

@@ -34,6 +34,7 @@ class OpenCLWrappedFunc {
   void operator()(TVMArgs args,
                   TVMRetValue* rv,
                   void** void_args) const {
+    CHECK(w_->context != nullptr) << "No OpenCL device";
     cl::OpenCLThreadEntry* t = w_->GetThreadEntry();
     // get the kernel from thread local kernel table.
     if (entry_.kernel_id >= t->kernel_table.size()) {
@@ -157,7 +158,6 @@ std::string OpenCLModuleNode::GetSource(const std::string& format) {
 void OpenCLModuleNode::Init() {
   workspace_ = GetGlobalWorkspace();
   workspace_->Init();
-  CHECK(workspace_->context != nullptr) << "No OpenCL device";
   device_built_flag_.resize(workspace_->devices.size(), false);
   // initialize the kernel id, need to lock global table.
   std::lock_guard<std::mutex> lock(workspace_->mu);