sfqcyy
/
mindspore
forked from OSSInnovation/mindspore
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

!4316 fix debugger d-chip watchpoints and ensure GPU updated tensor values 

Merge pull request !4316 from john_tzanakakis/master_ms1_grpc
This commit is contained in:
mindspore-ci-bot 2020-08-12 09:47:29 +08:00 committed by Gitee
parent 196fdbe16e 3569513232
commit 85364d59c2
3 changed files with 25 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
mindspore/ccsrc/debug/debugger/debugger.cc
View File

@ -172,8 +172,13 @@ void Debugger::PostExecute() {
return;
}
if (debugger_enabled_ && !is_dataset_graph_) {
MS_LOG(INFO) << "Debugger suspend at end of step; number of steps executed: " << num_step_;
CommandLoop();
if (device_target_ != kGPUDevice) {
num_step_++;
MS_LOG(INFO) << "Debugger suspend at end of step; number of steps executed: " << num_step_;
SendWatchpointsAndSuspend(CheckWatchpoints());
} else {
CommandLoop();
}
}
}

2
mindspore/ccsrc/debug/tensor_load.h
View File

@ -46,7 +46,7 @@ class TensorLoader {
}
}
tensor_list.push_back(tensor);
tensor_list_map.insert({tensor->GetName(), tensor});
tensor_list_map[tensor->GetName()] = tensor; // use [] instead of insert to ensure latest value
auto node_name = tensor->GetName();
node_name = node_name.substr(0, node_name.find_first_of(":"));
node_tensor_map.insert({node_name, tensor});

36
mindspore/ccsrc/runtime/device/gpu/gpu_kernel_runtime.cc
View File

@ -241,25 +241,23 @@ void LoadKernelData(Debugger *debugger, const CNodePtr &kernel,
}
// get inputs
if (!dump_enabled) {
auto input_size = AnfAlgo::GetInputTensorNum(kernel);
for (size_t j = 0; j < input_size; ++j) {
auto input_kernel = kernel->input(j + 1);
std::string input_kernel_name = input_kernel->fullname_with_scope();
auto addr = kernel_inputs[j];
auto type = AnfAlgo::GetOutputInferDataType(input_kernel, PARAMETER_OUTPUT_INDEX);
auto format = kOpFormat_DEFAULT;
auto gpu_addr = std::make_unique<GPUDeviceAddress>(addr->addr, addr->size, format, type);
string input_tensor_name = input_kernel_name + ':' + "0";
std::vector<int> int_shapes;
auto shape = AnfAlgo::GetOutputDeviceShape(input_kernel, PARAMETER_OUTPUT_INDEX);
(void)std::transform(shape.begin(), shape.end(), std::back_inserter(int_shapes),
[](size_t inner_item) { return SizeToInt(inner_item); });
auto ret = gpu_addr->LoadMemToHost(input_tensor_name, exec_order, format, int_shapes, type, 0, debugger, false);
if (!ret) {
MS_LOG(ERROR) << "LoadMemToHost:"
<< ", tensor_name:" << input_tensor_name << ", host_format:" << format << ".!";
}
auto input_size = AnfAlgo::GetInputTensorNum(kernel);
for (size_t j = 0; j < input_size; ++j) {
auto input_kernel = kernel->input(j + 1);
std::string input_kernel_name = input_kernel->fullname_with_scope();
auto addr = kernel_inputs[j];
auto type = AnfAlgo::GetOutputInferDataType(input_kernel, PARAMETER_OUTPUT_INDEX);
auto format = kOpFormat_DEFAULT;
auto gpu_addr = std::make_unique<GPUDeviceAddress>(addr->addr, addr->size, format, type);
string input_tensor_name = input_kernel_name + ':' + "0";
std::vector<int> int_shapes;
auto shape = AnfAlgo::GetOutputDeviceShape(input_kernel, PARAMETER_OUTPUT_INDEX);
(void)std::transform(shape.begin(), shape.end(), std::back_inserter(int_shapes),
[](size_t inner_item) { return SizeToInt(inner_item); });
auto ret = gpu_addr->LoadMemToHost(input_tensor_name, exec_order, format, int_shapes, type, 0, debugger, false);
if (!ret) {
MS_LOG(ERROR) << "LoadMemToHost:"
<< ", tensor_name:" << input_tensor_name << ", host_format:" << format << ".!";
}
}