misc: strip trailing whitespace (#235)

Remove trailing whitespace from all files.

Signed-off-by: Yulu Jia <[email protected]>

.github/ISSUE_TEMPLATE/bug_report.md

@@ -12,9 +12,9 @@ A clear and concise description of what the bug is.
 
 **To Reproduce**
 Steps to reproduce the behavior:
-1. 
-2. 
-3. 
+1.
+2.
+3.
 4. See error
 
 **Expected behavior**

README.md

@@ -65,10 +65,10 @@ python3 setup.py install
 ```
 
 Notes for above two options:
-- BytePS assumes that you have already installed one or more of the following frameworks: TensorFlow / PyTorch / MXNet. 
+- BytePS assumes that you have already installed one or more of the following frameworks: TensorFlow / PyTorch / MXNet.
 - BytePS depends on CUDA and NCCL. You should specify the NCCL path with `export BYTEPS_NCCL_HOME=/path/to/nccl`. By default it points to `/usr/local/nccl`.
 - The installation requires gcc>=4.9. If you are working on CentOS/Redhat and have gcc<4.9, you can try `yum install devtoolset-7` before everything else. In general, we recommend using gcc 4.9 for best compatibility ([an example](https://github.com/bytedance/byteps/blob/3fba75def0d81c1d3225f8f397cc985200f57de7/docker/Dockerfile.mxnet#L72-L80) to pin gcc).
-- RDMA support: During setup, the script will automatically detect the RDMA header file. If you want to use RDMA, make sure your RDMA environment has been properly installed and tested before install ([an example](https://github.com/bytedance/byteps/blob/3fba75def0d81c1d3225f8f397cc985200f57de7/docker/Dockerfile.mxnet#L29-L33) for Ubuntu-18.04). 
+- RDMA support: During setup, the script will automatically detect the RDMA header file. If you want to use RDMA, make sure your RDMA environment has been properly installed and tested before install ([an example](https://github.com/bytedance/byteps/blob/3fba75def0d81c1d3225f8f397cc985200f57de7/docker/Dockerfile.mxnet#L29-L33) for Ubuntu-18.04).
 
 
 ## Use BytePS in Your Code

byteps/common/communicator.cc

@@ -202,7 +202,7 @@ void BytePSCommSocket::startListenThread() {  // only root starts this in
                    << ", signal=" << message.signal << ", key=" << message.key
                    << ", myrank=" << _local_rank;
   }
-  BPS_LOG(DEBUG) << "listen thread joined" 
+  BPS_LOG(DEBUG) << "listen thread joined"
                  << " (rank=" << _local_rank << ")";
 }
 

byteps/common/core_loops.cc

@@ -80,7 +80,7 @@ void FinishOrProceed(std::shared_ptr<TensorTableEntry> task) {
                     << " tensor: " << task->tensor_name
                     << " task->key:" << task->key
                     << " type:" << this_op;
-                            
+
     task->context->part_comm_time[task->key][this_op].back()->dur = (long long)(us.count()) - _ts;
   }
 
@@ -102,7 +102,7 @@ void FinishOrProceed(std::shared_ptr<TensorTableEntry> task) {
       BPS_LOG(TRACE) << "Rank=" << BytePSGlobal::GetRank()
                      << " finish processing tensor: " << task->tensor_name;
       task->callback(Status::OK());
-      //* Add for profiling communication events     
+      //* Add for profiling communication events
       if (task->context->profile_flag) {
         BPS_CHECK(task->context->comm_time.back()->dur == 0)
                     << " tensor: " << task->tensor_name

byteps/common/cpu_reducer.cc

@@ -82,7 +82,7 @@ int CpuReducer::sum(void* dst, void* src, size_t len, DataType dtype) {
       BPS_CHECK(0) << "Unsupported data type: " << dtype;
   }
   return 0;
-}  
+}
 
 template <typename T>
 int CpuReducer::_sum(T* dst, T* src, size_t len) {

byteps/common/global.cc

@@ -117,7 +117,7 @@ void BytePSGlobal::Init() {
                     &_my_role);
 
   _is_root_device = (_my_role == LOCAL_ROOT) ? true : false;
-  
+
   // should round up partition bytes in order to be page aligned
   if (getenv("BYTEPS_PARTITION_BYTES")) {
     _partition_bytes = atoi(getenv("BYTEPS_PARTITION_BYTES"));
@@ -138,7 +138,7 @@ void BytePSGlobal::Init() {
   if (_is_distributed_job) {
     BPS_CHECK(getenv("DMLC_NUM_SERVER"))
         << "error: launch distributed job, but env DMLC_NUM_SERVER not set";
-    
+
     // set hash function
     _hash_knob = std::string(getenv("BYTEPS_KEY_HASH_FN") ? getenv("BYTEPS_KEY_HASH_FN") : "djb2");
     _mixed_mode = getenv("BYTEPS_ENABLE_MIXED_MODE") ? atoi(getenv("BYTEPS_ENABLE_MIXED_MODE")) : false;
@@ -408,7 +408,7 @@ void BytePSGlobal::ReDeclareTensor() {
 void BytePSGlobal::SetProfileFlag(BytePSContext *ctxt) {
   if (_is_trace == 1) {
     // Enable trace, check the start and end step
-    BPS_CHECK(_start_step >= 1 && _end_step > _start_step) 
+    BPS_CHECK(_start_step >= 1 && _end_step > _start_step)
                 << "BYTEPS_TRACE_START_STEP must be larger than 1, "
                 << "BYTEPS_TRACE_END_STEP must be larger than BYTEPS_TRACE_START_STEP.";
     if(ctxt->step_cnt == _start_step-1){
@@ -420,7 +420,7 @@ void BytePSGlobal::SetProfileFlag(BytePSContext *ctxt) {
         std::thread _t(BytePSGlobal::OutputTraces);
         _t.detach();
       }
-    } 
+    }
   } else {
     ctxt->profile_flag = false;
   }
@@ -470,7 +470,7 @@ void BytePSGlobal::OutputTraces(){
   file << "{" << std::endl;
   file << "    \"traceEvents\": [" << std::endl;
   auto first = true;
-  for(std::unordered_map<std::string, int>::iterator iter = _name2end.begin(); 
+  for(std::unordered_map<std::string, int>::iterator iter = _name2end.begin();
     iter != _name2end.end(); iter++){
     BPSContext *ctxt = &_name_to_cxt[iter->first];
     while (ctxt->comm_time.size() > 0) {
@@ -492,7 +492,7 @@ void BytePSGlobal::OutputTraces(){
           BPSCommTime *ret = _part_comm_time_queue.front();
           if (!first) file << ",\n";
           else first = false;
-          BytePSGlobal::EmitTrace(&file, ret, ctxt); 
+          BytePSGlobal::EmitTrace(&file, ret, ctxt);
           _part_comm_time_queue.pop();
         }
         type2part_comm_time.erase(type);
@@ -515,15 +515,15 @@ uint64_t BytePSGlobal::Hash_Mixed_Mode(uint64_t key) {
   size_t num_server_noncolocate = num_server_total-num_worker_total;
   size_t num_server_colocate = num_worker_total;
 
-  // The bound should be larger than num_server_total 
-  // in order to cover each server, but it also 
+  // The bound should be larger than num_server_total
+  // in order to cover each server, but it also
   // cannot be too large because it might cause unbalance
   auto bound = getenv("BYTEPS_MIXED_MODE_BOUND") ? atoi(getenv("BYTEPS_MIXED_MODE_BOUND")) : 101;
-  BPS_CHECK_GE(bound, num_server_total); 
-  auto ratio = (2.0 * num_server_noncolocate * (num_worker_total - 1)) / 
+  BPS_CHECK_GE(bound, num_server_total);
+  auto ratio = (2.0 * num_server_noncolocate * (num_worker_total - 1)) /
                   ((num_worker_total) * (num_worker_total+num_server_noncolocate) - 2 * num_server_noncolocate);
-  BPS_CHECK_LE(ratio, 1) 
-      << "number of (non-colocate servers) > number of (worker)" 
+  BPS_CHECK_LE(ratio, 1)
+      << "number of (non-colocate servers) > number of (worker)"
       << ", which is not permitted in the mixed mode";
   BPS_CHECK_GE(ratio, 0);
   auto threshold = ratio * bound;
@@ -549,7 +549,7 @@ uint64_t BytePSGlobal::Hash_DJB2(uint64_t key) {
   uint64_t hash = 5381;
   int c;
   while ((c = *str)) { // hash(i) = hash(i-1) * 33 ^ str[i]
-    hash = ((hash << 5) + hash) + c; 
+    hash = ((hash << 5) + hash) + c;
     str++;
   }
   return hash;
@@ -587,20 +587,20 @@ PSKV& BytePSGlobal::EncodeDefaultKey(uint64_t key, size_t len) {
     } else if (!_hash_knob.compare(std::string("sdbm"))) {
       server = Hash_SDBM(key) % num_servers;
     } else if (!_hash_knob.compare(std::string("mixed"))) {
-      BPS_CHECK(_mixed_mode) 
+      BPS_CHECK(_mixed_mode)
           << "mixed mode should also set: BYTEPS_ENABLE_MIXED_MODE";
       server = Hash_Mixed_Mode(key);
       CHECK_LT(server, num_servers);
     } else {
       BPS_CHECK(0) << "Unsupported BYTEPS_KEY_HASH_FN, "
                    << "must be one of [naive, built_in, djb2, sdbm]";
     }
-    
+
     _server_accumulated_len[server] += len;
     _total_accumulated_len += len;
     BPS_LOG(DEBUG) << "key " << key << " assigned to server " << server
                    << ", accumulated workload for this server is "
-                   << _server_accumulated_len[server] 
+                   << _server_accumulated_len[server]
                    << " (" << (100.0 * _server_accumulated_len[server] / _total_accumulated_len)
                    << "%)";
 

byteps/common/global.h

@@ -193,7 +193,7 @@ class BytePSGlobal {
   static uint64_t _sample_key;
 
   static int AlignTo(int input, int alignment) { return input / alignment * alignment; }
-  
+
   static int _pagesize;
   static int DivUp(int x, int y) { return (x + y - 1) / y; }
   static int RoundUp(int x, int y) { return DivUp(x, y) * y; }

byteps/common/scheduled_queue.cc

@@ -113,7 +113,7 @@ void BytePSScheduledQueue::recorderTs(std::shared_ptr<TensorTableEntry> task) {
     BPSCommTime *ret = new BPSCommTime;
     ret->start_t = (long long)(us.count());
     ret->key = task->key;
-    ret->type = this_op; 
+    ret->type = this_op;
     context->part_comm_time[task->key][this_op].push(ret);
   }
 }

byteps/misc/imagenet18/__init__.py

@@ -115,7 +115,7 @@ def  __init__(self, params, named_parameters, model, fp16_params, fp32_params, l
         # We use two loops for load-balancing
         for name in self._parameter_names.values():
             declare("Parameter."+name)
-        
+
         self.priorities = {}
         self.gradient_count = 0
 

byteps/mxnet/__init__.py

@@ -102,7 +102,7 @@ def update_multi_precision(self, index, weight, grad, state):
 
             # push delta weight, and pull weight back to the same tensor
             self._do_push_pull_param(index, weight)
-            
+
         else:
             self._do_push_pull(index, grad)
             self._optimizer.update_multi_precision(index, weight, grad, state)

byteps/server/queue.h

@@ -30,7 +30,7 @@ namespace server {
  */
 class PriorityQueue {
  public:
-  PriorityQueue(bool is_schedule) { 
+  PriorityQueue(bool is_schedule) {
     enable_schedule_ = is_schedule;
     if (enable_schedule_) {
       std::make_heap(queue_.begin(), queue_.end(),
@@ -50,8 +50,8 @@ class PriorityQueue {
     mu_.lock();
     queue_.push_back(std::move(new_value));
     if (enable_schedule_) {
-      ++push_cnt_[new_value.key]; 
-      std::push_heap(queue_.begin(), queue_.end(), 
+      ++push_cnt_[new_value.key];
+      std::push_heap(queue_.begin(), queue_.end(),
         [this](const BytePSEngineMessage& a, const BytePSEngineMessage& b) {
           return ComparePriority(a, b);
         }
@@ -69,15 +69,15 @@ class PriorityQueue {
     std::unique_lock<std::mutex> lk(mu_);
     cond_.wait(lk, [this]{return !queue_.empty();});
     if (enable_schedule_) {
-      std::pop_heap(queue_.begin(), queue_.end(), 
+      std::pop_heap(queue_.begin(), queue_.end(),
         [this](const BytePSEngineMessage& a, const BytePSEngineMessage& b) {
           return ComparePriority(a, b);
         }
       );
       *value = queue_.back();
       queue_.pop_back();
     } else {
-      *value = std::move(queue_.front());	
+      *value = std::move(queue_.front());
       queue_.erase(queue_.begin());
     }
   }