add

oneflow/core/kernel/image_decoder_random_crop_resize_kernel.cpp

@@ -175,12 +175,20 @@ DecodeHandleFactory CreateDecodeHandleFactory<DeviceType::kCPU>(int target_width
 
 #if defined(WITH_NVJPEG)
 
-int GpuDeviceMalloc(void** p, size_t s) { return (int)cudaMalloc(p, s); }
+int GpuDeviceMalloc(void** p, size_t s) { 
+  int err = (int)cudaMalloc(p, s);
+  LOG(ERROR)<<"cudaMalloc size: ";
+  LOG(ERROR)<<s<<" err: "<<err;
+  return err;
+}
 
 int GpuDeviceFree(void* p) { return (int)cudaFree(p); }
 
 int GpuPinnedMalloc(void** p, size_t s, unsigned int flags) {
-  return (int)cudaHostAlloc(p, s, flags);
+  int err = (int)cudaHostAlloc(p, s, flags);
+  LOG(ERROR)<<"cudaHostAlloc size: ";
+  LOG(ERROR)<<"cudaHostAlloc "<<s<<" err: "<<(int)err;
+  return (int)err;
 }
 
 int GpuPinnedFree(void* p) { return (int)cudaFreeHost(p); }
@@ -475,7 +483,11 @@ class Worker final {
       ChannelStatus status = work_queue_.Receive(&work);
       if (status == ChannelStatus::kChannelStatusErrorClosed) { break; }
       CHECK_EQ(status, ChannelStatus::kChannelStatusSuccess);
+      LOG(ERROR)<<"start warmup ";
+      LOG(ERROR)<<"warmup_size "<<warmup_size;
+      LOG(ERROR)<<"work->workspace_size "<<work->workspace_size;
       handle->WarmupOnce(warmup_size, work->workspace, work->workspace_size);
+      LOG(ERROR)<<"warmup done ";
       while (true) {
         const int task_id = work->task_counter->fetch_add(1, std::memory_order_relaxed);
         if (task_id >= work->tasks->size()) { break; }