[Convnets] 21.09 Fixes

Author: hXl3s

MxNet/Classification/RN50v1.5/dali.py

@@ -36,6 +36,7 @@ def add_dali_args(parser):
 
     group.add_argument('--dali-nvjpeg-width-hint', type=int, default=5980, help="Width hint value for nvJPEG (in pixels)")
     group.add_argument('--dali-nvjpeg-height-hint', type=int, default=6430, help="Height hint value for nvJPEG (in pixels)")
+    group.add_argument('--dali-dont-use-mmap', default=False, action='store_true', help="Use plain I/O instead of MMAP for datasets")
     return parser
 
 
@@ -47,7 +48,8 @@ def __init__(self, args, batch_size, num_threads, device_id, rec_path, idx_path,
                  ):
         super(HybridTrainPipe, self).__init__(batch_size, num_threads, device_id, seed=12 + device_id, prefetch_queue_depth = prefetch_queue)
         self.input = ops.MXNetReader(path=[rec_path], index_path=[idx_path],
-                                     random_shuffle=True, shard_id=shard_id, num_shards=num_shards)
+                                     random_shuffle=True, shard_id=shard_id, num_shards=num_shards,
+                                     dont_use_mmap=args.dali_dont_use_mmap)
 
         if dali_cpu:
             dali_device = "cpu"
@@ -101,7 +103,8 @@ def __init__(self, args, batch_size, num_threads, device_id, rec_path, idx_path,
                  nvjpeg_width_hint=5980, nvjpeg_height_hint=6430):
         super(HybridValPipe, self).__init__(batch_size, num_threads, device_id, seed=12 + device_id, prefetch_queue_depth=prefetch_queue)
         self.input = ops.MXNetReader(path=[rec_path], index_path=[idx_path],
-                                     random_shuffle=False, shard_id=shard_id, num_shards=num_shards)
+                                     random_shuffle=False, shard_id=shard_id, num_shards=num_shards,
+                                     dont_use_mmap=args.dali_dont_use_mmap)
 
         if dali_cpu:
             dali_device = "cpu"