HomeSort by relevance Sort by last modified time
    Searched refs:grabbed (Results 1 - 2 of 2) sorted by null

  /build/make/tools/
fat16copy.py 668 grabbed = 0
670 while grabbed < amount and len(free_zones) > 0:
672 grabbed += zone[1] * self.bytes_per_cluster
675 if grabbed < amount:
678 excess = (grabbed - amount) / self.bytes_per_cluster
  /external/tensorflow/tensorflow/contrib/verbs/
README.md 62 In order to allocate the result and proxy tensors, we need to know the tensor's meta-data, i.e. shape and data-type for DMAable tensors, and proto-size for serialized tensors. Unfortunately, this information is only available on the sender side which complicates manners. In order to avoid sending extra messages for querying the meta-data at each step, we store a local meta-data cache per tensor, which will only be update upon changes. Based on the assumption that the meta-data of a tensor rarely changes between steps, we expect that on most times the cache will only be updated once. The sender is responsible to detect changes in the meta-data, and update the receiver. In order for the sender to know that the meta-data had changed, each **RDMA_MESSAGE_TENSOR_REQUEST** will contain the meta-data that the receiver had grabbed from the local cache. The sender will then compare the meta-data from the message to the tensor's new meta-data.

Completed in 214 milliseconds