fix reshard of empty tensor and refine local shape calculation

[zhiqiu]

PR Category

Auto Parallel

PR Types

Bug fixes

Description

As the title

The following case raise Error as below,

a = paddle.rand([16, 32])
dist_a = dist.shard_tensor(a, mesh, [dist.Shard(0)])
b = paddle.rand([16, 32])
dist_b = dist.shard_tensor(b, mesh, [dist.Shard(1)])

a.stop_gradient = False
b.stop_gradient = False
dist_a.stop_gradient = False
dist_a.stop_gradient = False

sum = paddle.add(dist_a, dist_b)

#####
PreconditionNotMetError: Tensor's dimension is out of bound.Tensor's dimension must be equal or less than the size of its memory.But received Tensor's dimension is 1024, memory's size is 0.
  [Hint: Expected numel() * SizeOf(dtype()) <= memory_size(), but received numel() * SizeOf(dtype()):1024 > memory_size():0.] (at /paddle/paddle/phi/core/dense_tensor_impl.cc:55)

I0425 11:07:20.154877 21600 tcp_store.cc:293] receive shutdown event and so quit from MasterDaemon run loop
LAUNCH INFO 2024-04-25 11:07:21,250 Exit code 1

This PR fixes that.

Pcard-76459

[paddle-bot]

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[heavyrain-lzy]

LGTM