Delete Device2Host caused by comm with device and host

[zhaozheng09]

Delete some D2H ops

if not will be bring a sync within host and device, so replace it from if not to torch.where

Before submitting

• Was this discussed/agreed via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

before:

after:

Host will be block until device done, this D2H maybe block pipeline of host and device, so I use where op to replace if not

[Borda]

can we pls add tests for this fix?
and maybe pls update the title to better express what the PR does

[codecov]

Codecov Report

Attention: Patch coverage is 66.66667% with 4 lines in your changes missing coverage. Please review.

Project coverage is 41%. Comparing base (d528131) to head (5146a6d).

❗ There is a different number of reports uploaded between BASE (d528131) and HEAD (5146a6d). Click for more details.

HEAD has 76 uploads less than BASE

Flag	BASE (d528131)	HEAD (5146a6d)
gpu	2	0
unittest	2	0
torch2.0.1+cpu	6	3
python3.10	19	9
Windows	4	2
cpu	32	14
macOS	6	3
python3.12	7	3
torch2.5.0	2	1
torch2.0.1	4	2
torch2.5.0+cpu	2	1
Linux	22	9
python3.11	4	1
torch2.4.1+cu121	7	2
torch2.1.2+cpu	2	1
python3.9	2	1
torch2.2.2+cpu	2	1
torch2.3.1+cpu	2	1
torch2.5.0+cu124	5	2

Additional details and impacted files

@@           Coverage Diff            @@
##           master   #2840     +/-   ##
========================================
- Coverage      69%     41%    -28%     
========================================
  Files         346     332     -14     
  Lines       19129   18961    -168     
========================================
- Hits        13227    7739   -5488     
- Misses       5902   11222   +5320     

[zhaozheng09]

can we pls add tests for this fix? and maybe pls update the title to better express what the PR does

must be 100% cover ? this code will raise in old torch version, I don't know how to cover, please give me an example ? thx very much.

[Borda]

this code will raise in old torch version, I don't know how to cover, please give me an example ? thx very much

ok, so you say that this PR does not have any impact on functional metrics results but improves performances, correct?