Upgrade DirectML version to 1.6.0

[mingmingtasd]

I tried to upgrade to 1.8.0 or 1.7.0, but encounter some issues: some Conv2d end2end case will get wrong result value.

[  FAILED  ] Conv2dTests.FusedDepthwiseConv2dDefault
[  FAILED  ] Conv2dTests.FusedDepthwiseConv2dNchwOhwi
[  FAILED  ] Conv2dTests.DepthwiseConv2dWithNchwOihw

So I can only upgrade to 1.6.0 (DML_FEATURE_LEVEL_4_0). I will also raise an issue to DirectML repo. @fujunwei @BruceDai

[fujunwei]

LGTM.

[huningxin]

Is the failure of "check memory leak for DirectML backend (windows)" bot a known issue? @BruceDai @fujunwei

[huningxin]

@mingmingtasd

I will also raise an issue to DirectML repo

What's the issue? Are they issues of DirectML or WebNN-native implementation? Do you have any investigations on the root cause?

I am fine to upgrade to a version with no issues, but we eventually need to support the latest DirectML release.

[mingmingtasd]

@mingmingtasd

I will also raise an issue to DirectML repo

What's the issue? Are they issues of DirectML or WebNN-native implementation? Do you have any investigations on the root cause?

I am fine to upgrade to a version with no issues, but we eventually need to support the latest DirectML release.

I am doing investigations now, will try to re-produce with the DirectML samples. I think it looks like a DirectML issue. @huningxin

[fujunwei]

Is the failure of "check memory leak for DirectML backend (windows)" bot a known issue? @BruceDai @fujunwei

Should we revert the CI with #236?

[mingmingtasd]

I have confirmed that the conv2d issue on DirectML 1.7.0/1.80 should be their own issue. I have provided a case using their python sample to reproduce the issue and raise issue #234 to them. @fujunwei @huningxin

[fujunwei]

Merge it.