MIOpen Integrate conv bilinear ck solver lwp 633

Integrate CK's 3D bilinear solvers.

This PR is combination of https://github.com/ROCm/MIOpen/pull/2817 and https://github.com/ROCm/MIOpen/pull/2912

Apr 01 '24 16:04 bghimireamd

@bghimireamd GroupConv3D_BackwardWeights_half_Bilinear_Suite is producing NaN, please check

May 08 '24 19:05 junliume

@bghimireamd please help to resolve the conflicts. Thanks!

May 15 '24 04:05 junliume

@bghimireamd

This PR is combination of #2817 and #2912

Does this mean that all the review comments in #2817 and #2912 are resolved right there or the reviewers have to double their work in the new PR?

/cc @junliume

Jun 06 '24 13:06 atamazov

@atamazov yes issues in #2817 and #2912 were resolved.

Jun 06 '24 13:06 bghimireamd

⚠️ @bghimireamd Unfortuantely I see some bugs in this 3000+ lines PR. But this PR combines new code and NFC refactorings (like code moves), and this makes reviewing terribly hard to me. PLEASE newer do this in the future unless you provide clean instructions for the reviewers how to filter out refactorings.

/cc @amberhassaan

Jun 15 '24 21:06 atamazov

⚠️ @bghimireamd Unfortuantely I see some bugs in this 3000+ lines PR. But this PR combines new code and NFC refactorings (like code moves), and this makes reviewing terribly hard to me. PLEASE newer do this in the future unless you provide clean instructions for the reviewers how to filter out refactorings.

/cc @amberhassaan

I'm with you on that. This PR ran into some planning issues where we had 3 PRs initially but non of them could be merged stand-alone in a way where functionality would get tested, so it was either merge untested functionality or merge the three PRs into one. We went with the latter (due to time crunch). Hopefully we'll plan things better in the future.

Jun 18 '24 12:06 amberhassaan