MIOpen icon indicating copy to clipboard operation
MIOpen copied to clipboard
verified publisher icon ROCmSoftwarePlatform

Implement Fold and Unfold

Open DuongQLee opened this issue 1 year ago • 7 comments

  • Added Fold and Unfold op.
  • Full benchmark result compared to ROCm Here
  • Average performance:

Op Dtype Direction Time
Unfold fp32 fwd 16.43641979
fp32 bwd 1.458798342
fp16 fwd 15.83955361
fp16 bwd 1.459763543
bfp16 fwd 15.90593279
bfp16 bwd 1.455323877
Fold fp32 fwd 1.463731927
fp32 bwd 28.09828887
fp16 fwd 1.479315364
fp16 bwd 28.08557933
bfp16 fwd 1.47158127
bfp16 bwd 26.84660993

DuongQLee avatar Jul 29 '24 09:07 DuongQLee

Erro logs: error_log.txt

iq136boy avatar Jul 31 '24 01:07 iq136boy

@iq136boy Can I have the error log for Jenkins - Fp32 Hip All gfx90a? Thank you.

DuongQLee avatar Aug 07 '24 08:08 DuongQLee

@iq136boy Would you send us the build log of this PR?

long10024070 avatar Aug 20 '24 03:08 long10024070

Please check the error log: test_error_log.txt

iq136boy avatar Aug 28 '24 01:08 iq136boy

I will handle this PR since @DuongQLee is now part of another team.

hieule88 avatar Oct 16 '24 07:10 hieule88

@iq136boy Would you send us the build log of this PR?

hieule88 avatar Oct 21 '24 04:10 hieule88

Error log: 3167_log.txt

I fixed this error. Please send me the newest error log of this PR.

hieule88 avatar Nov 25 '24 07:11 hieule88

MIOpen is moving to the new monorepo setup and all older unmerged PR's are being closed. Please re-open this as part of the new repo if these changes are still needed.

BradPepersAMD avatar Jul 14 '25 06:07 BradPepersAMD