Bugfix for issue #153

[mjklemm]

While working on the new offload code for TCE CCSD(T), I have
removed some of the global variables that aparentely are being
used for MRCC. This commit reverts these changes and brings
the code back to a compilable state.

[edoapra]

@mjklemm This still does not make a successfull link when USE_OFFLOAD=1 & MRCC_METHODS=1

tce_mrcc_ccsdpt_subg.F:(.text+0x43c): undefined reference to `offl_alloc_'
tce_mrcc_ccsdpt_subg.F:(.text+0x928): undefined reference to `offl_zerofill_'
tce_mrcc_ccsdpt_subg.F:(.text+0x159f): undefined reference to `offl_free_'

[mjklemm]

What's your preference here? Rip out the code or make it functional again?

[chiensh]

It still won’t link when USE_OFFLOAD=0, unless comment out line 195 in tce_mrcc_ccsdpt_subg.F, which calls offs_zerofill regardless the value of USE_OFFLOAD 190 #if 0 191 call offl_zerofill(dbl_mb(k_singles),triplesx_mxlgth) 192 #endif 193 194 call dcopy(triplesx_mxlgth, 0.0d0, 0, dbl_mb(k_doubles), 1) 195 call offl_zerofill(dbl_mb(k_doubles),triplesx_mxlgth) 196 c -----------

…

On 20 Sep 2019, at 04:53, Edoardo Aprà ***@***.***> wrote: @mjklemm <https://github.com/mjklemm> This still does not make a successfull link when USE_OFFLOAD=1 & MRCC_METHODS=1 tce_mrcc_ccsdpt_subg.F:(.text+0x43c): undefined reference to `offl_alloc_' tce_mrcc_ccsdpt_subg.F:(.text+0x928): undefined reference to `offl_zerofill_' tce_mrcc_ccsdpt_subg.F:(.text+0x159f): undefined reference to `offl_free_' — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#154?email_source=notifications&email_token=ADLGF7EOSMICLVV2UPNKVELQKPREVA5CNFSM4IYNJCE2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD7EZOMY#issuecomment-533305139>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADLGF7G5DZWOUT5K6LMCC5TQKPREVANCNFSM4IYNJCEQ>.

[edoapra]

What's your preference here? Rip out the code or make it functional again?

@mjklemm We should use the same approach taken for ccsd_t
the mrcc offload was pretty much a clone of the ccsd_t offload

[mjklemm]

@mjklemm We should use the same approach taken for ccsd_t
the mrcc offload was pretty much a clone of the ccsd_t offload

OK, I can do that, but that will take a few days to work through the code. Would have a test case for me (pls send via email) that I can use to make sure that I'm not breaking anything? I would need something quick (less than 1-2 minutes).

[edoapra]

@chiensh
Dominic,
The fix I just pushd should address this problem ec60ebb

It still won’t link when USE_OFFLOAD=0, unless comment out line 195 in tce_mrcc_ccsdpt_subg.F, which calls offs_zerofill regardless the value of USE_OFFLOAD 190 #if 0 191 call offl_zerofill(dbl_mb(k_singles),triplesx_mxlgth) 192 #endif 193 194 call dcopy(triplesx_mxlgth, 0.0d0, 0, dbl_mb(k_doubles), 1) 195 call offl_zerofill(dbl_mb(k_doubles),triplesx_mxlgth) 196 c -----------
…

[edoapra]

This test takes less than a minute on a single process run on a Xeon W-2102
mrcc.nw.txt
mrcc.out.txt

OK, I can do that, but that will take a few days to work through the code. Would have a test case for me (pls send via email) that I can use to make sure that I'm not breaking anything? I would need something quick (less than 1-2 minutes).