Update GPU HCAL conditions framework

[fwyzard]

Address the comments from review of cms-sw#32039.

Remove duplicate GPU-related HCAL conditions records, and simplify the package dependencies moving the remaining ones to CondFormats/DataRecord.

Improve the handling of the GPU conditions payloads:

• use cms::cuda::device::unique_ptr to automatically deallocate the memory;
• use edm::propagate_const_array to ensure that the conditions data are not accidentally modified by client code;
• use cms::cuda::copyAsync(...) to simplify the copy of the conditions to the device.

[fwyzard]

If these changes look reasonable, I plan to extend them from HcalConvertedPedestalWidthsGPU to the other ESProducts.

[fwyzard]

@mariadalfonso @vkhristenko if you are fine with the changes, I'll work on the other ESProducts next.

[vkhristenko]

np from my side

On Thu, 19 Nov 2020 at 19:35, Andrea Bocci notifications@github.com wrote:

@mariadalfonso https://github.com/mariadalfonso @vkhristenko https://github.com/vkhristenko if you are fine with the changes, I'll work on the other ESProducts next.

— You are receiving this because you were mentioned.

Reply to this email directly, view it on GitHub https://github.com/cms-patatrack/cmssw/pull/574#issuecomment-730560879, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABSFUCKNU4YAAZ3NT5BTPZDSQVQQVANCNFSM4T3L3I3Q .

[mariadalfonso]

looks ok to me. I have not tested the changes but I assume are transparent

-  setup.get<HcalConvertedEffectivePedestalWidthsRcd>().get(effectivePedestalWidthsHandle);
+  setup.get<HcalConvertedPedestalWidthsRcd>().get(effectivePedestalWidthsHandle);

[mariadalfonso]

by the way similarly should be done for the conditions in ECAL

[fwyzard]

Validated on top of CMSSW_11_2_0_pre9

[fwyzard]

OK, I think I've addressed all the comments on edm::propage_const_array ?

[fwyzard]

According to StackOverflow the defaulted constructors are automatically constexpr, if possible.

I can add it explicitly if you prefer, it compiles fine.

[Dr15Jones]

According to StackOverflow the defaulted constructors are automatically constexpr, if possible.

I did a quick check on godbolt and StackOverflow appears correct :).

I can add it explicitly if you prefer, it compiles fine.

I guess I don't mind either way.

[fwyzard]

I'll add it - for consistency with https://en.cppreference.com/w/cpp/experimental/propagate_const/propagate_const :-)

.A