On Wed, 27 Jan 2021 at 10:59, Nilesh Patra <nil...@debian.org> wrote:
> Hi Michael > > On Wed, 27 Jan 2021 at 14:51, Michael R. Crusoe <cru...@debian.org> wrote: > >> Hello, >> >> After scheduling a binary rebuild[0] of packages using libsimde-dev >> > > First off, thanks a lot for doing this! > > >> I noticed that there are a few new packages using SIMDe (yay!) but aren't >> listed on https://wiki.debian.org/SIMDEverywhere >> > > It is at my end, I added simde support to plast, ngmlr and scrappie but > completely missed adding them to the wiki, sorry about that. > I actually wanted to know answers of a couple of questions before I open a > PR for them upstream: > > * In the simde wiki, point no.3, it is written to remove -msse3, > -march=native et al flags, however in a few PRs that you opened up, this[1] > for example, I do not see this change > Yeah, for Debian we need to remove them, but I don't want to enforce that upstream. > it is however appended with a -D__SIM_SSE3 - does this allow the > -msse3 flag to work well? And am I expected to do the same stuff in the > upstream build system? > That is a package specific flag. Most PRs are simpler than that one. This particular flag was removed later in the PR, so it is no longer needed: https://github.com/stamatak/standard-RAxML/pull/50/files#diff-1a92a17afd64b6b3dd4626377cd6bed402fbc85276bbcf60581a330fa6936187L5934 The PR is mainly to start the conversation with upstream. Some will be interested and will adapt and accept the PR. Some upstream maintainers will ignore it. > * There is no need to add in arguments: "-DSIMDE_ENABLE_OPENMP > -fopenmp-simd" into the upstream build system right? > That's a gcc specific invocation. If they already have code to differentiate between gcc, clang, icc, and other compilers, then slipping it into the gcc section makes sense. Otherwise it can be left out unless the project explicitly only supports gcc. > Just to be sure that I'm doing it the right way, could you please review > the PR I sent for plast[2] (sent this PR last month)? > (Building the upstream repository locally, it looked okay) > Sure! I apologize if these questions might somehow sound stupid, but I do not > know SIMDe's working well enough to take a judgement without being clear. > Not stupid at all, they are good questions. Feel free to update the wiki with what you learn. > > [1]: > https://github.com/stamatak/standard-RAxML/pull/50/files#diff-8bf5a567e3d917cc040b7bf12f499844b30961ed930b421b717d25039efdd94aR6 > [2]: https://github.com/PLAST-software/plast-library/pull/8 > > Thanks a lot, > Nilesh >
