Search Criteria
Package Details: translatelocally-git r512.1d0d382-1
Package Actions
| Git Clone URL: | https://aur.archlinux.org/translatelocally-git.git (read-only, click to copy) |
|---|---|
| Package Base: | translatelocally-git |
| Description: | Fast and secure translation on your local machine, powered by marian and Bergamot. |
| Upstream URL: | https://translatelocally.com |
| Licenses: | MIT |
| Submitter: | Dheart |
| Maintainer: | Dheart |
| Last Packager: | Dheart |
| Votes: | 1 |
| Popularity: | 0.000006 |
| First Submitted: | 2021-05-03 15:36 (UTC) |
| Last Updated: | 2025-08-11 20:42 (UTC) |
Dependencies (9)
- libarchive (libarchive-gitAUR)
- pcre2 (pcre2-gitAUR)
- qt6-base (qt6-base-gitAUR, qt6-base-scrollfixAUR, qt6-base-scrollfixAUR, qt6-xcb-private-headers-scrollfixAUR, qt6-xcb-private-headers-scrollfixAUR, qt6-base-headlessAUR, qt6-base-hifpsAUR)
- qt6-svg
- cmake (cmake3AUR, cmake-gitAUR) (make)
- gcc14AUR (make)
- git (git-gitAUR, git-glAUR) (make)
- intel-oneapi-mkl (intel-oneapi-hpckitAUR, intel-oneapi-base-toolkit, intel-oneapi-basekit) (make)
- qt6-tools (make)
Latest Comments
« First ‹ Previous 1 2 3 Next › Last »
yochananmarqos commented on 2023-12-07 15:07 (UTC)
@Dheart: I see. Using
Releaseit reduced the package size by 1.61 MiB which I wasn't expecting.My i7-12700H apparently doesn't support AVX512--at least support wasn't detected during build.
Dheart commented on 2023-12-07 02:06 (UTC)
This is a neural network inference toolkit. Everything will work fine even with O0, but it will be slow.
In particular O3 does more loop aggressive loop unrolls and vectorisations which are most apparent on AVX512 systems. If this was to make it to a repository, i totally agree with O2, but since it's self built and caters to the user's specific hardware, I would prefer to keep O3. These are our current build flags that we use everywhere:
https://github.com/browsermt/marian-dev/blob/master/CMakeLists.txt#L361
yochananmarqos commented on 2023-12-07 01:54 (UTC)
@Dheart: Why do you need
O3? Everything seems to be fine without it.Dheart commented on 2023-12-07 01:50 (UTC)
I don't have access to my main desktop for a few weeks, I'll update the package in late December. Thanks again for your explanations!
Dheart commented on 2023-12-07 01:48 (UTC)
Thank you. I will keep the release build as we definitely need O3 (I'm one of the main developers of the upstream project and some of its dependencies)
yochananmarqos commented on 2023-12-07 01:44 (UTC)
@Dheart: See Cmake package guidelines
Dheart commented on 2023-12-07 01:42 (UTC)
@yochananmarqos any reason
CMAKE_BUILD_TYPEis set to None? We set release so we can set the proper optimisation flags. Does this interfere with theCFLAGSandCXXFLAGSset inmakepkg.conf. I've seen other PKGBUILDs dounset CFLAGS && unset CXXFLAGS. Would that be a preferred approach?yochananmarqos commented on 2023-12-07 01:33 (UTC)
@Dheart: The goal is to not have anything downloaded during build. Sources should be in the source() array and anything else like submodules should be done in the prepare() function.
Since there are a lot of submodules and it would be more work to keep track of them all, I've updated my PKGBUILD to just fetch them recursively.
Dheart commented on 2023-12-07 01:09 (UTC)
@yochananmarqos thanks for the improved PKGBUILD. Before I update it, just a quick question: Why put the submodules inside the PKGBUILD? We use a cmake hook to fetch them precisely because we have many and we might add new ones at some point. THE VCS package guidelines do not explain well why the way you have done it is the preferred way of handling.
clangwas a thing because compilation was broken with GCC for a while, haven't checked if it's fixed.yochananmarqos commented on 2023-12-06 22:13 (UTC)
base-develare not required in makedepends(), it's assumed they're already installedintel-oneapi-mkl, it's only required to buildclangis not necessaryImproved PKGBUILD
« First ‹ Previous 1 2 3 Next › Last »