Opened 2 years ago
Last modified 2 years ago
#68507 new defect
OpenBLAS: some recent change in Macports broken finding OpenBLAS with Clang builds for some ports
| Reported by: | barracuda156 | Owned by: | |
|---|---|---|---|
| Priority: | Normal | Milestone: | |
| Component: | ports | Version: | 2.8.1 |
| Keywords: | Cc: | NicosPavlov, michaelld (Michael Dickens), catap (Kirill A. Korinsky), mascguy (Christopher Nielsen), Dave-Allured (Dave Allured) | |
| Port: | OpenBLAS, madness |
Description
Something very recently has broken OpenBLAS finding with clang builds for some port(s). See: https://github.com/macports/macports-ports/pull/20901
I tried to upgrade madness, which worked fine locally for me and worked on every system earlier for months: https://ports.macports.org/port/madness/details (well, okay, 10.6 Intel failed, it seems).
Upgrade failed with:
-- Found Gperftools: /opt/local/lib/libtcmalloc.dylib;/opt/local/lib/libprofiler.dylib found components: tcmalloc profiler
CMake Warning at /opt/local/share/cmake-3.24/Modules/FindBoost.cmake:1384 (message):
New Boost version may have incorrect or missing dependencies and imported
targets
Call Stack (most recent call first):
/opt/local/share/cmake-3.24/Modules/FindBoost.cmake:1507 (_Boost_COMPONENT_DEPENDENCIES)
/opt/local/share/cmake-3.24/Modules/FindBoost.cmake:2118 (_Boost_MISSING_DEPENDENCIES)
external/boost.cmake:3 (find_package)
CMakeLists.txt:641 (include)
-- Found Boost: /opt/local/libexec/boost/1.81/include (found suitable version "1.81.0", minimum required is "1.4") found components: math_tr1
-- Looking for sgemm_
-- Looking for sgemm_ - not found
-- Looking for sgemm_
-- Looking for sgemm_ - not found
-- Looking for sgemm_
-- Looking for sgemm_ - not found
-- Looking for sgemm
-- Looking for sgemm - not found
-- Looking for sgemm
-- Looking for sgemm - not found
-- Looking for sgemm
-- Looking for sgemm - not found
-- Looking for sgemm__
-- Looking for sgemm__ - not found
-- Looking for sgemm__
-- Looking for sgemm__ - not found
-- Looking for sgemm__
-- Looking for sgemm__ - not found
-- Looking for SGEMM
-- Looking for SGEMM - not found
-- Looking for SGEMM
-- Looking for SGEMM - not found
-- Looking for SGEMM
-- Looking for SGEMM - not found
-- Looking for SGEMM_
-- Looking for SGEMM_ - not found
-- Looking for SGEMM_
-- Looking for SGEMM_ - not found
-- Looking for SGEMM_
-- Looking for SGEMM_ - not found
CMake Error at external/lapack.cmake:83 (message):
Unable to link against BLAS function. Specify the BLAS library in
LAPACK_LIBRARIES.
Call Stack (most recent call first):
CMakeLists.txt:643 (include)
-- Configuring incomplete, errors occurred!
Initially I attributed that to a recent commit to upstream: https://github.com/m-a-d-n-e-s-s/madness/issues/504
However, when I tried revbumping our existing version of madness, it failed identically. It also fails now on Sonoma for me locally. Building with GCC on 10.6 PowerPC works fine.
I have no idea what goes wrong. Could someone help here?
Attachments (2)
Change History (7)
comment:1 follow-up: 2 Changed 2 years ago by ryandesign (Ryan Carsten Schmidt)
comment:2 Changed 2 years ago by barracuda156
Replying to ryandesign:
madness @2023.09.20 builds fine for me on Monterey x86_64.
Do your cmake logs say anything about why it failed for you?
The failure looks identical to what we saw on CI now.
Changed 2 years ago by barracuda156
| Attachment: | CMakeConfigureLog.txt added |
|---|
Changed 2 years ago by barracuda156
comment:3 Changed 2 years ago by Dave-Allured (Dave Allured)
| Cc: | Dave-Allured added |
|---|
comment:4 Changed 2 years ago by DanielO (Daniel O'Connor)
I think this is due to libopenblas having /opt/local/lib/libgcc listed twice as an RPATH.
I guess this would have to be fixed in the openblas build but I have no idea how hard that would be, or you can pass -Wl,-ld_classic to the link.
(Or you can hand edit the library and change one of the references but I was in a hurry..)
comment:5 Changed 2 years ago by catap (Kirill A. Korinsky)
See: https://github.com/macports/macports-ports/pull/21452 where we figuring out that to do with near the same issue.

madness @2023.09.20 builds fine for me on Monterey x86_64.
Do your cmake logs say anything about why it failed for you?